On Mon, Sep 02, 2002 at 02:48:47PM +0100, David Brownlee wrote:
> On Sun, 1 Sep 2002, Harold Gutch wrote:

[ NFS write corruption ]

> > The first 32 bytes of every 4096 byte block (except for the first
> > one) seem to be corrupted.
> 
> 	Very odd. Is 'sysctl hw.pagesize' 096 on the dreamcast?
> 	I'd be inclined to play with any of the available mount
> 	options (v2/v3 tcp/udp) to see if they affect matters.

hw.pagesize is set to 4096, yes.  

I now did some testing:

  1) NFSv2, UDP:
     0-bytes from offset 4096 to offset 8191, for every 8192 byte
     block (4096 - 8191, 12288 - 16383 etc.)
  2) NFSv2, TCP:
     the same as NFSv2/UDP
  3) NFSv3, UDP:
     32 byte file corruption at weird positions:
     4096, 4352, 4480, 4672, 4768, 4928, 5024, 5280, 5504
     (at this point i stopped comparing)
  4) NFSv3, TCP:
     the same as NFSv2/UDP

I used "-r=1024,-w=1024" for all of these tests, copied /netbsd
to /tmp/netbsd and then compared them on the NFS server.


> 	What OS is your NFS server running (I don't think that
> 	would be relevant, but it might be worth testing another).

NetBSD-current as of Aug 24 on i386.  For testing purposes I also
exported the Dreamcast-filesystem to the NFS-server itself and
mounted it via NFS - copying the file on the NFS-filesystem and
then comparing it on the local filesystem didn't give any
differences either.


> 	You could always nfs export a small mfs from the dreamcast to
> 	itself and test on that...

I'll give that a try later on tonight.


bye,
  Harold