Network File System (NFS) Version 4 Minor Version 1 Protocol

1.2.1. Reliance on NFSv4-wide Documents

In many cases, matters previously described within the NFSv4.1 specification, will be addressed in separate NFSv4-wide documents.¶

• The process of protocol extension and creation of new minor versions is described in a separate NFSv4-wide document, RFC8178 [32], dealing with the issue for the NFSv4 protocols as a whole.¶

• Internationalization will be described in a separate document describing internationalization for all of the NFSv4 protocols, currently [37]. The only v4.1-specific feature is the fs_charset_cap attribute, which is described in the current document rather than the internationalization document.¶

• Security will also be described in a separate document applying to all minor versions. The handling is different because there is a wider range of security-relevant features within v4.1, despite the fact that the fundamental approach is the same for all minor versions. As a result, for some features, the security document will have the lead role while, for others, the v4.1 specification will be the main source of information about the feature, although the basic security functionality will be as defined by the NFSv4-wide security document.¶

1.2.2. Adaptation of NFSv4-wide the NFSv4-wide Security Document to v4.1-specific Features

The v4.1-specific security-related features are dealt with as described below:¶

• Security issues regarding pNFS will be the responsibility of the v4.1 specification document. In doing this, it will rely, where it can, on the security facilities described in the NFSv4-wide security document.¶

  This is necessary because some layout types will access data without using the RPC-based facilities that discussed in the security document. In addition the requirements for security-related co-ordination between data server and metadata server are best dealt with in the security document, including cases in which RPC is used by both the data server and the metadata server, in which the necessary co-ordination requirements are defined by the layout type specification document¶

• The description of the SECINFO_NO_NAME operation, will remain in the v4.1 specification, even though the description of SECINFO will be consolidated in the security document.¶

  This shift is necessary because the description of SECINFO needs to be revised to allow negotiation of security-related transport characteristics in addition to authentication flavors, mechanisms and RPCSEC_GSS services.¶

  Since the results of SECINFO an SECINFO_NONAME are identical,¶

  The circumstances of use of SECINFO will be different for different minor version, because of the limited availability among other reasons. To deal with this situation, the security document will discuss these issue in separate sections, directed to v4.0 and v4.1+.¶

• The description of the MACH_CRED and SSV features will remain in the v4.1 specification document and will only be mentioned in passing in the security document.¶

  Instead, the focus with regard to state protection will be on client-peer authentication which applies to all minor versions.¶

• sessions¶

1.2.3. Conversion of the NFSv4.1 specification into a Document Suite

The main specification for this minor version will be divided into multiple parts. There will certainly be a division of the specification into two part with discussion/explanation of the protocol appearing in one document, while reference material, defining each error, request, operation, and callback appearing separately.¶

This transformation is being done to make the revision and publication progress more manageable since the size of the specification as a single document makes it hard to deal with. Given that need, it is likely that the discussion/explanation will be further subdivided as discussed below.¶

With regard to the proposed documents discussed below, it needs to be understood that each document will have additional material in addition to the cited core material but that we intend also to take steps to avoid unhelpful duplication. Specifically,¶

• Each document will have it's own abstract highlighting its role as part of the nfsv4.1 specification.¶

• Each document will have a short introductory section describing its relationships with other documents in the specification. Part of this will deal with the possible absence of some routine sections because matters are dealt with in other documents in the specification.¶

• We will try to avoid having a full Requirements Language section in each document, typically using a reference to the extended treatment of the subject in the introductory document.¶

• It doesn't seem that each document needs an IANA Considerations section, given that most of them will empty.¶

  However, if omitting them causes confusion in the RFC editing process, many documents will receive empty pro forma sections.¶

• Having a large number of Security Considerations sections would add confusion and needs to be avoided especially since the primary Security Considerations section will be in the NFSv4-wide security document rather than in the v4.1 specification per se.¶

  There will probably need to be a Security Considerations section in the pNFS document but no others.¶

  An explanation of how security is addressed for NFSv4.1 as whole will appear in the introductory document but it would not be a Security Considerations section.¶

• Each document will have it's own reference section. It is likely that (normative) intra-specification references will have their own sub-section.¶

• There will only be a single acknowledgements section, applying to the specification as a whole. It is expected that appear in the introductory document¶

