Re: 2.1.70: stange NFS modes

[Bill Hawes]

Finn Arne Gangstad wrote:
> I have the same problem here. NFS server is an aix 4.1.3 machine.
> Linux is 2.1.65 (sorry about that) SMP, dual ppro, 3c905 ethernet card
> 
> voldsboks:finnag:(710)$ mkdir urg; cd urg
> bash: cd: urg: Not a directory
> voldsboks:finnag:(711)$ cd urg
> voldsboks:urg:(712)$
> 
> This greatly complicates untaring tarfiles etc.
> 
> I made a test-program [see end of mail], which gave the following output.
> That it takes exactly 5 seconds before it acts like it should maybe gives
> a hint of what is wrong?
> 
> voldsboks:finnag:(715)$ ./dtst dilldall
> Total failures: 830249, time elapsed: 5.000005 sec
> voldsboks:finnag:(716)$ ./dtst dilldall
> Total failures: 842749, time elapsed: 5.000005 sec
> voldsboks:finnag:(717)$ ./dtst dilldall
> Total failures: 838027, time elapsed: 5.000005 sec

The only thing that comes to mind is that 5 seconds is the lifetime of a
file dentry under NFS -- if it's older than 5 seconds, a new server
lookup is performed. Directory dentries have a lifetime of 10 seconds.

I'm going to add a test to the nfs client mkdir code to check for bad
modes, but  this sounds like it might be a problem with the aix server.

The Linux NFS client doesn't censor the file mode returned by the server
in any way -- it just installs it in the inode. So if we get a trash
value ...

Regards,
Bill