bcache: plumb nr_replicas through disk_reservations

8 files changed, 94 insertions, 93 deletions

diff --git a/drivers/md/bcache/btree_update.c b/drivers/md/bcache/btree_update.c
index 32a3915fd118..48e062b7acbc 100644
--- a/drivers/md/bcache/btree_update.c
+++ b/drivers/md/bcache/btree_update.c
@@ -238,6 +238,7 @@ void btree_open_bucket_put(struct cache_set *c, struct btree *b)
 }
 
 static struct btree *__bch_btree_node_alloc(struct cache_set *c,
+					    struct disk_reservation *res,
 					    struct closure *cl)
 {
 	BKEY_PADDED(k) tmp;
@@ -263,7 +264,7 @@ retry:
 
 	ob = bch_alloc_sectors(c, &c->btree_write_point,
 			       bkey_i_to_extent(&tmp.k),
-			       c->opts.metadata_replicas, cl);
+			       res->nr_replicas, cl);
 	if (IS_ERR(ob))
 		return ERR_CAST(ob);
 
@@ -518,7 +519,8 @@ static struct btree_reserve *__bch_btree_reserve_get(struct cache_set *c,
 	struct btree *b;
 	struct disk_reservation disk_res = { 0, 0 };
 	unsigned sectors = nr_nodes * c->sb.btree_node_size;
-	int ret, flags = BCH_DISK_RESERVATION_GC_LOCK_HELD;
+	int ret, flags = BCH_DISK_RESERVATION_GC_LOCK_HELD|
+		BCH_DISK_RESERVATION_METADATA;
 
 	if (!check_enospc)
 		flags |= BCH_DISK_RESERVATION_NOFAIL;
@@ -544,7 +546,7 @@ static struct btree_reserve *__bch_btree_reserve_get(struct cache_set *c,
 	reserve->nr = 0;
 
 	while (reserve->nr < nr_nodes) {
-		b = __bch_btree_node_alloc(c, cl);
+		b = __bch_btree_node_alloc(c, &disk_res, cl);
 		if (IS_ERR(b)) {
 			ret = PTR_ERR(b);
 			goto err_free;
diff --git a/drivers/md/bcache/buckets.c b/drivers/md/bcache/buckets.c
index e45f128d1635..3fa40106c783 100644
--- a/drivers/md/bcache/buckets.c
+++ b/drivers/md/bcache/buckets.c
@@ -192,7 +192,7 @@ void bch_cache_set_stats_apply(struct cache_set *c,
 	 * Not allowed to reduce sectors_available except by getting a
 	 * reservation:
 	 */
-	BUG_ON(added > (disk_res ? disk_res->sectors : 0));
+	BUG_ON(added > (s64) (disk_res ? disk_res->sectors : 0));
 
 	if (added > 0) {
 		disk_res->sectors	-= added;
@@ -639,6 +639,8 @@ int bch_disk_reservation_add(struct cache_set *c,
 	s64 sectors_available;
 	int ret;
 
+	sectors *= res->nr_replicas;
+
 	lg_local_lock(&c->bucket_stats_lock);
 	stats = this_cpu_ptr(c->bucket_stats_percpu);
 
@@ -704,6 +706,9 @@ int bch_disk_reservation_get(struct cache_set *c,
 {
 	res->sectors = 0;
 	res->gen = c->capacity_gen;
+	res->nr_replicas = (flags & BCH_DISK_RESERVATION_METADATA)
+		? c->opts.metadata_replicas
+		: c->opts.data_replicas;
 
 	return bch_disk_reservation_add(c, res, sectors, flags);
 }
diff --git a/drivers/md/bcache/buckets.h b/drivers/md/bcache/buckets.h
index c96a398ca7bc..da077a3fe5a6 100644
--- a/drivers/md/bcache/buckets.h
+++ b/drivers/md/bcache/buckets.h
@@ -256,6 +256,7 @@ void bch_disk_reservation_put(struct cache_set *,
 
