idnits 2.17.1 draft-ietf-nfsv4-rfc5666bis-08.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (November 21, 2016) is 2712 days in the past. Is this intentional? Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) == Outdated reference: A later version (-08) exists of draft-ietf-nfsv4-rpcrdma-bidirection-05 -- Obsolete informational reference (is this intentional?): RFC 793 (Obsoleted by RFC 9293) -- Obsolete informational reference (is this intentional?): RFC 5661 (Obsoleted by RFC 8881) -- Obsolete informational reference (is this intentional?): RFC 5666 (Obsoleted by RFC 8166) -- Obsolete informational reference (is this intentional?): RFC 5667 (Obsoleted by RFC 8267) Summary: 0 errors (**), 0 flaws (~~), 2 warnings (==), 5 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network File System Version 4 C. Lever, Ed. 3 Internet-Draft Oracle 4 Obsoletes: 5666 (if approved) W. Simpson 5 Intended status: Standards Track DayDreamer 6 Expires: May 25, 2017 T. Talpey 7 Microsoft 8 November 21, 2016 10 Remote Direct Memory Access Transport for Remote Procedure Call, Version 11 One 12 draft-ietf-nfsv4-rfc5666bis-08 14 Abstract 16 This document specifies a protocol for conveying Remote Procedure 17 Call (RPC) messages on physical transports capable of Remote Direct 18 Memory Access (RDMA). It requires no revision to application RPC 19 protocols or the RPC protocol itself. This document obsoletes RFC 20 5666. 22 Status of This Memo 24 This Internet-Draft is submitted in full conformance with the 25 provisions of BCP 78 and BCP 79. 27 Internet-Drafts are working documents of the Internet Engineering 28 Task Force (IETF). Note that other groups may also distribute 29 working documents as Internet-Drafts. The list of current Internet- 30 Drafts is at http://datatracker.ietf.org/drafts/current/. 32 Internet-Drafts are draft documents valid for a maximum of six months 33 and may be updated, replaced, or obsoleted by other documents at any 34 time. It is inappropriate to use Internet-Drafts as reference 35 material or to cite them other than as "work in progress." 37 This Internet-Draft will expire on May 25, 2017. 39 Copyright Notice 41 Copyright (c) 2016 IETF Trust and the persons identified as the 42 document authors. All rights reserved. 44 This document is subject to BCP 78 and the IETF Trust's Legal 45 Provisions Relating to IETF Documents 46 (http://trustee.ietf.org/license-info) in effect on the date of 47 publication of this document. Please review these documents 48 carefully, as they describe your rights and restrictions with respect 49 to this document. Code Components extracted from this document must 50 include Simplified BSD License text as described in Section 4.e of 51 the Trust Legal Provisions and are provided without warranty as 52 described in the Simplified BSD License. 54 Table of Contents 56 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 57 1.1. Requirements Language . . . . . . . . . . . . . . . . . . 3 58 1.2. Remote Procedure Calls On RDMA Transports . . . . . . . . 3 59 2. Changes Since RFC 5666 . . . . . . . . . . . . . . . . . . . 4 60 2.1. Changes To The Specification . . . . . . . . . . . . . . 4 61 2.2. Changes To The Protocol . . . . . . . . . . . . . . . . . 4 62 3. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 5 63 3.1. Remote Procedure Calls . . . . . . . . . . . . . . . . . 5 64 3.2. Remote Direct Memory Access . . . . . . . . . . . . . . . 8 65 4. RPC-Over-RDMA Protocol Framework . . . . . . . . . . . . . . 10 66 4.1. Transfer Models . . . . . . . . . . . . . . . . . . . . . 10 67 4.2. Message Framing . . . . . . . . . . . . . . . . . . . . . 11 68 4.3. Managing Receiver Resources . . . . . . . . . . . . . . . 12 69 4.4. XDR Encoding With Chunks . . . . . . . . . . . . . . . . 14 70 4.5. Message Size . . . . . . . . . . . . . . . . . . . . . . 20 71 5. RPC-Over-RDMA In Operation . . . . . . . . . . . . . . . . . 23 72 5.1. XDR Protocol Definition . . . . . . . . . . . . . . . . . 24 73 5.2. Fixed Header Fields . . . . . . . . . . . . . . . . . . . 28 74 5.3. Chunk Lists . . . . . . . . . . . . . . . . . . . . . . . 30 75 5.4. Memory Registration . . . . . . . . . . . . . . . . . . . 32 76 5.5. Error Handling . . . . . . . . . . . . . . . . . . . . . 34 77 5.6. Protocol Elements No Longer Supported . . . . . . . . . . 36 78 5.7. XDR Examples . . . . . . . . . . . . . . . . . . . . . . 37 79 6. RPC Bind Parameters . . . . . . . . . . . . . . . . . . . . . 39 80 7. Upper Layer Binding Specifications . . . . . . . . . . . . . 40 81 7.1. DDP-Eligibility . . . . . . . . . . . . . . . . . . . . . 41 82 7.2. Maximum Reply Size . . . . . . . . . . . . . . . . . . . 42 83 7.3. Additional Considerations . . . . . . . . . . . . . . . . 42 84 7.4. Upper Layer Protocol Extensions . . . . . . . . . . . . . 43 85 8. Protocol Extensibility . . . . . . . . . . . . . . . . . . . 43 86 8.1. Conventional Extensions . . . . . . . . . . . . . . . . . 44 87 9. Security Considerations . . . . . . . . . . . . . . . . . . . 44 88 9.1. Memory Protection . . . . . . . . . . . . . . . . . . . . 44 89 9.2. RPC Message Security . . . . . . . . . . . . . . . . . . 45 90 10. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 48 91 11. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 49 92 12. References . . . . . . . . . . . . . . . . . . . . . . . . . 49 93 12.1. Normative References . . . . . . . . . . . . . . . . . . 49 94 12.2. Informative References . . . . . . . . . . . . . . . . . 50 95 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 52 97 1. Introduction 99 This document obsoletes RFC 5666. However, the protocol specified by 100 this document is based on existing interoperating implementations of 101 the RPC-over-RDMA Version One protocol. 103 The new specification clarifies text that is subject to multiple 104 interpretations, and removes support for unimplemented RPC-over-RDMA 105 Version One protocol elements. It clarifies the role of Upper Layer 106 Bindings and describes what they are to contain. 108 In addition, this document describes current practice using 109 RPCSEC_GSS [I-D.ietf-nfsv4-rpcsec-gssv3] on RDMA transports. 111 1.1. Requirements Language 113 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 114 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 115 document are to be interpreted as described in [RFC2119]. 117 1.2. Remote Procedure Calls On RDMA Transports 119 Remote Direct Memory Access (RDMA) [RFC5040] [RFC5041] [IB] is a 120 technique for moving data efficiently between end nodes. By 121 directing data into destination buffers as it is sent on a network, 122 and placing it via direct memory access by hardware, the benefits of 123 faster transfers and reduced host overhead are obtained. 125 Open Network Computing Remote Procedure Call (ONC RPC, or simply, 126 RPC) [RFC5531] is a remote procedure call protocol that runs over a 127 variety of transports. Most RPC implementations today use UDP 128 [RFC0768] or TCP [RFC0793]. On UDP, RPC messages are encapsulated 129 inside datagrams, while on a TCP byte stream, RPC messages are 130 delineated by a record marking protocol. An RDMA transport also 131 conveys RPC messages in a specific fashion that must be fully 132 described if RPC implementations are to interoperate. 134 RDMA transports present semantics different from either UDP or TCP. 135 They retain message delineations like UDP, but provide reliable and 136 sequenced data transfer like TCP. They also provide an offloaded 137 bulk transfer service not provided by UDP or TCP. RDMA transports 138 are therefore appropriately viewed as a new transport type by RPC. 140 In this context, the Network File System (NFS) protocols as described 141 in [RFC1094], [RFC1813], [RFC7530], [RFC5661], and future NFSv4 minor 142 verions are all obvious beneficiaries of RDMA transports. A complete 143 problem statement is presented in [RFC5532]. Many other RPC-based 144 protocols can also benefit. 146 Although the RDMA transport described herein can provide relatively 147 transparent support for any RPC application, this document also 148 describes mechanisms that can optimize data transfer even further, 149 given more active participation by RPC applications. 151 2. Changes Since RFC 5666 153 2.1. Changes To The Specification 155 The following alterations have been made to the RPC-over-RDMA Version 156 One specification. The section numbers below refer to [RFC5666]. 158 o Section 2 has been expanded to introduce and explain key RPC, XDR, 159 and RDMA terminology. These terms are now used consistently 160 throughout the specification. 162 o Section 3 has been re-organized and split into sub-sections to 163 help readers locate specific requirements and definitions. 165 o Sections 4 and 5 have been combined to improve the organization of 166 this information. 168 o The specification of the optional Connection Configuration 169 Protocol has been removed from the specification. 171 o A section consolidating requirements for Upper Layer Bindings has 172 been added. 174 o An XDR extraction mechanism is provided, along with full 175 copyright, matching the approach used in [RFC5662]. 177 o The "Security Considerations" section has been expanded to include 178 a discussion of how RPC-over-RDMA security depends on features of 179 the underlying RDMA transport. 181 o A subsection describing the use of RPCSEC_GSS with RPC-over-RDMA 182 Version One has been added. 184 2.2. Changes To The Protocol 186 Although the protocol described herein interoperates with existing 187 implementations of [RFC5666], the following changes have been made 188 relative to the protocol described in that document: 190 o Support for the Read-Read transfer model has been removed. Read- 191 Read is a slower transfer model than Read-Write. As a result, 192 implementers have chosen not to support it. Removal simplifies 193 explanatory text, and support for the RDMA_DONE procedure is no 194 longer necessary. 196 o The specification of RDMA_MSGP in [RFC5666] is not adequate, 197 although some incomplete implementations exist. Even if an 198 adequate specification were provided and an implementation was 199 produced, benefit for protocols such as NFSv4.0 [RFC7530] is 200 doubtful. Therefore the RDMA_MSGP message type is no longer 201 supported. 203 o Technical issues with regard to handling RPC-over-RDMA header 204 errors have been corrected. 206 o Specific requirements related to implicit XDR round-up and complex 207 XDR data types have been added. 209 o Explicit guidance is provided related to sizing Write chunks, 210 managing multiple chunks in the Write list, and handling unused 211 Write chunks. 213 o Clear guidance about Send and Receive buffer sizes has been 214 introduced. This enables better decisions about when a Reply 215 chunk must be provided. 217 The protocol version number has not been changed because the protocol 218 specified in this document fully interoperates with implementations 219 of the RPC-over-RDMA Version One protocol specified in [RFC5666]. 221 3. Terminology 223 3.1. Remote Procedure Calls 225 This section highlights key elements of the Remote Procedure Call 226 [RFC5531] and External Data Representation [RFC4506] protocols, upon 227 which RPC-over-RDMA Version One is constructed. Strong grounding 228 with these protocols is recommended before reading this document. 230 3.1.1. Upper Layer Protocols 232 Remote Procedure Calls are an abstraction used to implement the 233 operations of an "Upper Layer Protocol," or ULP. The term Upper 234 Layer Protocol refers to an RPC Program and Version tuple, which is a 235 versioned set of procedure calls that comprise a single well-defined 236 API. One example of an Upper Layer Protocol is the Network File 237 System Version 4.0 [RFC7530]. 239 3.1.2. Requesters And Responders 241 Like a local procedure call, every Remote Procedure Call (RPC) has a 242 set of "arguments" and a set of "results". A calling context is not 243 allowed to proceed until the procedure's results are available to it. 244 Unlike a local procedure call, the called procedure is executed 245 remotely rather than in the local application's context. 247 The RPC protocol as described in [RFC5531] is fundamentally a 248 message-passing protocol between one server and one or more clients. 249 ONC RPC transactions are made up of two types of messages: 251 CALL Message 252 A CALL message, or "Call", requests that work be done. A Call is 253 designated by the value zero (0) in the message's msg_type field. 254 An arbitrary unique value is placed in the message's xid field in 255 order to match this CALL message to a corresponding REPLY message. 257 REPLY Message 258 A REPLY message, or "Reply", reports the results of work requested 259 by a Call. A Reply is designated by the value one (1) in the 260 message's msg_type field. The value contained in the message's 261 xid field is copied from the Call whose results are being 262 reported. 264 The RPC client endpoint acts as a "requester". It serializes an RPC 265 Call's arguments and conveys them to a server endpoint via an RPC 266 Call message. This message contains an RPC protocol header, a header 267 describing the requested upper layer operation, and all arguments. 269 The RPC server endpoint acts as a "responder". It deserializes Call 270 arguments, and processes the requested operation. It then serializes 271 the operation's results into another byte stream. This byte stream 272 is conveyed back to the requester via an RPC Reply message. This 273 message contains an RPC protocol header, a header describing the 274 upper layer reply, and all results. 276 The requester deserializes the results and allows the original caller 277 to proceed. At this point the RPC transaction designated by the xid 278 in the Call message is complete, and the xid is retired. 