Troubleshooting 64-bit "cookie" / hash issue on Solaris 10

[Ray Van Dolson]

Apologies for cross-posting on the technical list.  After doing some
significant reading today though, I'm thinking this may be the more
appropriate audience.

Ran into an issue[1] with the vendor (Oracle) provided Samba stack in
Solaris 10 talking to an ext4-backed NFS export on a RHEL6 server.
Even after upgrading to the latest Samba provided by Oracle (3.6.15)
the issue persisted.

After much Googling, came across a tip that led us to disable dir_index
on our ext4 file system.  Lo and behold this fixed our issue!  But am
trying to understand which piece of the equation was breaking down
here (Linux kernel, Linux NFS daemon, Solaris or Samba).

Based on comments on LWN[2], am thinking that the issues we were seeing
could be due to how Samba is handling the readdir() results it gets
back.  Although I see mention of Samba being impacted by the ext4/ext3
changes on the LWN articles as well as mention of impact in a patch to
ext3 and ext4 by Eric Sandeen[3], I was unable to find a bug in Samba's
bug database nor anything on the samba-technical mailing list that led
me to believe this may have been fixed.  Perhaps my search-fu just
sucks though.

I'm also wondering if perhaps Oracle is just building their Samba
wrong or if we need to apply some other underlying parts of Solaris to
make things work.

Would any Samba developers be willing to comment on the state of this
issue and whether or not it still is affecting Samba?

Thanks,
Ray

[1] https://lists.samba.org/archive/samba/2013-August/175136.html
[2] https://lwn.net/Articles/544772/
[3] http://thread.gmane.org/gmane.comp.file-systems.ext4/37932