Network File System Version 4 | D. Noveck |
Internet-Draft | HPE |
Intended status: Standards Track | August 18, 2016 |
Expires: February 19, 2017 |
RPC-over-RDMA Extension to Manage Transport Characteristics
draft-dnoveck-nfsv4-rpcrdma-xcharext-02
This document specifies an extension to RPC-over-RDMA Version Two. The extension enables endpoints of an RPC-over-RDMA connection to exchange information which can be used to optimize message transfer.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on February 19, 2017.
Copyright (c) 2016 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
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 [RFC2119].
This document specifies an extension to RPC-over-RDMA Version Two. It allows each participating endpoint on a single connection to communicate various characteristics of its implementation, to request changes in characteristics of the other endpoint, and to effect changes and notify the other endpoint of changes to these characteristics during operation.
The extension described herein specifies OPTIONAL message header types to implement this mechanism. The means by which the implementation support status of these OPTIONAL types is ascertained is described in [rpcrmdav2].
Although this document specifies the new OPTIONAL message header types to implement these functions, the precise means by which the presence of support for these OPTIONAL functions will be ascertained is not described here, as would be done more appropriately by the RFC defining a version of RPC-over-RDMA which supports protocol extension.
This document is currently written to conform to the extension model for RPC-over-RDMA Version Two as described in [rpcrdmav2].
A number of different terms are used regarding the roles of the two participants in an RPC-over-RMA connection. Some of these roles last for the duration of a connection while others vary from request to request or from message to message.
The roles of the client and server are fixed for the lifetime of the connection, with the client defined as the endpoint which initiated the connection.
The roles of requester and responder often parallel those of client and server, although this is not always the case. Most requests are made in the forward direction, in which the client is the requester and the server is the responder. However, backward-direction requests are possible, in which case the server is the requester and the client is the responder. As a result, clients and servers may both act as requesters and responders.
The roles of sender and receiver vary from message. With regard to the messages described in this document, both the client and the server can act as sender and receiver. With regard to messages used to transfer RPC requests and replies, the requester sends requests and receives replies while the responder receives requests and sends replies.
An initial set of receiver and sender characteristics are specified in this document. An extensible approach is used, allowing new characteristics to be defined in future standards track documents.
Such characteristics are specified using:
<CODE BEGINS> typedef xcharid uint32; struct xcharval { xcharid xchv_which; opaque xchv_data<>; }; typedef xcharspec xcharval<>; typedef uint32 xcharsubset<>; <CODE ENDS>
The following XDR types are used by operations that deal with transport characteristics:
An xcharid specifies a particular transport characteristic. In order to allow easier XDR extension of the set of characteristics by concatenating XDR files, specific characteristics are defined as const values rather than as elements in an enum.
An xcharval specifies a value of a particular transport characteristic with the particular characteristic identified by xchv_which, while the associated value of that characteristic is contained within xchv_data.
While xchv_data is defined as opaque within the XDR, the contents are interpreted using the XDR typedef associated with the characteristic specified by xchv_which. The receiver of a message containing an xcharval MUST report an XDR error if the length of xchv_data is such that it extends beyond the bounds of the message transferred.
In cases in which the xcharid specified by xchv_which is understood by the receiver, the receiver also MUST report an XDR error if either of the following occur:
Note that no error is to be reported if xchv_which is unknown to the receiver. In that case, that xcharval is not processed and processing continues using the next xcharval, if any.
An xcharspec specifies a set of transport characteristics. No particular ordering of the xcharvals within it is imposed.
An xcharsubset identifies a subset of the characteristics in a previously specified xcharspec. Each bit in the mask denotes a particular element in a previously specified xcharspec. If a particular xcharval is at position N in the array, then bit number N mod 32 in word N div 32 specifies whether that particular xcharval is included in the defined subset. Words beyond the last one specified are treated as containing zero.
xcharsubsets are useful in a number of contexts:
Transport characteristics are divided into a number of groups
There are a number of operations defined in Section 4 which are used to communicate and manage transport characteristics.