280 In summary, CALL messages are sent by requesters to responders to 281 initiate an RPC transaction. REPLY messages are sent by responders 282 to requesters to complete the processing on an RPC transaction. 284 3.1.3. RPC Transports 286 The role of an "RPC transport" is to mediate the exchange of RPC 287 messages between requesters and responders. An RPC transport bridges 288 the gap between the RPC message abstraction and the native operations 289 of a particular network transport. 291 RPC-over-RDMA is a connection-oriented RPC transport. When a 292 connection-oriented transport is used, clients initiate transport 293 connections, while servers wait passively for incoming connection 294 requests. 296 3.1.4. External Data Representation 298 One cannot assume that all requesters and responders internally 299 represent data objects the same way. RPC uses eXternal Data 300 Representation, or XDR, to translate data types and serialize 301 arguments and results [RFC4506]. 303 The XDR protocol encodes data independent of the endianness or size 304 of host-native data types, allowing unambiguous decoding of data on 305 the receiving end. RPC Programs are specified by writing an XDR 306 definition of their procedures, argument data types, and result data 307 types. 309 XDR assumes that the number of bits in a byte (octet) and their order 310 are the same on both endpoints and on the physical network. The 311 smallest indivisible unit of XDR encoding is a group of four octets 312 in little-endian order. XDR also flattens lists, arrays, and other 313 complex data types so they can be conveyed as a stream of bytes. 315 A serialized stream of bytes that is the result of XDR encoding is 316 referred to as an "XDR stream." A sending endpoint encodes native 317 data into an XDR stream and then transmits that stream to a receiver. 318 A receiving endpoint decodes incoming XDR byte streams into its 319 native data representation format. 321 3.1.4.1. XDR Opaque Data 323 Sometimes a data item must be transferred as-is, without encoding or 324 decoding. The contents of such a data item are referred to as 325 "opaque data." XDR encoding places the content of opaque data items 326 directly into an XDR stream without altering it in any way. Upper 327 Layer Protocols or applications perform any needed data translation 328 in this case. Examples of opaque data items include the content of 329 files, or generic byte strings. 331 3.1.4.2. XDR Round-up 333 The number of octets in a variable-size opaque data item precedes 334 that item in an XDR stream. If the size of an encoded data item is 335 not a multiple of four octets, octets containing zero are added to 336 the end of the item as it is encoded so that the next encoded data 337 item starts on a four-octet boundary. The encoded size of the item 338 is not changed by the addition of the extra octets, and the zero 339 bytes are not exposed to the Upper Layer. 341 This technique is referred to as "XDR round-up," and the extra octets 342 are referred to as "XDR padding". 344 3.2. Remote Direct Memory Access 346 RPC requesters and responders can be made more efficient if large RPC 347 messages are transferred by a third party such as intelligent network 348 interface hardware (data movement offload), and placed in the 349 receiver's memory so that no additional adjustment of data alignment 350 has to be made (direct data placement). Remote Direct Memory Access 351 transports enable both optimizations. 353 3.2.1. Direct Data Placement 355 Typically, RPC implementations copy the contents of RPC messages into 356 a buffer before being sent. An efficient RPC implementation sends 357 bulk data without copying it into a separate send buffer first. 359 However, socket-based RPC implementations are often unable to receive 360 data directly into its final place in memory. Receivers often need 361 to copy incoming data to finish an RPC operation; sometimes, only to 362 adjust data alignment. 364 In this document, "RDMA" refers to the physical mechanism an RDMA 365 transport utilizes when moving data. Although this may not be 366 efficient, before an RDMA transfer a sender may copy data into an 367 intermediate buffer before an RDMA transfer. After an RDMA transfer, 368 a receiver may copy that data again to its final destination. 370 This document uses the term "direct data placement" (or DDP) to refer 371 specifically to an optimized data transfer where it is unnecessary 372 for a receiving host's CPU to copy transferred data to another 373 location after it has been received. Not all RDMA-based data 374 transfer qualifies as Direct Data Placement, and DDP can be achieved 375 using non-RDMA mechanisms. 377 3.2.2. RDMA Transport Requirements 379 The RPC-over-RDMA Version One protocol assumes the physical transport 380 provides the following abstract operations. A more complete 381 discussion of these operations is found in [RFC5040]. 383 Registered Memory 384 Registered memory is a segment of memory that is assigned a 385 steering tag that temporarily permits access by the RDMA provider 386 to perform data transfer operations. The RPC-over-RDMA Version 387 One protocol assumes that each segment of registered memory MUST 388 be identified with a steering tag of no more than 32 bits and 389 memory addresses of up to 64 bits in length. 391 RDMA Send 392 The RDMA provider supports an RDMA Send operation, with completion 393 signaled on the receiving peer after data has been placed in a 394 pre-posted memory segment. Sends complete at the receiver in the 395 order they were issued at the sender. The amount of data 396 transferred by an RDMA Send operation is limited by the size of 397 the remote pre-posted memory segment. 399 RDMA Receive 400 The RDMA provider supports an RDMA Receive operation to receive 401 data conveyed by incoming RDMA Send operations. To reduce the 402 amount of memory that must remain pinned awaiting incoming Sends, 403 the amount of pre-posted memory is limited. Flow-control to 404 prevent overrunning receiver resources is provided by the RDMA 405 consumer (in this case, the RPC-over-RDMA Version One protocol). 407 RDMA Write 408 The RDMA provider supports an RDMA Write operation to directly 409 place data in remote memory. The local host initiates an RDMA 410 Write, and completion is signaled there. No completion is 411 signaled on the remote. The local host provides a steering tag, 412 memory address, and length of the remote's memory segment. 414 RDMA Writes are not necessarily ordered with respect to one 415 another, but are ordered with respect to RDMA Sends. A subsequent 416 RDMA Send completion obtained at the write initiator guarantees 417 that prior RDMA Write data has been successfully placed in the 418 remote peer's memory. 420 RDMA Read 421 The RDMA provider supports an RDMA Read operation to directly 422 place peer source data in the read initiator's memory. The local 423 host initiates an RDMA Read, and completion is signaled there; no 424 completion is signaled on the remote. The local host provides 425 steering tags, memory addresses, and a length for the remote 426 source and local destination memory segments. 428 The remote peer receives no notification of RDMA Read completion. 429 The local host signals completion as part of a subsequent RDMA 430 Send message so that the remote peer can release steering tags and 431 subsequently free associated source memory segments. 433 The RPC-over-RDMA Version One protocol is designed to be carried over 434 RDMA transports that support the above abstract operations. This 435 protocol conveys to the RPC peer information sufficient for that RPC 436 peer to direct an RDMA layer to perform transfers containing RPC data 437 and to communicate their result(s). For example, it is readily 438 carried over RDMA transports such as Internet Wide Area RDMA Protocol 439 (iWARP) [RFC5040] [RFC5041]. 441 4. RPC-Over-RDMA Protocol Framework 443 4.1. Transfer Models 445 A "transfer model" designates which endpoint is responsible for 446 performing RDMA Read and Write operations. To enable these 447 operations, the peer endpoint first exposes segments of its memory to 448 the endpoint performing the RDMA Read and Write operations. 450 Read-Read 451 Requesters expose their memory to the responder, and the responder 452 exposes its memory to requesters. The responder employs RDMA Read 453 operations to pull RPC arguments or whole RPC calls from the 454 requester. Requesters employ RDMA Read operations to pull RPC 455 results or whole RPC relies from the responder. 457 Write-Write 458 Requesters expose their memory to the responder, and the responder 459 exposes its memory to requesters. Requesters employ RDMA Write 460 operations to push RPC arguments or whole RPC calls to the 461 responder. The responder employs RDMA Write operations to push 462 RPC results or whole RPC relies to the requester. 464 Read-Write 465 Requesters expose their memory to the responder, but the responder 466 does not expose its memory. The responder employs RDMA Read 467 operations to pull RPC arguments or whole RPC calls from the 468 requester. The responder employs RDMA Write operations to push 469 RPC results or whole RPC relies to the requester. 471 Write-Read 472 The responder exposes its memory to requesters, but requesters do 473 not expose their memory. Requesters employ RDMA Write operations 474 to push RPC arguments or whole RPC calls to the responder. 475 Requesters employ RDMA Read operations to pull RPC results or 476 whole RPC relies from the responder. 478 [RFC5666] specifies the use of both the Read-Read and the Read-Write 479 Transfer Model. All current RPC-over-RDMA Version One 480 implementations use only the Read-Write Transfer Model. Therefore 481 the use of the Read-Read Transfer Model within RPC-over-RDMA Version 482 One implementations is no longer supported. Transfer Models other 483 than the Read-Write model may be used in future versions of RPC-over- 484 RDMA. 486 4.2. Message Framing 488 On an RPC-over-RDMA transport, each RPC message is encapsulated by an 489 RPC-over-RDMA message. An RPC-over-RDMA message consists of two XDR 490 streams. 492 RPC Payload Stream 493 The "Payload stream" contains the encapsulated RPC message being 494 transferred by this RPC-over-RDMA message. This stream always 495 begins with the XID field of the encapsulated RPC message. 497 Transport Stream 498 The "Transport stream" contains a header that describes and 499 controls the transfer of the Payload stream in this RPC-over-RDMA 500 message. This header is analogous to the record marking used for 501 RPC over TCP but is more extensive, since RDMA transports support 502 several modes of data transfer. 504 In its simplest form, an RPC-over-RDMA message consists of a 505 Transport stream followed immediately by a Payload stream conveyed 506 together in a single RDMA Send. To transmit large RPC messages, a 507 combination of one RDMA Send operation and one or more RDMA Read or 508 Write operations is employed. 510 RPC-over-RDMA framing replaces all other RPC framing (such as TCP 511 record marking) when used atop an RPC-over-RDMA association, even 512 when the underlying RDMA protocol may itself be layered atop a 513 transport with a defined RPC framing (such as TCP). 515 It is however possible for RPC-over-RDMA to be dynamically enabled in 516 the course of negotiating the use of RDMA via an Upper Layer Protocol 517 exchange. Because RPC framing delimits an entire RPC request or 518 reply, the resulting shift in framing must occur between distinct RPC 519 messages, and in concert with the underlying transport. 521 4.3. Managing Receiver Resources 523 It is critical to provide RDMA Send flow control for an RDMA 524 connection. If any pre-posted receive buffer on the connection is 525 not large enough to accept an incoming RDMA Send, the RDMA Send 526 operation can fail. If a pre-posted receive buffer is not available 527 to accept an incoming RDMA Send, the RDMA Send operation can fail. 528 Repeated occurrences of such errors can be fatal to the connection. 529 This is different than conventional TCP/IP networking, in which 530 buffers are allocated dynamically as messages are received. 532 The longevity of an RDMA connection requires that sending endpoints 533 respect the resource limits of peer receivers. To ensure messages 534 can be sent and received reliably, there are two operational 535 parameters for each connection. 537 4.3.1. RPC-over-RDMA Credits 539 Flow control for RDMA Send operations directed to the responder is 540 implemented as a simple request/grant protocol in the RPC-over-RDMA 541 header associated with each RPC message. 543 An RPC-over-RDMA Version One credit is the capability to handle one 544 RPC-over-RDMA transaction. Each RPC-over-RDMA message sent from 545 requester to responder requests a number of credits from the 546 responder. Each RPC-over-RDMA message sent from responder to 547 requester informs the requester how many credits the responder has 548 granted. The requested and granted values are carried in each RPC- 549 over-RDMA message's rdma_credit field (see Section 5.2.3). 551 Practically speaking, the critical value is the granted value. A 552 requester MUST NOT send unacknowledged requests in excess of the 553 responder's granted credit limit. If the granted value is exceeded, 554 the RDMA layer may signal an error, possibly terminating the 555 connection. The granted value MUST NOT be zero, since such a value 556 would result in deadlock. 558 RPC calls complete in any order, but the current granted credit limit 559 at the responder is known to the requester from RDMA Send ordering 560 properties. The number of allowed new requests the requester may 561 send is then the lower of the current requested and granted credit 562 values, minus the number of requests in flight. Advertised credit 563 values are not altered when individual RPCs are started or completed. 