idnits 2.17.1 draft-ietf-nfsv4-federated-fs-reqts-01.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- ** The document seems to lack a License Notice according IETF Trust Provisions of 28 Dec 2009, Section 6.b.i or Provisions of 12 Sep 2009 Section 6.b -- however, there's a paragraph with a matching beginning. Boilerplate error? -- It seems you're using the 'non-IETF stream' Licence Notice instead Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document 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 (December 23, 2008) is 5603 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- ** Obsolete normative reference: RFC 3530 (Obsoleted by RFC 7530) ** Obsolete normative reference: RFC 5246 (Obsoleted by RFC 8446) Summary: 3 errors (**), 0 flaws (~~), 1 warning (==), 3 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 NFSv4 Working Group J. Lentini 3 Internet-Draft C. Everhart 4 Intended status: Informational NetApp 5 Expires: June 26, 2009 D. Ellard 6 BBN Technologies 7 R. Tewari 8 M. Naik 9 IBM Almaden 10 December 23, 2008 12 Requirements for Federated File Systems 13 draft-ietf-nfsv4-federated-fs-reqts-01 15 Status of this Memo 17 This Internet-Draft is submitted to IETF in full conformance with the 18 provisions of BCP 78 and BCP 79. 20 Internet-Drafts are working documents of the Internet Engineering 21 Task Force (IETF), its areas, and its working groups. Note that 22 other groups may also distribute working documents as Internet- 23 Drafts. 25 Internet-Drafts are draft documents valid for a maximum of six months 26 and may be updated, replaced, or obsoleted by other documents at any 27 time. It is inappropriate to use Internet-Drafts as reference 28 material or to cite them other than as "work in progress." 30 The list of current Internet-Drafts can be accessed at 31 http://www.ietf.org/ietf/1id-abstracts.txt. 33 The list of Internet-Draft Shadow Directories can be accessed at 34 http://www.ietf.org/shadow.html. 36 This Internet-Draft will expire on June 26, 2009. 38 Copyright Notice 40 Copyright (c) 2008 IETF Trust and the persons identified as the 41 document authors. All rights reserved. 43 This document is subject to BCP 78 and the IETF Trust's Legal 44 Provisions Relating to IETF Documents 45 (http://trustee.ietf.org/license-info) in effect on the date of 46 publication of this document. Please review these documents 47 carefully, as they describe your rights and restrictions with respect 48 to this document. 50 Abstract 52 This document describes and lists the functional requirements of a 53 federated file system and defines related terms. 55 Table of Contents 57 1. Requirements notation . . . . . . . . . . . . . . . . . . . . 3 58 2. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 59 3. Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 60 4. Examples and Discussion . . . . . . . . . . . . . . . . . . . 7 61 4.1. Create a Fileset and its FSL(s) . . . . . . . . . . . . . 7 62 4.1.1. Creating a Fileset and an FSN . . . . . . . . . . . . 7 63 4.1.2. Adding a Replica of a Fileset . . . . . . . . . . . . 8 64 4.2. Junction Resolution . . . . . . . . . . . . . . . . . . . 8 65 4.3. Junction Creation . . . . . . . . . . . . . . . . . . . . 10 66 5. Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 67 6. Proposed Requirements . . . . . . . . . . . . . . . . . . . . 15 68 6.1. Basic Assumptions . . . . . . . . . . . . . . . . . . . . 15 69 6.2. Requirements . . . . . . . . . . . . . . . . . . . . . . . 18 70 7. Non-Requirements . . . . . . . . . . . . . . . . . . . . . . . 24 71 8. Security Considerations . . . . . . . . . . . . . . . . . . . 25 72 9. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 26 73 10. References . . . . . . . . . . . . . . . . . . . . . . . . . . 27 74 10.1. Normative References . . . . . . . . . . . . . . . . . . . 27 75 10.2. Informational References . . . . . . . . . . . . . . . . . 27 76 Appendix A. Acknowledgements . . . . . . . . . . . . . . . . . . 28 77 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 29 79 1. Requirements notation 81 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 82 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 83 document are to be interpreted as described in [RFC2119]. 85 Note, that this is a requirements document, and in many instances 86 where these words are used in this document they refer to qualities 87 of a specification for a system that satisfies the document, or 88 requirements of a system that matches that specification. These 89 cases are distinguished when there is potential for ambiguity. 91 2. Overview 93 This document describes and lists the functional requirements of a 94 federated file system and defines related terms. 96 We do not describe the mechanisms that might be used to implement 97 this functionality except in cases where specific mechanisms, in our 98 opinion, follow inevitably from the requirements. Our focus is on 99 the interfaces between the entities of the system, not on the 100 protocols or their implementations. 102 Today, there are collections of fileservers that inter-operate to 103 provide a single namespace comprised of filesystem resources provided 104 by different members of the collection, joined together with inter- 105 filesystem references. The namespace can either be assembled at the 106 fileservers, the clients, or by an external namespace service, and is 107 often not easy or uniform to manage. The requirements in this draft 108 are meant to lead to a uniform server-based namespace that is capable 109 of spanning a whole enterprise and which is easy to manage. 