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.
6 * Copyright 2012 Google, Inc.
13 #include "writeback.h"
15 #include <linux/module.h>
16 #include <linux/hash.h>
17 #include <linux/random.h>
18 #include <linux/backing-dev.h>
20 #include <trace/events/bcache.h>
22 #define CUTOFF_CACHE_ADD 95
23 #define CUTOFF_CACHE_READA 90
25 struct kmem_cache *bch_search_cache;
27 static void bch_data_insert_start(struct closure *);
29 static unsigned cache_mode(struct cached_dev *dc)
31 return BDEV_CACHE_MODE(&dc->sb);
34 static bool verify(struct cached_dev *dc)
39 static void bio_csum(struct bio *bio, struct bkey *k)
42 struct bvec_iter iter;
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);
51 k->ptr[KEY_PTRS(k)] = csum & (~0ULL >> 1);
54 /* Insert data into cache */
56 static void bch_data_insert_keys(struct closure *cl)
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;
64 * If we're looping, might already be waiting on
65 * another journal write - can't wait on more than one journal write at
68 * XXX: this looks wrong
71 while (atomic_read(&s->cl.remaining) & CLOSURE_WAITING)
76 journal_ref = bch_journal(op->c, &op->insert_keys,
77 op->flush_journal ? cl : NULL);
79 ret = bch_btree_insert(op->c, &op->insert_keys,
80 journal_ref, replace_key);
82 op->replace_collision = true;
84 op->status = BLK_STS_RESOURCE;
85 op->insert_data_done = true;
89 atomic_dec_bug(journal_ref);
91 if (!op->insert_data_done) {
92 continue_at(cl, bch_data_insert_start, op->wq);
96 bch_keylist_free(&op->insert_keys);
100 static int bch_keylist_realloc(struct keylist *l, unsigned u64s,
103 size_t oldsize = bch_keylist_nkeys(l);
104 size_t newsize = oldsize + u64s;
107 * The journalling code doesn't handle the case where the keys to insert
108 * is bigger than an empty write: If we just return -ENOMEM here,
109 * bio_insert() and bio_invalidate() will insert the keys created so far
110 * and finish the rest when the keylist is empty.
112 if (newsize * sizeof(uint64_t) > block_bytes(c) - sizeof(struct jset))
115 return __bch_keylist_realloc(l, u64s);
118 static void bch_data_invalidate(struct closure *cl)
120 struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
121 struct bio *bio = op->bio;
123 pr_debug("invalidating %i sectors from %llu",
124 bio_sectors(bio), (uint64_t) bio->bi_iter.bi_sector);
126 while (bio_sectors(bio)) {
127 unsigned sectors = min(bio_sectors(bio),
128 1U << (KEY_SIZE_BITS - 1));
130 if (bch_keylist_realloc(&op->insert_keys, 2, op->c))
133 bio->bi_iter.bi_sector += sectors;
134 bio->bi_iter.bi_size -= sectors << 9;
136 bch_keylist_add(&op->insert_keys,
137 &KEY(op->inode, bio->bi_iter.bi_sector, sectors));
140 op->insert_data_done = true;
143 continue_at(cl, bch_data_insert_keys, op->wq);
146 static void bch_data_insert_error(struct closure *cl)
148 struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
151 * Our data write just errored, which means we've got a bunch of keys to
152 * insert that point to data that wasn't succesfully written.
154 * We don't have to insert those keys but we still have to invalidate
155 * that region of the cache - so, if we just strip off all the pointers
156 * from the keys we'll accomplish just that.
