Re: CVS commit: src/sys/nfs

[YAMAMOTO Takashi <yamt%mwd.biglobe.ne.jp@localhost>]

hi,

> The whole structure of the code is stupid. We go from:
> 
> nfs_readlink -> nfs_bioread -> nfs_doio -> nfs_doio_read -> nfs_readlinkrpc
>                  |
>                  \-> nfs_readlinkrpc
> 
> We have the proper credentials here in the beginning of the call chain,
> but we stupidly drop them between nfs_bioread and nfs_doio. The proper
> fix is to pass cred down the chain instead of relying on curproc.

prior to ubc, struct buf used to hold credentials.
so chuq might have some opinions.

maybe it isn't very important which credential is used here
because the result is cached and shared between users anyway.

> It would be nice to have the notion of "who are we doing the i/o on behalf
> of" in uio, but we don't, and yes I agree uio_procp->p_ucred is not it
> (since in the UIO_SYSSPACE case uio_procp is NULL).

i'm not sure if it belongs to uio.

how about letting nfs_readlink() hold a credential in n_rcred as
nfs_open() does, so that it can be used by nfs_doio_read() later?

YAMAMOTO Takashi