[lug] Sun Client Redhat AS3 NFS Probs
Jeffrey A. St. Pierre
Jasp2 at Colorado.EDU
Mon May 17 12:15:12 MDT 2004
On Mon, 17 May 2004, D. Stimits wrote:
> >
> > OK, so you are telling me that every exported filesystem requires
> > a quota file? So I have three choices 1. to turn rquotad off 2.
> > put every single exported filesystem under quota control, or 3.
> > deal with errors
> >
>
> At least for testing. Until you know for sure what is happening, it's
> one way to figure it out.
>
> > To me that would seem like a massive oversight in whoever wrote
> > 'quota'. I mean, what if I only want quotas on my home
> > directories but not my project directories? Anyway... let's give
> > it a try. Odds are you guys know more than I do about this
> > stuff.
> >
>
> Quota is a facet of security. Denial of service through remote
> filesystem exploit to fill up the drive with nonsense is one example.
> The whole point of quota is that you can't find a way around it if it is
> active. If either you have the feature added in, or if the filesystem
> type requires the feature, then a logged warning is pretty mild.
>
Well, it really could be a mess. I had just one of almost 100
projects moved to the new NFS server, and it was seriously
filling up my logs. Now, if this really is a security issue, am
I really better off having the quota files setup but no actual
quotas set? I only want quotas on the home directories (which
haven't been moved yet, but will be) and I want the researchers
(the users) to be able to fill their projects right to capacity
if they so desire. Quotas as part of protecting filesystem
overflows seems more applicable to /var which generally isn't
exported, so rquotad wouldn't notice it. Also, I don't know of
anyone who applies quotas to /var... Maybe I'm missing the point,
or not making my own clear. Here are the pieces of the puzzle as
I see it.
A. I want rquotad on because /export/home will have quotas on it
that people want to check from client systems.
B. I am also exporting /export/projXX which I don't want quotas
on them.
C. I still need setup quota files for the /export/projXX drives
and mount them with quota options turned on, because otherwise
everytime any user on any client runs a 'quota -v' every mounted
directory without quotas will generate errors on the server.
Don't make sense to me, and I'm not convinced quotas are a facet
of security. I think it is better classified as a facet of
controling the users, which may have security issues.
> > Thanks much to everyone who gave me input on this. However, I
> > still feel this is a bit of a bug. If there are no quotas setup
> > for a particular filesystem, doesn't it make more sense for the
> > daemon to just ignore the request for that filesystem, rather
> > than filling the messeges file with ambiguous errors? I mean
> > 'rpc.rquotad: Can't find filesystem mountpoint for directory
> > /export/rd02/diag' seems unrelated to the actual problem. If
> > it insists on giving an error, don't you think 'No quota files
> > for filesystem /mnt/filesystem' would be a better error
> > message? Do you think I should report this to the developers?
>
> Error messages being either misleading or lacking detail has been a
> problem since the beginning of computers. A better error message might
> be something worth mentioning to the developers. If this were something
> you'd seen before, you could go right to the fix, so to the developers
> the message probably isn't misleading until they hear from someone that
> had the problem while also unfamiliar with the setup.
>
> D. Stimits, stimits AT comcast DOT net
> _______________________________________________
>
Ok... That makes sense, I may submit a bug report just to improve
the error messages.
Thanks again,
-Jeff
More information about the LUG
mailing list