idnits 2.17.1 draft-ietf-nfsv4-rfc5664bis-02.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 == Line 1107 has weird spacing: '...stateid lor...' -- The document date (October 05, 2013) is 3856 days in the past. Is this intentional? -- Found something which looks like a code comment -- if you have code sections in the document, please surround them with '' and '' lines. Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) -- Possible downref: Non-RFC (?) normative reference: ref. '1' -- Possible downref: Non-RFC (?) normative reference: ref. '5' ** Obsolete normative reference: RFC 5661 (ref. '6') (Obsoleted by RFC 8881) ** Obsolete normative reference: RFC 3720 (ref. '8') (Obsoleted by RFC 7143) -- Possible downref: Non-RFC (?) normative reference: ref. '9' ** Obsolete normative reference: RFC 3980 (ref. '10') (Obsoleted by RFC 7143) -- Possible downref: Non-RFC (?) normative reference: ref. '11' -- Possible downref: Non-RFC (?) normative reference: ref. '13' Summary: 3 errors (**), 0 flaws (~~), 2 warnings (==), 7 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 NFSv4 B. Halevy 3 Internet-Draft PrimaryData 4 Intended status: Standards Track B. Harrosh 5 Expires: April 8, 2014 B. Welch 6 Panasas 7 October 05, 2013 9 Object-Based Parallel NFS (pNFS) Operations 10 draft-ietf-nfsv4-rfc5664bis-02 12 Abstract 14 Parallel NFS (pNFS) extends Network File System version 4 (NFSv4) to 15 allow clients to directly access file data on the storage used by the 16 NFSv4 server. This ability to bypass the server for data access can 17 increase both performance and parallelism, but requires additional 18 client functionality for data access, some of which is dependent on 19 the class of storage used, a.k.a. the Layout Type. The main pNFS 20 operations and data types in NFSv4 Minor version 1 specify a layout- 21 type-independent layer; layout-type-specific information is conveyed 22 using opaque data structures whose internal structure is further 23 defined by the particular layout type specification. This document 24 specifies the NFSv4.1 Object-Based pNFS Layout Type as a companion to 25 the main NFSv4 Minor version 1 specification. This document has been 26 updated since the initial version to clarify and fix some of the 27 RAID-related computations so they match current implementations. 29 Status of this Memo 31 This Internet-Draft is submitted in full conformance with the 32 provisions of BCP 78 and BCP 79. 34 Internet-Drafts are working documents of the Internet Engineering 35 Task Force (IETF). Note that other groups may also distribute 36 working documents as Internet-Drafts. The list of current Internet- 37 Drafts is at http://datatracker.ietf.org/drafts/current/. 39 Internet-Drafts are draft documents valid for a maximum of six months 40 and may be updated, replaced, or obsoleted by other documents at any 41 time. It is inappropriate to use Internet-Drafts as reference 42 material or to cite them other than as "work in progress." 44 This Internet-Draft will expire on April 8, 2014. 46 Copyright Notice 48 Copyright (c) 2013 IETF Trust and the persons identified as the 49 document authors. All rights reserved. 51 This document is subject to BCP 78 and the IETF Trust's Legal 52 Provisions Relating to IETF Documents 53 (http://trustee.ietf.org/license-info) in effect on the date of 54 publication of this document. Please review these documents 55 carefully, as they describe your rights and restrictions with respect 56 to this document. Code Components extracted from this document must 57 include Simplified BSD License text as described in Section 4.e of 58 the Trust Legal Provisions and are provided without warranty as 59 described in the Simplified BSD License. 61 Table of Contents 63 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4 64 1.1. Requirements Language . . . . . . . . . . . . . . . . . . 4 65 1.2. Overview of Changes . . . . . . . . . . . . . . . . . . . 4 66 2. XDR Description of the Objects-Based Layout Protocol . . . . . 4 67 2.1. Code Components Licensing Notice . . . . . . . . . . . . . 5 68 3. Basic Data Type Definitions . . . . . . . . . . . . . . . . . 6 69 3.1. pnfs_osd_objid4 . . . . . . . . . . . . . . . . . . . . . 6 70 3.2. pnfs_osd_version4 . . . . . . . . . . . . . . . . . . . . 7 71 3.3. pnfs_osd_object_cred4 . . . . . . . . . . . . . . . . . . 8 72 3.4. pnfs_osd_raid_algorithm4 . . . . . . . . . . . . . . . . . 9 73 4. Object Storage Device Addressing and Discovery . . . . . . . . 9 74 4.1. pnfs_osd_targetid_type4 . . . . . . . . . . . . . . . . . 10 75 4.2. pnfs_osd_deviceaddr4 . . . . . . . . . . . . . . . . . . . 11 76 4.2.1. SCSI Target Identifier . . . . . . . . . . . . . . . . 11 77 4.2.2. Device Network Address . . . . . . . . . . . . . . . . 12 78 5. Object-Based Layout . . . . . . . . . . . . . . . . . . . . . 13 79 5.1. pnfs_osd_data_map4 . . . . . . . . . . . . . . . . . . . . 13 80 5.2. pnfs_osd_layout4 . . . . . . . . . . . . . . . . . . . . . 14 81 5.3. Data Mapping Schemes . . . . . . . . . . . . . . . . . . . 15 82 5.3.1. Simple Striping . . . . . . . . . . . . . . . . . . . 15 83 5.3.2. Nested Striping . . . . . . . . . . . . . . . . . . . 16 84 5.3.3. Mirroring . . . . . . . . . . . . . . . . . . . . . . 18 85 5.4. RAID Algorithms . . . . . . . . . . . . . . . . . . . . . 19 86 5.4.1. PNFS_OSD_RAID_0 . . . . . . . . . . . . . . . . . . . 19 87 5.4.2. PNFS_OSD_RAID_4 . . . . . . . . . . . . . . . . . . . 19 88 5.4.3. PNFS_OSD_RAID_5 . . . . . . . . . . . . . . . . . . . 20 89 5.4.4. PNFS_OSD_RAID_PQ . . . . . . . . . . . . . . . . . . . 21 90 5.4.5. RAID Usage and Implementation Notes . . . . . . . . . 22 91 6. Object-Based Layout Update . . . . . . . . . . . . . . . . . . 22 92 6.1. pnfs_osd_deltaspaceused4 . . . . . . . . . . . . . . . . . 23 93 6.2. pnfs_osd_layoutupdate4 . . . . . . . . . . . . . . . . . . 23 94 7. Recovering from Client I/O Errors . . . . . . . . . . . . . . 24 95 8. Object-Based Layout Return . . . . . . . . . . . . . . . . . . 24 96 8.1. pnfs_osd_errno4 . . . . . . . . . . . . . . . . . . . . . 25 97 8.2. pnfs_osd_ioerr4 . . . . . . . . . . . . . . . . . . . . . 26 98 8.3. pnfs_osd_layoutreturn4 . . . . . . . . . . . . . . . . . . 27 99 9. Object-Based Creation Layout Hint . . . . . . . . . . . . . . 27 100 9.1. pnfs_osd_layouthint4 . . . . . . . . . . . . . . . . . . . 27 101 10. Layout Segments . . . . . . . . . . . . . . . . . . . . . . . 29 102 10.1. CB_LAYOUTRECALL and LAYOUTRETURN . . . . . . . . . . . . . 29 103 10.2. LAYOUTCOMMIT . . . . . . . . . . . . . . . . . . . . . . . 30 104 11. Recalling Layouts . . . . . . . . . . . . . . . . . . . . . . 30 105 11.1. CB_RECALL_ANY . . . . . . . . . . . . . . . . . . . . . . 30 106 12. Client Fencing . . . . . . . . . . . . . . . . . . . . . . . . 31 107 13. Security Considerations . . . . . . . . . . . . . . . . . . . 31 108 13.1. OSD Security Data Types . . . . . . . . . . . . . . . . . 32 109 13.2. The OSD Security Protocol . . . . . . . . . . . . . . . . 33 110 13.3. Protocol Privacy Requirements . . . . . . . . . . . . . . 34 111 13.4. Revoking Capabilities . . . . . . . . . . . . . . . . . . 34 112 14. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 35 113 15. References . . . . . . . . . . . . . . . . . . . . . . . . . . 35 114 15.1. Normative References . . . . . . . . . . . . . . . . . . . 35 115 15.2. Informative References . . . . . . . . . . . . . . . . . . 36 116 Appendix A. Acknowledgments . . . . . . . . . . . . . . . . . . . 37 117 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 37 119 1. Introduction 121 In pNFS, the file server returns typed layout structures that 122 describe where file data is located. There are different layouts for 123 different storage systems and methods of arranging data on storage 124 devices. This document describes the layouts used with object-based 125 storage devices (OSDs) that are accessed according to the OSD storage 126 protocol standard (ANSI INCITS 400-2004 [1]). 128 An "object" is a container for data and attributes, and files are 129 stored in one or more objects. The OSD protocol specifies several 130 operations on objects, including READ, WRITE, FLUSH, GET ATTRIBUTES, 131 SET ATTRIBUTES, CREATE, and DELETE. However, using the object-based 132 layout the client only uses the READ, WRITE, GET ATTRIBUTES, and 133 FLUSH commands. The other commands are only used by the pNFS server. 135 An object-based layout for pNFS includes object identifiers, 136 capabilities that allow clients to READ or WRITE those objects, and 137 various parameters that control how file data is striped across their 138 component objects. The OSD protocol has a capability-based security 139 scheme that allows the pNFS server to control what operations and 140 what objects can be used by clients. This scheme is described in 141 more detail in the "Security Considerations" section (Section 13). 143 1.1. Requirements Language 145 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 146 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 147 document are to be interpreted as described in RFC 2119 [2]. 149 1.2. Overview of Changes 151 This document is an update to the initial RFC. The primary area for 152 changes are the clarification and correction of the RAID-related 153 equations and algorithms in Section 5.3. The equations were restated 154 for clarity, and in a few places minor corrections were made to 155 ensure that this spec accurately matches current implementations. In 156 addition, minor corrections have been made to other sections. 