idnits 2.17.1 draft-ietf-nfsv4-designconsider-00.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- ** Looks like you're using RFC 2026 boilerplate. This must be updated to follow RFC 3978/3979, as updated by RFC 4748. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- ** Missing expiration date. The document expiration date should appear on the first and last page. ** The document seems to lack a 1id_guidelines paragraph about 6 months document validity -- however, there's a paragraph with a matching beginning. Boilerplate error? == No 'Intended status' indicated for this document; assuming Proposed Standard Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- ** The document seems to lack an Introduction section. ** The document seems to lack an IANA Considerations section. (See Section 2.2 of https://www.ietf.org/id-info/checklist for how to handle the case when there are no actions for IANA.) Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the RFC 3978 Section 5.4 Copyright Line does not match the current year == The document doesn't use any RFC 2119 keywords, yet seems to have RFC 2119 boilerplate text. -- The document seems to lack a disclaimer for pre-RFC5378 work, but may have content which was first submitted before 10 November 2008. If you have contacted all the original authors and they are all willing to grant the BCP78 rights to the IETF Trust, then this is fine, and you can ignore this comment. If not, you may need to add the pre-RFC5378 disclaimer. (See the Legal Provisions document at https://trustee.ietf.org/license-info for more information.) -- The document date (March 1999) is 9174 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) -- Missing reference section? 'RFC2054' on line 820 looks like a reference -- Missing reference section? 'RFC2055' on line 826 looks like a reference -- Missing reference section? 'RFC1831' on line 796 looks like a reference -- Missing reference section? 'XDSM' on line 933 looks like a reference -- Missing reference section? 'HPFS' on line 879 looks like a reference -- Missing reference section? 'Nagar' on line 898 looks like a reference -- Missing reference section? 'DCEACL' on line 874 looks like a reference -- Missing reference section? 'RFC2203' on line 844 looks like a reference -- Missing reference section? 'RFC2246' on line 862 looks like a reference -- Missing reference section? 'RFC2025' on line 814 looks like a reference -- Missing reference section? 'RFC1510' on line 784 looks like a reference -- Missing reference section? 'RFC2401' on line 868 looks like a reference -- Missing reference section? 'RFC1832' on line 802 looks like a reference -- Missing reference section? 'Unicode1' on line 920 looks like a reference -- Missing reference section? 'RFC1094' on line 778 looks like a reference -- Missing reference section? 'RFC1813' on line 790 looks like a reference -- Missing reference section? 'RFC1833' on line 808 looks like a reference -- Missing reference section? 'RFC2078' on line 832 looks like a reference -- Missing reference section? 'RFC2152' on line 838 looks like a reference -- Missing reference section? 'RFC2222' on line 850 looks like a reference -- Missing reference section? 'RFC2279' on line 856 looks like a reference -- Missing reference section? 'Hutson' on line 883 looks like a reference -- Missing reference section? 'Kistler' on line 888 looks like a reference -- Missing reference section? 'Mummert' on line 893 looks like a reference -- Missing reference section? 'Sandberg' on line 902 looks like a reference -- Missing reference section? 'Satyanarayanan1' on line 910 looks like a reference -- Missing reference section? 'Satyanarayanan2' on line 914 looks like a reference -- Missing reference section? 'Unicode2' on line 927 looks like a reference -- Missing reference section? 'XNFS' on line 938 looks like a reference Summary: 5 errors (**), 0 flaws (~~), 3 warnings (==), 31 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group Spencer Shepler 3 Internet Draft March 1999 4 Document: draft-ietf-nfsv4-designconsider-00.txt 6 NFS Version 4 Design Considerations 8 Status of this Memo 10 This document is an Internet-Draft and is in full conformance with 11 all provisions of Section 10 of RFC2026. 13 Internet-Drafts are working documents of the Internet Engineering 14 Task Force (IETF), its areas, and its working groups. Note that 15 other groups may also distribute working documents as Internet- 16 Drafts. 18 Internet-Drafts are draft documents valid for a maximum of six months 19 and may be updated, replaced, or obsoleted by other documents at any 20 time. It is inappropriate to use Internet- Drafts as reference 21 material or to cite them other than as "work in progress." 23 The list of current Internet-Drafts can be accessed at 24 http://www.ietf.org/ietf/1id-abstracts.txt 26 The list of Internet-Draft Shadow Directories can be accessed at 27 http://www.ietf.org/shadow.html. 29 Abstract 31 The main task of the NFS version 4 working group is to create a 32 protocol definition for a distributed file system that focuses on the 33 following items: improved access and good performance on the 34 Internet, strong security with negotiation built into the protocol, 35 better cross-platform interoperability, and designed for protocol 36 extensions. NFS version 4 will owe its general design to the 37 previous versions of NFS. It is expected, however, that many 38 features will be quite different in NFS version 4 than previous 39 versions to facilitate the goals of the working group and to address 40 areas that NFS version 2 and 3 have not. 42 This design considerations document is meant to present more detail 43 than the working group charter. Specifically, it presents the areas 44 that the working group will investigate and consider while developing 45 a protocol specification for NFS version 4. Based on this 46 investigation the working group will decide the features of the new 47 protocol based on the cost and benefits within the specific feature 48 areas. 50 Copyright 52 Copyright (C) The Internet Society (1999). All Rights Reserved. 54 Key Words 56 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 57 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 58 document are to be interpreted as described in RFC 2119. 60 Table of Contents 62 1. NFS Version 4 Design Considerations . . . . . . . . . . . . . 4 63 2. Ease of Implementation or Complexity of Protocol . . . . . . 4 64 2.1. Extensibility / layering . . . . . . . . . . . . . . . . . 4 65 2.2. Managed Extensions or Minor Versioning . . . . . . . . . . 