111 We define some terms to better describe the solution space. A 112 "fileset" is the abstract view of a filesystem in a uniform 113 namespace, and may be implemented behind that abstraction by one or 114 more physical filesystems at any given time. Each fileset has a name 115 called an "FSN" (fileset name), and each physical filesystem has a 116 fileset location ("FSL"). A fileset is a directory tree containing 117 files and directories, and it may also contain references to other 118 filesets. These references are called "junctions". To provide 119 location independence, a junction does not contain information about 120 the location of the real resource(s), but instead contains an FSN 121 that can be used to look up the location information. The service 122 that can be used to map from FSN to FSL(s) is called a namespace 123 database (NSDB) service. The NSDB provides a level of indirection 124 from the virtual paths in the uniform namespace to the actual 125 locations of files. 127 The servers direct clients to the proper locations by existing 128 mechanisms (e.g. the referrals mechanism within [RFC3530] and 129 [NFSv4.1 RFC TBD]). Updates to the locations make it possible to 130 support migration and replication of physical filesystems that 131 comprise the namespace, in a way that is transparent to filesystem 132 clients. 134 Figure 1 shows an example of a federation. This federation has two 135 members, named ALPHA and BETA. Federation members may contain an 136 arbitrary number of file servers and NSDB nodes; in this illustration 137 ALPHA and BETA each have three servers and one NSDB node. 139 +----------------------+ +----------------------+ 140 | Federation Member | | Federation Member | 141 | ALPHA | | BETA | 142 | | | | 143 | | | | 144 | +------------+ | | +------------+ | 145 | | NSDB | | | | NSDB | | 146 | | | | | | | | 147 | +------------+ | | +------------+ | 148 | | | | 149 | | | | 150 | | | | 151 | +----------+ | | +----------+ | 152 | | | | | | | | 153 | +-- | Servers | | | +-- | Servers | | 154 | | | | | | | | | | 155 | +-- | | | | | +-- | | | | 156 | | | +----------+ | | | | +----------+ | 157 | | | | | | | | | | 158 | | +----------+ | | | +----------+ | 159 | | | | | | | | 160 | +----------+ | | +----------+ | 161 +----------------------+ +----------------------+ 163 A federation with two members, ALPHA and BETA. ALPHA and BETA each 164 have their own NSDB node and several file servers, and they are 165 administered separately. 167 Figure 1 169 3. Purpose 171 Our objective is to specify a set of protocols by which fileservers 172 or collections of fileservers, with different administrators, can 173 form a federation of fileservers and NSDB nodes that provides a 174 namespace composed of the filesets hosted on the different 175 fileservers and fileserver collections. 177 It should be possible, using a system that implements the protocols, 178 to share a common namespace across all the fileservers in the 179 federation. It should also be possible for different fileservers in 180 the federation to project different namespaces and enable clients to 181 traverse them. 183 Such a federation may contain an arbitrary number of NSDB nodes, each 184 belonging to a different administrative entity, and each providing 185 the mappings that define a part of a namespace. Such a federation 186 may also have an arbitrary number of administrative entities, each 187 responsible for administering a subset of the fileservers and NSDB 188 nodes. Acting in concert, the administrators should be able to build 189 and administer this multi-fileserver, multi-collection namespace. 191 It is not the intent of the federation to guarantee namespace 192 consistency across all client views. Since different parts of the 193 namespace may be administered by different entities, it is possible 194 that a client could be accessing a stale area of the namespace 195 managed by one entity because a part of the namespace above it, 196 managed by another entity, has changed. 198 4. Examples and Discussion 200 In this section we provide examples and discussion of the basic 201 operations facilitated by the federated file system protocol: 202 creating a fileset, adding a replica of a fileset, resolving a 203 junction, and creating a junction. 205 4.1. Create a Fileset and its FSL(s) 207 A fileset is the abstraction of a set of files and their containing 208 directory tree. The fileset abstraction is the fundamental unit of 209 data management in the federation. This abstraction is implemented 210 by an actual directory tree whose root location is specified by a 211 fileset location (FSL). 213 In this section, we describe the basic requirements for starting with 214 a directory tree and creating a fileset that can be used in the 215 federation protocols. Note that we do not assume that the process of 216 creating a fileset requires any transformation of the files or the 217 directory hierarchy. The only thing that is required by this process 218 is assigning the fileset a fileset name (FSN) and expressing the 219 location(s) of the implementation of the fileset as FSL(s). 221 There are many possible variations to this procedure, depending on 222 how the FSN that binds the FSL is created, and whether other replicas 223 of the fileset exist, are known to the federation, and need to be 224 bound to the same FSN. 226 It is easiest to describe this in terms of how to create the initial 227 implementation of the fileset, and then describe how to add replicas. 229 4.1.1. Creating a Fileset and an FSN 231 1. Choose the NSDB node that will keep track of the FSL(s) and 232 related information for the fileset. 234 2. Request that the NSDB node register a new FSN for the fileset. 