Hello,
We have a problem with exporting a large and growing file system (currently 18 TB) using NFS.
We use several different Linux distribution in out institute, but mainly Suse Enterprise 10 (SP2).
The file system is built with the logical volume manager. Each time more data is needed,
another LUN of 2 TB is created on the SAN and added to the XFS file system.
This file system is used mainly for large (hundreds of MB) and very large (many GB) files,
although there are also some small file (a few kB).
In the XFS file system, the inode number is related to the location of the inode.
By default the inodes are created in the first 2 TB of the file system,
so that 32-bit inode numbers can be used. We have encountered a situation were
the file system still had a few TB free space, but no new files could be created.
We suspect that this was caused by having no free space in the first 2 TB
of the file system. There is a mount option (inode64), which allows the creation of inodes
in other regions of the file system, but then inodes are created with numbers which
use more than 32-bit. Of course, when such an inode is created, there is no way back.
Therefore, we are now experimenting with another mount option (ino64), on a test system,
which does not change the on-disk structure, but which always adds bits > 31 to the inode number,
so that we can check whether applications can handle 64-bits inodes.
On the system to which the file system is attached (a Lx86_64 system) this is not a problem.
However, we want to use the file system via NFS.
So, this file system is exported normally. We could not find any configuration option on
the NFS server
Then, if we want to mount the file system with NFS on another Lx86_64 system,
the mount command exits with the message:
Stale file handle. There is no information in /var/log/messages.
(Without the ino64 option in the XFS mount, there is no problem with the NFS mount.)
I have tried to experiment with the kernel parameter nfs.enable_ino64, but it has no effect.
I cannot find much more information on the Internet about this subject.
I thought NFSv3 was already capable of using 64-bit inodes.
Has someone a hint to look for solutions to use 64-bit inodes with NFS?
Regards,
Fred.Zwarts.
|