565 The requested and granted credit values MAY be adjusted to match the 566 needs or policies in effect on either peer. For instance, a 567 responder may reduce the granted credit value to accommodate the 568 available resources in a Shared Receive Queue. The responder MUST 569 ensure that an increase in receive resources is effected before the 570 next reply message is sent. 572 A requester MUST maintain enough receive resources to accommodate 573 expected replies. Responders have to be prepared for there to be no 574 receive resources available on requesters with no pending RPC 575 transactions. 577 Certain RDMA implementations may impose additional flow control 578 restrictions, such as limits on RDMA Read operations in progress at 579 the responder. Accommodation of such restrictions is considered the 580 responsibility of each RPC-over-RDMA Version One implementation. 582 4.3.2. Inline Threshold 584 An "inline threshold" value is the largest message size (in octets) 585 that can be conveyed in one direction between peer implementations 586 using RDMA Send and Receive. The inline threshold value is the 587 minimum of how large a message the sender can post via an RDMA Send 588 operation, and how large a message the receiver can accept via an 589 RDMA Receive operation. Each connection has two inline threshold 590 values: one for messages flowing from requester-to-responder 591 (referred to as the "call inline threshold"), and one for messages 592 flowing from responder-to-requester (referred to as the "reply inline 593 threshold"). 595 Unlike credit limits, inline threshold values are not advertised to 596 peers via the RPC-over-RDMA Version One protocol, and there is no 597 provision for inline threshold values to change during the lifetime 598 of an RPC-over-RDMA Version One connection. 600 4.3.3. Initial Connection State 602 When a connection is first established, peers might not know how many 603 receive resources the other has, nor how large the other peer's 604 inline thresholds are. 606 As a basis for an initial exchange of RPC requests, each RPC-over- 607 RDMA Version One connection provides the ability to exchange at least 608 one RPC message at a time, whose Call and Reply messages are no more 609 1024 bytes in size. A responder MAY exceed this basic level of 610 configuration, but a requester MUST NOT assume more than one credit 611 is available, and MUST receive a valid reply from the responder 612 carrying the actual number of available credits, prior to sending its 613 next request. 615 Receiver implementations MUST support inline thresholds of 1024 616 bytes, but MAY support larger inline thresholds values. A mechanism 617 for discovering a peer's inline thresholds before a connection is 618 established may be used to optimize the use of RDMA Send and Receive 619 operations. In the absense of such a mechanism, senders and receives 620 MUST assume the inline thresholds are 1024 bytes. 622 4.4. XDR Encoding With Chunks 624 When a direct data placement capability is available, it can be 625 determined during XDR encoding that the transport can efficiently 626 place the contents of one or more XDR data items directly into the 627 receiver's memory, separately from the transfer of other parts of the 628 containing XDR stream. 630 4.4.1. Reducing An XDR Stream 632 RPC-over-RDMA Version One provides a mechanism for moving part of an 633 RPC message via a data transfer separate from an RDMA Send/Receive. 634 The sender removes one or more XDR data items from the Payload 635 stream. They are conveyed via one or more RDMA Read or Write 636 operations. As the receiver decodes an incoming message, it skips 637 over directly placed data items. 639 The piece of memory containing the portion of the data stream that is 640 split out and placed directly is referred to as a "chunk". In some 641 contexts, data in the RPC-over-RDMA header that describes such pieces 642 of memory is also referred to as a "chunk". 644 A Payload stream after chunks have been removed is referred to as a 645 "reduced" Payload stream. Likewise, a data item that has been 646 removed from a Payload stream to be transferred separately is 647 referred to as a "reduced" data item. 649 4.4.2. DDP-Eligibility 651 Only an XDR data item that might benefit from Direct Data Placement 652 may be reduced. The eligibility of particular XDR data items to be 653 reduced is independent of RPC-over-RDMA, and thus is not specified by 654 this document. 656 To maintain interoperability on an RPC-over-RDMA transport, a 657 determination must be made of which XDR data items in each Upper 658 Layer Protocol are allowed to use Direct Data Placement. Therefore 659 an additional specification is needed that describes how an Upper 660 Layer Protocol enables Direct Data Placement. The set of 661 requirements for an Upper Layer Protocol to use an RPC-over-RDMA 662 transport is known as an "Upper Layer Binding specification," or ULB. 664 An Upper Layer Binding specification states which specific individual 665 XDR data items in an Upper Layer Protocol MAY be transferred via 666 Direct Data Placement. This document will refer to XDR data items 667 that are permitted to be reduced as "DDP-eligible". All other XDR 668 data items MUST NOT be reduced. RPC-over-RDMA Version One uses RDMA 669 Read and Write operations to transfer DDP-eligible data that has been 670 reduced. 672 Detailed requirements for Upper Layer Bindings are discussed in full 673 in Section 7. 675 4.4.3. RDMA Segments 677 When encoding a Payload stream that contains a DDP-eligible data 678 item, a sender may choose to reduce that data item. When it chooses 679 to do so, the sender does not place the item into the Payload stream. 680 Instead, the sender records in the RPC-over-RDMA header the location 681 and size of the memory region containing that data item. 683 The requester provides location information for DDP-eligible data 684 items in both RPC Calls and Replies. The responder uses this 685 information to initiate RDMA Read and Write operations to retrieve or 686 update the specified region of the requester's memory. 688 An "RDMA segment", or a "plain segment", is an RPC-over-RDMA header 689 data object that contains the precise co-ordinates of a contiguous 690 memory region that is to be conveyed via one or more RDMA Read or 691 RDMA Write operations. 693 Handle 694 Steering tag (STag) or handle obtained when the segment's memory 695 is registered for RDMA. Also known as an R_key, this value is 696 generated by registering this memory with the RDMA provider. 698 Length 699 The length of the memory segment, in octets. 701 Offset 702 The offset or beginning memory address of the segment. 704 See [RFC5040] for further discussion of the meaning of these fields. 706 4.4.4. Chunks 708 In RPC-over-RDMA Version One, a "chunk" refers to a portion of the 709 Payload stream that is moved via RDMA Read or Write operations. 710 Chunk data is removed from the sender's Payload stream, transferred 711 by separate RDMA operations, and then re-inserted into the receiver's 712 Payload stream. 714 Each chunk consists of one or more RDMA segments. Each segment 715 represents a single contiguous piece of that chunk. A requester MAY 716 divide a chunk into segments using any boundaries that are 717 convenient. 719 Except in special cases, a chunk contains exactly one XDR data item. 720 This makes it straightforward to remove chunks from an XDR stream 721 without affecting XDR alignment. 723 Many RPC-over-RDMA messages have no associated chunks. In this case, 724 all three chunk lists are marked empty. 726 4.4.4.1. Counted Arrays 728 If a chunk contains a counted array data type, the count of array 729 elements MUST remain in the Payload stream, while the array elements 730 MUST be moved to the chunk. For example, when encoding an opaque 731 byte array as a chunk, the count of bytes stays in the Payload 732 stream, while the bytes in the array are removed from the Payload 733 stream and transferred within the chunk. 735 Any byte count left in the Payload stream MUST match the sum of the 736 lengths of the segments making up the chunk. If they do not agree, 737 an RPC protocol encoding error results. 739 Individual array elements appear in a chunk in their entirety. For 740 example, when encoding an array of arrays as a chunk, the count of 741 items in the enclosing array stays in the Payload stream, but each 742 enclosed array, including its item count, is transferred as part of 743 the chunk. 745 4.4.4.2. Optional-data 747 If a chunk contains an optional-data data type, the "is present" 748 field MUST remain in the Payload stream, while the data, if present, 749 MUST be moved to the chunk. 751 4.4.4.3. XDR Unions 753 A union data type should never be made DDP-eligible, but one or more 754 of its arms may be DDP-eligible. 756 4.4.5. Read Chunks 758 A "Read chunk" represents an XDR data item that is to be pulled from 759 the requester to the responder using RDMA Read operations. 761 A Read chunk is a list of one or more RDMA read segments. Each RDMA 762 read segment consists of a Position field followed by a plain 763 segment. See Section 5.1.2 for details. 765 Position 766 The byte offset in the unreduced Payload stream where the receiver 767 re-inserts the data item conveyed in a chunk. The Position value 768 MUST be computed from the beginning of the unreduced Payload 769 stream, which begins at Position zero. All RDMA read segments 770 belonging to the same Read chunk have the same value in their 771 Position field. 773 While constructing an RPC-over-RDMA Call message, a requester 774 registers memory segments that contain data to be transferred via 775 RDMA Read operations. It advertises the co-ordinates of these 776 segments in the RPC-over-RDMA header of the RPC Call. 778 After receiving an RPC Call sent via an RDMA Send operation, a 779 responder transfers the chunk data from the requester using RDMA Read 780 operations. The responder reconstructs the transferred chunk data by 781 concatenating the contents of each segment, in list order, into the 782 received Payload stream at the Position value recorded in the 783 segment. 785 Put another way, the responder inserts the first segment in a Read 786 chunk into the Payload stream at the byte offset indicated by its 787 Position field. Segments whose Position field value match this 788 offset are concatenated afterwards, until there are no more segments 789 at that Position value. The next XDR data item in the Payload stream 790 follows. 792 4.4.5.1. Read Chunk Round-up 794 XDR requires each encoded data item to start on four-byte alignment. 795 When an odd-length data item is encoded, its length is encoded 796 literally, while the data is padded so the next data item in the XDR 797 stream can start on a four-byte boundary. Receivers ignore the 798 content of the pad bytes. 800 After an XDR data item has been reduced, all data items remaining in 801 the Payload stream must continue to adhere to these padding 802 requirements. Thus when an XDR data item is moved from the Payload 803 stream into a Read chunk, the requester MUST remove XDR padding for 804 that data item from the Payload stream as well. 806 The length of a Read chunk is the sum of the lengths of the read 807 segments that comprise it. If this sum is not a multiple of four, 808 the requester MAY choose to send a Read chunk without any XDR 809 padding. If the requester provides no actual round-up in a Read 810 chunk, the responder MUST be prepared to provide appropriate round-up 811 in the reconstructed call XDR stream 813 The Position field in a read segment indicates where the containing 814 Read chunk starts in the Payload stream. The value in this field 815 MUST be a multiple of four. Moreover, all segments in the same Read 816 chunk share the same Position value, even if one or more of the 817 segments have a non-four-byte aligned length. 819 4.4.5.2. Decoding Read Chunks 821 While decoding a received Payload stream, whenever the XDR offset in 822 the Payload stream matches that of a Read chunk, the responder 823 initiates an RDMA Read to pull the chunk's data content into 824 registered local memory. 826 The responder acknowledges its completion of use of Read chunk source 827 buffers when it sends an RPC Reply to the requester. The requester 828 may then release Read chunks advertised in the request. 830 4.4.6. Write Chunks 832 A "Write chunk" represents an XDR data item that is to be pushed from 833 a responder to a requester using RDMA Write operations. 835 A Write chunk is an array of one or more plain RDMA segments. Write 836 chunks are provided by a requester long before the responder has 837 prepared the reply Payload stream. In most cases, the byte offset of 838 a particular XDR data item in the reply is not predictable at the 839 time a request is issued. Therefore RDMA segments in a Write chunk 840 do not have a Position field. 842 While constructing an RPC Call message, a requester also prepares 843 memory regions to catch DDP-eligible reply data items. A requester 844 does not know the actual length of the result data item to be 845 returned, thus it MUST register a Write chunk long enough to 846 accommodate the maximum possible size of the returned data item. 848 The responder fills the segments contiguously in array order until 849 the result data item has been completely written into the Write 850 chunk. The responder copies the consumed Write chunk segments into 851 the Reply's RPC-over-RDMA header. As it does so, the responder 852 updates the segment length fields to reflect the actual amount of 853 data that is being returned in each segment, and updates the Write 854 chunk's segment count to reflect how many segments were consumed. 