Prime among these is ROPT_INITXCH (defined in Section 4.1 which serves as a means by which an endpoints transport characteristics may be presented to its peer, typically upon establishing a connection.
In addition, there are a set of related operations concerned with requesting, effecting and reporting changes in transport characteristics:
Unlike many other operation types, the above are not used to effect transfer of RPC requests but are internal one-way information transfers. However, a ROPT_REQXCH and the corresponding ROPT_RESPXCH do constitute an RPC-like remote call. The other operations are not part of a remote call transaction, although one or more asynchronous ROPT_UPDXCH operations may result from a ROPT_REQXCH.
Although the set of transport characteristics is subject to later extension, an initial set of transport characteristics is defined below in Table 1.
In that table, the columns contain the following information:
characteristic | code | XDR type | default | section |
---|---|---|---|---|
Receive Buffer Size | 1 | uint32 | 4096 | 3.1 |
Requester Remote Invalidation | 2 | bool | false | 3.2 |
Backward Request Support | 3 | enum bkreqsup | BKREQSUP_INLINE | 3.3 |
Note that there is no explicit indication regarding whether a particular characteristic can change or whether a change in the value may be requested (see Section 4.2). Such matters are not addressed by the protocol definition. A partner implementation can always request a change but peers MAY reject a request to change a characteristic for any reason. Implementations are always free to reject such requests if they cannot or do not wish to effect the requested change.
Either of the following will result in effective rejection requests to change specific characteristics:
With regard to unrequested changes in transport characteristics, it is the responsibility of the implementation making the change to do so in a fashion that which does not interfere with the other partner's continued correct operation (see Section 3.1).
The Receive Buffer Size specifies the minimum size, in octets, of pre-posted receive buffers. It is the responsibility of the participant sending this value to ensure that its pre-posted receives are at least the size specified, allowing the participant receiving this value to send messages that are of this size.
<CODE BEGINS> const uint32 XCHAR_RBSIZ = 1; typedef uint32 xchrbsiz; <CODE ENDS>
The sender may use his knowledge of the receiver's buffer size to determine when the message to be sent will fit in the preposted receive buffers that the receiver has set up. In particular,
Because there may be pre-posted receives with buffer sizes that reflect earlier values of the buffer size characteristic, changing this characteristics poses special difficulties:
The Requester Remote Invalidation characteristic indicates that the current endpoint, when in the role of a requester, is prepared for the responder to use RDMA Send With Invalidate when replying to an RPC-over-RDMA request containing non-empty chuck lists.
As RPC-over-RDMA is currently used, memory registrations exposed to peers are not established by the server and explicit RDMA operations are not done to satisfy backward direction requests. This makes it unlikely that servers will present non-default values of the XCHAR_REQREMINV characteristic or that clients will take note of that value when presented by servers.
<CODE BEGINS> const uint32 XCHAR_REQREMINV = 2; typedef bool xchrreqrem; <CODE ENDS>
When the Requester Remote Invalidate characteristic is set to false, a responder MUST use Send to convey RPC reply messages to the requester. When the Requester Remote Invalidate characteristic is set to true, a responder MAY use Send With Invalidate instead of Send to convey RPC replies to the requester.
The value of the Requester Remote Invalidate characteristic is not likely to change from the value reported by ROPT_INITXCH (see Section 4.2).
The value of this characteristic is used to indicate a client implementation's readiness to accept and process messages that are part of backward-direction RPC requests.
<CODE BEGINS> enum bkreqsup { BKREQSUP_NONE = 0, BKREQSUP_INLINE = 1, BKREQSUP_GENL = 2 }; const uint32 XCHAR_BRS = 3; typedef bkreqsup xchrbrs; <CODE ENDS>
Multiple levels of support are distinguished:
The support level of servers can be inferred from the backward- direction requests that they issue, assuming that issuing a request implicitly indicates support for receiving the corresponding reply. On this basis, support for receiving inline replies can be assumed when requests without read chunks, write chunks, or Reply chunks are issued, while requests with any of these elements allow the client to assume that general support for backward-direction replies is present on the server.
