Sequence ID calibration for
mis-ordered requests
Huawei Technologies Co.
China
zhangmingqian.zhang@huawei.com
Huawei Technologies Co.
China
yangjing8@huawei.com
Transport
Network File System Version 4
Mis-ordered request
This document updates RFC8881, Network File System (NFS) version 4
minor version 1, by adding two operations to prevent the client from
destroying session when getting the reply of a mis-ordered request with
NFS4ERR_SEQ_MISORDERED.
In NFSv4 minor version 1, sequence ID is used to ensure that the size
of the needed reply cache is tightly bounded. If the server gets a
mis-ordered request, the client will often break the session and
establish a new session with the server. This approach results in a
significant burden on the client and the server. During the process of
session rebuilding, IO performance will be affected. This is especially
troublesome when network latency is substantial, as, for example when
client and server are in different locations. This document will propose
extensions to NFSv4 that would allow client reconnection to be dispensed
with.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119.
In NFSv4 minor version 1, according to RFC 8881, Error code
NFS4ERR_SEQ_MISORDERED is returned by three operations.
The first operation is CREATE_SESSION. csa_sequence is one
argument of this operation, which is used for serializing
CREATE_SESSION via a per-client ID sequence number by the client. In
CREATE_SESSION request, csa_sequence should be equal to ether the
sequence ID in the client ID's slot(a retry), or the slot's sequence
ID + 1(correct normal request). Otherwise, NFS4ERR_SEQ_MISORDERED
will be returned from the server.
The second operation is SEQUENCE. sa_sequenceid is one argument
of this operation. In SEQUENCE request, If the difference between
sa_sequenceid and the server's cached sequence ID at the slot ID is
two (2) or more, or if sa_sequenceid is less than the cached
sequence ID (accounting for wraparound of the unsigned sequence ID
value), then the server MUST return NFS4ERR_SEQ_MISORDERED.
The third operation is CB_ SEQUENCE, which is similar to
SEQUENCE. csa_sequenceid is one argument of this operation. In CB_
SEQUENCE request, If the difference between csa_sequenceid and the
client’s cached sequence ID at the slot ID is two (2) or more,
or if csa_sequenceid is less than the cached sequence ID (accounting
for wraparound of the unsigned sequence ID value), then the client
MUST return NFS4ERR_SEQ_MISORDERED.
Mis-order requests may happen as a result of network partition,
software bug, etc. For such request, the operations subsequent to
SEQUENCE, if any, are not processed, and so slots state (sequence ID,
cached reply) are not changed. That means, requests before this
mis-ordered one were processed correctly and the session state was
correct. In the current implementation, for most of the clients, this
error code will trigger the requester breaking the session and creating
a new session. This process unacceptably interferes with ongoing IO
operations, especially for the IOs on the normal slots.
For example of a persistent session, there are several slots on a
session. Requests on the slots are being processed correctly and replies
are being received normally. Suppose on one slot of them, a mis-ordered
request is received by the server and a response with
NFS4ERR_SEQ_MISORDERED error returns to the client. Then, the client is
going to destroy the session and establish a new session. Before the new
session is ready, new requests will not be performed until the pending
operations finished. The effects on IOs of normal slots will become
dramatic especially in network latency is substantial, as, for example
when client and server are in different locations. The client has to
break the session because it does not know what sequence is expected for
that session and slot. This current cached sequence information would be
available to the client eliminating any need to break the session.
Two operations, SEQENCE_QUERY and CB_SEQENCE_QUERY, are added to
query sequence ID cached when getting NFS4ERR_SEQ_MISORDERED error.
struct SEQUENCE_QUERY4args {
sessionid4 sqa_sessionid;
slotid4 sqa_slotid;
};
struct SEQUENCE_QUERY4resok {
sessionid4 sqr_sessionid;
slotid4 sqr_slotid;
sequenceid4 sqr_sequenceid;
};
union SEQUENCE_QUERY4res switch (nfsstat4 sqr_status) {
case NFS4_OK: SEQUENCE_QUERY4resok sqr_resok4;
default: void;
};
The SEQUENCE_QUERY operation is used by the client to query
sequence ID cached for designated session and slot.
SEQUENCE_QUERY MUST appear as the sole operation type of any
COMPOUND in which it appears. Multiple SEQUENCE_QUERY operations can
be used in one COMPOUND to query multiple sequence IDs cached for
multiple slots. The error NFS4ERR_NOT_ONLY_OP will be returned when
that constraint is not met. Operations other than SEQUENCE,
SEQENCE_QUERY, BIND_CONN_TO_SESSION, EXCHANGE_ID, CREATE_SESSION,
and DESTROY_SESSION, MUST NOT appear as the first operation in a
COMPOUND.
If SEQUENCE_QUERY is received on a connection not associated with
the session via CREATE_SESSION or BIND_CONN_TO_SESSION, and
connection association enforcement is enabled (see Section 18.35),
then the server returns NFS4ERR_CONN_NOT_BOUND_TO_SESSION.
