On Mon, Nov 04, 2013 at 02:30:50PM -0700, Eric Blake wrote:

A qcow2 file with a backing file of 'gluster://host/vol/file' should not try to look for a directory named './gluster:/' in the file system.

* src/util/virstoragefile.c (virBackingStoreIsFile): Broaden check to include all protocols.

Signed-off-by: Eric Blake <eblake@redhat.com> ---

This one will be important to my work on gluster backing chains, but can be reviewed and applied independently now.

src/util/virstoragefile.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/src/util/virstoragefile.c b/src/util/virstoragefile.c index 1b4d4cf..8f04b78 100644 --- a/src/util/virstoragefile.c +++ b/src/util/virstoragefile.c @@ -671,8 +671,13 @@ virStorageFileMatchesVersion(int format, static bool virBackingStoreIsFile(const char *backing) { - /* Backing store is a network block device or Rados block device */ - if (STRPREFIX(backing, "nbd:") || STRPREFIX(backing, "rbd:")) + char *colon = strchr(backing, ':'); + char *slash = strchr(backing, '/'); + + /* Reject anything that looks like a protocol (such as nbd: or + * rbd:); if someone really does want a relative file name that + * includes ':', they can always prefix './'. */

This makes complete sense, I just can't figure out how qemu treats this because when I tried it on upstream now, it ended up with a segfault :)

+ if (colon && (!slash || colon < slash)) return false; return true; }

Anyway, ACK. Martin