It is likely that the discussion/explanation section could be further divided as listed below. Although this listing identifies the specific material to be assigned to individual documents, it should not be assumed that the set of top-level sections will remain the same. There will be the opportunity to sub-divide top-level sections to make tables of contents more useful, particularly in the case of documents derived from a single current top-level section.¶

• An Introductory document would be derived from sections 1 through 7, together with Appendix A and Acknowledgments.¶

• The main discussion document would be derived from sections 8 through 14 together with sections 25 and 26¶

• The document regarding multi-server namespace would be derived from Section 15.¶

• The document regarding pNFS would be derived from Section 16.¶

• The document regarding the file layout type would be derived from Section 17.¶

There is no similar opportunity to sub-divide the reference document since doing so would undercut its usefulness. That document will be derived from sections 19 through 19.3. It is likely that a replacement for [10] be treated as a second reference document within the NFSv4,1 specification document suite.¶

1.3.1. Changes Completed So Far in this Update

Work on the necessary changes discussed below has already been completed, although necessary review might not yet have occurred.¶

The discussion of minor versioning has been updated to refer to RFC8178 [32], instead of the former approach which allowed each minor version to make its own versioning rules.¶

The document has been updated to eliminate the current (erroneous) treatment of internationalization, derived from earlier NFSv4.1 specifications [65], [35]. The section dealing with internationalization has been deleted, since it was never implemented. In its place, the specification has been modified to reference an external document which is to define the appropriate handling for internationalization for the NFSv4 protocols as a whole. Currently, that document is the I-D draft-ietf-nfsv4-internationalization [37]. In addition, the treatment of the fs_charset_cap attribute has been revised to reflect the analysis presented in the internationalization document.¶

Despite the completion of the internationalization work within this document, there remains work to do, most of which involves the completion of review for the NFSv4-wide internationalization document¶

• The new document was based on the treatment of internationalization within RFC7530 [66], which served as a useful starting point, since implementation of all NFSv4 minor version followed the same approach to internationalization issues. However the following issues still need to be addressed:¶

• There are is a need to update the treatment within RFC7530 to reflect IDNA changes made soon after the document was published.¶

• There is a need to deal better with client name-caching issues, especially in the context of case-insensitive file systems. Text has been written and submitted but review is needed.¶

However, these issues will be addressed in the context of the internationalization document, rather than the NFSv4.1 specification. That document's transition to be a working group document is a good time to focus in order to get them addressed.¶

1.3.2. Changes Being Made in this Update to Address NFSv4.1 Errata Reports

Work is ongoing to deal with errata reports associated previous NFSv4.1 specifications. These include a large set of errata reports associated with RFC5661 and a few associated with RFC8881. This work can be categorized as follows:¶

• The following errata reports associated with RFC5661 were dealt with in RFC8881, either because their substance related to issues to be dealt with in RFC8881 or because the simplicity of the needed change and its non-controversial nature made it simple to address the report as part of the RFC editing process for RFC8881: 2062, 2280, 2324, 2330, 2548, 3558.¶

• Work needs to be done to address many errata reports relevant to RFC 5661 that were not addressed in RFC8881, because the change was too large or too controversial to address in the context of RFC editing for RFC8661.¶

• There are a number of errata reports associated with RFC8881, that also need to be addressed.¶

• A few of the existing errata reports appear to have implications for RFC5662 and would need to be addressed by an eventual rfc5662bis RFC.¶

The errata reports that remain and that will need to be addressed in this document include reports currently assigned a range of statuses in the errata reporting system, including reports marked Verified, Hold for Document Update, and Rejected. These statuses are relevant to the processing of the associated errata but not in a way as direct as might be anticipated since errata reports marked Rejected might be addressed, as a result of a justified change in working group consensus.¶

• The following errata reports associated with RFC5661, have already been incorporated into this document draft: 2005, 2006, 2249, 2299, 2327, 2328, 2722, 3653, 3714, 4215, 5040, 5467, 5476, 6015.¶

• The following errata reports associated with RFC5661, have not been incorporated and are being ignored because they deal with internationalization-related material which will be deleted in the handling of internationalization applying to rfc5661bis: 4914.¶

• The following errata reports associated with RFC88811 already been incorporated into this document draft: 6308, 6337.¶

• The following errata reports associated with RFC5661 have not yet been incorporated in this document draft but are expected to be resolved before initial submission of the descendant working group documents: 2291, 3208, 3379, 2556, 4118, 4119, 4492, 4712, 5212.¶

