> No, fix this _now_ please. This way you accomplish two things:
>
> 1) 64-bit people are happy and can test your new knfsd code
> 2) You avoid possibly having to go through a complete redesign
> to fit everything in there on 64-bit machines
OK, let me explain how I propose to make the filehandle fit, as
otherwise it would be a showstopper problem.
The plan is to remove the dhash and dlen fields, and change d_validate
so that it just verifies that the dentry is a valid dentry with the
given parent. This is necessary anyway, as one of the ways that
d_validate can fail is if the dentry has been renamed after the original
filehandle was created. Since renaming changes the dhash and dlen
values, the validation will fail, even though we want the filehandle to
follow the dentry to its new location. (This is one of the reasons to
use dentries in the first place.)
Assuming d_validate can be suitably modified, that brings the filehandle
budget to
2 pointers @ 8 bytes = 16 bytes
3 inot_t @ 4 bytes = 12 bytes
1 dev_t @ 4 bytes = 4 bytes
for 32 bytes worst case. The ino_t will eventually have to be turned
into nfs psi-style inode numbers if we want to use the full unfsd lookup
mechanism.
Regards,
Bill