855 Unconsumed segments are omitted in the returned Write chunk. 857 The responder then sends the RPC Reply via an RDMA Send operation. 858 After receiving the RPC Reply, the requester reconstructs the 859 transferred data by concatenating the contents of each segment, in 860 array order, into RPC Reply XDR stream. 862 4.4.6.1. Write Chunk Round-up 864 XDR requires each encoded data item to start on four-byte alignment. 865 When an odd-length data item is encoded, its length is encoded 866 literally, while the data is padded so the next data item in the XDR 867 stream can start on a four-byte boundary. Receivers ignore the 868 content of the pad bytes. 870 After a data item is reduced, data items remaining in the Payload 871 stream must continue to adhere to these padding requirements. Thus 872 when an XDR data item is moved from a reply Payload stream into a 873 Write chunk, the responder MUST remove XDR padding for that data item 874 from the reply Payload stream as well. 876 A requester SHOULD NOT provide extra length in a Write chunk to 877 accommodate XDR pad bytes. A responder MUST NOT write XDR pad bytes 878 for a Write chunk. 880 4.4.6.2. Unused Write Chunks 882 There are occasions when a requester provides a Write chunk but the 883 responder is not able to use it. 885 For example, an Upper Layer Protocol may define a union result where 886 some arms of the union contain a DDP-eligible data item while other 887 arms do not. The responder is REQUIRED to use requester-provided 888 Write chunks in this case, but if the responder returns a result that 889 uses an arm of the union that has no DDP-eligible data item, the 890 Write chunk remains unconsumed. 892 If there is a subsequent DDP-eligible data item, it MUST be placed in 893 that unconsumed Write chunk. The requester MUST provision each Write 894 chunk so it can be filled with the largest DDP-eligible data item 895 that can be placed in it. 897 However, if this is the last or only Write chunk available and it 898 remains unconsumed, The responder MUST set the Write chunk segment 899 count to zero, returning no segments in the Write chunk. 901 Unused write chunks, or unused bytes in write chunk segments, are not 902 returned as results. Their memory is returned to the Upper Layer as 903 part of RPC completion. However, the RPC layer MUST NOT assume that 904 the buffers have not been modified. 906 In other words, even if a responder indicates that a Write chunk is 907 not consumed (by setting all of the segment lengths in the chunk to 908 zero), the responder may have written some data into the segments 909 before deciding not to return that data item. For example, a problem 910 reading local storage might occur while an NFS server is filling 911 Write chunks. This would interrupt the stream of RDMA Write 912 operations that sends data back to the NFS client, but at that point 913 the NFS server needs to return an NFS error that reflects that the 914 Upper Layer NFS request has failed. 916 When there is a DDP-eligible result data item, and the requester 917 prefers the data item returned inline, the requester provides a Write 918 chunk for that item where either the segment count is zero, or the 919 length of each of the chunk's segments is zero. The responder MUST 920 return the corresponding data item inline. 922 4.5. Message Size 924 A receiver of RDMA Send operations is required by RDMA to have 925 previously posted one or more adequately sized buffers. Memory 926 savings are achieved on both requesters and responders by posting 927 small Receive buffers. However, not all RPC messages are small. 929 4.5.1. Short Messages 931 RPC messages are frequently smaller than typical inline thresholds. 932 For example, the NFS version 3 GETATTR operation is only 56 bytes: 20 933 bytes of RPC header, plus a 32-byte file handle argument and 4 bytes 934 for its length. The reply to this common request is about 100 bytes. 936 Since all RPC messages conveyed via RPC-over-RDMA require an RDMA 937 Send operation, the most efficient way to send an RPC message that is 938 smaller than the inline threshold is to append the Payload stream 939 directly to the Transport stream. An RPC-over-RDMA header with a 940 small RPC Call or Reply message immediately following is transferred 941 using a single RDMA Send operation. No RDMA Read or Write operations 942 are needed. 944 An RPC-over-RDMA transaction using Short Messages: 946 Requester Responder 947 | RDMA Send (RDMA_MSG) | 948 Call | ------------------------------> | 949 | | 950 | | Processing 951 | | 952 | RDMA Send (RDMA_MSG) | 953 | <------------------------------ | Reply 955 4.5.2. Chunked Messages 957 If DDP-eligible data items are present in a Payload stream, a sender 958 MAY reduce some or all of these items by removing them from the 959 Payload stream. The sender uses RDMA Read or Write operations to 960 transfer the reduced data items. The Transport stream with the 961 reduced Payload stream immediately following is then transferred 962 using a single RDMA Send operation 964 After receiving the Transport and Payload streams of a Chunked RPC- 965 over-RDMA Call message, the responder uses RDMA Read operations to 966 move reduced data items in Read chunks. Before sending the Transport 967 and Payload streams of a Chunked RPC-over-RDMA Reply message, the 968 responder uses RDMA Write operations to move reduced data items in 969 Write and Reply chunks. 971 An RPC-over-RDMA transaction with a Read chunk: 973 Requester Responder 974 | RDMA Send (RDMA_MSG) | 975 Call | ------------------------------> | 976 | RDMA Read | 977 | <------------------------------ | 978 | RDMA Response (arg data) | 979 | ------------------------------> | 980 | | 981 | | Processing 982 | | 983 | RDMA Send (RDMA_MSG) | 984 | <------------------------------ | Reply 986 An RPC-over-RDMA transaction with a Write chunk: 988 Requester Responder 989 | RDMA Send (RDMA_MSG) | 990 Call | ------------------------------> | 991 | | 992 | | Processing 993 | | 994 | RDMA Write (result data) | 995 | <------------------------------ | 996 | RDMA Send (RDMA_MSG) | 997 | <------------------------------ | Reply 999 4.5.3. Long Messages 1001 When a Payload stream is larger than the receiver's inline threshold, 1002 the Payload stream is reduced by removing DDP-eligible data items and 1003 placing them in chunks to be moved separately. If there are no DDP- 1004 eligible data items in the Payload stream, or the Payload stream is 1005 still too large after it has been reduced, the RDMA transport MUST 1006 use RDMA Read or Write operations to convey the Payload stream 1007 itself. This mechanism is referred to as a "Long Message." 1009 To transmit a Long Message, the sender conveys only the Transport 1010 stream with an RDMA Send operation. The Payload stream is not 1011 included in the Send buffer in this instance. Instead, the requester 1012 provides chunks that the responder uses to move the Payload stream. 1014 Long RPC Call 1015 To send a Long RPC-over-RDMA Call message, the requester provides 1016 a special Read chunk that contains the RPC Call's Payload stream. 1017 Every segment in this Read chunk MUST contain zero in its Position 1018 field. Thus this chunk is known as a "Position Zero Read chunk." 1020 Long RPC Reply 1021 To send a Long RPC-over-RDMA Reply message, the requester provides 1022 a single special Write chunk in advance, known as the "Reply 1023 chunk", that will contain the RPC Reply's Payload stream. The 1024 requester sizes the Reply chunk to accommodate the maximum 1025 expected reply size for that Upper Layer operation. 1027 Though the purpose of a Long Message is to handle large RPC messages, 1028 requesters MAY use a Long Message at any time to convey an RPC Call. 1030 A responder chooses which form of reply to use based on the chunks 1031 provided by the requester. If Write chunks were provided and the 1032 responder has a DDP-eligible result, it first reduces the reply 1033 Payload stream. If a Reply chunk was provided and the reduced 1034 Payload stream is larger than the reply inline threshold, the 1035 responder MUST use the requester-provided Reply chunk for the reply. 1037 Because these special chunks contain a whole RPC message, XDR data 1038 items appear in these special chunks without regard to their DDP- 1039 eligibility. 1041 An RPC-over-RDMA transaction using a Long Call: 1043 Requester Responder 1044 | RDMA Send (RDMA_NOMSG) | 1045 Call | ------------------------------> | 1046 | RDMA Read | 1047 | <------------------------------ | 1048 | RDMA Response (RPC call) | 1049 | ------------------------------> | 1050 | | 1051 | | Processing 1052 | | 1053 | RDMA Send (RDMA_MSG) | 1054 | <------------------------------ | Reply 1056 An RPC-over-RDMA transaction using a Long Reply: 1058 Requester Responder 1059 | RDMA Send (RDMA_MSG) | 1060 Call | ------------------------------> | 1061 | | 1062 | | Processing 1063 | | 1064 | RDMA Write (RPC reply) | 1065 | <------------------------------ | 1066 | RDMA Send (RDMA_NOMSG) | 1067 | <------------------------------ | Reply 1069 5. RPC-Over-RDMA In Operation 1071 Every RPC-over-RDMA Version One message has a header that includes a 1072 copy of the message's transaction ID, data for managing RDMA flow 1073 control credits, and lists of RDMA segments used for RDMA Read and 1074 Write operations. All RPC-over-RDMA header content is contained in 1075 the Transport stream, and thus MUST be XDR encoded. 1077 RPC message layout is unchanged from that described in [RFC5531] 1078 except for the possible reduction of data items that are moved by 1079 RDMA Read or Write operations. 1081 The RPC-over-RDMA protocol passes RPC messages without regard to 1082 their type (CALL or REPLY). Apart from restrictions imposed by 1083 upper-layer bindings, each endpoint of a connection MAY send RDMA_MSG 1084 or RDMA_NOMSG message header types at any time (subject to credit 1085 limits). 1087 5.1. XDR Protocol Definition 1089 This section contains a description of the core features of the RPC- 1090 over-RDMA Version One protocol, expressed in the XDR language 1091 [RFC4506]. 1093 This description is provided in a way that makes it simple to extract 1094 into ready-to-compile form. The reader can apply the following shell 1095 script to this document to produce a machine-readable XDR description 1096 of the RPC-over-RDMA Version One protocol. 1098 1100 #!/bin/sh 1101 grep '^ *///' | sed 's?^ /// ??' | sed 's?^ *///$??' 1103 1105 That is, if the above script is stored in a file called "extract.sh" 1106 and this document is in a file called "spec.txt" then the reader can 1107 do the following to extract an XDR description file: 1109 1111 sh extract.sh < spec.txt > rpcrdma_corev1.x 1113 1115 5.1.1. Code Component License 1117 Code components extracted from this document must include the 1118 following license text. When the extracted XDR code is combined with 1119 other complementary XDR code which itself has an identical license, 1120 only a single copy of the license text need be preserved. 1122 1124 /// /* 1125 /// * Copyright (c) 2010, 2016 IETF Trust and the persons 1126 /// * identified as authors of the code. All rights reserved. 1127 /// * 1128 /// * The authors of the code are: 1129 /// * B. Callaghan, T. Talpey, and C. Lever 1130 /// * 1131 /// * Redistribution and use in source and binary forms, with 1132 /// * or without modification, are permitted provided that the 1133 /// * following conditions are met: 1134 /// * 1135 /// * - Redistributions of source code must retain the above 1136 /// * copyright notice, this list of conditions and the 1137 /// * following disclaimer. 1138 /// * 1139 /// * - Redistributions in binary form must reproduce the above 1140 /// * copyright notice, this list of conditions and the 1141 /// * following disclaimer in the documentation and/or other 1142 /// * materials provided with the distribution. 1143 /// * 1144 /// * - Neither the name of Internet Society, IETF or IETF 1145 /// * Trust, nor the names of specific contributors, may be 1146 /// * used to endorse or promote products derived from this 1147 /// * software without specific prior written permission. 1148 /// * 1149 /// * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 1150 /// * AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED 1151 /// * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1152 /// * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 1153 /// * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 1154 /// * EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 1155 /// * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 1156 /// * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 1157 /// * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 1158 /// * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 1159 /// * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 1160 /// * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 1161 /// * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 1162 /// * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 1163 /// * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 1164 /// */ 1165 /// 1167 1169 5.1.2. RPC-Over-RDMA Version One XDR 1171 XDR data items defined in this section encodes the Transport Header 1172 Stream in each RPC-over-RDMA Version One message. Comments identify 1173 items that cannot be changed in subsequent versions. 