236 The FSN may either be chosen by the NSDB node or by the server. 237 The latter case is used if the fileset is being restored, perhaps 238 as part of disaster recovery, and the server wishes to specify 239 the FSN in order to permit existing junctions that reference that 240 FSN to work again. 242 At this point, the FSN exists, but its location is unspecified. 244 3. Send the FSN, the local volume path, the export path, and the 245 export options for the local implementation of the fileset to the 246 NSDB node. Annotations about the FSN or the location may also be 247 sent. 249 The NSDB node records this info and creates the initial FSL for 250 the fileset. 252 4.1.2. Adding a Replica of a Fileset 254 Adding a replica is straightforward: the NSDB node and the FSN are 255 already known. The only remaining step is to add another FSL. 257 Note that the federation protocols do not include methods for 258 creating or managing replicas: this is assumed to be a platform- 259 dependent operation (at least at this time). The only requirement is 260 that these fileset replicas be registered and unregistered with the 261 NSDB. 263 4.2. Junction Resolution 265 A fileset may contain references to other filesets. These references 266 are represented by junctions. If a client requests access to a 267 fileset object that is a junction, the server resolves the junction 268 to discover the FSL(s) that implements the referenced fileset. 270 There are many possible variations to this procedure, depending on 271 how the junctions are represented and how the information necessary 272 to perform resolution is represented by the server. 274 Step 4 is the only step that interacts directly with the federation 275 protocols. The rest of the steps may use platform-specific 276 interfaces. 278 1. The server determines that the object being accessed is a 279 junction. 281 2. Using the junction, the server does a local lookup to find the 282 FSN of the target fileset. 284 3. Using the FSN, the server finds the NSDB node responsible for the 285 target object. 287 4. The server contacts that NSDB node and asks for the set of FSLs 288 that implement the target FSN. The NSDB node responds with a set 289 of FSLs. 291 5. The server converts one of the FSLs to the location type used by 292 the client (e.g., fs_location for NFSv4, as described in 293 [RFC3530]). 295 6. The server redirects (in whatever manner is appropriate for the 296 client) the client to the location(s). 298 These steps are illustrated in Figure 2. The client sends request 1 299 to server X, in federation member ALPHA, in an attempt to reference 300 an object (which appears to the client as a directory). Server X 301 recognizes that the referenced object is actually a junction that 302 refers to a directory in a different fileset. Server X finds, from 303 the FSN in the junction, that the NSDB responsible for knowing the 304 location of the target of the junction is the NSDB of federation 305 member BETA. Server X sends request 2 to the NSDB of BETA, asking 306 for the current location of the directory. The NSDB sends response 3 307 to server X, telling the server that the directory is located on 308 server Y. Server X sends response 4 to the client, indicating that 309 the directory is in a "new" location on server Y. The client then 310 sends request 5 to server Y, repeating the initial request. 312 Given the current requirements and definitions, this resolution 313 method MUST work. However, there is no requirement that this is the 314 only resolution method that can be used. This method may be used as 315 the fallback when all else fails (or, for a simple implementation, it 316 could be the only method). This is a degenerate implementation of 317 the NSDB service as a simple composition of NSDB nodes; we expect 318 that large federations will use more sophisticated methods to share 319 the FSN and FSL information among multiple NSDB nodes. 321 +---------------+ 322 | | 323 | Client | >--------------------------+ 324 | | | 325 +---------------+ | 326 v ^ | 327 +-----+---+-------------+ +-----------------+-----+ 328 | | | Federation| |Federation | | 329 | | | member | |member | | 330 | | | ALPHA | |BETA | | 331 | | | | | | | 332 | | | | | | | 333 | | | | | | | 334 | | | | | | | 335 | | | | | +---------+ | | 336 | | | +---------+------+-> | | | | 337 | | | | | | | NSDB Y | | | 338 | | | | +-----+------+-< | | | | 339 | | | | | | | +---------+ | | 340 | | | | | | | | | 341 | | | | | | | | | 342 | | | | | | | | | 343 | 1| 4| 2| 3| | | 5| | 344 | v ^ ^ v | | v | 345 | +---------------+ | | +---------------+ | 346 | | | | | | | | 347 | | Server X | | | | Server Y | | 348 | | | | | | | | 349 | +---------------+ | | +---------------+ | 350 | | | | 351 +-----------------------+ +-----------------------+ 353 Figure 2 355 4.3. Junction Creation 357 Given a local path, a remote export and a path relative to that 358 export, create a junction from the local path to the path within the 359 remote export. 361 There are many possible variations to this procedure, depending on 362 how the junctions are represented and how the information necessary 363 to perform resolution is represented by the server. 365 Step 1 is the only step that uses the federation interfaces. The 366 rest of the steps may use platform-specific interfaces. 368 1. Contact the server named by the export and ask for the FSN for 369 the fileset, given its path relative to that export. 371 2. Insert the junction to the FSN, at the given path, into the local 372 filesystem. 374 5. Glossary 376 Administrator: user with the necessary authority to initiate 377 administrative tasks on one or more servers. 