158 2. XDR Description of the Objects-Based Layout Protocol 160 This document contains the external data representation (XDR [3]) 161 description of the NFSv4.1 objects layout protocol. The XDR 162 description is embedded in this document in a way that makes it 163 simple for the reader to extract into a ready-to-compile form. The 164 reader can feed this document into the following shell script to 165 produce the machine readable XDR description of the NFSv4.1 objects 166 layout protocol: 168 #!/bin/sh 169 grep '^ *///' $* | sed 's?^ */// ??' | sed 's?^ *///$??' 171 That is, if the above script is stored in a file called "extract.sh", 172 and this document is in a file called "spec.txt", then the reader can 173 do: 175 sh extract.sh < spec.txt > pnfs_osd_prot.x 177 The effect of the script is to remove leading white space from each 178 line, plus a sentinel sequence of "///". 180 The embedded XDR file header follows. Subsequent XDR descriptions, 181 with the sentinel sequence are embedded throughout the document. 183 Note that the XDR code contained in this document depends on types 184 from the NFSv4.1 nfs4_prot.x file ([4]). This includes both nfs 185 types that end with a 4, such as offset4, length4, etc., as well as 186 more generic types such as uint32_t and uint64_t. 188 2.1. Code Components Licensing Notice 190 The XDR description, marked with lines beginning with the sequence 191 "///", as well as scripts for extracting the XDR description are Code 192 Components as described in Section 4 of "Legal Provisions Relating to 193 IETF Documents" [5]. These Code Components are licensed according to 194 the terms of Section 4 of "Legal Provisions Relating to IETF 195 Documents". 197 /// /* 198 /// * Copyright (c) 2010 IETF Trust and the persons identified 199 /// * as authors of the code. All rights reserved. 200 /// * 201 /// * Redistribution and use in source and binary forms, with 202 /// * or without modification, are permitted provided that the 203 /// * following conditions are met: 204 /// * 205 /// * o Redistributions of source code must retain the above 206 /// * copyright notice, this list of conditions and the 207 /// * following disclaimer. 208 /// * 209 /// * o Redistributions in binary form must reproduce the above 210 /// * copyright notice, this list of conditions and the 211 /// * following disclaimer in the documentation and/or other 212 /// * materials provided with the distribution. 213 /// * 214 /// * o Neither the name of Internet Society, IETF or IETF 215 /// * Trust, nor the names of specific contributors, may be 216 /// * used to endorse or promote products derived from this 217 /// * software without specific prior written permission. 218 /// * 219 /// * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 220 /// * AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED 221 /// * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 222 /// * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 223 /// * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 224 /// * EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 225 /// * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 226 /// * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 227 /// * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 228 /// * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 229 /// * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 230 /// * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 231 /// * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 232 /// * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 233 /// * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 234 /// * 235 /// * This code was derived from draft-ietf-nfsv4-rfc5664bis-02. 236 [[RFC Editor: please insert RFC number if needed]] 237 /// * Please reproduce this note if possible. 238 /// */ 239 /// 240 /// /* 241 /// * pnfs_osd_prot.x 242 /// */ 243 /// 244 /// %#include 245 /// 247 3. Basic Data Type Definitions 249 The following sections define basic data types and constants used by 250 the Object-Based Layout protocol. 252 3.1. pnfs_osd_objid4 254 An object is identified by a number, somewhat like an inode number. 255 The object storage model has a two-level scheme, where the objects 256 within an object storage device are grouped into partitions. 258 /// struct pnfs_osd_objid4 { 259 /// deviceid4 oid_device_id; 260 /// uint64_t oid_partition_id; 261 /// uint64_t oid_object_id; 262 /// }; 263 /// 265 The pnfs_osd_objid4 type is used to identify an object within a 266 partition on a specified object storage device. "oid_device_id" 267 selects the object storage device from the set of available storage 268 devices. The device is identified with the deviceid4 type, which is 269 an index into addressing information about that device returned by 270 the GETDEVICELIST and GETDEVICEINFO operations. The deviceid4 data 271 type is defined in NFSv4.1 [6]. Within an OSD, a partition is 272 identified with a 64-bit number, "oid_partition_id". Within a 273 partition, an object is identified with a 64-bit number, 274 "oid_object_id". Creation and management of partitions is outside 275 the scope of this document, and is a facility provided by the object- 276 based storage file system. 278 3.2. pnfs_osd_version4 280 /// enum pnfs_osd_version4 { 281 /// PNFS_OSD_MISSING = 0, 282 /// PNFS_OSD_VERSION_1 = 1, 283 /// PNFS_OSD_VERSION_2 = 2 284 /// }; 285 /// 287 pnfs_osd_version4 is used to indicate the OSD protocol version used 288 to access an object, or whether an object is missing (i.e., 289 unavailable). Some of the RAID algorithms supported by object-based 290 layouts encode redundant information and can compensate for missing 291 components, but the data placement algorithms need to be aware of the 292 logical positions of the missing components. 294 The 1.0 version of the OSD standard has been ratified. The 2.0 295 version of the OSD standard has reached final draft status, but has 296 not been fully ratified. However, current object-based pNFS 297 implementations adhere to the OSD 2.0 protocol (SNIA T10/1729-D 298 [14]). The second generation OSD protocol has additional features to 299 support more robust error recovery, snapshots, and byte-range 300 capabilities. For completeness, and to allow for future revisions in 301 the OSD protocol, the OSD version is explicitly called out in the 302 information returned in the layout. (This information can also be 303 deduced by looking inside the capability type at the format field, 304 which is the first byte. The format value is 0x1 for an OSD v1 305 capability.) 307 3.3. pnfs_osd_object_cred4 309 /// enum pnfs_osd_cap_key_sec4 { 310 /// PNFS_OSD_CAP_KEY_SEC_NONE = 0, 311 /// PNFS_OSD_CAP_KEY_SEC_SSV = 1 312 /// }; 313 /// 314 /// struct pnfs_osd_object_cred4 { 315 /// pnfs_osd_objid4 oc_object_id; 316 /// pnfs_osd_version4 oc_osd_version; 317 /// pnfs_osd_cap_key_sec4 oc_cap_key_sec; 318 /// opaque oc_capability_key<>; 319 /// opaque oc_capability<>; 320 /// }; 321 /// 323 The pnfs_osd_object_cred4 structure is used to identify each 324 component comprising the file. The "oc_object_id" identifies the 325 component object, the "oc_osd_version" represents the osd protocol 326 version, or whether that component is unavailable, and the 327 "oc_capability" and "oc_capability_key", along with the 328 "oda_systemid" from the pnfs_osd_deviceaddr4, provide the OSD 329 security credentials needed to access that object. The 330 "oc_cap_key_sec" value denotes the method used to secure the 331 oc_capability_key (see Section 13.1 for more details). 333 To comply with the OSD security requirements, the capability key 334 SHOULD be transferred securely to prevent eavesdropping (see 335 Section 13). Therefore, a client SHOULD either issue the LAYOUTGET 336 or GETDEVICEINFO operations via RPCSEC_GSS with the privacy service 337 or previously establish a secret state verifier (SSV) for the 338 sessions via the NFSv4.1 SET_SSV operation. The 339 pnfs_osd_cap_key_sec4 type is used to identify the method used by the 340 server to secure the capability key. 342 o PNFS_OSD_CAP_KEY_SEC_NONE denotes that the oc_capability_key is 343 not encrypted, in which case the client SHOULD issue the LAYOUTGET 344 or GETDEVICEINFO operations with RPCSEC_GSS with the privacy 345 service or the NFSv4.1 transport should be secured by using 346 methods that are external to NFSv4.1 like the use of IPsec [15] 347 for transporting the NFSV4.1 protocol. 349 o PNFS_OSD_CAP_KEY_SEC_SSV denotes that the oc_capability_key 350 contents are encrypted using the SSV GSS context and the 351 capability key as inputs to the GSS_Wrap() function (see GSS-API 352 [7]) with the conf_req_flag set to TRUE. The client MUST use the 353 secret SSV key as part of the client's GSS context to decrypt the 354 capability key using the value of the oc_capability_key field as 355 the input_message to the GSS_unwrap() function. Note that to 356 prevent eavesdropping of the SSV key, the client SHOULD issue 357 SET_SSV via RPCSEC_GSS with the privacy service. 359 The actual method chosen depends on whether the client established a 360 SSV key with the server and whether it issued the operation with the 361 RPCSEC_GSS privacy method. Naturally, if the client did not 362 establish an SSV key via SET_SSV, the server MUST use the 363 PNFS_OSD_CAP_KEY_SEC_NONE method. Otherwise, if the operation was 364 not issued with the RPCSEC_GSS privacy method, the server SHOULD 365 secure the oc_capability_key with the PNFS_OSD_CAP_KEY_SEC_SSV 366 method. The server MAY use the PNFS_OSD_CAP_KEY_SEC_SSV method also 367 when the operation was issued with the RPCSEC_GSS privacy method. 