The proposed new operation are set forth in Table 2 below. In that table, the columns contain the following information:
operation | code | XDR type | msg | section |
---|---|---|---|---|
Specify Initial Characteristics | 1 | optinfo_initxch | No | 4.1 |
Request Characteristic Modification | 2 | optinfo_reqxch | No | 4.2 |
Respond to Modification Request | 3 | optinfo_respxch | No | 4.3 |
Report Updated Characteristics | 4 | optinfo_updxch | No | 4.4 |
Support for all of the operations above is OPTIONAL. RPC-over-RDMA Version Two implementations that receive an operation that is not supported MUST respond with RDMA_ERROR message with an error code of RDMA_ERR_INVAL_OPTION as specified in [rpcrdmav2]
The only operation support requirements are as follows:
The ROPT_INITXCH message type allows an RPC-over-RDMA participant, whether client or server, to indicate to its partner relevant transport characteristics that the partner might need to be aware of.
<CODE BEGINS> const uint32 ROPT_INITXCH = 1; struct optinfo_initxch { xcharspec optixch_start; xcharsubset optixch_nochg; }; <CODE ENDS>
The message definition for this operation is as follows:
All relevant transport characteristics that the sender is aware of should be included in optixch_start. Since support of this request is OPTIONAL, and since each of the characteristics is OPTIONAL as well, the sender cannot assume that the receiver will necessarily take note of these characteristics and so the sender should be prepared for cases in which the partner continues to assume that the default value for a particular characteristic is still in effect.
The subset of transport characteristic specified by optixch_nochg is not expected to change during the lifetime of the connection.
Generally, a participant will send a ROPT_INITXCH message as the first message after a connection is established. Given that fact, the sender should make sure that the message can be received by partners who use the default Receive Buffer Size. The connection's initial receive buffer size is typically 1KB, but it depends on the initial connection state of the RPC-over-RDMA version in use. See [rpcrdmav2] for details.
Those receiving an ROPT_INITXCH may encounter characteristics that they do not support or are unaware of. In such cases, these characteristics are simply ignored without any error response being generated.
The ROPT_REQXCH message type allows an RPC-over-RDMA participant, whether client or server, to request of its partner that relevant transport characteristics be changed.
The partner need not change the characteristics as requested by the sender but if it does support the message type, it will generate a ROPT_RESPXCH message, indicating the disposition of the request.
<CODE BEGINS> const uint32 ROPT_REQXCH = 2; struct optinfo_reqxch { xcharspec optreqxch_want; }; <CODE ENDS>
The message definition for this operation is as follows:
The xcharspec optreqxch_want is a set of transport characteristics together with the desired values requested by the sender.
The ROPT_RESPXCH message type allows an RPC-over-RDMA participant to respond to a request to change characteristics by its partner, indicating how the request was dealt with.
<CODE BEGINS> const uint32 ROPT_RESPXCH = 3; struct optinfo_respxch { xcharsubset optrespxch_done; xcharsubset optrespxch_rej; xcharsubset optrespxch_pend; }; <CODE ENDS>
The message definition for this operation is as follows:
The rdma_xid field of this message must match that used in the ROPT_REQXCH message to which this message is responding.
The optrespxch_done field indicates which of the requested transport characteristic changes have been immediately effected. For each such characteristic, the receiver is entitled to conclude that the requested change has been made and that future transmissions may be made based on the new value.
The optrespxch_rej field indicates which of the requested transport characteristic changes have been rejected by the sender. This may be because of any of the following reasons:
The optrespxch_pend field indicates which of the requested transport characteristic modifications remain pending, since they were neither rejected nor effected immediately. The receiver can expect the modification to be effected by a later ROPT_UPDXCH message, although there is no way to determine when this will happen. For each characteristic bit set in this field, one or more ROPT_UPXCH can be expected, the last of which will have optupdxch_pendclr flag set.