159 struct bkey *src = op->insert_keys.keys, *dst = op->insert_keys.keys;
161 while (src != op->insert_keys.top) {
162 struct bkey *n = bkey_next(src);
164 SET_KEY_PTRS(src, 0);
165 memmove(dst, src, bkey_bytes(src));
167 dst = bkey_next(dst);
171 op->insert_keys.top = dst;
173 bch_data_insert_keys(cl);
176 static void bch_data_insert_endio(struct bio *bio)
178 struct closure *cl = bio->bi_private;
179 struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
181 if (bio->bi_status) {
182 /* TODO: We could try to recover from this. */
184 op->status = bio->bi_status;
185 else if (!op->replace)
186 set_closure_fn(cl, bch_data_insert_error, op->wq);
188 set_closure_fn(cl, NULL, NULL);
191 bch_bbio_endio(op->c, bio, bio->bi_status, "writing data to cache");
194 static void bch_data_insert_start(struct closure *cl)
196 struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
197 struct bio *bio = op->bio, *n;
200 return bch_data_invalidate(cl);
202 if (atomic_sub_return(bio_sectors(bio), &op->c->sectors_to_gc) < 0)
206 * Journal writes are marked REQ_PREFLUSH; if the original write was a
207 * flush, it'll wait on the journal write.
209 bio->bi_opf &= ~(REQ_PREFLUSH|REQ_FUA);
214 struct bio_set *split = op->c->bio_split;
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),
220 continue_at(cl, bch_data_insert_keys, op->wq);
224 k = op->insert_keys.top;
226 SET_KEY_INODE(k, op->inode);
227 SET_KEY_OFFSET(k, bio->bi_iter.bi_sector);
229 if (!bch_alloc_sectors(op->c, k, bio_sectors(bio),
230 op->write_point, op->write_prio,
234 n = bio_next_split(bio, KEY_SIZE(k), GFP_NOIO, split);
236 n->bi_end_io = bch_data_insert_endio;
240 SET_KEY_DIRTY(k, true);
242 for (i = 0; i < KEY_PTRS(k); i++)
243 SET_GC_MARK(PTR_BUCKET(op->c, k, i),
247 SET_KEY_CSUM(k, op->csum);
251 trace_bcache_cache_insert(k);
252 bch_keylist_push(&op->insert_keys);
254 bio_set_op_attrs(n, REQ_OP_WRITE, 0);
255 bch_submit_bbio(n, op->c, k, 0);
258 op->insert_data_done = true;
259 continue_at(cl, bch_data_insert_keys, op->wq);
262 /* bch_alloc_sectors() blocks if s->writeback = true */
263 BUG_ON(op->writeback);
266 * But if it's not a writeback write we'd rather just bail out if
267 * there aren't any buckets ready to write to - it might take awhile and
268 * we might be starving btree writes for gc or something.
273 * Writethrough write: We can't complete the write until we've
274 * updated the index. But we don't want to delay the write while
275 * we wait for buckets to be freed up, so just invalidate the
279 return bch_data_invalidate(cl);
282 * From a cache miss, we can just insert the keys for the data
283 * we have written or bail out if we didn't do anything.
285 op->insert_data_done = true;
288 if (!bch_keylist_empty(&op->insert_keys))
289 continue_at(cl, bch_data_insert_keys, op->wq);
296 * bch_data_insert - stick some data in the cache
298 * This is the starting point for any data to end up in a cache device; it could
299 * be from a normal write, or a writeback write, or a write to a flash only
300 * volume - it's also used by the moving garbage collector to compact data in
301 * mostly empty buckets.
303 * It first writes the data to the cache, creating a list of keys to be inserted
304 * (if the data had to be fragmented there will be multiple keys); after the
305 * data is written it calls bch_journal, and after the keys have been added to
306 * the next journal write they're inserted into the btree.
308 * It inserts the data in s->cache_bio; bi_sector is used for the key offset,
309 * and op->inode is used for the key inode.
311 * If s->bypass is true, instead of inserting the data it invalidates the
312 * region of the cache represented by s->cache_bio and op->inode.