369 3.4. pnfs_osd_raid_algorithm4 371 /// enum pnfs_osd_raid_algorithm4 { 372 /// PNFS_OSD_RAID_0 = 1, 373 /// PNFS_OSD_RAID_4 = 2, 374 /// PNFS_OSD_RAID_5 = 3, 375 /// PNFS_OSD_RAID_PQ = 4 /* Reed-Solomon P+Q */ 376 /// }; 377 /// 379 pnfs_osd_raid_algorithm4 represents the data redundancy algorithm 380 used to protect the file's contents. See Section 5.4 for more 381 details. 383 4. Object Storage Device Addressing and Discovery 385 Data operations to an OSD require the client to know the "address" of 386 each OSD's root object. The root object is synonymous with the Small 387 Computer System Interface (SCSI) logical unit. The client specifies 388 SCSI logical units to its SCSI protocol stack using a representation 389 local to the client. Because these representations are local, 390 GETDEVICEINFO must return information that can be used by the client 391 to select the correct local representation. 393 In the block world, a set offset (logical block number or track/ 394 sector) contains a disk label. This label identifies the disk 395 uniquely. In contrast, an OSD has a standard set of attributes on 396 its root object. For device identification purposes, the OSD System 397 ID (root information attribute number 3) and the OSD Name (root 398 information attribute number 9) are used as the label. These appear 399 in the pnfs_osd_deviceaddr4 type below under the "oda_systemid" and 400 "oda_osdname" fields. 402 In some situations, SCSI target discovery may need to be driven based 403 on information contained in the GETDEVICEINFO response. One example 404 of this is Internet SCSI (iSCSI) targets that are not known to the 405 client until a layout has been requested. The information provided 406 as the "oda_targetid", "oda_targetaddr", and "oda_lun" fields in the 407 pnfs_osd_deviceaddr4 type described below (see Section 4.2) allows 408 the client to probe a specific device given its network address and 409 optionally its iSCSI Name (see iSCSI [8]), or when the device network 410 address is omitted, allows it to discover the object storage device 411 using the provided device name or SCSI Device Identifier (see SPC-3 412 [9].) 414 The oda_systemid is implicitly used by the client, by using the 415 object credential signing key to sign each request with the request 416 integrity check value. This method protects the client from 417 unintentionally accessing a device if the device address mapping was 418 changed (or revoked). The server computes the capability key using 419 its own view of the systemid associated with the respective deviceid 420 present in the credential. If the client's view of the deviceid 421 mapping is stale, the client will use the wrong systemid (which must 422 be system-wide unique) and the I/O request to the OSD will fail to 423 pass the integrity check verification. 425 To recover from this condition the client should report the error and 426 return the layout using LAYOUTRETURN, and invalidate all the device 427 address mappings associated with this layout. The client can then 428 ask for a new layout if it wishes using LAYOUTGET and resolve the 429 referenced deviceids using GETDEVICEINFO or GETDEVICELIST. 431 The server MUST provide the oda_systemid and SHOULD also provide the 432 oda_osdname. When the OSD name is present, the client SHOULD get the 433 root information attributes whenever it establishes communication 434 with the OSD and verify that the OSD name it got from the OSD matches 435 the one sent by the metadata server. To do so, the client uses the 436 root_obj_cred credentials. 438 4.1. pnfs_osd_targetid_type4 440 The following enum specifies the manner in which a SCSI target can be 441 specified. The target can be specified as a SCSI Name, or as an SCSI 442 Device Identifier. 444 /// enum pnfs_osd_targetid_type4 { 445 /// OBJ_TARGET_ANON = 1, 446 /// OBJ_TARGET_SCSI_NAME = 2, 447 /// OBJ_TARGET_SCSI_DEVICE_ID = 3 448 /// }; 449 /// 451 4.2. pnfs_osd_deviceaddr4 453 The "pnfs_osd_deviceaddr4" data structure is returned by the server 454 as the storage-protocol-specific opaque field da_addr_body in the 455 "device_addr4" structure by a successful GETDEVICEINFO operation 456 NFSv4.1 [6]. 458 The specification for an object device address is as follows: 460 /// union pnfs_osd_targetid4 switch (pnfs_osd_targetid_type4 oti_type) { 461 /// case OBJ_TARGET_SCSI_NAME: 462 /// string oti_scsi_name<>; 463 /// 464 /// case OBJ_TARGET_SCSI_DEVICE_ID: 465 /// opaque oti_scsi_device_id<>; 466 /// 467 /// default: 468 /// void; 469 /// }; 470 /// 471 /// union pnfs_osd_targetaddr4 switch (bool ota_available) { 472 /// case TRUE: 473 /// netaddr4 ota_netaddr; 474 /// case FALSE: 475 /// void; 476 /// }; 477 /// 478 /// struct pnfs_osd_deviceaddr4 { 479 /// pnfs_osd_targetid4 oda_targetid; 480 /// pnfs_osd_targetaddr4 oda_targetaddr; 481 /// opaque oda_lun[8]; 482 /// opaque oda_systemid<>; 483 /// pnfs_osd_object_cred4 oda_root_obj_cred; 484 /// opaque oda_osdname<>; 485 /// }; 486 /// 488 4.2.1. SCSI Target Identifier 490 When "oda_targetid" is specified as an OBJ_TARGET_SCSI_NAME, the 491 "oti_scsi_name" string MUST be formatted as an "iSCSI Name" as 492 specified in iSCSI [8] and [10]. Note that the specification of the 493 oti_scsi_name string format is outside the scope of this document. 494 Parsing the string is based on the string prefix, e.g., "iqn.", 495 "eui.", or "naa." and more formats MAY be specified in the future in 496 accordance with iSCSI Names properties. 498 Currently, the iSCSI Name provides for naming the target device using 499 a string formatted as an iSCSI Qualified Name (IQN) or as an Extended 500 Unique Identifier (EUI) [11] string. Those are typically used to 501 identify iSCSI or Secure Routing Protocol (SRP) [16] devices. The 502 Network Address Authority (NAA) string format (see [10]) provides for 503 naming the device using globally unique identifiers, as defined in 504 Fibre Channel Framing and Signaling (FC-FS) [17]. These are 505 typically used to identify Fibre Channel or SAS [18] (Serial Attached 506 SCSI) devices. In particular, such devices that are dual-attached 507 both over Fibre Channel or SAS and over iSCSI. 509 When "oda_targetid" is specified as an OBJ_TARGET_SCSI_DEVICE_ID, the 510 "oti_scsi_device_id" opaque field MUST be formatted as a SCSI Device 511 Identifier as defined in SPC-3 [9] VPD Page 83h (Section 7.6.3. 512 "Device Identification VPD Page"). If the Device Identifier is 513 identical to the OSD System ID, as given by oda_systemid, the server 514 SHOULD provide a zero-length oti_scsi_device_id opaque value. Note 515 that similarly to the "oti_scsi_name", the specification of the 516 oti_scsi_device_id opaque contents is outside the scope of this 517 document and more formats MAY be specified in the future in 518 accordance with SPC-3. 520 The OBJ_TARGET_ANON pnfs_osd_targetid_type4 MAY be used for providing 521 no target identification. In this case, only the OSD System ID, and 522 optionally the provided network address, are used to locate the 523 device. 525 4.2.2. Device Network Address 527 The optional "oda_targetaddr" field MAY be provided by the server as 528 a hint to accelerate device discovery over, e.g., the iSCSI transport 529 protocol. The network address is given with the netaddr4 type, which 530 specifies a TCP/IP based endpoint (as specified in NFSv4.1 [6]). 531 When given, the client SHOULD use it to probe for the SCSI device at 532 the given network address. The client MAY still use other discovery 533 mechanisms such as Internet Storage Name Service (iSNS) [12] to 534 locate the device using the oda_targetid. In particular, such an 535 external name service SHOULD be used when the devices may be attached 536 to the network using multiple connections, and/or multiple storage 537 fabrics (e.g., Fibre-Channel and iSCSI). 539 The "oda_lun" field identifies the OSD 64-bit Logical Unit Number, 540 formatted in accordance with SAM-3 [13]. The client uses the Logical 541 Unit Number to communicate with the specific OSD Logical Unit. Its 542 use is defined in detail by the SCSI transport protocol, e.g., iSCSI 543 [8]. 545 5. Object-Based Layout 547 The layout4 type is defined in the NFSv4.1 [6] as follows: 549 enum layouttype4 { 550 LAYOUT4_NFSV4_1_FILES = 1, 551 LAYOUT4_OSD2_OBJECTS = 2, 552 LAYOUT4_BLOCK_VOLUME = 3 553 }; 555 struct layout_content4 { 556 layouttype4 loc_type; 557 opaque loc_body<>; 558 }; 560 struct layout4 { 561 offset4 lo_offset; 562 length4 lo_length; 563 layoutiomode4 lo_iomode; 564 layout_content4 lo_content; 565 }; 567 This document defines structure associated with the layouttype4 568 value, LAYOUT4_OSD2_OBJECTS. The NFSv4.1 [6] specifies the loc_body 569 structure as an XDR type "opaque". The opaque layout is 570 uninterpreted by the generic pNFS client layers, but obviously must 571 be interpreted by the object storage layout driver. This section 572 defines the structure of this opaque value, pnfs_osd_layout4. 574 5.1. pnfs_osd_data_map4 576 /// struct pnfs_osd_data_map4 { 577 /// uint32_t odm_num_comps; 578 /// length4 odm_stripe_unit; 579 /// uint32_t odm_group_width; 580 /// uint32_t odm_group_depth; 581 /// uint32_t odm_mirror_cnt; 582 /// pnfs_osd_raid_algorithm4 odm_raid_algorithm; 583 /// }; 584 /// 586 The pnfs_osd_data_map4 structure parameterizes the algorithm that 587 maps a file's contents over the component objects. Instead of 588 limiting the system to simple striping scheme where loss of a single 589 component object results in data loss, the map parameters support 590 mirroring and more complicated schemes that protect against loss of a 591 component object. 