(1) Are dentry pointers safe and effective as nfs filehandles, and
(2) Given the above, are the dentry use counts tracked correctly?
I'm still convinced 2) is a yes. 1) is still up for grabs.
The issue is how to implement things without putting dentries (or
something which gets you to a dentry quickly) into the file handles?
All of the code was straightforward to do if I had the dentries right
there and then.
Yeah, we can go back to using inode's. But this is going to make
things very nontrivial to work again, even more so since in Linus's
latest 2.1.pre52 code the ability to go back from an inode to a dentry
has been completely erased. We'd need to design our own translation
mechanism, and it would be only for nfsd, this should be avoided if
possible.
Step 1 is to find out if the dentry based scheme can be made to work
at all. If you find that the answer to that is an absolute no, then
step 2 is to find out how to make it all work with inode's again.
Later,
David "Sparc" Miller
davem@caip.rutgers.edu