314 void bch_data_insert(struct closure *cl)
316 struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
318 trace_bcache_write(op->c, op->inode, op->bio,
319 op->writeback, op->bypass);
321 bch_keylist_init(&op->insert_keys);
323 bch_data_insert_start(cl);
328 unsigned bch_get_congested(struct cache_set *c)
333 if (!c->congested_read_threshold_us &&
334 !c->congested_write_threshold_us)
337 i = (local_clock_us() - c->congested_last_us) / 1024;
341 i += atomic_read(&c->congested);
348 i = fract_exp_two(i, 6);
350 rand = get_random_int();
351 i -= bitmap_weight(&rand, BITS_PER_LONG);
353 return i > 0 ? i : 1;
356 static void add_sequential(struct task_struct *t)
358 ewma_add(t->sequential_io_avg,
359 t->sequential_io, 8, 0);
361 t->sequential_io = 0;
364 static struct hlist_head *iohash(struct cached_dev *dc, uint64_t k)
366 return &dc->io_hash[hash_64(k, RECENT_IO_BITS)];
369 static bool check_should_bypass(struct cached_dev *dc, struct bio *bio)
371 struct cache_set *c = dc->disk.c;
372 unsigned mode = cache_mode(dc);
373 unsigned sectors, congested = bch_get_congested(c);
374 struct task_struct *task = current;
377 if (test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags) ||
378 c->gc_stats.in_use > CUTOFF_CACHE_ADD ||
379 (bio_op(bio) == REQ_OP_DISCARD))
382 if (mode == CACHE_MODE_NONE ||
383 (mode == CACHE_MODE_WRITEAROUND &&
384 op_is_write(bio_op(bio))))
388 * Flag for bypass if the IO is for read-ahead or background,
389 * unless the read-ahead request is for metadata (eg, for gfs2).
391 if (bio->bi_opf & (REQ_RAHEAD|REQ_BACKGROUND) &&
392 !(bio->bi_opf & REQ_META))
395 if (bio->bi_iter.bi_sector & (c->sb.block_size - 1) ||
396 bio_sectors(bio) & (c->sb.block_size - 1)) {
397 pr_debug("skipping unaligned io");
401 if (bypass_torture_test(dc)) {
402 if ((get_random_int() & 3) == 3)
408 if (!congested && !dc->sequential_cutoff)
411 spin_lock(&dc->io_lock);
413 hlist_for_each_entry(i, iohash(dc, bio->bi_iter.bi_sector), hash)
414 if (i->last == bio->bi_iter.bi_sector &&
415 time_before(jiffies, i->jiffies))
418 i = list_first_entry(&dc->io_lru, struct io, lru);
420 add_sequential(task);
423 if (i->sequential + bio->bi_iter.bi_size > i->sequential)
424 i->sequential += bio->bi_iter.bi_size;
426 i->last = bio_end_sector(bio);
427 i->jiffies = jiffies + msecs_to_jiffies(5000);
428 task->sequential_io = i->sequential;
431 hlist_add_head(&i->hash, iohash(dc, i->last));
432 list_move_tail(&i->lru, &dc->io_lru);
434 spin_unlock(&dc->io_lock);
436 sectors = max(task->sequential_io,
437 task->sequential_io_avg) >> 9;
439 if (dc->sequential_cutoff &&
440 sectors >= dc->sequential_cutoff >> 9) {
441 trace_bcache_bypass_sequential(bio);
445 if (congested && sectors >= congested) {
446 trace_bcache_bypass_congested(bio);
451 bch_rescale_priorities(c, bio_sectors(bio));
454 bch_mark_sectors_bypassed(c, dc, bio_sectors(bio));
461 /* Stack frame for bio_complete */
465 struct bio *orig_bio;
466 struct bio *cache_miss;
467 struct bcache_device *d;
469 unsigned insert_bio_sectors;
470 unsigned recoverable:1;
472 unsigned read_dirty_data:1;
473 unsigned cache_missed:1;
475 unsigned long start_time;
478 struct data_insert_op iop;
481 static void bch_cache_read_endio(struct bio *bio)
483 struct bbio *b = container_of(bio, struct bbio, bio);
484 struct closure *cl = bio->bi_private;
485 struct search *s = container_of(cl, struct search, cl);
488 * If the bucket was reused while our bio was in flight, we might have
489 * read the wrong data. Set s->error but not error so it doesn't get
490 * counted against the cache device, but we'll still reread the data
491 * from the backing device.