593 "odm_num_comps" is the number of component objects the file is 594 striped over. The server MAY grow the file by adding more components 595 to the stripe while clients hold valid layouts until the file has 596 reached its final stripe width. The file length in this case MUST be 597 limited to the number of bytes in a full stripe. 599 The "odm_stripe_unit" is the number of bytes placed on one component 600 before advancing to the next one in the list of components. The 601 number of bytes in a full stripe is odm_stripe_unit times the number 602 of components. In some RAID schemes, a stripe includes redundant 603 information (i.e., parity) that lets the system recover from loss or 604 damage to a component object. 606 The "odm_group_width" and "odm_group_depth" parameters allow a nested 607 striping pattern (see Section 5.3.2 for details). If there is no 608 nesting, then odm_group_width and odm_group_depth MUST be zero. The 609 size of the components array MUST be a multiple of odm_group_width. 611 The "odm_mirror_cnt" is used to replicate a file by replicating its 612 component objects. If there is no mirroring, then odm_mirror_cnt 613 MUST be 0. If odm_mirror_cnt is greater than zero, then the size of 614 the component array MUST be a multiple of (odm_mirror_cnt+1). 616 See Section 5.3 for more details. 618 5.2. pnfs_osd_layout4 620 /// struct pnfs_osd_layout4 { 621 /// pnfs_osd_data_map4 olo_map; 622 /// uint32_t olo_comps_index; 623 /// pnfs_osd_object_cred4 olo_components<>; 624 /// }; 625 /// 627 The pnfs_osd_layout4 structure specifies a layout over a set of 628 component objects. The "olo_components" field is an array of object 629 identifiers and security credentials that grant access to each 630 object. The organization of the data is defined by the 631 pnfs_osd_data_map4 type that specifies how the file's data is mapped 632 onto the component objects (i.e., the striping pattern). The data 633 placement algorithm that maps file data onto component objects 634 assumes that each component object occurs exactly once in the array 635 of components. Therefore, component objects MUST appear in the 636 olo_components array only once. The components array may represent 637 all objects comprising the file, in which case "olo_comps_index" is 638 set to zero and the number of entries in the olo_components array is 639 equal to olo_map.odm_num_comps. The server MAY return fewer 640 components than odm_num_comps, provided that the returned components 641 are sufficient to access any byte in the layout's data range (e.g., a 642 sub-stripe of "odm_group_width" components). In this case, 643 olo_comps_index represents the position of the returned components 644 array within the full array of components that comprise the file. 646 Note that the layout depends on the file size, which the client 647 learns from the generic return parameters of LAYOUTGET, by doing 648 GETATTR commands to the metadata server. The client uses the file 649 size to decide if it should fill holes with zeros or return a short 650 read. Striping patterns can cause cases where component objects are 651 shorter than other components because a hole happens to correspond to 652 the last part of the component object. 654 5.3. Data Mapping Schemes 656 This section describes the different data mapping schemes in detail. 657 The object layout always uses a "dense" layout as described in 658 NFSv4.1 [6]. This means that the second stripe unit of the file 659 starts at offset 0 of the second component, rather than at offset 660 stripe_unit bytes. After a full stripe has been written, the next 661 stripe unit is appended to the first component object in the list 662 without any holes in the component objects. 664 5.3.1. Simple Striping 666 The mapping from the logical offset within a file (L) to the 667 component object C and object-specific offset O is defined by the 668 following equations: 670 L: logical offset into the file 672 W: stripe width 673 W = size of olo_components array 675 S: number of bytes in a stripe 676 S = W * stripe_unit 678 N: stripe number 679 N = L / S 681 C: component index corresponding to L 682 C = (L % S) / stripe_unit 684 O: The component offset corresponding to L 685 O = (N * stripe_unit) + (L % stripe_unit) 687 Note that this computation does not accommodate the same object 688 appearing in the olo_components array multiple times. Therefore the 689 server may not return layouts with the same object appearing multiple 690 times. If needed the server can return multiple layout segments each 691 covering a single instance of the object. 693 For example, consider an object striped over four devices, . The stripe_unit is 4096 bytes. The stripe width S is thus 4 * 695 4096 = 16384. 697 Offset 0: 698 N = 0 / 16384 = 0 699 C = (0 % 16384) /4096 = 0 (D0) 700 O = 0*4096 + (0%4096) = 0 702 Offset 4096: 703 N = 4096 / 16384 = 0 704 C = (4096 % 16384) / 4096 = 1 (D1) 705 O = (0*4096)+(4096%4096) = 0 707 Offset 9000: 708 N = 9000 / 16384 = 0 709 C = (9000 % 16384) / 4096 = 2 (D2) 710 O = (0*4096)+(9000%4096) = 808 712 Offset 132000: 713 N = 132000 / 16384 = 8 714 C = (132000 % 16384) / 4096 = 0 (D0) 715 O = (8*4096) + (132000%4096) = 33696 717 5.3.2. Nested Striping 719 The odm_group_width and odm_group_depth parameters allow a nested 720 striping pattern. odm_group_width defines the width of a data stripe 721 and odm_group_depth defines how many stripes are written before 722 advancing to the next group of components in the list of component 723 objects for the file. The math used to map from a file offset to a 724 component object and offset within that object is shown below. The 725 computations map from the logical offset L to the component index C 726 and offset relative O within that component object. 728 L: logical offset into the file 730 FW: total number of components 731 FW = size of olo_components array 733 W: stripe width 734 W = group_width, if not zero, else FW 736 group_count: number of groups 737 group_count = FW / group_width, if group_width is not zero, else 1 739 D: number of data devices in a stripe 740 D = W 742 U: number of data bytes in a stripe within a group 743 U = D * stripe_unit 745 T: number of bytes striped within a group of component objects 746 (before advancing to the next group) 747 T = U * group_depth 749 S: number of bytes striped across all component objects 750 (before the pattern repeats) 751 S = T * group_count 753 M: The "major" (i.e., across all components) cycle number 754 M = L / S 756 G: group number from the beginning of the major cycle 757 G = (L % S) / T 759 H: byte offset within the last group 760 H = (L % S) % T 762 N: The "minor" (i.e., across the group) stripe number 763 N = H / U 765 C: component index corresponding to L 766 C = (G * D) + ((H % U) / stripe_unit) 768 O: The component offset corresponding to L 769 O = (M * group_depth * stripe_unit) + (N * stripe_unit) + 770 (L % stripe_unit) 772 For example, consider an object striped over 100 devices with a 773 group_width of 10, a group_depth of 50, and a stripe_unit of 1 MB. 774 In this scheme, 500 MB are written to the first 10 components, and 775 5000 MB are written before the pattern wraps back around to the first 776 component in the array. 778 Offset 0: 779 W = 100 780 group_count = 100 / 10 = 10 781 D = 10 782 U = 1 MB * 10 = 10 MB 783 T = 10 MB * 50 = 500 MB 784 S = 500 MB * 10 = 5000 MB 785 M = 0 / 5000 MB = 0 786 G = (0 % 5000 MB) / 500 MB = 0 787 H = (0 % 5000 MB) % 500 MB = 0 788 N = 0 / 10 MB = 0 789 C = (0 * 10) + ((0 % 10 MB) / 1 MB) = 0 790 O = (0 * 50 * 1 MB) + (0 * 1 MB) + (0 % 1 MB) = 0 792 Offset 27 MB: 793 M = 27 MB / 5000 MB = 0 794 G = (27 MB % 5000 MB) / 500 MB = 0 795 H = (27 MB % 5000 MB) % 500 MB = 27 MB 796 N = 27 MB / 10 MB = 2 797 C = (0 * 10) + ((27 MB % 10 MB) / 1 MB) = 7 798 O = (0 * 50 * 1 MB) + (2 * 1 MB) + (27 MB % 1 MB) = 2 MB 800 Offset 7232 MB: 801 M = 7232 MB / 5000 MB = 1 802 G = (7232 MB % 5000 MB) / 500 MB = 4 803 H = (7232 MB % 5000 MB) % 500 MB = 232 MB 804 N = 232 MB / 10 MB = 23 805 C = (4 * 10) + ((232 MB % 10 MB) / 1 MB) = 42 806 O = (1 * 50 * 1 MB) + (23 * 1 MB) + (7232 MB % 1 MB) = 73 MB 808 5.3.3. Mirroring 810 The odm_mirror_cnt is used to replicate a file by replicating its 811 component objects. If there is no mirroring, then odm_mirror_cnt 812 MUST be 0. If odm_mirror_cnt is greater than zero, then the size of 813 the olo_components array MUST be a multiple of (odm_mirror_cnt+1). 814 Thus, for a classic mirror on two objects, odm_mirror_cnt is one. 815 Note that mirroring can be defined over any RAID algorithm and 816 striping pattern (either simple or nested). If odm_group_width is 817 also non-zero, then the size of the olo_components array MUST be a 818 multiple of odm_group_width * (odm_mirror_cnt+1). Note that 819 odm_group_width does not account for mirrors. Replicas are adjacent 820 in the olo_components array, and the value C produced by the above 821 equations is not a direct index into the olo_components array. 