I have an nfs server and a diskless client, both running 2.2.19 under
Debian. If I update files on the server, sometimes the client gets
confused about some files, and stays confused. For instance:
[larsi@sparky ~]$ ls -l /lib/libe2*
ls: /lib/libe2p.so.2: Input/output error
-rw-r--r-- 1 root root 13400 Sep 22 17:15 /lib/libe2p.so.2.3
This is what it looks like on the server:
[larsi@quimbies ~]$ ls -l /tftpboot/sparky/lib/libe2p.so.2*
lrwxrwxrwx 1 root root 13 Oct 30 20:23 /tftpboot/sparky/lib/libe2p.so.2 -> libe2p.so.2.3
-rw-r--r-- 1 root root 13400 Sep 22 17:15 /tftpboot/sparky/lib/libe2p.so.2.3
So the error message is when trying to access the symlink.
If I tell the nfs client to output debug messages:
# echo 65535 > /proc/sys/sunrpc/nfs_debug
I then get the following output:
nfs_lookup_validate: lib/libe2p.so.2 has dud inode
NFS: put_inode(1/1912607313)
NFS: lookup(lib/libe2p.so.2)
NFS call lookup libe2p.so.2
NFS reply lookup: 0
NFS: nfs_fhget(lib/libe2p.so.2 fileid=1912607313)
NFS: __nfs_fhget(1/1912607313 ct=0)
NFS call lookup n
NFS reply lookup: 0
(I think this is the relevant part; it spews out quite a lot of data.)
There are no error messages in any of the log files on the server.
Is this a known bug?
--
(domestic pets only, the antidote for overdose, milk.)
[email protected] * Lars Magne Ingebrigtsen
>>>>> " " == Lars Magne Ingebrigtsen <[email protected]> writes:
> [larsi@quimbies ~]$ ls -l /tftpboot/sparky/lib/libe2p.so.2*
> lrwxrwxrwx 1 root root 13 Oct 30 20:23
> /tftpboot/sparky/lib/libe2p.so.2 -> libe2p.so.2.3 -rw-r--r-- 1
> root root 13400 Sep 22 17:15 /tftpboot/sparky/lib/libe2p.so.2.3
> So the error message is when trying to access the symlink.
<snip>
> Is this a known bug?
Known and fixed in the 2.2.20-pre series. It's just that the symlink
code in 2.2.19 doesn't call nfs_revalidate_inode() in order to verify
data cache consistency.
Cheers,
Trond