4 66 3. Reliable and Available . . . . . . . . . . . . . . . . . . . 6 67 4. Scalable Performance . . . . . . . . . . . . . . . . . . . . 6 68 4.1. Throughput and Latency via the Network . . . . . . . . . . 7 69 4.2. Client Caching . . . . . . . . . . . . . . . . . . . . . . 7 70 4.3. Disconnected Client Operation . . . . . . . . . . . . . . . 8 71 5. Interoperability . . . . . . . . . . . . . . . . . . . . . . 8 72 5.1. Platform Specific Behavior . . . . . . . . . . . . . . . . 9 73 5.2. Additional or Extended Attributes . . . . . . . . . . . . . 9 74 5.3. Access Control Lists . . . . . . . . . . . . . . . . . . 10 75 6. RPC Mechanism and Security . . . . . . . . . . . . . . . . 11 76 6.1. User identification . . . . . . . . . . . . . . . . . . . 11 77 6.2. Security . . . . . . . . . . . . . . . . . . . . . . . . 12 78 6.2.1. Transport Independence . . . . . . . . . . . . . . . . 12 79 6.2.2. Authentication . . . . . . . . . . . . . . . . . . . . 12 80 6.2.3. Data Integrity . . . . . . . . . . . . . . . . . . . . 12 81 6.2.4. Data Privacy . . . . . . . . . . . . . . . . . . . . . 13 82 6.2.5. Security Negotiation . . . . . . . . . . . . . . . . . 13 83 6.3. Summary . . . . . . . . . . . . . . . . . . . . . . . . . 14 84 7. Internet Accessibility . . . . . . . . . . . . . . . . . . 14 85 7.1. Congestion Control and Transport Selection . . . . . . . 14 86 7.2. Firewalls and Proxy Servers . . . . . . . . . . . . . . . 15 87 7.3. Multiple RPCs and Latency . . . . . . . . . . . . . . . . 15 88 8. File locking / recovery . . . . . . . . . . . . . . . . . . 16 89 9. Internationalization . . . . . . . . . . . . . . . . . . . 17 90 10. Security Considerations . . . . . . . . . . . . . . . . . 18 91 10.1. Denial of Service . . . . . . . . . . . . . . . . . . . 19 92 11. Bibliography . . . . . . . . . . . . . . . . . . . . . . . 20 93 12. Acknowledgments . . . . . . . . . . . . . . . . . . . . . 25 94 13. Author's Address . . . . . . . . . . . . . . . . . . . . . 25 95 14. Full Copyright Statement . . . . . . . . . . . . . . . . . 26 97 1. NFS Version 4 Design Considerations 99 As stated in the charter, the first deliverable for the NFS version 4 100 working group is this design considerations document. This document 101 is to cover the "limitations and deficiencies of NFS version 3". 102 This document will also be used as a mechanism to focus discussion 103 and avenues of investigation as the definition of NFS version 4 104 progresses. Therefore, the contents of this document cover the 105 general functional/feature areas that are anticipated for NFS version 106 4. Where appropriate, discussion of current NFS version 2 and 3 107 practice will be presented along with other appropriate references to 108 current distributed file system practice. 110 2. Ease of Implementation or Complexity of Protocol 112 One of the strengths of NFS has been the ability to implement a 113 client or server with relative ease. The eventual size of a basic 114 implementation is relatively small. The main reason for keeping NFS 115 as simple as possible is that a simple protocol design can be 116 described in a simple specification that promotes straightforward, 117 interoperable implementations. All protocols can run into problems 118 when deployed on real networks, but simple protocols yield problems 119 that are easier to diagnose and correct. 121 2.1. Extensibility / layering 123 With NFS' relative simplicity, the addition or layering of 124 functionality has been easy to accomplish. The addition of features 125 like the client automount or autofs, client side disk caching and 126 high availability servers are examples. This type of extensibility 127 is desirable in an environment where problem solutions do not require 128 protocol revision. This extensibility can also be helpful in the 129 future where unforeseen problems or opportunities can be solved by 130 layering functionality on an existing set of tools or protocol. 132 2.2. Managed Extensions or Minor Versioning 134 For those cases where the NFS protocol is deficient or where a minor 135 modification is the best solution for a problem, a minor version or a 136 managed extension could be helpful. There have been instances with 137 NFS version 2 and 3 where small straightforward functional additions 138 would have increased the overall value of the protocol immensely. 139 For instance, the PATHCONF procedure that was added to version 2 of 140 the MOUNT protocol would have been more appropriate for the NFS 141 protocol. WebNFS [RFC2054][RFC2055] overloading of the LOOKUP 142 procedure for NFS versions 2 and 3 would have been more cleanly 143 implemented in a new LOOKUP procedure. 145 However, the perceived size and burden of using a change of RPC 146 version number for the introduction of new functionality led to no or 147 slow change. It is possible that a new NFS protocol could allow for 148 the rare instance where protocol extension within the RPC version 149 number is the most prudent course and an RPC revision would be 150 unnecessary or impractical. 152 The areas of an NFS protocol which are most obviously volatile are 153 new orthogonal procedures, new well-defined file or directory 154 attributes and potentially new file types. As an example, potential 155 file types of the future could be a type such as "attribute" that 156 represents a named file stream or a "dynamic" file type that 157 generates dynamic data in response to a "query" procedure from the 158 client. 160 It is possible and highly desirable that these types of additions be 161 done without changing the overall design model of NFS without 162 significant effort or delay. 164 A strong consideration should be given to a NFS protocol mechanism 165 for the introduction of this type of new functionality. This is 166 obviously in contrast to using the standard RPC version mechanism to 167 provide minor changes. The process of using RPC version numbers to 168 introduce new functionality brings with it a lot of history which may 169 not technically prevent its use. However, the historical issues 170 involved will need to be addressed as part of the NFS version 4 171 protocol work; this should increase the ability for current and 172 future success of the protocol. 174 As background, the RPC protocol described in [RFC1831] uses a version 175 number to describe the set of procedure calls, replies, and their 176 semantics. Any change in this set must be reflected in a new version 177 number for the program. An example of this was the 178 MOUNTPROC_PATHCONF procedure added in version 2 of the MOUNT 179 protocol. Except for the addition of this new procedure, the 180 protocol was unchanged. Many thought this protocol revision was 181 unnecessary, since the RPC protocol already allows certain procedures 182 not be implemented and defines a PROC_UNAVAIL error. 