822 Instead, the following equations determine the replica component 823 index RCi, where i ranges from 0 to odm_mirror_cnt. 825 FW = size of olo_components array / (odm_mirror_cnt+1) 827 C = component index for striping or two-level striping 828 as calculated using above equations 830 i ranges from 0 to odm_mirror_cnt, inclusive 831 RCi = C * (odm_mirror_cnt+1) + i 833 5.4. RAID Algorithms 835 pnfs_osd_raid_algorithm4 determines the algorithm and placement of 836 redundant data. This section defines the different redundancy 837 algorithms. Note: The term "RAID" (Redundant Array of Independent 838 Disks) is used in this document to represent an array of component 839 objects that store data for an individual file. The objects are 840 stored on independent object-based storage devices. File data is 841 encoded and striped across the array of component objects using 842 algorithms developed for block-based RAID systems. 844 The use of per-file RAID encoding in the object-layout for pNFS 845 imposes an additional responsibility on the file system client. The 846 pNFS client SHOULD generate the redundant data and write it do 847 storage along with the file data according to the RAID parameters 848 returned in the layout. However, various error conditions may 849 prevent the client from meeting its obligations, and this is 850 supported by the error information in the pnfs_osd_ioerr4 structure 851 (see Section 8.1). An explicit error return from the client, or an 852 implicit error caused by a client's failure to return a layout MUST 853 trigger recovery action by the server to prevent access to invalid 854 data (see Section 7). It is the server's responsibility to only 855 grant layout information to files that can be safely accessed, and to 856 deny access to files that are in an inconsistent state. 858 5.4.1. PNFS_OSD_RAID_0 860 PNFS_OSD_RAID_0 means there is no parity data, so all bytes in the 861 component objects are data bytes located by the above equations for C 862 and O. If a component object is marked as PNFS_OSD_MISSING, an I/O 863 error MUST be returned if this component is accessed. In this case, 864 the generic NFS client layer MAY elect to retry this operation 865 against the pNFS server. 867 5.4.2. PNFS_OSD_RAID_4 869 PNFS_OSD_RAID_4 means that the last component object, or the last in 870 each group (if odm_group_width is greater than zero), contains parity 871 information computed over the rest of the stripe with an XOR 872 operation. If a component object is unavailable, the client can read 873 the rest of the stripe units in the damaged stripe and recompute the 874 missing stripe unit by XORing the other stripe units in the stripe. 875 Or the client can replay the READ against the pNFS server that will 876 presumably perform the reconstructed read on the client's behalf. 878 When parity is present in the file, then the number of parity devices 879 is taken into account in the above equations when calculating (D), 880 the number of data devices in a stripe, as follows: 882 P: number of parity devices in each stripe 883 P = 1 885 D: number of data devices in a stripe 886 D = W - P 888 I: parity device index 889 I = D 891 5.4.3. PNFS_OSD_RAID_5 893 PNFS_OSD_RAID_5 means that the position of the parity data is rotated 894 on each stripe or each group (if odm_group_width is greater than 895 zero). In the first stripe, the last component holds the parity. In 896 the second stripe, the next-to-last component holds the parity, and 897 so on. In this scheme, all stripe units are rotated so that I/O is 898 evenly spread across objects as the file is read sequentially. The 899 rotated parity layout is illustrated here, with hexadecimal numbers 900 indicating the stripe unit. 902 0 1 2 P 903 4 5 P 3 904 8 P 6 7 905 P 9 a b 907 Note that the math for RAID_5 is similar to RAID_4 only that the 908 device indices for each stripe are rotated backwards. So start with 909 the equations above for RAID_4, then compute the rotation as 910 described below. Also note that the parity rotation cycle always 911 starts on group boundaries so the first stripe in a group has its 912 parity at device D. 914 P: number of parity devices in each stripe 915 P = 1 917 PC: Parity Cycle 918 PC = W 920 R: The parity rotation index 921 (N is as computed in above equations for RAID-4) 922 R = N % PC 924 I: parity device index 925 I = (W + W - (R + 1) * P) % W 927 Cr: The rotated device index 928 (C is as computed in the above equations for RAID-4) 929 Cr = (W + C - (R * P)) % W 931 Note: W is added above to avoid negative numbers modulo math. 933 5.4.4. PNFS_OSD_RAID_PQ 935 PNFS_OSD_RAID_PQ is a double-parity scheme that uses the Reed-Solomon 936 P+Q encoding scheme [19]. In this layout, the last two component 937 objects hold the P and Q data, respectively. P is parity computed 938 with XOR. The Q computation is described in detail by Anvin [20]. 939 The same polynomial "x^8+x^4+x^3+x^2+1" and Galois field size of 2^8 940 are used here. Clients may simply choose to read data through the 941 metadata server if two or more components are missing or damaged. 943 The equations given above for embedded parity can be used to map a 944 file offset to the correct component object by setting the number of 945 parity components (P) to 2 instead of 1 for RAID-5 and computing the 946 Parity Cycle length as the Lowest Common Multiple [21] of 947 odm_group_width and P, devided by P, as described below. Note: This 948 algorithm can be used also for RAID-5 where P=1. 950 P: number of parity devices 951 P = 2 953 PC: Parity cycle: 954 PC = LCM(W, P) / P 956 Q: The device index holding the Q component 957 (I is as computed in the above equations for RAID-5) 958 Qdev = (I + 1) % W 960 5.4.5. RAID Usage and Implementation Notes 962 RAID layouts with redundant data in their stripes require additional 963 serialization of updates to ensure correct operation. Otherwise, if 964 two clients simultaneously write to the same logical range of an 965 object, the result could include different data in the same ranges of 966 mirrored tuples, or corrupt parity information. It is the 967 responsibility of the metadata server to enforce serialization 968 requirements. Serialization MUST occur at the RAID stripe boundary 969 for write operations to avoid corrupting parity by concurrent updates 970 to the same stripe. Mirrors do not have explicit stripe boundaries, 971 so it is sufficient to serialize writes to the same byte ranges. 973 Many alternative encoding schemes exist for P>=2 [22]. These involve 974 P or Q equations different than the Reed-Solomon encoding used in 975 PNFS_OSD_RAID_PQ. Thus, if one of these schemes is to be used in the 976 future, a distinct value must be added to pnfs_osd_raid_algorithm4 977 for it. 979 6. Object-Based Layout Update 981 layoutupdate4 is used in the LAYOUTCOMMIT operation to convey updates 982 to the layout and additional information to the metadata server. It 983 is defined in the NFSv4.1 [6] as follows: 985 struct layoutupdate4 { 986 layouttype4 lou_type; 987 opaque lou_body<>; 988 }; 990 The layoutupdate4 type is an opaque value at the generic pNFS client 991 level. If the lou_type layout type is LAYOUT4_OSD2_OBJECTS, then the 992 lou_body opaque value is defined by the pnfs_osd_layoutupdate4 type. 994 Object-Based pNFS clients are not allowed to modify the layout. 995 Therefore, the information passed in pnfs_osd_layoutupdate4 is used 996 only to update the file's attributes. In addition to the generic 997 information the client can pass to the metadata server in 998 LAYOUTCOMMIT such as the highest offset the client wrote to and the 999 last time it modified the file, the client MAY use 1000 pnfs_osd_layoutupdate4 to convey the capacity consumed (or released) 1001 by writes using the layout, and to indicate that I/O errors were 1002 encountered by such writes. 1004 6.1. pnfs_osd_deltaspaceused4 1006 /// union pnfs_osd_deltaspaceused4 switch (bool dsu_valid) { 1007 /// case TRUE: 1008 /// int64_t dsu_delta; 1009 /// case FALSE: 1010 /// void; 1011 /// }; 1012 /// 1014 pnfs_osd_deltaspaceused4 is used to convey space utilization 1015 information at the time of LAYOUTCOMMIT. For the file system to 1016 properly maintain capacity-used information, it needs to track how 1017 much capacity was consumed by WRITE operations performed by the 1018 client. In this protocol, the OSD returns the capacity consumed by a 1019 write (*), which can be different than the number of bytes written 1020 because of internal overhead like block-level allocation and indirect 1021 blocks, and the client reflects this back to the pNFS server so it 1022 can accurately track quota. The pNFS server can choose to trust this 1023 information coming from the clients and therefore avoid querying the 1024 OSDs at the time of LAYOUTCOMMIT. If the client is unable to obtain 1025 this information from the OSD, it simply returns invalid 1026 olu_delta_space_used. 1028 6.2. pnfs_osd_layoutupdate4 1030 /// struct pnfs_osd_layoutupdate4 { 1031 /// pnfs_osd_deltaspaceused4 olu_delta_space_used; 1032 /// bool olu_ioerr_flag; 1033 /// }; 1034 /// 1036 "olu_delta_space_used" is used to convey capacity usage information 1037 back to the metadata server. 1039 The "olu_ioerr_flag" is used when I/O errors were encountered while 1040 writing the file. The client MUST report the errors using the 1041 pnfs_osd_ioerr4 structure (see Section 8.1) at LAYOUTRETURN time. 1043 If the client updated the file successfully before hitting the I/O 1044 errors, it MAY use LAYOUTCOMMIT to update the metadata server as 1045 described above. Typically, in the error-free case, the server MAY 1046 turn around and update the file's attributes on the storage devices. 