379 Admin entity: A server or agent that administers a collection of 380 fileservers and persistently stores the namespace information. 382 Client: Any client that accesses the fileserver data using a 383 supported filesystem access protocol. 385 Federation: A set of server collections and singleton servers that 386 use a common set of interfaces and protocols in order to provide 387 to their clients a federated namespace accessible through a 388 filesystem access protocol. 390 Fileserver: A server exporting a filesystem via a network filesystem 391 access protocol. 393 Fileset: The abstraction of a set of files and their containing 394 directory tree. A fileset is the fundamental unit of data 395 management in the federation. 397 Note that all files within a fileset are descendants of one 398 directory, and that filesets do not span filesystems. 400 Filesystem: A self-contained unit of export for a fileserver, and 401 the mechanism used to implement filesets. The fileset does not 402 need to be rooted at the root of the filesystem, nor at the export 403 point for the filesystem. 405 A single filesystem MAY implement more than one fileset, if the 406 client protocol and the fileserver permit this. 408 Filesystem access protocol: A network filesystem access protocol 409 such as NFSv2 [RFC1094], NFSv3 [RFC1813], NFSv4 [RFC3530], or 410 CIFS. 412 FSL (Fileset location): The location of the implementation of a 413 fileset at a particular moment in time. A FSL MUST be something 414 that can be translated into a protocol-specific description of a 415 resource that a client can access directly, such as a fs_location 416 (for NFSv4), or share name (for CIFS). Note that not all FSLs 417 need to be explicitly exported as long as they are contained 418 within an exported path on the fileserver. 420 FSN (Fileset name): A platform-independent and globally unique name 421 for a fileset. Two FSLs that implement replicas of the same 422 fileset MUST have the same FSN, and if a fileset is migrated from 423 one location to another, the FSN of that fileset MUST remain the 424 same. 426 Junction: A filesystem object used to link a directory name in the 427 current fileset with an object within another fileset. The 428 server-side "link" from a leaf node in one fileset to the root of 429 another fileset. 431 Namespace: A filename/directory tree that a sufficiently-authorized 432 client can observe. 434 NSDB (Namespace Database Service): A service that maps FSNs to FSLs. 435 The NSDB may also be used to store other information, such as 436 annotations for these mappings and their components. 438 NSDB Node: The name or location of a server that implements part of 439 the NSDB service and is responsible for keeping track of the FSLs 440 (and related info) that implement a given partition of the FSNs. 442 Referral: A server response to a client access that directs the 443 client to evaluate the current object as a reference to an object 444 at a different location (specified by an FSL) in another fileset, 445 and possibly hosted on another fileserver. The client re-attempts 446 the access to the object at the new location. 448 Replica: A replica is a redundant implementation of a fileset. Each 449 replica shares the same FSN, but has a different FSL. 451 Replicas may be used to increase availability or performance. 452 Updates to replicas of the same fileset MUST appear to occur in 453 the same order, and therefore each replica is self-consistent at 454 any moment. 456 We do not assume that updates to each replica occur simultaneously 457 If a replica is offline or unreachable, the other replicas may be 458 updated. 460 Server Collection: A set of fileservers administered as a unit. A 461 server collection may be administered with vendor-specific 462 software. 464 The namespace provided by a server collection could be part of the 465 federated namespace. 467 Singleton Server: A server collection containing only one server; a 468 stand-alone fileserver. 470 6. Proposed Requirements 472 The phrase "USING THE FEDERATION INTERFACES" implies that the 473 subsequent requirement must be satisfied, in its entirety, via the 474 federation interfaces. 476 Note that the requirements are described in terms of correct behavior 477 by all entities. We do not address the requirements of the system in 478 the presence of faults. 480 6.1. Basic Assumptions 482 Several of the requirements are so fundamental that we treat them as 483 basic assumptions; if any of these assumptions are violated, the rest 484 of the requirements must be reviewed in their entirety. 486 A1: The federation protocols do not require any changes to existing 487 client-facing protocols, and MAY be extended to incorporate new 488 client-facing protocols. 490 A2: A client SHOULD NOT require any a priori knowledge of the 491 general structure or composition of the federation. 493 The client may require some specific knowledge in order to find 494 and access an instance of the fileset that defines the root of 495 its view of the namespace. As the client traverses the 496 namespace, the client discovers the information it needs in 497 order to locate the filesets it accesses. 499 A3: All requirements MUST be satisfiable via the federation 500 protocols and the standard protocols used by the fileservers 501 (i.e., NFS, CIFS, DNS, etc). 503 USING THE FEDERATION INTERFACES, a federation operation that 504 requires an interaction between two (or more) entities that are 505 members of the federation MUST be possible without requiring any 506 proprietary protocols. 508 A4: All the entities participating in a federation operation MUST be 509 able to authenticate each other. 