The sqa_sessionid argument identifies the session to which this
request applies. The sqr_sessionid result MUST equal
sqa_sessionid.
The sqa_slotid argument is the index in the reply cache for the
request. The sqr_slotid result MUST equal sqa_slotid.
The sqr_sequenceid field is the cached sequence ID on the slot.
The client SHOULD use this value to calibrate sa_sequenceid in the
next SEQUENCE operation, that is, sqr_sequenceid+1 SHOULD be used as
the sequence ID of the next request on this slot.
For CREATE_SESSION, SEQUENCE operations, if the sequence ID in
the request is mis-ordered(see RFC8881 18.46.3 Section), the replier
will fail the request by NFS4ERR_SEQ_MISORDERED and keep the reply
cache unchanged on the slot of this session. When getting
NFS4ERR_SEQ_MISORDERED error code in the response, the client SHOULD
query the cached sequence ID of the slot and session by
SEQUENCE_QUERY to calibrate its sequence ID for the subsequent
requests. That is, the sequence ID in next request on this slot
SHOULD be sqr_sequenceid+1.
SEQUENCE_QUERY will leave the state of the slot (sequence ID,
cached reply) unchanged and lease of state related to the client ID
not renewed.
If the client is querying an unknown session ID to the server,
the server SHOULD return NFS4ERR_BADSESSION in the response.
If the client is attempting to access a slot the replier does not
have in its slot table (It is possible the slot may have been
retired), NFS4ERR_BADSLOT SHOULD be returned in the response.
struct CB_SEQUENCE_QUERY4args {
sessionid4 csqa_sessionid;
slotid4 csqa_slotid;
};
struct CB_SEQUENCE_QUERY4resok {
sessionid4 csqr_sessionid;
slotid4 csqr_slotid;
sequenceid4 csqr_sequenceid;
};
union CB_SEQUENCE_QUERY4res switch (nfsstat4 csqr_status) {
case NFS4_OK: CB_SEQUENCE_QUERY4resok csqr_resok4;
default: void;
};
CB_SEQUENCE_QUERY is used to calibrate sequence ID of the call
back request of the server for the backchannel of the session.
For each CB_COMPOUND request, the first operation MUST be
CB_SEQUENCE or CB_SEQUENCE_QUERY. If any other operation is in the
first position of CB_COMPOUND except CB_SEQUENCE_QUERY and
CB_SEQUENCE, NFS4ERR_OP_NOT_IN_SESSION MUST be returned.
If the first operation is CB_SEQUENCE, CB_SEQUENCE MUST appear
once. The error NFS4ERR_SEQUENCE_POS MUST be returned when
CB_SEQUENCE is found in any position in a CB_COMPOUND beyond the
first. If the first operation of a CB_COMPOUND is CB_SEQUENCE_QUERY,
CB_SEQUENCE_QUERY MUST be the sole operation type. There can be
multiple CB_SEQUENCE_QUERY in this CB_COMPOUND to request multiple
cached sequence IDs of designated sessions and slots. If any other
operations are found in this CB_COMPOUND, NFS4ERR_NOT_ONLY_OP MUST
be returned.
The csqa_sessionid argument identifies the session to which this
request applies. The csqr_sessionid result MUST equal
csqa_sessionid.
The csqa_slotid argument is the index in the reply cache for the
request. The csqr_slotid result MUST equal sqa_slotid.
The csqr_sequenceid field is the cached sequence ID on the slot.
The server SHOULD use this value to calibrate csa_sequenceid in the
next SEQUENCE operation, that is, csqr_sequenceid+1 SHOULD be used
as the sequence ID of the next request on this slot.
For CB_SEQUENCE operations, if the sequence ID in the call back
request is mis-ordered(see RFC8881 20.9 Section), the client will
fail this request by NFS4ERR_SEQ_MISORDERED and keep the reply cache
unchanged on the slot of this session. When getting
NFS4ERR_SEQ_MISORDERED error code in the response, the server SHOULD
query the cached sequence ID of the slot and session by
CB_SEQUENCE_QUERY to calibrate its sequence ID for the subsequent
requests. That is, the sequence ID in next call back request on this
slot should be csqr_sequenceid+1. CB_SEQUENCE_QUERY will leave the
state of the slot (sequence ID, cached reply) unchanged and the
reply of CB_SEQUENCE_QUERY will not renew the lease of state related
to the client ID on the server side.
If the server is querying an unknown session ID to the client,
the client SHOULD return NFS4ERR_BADSESSION in the response.
If the server is attempting to access a slot the client does not
have in its slot table (It is possible the slot may have been
retired), NFS4ERR_BADSLOT SHOULD be returned in the response.
This memo includes no request to IANA.
Note to RFC Editor: this section may be removed on publication as an
RFC.
All considerations from RFC 8881 security relative sections
[RFC8881].
The authors would like to acknowledge David Noveck for reviews of the
various versions of the draft.