> On Aug 28, 2024, at 2:30 AM, NeilBrown <neilb@xxxxxxx> wrote: > > On Wed, 28 Aug 2024, Mike Snitzer wrote: >> On Wed, Aug 28, 2024 at 11:12:00AM +1000, NeilBrown wrote: >>> >>> The "AUTH_UNIX check below" only applies if exp->ex_flavours == 0. >>> To make "rqstp == NULL" mean "treat like AUTH_UNIX" I think we need >>> to confirm that >>> exp->ex_xprtsec_mods & NFSEXP_XPRTSEC_NONE >>> and either >>> exp->ex_nflavours == 0 >>> or >>> one for the exp->ex_flavors->pseudoflavor values is RPC_AUTH_UNIX >>> >>> I'm not sure that is all really necessary, but if not then we probably >>> need a better comment... >> >> Think extra checks aren't needed (unless you think a NULL rqstp >> _without_ the use of LOCALIO possible? which could trigger a false >> positive granting of access? seems unlikely but...) >> > > I agree they aren't needed. I think we need to have a clear > understanding of why that aren't needed, and to write that understanding > down. So that if some day someone wants to change this code, they can > understand the consequences. > I don't know what is best, but I think we should have a comment > justifying the short-circuit, and I don't think the current proposed > comment does that correctly. My goal is to document that understanding here, as you stated. I will leave it to you and Mike to adjust the wording to your liking. -- Chuck Lever
