I have now tried 2.5.31-1um twice and have the same problem on both.
I have been having problems with hostfs failing when I try to mount it.
if I have slab poisoning on it faults on a 0x5a5a5aXX style address
if it is off it is kernel mode fault at 0xfffffff4...
what appears to me to be going on is that
where it looks like the iget is trying to dereference the dentry parent
which is not initialized yet or if slab poisoning is off it notices that it
does not have a parent and dies at 68 where it does a dereference from the
Also is the HOST_I used for some special reason? it is used later in about 6
places in one function to allow it to free the name and other kmalloced
structs. To me it looks like it could be stuffed into a local var and done
once rather the being called repeatedly.
It also looks like lines 872 and 865 have the definitions of the
hostfs_types, and hostfs_root_types cases swapped and while it should not
hurt anything it looks strange.
Further while debugging I noticed that the device name is not checked which
is how "none" works getsbnodev passes it around so it could be possible to
use mount -t hostfs /host/path /uml/mnt/point by changing from using the
data pointer to using the devname pointer why does it not do this now? It
would then have syntax like all other mount commands if necessary I suppose
a "HOSTFS:/local/uml/filesystem" syntax could be used rather than just
putting the host path into the device. Yeilding a almost NFS style