AIX

Hi,

I have a server running AIX 5.3.0.0 ML 5, and on two occasions have seen issues where when the client side executes an 'ls', the output doesn't return all files (there is no difference in permissions between a file that can be listed, and a file that can't be listed), and I'm using the anonymous uid to access. At the moment I believe that both issues, although slightly different have the same root cause (feel free to disgree!):

First incident:

NFS client side = Services For Unix 3.5 NFS gateway.

The NFS gateway was able to mount the file system and list absolutely fine, however clients accessing the NFS gateway server were unable to list all files. At the time I mounted the same file system onto another AIX (5.3.0.0. ML 5) and could list all files as normal. This appears to have been resolved at the time by re-exporting the file system and re-mounting the file system on the NFS gateway server.

Second incident:

NFS client side = AIX 5.3.0.0

The client side in this incident was the same server that I was unable to replicate incident 1 with. The client side was able to list all files, apart from one file on the export. I unmounted and remounted the file system, and the server was able to list it.

In both of the above, I was able to access the file from the NFS client by using the file name (i.e. ls -l <filename>), however the file wasn't listed using 'ls'.

I did find APAR IY73615 which appears to be a similar problem, however this concerns non AIX clients (so only applicable to incident 1). By the time incident 2 occured (a week later) I was aware of the above, and touched the affected file, but was still unable to list it (though this was AIX to AIX). Both AIX servers are also on ML 5, and this problem appears to have resolved in ML 3.

I've spent quite a bit of time reading (redbooks, IBM's documentation site, forums) but haven't been able to find anything concerning this problem. The next time it happens I'll try to use syslog to generate some output, but am keen to know if anyone has come across this, or any thoguhts on how best to tackle it.

Many thanks,
Rich

Tbh, I do not trust/like NFS much cause of similar effects described by you. So if possible, maybe upgrade to the latest TL available and hopefully have it fixed by this. TL 5 is rather old.

TL 12, ie. Service-Pack 5300-12-03-1107 is the latest.

Thanks - that's an option I can certainly look at. I'm sure you'll appreciate that I'll need some firm evidence of a resolution in the most recent TL before upgrading to it. Will wait with anticipation for the next occurance!

Not sure, but I would think this is normal behaviour (at least we are treating it as normal behaviour in our company - and at least it still occurs with TL11) as the client always only can see what had been within the exported filesystem during the time of the export - except new content has been added by the client himself (in case of a rw mount) - I was thinking that the export puts something into the share that is comparable to a .toc file in an AIX filesystem where you can for example install only packages from that had been there already when the .toc has been written - but when you try to install something that came afterwards into the same filesystem, you cannot?

Regards
zxmaus

Thanks zxmaus,

NFS is designed to allow both the server and client to update remote file systems and retrieve their updated contents on the fly. I've just signed up, so can't post links, but googling RFC 1813 will allow you to confirm the protocol specification on faqs.org. The paragraph I'm looking at says:

"Procedure READDIR retrieves a variable number of entries, in sequence, from a directory and returns the name and file identifier for each, with information to allow the client to request additional directory entries in a subsequent READDIR request"

I believe that should allow me to list the contents of a directory if it's changed by the server. This is the way that I've always experienced NFS working in the past.