511 All principals (clients, users, administrator of a singleton or 512 server collection, hosts, NSDB nodes, etc) that can assume a 513 role defined by the federation protocol can identify themselves 514 to each other via an authentication mechanism. This mechanism 515 is not defined or further described in this document. 517 The authority of a principal to request that a second principal 518 perform a specific operation is ultimately determined by the 519 second. Authorization may be partitioned by server collection 520 or set of servers as well as by operation. For example, if a 521 user has administrative privileges on one server in the 522 federation, this does not imply that they have administrative 523 privileges (or, for that matter, any privileges whatsoever) on 524 any other server in the federation. 526 In order to access the functionality provided by the federation 527 interfaces, it may be necessary to have elevated privileges or 528 authorization. The authority required by different operations 529 may be different. For example, the authority required to query 530 the NSDB about the FSLs bound to an FSN may be different than 531 the authority required to change the bindings of that FSN. 533 An operation attempted by an unauthorized entity MUST fail in a 534 manner that indicates that the failure was due to insufficient 535 authorization. 537 This document does not enumerate the authorization necessary for 538 any operation. 540 A5: The federation protocols MUST NOT require changes to existing 541 authentication/authorization mechanisms in use at the 542 fileservers for client-facing protocols. 544 A user's view of the namespace may be limited by the 545 authentication and authorization privileges it has on the 546 different fileservers in the federation. As such, users may 547 only be able to traverse the parts of the namespace that they 548 have access to. 550 The federation protocols do not impose any restrictions on how 551 users are represented within the federation. For example, a 552 single enterprise could employ a common identity for users 553 across the federation. A grid environment could utilize user 554 mapping or translations across different administrative domains. 556 A6: In a federated system, we assume that an FSN MUST express, or 557 can be used to discover, the following two pieces of 558 information: 560 1. The location of the NSDB node that is responsible for 561 knowing the filesystem location(s) (FSLs) of the named 562 fileset. 564 The NSDB node must be specified because there may be many 565 NSDB nodes in a federation. We do not assume that any 566 single entity knows the location of all of the NSDB nodes, 567 and therefore exhaustive search is not an option. 569 There are several ways in which a fileserver can locate the 570 NSDB node responsible for a given fileset. One approach, 571 given a DNS infrastructure, is to specify the location of 572 the NSDB node by the FQDN of the server hosting the NSDB 573 node. Another approach is to use a separate DNS-style 574 hierarchy to resolve the location of the NSDB node. 576 2. The FSN identifier. 578 The FSN identifier is the index used by the NSDB node to 579 identify the target fileset. 581 There are several ways to represent FSN identifiers. One 582 approach could use 128-bit UUIDs as described described in 583 [RFC4122]. 585 As an example, an FSN could be represented by a URL of the form 586 nsdb.example.com/UUID where nsdb.example.com is the FQDN of the 587 server hosting the NSDB node and UUID is the string 588 representation of the identifier. 590 Note that it is not assumed that it is always required for a 591 server to contact the NSDB node specified by the FSN in order to 592 find the FSLs. The relevant information stored in that NSDB 593 node may also be cached local to the server or on a proxy NSDB 594 node "near" the server. 596 A7: All federation servers and NSDB nodes are assumed to execute the 597 federation protocols correctly. The behavior of the federation 598 is undefined in the case of Byzantine behavior by any federation 599 server or NSDB node. 601 A8: The locations of federation services (such as NSDBs and FSLs) 602 can be specified in a manner such that they can be correctly 603 interpreted by all members of the federation that will access 604 them. 606 For example, if an NSDB node is specified by a FQDN, then this 607 implies that every member of the federation that needs to access 608 this NSDB node can resolve this FQDN to an IP address for that 609 NSDB node. (It is not necessary that the FQDN always resolve to 610 the same address; the same service may appear at different 611 addresses on different networks.) 613 It is the responsibility of each federation member to ensure 614 that the resources it wishes to expose have accessible network 615 locations and that the necessary resolution mechanisms (i.e., 616 DNS) are given the necessary data to perform the resolution 617 correctly. 619 6.2. Requirements 621 R1: Requirements of each FSN: 623 a. Each FSN MUST be unique within the scope of its NSDB (so 624 that the FSN is globally unique). 626 b. The FSN MUST be sufficiently descriptive to locate an 627 instance of the fileset it names within the federation at 628 any time. 630 c. All FSNs MUST be invariant when their underlying 631 filesystems move or are replicated; only mappings from FSN 632 to FSL(s) change under these transformations. 