 #define BCH_DISK_RESERVATION_NOFAIL		(1 << 0)
 #define BCH_DISK_RESERVATION_GC_LOCK_HELD	(1 << 1)
+#define BCH_DISK_RESERVATION_METADATA		(1 << 2)
 
 int bch_disk_reservation_add(struct cache_set *,
 			     struct disk_reservation *,
diff --git a/drivers/md/bcache/buckets_types.h b/drivers/md/bcache/buckets_types.h
index 256f6fe7f272..7712ff2b6d1a 100644
--- a/drivers/md/bcache/buckets_types.h
+++ b/drivers/md/bcache/buckets_types.h
@@ -73,8 +73,9 @@ struct bucket_heap_entry {
  * A reservation for space on disk:
  */
 struct disk_reservation {
-	u32		sectors;
+	u64		sectors;
 	u32		gen;
+	unsigned	nr_replicas;
 };
 
 #endif /* _BUCKETS_TYPES_H */
diff --git a/drivers/md/bcache/fs-io.c b/drivers/md/bcache/fs-io.c
index d12d1c2e6e2e..6062da0a8622 100644
--- a/drivers/md/bcache/fs-io.c
+++ b/drivers/md/bcache/fs-io.c
@@ -856,7 +856,10 @@ alloc_io:
 		w->io->op.sectors_added	= 0;
 		w->io->op.is_dio	= false;
 		bch_write_op_init(&w->io->op.op, w->c, &w->io->bio,
-				  (struct disk_reservation) { 0 }, NULL,
+				  (struct disk_reservation) {
+					.nr_replicas = w->c->opts.data_replicas,
+				  },
+				  NULL,
 				  bkey_to_s_c(&KEY(w->inum, 0, 0)),
 				  NULL, &ei->journal_seq, 0);
 		w->io->op.op.index_update_fn = bchfs_write_index_update;
@@ -1305,10 +1308,8 @@ static void bch_do_direct_IO_write(struct dio_write *dio)
 	dio->iop.is_dio		= true;
 	dio->iop.new_i_size	= U64_MAX;
 	bch_write_op_init(&dio->iop.op, dio->c, &dio->bio,
-			  (struct disk_reservation) {
-			  .sectors = bio_sectors(bio),
-			  .gen = dio->res.gen
-			  }, NULL,
+			  dio->res,
+			  NULL,
 			  bkey_to_s_c(&KEY(inode->i_ino,
 					   bio_end_sector(bio),
 					   bio_sectors(bio))),
@@ -1316,6 +1317,7 @@ static void bch_do_direct_IO_write(struct dio_write *dio)
 	dio->iop.op.index_update_fn = bchfs_write_index_update;
 
 	dio->res.sectors -= bio_sectors(bio);
+	dio->iop.op.res.sectors = bio_sectors(bio);
 
 	task_io_account_write(bio->bi_iter.bi_size);
 
diff --git a/drivers/md/bcache/io.c b/drivers/md/bcache/io.c
index 9e4dae1c95d6..8b560c0ff5a7 100644
--- a/drivers/md/bcache/io.c
+++ b/drivers/md/bcache/io.c
@@ -1163,16 +1163,6 @@ void bch_write(struct closure *cl)
 		closure_return(cl);
 	}
 
-	if (version_stress_test(c))
-		op->insert_key.k.version = bch_rand_range(UINT_MAX);
-
-	/*
-	 * This ought to be initialized in bch_write_op_init(), but struct
-	 * cache_set isn't exported
-	 */
-	if (!op->io_wq)
-		op->io_wq = op->c->wq;
-
 	if (!(op->flags & BCH_WRITE_DISCARD))
 		bch_increment_clock(c, bio_sectors(bio), WRITE);
 
@@ -1239,13 +1229,13 @@ void bch_write_op_init(struct bch_write_op *op, struct cache_set *c,
 	}
 
 	op->c		= c;
-	op->io_wq	= NULL;
+	op->io_wq	= op->c->wq;
 	op->bio		= bio;
 	op->written	= 0;
 	op->error	= 0;
 	op->flags	= flags;
 	op->compression_type = c->opts.compression;
-	op->nr_replicas	= c->opts.data_replicas;
+	op->nr_replicas	= res.nr_replicas;
 	op->res		= res;
 	op->wp		= wp;
 