495 s->iop.status = bio->bi_status;
496 else if (!KEY_DIRTY(&b->key) &&
497 ptr_stale(s->iop.c, &b->key, 0)) {
498 atomic_long_inc(&s->iop.c->cache_read_races);
499 s->iop.status = BLK_STS_IOERR;
502 bch_bbio_endio(s->iop.c, bio, bio->bi_status, "reading from cache");
506 * Read from a single key, handling the initial cache miss if the key starts in
507 * the middle of the bio
509 static int cache_lookup_fn(struct btree_op *op, struct btree *b, struct bkey *k)
511 struct search *s = container_of(op, struct search, op);
512 struct bio *n, *bio = &s->bio.bio;
513 struct bkey *bio_key;
516 if (bkey_cmp(k, &KEY(s->iop.inode, bio->bi_iter.bi_sector, 0)) <= 0)
519 if (KEY_INODE(k) != s->iop.inode ||
520 KEY_START(k) > bio->bi_iter.bi_sector) {
521 unsigned bio_sectors = bio_sectors(bio);
522 unsigned sectors = KEY_INODE(k) == s->iop.inode
523 ? min_t(uint64_t, INT_MAX,
524 KEY_START(k) - bio->bi_iter.bi_sector)
527 int ret = s->d->cache_miss(b, s, bio, sectors);
528 if (ret != MAP_CONTINUE)
531 /* if this was a complete miss we shouldn't get here */
532 BUG_ON(bio_sectors <= sectors);
538 /* XXX: figure out best pointer - for multiple cache devices */
541 PTR_BUCKET(b->c, k, ptr)->prio = INITIAL_PRIO;
544 s->read_dirty_data = true;
546 n = bio_next_split(bio, min_t(uint64_t, INT_MAX,
547 KEY_OFFSET(k) - bio->bi_iter.bi_sector),
548 GFP_NOIO, s->d->bio_split);
550 bio_key = &container_of(n, struct bbio, bio)->key;
551 bch_bkey_copy_single_ptr(bio_key, k, ptr);
553 bch_cut_front(&KEY(s->iop.inode, n->bi_iter.bi_sector, 0), bio_key);
554 bch_cut_back(&KEY(s->iop.inode, bio_end_sector(n), 0), bio_key);
556 n->bi_end_io = bch_cache_read_endio;
557 n->bi_private = &s->cl;
560 * The bucket we're reading from might be reused while our bio
561 * is in flight, and we could then end up reading the wrong
564 * We guard against this by checking (in cache_read_endio()) if
565 * the pointer is stale again; if so, we treat it as an error
566 * and reread from the backing device (but we don't pass that
567 * error up anywhere).