1175 1177 /// /* 1178 /// * Plain RDMA segment (Section 4.4.3) 1179 /// */ 1180 /// struct xdr_rdma_segment { 1181 /// uint32 handle; /* Registered memory handle */ 1182 /// uint32 length; /* Length of the chunk in bytes */ 1183 /// uint64 offset; /* Chunk virtual address or offset */ 1184 /// }; 1185 /// 1186 /// /* 1187 /// * Read segment (Section 4.4.5) 1188 /// */ 1189 /// struct xdr_read_chunk { 1190 /// uint32 position; /* Position in XDR stream */ 1191 /// struct xdr_rdma_segment target; 1192 /// }; 1193 /// 1194 /// /* 1195 /// * Read list (Section 5.3.1) 1196 /// */ 1197 /// struct xdr_read_list { 1198 /// struct xdr_read_chunk entry; 1199 /// struct xdr_read_list *next; 1200 /// }; 1201 /// 1202 /// /* 1203 /// * Write chunk (Section 4.4.6) 1204 /// */ 1205 /// struct xdr_write_chunk { 1206 /// struct xdr_rdma_segment target<>; 1207 /// }; 1208 /// 1209 /// /* 1210 /// * Write list (Section 5.3.2) 1211 /// */ 1212 /// struct xdr_write_list { 1213 /// struct xdr_write_chunk entry; 1214 /// struct xdr_write_list *next; 1215 /// }; 1216 /// 1217 /// /* 1218 /// * Chunk lists (Section 5.3) 1219 /// */ 1220 /// struct rpc_rdma_header { 1221 /// struct xdr_read_list *rdma_reads; 1222 /// struct xdr_write_list *rdma_writes; 1223 /// struct xdr_write_chunk *rdma_reply; 1224 /// /* rpc body follows */ 1225 /// }; 1226 /// 1227 /// struct rpc_rdma_header_nomsg { 1228 /// struct xdr_read_list *rdma_reads; 1229 /// struct xdr_write_list *rdma_writes; 1230 /// struct xdr_write_chunk *rdma_reply; 1231 /// }; 1232 /// 1233 /// /* Not to be used */ 1234 /// struct rpc_rdma_header_padded { 1235 /// uint32 rdma_align; 1236 /// uint32 rdma_thresh; 1237 /// struct xdr_read_list *rdma_reads; 1238 /// struct xdr_write_list *rdma_writes; 1239 /// struct xdr_write_chunk *rdma_reply; 1240 /// /* rpc body follows */ 1241 /// }; 1242 /// 1243 /// /* 1244 /// * Error handling (Section 5.5) 1245 /// */ 1246 /// enum rpc_rdma_errcode { 1247 /// ERR_VERS = 1, /* Value fixed for all versions */ 1248 /// ERR_CHUNK = 2 1249 /// }; 1250 /// 1251 /// /* Structure fixed for all versions */ 1252 /// struct rpc_rdma_errvers { 1253 /// uint32 rdma_vers_low; 1254 /// uint32 rdma_vers_high; 1255 /// }; 1256 /// 1257 /// union rpc_rdma_error switch (rpc_rdma_errcode err) { 1258 /// case ERR_VERS: 1259 /// rpc_rdma_errvers range; 1260 /// case ERR_CHUNK: 1261 /// void; 1262 /// }; 1263 /// 1264 /// /* 1265 /// * Procedures (Section 5.2.4) 1266 /// */ 1267 /// enum rdma_proc { 1268 /// RDMA_MSG = 0, /* Value fixed for all versions */ 1269 /// RDMA_NOMSG = 1, /* Value fixed for all versions */ 1270 /// RDMA_MSGP = 2, /* Not to be used */ 1271 /// RDMA_DONE = 3, /* Not to be used */ 1272 /// RDMA_ERROR = 4 /* Value fixed for all versions */ 1273 /// }; 1274 /// 1275 /// /* The position of the proc discriminator field is 1276 /// * fixed for all versions */ 1277 /// union rdma_body switch (rdma_proc proc) { 1278 /// case RDMA_MSG: 1279 /// rpc_rdma_header rdma_msg; 1280 /// case RDMA_NOMSG: 1281 /// rpc_rdma_header_nomsg rdma_nomsg; 1282 /// case RDMA_MSGP: /* Not to be used */ 1283 /// rpc_rdma_header_padded rdma_msgp; 1284 /// case RDMA_DONE: /* Not to be used */ 1285 /// void; 1286 /// case RDMA_ERROR: 1287 /// rpc_rdma_error rdma_error; 1288 /// }; 1289 /// 1290 /// /* 1291 /// * Fixed header fields (Section 5.2) 1292 /// */ 1293 /// struct rdma_msg { 1294 /// uint32 rdma_xid; /* Position fixed for all versions */ 1295 /// uint32 rdma_vers; /* Position fixed for all versions */ 1296 /// uint32 rdma_credit; /* Position fixed for all versions */ 1297 /// rdma_body rdma_body; 1298 /// }; 1300 1302 5.2. Fixed Header Fields 1304 The RPC-over-RDMA header begins with four fixed 32-bit fields that 1305 control the RDMA interaction. 1307 The first three words are individual fields in the rdma_msg 1308 structure. The fourth word is the first word of the rdma_body union 1309 which acts as the discriminator for the switched union. The contents 1310 of this field are described in Section 5.2.4. 1312 These four fields must remain with the same meanings and in the same 1313 positions in all subsequent versions of the RPC-over-RDMA protocol. 1315 5.2.1. Transaction ID (XID) 1317 The XID generated for the RPC Call and Reply. Having the XID at a 1318 fixed location in the header makes it easy for the receiver to 1319 establish context as soon as each RPC-over-RDMA message arrives. 1320 This XID MUST be the same as the XID in the RPC message. The 1321 receiver MAY perform its processing based solely on the XID in the 1322 RPC-over-RDMA header, and thereby ignore the XID in the RPC message, 1323 if it so chooses. 1325 5.2.2. Version Number 1327 For RPC-over-RDMA Version One, this field MUST contain the value one 1328 (1). Rules regarding changes to this transport protocol version 1329 number can be found in Section 8. 1331 5.2.3. Credit Value 1333 When sent with an RPC Call message, the requested credit value is 1334 provided. When sent with an RPC Reply message, the granted credit 1335 value is returned. Further discussion of how the credit value is 1336 determined can be found in Section 4.3. 1338 5.2.4. Procedure Number 1340 o RDMA_MSG = 0 indicates that chunk lists and a Payload stream 1341 follow. The format of the chunk lists is discussed below. 1343 o RDMA_NOMSG = 1 indicates that after the chunk lists there is no 1344 Payload stream. In this case, the chunk lists provide information 1345 to allow the responder to transfer the Payload stream using RDMA 1346 Read or Write operations. 1348 o RDMA_MSGP = 2 is reserved. 1350 o RDMA_DONE = 3 is reserved. 1352 o RDMA_ERROR = 4 is used to signal an encoding error in the RPC- 1353 over-RDMA header. 1355 An RDMA_MSG procedure conveys the Transport stream and the Payload 1356 stream via an RDMA Send operation. The Transport stream contains the 1357 four fixed fields, followed by the Read and Write lists and the Reply 1358 chunk, though any or all three MAY be marked as not present. The 1359 Payload stream then follows, beginning with its XID field. If a Read 1360 or Write chunk list is present, a portion of the Payload stream has 1361 been excised and is conveyed separately via RDMA Read or Write 1362 operations. 1364 An RDMA_NOMSG procedure conveys the Transport stream via an RDMA Send 1365 operation. The Transport stream contains the four fixed fields, 1366 followed by the Read and Write chunk lists and the Reply chunk. 1367 Though any of these MAY be marked as not present, one MUST be present 1368 and MUST hold the Payload stream for this RPC-over-RDMA message. If 1369 a Read or Write chunk list is present, a portion of the Payload 1370 stream has been excised and is conveyed separately via RDMA Read or 1371 Write operations. 1373 An RDMA_ERROR procedure conveys the Transport stream via an RDMA Send 1374 operation. The Transport stream contains the four fixed fields, 1375 followed by formatted error information. No Payload stream is 1376 conveyed in this type of RPC-over-RDMA message. 1378 A requester MUST NOT send an RPC-over-RDMA header with the RDMA_ERROR 1379 procedure. A responder MUST silently discard RDMA_ERROR procedures. 1381 A gather operation on each RDMA Send operation can be used to combine 1382 the Transport and Payload streams, which might have been constructed 1383 in separate buffers. However, the total length of the gathered send 1384 buffers MUST NOT exceed the inline threshold. 1386 5.3. Chunk Lists 1388 The chunk lists in an RPC-over-RDMA Version One header are three XDR 1389 optional-data fields that follow the fixed header fields in RDMA_MSG 1390 and RDMA_NOMSG procedures. Read Section 4.19 of [RFC4506] carefully 1391 to understand how optional-data fields work. Examples of XDR encoded 1392 chunk lists are provided in Section 5.7 as an aid to understanding. 1394 5.3.1. Read List 1396 Each RDMA_MSG or RDMA_NOMSG procedure has one "Read list." The Read 1397 list is a list of zero or more Read segments, provided by the 1398 requester, that are grouped by their Position fields into Read 1399 chunks. Each Read chunk advertises the location of argument data the 1400 responder is to retrieve via RDMA Read operations. The requester has 1401 removed the data in these chunks from the call's Payload stream. 1403 Via a Position Zero Read Chunk, a requester may provide an RPC Call 1404 message as a chunk in the Read list. 1406 If the RPC Call has no argument data that is DDP-eligible and the 1407 Position Zero Read Chunk is not being used, the requester leaves the 1408 Read list empty. 1410 Responders MUST leave the Read list empty in all replies. 1412 5.3.2. Write List 1414 Each RDMA_MSG or RDMA_NOMSG procedure has one "Write list." The 1415 Write list is a list of zero or more Write chunks, provided by the 1416 requester. Each Write chunk is an array of RDMA segments, thus the 1417 Write list is a list of counted arrays. Each Write chunk advertises 1418 receptacles for DDP-eligible data to be pushed by the responder via 1419 RDMA Write operations. If the RPC Reply has no possible DDP-eligible 1420 result data items, the requester leaves the Write list empty. 1422 When a Write list is provided for the results of an RPC Call, the 1423 responder MUST provide data corresponding to DDP-eligible XDR data 1424 items via RDMA Write operations to the memory referenced in the Write 1425 list. The responder removes the data in these chunks from the 1426 reply's Payload stream. 1428 When multiple Write chunks are present, the responder fills in each 1429 Write chunk with a DDP-eligible result until either there are no more 1430 results or no more Write chunks. The requester may not be able to 1431 predict which DDP-eligible data item goes in which chunk. Thus the 1432 requester is responsible for allocating and registering Write chunks 1433 large enough to accommodate the largest XDR data item that might be 1434 associated with each chunk in the list. 1436 The RPC Reply conveys the size of result data items by returning each 1437 Write chunk to the requester with the segment lengths rewritten to 1438 match the actual data transferred. Decoding the reply therefore 1439 performs no local data copying but merely returns the length obtained 1440 from the reply. 1442 Each decoded result consumes one entry in the Write list, which in 1443 turn consists of an array of RDMA segments. The length of a Write 1444 chunk is therefore the sum of all returned lengths in all segments 1445 comprising the corresponding list entry. As each Write chunk is 1446 decoded, the entire Write list entry is consumed. 1448 A requester constructs the Write list for an RPC transaction before 1449 the responder has formulated its reply. When there is only one DDP- 1450 eligible result data item, the requester inserts only a single Write 1451 chunk in the Write list. If the responder populates that chunk with 1452 data, the requester knows with certainty which result data item is 1453 contained in it. 1455 However, Upper Layer Protocol procedures may allow replies where more 1456 than one result data item is DDP-eligible. For example, an NFSv4 1457 COMPOUND procedure is composed of individual NFSv4 operations, more 1458 than one of which may have a reply containing a DDP-eligible result. 1460 As stated above, when multiple Write chunks are present, the 1461 responder reduces DDP-eligible results until either there are no more 1462 results or no more Write chunks. Then, as the requester decodes the 1463 reply Payload stream, it is clear from the contents of the reply 1464 which Write chunk contains which data item. 1466 When a requester has provided a Write list in a Call message, the 1467 responder MUST copy that list into the associated Reply. The copied 1468 Write list in the Reply is modified as above to reflect the actual 1469 amount of data that is being returned in the Write list. 1471 5.3.3. Reply Chunk 1473 Each RDMA_MSG or RDMA_NOMSG procedure has one "Reply chunk." The 1474 Reply chunk is a Write chunk, provided by the requester. The Reply 1475 chunk is a single counted array of RDMA segments. 1477 A requester MUST provide a Reply chunk whenever the maximum possible 1478 size of the reply message is larger than the inline threshold for 1479 messages from responder to requester. The Reply chunk MUST be large 1480 enough to contain a Payload stream (RPC message) of this maximum 1481 size. If the Transport stream and reply Payload stream together are 1482 smaller than the reply inline threshold, the responder MAY return it 1483 as a Short message rather than using the requester-provided Reply 1484 chunk. 1486 When a requester has provided a Reply chunk in a Call message, the 1487 responder MUST copy that chunk into the associated Reply. The copied 1488 Reply chunk in the Reply is modified to reflect the actual amount of 1489 data that is being returned in the Reply chunk. 1491 5.4. Memory Registration 1493 RDMA requires that data is transferred between only registered memory 1494 segments at the source and destination. All protocol headers as well 1495 as separately transferred data chunks must reside in registered 1496 memory. 1498 Since the cost of registering and de-registering memory can be a 1499 significant proportion of the RDMA transaction cost, it is important 1500 to minimize registration activity. For memory that is targeted by 1501 RDMA Send and Receive operations, a local-only registration is 1502 sufficient and can be left in place during the life of a connection 1503 without any risk of data exposure. 1505 5.4.1. Registration Longevity 1507 Data transferred via RDMA Read and Write can reside in a memory 1508 allocation not in the control of the RPC-over-RDMA transport. These 1509 memory allocations can persist outside the bounds of an RPC 1510 transaction. They are registered and invalidated as needed, as part 1511 of each RPC transaction. 