634 d. All files accessible from the global namespace MUST be part 635 of a fileset that has an assigned FSN. 637 Not all filesets in the federation are required to have an FSN 638 or be reachable by an FSL. Only those filesets that are the 639 target of a junction (as described in R3) are required to have 640 an FSN. 642 R2: USING THE FEDERATION INTERFACES, it MUST be possible to create 643 an FSN for a fileset, and it must be possible to bind an FSL to 644 that FSN. These operations are NSDB operations and do not 645 require any action on the part of a file server. 647 It is possible to create an FSN for a fileset that has not 648 actually been created. It is also possible to bind a 649 nonexistant FSL to an FSN. It is also possible to create a 650 fileset without assigning it an FSN. The binding between an 651 FSN and an FSL is defined entirely within the context of the 652 NSDB; the servers do not "know" whether the filesets they host 653 have been assigned FSNs (or, if so, what those FSNs are). 655 The requirement that filesets can exist prior to being assigned 656 an FSN, and the requirement that FSNs can exist independent of 657 filesets are intended to simplify the construction of the 658 namespace in a convenient manner. For example, they permit an 659 admin to assign FSNs to existing filesets and thereby 660 incorporate existing filesets into the namespace. They also 661 permit the structure of the namespace to be defined prior to 662 creation of the component filesets. In either case, it is the 663 responsibility of the entity updating the NSDB with FSNs and 664 FSN-to-FSL mappings to ensure that the namespace is constructed 665 in a consistent manner. (The simplest way to accomplish this 666 is to ensure that the FSN and FSN-to-FSL mappings are always 667 recorded in the NSDB prior to the creation of any junctions 668 that refer to that FSN.) 670 a. An administrator MAY specify the entire FSN (including both 671 the NSDB node location and the identifier) of the newly- 672 created FSL, or the administrator MAY specify only the NSDB 673 node and have the system choose the identifier. 675 The admin can choose to specify the FSN explicitly in order 676 to recreate a lost fileset with a given FSN (for example, 677 as part of disaster recovery). It is an error to assign an 678 FSN that is already in use by an active fileset. 680 Note that creating a replica of an existing filesystem is 681 NOT accomplished by assigning the FSN of the filesystem you 682 wish to replicate to a new filesystem. 684 b. USING THE FEDERATION INTERFACES, it MUST be possible to 685 create a federation FSL by specifying a specific local 686 volume, path, export path, and export options. 688 R3: USING THE FEDERATION INTERFACES, and given the FSN of a target 689 fileset, it MUST be possible to create a junction to that 690 fileset at a named place in another fileset. 692 After a junction has been created, clients that access the 693 junction transparently interpret it as a reference to the 694 FSL(s) that implement the FSN associated with the junction. 696 a. It SHOULD be possible to have more than one junction whose 697 target is a given fileset. In other words, it SHOULD be 698 possible to mount a fileset at multiple named places. 700 b. If the fileset in which the junction is created is 701 replicated, then the junction MUST eventually appear in all 702 of its replicas. 704 The operation of creating a junction within a fileset is 705 treated as an update to the fileset, and therefore obey the 706 general rules about updates to replicated filesets. 708 R4: USING THE FEDERATION INTERFACES, it MUST be possible to delete 709 a specific junction from a fileset. 711 If a junction is deleted, clients who are already viewing the 712 fileset referred to by the junction after traversing the 713 junction MAY continue to view the old namespace. They might 714 not discover that the junction no longer exists (or has been 715 deleted and replaced with a new junction, possibly referring to 716 a different FSN). 718 After a junction is deleted, another object with the same name 719 (another junction, or an ordinary filesystem object) may be 720 created. 722 The operation of deleting a junction within a fileset is 723 treated as an update to the fileset, and therefore obey the 724 general rules about updates to replicated filesets. 726 R5: USING THE FEDERATION INTERFACES, it MUST be possible to 727 invalidate an FSN. 729 a. If a junction refers to an FSN that is invalid, attempting 730 to traverse the junction MUST fail. 732 An FSN that has been invalidated MAY become valid again if the 733 FSN is recreated (i.e., as part of a disaster recovery 734 process). 736 If an FSN is invalidated, clients who are already viewing the 737 fileset named by the FSN MAY continue to view the old 738 namespace. They might not discover that the FSN is no longer 739 valid until they try to traverse a junction that refers to it. 741 R6: USING THE FEDERATION INTERFACES, it MUST be possible to 742 invalidate an FSL. 744 a. An invalid FSL MUST NOT be returned as the result of 745 resolving a junction. 747 An FSL that has been invalidated MAY become valid again if the 748 FSL is recreated (i.e., as part of a disaster recovery 749 process). 751 If an FSL is invalidated, clients who are already viewing the 752 fileset implemented by the FSL MAY continue to use that FSL. 753 They might not discover that the FSL is no longer valid until 754 they try to traverse a junction that refers to the fileset 755 implemented by the FSL. 757 Note that invalidating an FSL does not imply that the 758 underlying export or share (depending on the file access 759 protocol in use) is changed in any way -- it only changes the 760 mappings from FSNs to FSLs on the NSDB. 762 R7: It MUST be possible for the federation of servers to provide 763 multiple namespaces. 