• The remaining errata reports, all associated with RFC56661, need to be resolved considerably before WGLC for the descendent working group documents.¶

1.3.3. Changes Being Made Now in this Update

Work on the necessary changes discussed below has started but is not yet complete. This includes cases in which work to be completed is not within this document, but in a document referred to by this document. In such cases, matters formerly dealt within the NFSv4.1 specification, in the form of a single document, need to be addressed in a number of documents, each dealing with all NFSv4 minor versions together.¶

As noted previously, there are significant problems with the treatment of security within previous NFSv4.1 specifications [39], [65], and within other current NFSv4 specifications (e.g. RFC7530 [66], RFC8881 [35]). These are listed in Section 25.1. Work has started on these issues, although it is not as advanced as that for internationalization, since many important decisions need to be made. There is a preliminary security I-D [69] which serves as a guide to the decisions that will need to be made to guide the writing of of a working group standards-track document discussing security for all the NFSv4 protocols, which this document will eventually refer to normatively.¶

Work has begun in Sections 2.7 and 5.3 to make the presentation more suitable to an environment in which RPC makes transport-level encryption and client-host authentication available. However, there is a need for some working group decisions to be made before completion of the transition to a security framework that fully embraces these new elements. In addition, the writing of a new Security Considerations section will require substantial progress on a standards-track security document for NFSv4 as a whole. Once that work is done, there will need to be a re-organization of those sections and their role will primarily be to refer to the standards-track security document.¶

In addition, there is work being done to address security issues for NFSv4.1-specific features:¶

• Significant work has been done to clarify security implications of pNFS.¶

  This work has primarily consisted of a major revision of Section 16.9 although there are significant updates to Sections 2.7 and 2.8.2.¶

  It has been made clear that the only cases in which there are essentially no security consequences from the use of pNFS, are those in which RPC is used by the storage protocol, correcting text in previous specifications which gives a contrary impression.¶

  The text has been revised to take account of the existence of services provided by rpc-tls including encryption and client host authentication.¶

  There has been a re-organization of Section 16.9, including separate subsections dealing with non-RPC-based storage protocols and RPC-based storage protocols with either loose or tight coupling between storage server and metadata server.¶

• Significant work has been done to provide rpc-tls-based state protection which can be taken advantage even by clients who have not implemented SP4_MACH_CRED or SP4_SSV or who are using AUTH_SYS.¶

  The Section 5.5.3 has been revised to allow, when SP4_NONE is used, to allow client host authentication to be used for state protection.¶

  It is made clear in Section 7.8.3 that the use of SP4_NONE, when host-client authentication is active provides state protection against other clients rather than waiving state protection.¶

  For many of the changes mentioned above, the definitive treatment will appear in the NFSv4-wide security document and there might also be a temporary references to the preliminary security I-D.¶

  Further changes along these lines will most likely be necessary wherever in the document the SP4_* values are referred to.¶

There are a number of issues relating to the use of the key words defined in [1]. While the issues below could be treated individually and distributed among Sections 1.3.1 through 1.3.4, for now, we will treat them together.¶

• A shift has been made from only citing RFC2119 [1] to citing RFC8174 [31]. While it is sometimes said that, in the absence of RFC8174, "must" and "MUST" are to be considered synonymous, the working group has never interpreted RFC2119 in that way, although the clarification provided by RFC8174 was helpful. In light of this, it might be considered that all the necessary work has been done, apart from necessary review. However, given the working group discussion about this issue in connection with RFC8881, it appears that the working group will need to further discuss this issue soon after this document becomes a working group document. That would enable us to consider this aspect of the work complete.¶

• The use of the term "RECOMMENDED" to describe NFSv4 attributes which are not MANDATORY as been addressed by switching to the term "Recommended", since "RECOMMENDED" is not in accord with RFC2119 [1]. This work is considered complete.¶

• For many of the existing uses of the terms "SHOULD" and "SHOULD NOT", it is not clear that the meaning is compatible with RFC2119. The difficulty is compounded by uncertainty left as to the proper use of these terms, about which there may be disagreement within the working group. See Appendix A.1.1 for a detailed discussion of issues that need to be resolved. Some work has started on this area by adjusting text in certain sections but the work cannot be completed until there is agreement about the proper use of these terms in this document.¶

  These issues have been addressed by changes in Sections 5.7.2, 5.7.3, and 7.6.1). In addition, similar changes will be made in [69].¶

