We just noticed problems with NFS mounting directories from SGI servers
that contain directories that are exported with the SGI-specific
"nohide" option. Here is our NFS configuration:
NFS-Server: SGI machine "lumen" with IRIX 6.5.3m:
/etc/exports:
/ -nohide <list of machines>
/nobackup -nohide <list of machines>
NFS-clients: Linux machine:
RedHat6.0, kernel 2.2.5, (with knfsd-1.2.2-4 in case it should be
relevant) or
Debian2.1, kernel 2.2.7, with (nfs-server 2.2beta37-1)
And here is what happens:
-------------------------------------
# mount lumen:/ /mnt
# ls -F /mnt
/mnt:
6.2/ dev.index hw/ lib64 nobackup/
sbin/
Desktop dev/ irix65 licensed ns/
usr/
aw/ dumpster/ irix@ local opt/
var/
bin@ etc/ lib/ mnt/ proc/ u/
debug home/ lib32 sat/ unix*
root@necropolis / # ls -F /mnt
/mnt: Input/output error.
-------------------------------------
It is as simple as mounting the directory and having ls(1) stat(2) the
files in there (a simple ls(1) without the -l or -F option does not
trigger the bug).
The same test on a system with generic RedHat5.2 does not show the bug,
though. To me this seems to indicate a bug in the 2.2.X kernel.
Everything works just fine with an SGI as a client, as well.
Here are the linux kernel messages related to the bug
-------------------------------------
Jun 4 15:46:45 necropolis kernel: __nfs_fhget: inode 128 busy,
i_count=2, i_nlink=34
Jun 4 15:46:45 necropolis kernel: nfs_free_dentries: found ///,
d_count=26, hashed=0
Jun 4 15:46:45 necropolis kernel: __nfs_fhget: inode 128 still busy,
i_count=2
Jun 4 15:46:45 necropolis kernel: __nfs_fhget: killing /// filehandle
Jun 4 15:46:45 necropolis kernel: nfs_readdir: inode is NULL or not a
directory
-------------------------------------
We were under the impression that the nohide option is transparent to
the client, but apparently the Linux implementation does see a
difference on the client side.
Anyone out there who has seen this before and has an idea of how to fix
this bug?
Thanks,
Philipp Slusallek
P.S.: Please also reply directly by email as I do not normally read
these newsgroups/mailing lists.
P.S.: If required we could make such a mount point available to test
this bug. problem.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/