From: Brent Callaghan (Brent.Callaghan@eng.sun.com)
Date: 07/16/98-05:44:31 AM Z
Date: Thu, 16 Jul 1998 03:44:31 -0700 (Pacific Daylight Time) From: Brent Callaghan <Brent.Callaghan@eng.sun.com> Subject: Re: nfs-v4 novice questions Message-ID: <Roam2.0.6.900585871.19201.brent@jurassic.eng.sun.com> >After I read Strawman document, I have the following questions: > >1) In NFS version 3 RFC, it has attributes returned for almost procedures > which reduces a lot of GETATTR calls. Why do you take this ability out >of nfs-v4? The ability is still there - a client can add GETATTR calls to any and every compound request if it wants attributes updated. >2) In page 8, there exists bitmap for attributes returned. Where is the >information > in this document to describe the mapping for multiple attributes for >files >and > filesystems? What's the procedures to get information for filesystems? GETATTR. Why should filesystems have their own set of attributes ? It's a little more space efficient if you can get filesystem attributes at the same time as you retrieve file attributes. >3) Through this document, I have impression that there will be no LOCKD and > STATD programs for nfs-v4 since there exist LOCKxx procedures. Is it >true? Yes. >4) Will there exist MOUNT program for nfs-v4? If yes, where is the document? > If not, how does MOUNT program/version fit in nfs-v4? It doesn't. There is no need for MOUNT if you can do a multicomponent lookup from a root filehandle as WebNFS clients can now. >5) What's the difference between NFS4ERR_FHEXPIRED and NFS4ERR_STALE? > What's the situation to return the return code for each one? These need further explanation. EXPIRED indicates that the filehandle (not the file) has changed. STALE indicates that the file has been removed. >6) There exist some procedures have the same function with nfs-v3 RFC. > Why can they have the same procedure number? It will be much easier > for the comparison. They're in alphabetical order now. You can still use the names for comparison. >7) Is there any plan to put the definition for arguments and results with >procedures > instead of lump on the back of document? Most of the arguments and results are simple sequences of scalar values. I don't think enclosing them in an XDR struct adds any real value - though if the consensus is that they should be added then I don't think that would be a problem. >8) I thought one of the purpose for nfs-v4 is to improve performance. That >is why nfs-v3 lump attributes with procedures. But nfs-v4 seems to me is one >procedure > for one purpose; e.g. in page 35, it uses PUTFH, LOOKUP and GETFH. > Why do you use one procedure to do these 3 jobs instead of one? Do I miss > something here? Yep, I think you've missed the rationale for compound ops - discussed at the beginning of the document. >9) Why can you combine LOOKUP and LOOKUPP into one procedure? What pathname would you suggest to represent a "parent" directory ? >10) Why do you need NVERIFY and VERIFY procedures in server since the client >is likely do it anyway? They are useful for aborting a request that should not proceed. How do you propose the server second-guess the client's criteria for aborting a request ? >11) What is the purpose of RESTOREFH, SAVEFH, PUTFH, and PUTROOTFH > since NFS is stateless protocol and FH in procedure is the one you >are working on? They're for putting a filehandle aside temporarily. The filehandle is not saved beyond the completion of the compound request. For instance, after a CREATE the post-op attributes might be obtained from the saved directory filehandle. >12) In page 79, AUTH_RPCSEC_GSS in rpc_flavor4 = 4. > Based on RFC2203, RPCSEC_GSS in auth_flavor = 6. > What is the difference between these two? Yes, that's a typo - we should fix it. Thanks Brent
This archive was generated by hypermail 2.1.2 : 03/04/05-01:45:58 AM Z CST