• There are some cases in which the terms "MUST" and "MUST NOT" have essentially been ignored by implementations or other reasons to believe that these terms may have been used inappropriately. See Appendix A.1.2 for a detailed discussion. Some work has been done toward addressing these issues but it is not complete, because further discussion is needed regarding changes to be made in Section 7.6.2.¶

  These issues have been addressed by changes in Sections 5.7.1 and 5.7.2.¶

1.3.4. Changes That Will Need to be Made Later in this Update

Work on necessary changes discussed below has not started yet, although some discussion and planning may have occurred, possibly together with preliminary specification text within Appendix A proposing likely changes to be made late in the specification proper.¶

Work also has to be done to divide the specification into two parts, dealing with protocol discussion/explanation and reference material respectively. That work is expected to be done as part of the transition of this effort into a set of working group documents.¶

There is a significant amount of pNFS-related work which it may make sense to address together.¶

• Work needs to incorporate the material within RFC 8434 [68], which establishes the requirements for parallel NFS (pNFS) layout types, which are not clearly specified in RFC 5661, leading to confusion.¶

• There are a significant number of pNFS-related errata that need to be addressed: 2505, 2751, 3064-3068, 3208, 3901, 4118, 4119.¶

• ¶

There are a number of issues that need to be addressed regarding the treatment of Directory Delegations. As initial discussion of these issues has resulted in no clear consensus on how these issues should be addressed in this minor version, further working group discussion will be needed. See Appendix A.3. for details.¶

There are issues that need to be addressed regarding how retried requests are to be terminated, including the fact the most common client handling of this situation violates a "MUST" in the existing specification. This expected to take the form of a revised Section 7.6.2, as discussed in Appendix A.2.¶

2.8.1. Core Protocol

Unlike NFSv3, which relied on a series of ancillary protocols (e.g., NLM, NSM (Network Status Monitor), MOUNT), within all minor versions of NFSv4 a single RPC protocol is available to make requests to the server. Facilities that had been separate protocols, such as locking, are now integrated within a single unified protocol, although, to implement pNFS, different data access protocols may also be used.¶

2.8.2. Parallel Access

Minor version 1 supports high-performance data access to a clustered server implementation by enabling a separation of metadata access and data access, with the latter able to be done to multiple servers in parallel.¶

Such parallel data access is controlled by recallable objects known as "layouts", which are integrated into the protocol locking model. Clients direct requests for data access to a set of data servers specified by the layout via a data storage protocol which may be NFSv4.1 or may be another protocol.¶

Because the protocols used for parallel data access are not necessarily RPC-based, the RPC-based security model (Section 2.7) is impacted (see Section 16.9). The degree of impact varies with the storage protocol (see Section 16.2.5) used for data access, and can be as low as zero for some RPC-based storage protocols (see Section 17.12).¶

2.9.1. Filehandles

As in previous versions of the NFS protocol, opaque filehandles are used to identify individual files and directories. Lookup-type and create operations translate file and directory names to filehandles, which are then used to identify objects in subsequent operations.¶

The NFSv4.1 protocol provides support for persistent filehandles, guaranteed to be valid for the lifetime of the file system object designated. In addition, it provides support to servers to provide filehandles with more limited validity guarantees, called volatile filehandles.¶

2.9.2. File Attributes

The NFSv4.1 protocol has a rich and extensible file object attribute structure, which is divided into REQUIRED, Recommended, and named attributes (see Section 10).¶

Several (but not all) of the REQUIRED attributes are derived from the attributes of NFSv3 (see the definition of the fattr3 data type in [40]). An example of a REQUIRED attribute is the file object's type (Section 10.8.1.2) so that regular files can be distinguished from directories (also known as folders in some operating environments) and other types of objects. REQUIRED attributes are discussed in Section 10.1.¶

Three example of Recommended attributes are acl, sacl, and dacl. Each of these attributes defines an Access Control List (ACL) on a file object Section 10.14. An ACL provides directory and file access control more flexible than the model used in NFSv3. The ACL definition allows for specification of specific sets of permissions for individual users and groups. In addition, ACL inheritance allows propagation of access permissions and restrictions down a directory tree as file system objects are created. Recommended attributes are discussed in Section 10.2.¶

A named attribute is an opaque byte stream that is associated with a directory or file and referred to by a string name. Named attributes are meant to be used by client applications as a method to associate application-specific data with a regular file or directory. NFSv4.1 modifies named attributes relative to NFSv4.0 by tightening the allowed operations in order to prevent the development of non-interoperable implementations. Named attributes are discussed in Section 10.3.¶