570 __bch_submit_bbio(n, b->c);
571 return n == bio ? MAP_DONE : MAP_CONTINUE;
574 static void cache_lookup(struct closure *cl)
576 struct search *s = container_of(cl, struct search, iop.cl);
577 struct bio *bio = &s->bio.bio;
580 bch_btree_op_init(&s->op, -1);
582 ret = bch_btree_map_keys(&s->op, s->iop.c,
583 &KEY(s->iop.inode, bio->bi_iter.bi_sector, 0),
584 cache_lookup_fn, MAP_END_KEY);
585 if (ret == -EAGAIN) {
586 continue_at(cl, cache_lookup, bcache_wq);
593 /* Common code for the make_request functions */
595 static void request_endio(struct bio *bio)
597 struct closure *cl = bio->bi_private;
599 if (bio->bi_status) {
600 struct search *s = container_of(cl, struct search, cl);
601 s->iop.status = bio->bi_status;
602 /* Only cache read errors are recoverable */
603 s->recoverable = false;
610 static void bio_complete(struct search *s)
613 struct request_queue *q = s->orig_bio->bi_disk->queue;
614 generic_end_io_acct(q, bio_data_dir(s->orig_bio),
615 &s->d->disk->part0, s->start_time);
617 trace_bcache_request_end(s->d, s->orig_bio);
618 s->orig_bio->bi_status = s->iop.status;
619 bio_endio(s->orig_bio);
624 static void do_bio_hook(struct search *s, struct bio *orig_bio)
626 struct bio *bio = &s->bio.bio;
628 bio_init(bio, NULL, 0);
629 __bio_clone_fast(bio, orig_bio);
630 bio->bi_end_io = request_endio;
631 bio->bi_private = &s->cl;
636 static void search_free(struct closure *cl)
638 struct search *s = container_of(cl, struct search, cl);
644 closure_debug_destroy(cl);
645 mempool_free(s, s->d->c->search);
648 static inline struct search *search_alloc(struct bio *bio,
649 struct bcache_device *d)
653 s = mempool_alloc(d->c->search, GFP_NOIO);
655 closure_init(&s->cl, NULL);
659 s->cache_miss = NULL;
663 s->write = op_is_write(bio_op(bio));
664 s->read_dirty_data = 0;
665 s->start_time = jiffies;
669 s->iop.inode = d->id;
670 s->iop.write_point = hash_long((unsigned long) current, 16);
671 s->iop.write_prio = 0;
674 s->iop.flush_journal = op_is_flush(bio->bi_opf);
675 s->iop.wq = bcache_wq;
682 static void cached_dev_bio_complete(struct closure *cl)
684 struct search *s = container_of(cl, struct search, cl);
685 struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
693 static void cached_dev_cache_miss_done(struct closure *cl)
695 struct search *s = container_of(cl, struct search, cl);
697 if (s->iop.replace_collision)
698 bch_mark_cache_miss_collision(s->iop.c, s->d);
701 bio_free_pages(s->iop.bio);
703 cached_dev_bio_complete(cl);
706 static void cached_dev_read_error(struct closure *cl)
708 struct search *s = container_of(cl, struct search, cl);
709 struct bio *bio = &s->bio.bio;
710 struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
713 * If cache device is dirty (dc->has_dirty is non-zero), then
714 * recovery a failed read request from cached device may get a
715 * stale data back. So read failure recovery is only permitted
716 * when cache device is clean.
718 if (s->recoverable &&
719 (dc && !atomic_read(&dc->has_dirty))) {
720 /* Retry from the backing device: */
721 trace_bcache_read_retry(s->orig_bio);
724 do_bio_hook(s, s->orig_bio);
726 /* XXX: invalidate cache */
728 closure_bio_submit(bio, cl);
731 continue_at(cl, cached_dev_cache_miss_done, NULL);
734 static void cached_dev_read_done(struct closure *cl)
736 struct search *s = container_of(cl, struct search, cl);
737 struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
740 * We had a cache miss; cache_bio now contains data ready to be inserted
743 * First, we copy the data we just read from cache_bio's bounce buffers
744 * to the buffers the original bio pointed to:
748 bio_reset(s->iop.bio);
749 s->iop.bio->bi_iter.bi_sector = s->cache_miss->bi_iter.bi_sector;
750 bio_copy_dev(s->iop.bio, s->cache_miss);
751 s->iop.bio->bi_iter.bi_size = s->insert_bio_sectors << 9;
752 bch_bio_map(s->iop.bio, NULL);
754 bio_copy_data(s->cache_miss, s->iop.bio);
756 bio_put(s->cache_miss);
757 s->cache_miss = NULL;
