Hello,
in the current kernel nfsd, a buffer of (8+1)k is allocated for each
kernel nfsd thread. Since the NFS ACL protocol requests arrive at the
same transport endpoint, in the short term (i.e., for the NFS ACL
protocol) I need a limit of (24+1)k for each nfsd thread. While this
wastes memory that will not be used most of the time, the amount of
wasted memory is still not totally crazy.
However, the NFS ACL protocol still does not support Extended Attributes
in general, so somebody will have to implement one of the EA NFS
protocols in addition. (I'm leaning towards the one protocol extension
from SGI OB1). This will require a buffer of 64k. So I'm wondering how
we could solve the problem of buffer space more flexibly. Any ideas?
Thanks,
Andreas.