bcache: kill BCH_WRITE_CHECK_ENOSPC

13 files changed, 65 insertions, 67 deletions

diff --git a/drivers/md/bcache/alloc.c b/drivers/md/bcache/alloc.c
index 59d3574465df..6e732325df65 100644
--- a/drivers/md/bcache/alloc.c
+++ b/drivers/md/bcache/alloc.c
@@ -971,8 +971,7 @@ static enum bucket_alloc_ret __bch_bucket_alloc_set(struct cache_set *c,
 						    enum alloc_reserve reserve,
 						    unsigned nr_replicas,
 						    struct cache_group *devs,
-						    long *caches_used,
-						    bool check_enospc)
+						    long *caches_used)
 {
 	enum bucket_alloc_ret ret;
 
@@ -1037,19 +1036,11 @@ static enum bucket_alloc_ret __bch_bucket_alloc_set(struct cache_set *c,
 			.offset	= bucket_to_sector(ca, r),
 			.dev	= ca->sb.nr_this_dev,
 		};
-
-		if (check_enospc && cache_set_full(c)) {
-			ret = CACHE_SET_FULL;
-			goto err_nocheck;
-		}
 	}
 
 	rcu_read_unlock();
 	return ALLOC_SUCCESS;
 err:
-	if (check_enospc && cache_set_full(c))
-		ret = CACHE_SET_FULL;
-err_nocheck:
 	rcu_read_unlock();
 	return ret;
 }
@@ -1058,15 +1049,14 @@ static int bch_bucket_alloc_set(struct cache_set *c, struct open_bucket *ob,
 				enum alloc_reserve reserve,
 				unsigned nr_replicas,
 				struct cache_group *devs, long *caches_used,
-				bool check_enospc, struct closure *cl)
+				struct closure *cl)
 {
 	struct closure_waitlist *waitlist = NULL;
 	bool waiting = false;
 
 	while (1) {
 		switch (__bch_bucket_alloc_set(c, ob, reserve, nr_replicas,
-					       devs, caches_used,
-					       check_enospc)) {
+					       devs, caches_used)) {
 		case ALLOC_SUCCESS:
 			if (waitlist)
 				closure_wake_up(waitlist);
@@ -1271,7 +1261,6 @@ static int open_bucket_add_buckets(struct cache_set *c,
 				   struct open_bucket *ob,
 				   struct bkey_i_extent *e,
 				   unsigned nr_replicas,
-				   bool check_enospc,
 				   struct closure *cl)
 {
 	const struct bch_extent_ptr *ptr;
@@ -1317,8 +1306,7 @@ static int open_bucket_add_buckets(struct cache_set *c,
 	}
 
 	return bch_bucket_alloc_set(c, ob, wp->reserve, nr_replicas,
-				    wp->group, caches_used,
-				    check_enospc, cl);
+				    wp->group, caches_used, cl);
 }
 
 /*
@@ -1328,7 +1316,6 @@ struct open_bucket *bch_alloc_sectors_start(struct cache_set *c,
 					    struct write_point *wp,
 					    struct bkey_i_extent *e,
 					    unsigned nr_replicas,
-					    bool check_enospc,
 					    struct closure *cl)
 {
 	struct open_bucket *ob;
@@ -1382,8 +1369,7 @@ retry:
 		ob = new_ob;
 	}
 
-	ret = open_bucket_add_buckets(c, wp, ob, e, nr_replicas,
-				      check_enospc, cl);
+	ret = open_bucket_add_buckets(c, wp, ob, e, nr_replicas, cl);
 	if (ret) {
 		mutex_unlock(&ob->lock);
 		return ERR_PTR(ret);
@@ -1476,12 +1462,11 @@ struct open_bucket *bch_alloc_sectors(struct cache_set *c,
 				      struct write_point *wp,
 				      struct bkey_i_extent *e,
 				      unsigned nr_replicas,
-				      bool check_enospc,
 				      struct closure *cl)
 {
 	struct open_bucket *ob;
 