1513 The requester endpoint must ensure that memory segments associated 1514 with each RPC transaction are properly fenced from responders before 1515 allowing Upper Layer access to the data contained in them. Moreover, 1516 the requester must not access these memory segments while the 1517 responder has access to them. 1519 This includes segments that are associated with canceled RPCs. A 1520 responder cannot know that the requester is no longer waiting for a 1521 reply, and might proceed to read or even update memory that the 1522 requester might have released for other use. 1524 5.4.2. Communicating DDP-Eligibility 1526 The interface by which an Upper Layer Protocol implementation 1527 communicates the eligibility of a data item locally to its local RPC- 1528 over-RDMA endpoint is not described by this specification. 1530 Depending on the implementation and constraints imposed by Upper 1531 Layer Bindings, it is possible to implement reduction transparently 1532 to upper layers. Such implementations may lead to inefficiencies, 1533 either because they require the RPC layer to perform expensive 1534 registration and de-registration of memory "on the fly", or they may 1535 require using RDMA chunks in reply messages, along with the resulting 1536 additional handshaking with the RPC-over-RDMA peer. 1538 However, these issues are internal and generally confined to the 1539 local interface between RPC and its upper layers, one in which 1540 implementations are free to innovate. The only requirement, beyond 1541 constraints imposed by the Upper Layer Binding, is that the resulting 1542 RPC-over-RDMA protocol sent to the peer is valid for the upper layer. 1544 5.4.3. Registration Strategies 1546 The choice of which memory registration strategies to employ is left 1547 to requester and responder implementers. To support the widest array 1548 of RDMA implementations, as well as the most general steering tag 1549 scheme, an Offset field is included in each segment. 1551 While zero-based offset schemes are available in many RDMA 1552 implementations, their use by RPC requires individual registration of 1553 each segment. For such implementations, this can be a significant 1554 overhead. By providing an offset in each chunk, many pre- 1555 registration or region-based registrations can be readily supported. 1556 By using a single, universal chunk representation, the RPC-over-RDMA 1557 protocol implementation is simplified to its most general form. 1559 5.5. Error Handling 1561 A receiver performs basic validity checks on the RPC-over-RDMA header 1562 and chunk contents before it passes the RPC message to the RPC 1563 consumer. If an incoming RPC-over-RDMA message is not as long as a 1564 minimal size RPC-over-RDMA header (28 bytes), the receiver cannot 1565 trust the value of the XID field, and therefore MUST silently discard 1566 the message before performing any parsing. If other errors are 1567 detected in the RPC-over-RDMA header of a Call message, a responder 1568 MUST send an RDMA_ERROR message back to the requester. If errors are 1569 detected in the RPC-over-RDMA header of a Reply message, a requester 1570 MUST silently discard the message. 1572 To form an RDMA_ERROR procedure: The rdma_xid field MUST contain the 1573 same XID that was in the rdma_xid field in the failing request; The 1574 rdma_vers field MUST contain the same version that was in the 1575 rdma_vers field in the failing request; The rdma_proc field MUST 1576 contain the value RDMA_ERROR; The rdma_err field contains a value 1577 that reflects the type of error that occurred, as described below. 1579 An RDMA_ERROR procedure indicates a permanent error. Receipt of this 1580 procedure completes the RPC transaction associated with XID in the 1581 rdma_xid field. A receiver MUST silently discard an RDMA_ERROR 1582 procedure that it cannot decode. 1584 5.5.1. Header Version Mismatch 1586 When a responder detects an RPC-over-RDMA header version that it does 1587 not support (currently this document defines only Version One), it 1588 MUST reply with an RDMA_ERROR procedure and set the rdma_err value to 1589 ERR_VERS, also providing the low and high inclusive version numbers 1590 it does, in fact, support. 1592 5.5.2. XDR Errors 1594 A receiver might encounter an XDR parsing error that prevents it from 1595 processing the incoming Transport stream. Examples of such errors 1596 include an invalid value in the rdma_proc field, an RDMA_NOMSG 1597 message that has no chunk lists, or the contents of the rdma_xid 1598 field might not match the contents of the XID field in the 1599 accompanying RPC message. If the rdma_vers field contains a 1600 recognized value, but an XDR parsing error occurs, the responder MUST 1601 reply with an RDMA_ERROR procedure and set the rdma_err value to 1602 ERR_CHUNK. 1604 When a responder receives a valid RPC-over-RDMA header but the 1605 responder's Upper Layer Protocol implementation cannot parse the RPC 1606 arguments in the RPC Call message, the responder SHOULD return a 1607 RPC_GARBAGEARGS reply, using an RDMA_MSG procedure. This type of 1608 parsing failure might be due to mismatches between chunk sizes or 1609 offsets and the contents of the Payload stream, for example. A 1610 responder MAY also report the presence of a non-DDP-eligible data 1611 item in a Read or Write chunk using RPC_GARBAGEARGS. 1613 5.5.3. Responder RDMA Operational Errors 1615 In RPC-over-RDMA Version One, it is the responder which drives RDMA 1616 Read and Write operations that target the requester's memory. 1617 Problems might arise as the responder attempts to use requester- 1618 provided resources for RDMA operations. For example: 1620 o Chunks can be validated only by using their contents to form RDMA 1621 Read or Write operations. If chunk contents are invalid (say, a 1622 segment is no longer registered, or a chunk length is too long), a 1623 Remote Access error occurs. 1625 o If a requester's receive buffer is too small, the responder's Send 1626 operation completes with a Local Length Error. 1628 o If the requester-provided Reply chunk is too small to accommodate 1629 a large RPC Reply, a Remote Access error occurs. A responder can 1630 detect this problem before attempting to write past the end of the 1631 Reply chunk. 1633 RDMA operational errors are typically fatal to the connection. To 1634 avoid a retransmission loop and repeated connection loss that 1635 deadlocks the connection, once the requester has re-established a 1636 connection, the responder should send an RDMA_ERROR reply with an 1637 rdma_err value of ERR_CHUNK to indicate that no RPC-level reply is 1638 possible for that XID. 1640 5.5.4. Other Operational Errors 1642 While a requester is constructing a Call message, an unrecoverable 1643 problem might occur that prevents the requester from posting further 1644 RDMA Work Requests on behalf of that message. As with other 1645 transports, if a requester is unable to construct and transmit a Call 1646 message, the associated RPC transaction fails immediately. 1648 After a requester has received a reply, if it is unable to invalidate 1649 a memory region due to an unrecoverable problem, the requester MUST 1650 close the connection to fence that memory from the responder before 1651 the associated RPC transaction is complete. 1653 While a responder is constructing a Reply message or error message, 1654 an unrecoverable problem might occur that prevents the responder from 1655 posting further RDMA Work Requests on behalf of that message. If a 1656 responder is unable to construct and transmit a Reply or error 1657 message, the responder MUST close the connection to signal to the 1658 requester that a reply was lost. 1660 5.5.5. RDMA Transport Errors 1662 The RDMA connection and physical link provide some degree of error 1663 detection and retransmission. iWARP's Marker PDU Aligned (MPA) layer 1664 (when used over TCP), Stream Control Transmission Protocol (SCTP), as 1665 well as the InfiniBand link layer all provide Cyclic Redundancy Check 1666 (CRC) protection of the RDMA payload, and CRC-class protection is a 1667 general attribute of such transports. 1669 Additionally, the RPC layer itself can accept errors from the 1670 transport, and recover via retransmission. RPC recovery can handle 1671 complete loss and re-establishment of a transport connection. 1673 The details of reporting and recovery from RDMA link layer errors are 1674 outside the scope of this protocol specification. See Section 9 for 1675 further discussion of the use of RPC-level integrity schemes to 1676 detect errors. 1678 5.6. Protocol Elements No Longer Supported 1680 The following protocol elements are no longer supported in RPC-over- 1681 RDMA Version One. Related enum values and structure definitions 1682 remain in the RPC-over-RDMA Version One protocol for backwards 1683 compatibility. 1685 5.6.1. RDMA_MSGP 1687 The specification of RDMA_MSGP in Section 3.9 of [RFC5666] is 1688 incomplete. To fully specify RDMA_MSGP would require: 1690 o Updating the definition of DDP-eligibility to include data items 1691 that may be transferred, with padding, via RDMA_MSGP procedures 1693 o Adding full operational descriptions of the alignment and 1694 threshold fields 1696 o Discussing how alignment preferences are communicated between two 1697 peers without using CCP 1699 o Describing the treatment of RDMA_MSGP procedures that convey Read 1700 or Write chunks 1702 The RDMA_MSGP message type is beneficial only when the padded data 1703 payload is at the end of an RPC message's argument or result list. 1704 This is not typical for NFSv4 COMPOUND RPCs, which often include a 1705 GETATTR operation as the final element of the compound operation 1706 array. 1708 Without a full specification of RDMA_MSGP, there has been no fully 1709 implemented prototype of it. Without a complete prototype of 1710 RDMA_MSGP support, it is difficult to assess whether this protocol 1711 element has benefit, or can even be made to work interoperably. 1713 Therefore, senders MUST NOT send RDMA_MSGP procedures. When 1714 receiving an RDMA_MSGP procedure, responders SHOULD reply with an 1715 RDMA_ERROR procedure, setting the rdma_err field to ERR_CHUNK; 1716 requesters MUST silently discard the message. 1718 5.6.2. RDMA_DONE 1720 Because no implementation of RPC-over-RDMA Version One uses the Read- 1721 Read transfer model, there is never a need to send an RDMA_DONE 1722 procedure. 1724 Therefore, senders MUST NOT send RDMA_DONE messages. Receivers MUST 1725 silently discard RDMA_DONE messages. 1727 5.7. XDR Examples 1729 RPC-over-RDMA chunk lists are complex data types. In this section, 1730 illustrations are provided to help readers grasp how chunk lists are 1731 represented inside an RPC-over-RDMA header. 1733 An RDMA segment is the simplest component, being made up of a 32-bit 1734 handle (H), a 32-bit length (L), and 64-bits of offset (OO). Once 1735 flattened into an XDR stream, RDMA segments appear as 1737 HLOO 1739 A Read segment has an additional 32-bit position field. Read 1740 segments appear as 1741 PHLOO 1743 A Read chunk is a list of Read segments. Each segment is preceded by 1744 a 32-bit word containing a one if there is a segment, or a zero if 1745 there are no more segments (optional-data). In XDR form, this would 1746 look like 1748 1 PHLOO 1 PHLOO 1 PHLOO 0 1750 where P would hold the same value for each segment belonging to the 1751 same Read chunk. 1753 The Read List is also a list of Read segments. In XDR form, this 1754 would look like a Read chunk, except that the P values could vary 1755 across the list. An empty Read List is encoded as a single 32-bit 1756 zero. 1758 One Write chunk is a counted array of segments. In XDR form, the 1759 count would appear as the first 32-bit word, followed by an HLOO for 1760 each element of the array. For instance, a Write chunk with three 1761 elements would look like 1763 3 HLOO HLOO HLOO 1765 The Write List is a list of counted arrays. In XDR form, this is a 1766 combination of optional-data and counted arrays. To represent a 1767 Write List containing a Write chunk with three segments and a Write 1768 chunk with two segments, XDR would encode 1770 1 3 HLOO HLOO HLOO 1 2 HLOO HLOO 0 1772 An empty Write List is encoded as a single 32-bit zero. 1774 The Reply chunk is a Write chunk. Since it is an optional-data 1775 field, however, there is a 32-bit field in front of it that contains 1776 a one if the Reply chunk is present, or a zero if it is not. After 1777 encoding, a Reply chunk with 2 segments would look like 1779 1 2 HLOO HLOO 1781 Frequently a requester does not provide any chunks. In that case, 1782 after the four fixed fields in the RPC-over-RDMA header, there are 1783 simply three 32-bit fields that contain zero. 1785 6. RPC Bind Parameters 1787 In setting up a new RDMA connection, the first action by a requester 1788 is to obtain a transport address for the responder. The mechanism 1789 used to obtain this address, and to open an RDMA connection is 1790 dependent on the type of RDMA transport, and is the responsibility of 1791 each RPC protocol binding and its local implementation. 1793 RPC services normally register with a portmap or rpcbind [RFC1833] 1794 service, which associates an RPC Program number with a service 1795 address. (In the case of UDP or TCP, the service address for NFS is 1796 normally port 2049.) This policy is no different with RDMA 1797 transports, although it may require the allocation of port numbers 1798 appropriate to each Upper Layer Protocol that uses the RPC framing 1799 defined here. 1801 When mapped atop the iWARP transport [RFC5040] [RFC5041], which uses 1802 IP port addressing due to its layering on TCP and/or SCTP, port 1803 mapping is trivial and consists merely of issuing the port in the 1804 connection process. The NFS/RDMA protocol service address has been 1805 assigned port 20049 by IANA, for both iWARP/TCP and iWARP/SCTP. 1807 When mapped atop InfiniBand [IB], which uses a Group Identifier 1808 (GID)-based service endpoint naming scheme, a translation MUST be 1809 employed. One such translation is defined in the InfiniBand Port 1810 Addressing Annex [IBPORT], which is appropriate for translating IP 1811 port addressing to the InfiniBand network. Therefore, in this case, 1812 IP port addressing may be readily employed by the upper layer. 