The subsets of characteristics specified by optrespxch_done, optrespxch_rej, optrespxch_pend should not overlap and, when ored together, should cover the entire set of characteristics specified by optreqxch_want in the corresponding request.
The ROPT_UPDXCH message type allows an RPC-over-RDMA participant to notify the other participant that a change to the transport characteristics has occurred.
This may be because:
One should pay particular attention to the fact that there is no there no way to tie a message reporting a change to the specific request which asked for the change. In particular, the rdma_xid field in this message is independent of that for any earlier ROPT_REQXCH message.
<CODE BEGINS> const uint32 ROPT_UPDXCH = 4; struct optinfo_updxch { xcharval optupdxch_now; bool optupdxch_pendclr; }; <CODE ENDS>
The message definition for this operation is as follows:
optupdxch_now defines the new characteristic value to be used.
optupdxch_pendclr, if true, indicates that a previous request to update the characteristic specified by optupdxch_now.xchv_which is no longer to be considered pending. This may be set true even if the characteristic value is not changed from the previous value.
Some instances of ROPT_UPDXCH are the result of a previous a previous ROPT_REQXCH while others are unsolicited. This distinction relates to the setting of optupdxch_pendclr as follows:
This section contains an XDR [RFC4506] description of the proposed extension.
<CODE BEGINS> #!/bin/sh grep '^ *///' | sed 's?^ /// ??' | sed 's?^ *///$??' <CODE ENDS>
This description is provided in a way that makes it simple to extract into ready-to-use form. The reader can apply the following shell script to this document to produce a machine-readable XDR description of extension which can be combined with XDR for the base protocol to produce an XDR that combines the base protocol with the optional extensions.
<CODE BEGINS> sh extract.sh < ext.txt > charext.x <CODE ENDS>
That is, if the above script is stored in a file called "extract.sh" and this document is in a file called "ext.txt" then the reader can do the following to extract an XDR description file for this extension:
<CODE BEGINS> /// /* /// * Copyright (c) 2010, 2016 IETF Trust and the persons /// * identified as authors of the code. All rights reserved. /// * /// * The author of the code is: D. Noveck. /// * /// * Redistribution and use in source and binary forms, with /// * or without modification, are permitted provided that the /// * following conditions are met: /// * /// * - Redistributions of source code must retain the above /// * copyright notice, this list of conditions and the /// * following disclaimer. /// * /// * - Redistributions in binary form must reproduce the above /// * copyright notice, this list of conditions and the /// * following disclaimer in the documentation and/or other /// * materials provided with the distribution. /// * /// * - Neither the name of Internet Society, IETF or IETF /// * Trust, nor the names of specific contributors, may be /// * used to endorse or promote products derived from this /// * software without specific prior written permission. /// * /// * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS /// * AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED /// * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE /// * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS /// * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO /// * EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE /// * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, /// * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT /// * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR /// * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS /// * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF /// * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, /// * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING /// * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF /// * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. /// */ <CODE ENDS>
Code components extracted from this document must include the following license text. When the extracted XDR code is combined with other complementary XDR code which itself has an identical license, only a single copy of the license text need be preserved.
