Mark Belanger a �crit :
>
> "Joseph S. Gardner" wrote:
> >
> > Does ReiserFS support NFS??
>
> There is no relationship between the local
> filesystem and NFS.
>
> If you are using the reiserfs, and have NFS configured
> properly, clients can mount your filesystem via NFS.
> Likewise, you can mount remote NFS filesystems.
Extract from www.reiserfs.org :
What's up with NFS and ReiserFS?
First you should note that NFSD under Linux is weak overall, there are
several ugly problems independent of reiserfs.
There are two nfsd implementations: user mode nfsd and kernel mode nfsd
(both have problems independent of reiserfs). The knfsd is under active
development in a www.sourceforge.net project and in 2.4; unfsd is
"legacy", but stable.
Reiserfs specific problems:
knfsd: has problems to find files again when you run out of the file
handle cache. It works but is very slow. Solution: for 2.4 apply this
patches or for 2.2 apply patches from ftp.suse.com:/pub/people/ak/nfs.
For both you have to let all clients remount. (I have not tried if
Lexa's patches work). For the sourceforge 2.2 backport of the 2.4 server
there is currently no reiserfs patchkit, but Lexa's patch could be
probably fixed for that. It is likely that the sourceforge version will
be merged into future 2.2 kernels (.18 or .19), so someone will have to
do that.
They both have problems with extensive renaming of special files (named
pipes, sockets, etc.) on the exported fs that are not solvable without
major VFS changes in Linux. It is very unlikely that you will hit that
problem, is rather obscure.
unfsd: works fine at first in its limits (no file locking etc.), but
when your file system is big and busy it'll complain after some time
about conflicts in the inodes and some files will behave strangely. The
reason is that it uses the high bits of the 32bit inode number to store
dev_t for multiple file system exports, and reiserfs object id selection
runs into these high bits after some time. Solution is to apply the
unfsd* patch from ftp.suse.com:/pub/people/ak/nfs and use the
--no-cross-mounts option. This implies that you can only export single
file systems, not full file system trees [this feature is not salvagable
with reiserfs and NFSv2, do not ask]
There is a known bug in the patch that is triggered by --no-cross-mounts
that sometimes causes spurious EIO on the root dir of client mounts,
I'll release an update for that later.