1814 When a mapping standard or convention exists for IP ports on an RDMA 1815 interconnect, there are several possibilities for each upper layer to 1816 consider: 1818 o One possibility is to have responder register its mapped IP port 1819 with the rpcbind service, under the netid (or netid's) defined 1820 here. An RPC-over-RDMA-aware requester can then resolve its 1821 desired service to a mappable port, and proceed to connect. This 1822 is the most flexible and compatible approach, for those upper 1823 layers that are defined to use the rpcbind service. 1825 o A second possibility is to have the responder's portmapper 1826 register itself on the RDMA interconnect at a "well known" service 1827 address (on UDP or TCP, this corresponds to port 111). A 1828 requester could connect to this service address and use the 1829 portmap protocol to obtain a service address in response to a 1830 program number, e.g., an iWARP port number, or an InfiniBand GID. 1832 o Alternatively, the requester could simply connect to the mapped 1833 well-known port for the service itself, if it is appropriately 1834 defined. By convention, the NFS/RDMA service, when operating atop 1835 such an InfiniBand fabric, will use the same 20049 assignment as 1836 for iWARP. 1838 Historically, different RPC protocols have taken different approaches 1839 to their port assignment; therefore, the specific method is left to 1840 each RPC-over-RDMA-enabled Upper Layer binding, and not addressed 1841 here. 1843 In Section 10, this specification defines two new "netid" values, to 1844 be used for registration of upper layers atop iWARP [RFC5040] 1845 [RFC5041] and (when a suitable port translation service is available) 1846 InfiniBand [IB]. Additional RDMA-capable networks MAY define their 1847 own netids, or if they provide a port translation, MAY share the one 1848 defined here. 1850 7. Upper Layer Binding Specifications 1852 An Upper Layer Protocol is typically defined independently of any 1853 particular RPC transport. An Upper Layer Binding specification (ULB) 1854 provides guidance that helps the Upper Layer Protocol interoperate 1855 correctly and efficiently over a particular transport. For RPC-over- 1856 RDMA Version One, an Upper Layer Binding may provide: 1858 o A taxonomy of XDR data items that are eligible for Direct Data 1859 Placement 1861 o Constraints on which Upper Layer procedures may be reduced, and on 1862 how many chunks may appear in a single RPC request 1864 o A method for determining the maximum size of the reply Payload 1865 stream for all procedures in the Upper Layer Protocol 1867 o An rpcbind port assignment for operation of the RPC Program and 1868 Version on an RPC-over-RDMA transport 1870 Each RPC Program and Version tuple that utilizes RPC-over-RDMA 1871 Version One needs to have an Upper Layer Binding specification. 1873 7.1. DDP-Eligibility 1875 An Upper Layer Binding designates some XDR data items as eligible for 1876 Direct Data Placement. As an RPC-over-RDMA message is formed, DDP- 1877 eligible data items can be removed from the Payload stream and placed 1878 directly in the receiver's memory. 1880 An XDR data item should be considered for DDP-eligibility if there is 1881 a clear benefit to moving the contents of the item directly from the 1882 sender's memory to the receiver's memory. Criteria for DDP- 1883 eligibility include: 1885 o The XDR data item is frequently sent or received, and its size is 1886 often much larger than typical inline thresholds. 1888 o Transport-level processing of the XDR data item is not needed. 1889 For example, the data item is an opaque byte array, which requires 1890 no XDR encoding and decoding of its content. 1892 o The content of the XDR data item is sensitive to address 1893 alignment. For example, pullup would be required on the receiver 1894 before the content of the item can be used. 1896 o The XDR data item does not contain DDP-eligible data items. 1898 In addition to defining the set of data items that are DDP-eligible, 1899 an Upper Layer Binding may also limit the use of chunks to particular 1900 Upper Layer procedures. If more than one data item in a procedure is 1901 DDP-eligible, the Upper Layer Binding may also limit the number of 1902 chunks that a requester can provide for a particular Upper Layer 1903 procedure. 1905 Senders MUST NOT reduce data items that are not DDP-eligible. Such 1906 data items MAY, however, be moved as part of a Position Zero Read 1907 Chunk or a Reply chunk. 1909 The programming interface by which an Upper Layer implementation 1910 indicates the DDP-eligibility of a data item to the RPC transport is 1911 not described by this specification. The only requirements are that 1912 the receiver can re-assemble the transmitted RPC-over-RDMA message 1913 into a valid XDR stream, and that DDP-eligibility rules specified by 1914 the Upper Layer Binding are respected. 1916 There is no provision to express DDP-eligibility within the XDR 1917 language. The only definitive specification of DDP-eligibility is an 1918 Upper Layer Binding. 1920 7.1.1. DDP-Eligibility Violation 1922 A DDP-eligibility violation occurs when a requester forms a Call 1923 message with a non-DDP-eligible data item in a Read chunk. A 1924 violation occurs when a responder forms a Reply message without 1925 reducing a DDP-eligible data item when there is a Write list provided 1926 by the requester. 1928 In the first case, a responder MUST NOT process the Call message. 1930 In the second case, as a requester parses a Reply message, it must 1931 assume that the responder has correctly reduced a DDP-eligible result 1932 data item. If the responder has not done so, it is likely that the 1933 requester cannot finish parsing the Payload stream and that an XDR 1934 error would result. 1936 Both types of violations MUST be reported as described in 1937 Section 5.5.2. 1939 7.2. Maximum Reply Size 1941 A requester provides resources for both a Call message and its 1942 matching Reply message. A requester forms the Call message itself, 1943 thus can compute the exact resources needed for it. 1945 A requester must allocate resources for the Reply message (an RPC- 1946 over-RDMA credit, a Receive buffer, and possibly a Write list and 1947 Reply chunk) before the responder has formed the actual reply. To 1948 accommodate all possible replies for the procedure in the Call 1949 message, a requester must allocate reply resources based on the 1950 maximum possible size of the expected Reply message. 1952 If there are procedures in the Upper Layer Protocol for which there 1953 is no clear reply size maximum, the Upper Layer Binding needs to 1954 specify a dependable means for determining the maximum. 1956 7.3. Additional Considerations 1958 There may be other details provided in an Upper Layer Binding. 1960 o An Upper Layer Binding may recommend inline threshold values or 1961 other transport-related parameters for RPC-over-RDMA Version One 1962 connections bearing that Upper Layer Protocol. 1964 o An Upper Layer Protocol may provide a means to communicate these 1965 transport-related parameters between peers. Note that RPC-over- 1966 RDMA Version One does not specify any mechanism for changing any 1967 transport-related parameter after a connection has been 1968 established. 1970 o Multiple Upper Layer Protocols may share a single RPC-over-RDMA 1971 Version One connection when their Upper Layer Bindings allow the 1972 use of RPC-over-RDMA Version One and the rpcbind port assignments 1973 for the Protocols allow connection sharing. In this case, the 1974 same transport parameters (such as inline threshold) apply to all 1975 Protocols using that connection. 1977 Each Upper Layer Binding needs to be designed to allow correct 1978 interoperation without regard to the transport parameters actually in 1979 use. Furthermore, implementations of Upper Layer Protocols must be 1980 designed to interoperate correctly regardless of the connection 1981 parameters in effect on a connection. 1983 7.4. Upper Layer Protocol Extensions 1985 An RPC Program and Version tuple may be extensible. For instance, 1986 there may be a minor versioning scheme that is not reflected in the 1987 RPC version number. Or, the Upper Layer Protocol may allow 1988 additional features to be specified after the original RPC program 1989 specification was ratified. 1991 Upper Layer Bindings are provided for interoperable RPC Programs and 1992 Versions by extending existing Upper Layer Bindings to reflect the 1993 changes made necessary by each addition to the existing XDR. 1995 8. Protocol Extensibility 1997 The RPC-over-RDMA header format is specified using XDR, unlike the 1998 message header used with RPC over TCP. To maintain a high degree of 1999 interoperability among implementations of RPC-over-RDMA, any change 2000 to this XDR requires a protocol version number change. New versions 2001 of RPC-over-RDMA may be published as separate protocol specifications 2002 without updating this document. 2004 The first four fields in every RPC-over-RDMA header must remain 2005 aligned at the same fixed offsets for all versions of the RPC-over- 2006 RDMA protocol. The version number must be in a fixed place to enable 2007 implementations to detect protocol version mismatches. 2009 For version mismatches to be reported in a fashion that all future 2010 version implementations can reliably decode, the rdma_proc field must 2011 remain in a fixed place, the value of ERR_VERS must always remain the 2012 same, and the field placement in struct rpc_rdma_errvers must always 2013 remain the same. 2015 8.1. Conventional Extensions 2017 Introducing new capabilities to RPC-over-RDMA Version One is limited 2018 to the adoption of conventions that make use of existing XDR (defined 2019 in this document) and allowed abstract RDMA operations. Because no 2020 mechanism for detecting optional features exists in RPC-over-RDMA 2021 Version One, implementations must rely on Upper Layer Protocols to 2022 communicate the existence of such extensions. 2024 Such extensions must be specified in a Standards Track document with 2025 appropriate review by the nfsv4 Working Group and the IESG. An 2026 example of a conventional extension to RPC-over-RDMA Version One is 2027 the specification of backward direction message support to enable 2028 NFSv4.1 callback operations, described in 2029 [I-D.ietf-nfsv4-rpcrdma-bidirection]. 2031 9. Security Considerations 2033 9.1. Memory Protection 2035 A primary consideration is the protection of the integrity and 2036 privacy of local memory by an RPC-over-RDMA transport. The use of 2037 RPC-over-RDMA MUST NOT introduce any vulnerabilities to system memory 2038 contents, nor to memory owned by user processes. 2040 It is REQUIRED that any RDMA provider used for RPC transport be 2041 conformant to the requirements of [RFC5042] in order to satisfy these 2042 protections. These protections are provided by the RDMA layer 2043 specifications, and in particular, their security models. 2045 9.1.1. Protection Domains 2047 The use of Protection Domains to limit the exposure of memory 2048 segments to a single connection is critical. Any attempt by an 2049 endpoint not participating in that connection to re-use memory 2050 handles needs to result in immediate failure of that connection. 2051 Because Upper Layer Protocol security mechanisms rely on this aspect 2052 of Reliable Connection behavior, strong authentication of remote 2053 endpoints is recommended. 2055 9.1.2. Handle Predictability 2057 Unpredictable memory handles should be used for any operation 2058 requiring advertised memory segments. Advertising a continuously 2059 registered memory region allows a remote host to read or write to 2060 that region even when an RPC involving that memory is not under way. 2061 Therefore implementations should avoid advertising persistently 2062 registered memory. 2064 9.1.3. Memory Fencing 2066 Requesters should register memory segments for remote access only 2067 when they are about to be the target of an RPC operation that 2068 involves an RDMA Read or Write. 2070 Registered memory segments should be invalidated as soon as related 2071 RPC operations are complete. Invalidation and DMA unmapping of RDMA 2072 segments should be complete before message integrity checking is 2073 done, and before the RPC consumer is allowed to continue execution 2074 and use or alter the contents of a memory region. 2076 An RPC transaction on a requester might be terminated before a reply 2077 arrives if the RPC consumer exits unexpectedly (for example it is 2078 signaled or a segmentation fault occurs). When an RPC terminates 2079 abnormally, memory segments associated with that RPC should be 2080 invalidated appropriately before the segments are released to be 2081 reused for other purposes on the requester. 2083 9.2. RPC Message Security 2085 ONC RPC provides cryptographic security via the RPCSEC_GSS framework 2086 [I-D.ietf-nfsv4-rpcsec-gssv3]. RPCSEC_GSS implements message 2087 authentication, per-message integrity checking, and per-message 2088 confidentiality. However, integrity and privacy services require 2089 significant movement of data on each endpoint host. Some performance 2090 benefits enabled by RDMA transports can be lost. 2092 9.2.1. RPC-Over-RDMA Protection At Lower Layers 2094 Note that performance loss is expected when RPCSEC_GSS integrity or 2095 privacy is in use on any RPC transport. Protection below the RDMA 2096 layer is a more appropriate security mechanism for RDMA transports in 2097 performance-sensitive deployments. Certain configurations of IPsec 2098 can be co-located in RDMA hardware, for example, without any change 2099 to RDMA consumers or loss of data movement efficiency. 