-	ob = bch_alloc_sectors_start(c, wp, e, nr_replicas, check_enospc, cl);
+	ob = bch_alloc_sectors_start(c, wp, e, nr_replicas, cl);
 	if (IS_ERR_OR_NULL(ob))
 		return ob;
 
diff --git a/drivers/md/bcache/alloc.h b/drivers/md/bcache/alloc.h
index 4a7afd213958..eb1bd5e3b6a3 100644
--- a/drivers/md/bcache/alloc.h
+++ b/drivers/md/bcache/alloc.h
@@ -19,14 +19,13 @@ void bch_open_bucket_put(struct cache_set *, struct open_bucket *);
 struct open_bucket *bch_alloc_sectors_start(struct cache_set *,
 					    struct write_point *,
 					    struct bkey_i_extent *,
-					    unsigned, bool,
-					    struct closure *);
+					    unsigned, struct closure *);
 void bch_alloc_sectors_done(struct cache_set *, struct write_point *,
 			    struct bkey_i_extent *, unsigned,
 			    struct open_bucket *, unsigned);
 
 struct open_bucket *bch_alloc_sectors(struct cache_set *, struct write_point *,
-				      struct bkey_i_extent *, unsigned, bool,
+				      struct bkey_i_extent *, unsigned,
 				      struct closure *);
 
 static inline void bch_wake_allocator(struct cache *ca)
diff --git a/drivers/md/bcache/btree_update.c b/drivers/md/bcache/btree_update.c
index aea66cc8e68c..d3322ce47297 100644
--- a/drivers/md/bcache/btree_update.c
+++ b/drivers/md/bcache/btree_update.c
@@ -229,7 +229,6 @@ void btree_open_bucket_put(struct cache_set *c, struct btree *b)
 }
 
 static struct btree *__bch_btree_node_alloc(struct cache_set *c,
-					    bool check_enospc,
 					    struct closure *cl)
 {
 	BKEY_PADDED(k) tmp;
@@ -255,8 +254,7 @@ retry:
 
 	ob = bch_alloc_sectors(c, &c->btree_write_point,
 			       bkey_i_to_extent(&tmp.k),
-			       c->opts.metadata_replicas,
-			       check_enospc, cl);
+			       c->opts.metadata_replicas, cl);
 	if (IS_ERR(ob))
 		return ERR_CAST(ob);
 
@@ -447,6 +445,8 @@ static struct btree *__btree_root_alloc(struct cache_set *c, unsigned level,
 
 void bch_btree_reserve_put(struct cache_set *c, struct btree_reserve *reserve)
 {
+	atomic64_sub_bug(reserve->sectors_reserved, &c->sectors_reserved);
+
 	mutex_lock(&c->btree_reserve_cache_lock);
 
 	while (reserve->nr) {
@@ -484,8 +484,16 @@ static struct btree_reserve *__bch_btree_reserve_get(struct cache_set *c,
 {
 	struct btree_reserve *reserve;
 	struct btree *b;
+	unsigned sectors_reserved = 0;
 	int ret;
 
+	if (check_enospc) {
+		sectors_reserved = nr_nodes * c->sb.btree_node_size;
+
+		if (bch_reserve_sectors(c, sectors_reserved))
+			return ERR_PTR(-ENOSPC);
+	}
+
 	BUG_ON(nr_nodes > BTREE_RESERVE_MAX);
 
 	/*
@@ -498,10 +506,11 @@ static struct btree_reserve *__bch_btree_reserve_get(struct cache_set *c,
 
 	reserve = mempool_alloc(&c->btree_reserve_pool, GFP_NOIO);
 