184 Another historical data-point from NFS version 2 and 3 is the support 185 (or lack) of symbolic links. Servers that cannot support this 186 feature will simply reject calls to the SYMLINK and READLINK 187 procedures. Additionally, NFS version 4 may describe many file 188 attributes which cannot be supported by the server or file systems on 189 the server. Therefore, the protocol must support a discovery 190 mechanism that allows clients to determine which features of the 191 protocol are supported by a server. 193 3. Reliable and Available 195 Current NFS protocol design, while placing an emphasis on simple 196 server design, has led to timely recovery from server and client 197 failure. This and other aspects to the design have provided a basis 198 for layered technologies like high availability and clustered 199 servers. Providing a protocol design approach that lends itself to 200 these types of reliability and availability features is very 201 desirable. 203 For the next version of NFS, consideration should be given to client 204 side availability schemes such as client switching between or fail- 205 over to available server replicas. NFS currently requires that file 206 handles be immutable; this requirement adds unnecessarily to the 207 complexity of building fail-over configurations. If possible, the 208 protocol should allow for or ease the building of such layered 209 solutions. 211 For the next version of NFS, consideration should be given to schemes 212 that support client switching between server replicas or highly 213 available NFS servers that provide paths to data through multiple 214 servers. For example: NFS currently requires that filehandles be 215 unchanging for any instance of a file or directory. This requirement 216 makes it more difficult for a client to switch from one server to 217 another, since each server may construct filehandles differently. 218 Protocol support could allow the client to handle a filehandle 219 change. 221 4. Scalable Performance 223 In designing and developing an NFS protocol from a performance 224 viewpoint there are several different points to consider. Each can 225 play a significant role in perceived and real performance from the 226 user's perspective. The three main areas of interest are: throughput 227 and latency via the network, server work load or scalability and 228 client side caching. 230 4.1. Throughput and Latency via the Network 232 NFS currently has characteristics that provide good throughput for 233 reading and writing file data. This is commonly achieved by the 234 client's use of pipelining or windowing multiple RPC READ/WRITE 235 requests to the server. The flexibility of the NFS and ONCRPC 236 protocols allow for implementations to use this type of adaptation to 237 provide efficient use of the network connection. 239 However, the number of RPCs required to accomplish some tasks 240 combined with high latency network environments may lead to sluggish 241 single user or single client response. The protocol should continue 242 to provide good raw read and write throughput while addressing the 243 issue of network latency. This issue is discussed further in the 244 section on Internet Accessibility. 246 4.2. Client Caching 248 In an attempt to speed response time and to reduce network and server 249 load, NFS clients have always cached directory and file data. 250 However, this has usually been done as memory cache and in relatively 251 recent history, local disk caching has been added. 253 It is very desirable to have the NFS client cache directory and file 254 data. Other distributed file systems have shown that aggressive 255 client side caching can be very visible to the end user in the form 256 of decreasing overall response time. For AFS and DCE/DFS, caching is 257 accomplished by the utilization of server call backs to notify the 258 client of potential cache invalidation. CIFS and its opportunistic 259 locks provide a similar call back mechanism. Clients in both of 260 these environments are able to cache data while avoiding interaction 261 with the network and server. 263 With these protocols it is also possible to cache or delay certain 264 protocol requests at the client which further reduces the protocol 265 traffic flowing between client and server. In the case of CIFS, it 266 is possible for a client to obtain an opportunistic lock for a file 267 and subsequently process file lock requests completely at the client. 268 If there are no conflicts with other clients for file data access, 269 the server is never contacted for the file locking traffic generated 270 by the user application. This behavior is not a protocol requirement 271 but is allowed by the protocol as an implementation option to improve 272 performance. 274 NFS versions 2 and 3 make no caching requirements. Implementations 275 typically implement close-to-open cache consistency which requires 276 clients flush all changes to the server on each file close, and check 277 for file changes on the server on each file open. The consistency 278 check required on each file open can generate a large amount of 279 GETATTR traffic. With this approach, there are windows when the 280 client can still be acting with stale data between the open and close 281 of a file. 283 Client caching is increasingly important for Internet environments 284 where throughput can be limited and response time can grow 285 significantly. Therefore the NFS version 4 caching design will need 286 to take into account the full spectrum of caching designs as 287 exemplified by the current technologies of NFS, AFS, DCE/DFS, CIFS, 288 etc. in determining an appropriate design. This will need to be done 289 while weighing the complexity of each possible approach with the need 290 of the eventual users and operating environments into which NFS 291 version 4 may be deployed. Some of these considerations are: 292 Internet accessibility, firewall traversal (call back availability), 293 proxy caching, low-overhead or simple clients. 295 4.3. Disconnected Client Operation 297 An extension of client caching is the provision for disconnected 298 operation at the client. With the ability to cache directory and 299 file data aggressively, a client could then provide service to the 300 end user while disconnected from the server or network. 