760 if (verify(dc) && s->recoverable && !s->read_dirty_data)
761 bch_data_verify(dc, s->orig_bio);
766 !test_bit(CACHE_SET_STOPPING, &s->iop.c->flags)) {
767 BUG_ON(!s->iop.replace);
768 closure_call(&s->iop.cl, bch_data_insert, NULL, cl);
771 continue_at(cl, cached_dev_cache_miss_done, NULL);
774 static void cached_dev_read_done_bh(struct closure *cl)
776 struct search *s = container_of(cl, struct search, cl);
777 struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
779 bch_mark_cache_accounting(s->iop.c, s->d,
780 !s->cache_missed, s->iop.bypass);
781 trace_bcache_read(s->orig_bio, !s->cache_miss, s->iop.bypass);
784 continue_at_nobarrier(cl, cached_dev_read_error, bcache_wq);
785 else if (s->iop.bio || verify(dc))
786 continue_at_nobarrier(cl, cached_dev_read_done, bcache_wq);
788 continue_at_nobarrier(cl, cached_dev_bio_complete, NULL);
791 static int cached_dev_cache_miss(struct btree *b, struct search *s,
792 struct bio *bio, unsigned sectors)
794 int ret = MAP_CONTINUE;
796 struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
797 struct bio *miss, *cache_bio;
801 if (s->cache_miss || s->iop.bypass) {
802 miss = bio_next_split(bio, sectors, GFP_NOIO, s->d->bio_split);
803 ret = miss == bio ? MAP_DONE : MAP_CONTINUE;
807 if (!(bio->bi_opf & REQ_RAHEAD) &&
808 !(bio->bi_opf & REQ_META) &&
809 s->iop.c->gc_stats.in_use < CUTOFF_CACHE_READA)
810 reada = min_t(sector_t, dc->readahead >> 9,
811 get_capacity(bio->bi_disk) - bio_end_sector(bio));
813 s->insert_bio_sectors = min(sectors, bio_sectors(bio) + reada);
815 s->iop.replace_key = KEY(s->iop.inode,
816 bio->bi_iter.bi_sector + s->insert_bio_sectors,
817 s->insert_bio_sectors);
819 ret = bch_btree_insert_check_key(b, &s->op, &s->iop.replace_key);
823 s->iop.replace = true;
825 miss = bio_next_split(bio, sectors, GFP_NOIO, s->d->bio_split);
827 /* btree_search_recurse()'s btree iterator is no good anymore */
828 ret = miss == bio ? MAP_DONE : -EINTR;
830 cache_bio = bio_alloc_bioset(GFP_NOWAIT,
831 DIV_ROUND_UP(s->insert_bio_sectors, PAGE_SECTORS),
836 cache_bio->bi_iter.bi_sector = miss->bi_iter.bi_sector;
837 bio_copy_dev(cache_bio, miss);
838 cache_bio->bi_iter.bi_size = s->insert_bio_sectors << 9;
840 cache_bio->bi_end_io = request_endio;
841 cache_bio->bi_private = &s->cl;
843 bch_bio_map(cache_bio, NULL);
844 if (bio_alloc_pages(cache_bio, __GFP_NOWARN|GFP_NOIO))
848 bch_mark_cache_readahead(s->iop.c, s->d);
850 s->cache_miss = miss;
851 s->iop.bio = cache_bio;
853 closure_bio_submit(cache_bio, &s->cl);
859 miss->bi_end_io = request_endio;
860 miss->bi_private = &s->cl;
861 closure_bio_submit(miss, &s->cl);
865 static void cached_dev_read(struct cached_dev *dc, struct search *s)
867 struct closure *cl = &s->cl;
869 closure_call(&s->iop.cl, cache_lookup, NULL, cl);
870 continue_at(cl, cached_dev_read_done_bh, NULL);
875 static void cached_dev_write_complete(struct closure *cl)
877 struct search *s = container_of(cl, struct search, cl);
878 struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
880 up_read_non_owner(&dc->writeback_lock);
881 cached_dev_bio_complete(cl);
884 static void cached_dev_write(struct cached_dev *dc, struct search *s)
886 struct closure *cl = &s->cl;
887 struct bio *bio = &s->bio.bio;
888 struct bkey start = KEY(dc->disk.id, bio->bi_iter.bi_sector, 0);
889 struct bkey end = KEY(dc->disk.id, bio_end_sector(bio), 0);
891 bch_keybuf_check_overlapping(&s->iop.c->moving_gc_keys, &start, &end);
893 down_read_non_owner(&dc->writeback_lock);
894 if (bch_keybuf_check_overlapping(&dc->writeback_keys, &start, &end)) {
896 * We overlap with some dirty data undergoing background
897 * writeback, force this write to writeback
899 s->iop.bypass = false;
900 s->iop.writeback = true;
904 * Discards aren't _required_ to do anything, so skipping if
905 * check_overlapping returned true is ok
907 * But check_overlapping drops dirty keys for which io hasn't started,
908 * so we still want to call it.