2101 The use of protection in a lower layer MAY be negotiated through the 2102 use of an RPCSEC_GSS security flavor defined in 2103 [I-D.ietf-nfsv4-rpcsec-gssv3] in conjunction with the Channel Binding 2104 mechanism [RFC5056] and IPsec Channel Connection Latching [RFC5660]. 2105 Use of such mechanisms is REQUIRED where integrity and/or privacy is 2106 desired and where efficiency is required. 2108 9.2.2. RPCSEC_GSS On RPC-Over-RDMA Transports 2110 Not all RDMA devices and fabrics support the above protection 2111 mechanisms. Also, per-message authentication is still required on 2112 NFS clients where multiple users access NFS files. In these cases, 2113 RPCSEC_GSS can protect NFS traffic conveyed on RPC-over-RDMA 2114 connections. 2116 RPCSEC_GSS extends the ONC RPC protocol [RFC5531] without changing 2117 the format of RPC messages. By observing the conventions described 2118 in this section, an RPC-over-RDMA transport can convey RPCSEC_GSS- 2119 protected RPC messages interoperably. 2121 As part of the ONC RPC protocol, protocol elements of RPCSEC_GSS that 2122 appear in the Payload stream of an RPC-over-RDMA message (such as 2123 control messages exchanged as part of establishing or destroying a 2124 security context, or data items that are part of RPCSEC_GSS 2125 authentication material) MUST NOT be reduced. 2127 9.2.2.1. RPCSEC_GSS Context Negotiation 2129 Some NFS client implementations use a separate connection to 2130 establish a GSS context for NFS operation. These clients use TCP and 2131 the standard NFS port (2049) for context establishment. However 2132 there is no guarantee that an NFS/RDMA server provides a TCP-based 2133 NFS server on port 2049. 2135 9.2.2.2. RPC-Over-RDMA With RPCSEC_GSS Authentication 2137 The RPCSEC_GSS authentication service has no impact on the DDP- 2138 eligibity of data items in an Upper Layer Protocol. 2140 However, RPCSEC_GSS authentication material appearing in an RPC 2141 message header can be larger than, say, an AUTH_SYS authenticator. 2142 In particular, when an RPCSEC_GSS pseudoflavor is in use, a requester 2143 needs to accommodate a larger RPC credential when marshaling Call 2144 messages, and to provide for a maximum size RPCSEC_GSS verifier when 2145 allocating reply buffers and Reply chunks. 2147 RPC messages, and thus Payload streams, are made larger as a result. 2148 Upper Layer Protocol operations that fit in a Short Message when a 2149 simpler form of authentication is in use might need to be reduced, or 2150 conveyed via a Long Message, when RPCSEC_GSS authentication is in 2151 use. It is more likely that a requester provides both a Read list 2152 and a Reply chunk in the same RPC-over-RDMA header to convey a Long 2153 call and provision a receptacle for a Long reply. More frequent use 2154 of Long messages can impact transport efficiency. 2156 9.2.2.3. RPC-Over-RDMA With RPCSEC_GSS Integrity Or Privacy 2158 The RPCSEC_GSS integrity service enables endpoints to detect 2159 modification of RPC messages in flight. The RPCSEC_GSS privacy 2160 service prevents all but the intended recipient from viewing the 2161 cleartext content of RPC arguments and results. RPCSEC_GSS integrity 2162 and privacy are end-to-end. They protect RPC arguments and results 2163 from application to server endpoint, and back. 2165 The RPCSEC_GSS integrity and encryption services operate on whole RPC 2166 messages after they have been XDR encoded for transmit, and before 2167 they have been XDR decoded after receipt. Both sender and receiver 2168 endpoints use intermediate buffers to prevent exposure of encrypted 2169 data or unverified cleartext data to RPC consumers. After 2170 verification, encryption, and message wrapping has been performed, 2171 the transport layer MAY use RDMA data transfer between these 2172 intermediate buffers. 2174 The process of reducing a DDP-eligible data item removes the data 2175 item and its XDR padding from the encoded XDR stream. XDR padding of 2176 a reduced data item is not transferred in an RPC-over-RDMA message. 2177 After reduction, the Payload stream contains fewer octets then the 2178 whole XDR stream did beforehand. XDR padding octets are often zero 2179 bytes, but they don't have to be. Thus reducing DDP-eligible items 2180 affects the result of message integrity verification or encryption. 2182 Therefore a sender MUST NOT reduce a Payload stream when RPCSEC_GSS 2183 integrity or encryption services are in use. Effectively, no data 2184 item is DDP-eligible in this situation, and Chunked Messages cannot 2185 be used. In this mode, an RPC-over-RDMA transport operates in the 2186 same manner as a transport that does not support direct data 2187 placement. 2189 When RPCSEC_GSS integrity or privacy is in use, a requester provides 2190 both a Read list and a Reply chunk in the same RPC-over-RDMA header 2191 to convey a Long call and provision a receptacle for a Long reply. 2193 9.2.2.4. Protecting RPC-Over-RDMA Transport Headers 2195 Like the base fields in an ONC RPC message (XID, call direction, and 2196 so on), the contents of an RPC-over-RDMA message's Transport stream 2197 are not protected by RPCSEC_GSS. This exposes XIDs, connection 2198 credit limits, and chunk lists (but not the content of the data items 2199 they refer to) to malicious behavior, which could redirect data that 2200 is transferred by the RPC-over-RDMA message, result in spurious 2201 retransmits, or trigger connection loss. 2203 In particular, if an attacker alters the information contained in the 2204 chunk lists of an RPC-over-RDMA header, data contained in those 2205 chunks can be redirected to other registered memory segments on 2206 requesters. An attacker might alter the arguments of RDMA Read and 2207 RDMA Write operations on the wire to similar effect. The use of 2208 RPCSEC_GSS integrity or privacy services enable the requester to 2209 detect if such tampering has been done and reject the RPC message. 2211 Encryption at lower layers, as described in Section 9.2.1, protects 2212 the content of the Transport stream. To address attacks on RDMA 2213 protocols themselves, RDMA transport implementations should conform 2214 to [RFC5042]. 2216 10. IANA Considerations 2218 Three assignments are specified by this document. These are 2219 unchanged from [RFC5666]: 2221 o A set of RPC "netids" for resolving RPC-over-RDMA services 2223 o Optional service port assignments for Upper Layer Bindings 2225 o An RPC program number assignment for the configuration protocol 2227 These assignments have been established, as below. 2229 The new RPC transport has been assigned an RPC "netid", which is an 2230 rpcbind [RFC1833] string used to describe the underlying protocol in 2231 order for RPC to select the appropriate transport framing, as well as 2232 the format of the service addresses and ports. 2234 The following "Netid" registry strings are defined for this purpose: 2236 NC_RDMA "rdma" 2237 NC_RDMA6 "rdma6" 2239 These netids MAY be used for any RDMA network satisfying the 2240 requirements of Section 3.2.2, and able to identify service endpoints 2241 using IP port addressing, possibly through use of a translation 2242 service as described above in Section 6. The "rdma" netid is to be 2243 used when IPv4 addressing is employed by the underlying transport, 2244 and "rdma6" for IPv6 addressing. 2246 The netid assignment policy and registry are defined in [RFC5665]. 2248 As a new RPC transport, this protocol has no effect on RPC Program 2249 numbers or existing registered port numbers. However, new port 2250 numbers MAY be registered for use by RPC-over-RDMA-enabled services, 2251 as appropriate to the new networks over which the services will 2252 operate. 2254 For example, the NFS/RDMA service defined in [RFC5667] has been 2255 assigned the port 20049, in the IANA registry: 2257 nfsrdma 20049/tcp Network File System (NFS) over RDMA 2258 nfsrdma 20049/udp Network File System (NFS) over RDMA 2259 nfsrdma 20049/sctp Network File System (NFS) over RDMA 2261 The RPC program number assignment policy and registry are defined in 2262 [RFC5531]. 2264 11. Acknowledgments 2266 The editor gratefully acknowledges the work of Brent Callaghan and 2267 Tom Talpey on the original RPC-over-RDMA Version One specification 2268 [RFC5666]. 2270 Dave Noveck provided excellent review, constructive suggestions, and 2271 consistent navigational guidance throughout the process of drafting 2272 this document. Dave also contributed much of the organization and 2273 content of Section 8 and helped the authors understand the 2274 complexities of XDR extensibility. 2276 The comments and contributions of Karen Deitke, Dai Ngo, Chunli 2277 Zhang, Dominique Martinet, and Mahesh Siddheshwar are accepted with 2278 great thanks. The editor also wishes to thank Bill Baker, Greg 2279 Marsden, and Matt Benjamin for their support of this work. 2281 The extract.sh shell script and formatting conventions were first 2282 described by the authors of the NFSv4.1 XDR specification [RFC5662]. 2284 Special thanks go to nfsv4 Working Group Chair Spencer Shepler and 2285 nfsv4 Working Group Secretary Thomas Haynes for their support. 2287 12. References 2289 12.1. Normative References 2291 [I-D.ietf-nfsv4-rpcsec-gssv3] 2292 Adamson, A. and N. Williams, "Remote Procedure Call (RPC) 2293 Security Version 3", draft-ietf-nfsv4-rpcsec-gssv3-17 2294 (work in progress), January 2016. 2296 [RFC1833] Srinivasan, R., "Binding Protocols for ONC RPC Version 2", 2297 RFC 1833, DOI 10.17487/RFC1833, August 1995, 2298 . 2300 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 2301 Requirement Levels", BCP 14, RFC 2119, 2302 DOI 10.17487/RFC2119, March 1997, 2303 . 2305 [RFC4506] Eisler, M., Ed., "XDR: External Data Representation 2306 Standard", STD 67, RFC 4506, DOI 10.17487/RFC4506, May 2307 2006, . 2309 [RFC5042] Pinkerton, J. and E. Deleganes, "Direct Data Placement 2310 Protocol (DDP) / Remote Direct Memory Access Protocol 2311 (RDMAP) Security", RFC 5042, DOI 10.17487/RFC5042, October 2312 2007, . 2314 [RFC5056] Williams, N., "On the Use of Channel Bindings to Secure 2315 Channels", RFC 5056, DOI 10.17487/RFC5056, November 2007, 2316 . 2318 [RFC5531] Thurlow, R., "RPC: Remote Procedure Call Protocol 2319 Specification Version 2", RFC 5531, DOI 10.17487/RFC5531, 2320 May 2009, . 2322 [RFC5660] Williams, N., "IPsec Channels: Connection Latching", 2323 RFC 5660, DOI 10.17487/RFC5660, October 2009, 2324 . 2326 [RFC5665] Eisler, M., "IANA Considerations for Remote Procedure Call 2327 (RPC) Network Identifiers and Universal Address Formats", 2328 RFC 5665, DOI 10.17487/RFC5665, January 2010, 2329 . 2331 12.2. Informative References 2333 [I-D.ietf-nfsv4-rpcrdma-bidirection] 2334 Lever, C., "Bi-directional Remote Procedure Call On RPC- 2335 over-RDMA Transports", draft-ietf-nfsv4-rpcrdma- 2336 bidirection-05 (work in progress), June 2016. 2338 [IB] InfiniBand Trade Association, "InfiniBand Architecture 2339 Specifications", . 2341 [IBPORT] InfiniBand Trade Association, "IP Addressing Annex", 2342 . 2344 [RFC0768] Postel, J., "User Datagram Protocol", STD 6, RFC 768, 2345 DOI 10.17487/RFC0768, August 1980, 2346 . 2348 [RFC0793] Postel, J., "Transmission Control Protocol", STD 7, 2349 RFC 793, DOI 10.17487/RFC0793, September 1981, 2350 . 2352 [RFC1094] Nowicki, B., "NFS: Network File System Protocol 2353 specification", RFC 1094, DOI 10.17487/RFC1094, March 2354 1989, . 2356 [RFC1813] Callaghan, B., Pawlowski, B., and P. Staubach, "NFS 2357 Version 3 Protocol Specification", RFC 1813, 2358 DOI 10.17487/RFC1813, June 1995, 2359 . 2361 [RFC5040] Recio, R., Metzler, B., Culley, P., Hilland, J., and D. 2362 Garcia, "A Remote Direct Memory Access Protocol 2363 Specification", RFC 5040, DOI 10.17487/RFC5040, October 2364 2007, . 2366 [RFC5041] Shah, H., Pinkerton, J., Recio, R., and P. Culley, "Direct 2367 Data Placement over Reliable Transports", RFC 5041, 2368 DOI 10.17487/RFC5041, October 2007, 2369 . 2371 [RFC5532] Talpey, T. and C. Juszczak, "Network File System (NFS) 2372 Remote Direct Memory Access (RDMA) Problem Statement", 2373 RFC 5532, DOI 10.17487/RFC5532, May 2009, 2374 . 2376 [RFC5661] Shepler, S., Ed., Eisler, M., Ed., and D. Noveck, Ed., 2377 "Network File System (NFS) Version 4 Minor Version 1 2378 Protocol", RFC 5661, DOI 10.17487/RFC5661, January 2010, 2379 . 2381 [RFC5662] Shepler, S., Ed., Eisler, M., Ed., and D. Noveck, Ed., 2382 "Network File System (NFS) Version 4 Minor Version 1 2383 External Data Representation Standard (XDR) Description", 2384 RFC 5662, DOI 10.17487/RFC5662, January 2010, 2385 . 2387 [RFC5666] Talpey, T. and B. Callaghan, "Remote Direct Memory Access 2388 Transport for Remote Procedure Call", RFC 5666, 2389 DOI 10.17487/RFC5666, January 2010, 2390 . 2392 [RFC5667] Talpey, T. and B. Callaghan, "Network File System (NFS) 2393 Direct Data Placement", RFC 5667, DOI 10.17487/RFC5667, 2394 January 2010, . 2396 [RFC7530] Haynes, T., Ed. and D. Noveck, Ed., "Network File System 2397 (NFS) Version 4 Protocol", RFC 7530, DOI 10.17487/RFC7530, 2398 March 2015, . 2400 Authors' Addresses 2402 Charles Lever (editor) 2403 Oracle Corporation 2404 1015 Granger Avenue 2405 Ann Arbor, MI 48104 2406 USA 2408 Phone: +1 734 274 2396 2409 Email: chuck.lever@oracle.com 2411 William Allen Simpson 2412 DayDreamer 2413 1384 Fontaine 2414 Madison Heights, MI 48071 2415 USA 2417 Email: william.allen.simpson@gmail.com 2419 Tom Talpey 2420 Microsoft Corp. 2421 One Microsoft Way 2422 Redmond, WA 98052 2423 USA 2425 Phone: +1 425 704-9945 2426 Email: ttalpey@microsoft.com