302 While very desirable, disconnected operation has the potential to 303 inflict itself upon the NFS protocol in an undesirable way as 304 compared to traditional client caching. Given the complexities of 305 disconnected client operation and subsequent resolution of client 306 data modification through various playback or data selection 307 mechanisms, disconnected operation should not be a requirement for 308 the NFS effort. Even so, the NFS protocol should consider the 309 potential layering of disconnected operation solutions on top of the 310 NFS protocol (as with other server and client solutions). The 311 experiences with Coda, disconnected AFS and others should be helpful 312 in this area. (see references) 314 5. Interoperability 316 The NFS protocols are available for many different operating 317 environments. Even though this shows the protocol's ability to 318 provide distributed file system service for more than a single 319 operating system, the design of NFS is certainly Unix-centric. The 320 next NFS protocol needs to be more inclusive of platform or file 321 system features beyond those of traditional Unix. 323 5.1. Platform Specific Behavior 325 Because of Unix-centric origins, NFS version 2 and 3 protocol 326 requirements have been difficult to implement in some environments. 327 For example, persistent file handles (unique identifiers of file 328 system objects), Unix uid/gid mappings, directory modification time, 329 accurate file sizes, file/directory locking semantics (SHAREs, PC- 330 style locking). In the design of NFS version 4, these areas and 331 others not mentioned will need to be considered and, if possible, 332 cross-platform solutions developed. 334 5.2. Additional or Extended Attributes 336 NFS versions 2 and 3 do not provide for file or directory attributes 337 beyond those that are found in the traditional Unix environment. For 338 example the user identifier/owner of the file, a permission or access 339 bitmap, time stamps for modification of the file or directory and 340 file size to name a few. While the current set of attributes has 341 usually been sufficient, the file system's ability to manage 342 additional information associated with a file or directory can be 343 useful. 345 There are many possibilities for additional attributes in the next 346 version of NFS. Some of these include: object creation timestamp, 347 user identifier of file's creator, timestamp of last backup or 348 archival bit, version number, file content type (MIME type), 349 existence of data management involvement (i.e. DMAPI [XDSM]). 351 This list is representative of the possibilities and is meant to show 352 the need for an additional attribute set. Enumerating the 'correct' 353 set of attributes, however, is difficult. This is one of the reasons 354 for looking towards a minor versioning mechanism as a way to provide 355 needed extensibility. Another way to provide some extensibility is 356 to support a generalized named attribute mechanism. This mechanism 357 would allow a client to name, store and retrieve arbitrary data and 358 have it associated as an attribute of a file or directory. 360 One difficulty in providing named attributes is determining if the 361 protocol should specify the names for the attributes, their type or 362 structure. How will the protocol determine or allow for attributes 363 that can be read but not written is another issue. Yet another could 364 be the side effects that these attributes have on the core set of 365 file properties such as setting a size attribute to 0 and having 366 associated file data deleted. 368 As these brief examples show, this type of extended attribute 369 mechanism brings with it a large set of issues that will need to be 370 addressed in the protocol specification while keeping the overall 371 goal of simplicity in mind. 373 There are operating environments that provide named or extended 374 attribute mechanisms. Digital Unix provides for the storage of 375 extended attributes with some generalized format. HPFS[HPFS] and 376 NTFS [Nagar] also provide for named data associated with traditional 377 files. SGI's local file system, XFS, also provides for this type of 378 name/value extended attributes. However, there does not seem to be a 379 clear direction that can be taken from these or other environments. 381 5.3. Access Control Lists 383 Access Control Lists (ACL) can be viewed as one specific type of 384 extended attribute. This attribute is a designation of user access 385 to a file or directory. Many vendors have created ancillary 386 protocols to NFS to extend the server's ACL mechanism across the 387 network. Generally this has been done for homogeneous operating 388 environments. Even though the server still interprets the ACL and has 389 final control over access to a file system object, the client is able 390 to manipulate the ACL via these additional protocols. Other 391 distributed file systems have also provided ACL support (DFS, AFS and 392 CIFS). 394 The basic factor driving the requirement for ACL support in all of 395 these file systems has been the user's desire to grant and restrict 396 access to file system data on a per user basis. Based on the desire 397 of the user and current distributed file system support, it seems to 398 be a requirement that NFS provide this capability as well. 400 Because many local and distributed file system ACL implementations 401 have been done without a common architecture, the major issue is one 402 of compatibility. Although the POSIX draft, DCE/DFS [DCEACL] and 403 Windows NT ACLs have a similar structure in an array of Access 404 Control Entries consisting of a type field, identity, and permission 405 bits, the similarity ends there. Each model defines its own variants 406 of entry types, identifies users and groups differently, provides 407 different kinds of permission bits, and describes different 408 procedures for ACL creation, defaults, and evaluation. 410 In the least it will be problematic to create a workable ACL 411 mechanism that will encompass a reasonable set of functionality for 412 all operating environments. Even with the complicated nature of ACL 413 support it is still worthwhile to work towards a solution that can at 414 least provide basic functionality for the user. 416 6. RPC Mechanism and Security 418 NFS relies on the security mechanisms provided by the ONCRPC 419 [RFC1831] protocol. Until the introduction of the ONCRPC RPCSEC_GSS 420 security flavor [RFC2203], NFS security was generally limited to none 421 (AUTH_SYS) or DES (AUTH_DH). The AUTH_DH security flavor was not 422 successful in providing readily available security for NFS because of 423 a lack of widespread implementation which precluded widespread 424 deployment. Also the Diffie-Hellman 192 bit public key modulus used 425 for the AUTH_DH security flavor quickly became too small for 426 reasonable security. 