@@ -1275,6 +1265,9 @@ void bch_write_op_init(struct bch_write_op *op, struct cache_set *c,
 			(op->flags & BCH_WRITE_CACHED) ? BCH_EXTENT_CACHED :
 			BCH_EXTENT;
 	}
+
+	if (version_stress_test(c))
+		op->insert_key.k.version = bch_rand_range(UINT_MAX);
 }
 
 void bch_replace_init(struct bch_replace_info *r, struct bkey_s_c old)
@@ -1320,73 +1313,6 @@ struct cache_promote_op {
 	struct bch_write_bio	bio; /* must be last */
 };
 
-/**
- * __cache_promote -- insert result of read bio into cache
- *
- * Used for backing devices and flash-only volumes.
- *
- * @orig_bio must actually be a bbio with a valid key.
- */
-void __cache_promote(struct cache_set *c, struct bbio *orig_bio,
-		     struct bkey_s_c old,
-		     struct bkey_s_c new,
-		     unsigned write_flags)
-{
-#if 0
-	struct cache_promote_op *op;
-	struct bio *bio;
-	unsigned pages = DIV_ROUND_UP(orig_bio->bio.bi_iter.bi_size, PAGE_SIZE);
-
-	/* XXX: readahead? */
-
-	op = kmalloc(sizeof(*op) + sizeof(struct bio_vec) * pages, GFP_NOIO);
-	if (!op)
-		goto out_submit;
-
-	/* clone the bbio */
-	memcpy(&op->bio, orig_bio, offsetof(struct bbio, bio));
-
-	bio = &op->bio.bio.bio;
-	bio_init(bio);
-	bio_get(bio);
-	bio->bi_bdev		= orig_bio->bio.bi_bdev;
-	bio->bi_iter.bi_sector	= orig_bio->bio.bi_iter.bi_sector;
-	bio->bi_iter.bi_size	= orig_bio->bio.bi_iter.bi_size;
-	bio->bi_end_io		= cache_promote_endio;
-	bio->bi_private		= &op->cl;
-	bio->bi_io_vec		= bio->bi_inline_vecs;
-	bch_bio_map(bio, NULL);
-
-	if (bio_alloc_pages(bio, __GFP_NOWARN|GFP_NOIO))
-		goto out_free;
-
-	orig_bio->ca = NULL;
-
-	closure_init(&op->cl, &c->cl);
-	op->orig_bio		= &orig_bio->bio;
-	op->stale		= 0;
-
-	bch_write_op_init(&op->iop, c, &op->bio, &c->promote_write_point,
-			  new, old,
-			  BCH_WRITE_ALLOC_NOWAIT|write_flags);
-	op->iop.nr_replicas = 1;
-
-	//bch_cut_front(bkey_start_pos(&orig_bio->key.k), &op->iop.insert_key);
-	//bch_cut_back(orig_bio->key.k.p, &op->iop.insert_key.k);
-
-	trace_bcache_promote(&orig_bio->bio);
-
-	op->bio.bio.submit_time_us = local_clock_us();
-	closure_bio_submit(bio, &op->cl);
-
-	continue_at(&op->cl, cache_promote_write, c->wq);
-out_free:
-	kfree(op);
-out_submit:
-	generic_make_request(&orig_bio->bio);
-#endif
-}
-
 /* Read */
 
 static int bio_checksum_uncompress(struct cache_set *c,
@@ -1715,6 +1641,7 @@ void bch_read_extent_iter(struct cache_set *c, struct bch_read_bio *orig,
 				  &c->promote_write_point, k,
 				  &promote_op->replace.hook, NULL,
 				  BCH_WRITE_ALLOC_NOWAIT);
+		promote_op->iop.nr_replicas = 1;
 
 		if (rbio->crc.compression_type) {
 			promote_op->iop.flags |= BCH_WRITE_DATA_COMPRESSED;
diff --git a/drivers/md/bcache/io.h b/drivers/md/bcache/io.h
index b37d9845b17e..50172477fcb8 100644
--- a/drivers/md/bcache/io.h
+++ b/drivers/md/bcache/io.h
@@ -89,10 +89,6 @@ int bch_discard(struct cache_set *, struct bpos, struct bpos,
 		u64, struct disk_reservation *,
 		struct extent_insert_hook *, u64 *);
 
