On 5/29/13 2:01 PM, Jeremy Allison wrote:
On Wed, May 29, 2013 at 01:58:08PM -0700, Jeremy Allison wrote:
On Wed, May 29, 2013 at 01:54:20PM -0700, Anand Avati wrote:
3. In our test case the fd number returned by vfs_gluster for the
initial open file _was_ 0. That dirfd() condition needs a fix, I
think?
Yep. That's why I fixed it in master and 4.0.x, it already came up in
a breakage of a different vfs module that used the same trick.
I don't know if the master branch has equivalent code elsewhere,
couldn't find it easily..
Oh now I understand, that's the broken code I fixed in master :-).
It's already been back-ported to 4.0.x.
You'll need to back port the dirfd removal fixes from 4.0.x to
3.6.x in order for this to wotk.
The bugzilla entry is #9778 - Samba directory code uses dirfd() without vectoring through a VFS call.
Re-open it and request a back-port of this code to 3.6.x and I'll work on it.
Sorry, not easy for me to do right now.
Please find the backported patches (against v3-6-stable) here -
https://github.com/avati/samba/commits/vfs_glusterfs-v3-6-stable
There was only one conflict in "Remove dependency on detection of
HAVE_DIRFD for use of fdopendir()." which is resolved in there. Those
set of patches should cleanly apply on v3-6-stable.
Thanks,
Avati