428 6.1. User identification 430 NFS has been limited to the use of the Unix-centric user 431 identification mechanism of numeric user id based on the available 432 file system attributes and the use of the ONCRPC. However, for NFS 433 to move beyond the limits of large work groups, user identification 434 should be string based and the definition of the user identifier 435 should allow for integration into an external naming service or 436 services. 438 Internet scaling should also be considered for this as well. The 439 identification mechanism should take into account multiple naming 440 domains and multiple naming mechanisms. Flexibility is the key to a 441 solution that can grow with the needs of the user and administrator. 443 If NFS is to move among various naming and security services, it may 444 be necessary to stay with a string based identification. This would 445 allow for servers and clients to translate between the external 446 string representation to a local or internal numeric (or other 447 identifier) which matches internal implementation needs. 449 As an example, DFS uses a string based naming scheme but translates 450 the name to a UUID (16 byte identifier) that is used for internal 451 protocol representations. The DCE/DFS string name is a combination of 452 cell (administrative domain) and user name. As mentioned, NFS 453 clients and servers map a Unix user name to a 32 bit user identifier 454 that is then used for ONCRPC and NFS protocol fields requiring the 455 user identifier. 457 6.2. Security 459 Because of the aforementioned problems, user authentication has been 460 a major issue for ONCRPC and hence NFS. To satisfy requirements of 461 the IETF and to address concerns and requirements from users, NFS 462 version 4 must provide for the use of acceptable security mechanisms. 463 The various mechanisms currently available should be explored for 464 their appropriate use with NFS version 4 and ONCRPC. Some of these 465 mechanisms are: TLS [RFC2246], SPKM [RFC2025], KerberbosV5 [RFC1510], 466 IPSEC [RFC2401]. Since ONCRPC is the basis for NFS client and server 467 interaction, the RPCSEC_GSS [RFC2203] framework should be strongly 468 considered since it provides a method to employ mechanisms like SPKM 469 and KerberosV5. Before a security mechanism can be evaluated, the 470 NFS environment and requirements must be discussed. 472 6.2.1. Transport Independence 474 As mentioned later in this document in the section "Internet 475 Accessibility", transport independence is an asset for NFS and ONCRPC 476 and is a general requirement. This allows for transport choice based 477 on the target environment and specific application of NFS. The most 478 common transports in use with NFS are UDP and TCP. This ability to 479 choose transport should be maintained in combination with the user's 480 choice of a security mechanism. This implies that "mandatory to 481 implement" security mechanisms for NFS should allow for both 482 connection-less and connection-oriented transports. 484 6.2.2. Authentication 486 As should be expected, strong authentication is a requirement for NFS 487 version 4. Each operation generated via ONCRPC contains user 488 identification and authentication information. It is common in NFS 489 version 2 and 3 implementations to multiplex various users' requests 490 over a single or few connections to the NFS server. This allows for 491 scalability in the number of clients systems. Security mechanisms or 492 frameworks should allow for this multiplexing of requests to sustain 493 the implementation model that is available today. 495 6.2.3. Data Integrity 497 Until the introduction of RPCSEC_GSS, the ability to provide data 498 integrity over ONCRPC and to NFS was not available. Since file and 499 directory data is the essence of a distributed file service, the NFS 500 protocol should provide to the users of the file service a reasonable 501 level of data integrity. The security mechanisms chosen must provide 502 for NFS data protection with a cryptographically strong checksum. As 503 with other aspects within NFS version 4, the user or administrator 504 should be able to choose whether data integrity is employed. This 505 will provide needed flexibility for a variety of NFS version 4 506 solutions. 508 6.2.4. Data Privacy 510 Data privacy, while desirable, is not as important in all 511 environments as authentication and integrity. For example, in a LAN 512 environment the performance overhead of data privacy may not be 513 required to meet an organization's data protection policies. It may 514 also be the case that the performance of the distributed file system 515 solution is more important than the data privacy of that solution. 516 Even with these considerations, the user or administrator must have 517 the choice of data privacy and therefore it must be included in NFS 518 version 4. 520 6.2.5. Security Negotiation 522 With the ability to provide security mechanism choices to the user or 523 administrator, NFS version 4 should offer reasonable flexibility for 524 application of local security policies. However, this presents the 525 problem of negotiating the appropriate security mechanism between 526 client and server. It is unreasonable to require the client know the 527 server's chosen mechanism before initial contact. The issue is 528 further complicated by an administrator who may choose more than one 529 security mechanism for the various file system resources being shared 530 by an NFS server. These types of choices and policies require that 531 NFS version 4 deal with negotiating the appropriate security 532 mechanism based on mechanism availability and policy deployment at 533 client and server. This negotiation will need to take into account 534 the possibility of a change in policy as an NFS client crosses 535 certain file system boundaries at the server. The security 536 mechanisms required may change at these boundaries and therefore the 537 negotiation must be included within the NFS protocol itself to be 538 done properly (i.e. securely). 540 6.3. Summary 542 Other distributed file system solutions such as AFS and DFS provide 543 strong authentication mechanisms. CIFS does provide authentication 544 at initial server contact and a message signing option for subsequent 545 interaction. Recent NFS version 2 and 3 implementations, with the 546 use of RPCSEC_GSS, provide strong authentication, integrity, and 547 privacy. 