+	reserve->sectors_reserved = sectors_reserved;
 	reserve->nr = 0;
 
 	while (reserve->nr < nr_nodes) {
-		b = __bch_btree_node_alloc(c, check_enospc, cl);
+		b = __bch_btree_node_alloc(c, cl);
 		if (IS_ERR(b)) {
 			ret = PTR_ERR(b);
 			goto err_free;
diff --git a/drivers/md/bcache/btree_update.h b/drivers/md/bcache/btree_update.h
index b3debb9d4784..e4ccd1ddb7e7 100644
--- a/drivers/md/bcache/btree_update.h
+++ b/drivers/md/bcache/btree_update.h
@@ -10,6 +10,7 @@ struct bkey_format;
 struct btree;
 
 struct btree_reserve {
+	unsigned		sectors_reserved;
 	unsigned		nr;
 	struct btree		*b[];
 };
diff --git a/drivers/md/bcache/buckets.c b/drivers/md/bcache/buckets.c
index 7f9fb8092e44..a78d883ee656 100644
--- a/drivers/md/bcache/buckets.c
+++ b/drivers/md/bcache/buckets.c
@@ -436,3 +436,25 @@ void bch_unmark_open_bucket(struct cache *ca, struct bucket *g)
 		new.owned_by_allocator = 0;
 	}));
 }
+
+#define SECTORS_CACHE	1024
+
+int bch_reserve_sectors(struct cache_set *c, unsigned sectors)
+{
+	u64 sectors_to_get = SECTORS_CACHE + sectors;
+
+	if (likely(atomic64_sub_return(sectors,
+				       &c->sectors_reserved_cache) >= 0))
+		return 0;
+
+	atomic64_add(sectors_to_get, &c->sectors_reserved);
+
+	if (likely(!cache_set_full(c))) {
+		atomic64_add(sectors_to_get, &c->sectors_reserved_cache);
+		return 0;
+	}
+
+	atomic64_sub_bug(sectors_to_get, &c->sectors_reserved);
+	atomic64_add(sectors, &c->sectors_reserved_cache);
+	return -ENOSPC;
+}
diff --git a/drivers/md/bcache/buckets.h b/drivers/md/bcache/buckets.h
index ac3f1b59d957..0c68ad57a23b 100644
--- a/drivers/md/bcache/buckets.h
+++ b/drivers/md/bcache/buckets.h
@@ -244,4 +244,6 @@ void bch_unmark_open_bucket(struct cache *, struct bucket *);
 int bch_mark_pointers(struct cache_set *, struct bkey_s_c_extent,
 		      int, bool, bool, bool, struct gc_pos);
 
+int bch_reserve_sectors(struct cache_set *, unsigned);
+
 #endif /* _BUCKETS_H */
diff --git a/drivers/md/bcache/fs-io.c b/drivers/md/bcache/fs-io.c
index 7efe53c563a8..173d5c9f4c65 100644
--- a/drivers/md/bcache/fs-io.c
+++ b/drivers/md/bcache/fs-io.c
@@ -360,7 +360,7 @@ union { struct {
 	 *
 	 * BCH_PAGE_UNALLOCATED: page is not fully written on disk, or is
 	 * compressed - before writing we have to reserve space with
-	 * reserve_sectors()
+	 * bch_reserve_sectors()
 	 *
 	 * BCH_PAGE_RESERVED: page has space reserved on disk (reservation will
 	 * be consumed when the page is written).
@@ -411,28 +411,6 @@ union { struct {
 	_old;								\
 })
 