910 if (bio_op(bio) == REQ_OP_DISCARD)
911 s->iop.bypass = true;
913 if (should_writeback(dc, s->orig_bio,
916 s->iop.bypass = false;
917 s->iop.writeback = true;
921 s->iop.bio = s->orig_bio;
924 if ((bio_op(bio) != REQ_OP_DISCARD) ||
925 blk_queue_discard(bdev_get_queue(dc->bdev)))
926 closure_bio_submit(bio, cl);
927 } else if (s->iop.writeback) {
928 bch_writeback_add(dc);
931 if (bio->bi_opf & REQ_PREFLUSH) {
932 /* Also need to send a flush to the backing device */
933 struct bio *flush = bio_alloc_bioset(GFP_NOIO, 0,
936 bio_copy_dev(flush, bio);
937 flush->bi_end_io = request_endio;
938 flush->bi_private = cl;
939 flush->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
941 closure_bio_submit(flush, cl);
944 s->iop.bio = bio_clone_fast(bio, GFP_NOIO, dc->disk.bio_split);
946 closure_bio_submit(bio, cl);
949 closure_call(&s->iop.cl, bch_data_insert, NULL, cl);
950 continue_at(cl, cached_dev_write_complete, NULL);
953 static void cached_dev_nodata(struct closure *cl)
955 struct search *s = container_of(cl, struct search, cl);
956 struct bio *bio = &s->bio.bio;
958 if (s->iop.flush_journal)
959 bch_journal_meta(s->iop.c, cl);
961 /* If it's a flush, we send the flush to the backing device too */
962 closure_bio_submit(bio, cl);
964 continue_at(cl, cached_dev_bio_complete, NULL);
967 /* Cached devices - read & write stuff */
969 static blk_qc_t cached_dev_make_request(struct request_queue *q,
973 struct bcache_device *d = bio->bi_disk->private_data;
974 struct cached_dev *dc = container_of(d, struct cached_dev, disk);
975 int rw = bio_data_dir(bio);
977 generic_start_io_acct(q, rw, bio_sectors(bio), &d->disk->part0);
979 bio_set_dev(bio, dc->bdev);
980 bio->bi_iter.bi_sector += dc->sb.data_offset;
982 if (cached_dev_get(dc)) {
983 s = search_alloc(bio, d);
984 trace_bcache_request_start(s->d, bio);
986 if (!bio->bi_iter.bi_size) {
988 * can't call bch_journal_meta from under
989 * generic_make_request
991 continue_at_nobarrier(&s->cl,
995 s->iop.bypass = check_should_bypass(dc, bio);
998 cached_dev_write(dc, s);
1000 cached_dev_read(dc, s);
1003 if ((bio_op(bio) == REQ_OP_DISCARD) &&
1004 !blk_queue_discard(bdev_get_queue(dc->bdev)))
1007 generic_make_request(bio);
1010 return BLK_QC_T_NONE;
1013 static int cached_dev_ioctl(struct bcache_device *d, fmode_t mode,
1014 unsigned int cmd, unsigned long arg)
1016 struct cached_dev *dc = container_of(d, struct cached_dev, disk);
1017 return __blkdev_driver_ioctl(dc->bdev, mode, cmd, arg);
1020 static int cached_dev_congested(void *data, int bits)
1022 struct bcache_device *d = data;
1023 struct cached_dev *dc = container_of(d, struct cached_dev, disk);
1024 struct request_queue *q = bdev_get_queue(dc->bdev);
1027 if (bdi_congested(q->backing_dev_info, bits))
1030 if (cached_dev_get(dc)) {
