Subject: nfs bug?
To: None <current-users@netbsd.org>
From: None <uwe@6bone.informatik.uni-leipzig.de>
List: current-users
Date: 05/25/2004 22:37:20
hello,

if you mount an nfs directory from a SFU 3.5 nfs-server you cannot
execute a "rm -rf" command at a non-empty directory. the error code is
ERR_NOTEMPTY. if you try "rm -f" all works well.

example at a SFU directory:

mkdir -p foo/foo/foo
rm -rf foo --> Error
rm -f foo  --> work fine

the microsoft support tells me the following:

#######################################################

So the issue here is that you're seeing this problem on SFU 3.5 NFS Server? Or
is it still SFU 2.3? For the record, we don't test with NetBSD at all.
Only supported flavors are HP-UX, AIX, Solaris and RH Linux.

Your network traces tell me that nothing is wrong with NFS Server. The two
captures where you have the rm succeeding show that the directory
hierarchy is like this - .\foo\foo\foo and these 4 directories have
handles like 000A - C522 - C622 - C722.

If you look at the traces closely, the two cases where the experiment
succeeded show that you rightly get three RMDIR (remove directory) calls
in the order C722 - C622 - C522 and everything is as expected.

In the failure trace, you see that the client knows that the directory
hierarchy is like 000A - C5222 - C622 and is never sending an RMDIR on
C622 but is sending a RMDIR directly on C522. This is rightly failed by
the NFS Server with ERR_NOTEMPTY.

Seems to me like the client is goofing up here

#############################################################

Can anyone confirm, if it is really a NetBSD nfs problem?
In the other case I will call the microsoft support again.


Thank you for your efforts
Uwe