765 R8: USING THE FEDERATION INTERFACES: 767 a. It MUST be possible to query the fileserver named in an FSL 768 to discover whether a junction exists at a given path 769 within that FSL. 771 b. It MAY be possible to query the fileserver named in an FSL 772 to discover the junctions, if any, in that FSL. If this 773 feature is implemented, the fileserver SHOULD report each 774 junction's path within the FSL and the targeted FSN. 776 R9: The projected namespace (and the objects named by the 777 namespace) MUST be accessible to clients via at least one 778 standard filesystem access protocol. 780 a. The namespace SHOULD be accessible to clients via versions 781 of the CIFS (SMB) protocol. 783 b. The namespace SHOULD be accessible to clients via the NFSv4 784 protocol as described in [RFC3530]. 786 c. The namespace SHOULD be accessible to clients via the NFSv3 787 protocol as described in [RFC1813]. 789 d. The namespace SHOULD be accessible to clients via the NFSv2 790 protocol as described in [RFC1094]. 792 It must be understood that some of these protocols, such as 793 NFSv3 and NFSv2, have no innate ability to access a namespace 794 of this kind. Where such protocols have been augmented with 795 other protocols and mechanisms (such as autofs or amd for 796 NFSv3) to provide an extended namespace, we propose that these 797 protocols and mechanisms may be used, or extended, in order to 798 satisfy the requirements given in this draft, and different 799 clients may use different mechanisms. 801 R10: USING THE FEDERATION INTERFACES, it MUST be possible to modify 802 the NSDB mapping from an FSN to a set of FSLs to reflect the 803 migration from one FSL to another. 805 R11: FSL migration SHOULD have little or no impact on the clients, 806 but this is not guaranteed across all federation members. 808 Whether FSL migration is performed transparently depends on 809 whether the source and destination servers are able to do so. 810 It is the responsibility of the administrator to recognize 811 whether or not the migration will be transparent, and advise 812 the system accordingly. The federation, in turn, MUST advise 813 the servers to notify their clients, if necessary. 815 For example, on some systems, it may be possible to migrate a 816 fileset from one system to another with minimal client impact 817 because all client-visible metadata (inode numbers, etc) are 818 preserved during migration. On other systems, migration might 819 be quite disruptive. 821 R12: USING THE FEDERATION INTERFACES, it MUST be possible to modify 822 the NSDB mapping from an FSN to a set of FSLs to reflect the 823 addition/removal of a replica at a given FSL. 825 R13: Replication SHOULD have little or no negative impact on the 826 clients. 828 Whether FSL replication is performed transparently depends on 829 whether the source and destination servers are able to do so. 830 It is the responsibility of the administrator initiating the 831 replication to recognize whether or not the replication will be 832 transparent, and advise the federation accordingly. The 833 federation MUST advise the servers to notify their clients, if 834 necessary. 836 For example, on some systems, it may be possible to mount any 837 FSL of an FSN read/write, while on other systems, there may be 838 any number of read-only replicas but only one FSL that can be 839 mounted read-write. 841 R14: USING THE FEDERATION INTERFACES, it SHOULD be possible to 842 annotate the objects and relations managed by the federation 843 protocol with arbitrary name/value pairs. 845 These annotations are not used by the federation protocols -- 846 they are intended for use by higher-level protocols. For 847 example, an annotation that might be useful for a system 848 administrator browsing the federation would be the "owner" of 849 each FSN (i.e., "this FSN is for the home directory of Joe 850 Smith."). As another example, the annotations may express 851 hints used by the clients (such as priority information for 852 NFSv4.1). 854 Both FSNs and FSLs may be annotated. For example, an FSN 855 property might be "This is Joe Smith's home directory", and an 856 FSL property might be "This instance of the FSN is at the 857 remote backup site." 859 a. USING THE FEDERATION INTERFACES, it MUST be possible to 860 query the system to find the annotations for a junction. 862 b. USING THE FEDERATION INTERFACES, it MUST be possible to 863 query the system to find the annotations for an FSN. 865 c. USING THE FEDERATION INTERFACES, it MUST be possible to 866 query the system to find the annotations for an FSL. 868 R15: It MUST be possible for the federation to project a namespace 869 with a common root. 871 a. It SHOULD be possible to define a root fileset that is 872 exported by one or more fileservers in the federation as 873 the top level of a namespace. [Corollary: There is one 874 root fileset per namespace and it is possible to support 875 multiple namespaces per federation.] 877 b. It SHOULD be possible for a fileserver to locate an NSDB 878 that stores the layout of a root fileset. 880 c. It SHOULD be possible to access, store and update 881 information related to a root fileset using the federation 882 protocols. 884 d. It SHOULD be possible to replicate root fileset information 885 across multiple repositories. 887 e. If a root fileset is defined it SHOULD be possible to 888 enable a file server to export that root fileset for client 889 access. 891 f. If a root fileset is defined it SHOULD be possible for 892 multiple file servers to project a common root with defined 893 consistency semantics. 895 7. Non-Requirements 897 N1: It is not necessary for the namespace to be known by any 898 specific fileserver. 