<CODE BEGINS> /// ////* /// * Basic transport characteristic types /// */ ///typedef xcharid uint32; /// ///struct xcharval { /// xcharid xchv_which; /// opaque xchv_data<>; ///}; /// ///typedef xcharspec xcharval<>; /// ///typedef xcharsubset uint32<>; /// ////* /// * Transport characteristic codes /// */ ///const uint32 XCHAR_RBSIZ = 1; ///const uint32 XCHAR_REQREMINV = 2; ///const uint32 XCHAR_BRS = 3; /// ////* /// * Other transport characteristic types /// */ ///enum bkreqsup { /// BKREQSUP_NONE = 0, /// BKREQSUP_INLINE = 1, /// BKREQSUP_GENL = 2 ///}; /// ////* /// * Transport characteristic typedefs /// */ ///typedef uint32 xchrbsiz; ///typedef bool xchrreqrem; ///typedef bkreqsup xchrbrs; /// ////* /// * Optional operation codes /// */ ///const uint32 ROPT_INITXCH = 1; ///const uint32 ROPT_REQXCH = 2; ///const uint32 ROPT_RESPXCH = 3; ///const uint32 ROPT_UPDXCH = 4; /// ////* /// * Optional operation message structures /// */ ///struct optinfo_initxch { /// xcharspec optixch_start; /// xcharsubset optixch_nochg; ///}; /// ///struct optinfo_reqxch { /// xcharspec optreqxch_want; ///}; /// ///struct optinfo_respxch { /// xcharsubset optrespxch_done; /// xcharsubset optrespxch_rej; /// xcharsubset optrespxch_pend; ///}; /// ///struct optinfo_updxch { /// xcharval optupdxch_now; /// bool optupdxch_pendclr; ///}; <CODE ENDS>
The set of transport characteristics is designed to be extensible. As a result, once new characteristics are defined in standards track documents, the operations defined in this document may reference these new transport characteristics, as well as the ones described in this document.
A standards track document defining a new transport characteristic should include the following information paralleling that provided in this document for the transport characteristics defined herein.
The definition of transport characteristic structures is such as to make it easy to assign unique values. There is no requirement that a continuous set of values be used and implementations should not rely on all such values being small integers. A unique value should be selected when the defining document is first published as an internet draft. When the document becomes a standards track document working group should insure that:
Documents defining new characteristics fall into a number of categories.
When additional transport characteristics are proposed, the review of the associated standards track document should deal with possible security issues raised by those new transport characteristics.
Given the design of the transport characteristics data structure, it possible to use the operations to implement experimental, possibly unpublished, transport characteristics.
xcharids in the range from 4,294,967,040 to 4,294,967,295 are reserved for experimental use and these values should not be assigned to new characteristics in standards track documents.
When values in this range are used there is no guarantee if successful interoperation among independent implementations.
Like other fields that appear in each RPC-over-RDMA header, characteristic information is sent in the clear on the fabric with no integrity protection, making it vulnerable to man-in-the-middle attacks.
For example, if a man-in-the-middle were to change the value of the Receive buffer size or the Requester Remote Invalidation boolean, it could reduce connection performance or trigger loss of connection. Repeated connection loss can impact performance or even prevent a new connection from being established. Recourse is to deploy on a private network or use link-layer encryption.
This document does not require any actions by IANA.
[bidir] | Lever, C., "Size-Limited Bi-directional Remote Procedure Call On Remote Direct Memory Access Transports", April 2016. Work in progress. |
[RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |
[RFC4506] | Eisler, M., "XDR: External Data Representation Standard", STD 67, RFC 4506, DOI 10.17487/RFC4506, May 2006. |
[rfc5666bis] | Lever, C., Simpson, W. and T. Talpey, "Remote Direct Memory Access Transport for Remote Procedure Call", May 2016. Work in progress. |
[rpcrdmav2] | Lever, C. and D. Noveck, "RPC-over-RDMA Version Two", June 2016. Work in progress. |
[RFC5662] | Shepler, S., Eisler, M. and D. Noveck, "Network File System (NFS) Version 4 Minor Version 1 External Data Representation Standard (XDR) Description", RFC 5662, DOI 10.17487/RFC5662, January 2010. |
[RFC5666] | Talpey, T. and B. Callaghan, "Remote Direct Memory Access Transport for Remote Procedure Call", RFC 5666, DOI 10.17487/RFC5666, January 2010. |
The author gratefully acknowledges the work of Brent Callaghan and Tom Talpey producing the original RPC-over-RDMA Version One specification [RFC5666] and also Tom's work in helping to clarify that specification.
The author also wishes to thank Chuck Lever for his work resurrecting NFS support for RDMA in [rfc5666bis] and for his helpful review of and suggestions for this document.
The extract.sh shell script and formatting conventions were first described by the authors of the NFSv4.1 XDR specification [RFC5662].