1034 for_each_cache(ca, d->c, i) {
1035 q = bdev_get_queue(ca->bdev);
1036 ret |= bdi_congested(q->backing_dev_info, bits);
1045 void bch_cached_dev_request_init(struct cached_dev *dc)
1047 struct gendisk *g = dc->disk.disk;
1049 g->queue->make_request_fn = cached_dev_make_request;
1050 g->queue->backing_dev_info->congested_fn = cached_dev_congested;
1051 dc->disk.cache_miss = cached_dev_cache_miss;
1052 dc->disk.ioctl = cached_dev_ioctl;
1055 /* Flash backed devices */
1057 static int flash_dev_cache_miss(struct btree *b, struct search *s,
1058 struct bio *bio, unsigned sectors)
1060 unsigned bytes = min(sectors, bio_sectors(bio)) << 9;
1062 swap(bio->bi_iter.bi_size, bytes);
1064 swap(bio->bi_iter.bi_size, bytes);
1066 bio_advance(bio, bytes);
1068 if (!bio->bi_iter.bi_size)
1071 return MAP_CONTINUE;
1074 static void flash_dev_nodata(struct closure *cl)
1076 struct search *s = container_of(cl, struct search, cl);
1078 if (s->iop.flush_journal)
1079 bch_journal_meta(s->iop.c, cl);
1081 continue_at(cl, search_free, NULL);
1084 static blk_qc_t flash_dev_make_request(struct request_queue *q,
1089 struct bcache_device *d = bio->bi_disk->private_data;
1090 int rw = bio_data_dir(bio);
1092 generic_start_io_acct(q, rw, bio_sectors(bio), &d->disk->part0);
1094 s = search_alloc(bio, d);
1098 trace_bcache_request_start(s->d, bio);
1100 if (!bio->bi_iter.bi_size) {
1102 * can't call bch_journal_meta from under
1103 * generic_make_request
1105 continue_at_nobarrier(&s->cl,
1108 return BLK_QC_T_NONE;
1110 bch_keybuf_check_overlapping(&s->iop.c->moving_gc_keys,
1111 &KEY(d->id, bio->bi_iter.bi_sector, 0),
1112 &KEY(d->id, bio_end_sector(bio), 0));
1114 s->iop.bypass = (bio_op(bio) == REQ_OP_DISCARD) != 0;
1115 s->iop.writeback = true;
1118 closure_call(&s->iop.cl, bch_data_insert, NULL, cl);
1120 closure_call(&s->iop.cl, cache_lookup, NULL, cl);
1123 continue_at(cl, search_free, NULL);
1124 return BLK_QC_T_NONE;
1127 static int flash_dev_ioctl(struct bcache_device *d, fmode_t mode,
1128 unsigned int cmd, unsigned long arg)
1133 static int flash_dev_congested(void *data, int bits)
1135 struct bcache_device *d = data;
1136 struct request_queue *q;
1141 for_each_cache(ca, d->c, i) {
1142 q = bdev_get_queue(ca->bdev);
1143 ret |= bdi_congested(q->backing_dev_info, bits);
1149 void bch_flash_dev_request_init(struct bcache_device *d)
1151 struct gendisk *g = d->disk;
1153 g->queue->make_request_fn = flash_dev_make_request;
1154 g->queue->backing_dev_info->congested_fn = flash_dev_congested;
1155 d->cache_miss = flash_dev_cache_miss;
1156 d->ioctl = flash_dev_ioctl;
1159 void bch_request_exit(void)
1161 if (bch_search_cache)
1162 kmem_cache_destroy(bch_search_cache);
1165 int __init bch_request_init(void)
1167 bch_search_cache = KMEM_CACHE(search, 0);
1168 if (!bch_search_cache)