900 In the same manner that clients do not need to have a priori 901 knowledge of the structure of the namespace or its mapping onto 902 federation members, the projected namespace can exist without 903 individual fileservers knowing the entire organizational 904 structure, or, indeed, without knowing exactly where in the 905 projected namespace the filesets they host exist. 907 Fileservers do need to be able to handle referrals from other 908 fileservers, but they do not need to know what path the client 909 was accessing when the referral was generated. 911 N2: It is not necessary for updates and accesses to the federation 912 data to occur in transaction or transaction-like contexts. 914 One possible requirement that is omitted from our current list 915 is that updates and accesses to the data stored in the NSDB (or 916 individual NSDB nodes) occur within a transaction context. We 917 were not able to agree whether the benefits of transactions are 918 worth the complexity they add (both to the specification and its 919 eventual implementation) but this topic is open for discussion. 921 Below is the the draft of a proposed requirement that provides 922 transactional semantics: 924 "There MUST be a way to ensure that sequences of operations, 925 including observations of the namespace (including finding 926 the locations corresponding to a set of FSNs) and changes to 927 the namespace or related data stored in the system (including 928 the creation, renaming, or deletion of junctions, and the 929 creation, altering, or deletion of mappings between FSN and 930 filesystem locations), can be performed in a manner that 931 provides predictable semantics for the relationship between 932 the observed values and the effect of the changes." 934 "It MUST be possible to protect sequences of operations by 935 transactions with NSDB-wide or server-wide ACID semantics." 937 8. Security Considerations 939 Assuming the Internet threat model, the federated resolution 940 mechanism described in this document MUST be implemented in such a 941 way to prevent loss of CONFIDENTIALITY, DATA INTEGRITY and PEER 942 ENTITY AUTHENTICATION, as described in [RFC3552]. 944 CONFIDENTIALITY may be violated if an unauthorized party is able to 945 eavesdrop on the communication between authorized servers and NSDB 946 nodes and thereby learn the locations or other information about FSNs 947 that they would not be authorized to discover via direct queries. 948 DATA INTEGRITY may be compromised if a third party is able to 949 undetectably alter the contents of the communication between servers 950 and NSDB nodes. PEER ENTITY AUTHENTICATION is defeated if one server 951 can masquerade as another server without proper authority, or if an 952 arbitrary host can masquerade as a NSDB node. 954 Well-established techniques for providing authenticated channels may 955 be used to defeat these attacks, and the protocol MUST support at 956 least one of them. 958 For example, if LDAP is used to implement the query mechanism 959 [RFC4511], then TLS may be used to provide both authentication and 960 integrity [RFC5246] [RFC4513]. If the query protocol is implemented 961 on top of ONC/RPC, then RPCSEC_GSS may be used to fill the same role 962 [RFC2203] [RFC2743]. 964 9. IANA Considerations 966 This document has no actions for IANA. 968 10. References 970 10.1. Normative References 972 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 973 Requirement Levels", BCP 14, RFC 2119, March 1997. 975 [RFC2203] Eisler, M., Chiu, A., and L. Ling, "RPCSEC_GSS Protocol 976 Specification", RFC 2203, September 1997. 978 [RFC2743] Linn, J., "Generic Security Service Application Program 979 Interface Version 2, Update 1", RFC 2743, January 2000. 981 [RFC3530] Shepler, S., Callaghan, B., Robinson, D., Thurlow, R., 982 Beame, C., Eisler, M., and D. Noveck, "Network File System 983 (NFS) version 4 Protocol", RFC 3530, April 2003. 985 [RFC3552] Rescorla, E. and B. Korver, "Guidelines for Writing RFC 986 Text on Security Considerations", BCP 72, RFC 3552, 987 July 2003. 989 [RFC4122] Leach, P., Mealling, M., and R. Salz, "A Universally 990 Unique IDentifier (UUID) URN Namespace", RFC 4122, 991 July 2005. 993 [RFC4511] Sermersheim, J., "Lightweight Directory Access Protocol 994 (LDAP): The Protocol", RFC 4511, June 2006. 996 [RFC4513] Harrison, R., "Lightweight Directory Access Protocol 997 (LDAP): Authentication Methods and Security Mechanisms", 998 RFC 4513, June 2006. 1000 [RFC5246] Dierks, T. and E. Rescorla, "The Transport Layer Security 1001 (TLS) Protocol Version 1.2", RFC 5246, August 2008. 1003 10.2. Informational References 1005 [RFC1094] Nowicki, B., "NFS: Network File System Protocol 1006 specification", RFC 1094, March 1989. 1008 [RFC1813] Callaghan, B., Pawlowski, B., and P. Staubach, "NFS 1009 Version 3 Protocol Specification", RFC 1813, June 1995. 1011 Appendix A. Acknowledgements 1013 We would like to thank Robert Thurlow of Sun Microsystems for helping 1014 to author this document. 1016 Authors' Addresses 1018 James Lentini 1019 NetApp 1020 1601 Trapelo Rd, Suite 16 1021 Waltham, MA 02451 1022 US 1024 Phone: +1 781-768-5359 1025 Email: jlentini@netapp.com 1027 Craig Everhart 1028 NetApp 1029 7301 Kit Creek Rd 1030 Research Triangle Park, NC 27709 1031 US 1033 Phone: +1 919-476-5320 1034 Email: everhart@netapp.com 1036 Daniel Ellard 1037 BBN Technologies 1038 10 Moulton Street 1039 Cambridge, MA 02138 1040 US 1042 Phone: +1 617-873-8000 1043 Email: ellard@gmail.com 1045 Renu Tewari 1046 IBM Almaden 1047 650 Harry Rd 1048 San Jose, CA 95120 1049 US 1051 Email: tewarir@us.ibm.com 1052 Manoj Naik 1053 IBM Almaden 1054 650 Harry Rd 1055 San Jose, CA 95120 1056 US 1058 Email: manoj@almaden.ibm.com