On Gluster storage with sector size of 4096 bytes, buf_align may be
wrong; reading 4096 bytes into unaligned buffer succeeds. This probably
happens because the actual read happens on the Gluster node with aligned
buffer, and Gluster client does not enforce any alignment on the host.

However request_alignment is always right, since the same size is use on
the Gluster node to perform the actual I/O. Use the maximum value for
setting min_mem_alignment.

With this change we can provision a virtual machine with Gluster storage
using VDO device and fuse mount.

This is a partial fix for https://bugzilla.redhat.com/1737256. To make
this work, the management system must ensure that the first block of the
image is allocated, for example:

    qemu-img create -f raw test.img 1g
    dd if=/dev/zero bs=4096 count=1 of=test.img conv=nortunc

Signed-off-by: Nir Soffer <nsoffer@redhat.com>
---
 block/file-posix.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/block/file-posix.c b/block/file-posix.c
index 4479cc7ab4..d29b9e5229 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -1122,7 +1122,7 @@ static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
     }
 
     raw_probe_alignment(bs, s->fd, errp);
-    bs->bl.min_mem_alignment = s->buf_align;
+    bs->bl.min_mem_alignment = MAX(s->buf_align, bs->bl.request_alignment);
     bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
 }
 
-- 
2.20.1