549 NFS version 4 must provide for strong authentication, integrity, and 550 privacy. This must be part of the protocol so that users have the 551 choice to use strong security if their environment or policies 552 warrant such use. 554 Based on the requirements presented, the ONCRPC RPCSEC_GSS security 555 flavor seems to provide an appropriate framework for satisfying these 556 requirements. RPCSEC_GSS provides for authentication, integrity, and 557 privacy. The RPCSEC_GSS is also extensible in that it provides for 558 both public and private key security mechanisms along with the 559 ability to plug in various mechanisms in such a way that it does not 560 significantly disrupt ONCRPC or NFS implementations. 562 With RPCSEC_GSS' ability to support both public and private key 563 mechanisms, NFS version 4 should consider "mandatory to implement" 564 choices from both. The intent is to provide a security solution that 565 will flexibly scale to match the needs of end users. Providing this 566 range of solutions will allow for appropriate usage based on policy 567 and available resources for deployment. Note that, in the end, the 568 user must have a choice and that choice may be to use all of the 569 available mechanisms in NFS version 4 or none of them. 571 7. Internet Accessibility 573 Being a product of an IETF working group, the NFS protocol should not 574 only be built upon IETF technologies where possible but should also 575 work well within the broader Internet environment. 577 7.1. Congestion Control and Transport Selection 579 As with any network protocol, congestion control is a major issue and 580 the transport mechanisms that are chosen for NFS should take this 581 into account. Traditionally, implementations of NFS have been 582 deployed using both UDP and TCP. With the use of UDP, most 583 implementations provide a rudimentary attempt control congestion with 584 simple back-off algorithms and round trip timers. While this may be 585 sufficient in today's NFS deployments, as an Internet protocol NFS 586 will need to ensure sufficient congestion control or management. 588 With congestion control in mind, NFS must use TCP as a transport (via 589 ONCRPC). The UDP transport provides its own advantages in certain 590 circumstances. In today's NFS implementations, UDP has been shown to 591 produce greater throughput as compared to similarly configured 592 systems that use TCP. This issue will need to be investigated such 593 that a determination can be made as to whether the differences are 594 within implementation details. If UDP is supplied as an NFS 595 transport mechanism, then the congestion controls issues will need 596 resolution to make its use suitable. 598 7.2. Firewalls and Proxy Servers 600 NFS's protocol design should allow its use via Internet firewalls. 601 The protocol should also allow for the use of file system proxy/cache 602 servers. Proxy servers can be very useful for scalability and other 603 reasons. The NFS protocol needs to address the need of proxy servers 604 in a way that will deal with the issues of security, access control, 605 and content control. It is possible that these issues can be 606 addressed by documenting the related issues of proxy server usage. 607 However, it is likely that the NFS protocol will need to support 608 proxy servers directly through the NFS protocol. 610 The protocol could allow a request to be sent to a proxy that 611 contains the name of the target NFS server to which the request might 612 be forwarded, or from which a response might be cached. In any case, 613 the NFS proxy server should be considered during protocol 614 development. 616 The problems encountered in making the NFS protocol work through 617 firewalls are described in detail in [RFC2054] and [RFC2055]. 619 7.3. Multiple RPCs and Latency 621 As an application at the NFS client performs simple file system 622 operations, multiple NFS operations or RPCs may be executed to 623 accomplish the work for the application. While the NFS version 3 624 protocol addressed some of this by returning file and directory 625 attributes for most procedures, hence reducing follow up GETATTR 626 requests, there is still room for improvement. Reducing the number 627 of RPCs will lead to a reduction of processing overhead on the server 628 (transport and security processing) along with reducing the time 629 spent at the client waiting for the server's individual responses. 630 This issue is more prominent in environments with larger degrees of 631 latency. 633 The CIFS file access protocol supports 'batched requests' that allow 634 multiple requests to be batched, therefore reducing the number of 635 round trip messages between client and server. 637 This same approach can be used by NFS to allow the grouping of 638 multiple procedure calls together in a traditional RPC request. Not 639 only would this reduce protocol imposed latency but it would reduce 640 transport and security processing overhead and could allow a client 641 to complete more complex tasks by combining procedures. 643 8. File locking / recovery 645 NFS provided Unix file locking and DOS SHARE capability with the use 646 of an ancillary protocol (Network Lock Manager / NLM). The DOS SHARE 647 mechanism is the DOS equivalent of file locking in that it provides 648 the basis for sharing or exclusive access to file and directory data 649 without risk of data corruption. The NLM protocol provides file 650 locking and recovery of those locks in the event of client or server 651 failure. The NLM protocol requires that the server make call backs 652 to the client for certain scenarios and therefore is not necessarily 653 well suited for Internet firewall traversal. 655 Available and correct file locking support for NFS version 2 and 3 656 clients and servers has historically been problematic. The 657 availability of NLM support has traditionally been a problem and 658 seems to be most related to the fact that NFS and NLM are two 659 separate protocols. It is easy to deliver an NFS client and server 660 implementation and then add NLM support later. This led to a general 661 lack of NLM support early on in NFS' lifetime. One of the reasons 662 that NLM was delivered separately has been its relative complexity 663 which has in turn led to poor implementations and testing 664 difficulties. Even in later implementations where reliability and 665 performance had been increased to acceptable levels for NLM, users 666 still chose to avoid the use of the protocol and its support. The 667 last issue with NLM is the presence of minor protocol design flaws 668 that relate to high network load and recovery. 