Basically, it all sounds reasonable to me. What feedback I've seen from Sun
basically recognizes that directory offset cookies can't be implemented
correctly and were a mistake. (Servers should just always send a 0 cookie
and ignore ones returned from clients.) You might try email to
nfs3@sun.eng.com and see if they give you the same response?

As for putting offsets in the buf structure, I think it is the best place
to keep them (and you noted, when I wrote the code I was trying to avoid
buf structure changes, etc.). At that point, you are back to handling
directories just like v2 did, which seems to be about as good as you can
do. The idea of doing a lot of client readahead for directories sounds
good, too. (I would set an upper bound on "all", just in case someone
exports a directory with many thousands of entries. Not a particularily
good idea, but somebody will do it anyhow:-)

As for the BAD_COOKIE error return from a server, if Sun considers it
fatal, you probably should too.

Sounds like a good little project, rick