-void __cache_promote(struct cache_set *, struct bbio *,
-		     struct bkey_s_c, struct bkey_s_c, unsigned);
-bool cache_promote(struct cache_set *, struct bbio *, struct bkey_s_c);
-
 void bch_read_retry_work(struct work_struct *);
 void bch_wake_delayed_writes(unsigned long data);
 
diff --git a/drivers/md/bcache/request.c b/drivers/md/bcache/request.c
index dc3fa5477d6c..fc760de6d997 100644
--- a/drivers/md/bcache/request.c
+++ b/drivers/md/bcache/request.c
@@ -325,6 +325,73 @@ static void cached_dev_read_done_bh(struct closure *cl)
 }
 
 /**
+ * __cache_promote -- insert result of read bio into cache
+ *
+ * Used for backing devices and flash-only volumes.
+ *
+ * @orig_bio must actually be a bbio with a valid key.
+ */
+void __cache_promote(struct cache_set *c, struct bbio *orig_bio,
+		     struct bkey_s_c old,
+		     struct bkey_s_c new,
+		     unsigned write_flags)
+{
+#if 0
+	struct cache_promote_op *op;
+	struct bio *bio;
+	unsigned pages = DIV_ROUND_UP(orig_bio->bio.bi_iter.bi_size, PAGE_SIZE);
+
+	/* XXX: readahead? */
+
+	op = kmalloc(sizeof(*op) + sizeof(struct bio_vec) * pages, GFP_NOIO);
+	if (!op)
+		goto out_submit;
+
+	/* clone the bbio */
+	memcpy(&op->bio, orig_bio, offsetof(struct bbio, bio));
+
+	bio = &op->bio.bio.bio;
+	bio_init(bio);
+	bio_get(bio);
+	bio->bi_bdev		= orig_bio->bio.bi_bdev;
+	bio->bi_iter.bi_sector	= orig_bio->bio.bi_iter.bi_sector;
+	bio->bi_iter.bi_size	= orig_bio->bio.bi_iter.bi_size;
+	bio->bi_end_io		= cache_promote_endio;
+	bio->bi_private		= &op->cl;
+	bio->bi_io_vec		= bio->bi_inline_vecs;
+	bch_bio_map(bio, NULL);
+
+	if (bio_alloc_pages(bio, __GFP_NOWARN|GFP_NOIO))
+		goto out_free;
+
+	orig_bio->ca = NULL;
+
+	closure_init(&op->cl, &c->cl);
+	op->orig_bio		= &orig_bio->bio;
+	op->stale		= 0;
+
+	bch_write_op_init(&op->iop, c, &op->bio, &c->promote_write_point,
+			  new, old,
+			  BCH_WRITE_ALLOC_NOWAIT|write_flags);
+	op->iop.nr_replicas = 1;
+
+	//bch_cut_front(bkey_start_pos(&orig_bio->key.k), &op->iop.insert_key);
+	//bch_cut_back(orig_bio->key.k.p, &op->iop.insert_key.k);
+
+	trace_bcache_promote(&orig_bio->bio);
+
+	op->bio.bio.submit_time_us = local_clock_us();
+	closure_bio_submit(bio, &op->cl);
+
+	continue_at(&op->cl, cache_promote_write, c->wq);
+out_free:
+	kfree(op);
+out_submit:
+	generic_make_request(&orig_bio->bio);
+#endif
+}
+
+/**
  * cached_dev_cache_miss - populate cache with data from backing device
  *
  * We don't write to the cache if s->bypass is set.