mountd eats CPU, issues tons of ioctls

[J. Bruce Fields]

On Thu, Jul 05, 2007 at 07:49:22PM +0200, Marcus Better wrote:
> the rpc.mountd on my NFSv4 server stalls for several minutes at the first mount attempt from the client. During this time, "top" shows mountd eating some 90% of the CPU time. After a while it calms down, the mount succeeds and everything works. strace shows that mountd is issuing lots of ioctls during this time (over 20k calls in a minute or so).
> 
> The system is a Linksys NSLU2 running Debian etch (arm), nfs-kernel-server 1.1.0-4, Debian kernel 2.6.21-1-ixp4xx.
> 
> I have about the same setup on other i386 servers that don't show this behaviour. Of course the NSLU2 is a really slow device, but it still seems suspicious. The server uses Kerberos authentication.
> 
> Below are parts of the strace output. Any idea what might be wrong?
> 
> Regards,
> 
> Marcus
> 
> ioctl(10, DM_TABLE_DEPS, 0x3b418)       = 0
> ioctl(10, DM_TABLE_DEPS, 0x3b418)       = 0
> ioctl(10, DM_TABLE_DEPS, 0x3b418)       = 0
> ioctl(10, DM_TABLE_DEPS, 0x3b418)       = 0
> ioctl(10, DM_LIST_DEVICES, 0x37410)     = 0

I have no idea what those ioctl's are.... With some help from google and
grep--it seems to be some problem with libblkid and md.  I seem to
recall this being discussed on either this list or the sourceforge.net
list recently, but can't find it now.  This might be relevant:

	http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=413661

(And let us know how the NSLU2 works; I made one attempt to get debian
installed on one and failed.  Tom Talpey got much further, but I don't
recall whether he said he got krb5 working....)

--b.