1047 However, if I/O errors were encountered, the server better not 1048 attempt to write the new attributes on the storage devices until it 1049 receives the I/O error report; therefore, the client MUST set the 1050 olu_ioerr_flag to true. Note that in this case, the client SHOULD 1051 send both the LAYOUTCOMMIT and LAYOUTRETURN operations in the same 1052 COMPOUND RPC. 1054 7. Recovering from Client I/O Errors 1056 The pNFS client may encounter errors when directly accessing the 1057 object storage devices. A well behaved client will report any such 1058 errors promptly by executing a LAYOUTRETURN. When the 1059 LAYOUT4_OSD2_OBJECTS layout type is used, the client MUST report the 1060 I/O errors to the server at LAYOUTRETURN time using the 1061 pnfs_osd_ioerr4 structure (see Section 8.1). 1063 It is the responsibility of the metadata server to handle the I/O 1064 errors. The server MUST analyze the error and perform the required 1065 recovery operations such as repairing any parity inconsistencies, 1066 recovering media failures, or reconstructing missing objects. 1068 The metadata server SHOULD recall any outstanding layouts to allow it 1069 exclusive write access to the stripes being recovered and to prevent 1070 other clients from hitting the same error condition. In these cases, 1071 the server MUST complete recovery before handing out any new layouts 1072 to the affected byte ranges. 1074 The client SHOULD attempt to compensate for the error before giving 1075 up and reflecting an error to the application. The first step in 1076 error recovery is to return the layout with LAYOUTRETURN and the 1077 associated error information. The second step is to request a new 1078 layout using LAYOUTGET and then retry the I/O operation with the new 1079 layout. Finally, if the error persists, the client may choose to 1080 retry the I/O operation using regular NFS READ or WRITE operations 1081 via the metadata server. 1083 8. Object-Based Layout Return 1085 layoutreturn_file4 is used in the LAYOUTRETURN operation to convey 1086 layout-type specific information to the server. It is defined in the 1087 NFSv4.1 [6] as follows: 1089 struct layoutreturn_file4 { 1090 offset4 lrf_offset; 1091 length4 lrf_length; 1092 stateid4 lrf_stateid; 1093 /* layouttype4 specific data */ 1094 opaque lrf_body<>; 1095 }; 1097 union layoutreturn4 switch(layoutreturn_type4 lr_returntype) { 1098 case LAYOUTRETURN4_FILE: 1099 layoutreturn_file4 lr_layout; 1100 default: 1101 void; 1102 }; 1104 struct LAYOUTRETURN4args { 1105 /* CURRENT_FH: file */ 1106 bool lora_reclaim; 1107 layoutreturn_stateid lora_recallstateid; 1108 layouttype4 lora_layout_type; 1109 layoutiomode4 lora_iomode; 1110 layoutreturn4 lora_layoutreturn; 1111 }; 1113 If the lora_layout_type layout type is LAYOUT4_OSD2_OBJECTS, then the 1114 lrf_body opaque value is defined by the pnfs_osd_layoutreturn4 type. 1116 The pnfs_osd_layoutreturn4 type allows the client to report I/O error 1117 information back to the metadata server as defined below. 1119 8.1. pnfs_osd_errno4 1121 /// enum pnfs_osd_errno4 { 1122 /// PNFS_OSD_ERR_EIO = 1, 1123 /// PNFS_OSD_ERR_NOT_FOUND = 2, 1124 /// PNFS_OSD_ERR_NO_SPACE = 3, 1125 /// PNFS_OSD_ERR_BAD_CRED = 4, 1126 /// PNFS_OSD_ERR_NO_ACCESS = 5, 1127 /// PNFS_OSD_ERR_UNREACHABLE = 6, 1128 /// PNFS_OSD_ERR_RESOURCE = 7 1129 /// }; 1130 /// 1132 pnfs_osd_errno4 is used to represent error types when read/write 1133 errors are reported to the metadata server. The error codes serve as 1134 hints to the metadata server that may help it in diagnosing the exact 1135 reason for the error and in repairing it. 1137 o PNFS_OSD_ERR_EIO indicates the operation failed because the object 1138 storage device experienced a failure trying to access the object. 1139 The most common source of these errors is media errors, but other 1140 internal errors might cause this as well. In this case, the 1141 metadata server should go examine the broken object more closely; 1142 hence, it should be used as the default error code. 1144 o PNFS_OSD_ERR_NOT_FOUND indicates the object ID specifies an object 1145 that does not exist on the object storage device. 1147 o PNFS_OSD_ERR_NO_SPACE indicates the operation failed because the 1148 object storage device ran out of free capacity during the 1149 operation. 1151 o PNFS_OSD_ERR_BAD_CRED indicates the security parameters are not 1152 valid. The primary cause of this is that the capability has 1153 expired, or the access policy tag (a.k.a., capability version 1154 number) has been changed to revoke capabilities. The client will 1155 need to return the layout and get a new one with fresh 1156 capabilities. 1158 o PNFS_OSD_ERR_NO_ACCESS indicates the capability does not allow the 1159 requested operation. This should not occur in normal operation 1160 because the metadata server should give out correct capabilities, 1161 or none at all. 1163 o PNFS_OSD_ERR_UNREACHABLE indicates the client did not complete the 1164 I/O operation at the object storage device due to a communication 1165 failure. Whether or not the I/O operation was executed by the OSD 1166 is undetermined. 1168 o PNFS_OSD_ERR_RESOURCE indicates the client did not issue the I/O 1169 operation due to a local problem on the initiator (i.e., client) 1170 side, e.g., when running out of memory. The client MUST guarantee 1171 that the OSD command was never dispatched to the OSD. 1173 8.2. pnfs_osd_ioerr4 1175 /// struct pnfs_osd_ioerr4 { 1176 /// pnfs_osd_objid4 oer_component; 1177 /// length4 oer_comp_offset; 1178 /// length4 oer_comp_length; 1179 /// bool oer_iswrite; 1180 /// pnfs_osd_errno4 oer_errno; 1181 /// }; 1182 /// 1184 The pnfs_osd_ioerr4 structure is used to return error indications for 1185 objects that generated errors during data transfers. These are hints 1186 to the metadata server that there are problems with that object. For 1187 each error, "oer_component", "oer_comp_offset", and "oer_comp_length" 1188 represent the object and byte range within the component object in 1189 which the error occurred; "oer_iswrite" is set to "true" if the 1190 failed OSD operation was data modifying, and "oer_errno" represents 1191 the type of error. 1193 Component byte ranges in the optional pnfs_osd_ioerr4 structure are 1194 used for recovering the object and MUST be set by the client to cover 1195 all failed I/O operations to the component. 1197 8.3. pnfs_osd_layoutreturn4 1199 /// struct pnfs_osd_layoutreturn4 { 1200 /// pnfs_osd_ioerr4 olr_ioerr_report<>; 1201 /// }; 1202 /// 1204 When OSD I/O operations failed, "olr_ioerr_report<>" is used to 1205 report these errors to the metadata server as an array of elements of 1206 type pnfs_osd_ioerr4. Each element in the array represents an error 1207 that occurred on the object specified by oer_component. If no errors 1208 are to be reported, the size of the olr_ioerr_report<> array is set 1209 to zero. 1211 9. Object-Based Creation Layout Hint 1213 The layouthint4 type is defined in the NFSv4.1 [6] as follows: 1215 struct layouthint4 { 1216 layouttype4 loh_type; 1217 opaque loh_body<>; 1218 }; 1220 The layouthint4 structure is used by the client to pass a hint about 1221 the type of layout it would like created for a particular file. If 1222 the loh_type layout type is LAYOUT4_OSD2_OBJECTS, then the loh_body 1223 opaque value is defined by the pnfs_osd_layouthint4 type. 1225 9.1. pnfs_osd_layouthint4 1227 /// union pnfs_osd_max_comps_hint4 switch (bool omx_valid) { 1228 /// case TRUE: 1229 /// uint32_t omx_max_comps; 1230 /// case FALSE: 1231 /// void; 1232 /// }; 1233 /// 1234 /// union pnfs_osd_stripe_unit_hint4 switch (bool osu_valid) { 1235 /// case TRUE: 1236 /// length4 osu_stripe_unit; 1237 /// case FALSE: 1238 /// void; 1239 /// }; 1240 /// 1241 /// union pnfs_osd_group_width_hint4 switch (bool ogw_valid) { 1242 /// case TRUE: 1243 /// uint32_t ogw_group_width; 1244 /// case FALSE: 1245 /// void; 1246 /// }; 1247 /// 1248 /// union pnfs_osd_group_depth_hint4 switch (bool ogd_valid) { 1249 /// case TRUE: 1250 /// uint32_t ogd_group_depth; 1251 /// case FALSE: 1252 /// void; 1253 /// }; 1254 /// 1255 /// union pnfs_osd_mirror_cnt_hint4 switch (bool omc_valid) { 1256 /// case TRUE: 1257 /// uint32_t omc_mirror_cnt; 1258 /// case FALSE: 1259 /// void; 1260 /// }; 1261 /// 1262 /// union pnfs_osd_raid_algorithm_hint4 switch (bool ora_valid) { 1263 /// case TRUE: 1264 /// pnfs_osd_raid_algorithm4 ora_raid_algorithm; 1265 /// case FALSE: 1266 /// void; 1267 /// }; 1268 /// 1269 /// struct pnfs_osd_layouthint4 { 1270 /// pnfs_osd_max_comps_hint4 olh_max_comps_hint; 1271 /// pnfs_osd_stripe_unit_hint4 olh_stripe_unit_hint; 1272 /// pnfs_osd_group_width_hint4 olh_group_width_hint; 1273 /// pnfs_osd_group_depth_hint4 olh_group_depth_hint; 1274 /// pnfs_osd_mirror_cnt_hint4 olh_mirror_cnt_hint; 1275 /// pnfs_osd_raid_algorithm_hint4 olh_raid_algorithm_hint; 1276 /// }; 1277 /// 1279 This type conveys hints for the desired data map. All parameters are 1280 optional so the client can give values for only the parameters it 1281 cares about, e.g. it can provide a hint for the desired number of 1282 mirrored components, regardless of the RAID algorithm selected for 1283 the file. The server should make an attempt to honor the hints, but 1284 it can ignore any or all of them at its own discretion and without 1285 failing the respective CREATE operation. 1287 The "olh_max_comps_hint" can be used to limit the total number of 1288 component objects comprising the file. All other hints correspond 1289 directly to the different fields of pnfs_osd_data_map4. 