-#define SECTORS_CACHE	1024
-
-static int reserve_sectors(struct cache_set *c, unsigned sectors)
-{
-	u64 sectors_to_get = SECTORS_CACHE + sectors;
-
-	if (likely(atomic64_sub_return(sectors,
-				       &c->sectors_reserved_cache) >= 0))
-		return 0;
-
-	atomic64_add(sectors_to_get, &c->sectors_reserved);
-
-	if (likely(!cache_set_full(c))) {
-		atomic64_add(sectors_to_get, &c->sectors_reserved_cache);
-		return 0;
-	}
-
-	atomic64_sub_bug(sectors_to_get, &c->sectors_reserved);
-	atomic64_add(sectors, &c->sectors_reserved_cache);
-	return -ENOSPC;
-}
-
 static inline struct bch_page_state *page_state(struct page *page)
 {
 	struct bch_page_state *s = (void *) &page->private;
@@ -466,7 +444,7 @@ static int bch_get_page_reservation(struct cache_set *c, struct page *page)
 	if (s->alloc_state != BCH_PAGE_UNALLOCATED)
 		return 0;
 
-	ret = reserve_sectors(c, PAGE_SECTORS);
+	ret = bch_reserve_sectors(c, PAGE_SECTORS);
 	if (ret)
 		return ret;
 
@@ -1491,7 +1469,7 @@ static int bch_direct_IO_write(struct cache_set *c, struct kiocb *req,
 	 * Have to then guard against racing with truncate (deleting data that
 	 * we would have been overwriting)
 	 */
-	ret = reserve_sectors(c, dio->nr_sectors);
+	ret = bch_reserve_sectors(c, dio->nr_sectors);
 	if (ret)
 		goto err_put_sectors_dirty;
 
@@ -2325,7 +2303,7 @@ static long bch_fallocate(struct inode *inode, int mode,
 
 		sectors = reservation.k.size;
 
-		ret = reserve_sectors(c, sectors);
+		ret = bch_reserve_sectors(c, sectors);
 		if (ret)
 			goto err_put_sectors_dirty;
 
diff --git a/drivers/md/bcache/io.c b/drivers/md/bcache/io.c
index 19bd1b44f66d..aac4eaa33f08 100644
--- a/drivers/md/bcache/io.c
+++ b/drivers/md/bcache/io.c
@@ -1012,7 +1012,6 @@ static void __bch_write(struct closure *cl)
 		b = bch_alloc_sectors_start(op->c, op->wp,
 					    bkey_i_to_extent(k),
 					    op->nr_replicas,
-					    op->check_enospc,
 					    op->nowait ? NULL : cl);
 		BUG_ON(!b);
 
@@ -1226,7 +1225,6 @@ void bch_write_op_init(struct bch_write_op *op, struct cache_set *c,
 	op->written	= 0;
 	op->error	= 0;
 	op->flags	= 0;
-	op->check_enospc = (flags & BCH_WRITE_CHECK_ENOSPC) != 0;
 	op->nowait	= (flags & BCH_WRITE_ALLOC_NOWAIT) != 0;
 	op->discard	= (flags & BCH_WRITE_DISCARD) != 0;
 	op->cached	= (flags & BCH_WRITE_CACHED) != 0;
@@ -1351,7 +1349,6 @@ void __cache_promote(struct cache_set *c, struct bbio *orig_bio,
 
 	bch_write_op_init(&op->iop, c, &op->bio, &c->promote_write_point,
 			  new, old,
-			  BCH_WRITE_CHECK_ENOSPC|
 			  BCH_WRITE_ALLOC_NOWAIT|write_flags);
 	op->iop.nr_replicas = 1;
 
@@ -1729,7 +1726,6 @@ void bch_read_extent_iter(struct cache_set *c, struct bio *orig,
 				  &promote_op->bio,
 				  &c->promote_write_point,
 				  k, NULL, NULL,
-				  BCH_WRITE_CHECK_ENOSPC|
 				  BCH_WRITE_ALLOC_NOWAIT);
 
 		if (!read_full) {
diff --git a/drivers/md/bcache/io.h b/drivers/md/bcache/io.h
index 490742660193..d165fc468f78 100644
--- a/drivers/md/bcache/io.h
+++ b/drivers/md/bcache/io.h
@@ -17,12 +17,11 @@
 	container_of((_bio), struct bch_write_bio, bio.bio)
 
