Hi Jan, Miklos,

On Wed, Apr 12, 2017 at 12:24:40PM +0200, Jan Kara wrote:

Allocate struct backing_dev_info separately instead of embedding it
inside the superblock. This unifies handling of bdi among users.

CC: Miklos Szeredi <miklos@szeredi.hu>
CC: linux-fsdevel@vger.kernel.org
Acked-by: Miklos Szeredi <redacted>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>

...

 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 {
 	int err;
+	char *suffix = "";
 
-	fc->bdi.name = "fuse";
-	fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
-	/* fuse does it's own writeback accounting */
-	fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
-
-	err = bdi_init(&fc->bdi);
+	if (sb->s_bdev)
+		suffix = "-fuseblk";
+	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
+				   MINOR(fc->dev), suffix);
 	if (err)
 		return err;

This call to super_setup_bdi_name would only work with "fuse" but not
with "fuseblk" as mounting a block device in userspace triggers
mount_bdev call which results in set_bdev_super taking a reference
from block device's BDI.  But super_setup_bdi_name allocates a new bdi
and ignores the already existing reference which triggers:

WARN_ON(sb->s_bdi != &noop_backing_dev_info);

as sb->s_bdi already has a reference from set_bdev_super.  This works
for "fuse" (without a blocking device) for obvious reasons.  I can
reproduce this on -rc1 and also found a report on lkml:
https://lkml.org/lkml/2017/5/2/445

Only sane solution seems to be maintaining a private bdi instace just
for fuseblk and let fuse use the common new infrastructure.

Best regards,
Rakesh