1291 10. Layout Segments 1293 The pnfs layout operations operate on logical byte ranges. There is 1294 no requirement in the protocol for any relationship between byte 1295 ranges used in LAYOUTGET to acquire layouts and byte ranges used in 1296 CB_LAYOUTRECALL, LAYOUTCOMMIT, or LAYOUTRETURN. However, using OSD 1297 byte-range capabilities poses limitations on these operations since 1298 the capabilities associated with layout segments cannot be merged or 1299 split. The following guidelines should be followed for proper 1300 operation of object-based layouts. 1302 10.1. CB_LAYOUTRECALL and LAYOUTRETURN 1304 In general, the object-based layout driver should keep track of each 1305 layout segment it got, keeping record of the segment's iomode, 1306 offset, and length. The server should allow the client to get 1307 multiple overlapping layout segments but is free to recall the layout 1308 to prevent overlap. 1310 In response to CB_LAYOUTRECALL, the client should return all layout 1311 segments matching the given iomode and overlapping with the recalled 1312 range. When returning the layouts for this byte range with 1313 LAYOUTRETURN, the client MUST NOT return a sub-range of a layout 1314 segment it has; each LAYOUTRETURN sent MUST completely cover at least 1315 one outstanding layout segment. 1317 The server, in turn, should release any segment that exactly matches 1318 the clientid, iomode, and byte range given in LAYOUTRETURN. If no 1319 exact match is found, then the server should release all layout 1320 segments matching the clientid and iomode and that are fully 1321 contained in the returned byte range. If none are found and the byte 1322 range is a subset of an outstanding layout segment with for the same 1323 clientid and iomode, then the client can be considered malfunctioning 1324 and the server SHOULD recall all layouts from this client to reset 1325 its state. If this behavior repeats, the server SHOULD deny all 1326 LAYOUTGETs from this client. 1328 10.2. LAYOUTCOMMIT 1330 LAYOUTCOMMIT is only used by object-based pNFS to convey modified 1331 attributes hints and/or to report the presence of I/O errors to the 1332 metadata server (MDS). Therefore, the offset and length in 1333 LAYOUTCOMMIT4args are reserved for future use and should be set to 0. 1335 11. Recalling Layouts 1337 The object-based metadata server should recall outstanding layouts in 1338 the following cases: 1340 o When the file's security policy changes, i.e., Access Control 1341 Lists (ACLs) or permission mode bits are set. 1343 o When the file's aggregation map changes, rendering outstanding 1344 layouts invalid. 1346 o When there are sharing conflicts. For example, the server will 1347 issue stripe-aligned layout segments for RAID-5 objects. To 1348 prevent corruption of the file's parity, multiple clients must not 1349 hold valid write layouts for the same stripes. An outstanding 1350 READ/WRITE (RW) layout should be recalled when a conflicting 1351 LAYOUTGET is received from a different client for LAYOUTIOMODE4_RW 1352 and for a byte range overlapping with the outstanding layout 1353 segment. 1355 11.1. CB_RECALL_ANY 1357 The metadata server can use the CB_RECALL_ANY callback operation to 1358 notify the client to return some or all of its layouts. The NFSv4.1 1359 [6] defines the following types: 1361 const RCA4_TYPE_MASK_OBJ_LAYOUT_MIN = 8; 1362 const RCA4_TYPE_MASK_OBJ_LAYOUT_MAX = 9; 1364 struct CB_RECALL_ANY4args { 1365 uint32_t craa_objects_to_keep; 1366 bitmap4 craa_type_mask; 1367 }; 1369 Typically, CB_RECALL_ANY will be used to recall client state when the 1370 server needs to reclaim resources. The craa_type_mask bitmap 1371 specifies the type of resources that are recalled and the 1372 craa_objects_to_keep value specifies how many of the recalled objects 1373 the client is allowed to keep. The object-based layout type mask 1374 flags are defined as follows. They represent the iomode of the 1375 recalled layouts. In response, the client SHOULD return layouts of 1376 the recalled iomode that it needs the least, keeping at most 1377 craa_objects_to_keep object-based layouts. 1379 /// enum pnfs_osd_cb_recall_any_mask { 1380 /// PNFS_OSD_RCA4_TYPE_MASK_READ = 8, 1381 /// PNFS_OSD_RCA4_TYPE_MASK_RW = 9 1382 /// }; 1383 /// 1385 The PNFS_OSD_RCA4_TYPE_MASK_READ flag notifies the client to return 1386 layouts of iomode LAYOUTIOMODE4_READ. Similarly, the 1387 PNFS_OSD_RCA4_TYPE_MASK_RW flag notifies the client to return layouts 1388 of iomode LAYOUTIOMODE4_RW. When both mask flags are set, the client 1389 is notified to return layouts of either iomode. 1391 12. Client Fencing 1393 In cases where clients are uncommunicative and their lease has 1394 expired or when clients fail to return recalled layouts within a 1395 lease period at the least (see "Recalling a Layout"[6]), the server 1396 MAY revoke client layouts and/or device address mappings and reassign 1397 these resources to other clients. To avoid data corruption, the 1398 metadata server MUST fence off the revoked clients from the 1399 respective objects as described in Section 13.4. 1401 13. Security Considerations 1403 The pNFS extension partitions the NFSv4 file system protocol into two 1404 parts, the control path and the data path (storage protocol). The 1405 control path contains all the new operations described by this 1406 extension; all existing NFSv4 security mechanisms and features apply 1407 to the control path. The combination of components in a pNFS system 1408 is required to preserve the security properties of NFSv4 with respect 1409 to an entity accessing data via a client, including security 1410 countermeasures to defend against threats that NFSv4 provides 1411 defenses for in environments where these threats are considered 1412 significant. 1414 The metadata server enforces the file access-control policy at 1415 LAYOUTGET time. The client should use suitable authorization 1416 credentials for getting the layout for the requested iomode (READ or 1417 RW) and the server verifies the permissions and ACL for these 1418 credentials, possibly returning NFS4ERR_ACCESS if the client is not 1419 allowed the requested iomode. If the LAYOUTGET operation succeeds 1420 the client receives, as part of the layout, a set of object 1421 capabilities allowing it I/O access to the specified objects 1422 corresponding to the requested iomode. When the client acts on I/O 1423 operations on behalf of its local users, it MUST authenticate and 1424 authorize the user by issuing respective OPEN and ACCESS calls to the 1425 metadata server, similar to having NFSv4 data delegations. If access 1426 is allowed, the client uses the corresponding (READ or RW) 1427 capabilities to perform the I/O operations at the object storage 1428 devices. When the metadata server receives a request to change a 1429 file's permissions or ACL, it SHOULD recall all layouts for that file 1430 and it MUST change the capability version attribute on all objects 1431 comprising the file to implicitly invalidate any outstanding 1432 capabilities before committing to the new permissions and ACL. Doing 1433 this will ensure that clients re-authorize their layouts according to 1434 the modified permissions and ACL by requesting new layouts. 1435 Recalling the layouts in this case is courtesy of the server intended 1436 to prevent clients from getting an error on I/Os done after the 1437 capability version changed. 1439 The object storage protocol MUST implement the security aspects 1440 described in version 1 of the T10 OSD protocol definition [1]. The 1441 standard defines four security methods: NOSEC, CAPKEY, CMDRSP, and 1442 ALLDATA. To provide minimum level of security allowing verification 1443 and enforcement of the server access control policy using the layout 1444 security credentials, the NOSEC security method MUST NOT be used for 1445 any I/O operation. The remainder of this section gives an overview 1446 of the security mechanism described in that standard. The goal is to 1447 give the reader a basic understanding of the object security model. 1448 Any discrepancies between this text and the actual standard are 1449 obviously to be resolved in favor of the OSD standard. 1451 13.1. OSD Security Data Types 1453 There are three main data types associated with object security: a 1454 capability, a credential, and security parameters. The capability is 1455 a set of fields that specifies an object and what operations can be 1456 performed on it. A credential is a signed capability. Only a 1457 security manager that knows the secret device keys can correctly sign 1458 a capability to form a valid credential. In pNFS, the file server 1459 acts as the security manager and returns signed capabilities (i.e., 1460 credentials) to the pNFS client. The security parameters are values 1461 computed by the issuer of OSD commands (i.e., the client) that prove 1462 they hold valid credentials. The client uses the credential as a 1463 signing key to sign the requests it makes to OSD, and puts the 1464 resulting signatures into the security_parameters field of the OSD 1465 command. The object storage device uses the secret keys it shares 1466 with the security manager to validate the signature values in the 1467 security parameters. 1469 The security types are opaque to the generic layers of the pNFS 1470 client. The credential contents are defined as opaque within the 1471 pnfs_osd_object_cred4 type. Instead of repeating the definitions 1472 here, the reader is referred to Section 4.9.2.2 of the OSD standard. 1474 13.2. The OSD Security Protocol 1476 The object storage protocol relies on a cryptographically secure 1477 capability to control accesses at the object storage devices. 