Return-Path: Received: from fieldses.org ([173.255.197.46]:52762 "EHLO fieldses.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751367AbdITSPo (ORCPT ); Wed, 20 Sep 2017 14:15:44 -0400 Date: Wed, 20 Sep 2017 14:15:44 -0400 From: "J. Bruce Fields" To: Chuck Lever Cc: Trond Myklebust , Anna Schumaker , Linux NFS Mailing List Subject: Re: [PATCH] nfs: RPC_MAX_AUTH_SIZE is in bytes Message-ID: <20170920181544.GH14329@fieldses.org> References: <20170920164213.GF14329@fieldses.org> <20170920164951.GG14329@fieldses.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: Sender: linux-nfs-owner@vger.kernel.org List-ID: On Wed, Sep 20, 2017 at 12:56:07PM -0400, Chuck Lever wrote: > > > On Sep 20, 2017, at 12:49 PM, J. Bruce Fields wrote: > > > > On Wed, Sep 20, 2017 at 12:42:13PM -0400, bfields wrote: > >> From: "J. Bruce Fields" > >> > >> The units of RPC_MAX_AUTH_SIZE is bytes, not 4-byte words. This causes > >> the client to request a larger-than-necessary session replay slot size. > > > > By the way, the client's still asking for 3428 bytes after that, which > > seems high. It's mostly the fault of NFS4_MAXLABELLEN, which is 2048. > > The maximum size of NFSv4 LOOKUP replies is also quite large, thanks > to label support (IIRC). It would be nicer for NFS/RDMA, at least, if > the maximum reply sizes of common operations were small (say, less > than 1024 bytes). > > > > I haven't spotted an improvement there yet. > > I haven't either. Session (and I assume RDMA) limits have to be negotiated pretty early, before we know server capabilities. And even if we renegotiated later--I think security label support is per-filesystem, while the connection parameters are server-wide. So, I dunno, all I can think of is putting off fetching the security label till it's needed, but maybe that's not practical--I don't know where it's used. Well, whatever, we can live with it. --b. > > > > Noticed because knfsd was hitting session drc cache limits much to early > > and failing CREATE_SESSION (hence mount). The main trouble is on the > > server side, so I'm relaxing the limits there (and revisiting Trond's > > dynamic slot renegotiation patches). But I thought I should check for > > any easy improvements here too. > > > > --b. > > > >> > >> Signed-off-by: J. Bruce Fields > >> --- > >> fs/nfs/nfs4xdr.c | 4 ++-- > >> 1 file changed, 2 insertions(+), 2 deletions(-) > >> > >> diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c > >> index 37c8af003275..14ed9791ec9c 100644 > >> --- a/fs/nfs/nfs4xdr.c > >> +++ b/fs/nfs/nfs4xdr.c > >> @@ -1842,8 +1842,8 @@ static void encode_create_session(struct xdr_stream *xdr, > >> * Assumes OPEN is the biggest non-idempotent compound. > >> * 2 is the verifier. > >> */ > >> - max_resp_sz_cached = (NFS4_dec_open_sz + RPC_REPHDRSIZE + > >> - RPC_MAX_AUTH_SIZE + 2) * XDR_UNIT; > >> + max_resp_sz_cached = (NFS4_dec_open_sz + RPC_REPHDRSIZE + 2) > >> + * XDR_UNIT + RPC_MAX_AUTH_SIZE; > >> > >> encode_op_hdr(xdr, OP_CREATE_SESSION, decode_create_session_maxsz, hdr); > >> p = reserve_space(xdr, 16 + 2*28 + 20 + clnt->cl_nodelen + 12); > >> -- > >> 2.13.5 > >> > > -- > > To unsubscribe from this list: send the line "unsubscribe linux-nfs" in > > the body of a message to majordomo@vger.kernel.org > > More majordomo info at http://vger.kernel.org/majordomo-info.html > > -- > Chuck Lever > >