670 Based on the experiences with NLM, locking support for NFS version 4 671 should strive to meet or at least consider the following (in order of 672 importance): 674 o Integration with the NFS protocol and ease of implementation. 676 o Interoperability between operating environments. The protocol 677 should make a reasonable effort to support the locking semantics 678 of both PC and Unix clients and servers. This will allow for 679 greater integration of all environments. 681 o Scalable solutions - thousands of clients. The server should 682 not be required to maintain significant client file locking 683 state between server instantiations. 685 o Internet capable (firewall traversal, latency sensitive). The 686 server should not be required to initiate TCP connections to 687 clients. 689 o Timely recovery in the event of client/server or network 690 failure. Server recovery should be rapid. The protocol should 691 allow clients to detect the loss of a lock. 693 9. Internationalization 695 NFS version 2 and 3 are currently limited in the character encoding 696 of strings. In the NFS protocols, strings are used for file and 697 directory names, and symbolic link contents. Although the XDR 698 definition [RFC1832] limits strings in the NFS protocol to 7-bit US- 699 ASCII, common usage is to encode filenames in 8-bit ISO-Latin-1. 700 However, there is no mechanism available to tag XDR character strings 701 to indicate the character encoding used by the client or server. 702 Obviously this limits NFS' usefulness in an environment with clients 703 that may operate with various character sets. 705 One approach to address this deficiency is to use the Unicode 706 Standard [Unicode1] as the means to exchange character strings for 707 the NFS version 4 protocol. The Unicode Standard is a 16 bit encoding 708 that supports full multilingual text. The Unicode Standard is code- 709 for-code identical with International Standard ISO/IEC 10646-1:1993. 710 "Information Technology -- Universal Multiple-Octet Coded Character 711 Set (UCS)-Part 1: Architecture and Basic Multilingual Plane." Because 712 Unicode is a 16 bit encoding, it may be more efficient for the NFS 713 version 4 protocol to use an encoding for wire transfer that will be 714 useful for a majority of usage. One possible encoding is the UCS 715 transformation format (UTF). UTF-8 is an encoding method for UCS-4 716 characters which allows for the direct encoding of US-ASCII 717 characters but expands for the correct encoding of the full UCS-4 31 718 bit definitions. Currently, the UCS-4 and Unicode standards do not 719 diverge. 721 This Unicode/UTF-8 encoding can be used for places in the protocol 722 that a traditional string representation is needed. This includes 723 file and directory names along with symlink contents. This should 724 also include other file and directory attributes that are eventually 725 defined as strings. 727 The Unicode standard is applicable to the well defined strings within 728 the protocol. Dealing with file content is much more difficult. NFS 729 has traditionally dealt with file data as an opaque byte stream. No 730 other structure or content specification has been levied upon the 731 file content. The main advantage to this approach is its flexibility. 732 This byte stream can contain any data content and may be accessed in 733 any sequential or random fashion. Unfortunately, it is left to the 734 application or user to make the determination of file content and 735 format. It is possible to construct a mechanism in the protocol that 736 specifies file data type while maintaining the byte stream model for 737 data access. However, this approach may be limiting in ways unclear 738 to the designers of the NFS version 4 protocol. An expandable and 739 adaptable approach is to use the previously discussed extended 740 attributes as the mechanism to specify file content and format. The 741 use of extended attributes allows for future definition and growth as 742 various data types are created and allows for maintaining a simple 743 file data model for the NFS protocol. 745 It should be noted that as the Unicode standards are currently 746 defined there is the possibility for minor inconsistencies when 747 converting from local character representations to Unicode and then 748 back again. This should not be a problem with single client and 749 server interaction but may become apparent with the interaction of 750 two or more clients with separate conversion implementations. 751 Therefore, as NFS version 4 progresses in its development, these 752 types of Unicode issues need to be tracked and understood for their 753 potential impact on client/server interaction. In any case, Unicode 754 seems to be the best selection for NFS version 4 based on its 755 standards background and apparent future direction. 757 10. Security Considerations 759 Two previous sections within this document deal with security issues. 760 The section covering 'Access Control Lists' covers the mechanisms 761 that need to be investigated for file system level control. The 762 section that covers RPC security deals with individual user 763 authentication along with data integrity and privacy issues. This 764 section also covers negotiation of security mechanisms. These 765 sections should be consulted for additional discussion and detail. 767 10.1. Denial of Service 769 As with all services, the denial of service by either incorrect 770 implementations or malicious agents is always a concern. With the 771 target of providing NFS version 4 for Internet use, it is all the 772 more important that all aspects of the NFS version 4 protocol be 773 reviewed for potential denial of service scenarios. When found these 774 potential problems should be mitigated as much as possible. 776 11. Bibliography 778 [RFC1094] 779 Sun Microsystems, Inc., "NFS: Network File System Protocol 780 Specification", RFC1094, March 1989. 782 http://www.ietf.org/rfc/rfc1094.txt 784 [RFC1510] 785 Kohl, J., Neuman, C., "The Kerberos Network Authentication Service 786 (V5)", RFC1510, Digital Equipment Corporation, ISI, September 1993. 788 http://www.ietf.org/rfc/rfc1510.txt 790 [RFC1813] 791 Callaghan, B., Pawlowski, B., Staubach, P., "NFS Version 3 Protocol 792 Specification", RFC1813, Sun Microsystems, Inc., June 1995. 794 http://www.ietf.org/rfc/rfc1813.txt 796 [RFC1831] 797 Srinivasan, R., "RPC: Remote Procedure Call Protocol Specification 798 Version 2", RFC1831, Sun Microsystems, Inc., August 1995. 