From: "J. Bruce Fields" Subject: Re: [PATCH 05/31] nfsd41: encode create_session result into cache Date: Fri, 15 May 2009 19:05:18 -0400 Message-ID: <20090515230518.GG26389@fieldses.org> References: <1240938005-23778-1-git-send-email-andros@netapp.com> <1240938005-23778-2-git-send-email-andros@netapp.com> <1240938005-23778-3-git-send-email-andros@netapp.com> <1240938005-23778-4-git-send-email-andros@netapp.com> <1240938005-23778-5-git-send-email-andros@netapp.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: pnfs@linux-nfs.org, linux-nfs@vger.kernel.org To: andros@netapp.com Return-path: Received: from mail.fieldses.org ([141.211.133.115]:52094 "EHLO pickle.fieldses.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754036AbZEOXFS (ORCPT ); Fri, 15 May 2009 19:05:18 -0400 In-Reply-To: <1240938005-23778-5-git-send-email-andros@netapp.com> Sender: linux-nfs-owner@vger.kernel.org List-ID: On Tue, Apr 28, 2009 at 12:59:39PM -0400, andros@netapp.com wrote: > From: Andy Adamson > > CREATE_SESSION can be preceeded by a SEQUENCE operation and the > create session single slot cache must be maintained. Encode the results of > a create session call into the cache at the end of processing. > > The create session result will also be encoded into the RPC response, and if > it is preceeded by a SEQUENCE operation, will also be encoded into the > session slot table cache. > > Errors that do not change the create session cache: > A create session NFS4ERR_STALE_CLIENTID error means that a client record > (and associated create session slot) could not be found and therefore can't > be changed. NFSERR_SEQ_MISORDERED errors do not change the slot cache. > > All other errors get cached. > > Signed-off-by: Andy Adamson > --- > fs/nfsd/nfs4state.c | 6 ++++-- > fs/nfsd/nfs4xdr.c | 18 ++++++++++++++++++ > include/linux/nfsd/xdr4.h | 2 ++ > 3 files changed, 24 insertions(+), 2 deletions(-) > > diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c > index fee6bf0..142c460 100644 > --- a/fs/nfsd/nfs4state.c > +++ b/fs/nfsd/nfs4state.c > @@ -1376,7 +1376,7 @@ nfsd4_create_session(struct svc_rqst *rqstp, > if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) || > (ip_addr != unconf->cl_addr)) { > status = nfserr_clid_inuse; > - goto out; > + goto out_cache; > } > > slot = &unconf->cl_slot; > @@ -1418,7 +1418,9 @@ nfsd4_create_session(struct svc_rqst *rqstp, > memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data, > NFS4_MAX_SESSIONID_LEN); > cr_ses->seqid = slot->sl_seqid; > - > +out_cache: > + /* cache solo and embedded create sessions under the state lock */ > + nfsd4_cache_create_session(cr_ses, slot, status); > out: > nfs4_unlock_state(); > dprintk("%s returns %d\n", __func__, ntohl(status)); > diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c > index b2f8d74..1dfec1d 100644 > --- a/fs/nfsd/nfs4xdr.c > +++ b/fs/nfsd/nfs4xdr.c > @@ -3046,6 +3046,24 @@ nfsd4_encode_create_session(struct nfsd4_compoundres *resp, int nfserr, > return 0; > } > > +/* > + * Encode the create_session result into the create session single DRC > + * slot cache. Do this for solo or embedded create session operations. > + */ > +void > +nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses, > + struct nfsd4_clid_slot *slot, int nfserr) > +{ > + __be32 *p = (__be32 *)slot->sl_data; > + struct nfsd4_compoundres tmp = { > + .p = p, > + .end = p + XDR_QUADLEN(CS_MAX_ENC_SZ), > + }, *resp = &tmp; bfields@pig:~/local/linux-2.6$ gdb vmlinux GNU gdb 6.8-debian ... This GDB was configured as "i486-linux-gnu"... (gdb) p sizeof(struct nfsd4_compoundres) $1 = 588 So that's a 588 byte structure on my machine--better not to put it on the stack if we can avoid it. This trick of faking up a temporary nfsd4_compoundres to be able to use nfsd4_encode_create_session() unmodified is sort of brilliant but possibly also sort of nuts? I don't know. Anyway, maybe: - rewrite nfsd4_encode_create_session() to call a common encode_create_session that takes just the start and end pointers, then call that? Or - forget the xdr encoding entirely and just save the values we care about from nfsd4_create_session? (Heck, there's no pointers in there or anything--you could just make a copy of the whole struct.) I'm leaning towards the latter--the xdr encoding seems unnecessary, and introduces error-handling for a case (overflowing the CS_MAX_ENC_SZ-size buffer) that we'll never hit. And if we're making the clientid slot such a special case then I suppose there's no need to xdr-encode for uniform handling with the regular sessions slot case. --b. > + > + slot->sl_status = nfsd4_encode_create_session(resp, nfserr, cr_ses); > + slot->sl_datalen = (char *)resp->p - (char *)p; > +} > + > static __be32 > nfsd4_encode_destroy_session(struct nfsd4_compoundres *resp, int nfserr, > struct nfsd4_destroy_session *destroy_session) > diff --git a/include/linux/nfsd/xdr4.h b/include/linux/nfsd/xdr4.h > index afd4464..21e0b73 100644 > --- a/include/linux/nfsd/xdr4.h > +++ b/include/linux/nfsd/xdr4.h > @@ -517,6 +517,8 @@ extern __be32 nfsd4_setclientid_confirm(struct svc_rqst *rqstp, > extern void nfsd4_store_cache_entry(struct nfsd4_compoundres *resp); > extern __be32 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp, > struct nfsd4_sequence *seq); > +extern void nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses, > + struct nfsd4_clid_slot *slot, int nfserr); > extern __be32 nfsd4_exchange_id(struct svc_rqst *rqstp, > struct nfsd4_compound_state *, > struct nfsd4_exchange_id *); > -- > 1.5.4.3 >