 enum bch_write_flags {
-	BCH_WRITE_CHECK_ENOSPC		= (1 << 0),
-	BCH_WRITE_ALLOC_NOWAIT		= (1 << 1),
-	BCH_WRITE_DISCARD		= (1 << 2),
-	BCH_WRITE_CACHED		= (1 << 3),
-	BCH_WRITE_FLUSH			= (1 << 4),
-	BCH_WRITE_DISCARD_ON_ERROR	= (1 << 5),
+	BCH_WRITE_ALLOC_NOWAIT		= (1 << 0),
+	BCH_WRITE_DISCARD		= (1 << 1),
+	BCH_WRITE_CACHED		= (1 << 2),
+	BCH_WRITE_FLUSH			= (1 << 3),
+	BCH_WRITE_DISCARD_ON_ERROR	= (1 << 4),
 };
 
 void bch_write_op_init(struct bch_write_op *, struct cache_set *,
diff --git a/drivers/md/bcache/io_types.h b/drivers/md/bcache/io_types.h
index 88d35b8d3590..3946f2b7ba22 100644
--- a/drivers/md/bcache/io_types.h
+++ b/drivers/md/bcache/io_types.h
@@ -82,8 +82,6 @@ struct bch_write_op {
 		u16		flags;
 
 	struct {
-		/* Return -ENOSPC if cache set is full? */
-		unsigned	check_enospc:1;
 		/* Return -ENOSPC if no buckets immediately available? */
 		unsigned	nowait:1;
 		/* Discard key range? */
diff --git a/drivers/md/bcache/migrate.c b/drivers/md/bcache/migrate.c
index 685e96389e28..18c97ba8ec7a 100644
--- a/drivers/md/bcache/migrate.c
+++ b/drivers/md/bcache/migrate.c
@@ -39,10 +39,15 @@ static int issue_migration_move(struct cache *ca,
 	struct cache_set *c = ca->set;
 	struct moving_io *io;
 
+	if (bch_reserve_sectors(c, k.k->size))
+		return -ENOSPC;
+
 	io = moving_io_alloc(k);
 	if (io == NULL)
 		return -ENOMEM;
 
+	io->has_reservation = true;
+
 	/* This also copies k into the write op's replace_key and insert_key */
 
 	bch_replace_init(&io->replace, k);
@@ -50,7 +55,7 @@ static int issue_migration_move(struct cache *ca,
 	bch_write_op_init(&io->op, c, &io->bio,
 			  &c->migration_write_point,
 			  k, &io->replace.hook, NULL,
-			  BCH_WRITE_CHECK_ENOSPC);
+			  0);
 	io->op.nr_replicas = 1;
 
 	io->op.io_wq = q->wq;
diff --git a/drivers/md/bcache/move.c b/drivers/md/bcache/move.c
index 5c3750fc5c23..003865c596a3 100644
--- a/drivers/md/bcache/move.c
+++ b/drivers/md/bcache/move.c
@@ -112,6 +112,9 @@ struct moving_io *moving_io_alloc(struct bkey_s_c k)
 
 void moving_io_free(struct moving_io *io)
 {
+	if (io->has_reservation)
+		atomic64_sub_bug(io->key.k.size, &io->op.c->sectors_reserved);
+
 	bch_bio_free_pages(&io->bio.bio.bio);
 	kfree(io);
 }
diff --git a/drivers/md/bcache/move.h b/drivers/md/bcache/move.h
index a265578b69cd..6d5a96b06b07 100644
--- a/drivers/md/bcache/move.h
+++ b/drivers/md/bcache/move.h
@@ -91,6 +91,7 @@ struct moving_io {
 	unsigned		read_issued:1;
 	unsigned		read_completed:1;
 	unsigned		write_issued:1;
+	unsigned		has_reservation:1;
 	/* Must be last since it is variable size */
 	struct bch_write_bio	bio;
 };