800 http://www.ietf.org/rfc/rfc1831.txt 802 [RFC1832] 803 Srinivasan, R., "XDR: External Data Representation Standard", 804 RFC1832, Sun Microsystems, Inc., August 1995. 806 http://www.ietf.org/rfc/rfc1832.txt 808 [RFC1833] 809 Srinivasan, R., "Binding Protocols for ONC RPC Version 2", RFC1833, 810 Sun Microsystems, Inc., August 1995. 812 http://www.ietf.org/rfc/rfc1833.txt 814 [RFC2025] 815 Adams, C., "The Simple Public-Key GSS-API Mechanism (SPKM)", RFC2025, 816 Bell-Northern Research, October 1996. 818 http://www.ietf.org/rfc/rfc2025.txt 820 [RFC2054] 821 Callaghan, B., "WebNFS Client Specification", RFC2054, Sun 822 Microsystems, Inc., October 1996 824 http://www.ietf.org/rfc/rfc2054.txt 826 [RFC2055] 827 Callaghan, B., "WebNFS Server Specification", RFC2054, Sun 828 Microsystems, Inc., October 1996 830 http://www.ietf.org/rfc/rfc2055.txt 832 [RFC2078] 833 Linn, J., "Generic Security Service Application Program Interface, 834 Version 2", RFC2078, OpenVision Technologies, January 1997. 836 http://www.ietf.org/rfc/rfc2078.txt 838 [RFC2152] 839 Goldsmith, D., "UTF-7 A Mail-Safe Transformation Format of Unicode", 840 RFC2152, Apple Computer, Inc., May 1997 842 http://www.ietf.org/rfc/rfc2152.txt 844 [RFC2203] 845 Eisler, M., Chiu, A., Ling, L., "RPCSEC_GSS Protocol Specification", 846 RFC2203, Sun Microsystems, Inc., August 1995. 848 http://www.ietf.org/rfc/rfc2203.txt 850 [RFC2222] 851 Myers, J., "Simple Authentication and Security Layer (SASL)", 852 RFC2222, Netscape Communications, October 1997. 854 http://www.ietf.org/rfc/rfc2222.txt 856 [RFC2279] 857 Yergeau, F., "UTF-8, a transformation format of ISO 10646", RFC2279, 858 Alis Technologies, January 1998. 860 http://www.ietf.org/rfc/rfc2279.txt 862 [RFC2246] 863 Dierks, T., Allen, C. "The TLS Protocols Version 1.0", RFC 2246, 864 Certicom, January 1999. 866 http://www.ietf.org/rfc/rfc2246.txt 868 [RFC2401] 869 Kent, S., Atkinson, R., "Security Architecture for the Internet 870 Protocol", RFC2401, BBN Corp., @Home Network, November 1998. 872 http://www.ietf.org/rfc/rfc2401.txt 874 [DCEACL] 875 The Open Group, Open Group Technical Standard, "DCE 1.1: 876 Authentication and Security Services," Document Number C311, August 877 1997. Provides a discussion of DEC ACL structure and semantics. 879 [HPFS] 880 Les Bell and Associates Pty Ltd, "The HPFS FAQ," 881 http://www.lesbell.com.au/hpfsfaq.html 883 [Hutson] 884 Huston, L.B., Honeyman, P., "Disconnected Operation for AFS," June 885 1993. Proc. USENIX Symp. on Mobile and Location-Independent 886 Computing, Cambridge, August 1993. 888 [Kistler] 889 Kistler, James J., Satyanarayanan, M., "Disconnected Operations in 890 the Coda File System," ACM Trans. on Computer Systems, vol. 10, no. 891 1, pp. 3-25, Feb. 1992. 893 [Mummert] 894 Mummert, L. B., Ebling, M. R., Satyanarayanan, M., "Exploiting Weak 895 Connectivity for Mobile File Access," Proc. of the 15th ACM Symp. on 896 Operating Systems Principles Dec. 1995. 898 [Nagar] 899 Nagar, R., "Windows NT File System Internals," ISBN 1565922492, 900 O`Reilly & Associates, Inc. 902 [Sandberg] 903 Sandberg, R., D. Goldberg, S. Kleiman, D. Walsh, B. Lyon, "Design 904 and Implementation of the Sun Network Filesystem," USENIX Conference 905 Proceedings, USENIX Association, Berkeley, CA, Summer 1985. The 906 basic paper describing the SunOS implementation of the NFS version 2 907 protocol, and discusses the goals, protocol specification and trade- 908 offs. 910 [Satyanarayanan1] 911 Satyanarayanan, M., "Fundamental Challenges in Mobile Computing," 912 Proc. of the ACM Principles of Distributed Computing, 1995. 914 [Satyanarayanan2] 915 Satyanarayanan, M., Kistler, J. J., Mummert L. B., Ebling M. R., 916 Kumar, P. , Lu, Q., "Experience with disconnected operation in 917 mobile computing environment," Proc. of the USENIX Symp. on Mobile 918 and Location-Independent Computing, Jun. 1993. 920 [Unicode1] 921 "Unicode Technical Report #8 - The Unicode Standard, Version 2.1", 922 Unicode, Inc., The Unicode Consortium, P.O. Box 700519, San Jose, CA 923 95710-0519 USA, September 1998 925 http://www.unicode.org/unicode/reports/tr8.html 927 [Unicode2] 928 "Unsupported Scripts" Unicode, Inc., The Unicode Consortium, P.O. Box 929 700519, San Jose, CA 95710-0519 USA, October 1998 931 http://www.unicode.org/unicode/standard/unsupported.html 933 [XDSM] 934 The Open Group, Open Group Technical Standard, "Systems Management: 935 Data Storage Management (XDSM) API," ISBN 1-85912-190-X, January 936 1997. 938 [XNFS] 939 The Open Group, Protocols for Interworking: XNFS, Version 3W, The 940 Open Group, 1010 El Camino Real Suite 380, Menlo Park, CA 94025, ISBN 941 1-85912-184-5, February 1998. 943 HTML version available: http://www.opengroup.org 945 12. Acknowledgments 947 o Brent Callaghan for content contributions. 949 13. Author's Address 951 Address comments related to this memorandum to: 953 spencer.shepler@eng.sun.com -or- nfsv4-wg@sunroof.eng.sun.com 955 Spencer Shepler 956 Sun Microsystems, Inc. 957 7808 Moonflower Drive 958 Austin, Texas 78750 960 Phone: (512) 349-9376 961 E-mail: spencer.shepler@eng.sun.com 963 14. Full Copyright Statement 965 "Copyright (C) The Internet Society (1999). All Rights Reserved. 967 This document and translations of it may be copied and furnished to 968 others, and derivative works that comment on or otherwise explain it 969 or assist in its implmentation may be prepared, copied, published and 970 distributed, in whole or in part, without restriction of any kind, 971 provided that the above copyright notice and this paragraph are 972 included on all such copies and derivative works. However, this 973 document itself may not be modified in any way, such as by removing 974 the copyright notice or references to the Internet Society or other 975 Internet organizations, except as needed for the purpose of 976 developing Internet standards in which case the procedures for 977 copyrights defined in the Internet Standards process must be 978 followed, or as required to translate it into languages other than 979 English. 981 The limited permissions granted above are perpetual and will not be 982 revoked by the Internet Society or its successors or assigns. 984 This document and the information contained herein is provided on an 985 "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING 986 TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING 987 BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION 988 HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF 989 MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE."