Re: Clarification on delegation behavior for same-client conflicting OPEN
On Thu, 2025-10-30 at 17:35 +0000, Suhas Athani via Devel wrote:
Hello NFS community,
We’re seeking clarification on server behavior for OPEN delegations when the same client issues a potentially conflicting OPEN on a file for which it already holds a delegation. Context and RFC references
* RFC 8881(10.4 Open Delegation) - “There must be no current OPEN conflicting with the requested delegation.” - “There should be no current delegation that conflicts with the delegation being requested.” * RFC 8881(10.4.1 Open Delegation and Data Caching) - For delegation handling, READs/WRITEs without OPEN are treated as the functional equivalents of a corresponding type of OPEN, and the server “can use the client ID associated with the current session to determine if the operation has been done by the holder of the delegation (in which case, no recall is necessary) or by another client (in which case, the delegation must be recalled and I/O not proceed until the delegation is returned or revoked).” * Historical reference: RFC 5661 (obsoleted by RFC 8881) carries the same sections 10.4 and 10.4.1 Questions 1) Same-client conflicting OPEN:
* If a client holds an OPEN_DELEGATE_READ on a file and then the same client issues an OPEN that requires write access (or otherwise conflicts), should the server: *
- Allow the OPEN to complete immediately without recalling the delegation (i.e., no recall necessary for same-client), per RFC 8881 10.4.1; or * Recall the delegation anyway and delay the operation until DELEGRETURN?
The Linux NFS server allows the open to complete, which I think has been the consensus around this point in earlier discussions. Basically, activity from the holder of a delegation is not considered "conflicting". That client presumably knows about any changes and can update its cache accordingly, so we don't need to recall the delegation in this case.
2) OPEN_DELEGATE_WRITE symmetry:
* If a client holds an OPEN_DELEGATE_WRITE and then the same client issues an OPEN that requires read access (or otherwise changes share access/deny modes), should the server similarly allow the operation to proceed without recall, or recall and delay?
WRITE delegations should probably have been called READ_WRITE. The Linux NFS server and the spec treat them as a superset of a READ delegation. So, opening the file for READ when you hold a WRITE deleg is not considered conflicting activity.
3) Any updates since RFC 5661:
* Are there clarifications or consensus updates in RFC 8881 (vs. RFC 5661) or later documents that alter expected behavior in the same-client case?
Thank you in advance for your time and insights. Looking forward to your guidance and clarification on these points.
Regards, Suhas Athani NFS-ganesha Team
_______________________________________________ Devel mailing list -- devel@lists.nfs-ganesha.org To unsubscribe send an email to devel-leave@lists.nfs-ganesha.org
-- Jeff Layton <jlayton@poochiereds.net>
On Fri, Oct 31, 2025 at 7:29 AM Jeff Layton <jlayton@poochiereds.net> wrote:
On Thu, 2025-10-30 at 17:35 +0000, Suhas Athani via Devel wrote:
Hello NFS community,
We’re seeking clarification on server behavior for OPEN delegations when the same client issues a potentially conflicting OPEN on a file for which it already holds a delegation.
Context and RFC references
* RFC 8881(10.4 Open Delegation) - “There must be no current OPEN conflicting with the requested delegation.” - “There should be no current delegation that conflicts with the delegation being requested.”
* RFC 8881(10.4.1 Open Delegation and Data Caching) - For delegation handling, READs/WRITEs without OPEN are treated as the functional equivalents of a corresponding type of OPEN, and the server “can use the client ID associated with the current session to determine if the operation has been done by the holder of the delegation (in which case, no recall is necessary) or by another client (in which case, the delegation must be recalled and I/O not proceed until the delegation is returned or revoked).”
* Historical reference: RFC 5661 (obsoleted by RFC 8881) carries the same sections 10.4 and 10.4.1 Questions 1) Same-client conflicting OPEN:
* If a client holds an OPEN_DELEGATE_READ on a file and then the same client issues an OPEN that requires write access (or otherwise conflicts), should the server: *
- Allow the OPEN to complete immediately without recalling the delegation (i.e., no recall necessary for same-client), per RFC 8881 10.4.1; or
* Recall the delegation anyway and delay the operation until DELEGRETURN?