1478 Capabilities are generated by the metadata server, returned to the 1479 client, and used by the client as described below to authenticate 1480 their requests to the object-based storage device. Capabilities 1481 therefore achieve the required access and open mode checking. They 1482 allow the file server to define and check a policy (e.g., open mode) 1483 and the OSD to enforce that policy without knowing the details (e.g., 1484 user IDs and ACLs). 1486 Since capabilities are tied to layouts, and since they are used to 1487 enforce access control, when the file ACL or mode changes the 1488 outstanding capabilities MUST be revoked to enforce the new access 1489 permissions. The server SHOULD recall layouts to allow clients to 1490 gracefully return their capabilities before the access permissions 1491 change. 1493 Each capability is specific to a particular object, an operation on 1494 that object, a byte range within the object (in OSDv2), and has an 1495 explicit expiration time. The capabilities are signed with a secret 1496 key that is shared by the object storage devices and the metadata 1497 managers. Clients do not have device keys so they are unable to 1498 forge the signatures in the security parameters. The combination of 1499 a capability, the OSD System ID, and a signature is called a 1500 "credential" in the OSD specification. 1502 The details of the security and privacy model for object storage are 1503 defined in the T10 OSD standard. The following sketch of the 1504 algorithm should help the reader understand the basic model. 1506 LAYOUTGET returns a CapKey and a Cap, which, together with the OSD 1507 SystemID, are also called a credential. It is a capability and a 1508 signature over that capability and the SystemID. The OSD Standard 1509 refers to the CapKey as the "Credential integrity check value" and to 1510 the ReqMAC as the "Request integrity check value". 1512 CapKey = MAC(Cap, SystemID) 1513 Credential = {Cap, SystemID, CapKey} 1515 The client uses CapKey to sign all the requests it issues for that 1516 object using the respective Cap. In other words, the Cap appears in 1517 the request to the storage device, and that request is signed with 1518 the CapKey as follows: 1520 ReqMAC = MAC(Req, ReqNonce) 1521 Request = {Cap, Req, ReqNonce, ReqMAC} 1523 The following is sent to the OSD: {Cap, Req, ReqNonce, ReqMAC}. The 1524 OSD uses the SecretKey it shares with the metadata server to compare 1525 the ReqMAC the client sent with a locally computed value: 1527 LocalCapKey = MAC(Cap, SystemID) 1528 LocalReqMAC = MAC(Req, ReqNonce) 1530 and if they match the OSD assumes that the capabilities came from an 1531 authentic metadata server and allows access to the object, as allowed 1532 by the Cap. 1534 13.3. Protocol Privacy Requirements 1536 Note that if the server LAYOUTGET reply, holding CapKey and Cap, is 1537 snooped by another client, it can be used to generate valid OSD 1538 requests (within the Cap access restrictions). 1540 To provide the required privacy requirements for the capability key 1541 returned by LAYOUTGET, the GSS-API [7] framework can be used, e.g., 1542 by using the RPCSEC_GSS privacy method to send the LAYOUTGET 1543 operation or by using the SSV key to encrypt the oc_capability_key 1544 using the GSS_Wrap() function. Two general ways to provide privacy 1545 in the absence of GSS-API that are independent of NFSv4 are either an 1546 isolated network such as a VLAN or a secure channel provided by IPsec 1547 [15]. 1549 13.4. Revoking Capabilities 1551 At any time, the metadata server may invalidate all outstanding 1552 capabilities on an object by changing its POLICY ACCESS TAG 1553 attribute. The value of the POLICY ACCESS TAG is part of a 1554 capability, and it must match the state of the object attribute. If 1555 they do not match, the OSD rejects accesses to the object with the 1556 sense key set to ILLEGAL REQUEST and an additional sense code set to 1557 INVALID FIELD IN CDB. When a client attempts to use a capability and 1558 is rejected this way, it should issue a LAYOUTCOMMIT for the object 1559 and specify PNFS_OSD_BAD_CRED in the olr_ioerr_report parameter. The 1560 client may elect to issue a compound LAYOUTRETURN/LAYOUTGET (or 1561 LAYOUTCOMMIT/LAYOUTRETURN/LAYOUTGET) to attempt to fetch a refreshed 1562 set of capabilities. 1564 The metadata server may elect to change the access policy tag on an 1565 object at any time, for any reason (with the understanding that there 1566 is likely an associated performance penalty, especially if there are 1567 outstanding layouts for this object). The metadata server MUST 1568 revoke outstanding capabilities when any one of the following occurs: 1570 o the permissions on the object change, 1572 o a conflicting mandatory byte-range lock is granted, or 1574 o a layout is revoked and reassigned to another client. 1576 A pNFS client will typically hold one layout for each byte range for 1577 either READ or READ/WRITE. The client's credentials are checked by 1578 the metadata server at LAYOUTGET time and it is the client's 1579 responsibility to enforce access control among multiple users 1580 accessing the same file. It is neither required nor expected that 1581 the pNFS client will obtain a separate layout for each user accessing 1582 a shared object. The client SHOULD use OPEN and ACCESS calls to 1583 check user permissions when performing I/O so that the server's 1584 access control policies are correctly enforced. The result of the 1585 ACCESS operation may be cached while the client holds a valid layout 1586 as the server is expected to recall layouts when the file's access 1587 permissions or ACL change. 1589 14. IANA Considerations 1591 As described in NFSv4.1 [6], new layout type numbers have been 1592 assigned by IANA. This document defines the protocol associated with 1593 the existing layout type number, LAYOUT4_OSD2_OBJECTS, and it 1594 requires no further actions for IANA. 1596 15. References 1598 15.1. Normative References 1600 [1] Weber, R., "Information Technology - SCSI Object-Based Storage 1601 Device Commands (OSD)", ANSI INCITS 400-2004, December 2004. 1603 [2] Bradner, S., "Key words for use in RFCs to Indicate Requirement 1604 Levels", BCP 14, RFC 2119, March 1997. 1606 [3] Eisler, M., "XDR: External Data Representation Standard", 1607 STD 67, RFC 4506, May 2006. 1609 [4] Shepler, S., Ed., Eisler, M., Ed., and D. Noveck, Ed., "Network 1610 File System (NFS) Version 4 Minor Version 1 External Data 1611 Representation Standard (XDR) Description", RFC 5662, 1612 January 2010. 1614 [5] IETF Trust, "Legal Provisions Relating to IETF Documents", 1615 November 2008, 1616 . 1618 [6] Shepler, S., Ed., Eisler, M., Ed., and D. Noveck, Ed., "Network 1619 File System (NFS) Version 4 Minor Version 1 Protocol", 1620 RFC 5661, January 2010. 1622 [7] Linn, J., "Generic Security Service Application Program 1623 Interface Version 2, Update 1", RFC 2743, January 2000. 1625 [8] Satran, J., Meth, K., Sapuntzakis, C., Chadalapaka, M., and E. 1626 Zeidner, "Internet Small Computer Systems Interface (iSCSI)", 1627 RFC 3720, April 2004. 1629 [9] Weber, R., "SCSI Primary Commands - 3 (SPC-3)", ANSI 1630 INCITS 408-2005, October 2005. 1632 [10] Krueger, M., Chadalapaka, M., and R. Elliott, "T11 Network 1633 Address Authority (NAA) Naming Format for iSCSI Node Names", 1634 RFC 3980, February 2005. 1636 [11] IEEE, "Guidelines for 64-bit Global Identifier (EUI-64) 1637 Registration Authority", 1638 . 1640 [12] Tseng, J., Gibbons, K., Travostino, F., Du Laney, C., and J. 1641 Souza, "Internet Storage Name Service (iSNS)", RFC 4171, 1642 September 2005. 1644 [13] Weber, R., "SCSI Architecture Model - 3 (SAM-3)", ANSI 1645 INCITS 402-2005, February 2005. 1647 15.2. Informative References 1649 [14] Weber, R., "SCSI Object-Based Storage Device Commands -2 1650 (OSD-2)", January 2009, 1651 . 1653 [15] Kent, S. and K. Seo, "Security Architecture for the Internet 1654 Protocol", RFC 4301, December 2005. 1656 [16] T10 1415-D, "SCSI RDMA Protocol (SRP)", ANSI INCITS 365-2002, 1657 December 2002. 1659 [17] T11 1619-D, "Fibre Channel Framing and Signaling - 2 1660 (FC-FS-2)", ANSI INCITS 424-2007, February 2007. 1662 [18] T10 1601-D, "Serial Attached SCSI - 1.1 (SAS-1.1)", ANSI 1663 INCITS 417-2006, June 2006. 1665 [19] MacWilliams, F. and N. Sloane, "The Theory of Error-Correcting 1666 Codes, Part I", 1977. 1668 [20] Anvin, H., "The Mathematics of RAID-6", May 2009, 1669 . 1671 [21] The free encyclopedia, Wikipedia., "Least common multiple", 1672 April 2011, 1673 . 1675 [22] Plank, James S., and Luo, Jianqiang and Schuman, Catherine D. 1676 and Xu, Lihao and Wilcox-O'Hearn, Zooko, "A Performance 1677 Evaluation and Examination of Open-source Erasure Coding 1678 Libraries for Storage", 2007. 1680 Appendix A. Acknowledgments 1682 Todd Pisek was a co-editor of the initial versions of this document. 1683 Daniel E. Messinger, Pete Wyckoff, Mike Eisler, Sean P. Turner, Brian 1684 E. Carpenter, Jari Arkko, David Black, and Jason Glasgow reviewed and 1685 commented on this document. 1687 Authors' Addresses 1689 Benny Halevy 1690 Primary Data 1692 Email: bhalevy@primarydata.com 1693 URI: http://www.primarydata.com/ 1694 Boaz Harrosh 1695 Panasas, Inc. 1696 1501 Reedsdale St. Suite 400 1697 Pittsburgh, PA 15233 1698 USA 1700 Phone: +1-412-323-3500 1701 Email: bharrosh@panasas.com 1702 URI: http://www.panasas.com/ 1704 Brent Welch 1705 Panasas, Inc. 1706 969 W. Maude Ave 1707 Sunnyvale, CA 94095 1708 USA 1710 Phone: +1-408-215-6715 1711 Email: welch@acm.org 1712 URI: http://www.panasas.com/