The only thing I'll add to what Jeff said is.. for the case of CLAIM_DELEGATE_CUR or CLAIM_DELEG_CUR_FH you do not want to recall and wait for a DELEGRETURN, since these Opens need to be done before the client can DELEGRETURN. Also, a client is being dumb if it does any Opens on the FH other than the above 2 Claim types while it holds a Write delegation. However, I don't the the RFC forbids it, so I'd say just do it. (A Write delegation allows Reading/Writing, opening for any acces/deny case locally in the client. The server recalls the write delegation when another client requests any Open for the FH. rick
The Linux NFS server allows the open to complete, which I think has been the consensus around this point in earlier discussions. Basically, activity from the holder of a delegation is not considered "conflicting". That client presumably knows about any changes and can update its cache accordingly, so we don't need to recall the delegation in this case.
2) OPEN_DELEGATE_WRITE symmetry:
* If a client holds an OPEN_DELEGATE_WRITE and then the same client issues an OPEN that requires read access (or otherwise changes share access/deny modes), should the server similarly allow the operation to proceed without recall, or recall and delay?
WRITE delegations should probably have been called READ_WRITE. The Linux NFS server and the spec treat them as a superset of a READ delegation. So, opening the file for READ when you hold a WRITE deleg is not considered conflicting activity.
3) Any updates since RFC 5661:
* Are there clarifications or consensus updates in RFC 8881 (vs. RFC 5661) or later documents that alter expected behavior in the same-client case?
Thank you in advance for your time and insights. Looking forward to your guidance and clarification on these points.
Regards, Suhas Athani NFS-ganesha Team
_______________________________________________ Devel mailing list -- devel@lists.nfs-ganesha.org To unsubscribe send an email to devel-leave@lists.nfs-ganesha.org
-- Jeff Layton <jlayton@poochiereds.net>
_______________________________________________ nfsv4 mailing list -- nfsv4@ietf.org To unsubscribe send an email to nfsv4-leave@ietf.org
On Wed, 2025-11-05 at 12:18 -0800, Rick Macklem wrote:
On Fri, Oct 31, 2025 at 7:29 AM Jeff Layton <jlayton@poochiereds.net> wrote:
On Thu, 2025-10-30 at 17:35 +0000, Suhas Athani via Devel wrote:
Hello NFS community,
We’re seeking clarification on server behavior for OPEN delegations when the same client issues a potentially conflicting OPEN on a file for which it already holds a delegation.
Context and RFC references
* RFC 8881(10.4 Open Delegation) - “There must be no current OPEN conflicting with the requested delegation.” - “There should be no current delegation that conflicts with the delegation being requested.”
* RFC 8881(10.4.1 Open Delegation and Data Caching) - For delegation handling, READs/WRITEs without OPEN are treated as the functional equivalents of a corresponding type of OPEN, and the server “can use the client ID associated with the current session to determine if the operation has been done by the holder of the delegation (in which case, no recall is necessary) or by another client (in which case, the delegation must be recalled and I/O not proceed until the delegation is returned or revoked).”
* Historical reference: RFC 5661 (obsoleted by RFC 8881) carries the same sections 10.4 and 10.4.1 Questions 1) Same-client conflicting OPEN:
* If a client holds an OPEN_DELEGATE_READ on a file and then the same client issues an OPEN that requires write access (or otherwise conflicts), should the server: *
- Allow the OPEN to complete immediately without recalling the delegation (i.e., no recall necessary for same-client), per RFC 8881 10.4.1; or
* Recall the delegation anyway and delay the operation until DELEGRETURN?
The only thing I'll add to what Jeff said is.. for the case of CLAIM_DELEGATE_CUR or CLAIM_DELEG_CUR_FH you do not want to recall and wait for a DELEGRETURN, since these Opens need to be done before the client can DELEGRETURN.
Also, a client is being dumb if it does any Opens on the FH other than the above 2 Claim types while it holds a Write delegation. However, I don't the the RFC forbids it, so I'd say just do it. (A Write delegation allows Reading/Writing, opening for any acces/deny case locally in the client. The server recalls the write delegation when another client requests any Open for the FH.
RFC8881 strongly encourages clients to send OPENs in parallel. If you want to build servers that don't support doing so, then that is of course your right... -- Trond Myklebust Linux NFS client maintainer, Hammerspace trondmy@kernel.org, trond.myklebust@hammerspace.com
participants (3)
-
Jeff Layton -
Rick Macklem -
Trond Myklebust