2.9.3. Multi-Server Namespace

NFSv4.1 contains a number of features to allow implementation of namespaces that cross server boundaries and that allow and facilitate a nondisruptive transfer of support for individual file systems between servers. They are all based upon attributes that allow one file system to specify alternate, additional, and new location information that specifies how the client may access that file system.¶

These attributes can be used to provide for individual active file systems:¶

• Alternate network addresses to access the current file system instance.¶
• The locations of alternate file system instances or replicas to be used in the event that the current file system instance becomes unavailable.¶

These file system location attributes may be used together with the concept of absent file systems, in which a position in the server namespace is associated with locations on other servers without there being any corresponding file system instance on the current server. For example,¶

• These attributes may be used with absent file systems to implement referrals whereby one server may direct the client to a file system provided by another server. This allows extensive multi-server namespaces to be constructed.¶
• These attributes may be provided when a previously present file system becomes absent. This allows nondisruptive migration of file systems to alternate servers.¶

5.3.1. RPCSEC_GSS and Security Services

RPCSEC_GSS [4] uses the functionality of GSS-API [7]. This allows for the use of various security mechanisms by the RPC layer without the additional implementation overhead of adding RPC security flavors.¶

     service@hostname

     nfs

     nfs/hostname

5.5.3. Resolving Client Owner Conflicts

When the server gets an EXCHANGE_ID for a client owner that currently has no state, or that has state but the lease has expired, the server MUST allow the EXCHANGE_ID and confirm the new client ID if followed by the appropriate CREATE_SESSION.¶

When the server gets an EXCHANGE_ID for a new incarnation of a client owner that currently has an old incarnation with state and an unexpired lease, the server is allowed to dispose of the state of the previous incarnation of the client owner if one of the following is true:¶

• The client ID was created without explicit state protection (i.e. SP4_NONE was used) and without client host authentication while the current EXCHANGE_ID shares those characteristics and the principals used for client Id creation and the current EXCHANGE_ID match as well.¶
• The client ID was created without explicit state protection (i.e. SP4_NONE was used) and with client host authentication while the current EXCHANGE_ID shares those characteristics with the EXCHANGE_ID used to create the client ID while the authenticated client hosts match as well.¶
• The principal that created the client ID for the client owner is the same as the principal that is sending the EXCHANGE_ID operation. Note that if the client ID was created with SP4_MACH_CRED state protection (Section 22.35), the principal MUST be based on RPCSEC_GSS authentication, the RPCSEC_GSS service used MUST be integrity or privacy, and the same GSS mechanism and principal MUST be used as that used when the client ID was created.¶
• The client ID was established with SP4_SSV protection (Section 22.35, Section 7.8.3) and the client sends the EXCHANGE_ID with the security flavor set to RPCSEC_GSS using the GSS SSV mechanism (Section 7.9).¶
• The client ID was established with SP4_SSV protection, and under the conditions described herein, the EXCHANGE_ID was sent with SP4_MACH_CRED state protection. Because the SSV might not persist across client and server restart, and because the first time a client sends EXCHANGE_ID to a server it does not have an SSV, the client MAY send the subsequent EXCHANGE_ID without an SSV RPCSEC_GSS handle. Instead, as with SP4_MACH_CRED protection, the principal MUST be based on RPCSEC_GSS authentication, the RPCSEC_GSS service used MUST be integrity or privacy, and the same GSS mechanism and principal MUST be used as that used when the client ID was created.¶

If none of the above situations apply, the server MUST return NFS4ERR_CLID_INUSE.¶

If the server accepts the principal and co_ownerid as matching that which created the client ID, and the co_verifier in the EXCHANGE_ID differs from the co_verifier used when the client ID was created, then after the server receives a CREATE_SESSION that confirms the client ID, the server deletes state. If the co_verifier values are the same (e.g., the client either is updating properties of the client ID (Section 22.35) or is attempting trunking (Section 7.5), the server MUST NOT delete state.¶

5.7.1. REQUIRED and RECOMMENDED Properties of Transports

NFSv4.1 works over Remote Direct Memory Access (RDMA) and non-RDMA-based transports with the following attributes:¶

• The transport supports reliable delivery of data, which NFSv4.1 requires. However the possibility of connections breaking is addressed in NFSv4.1 by a session-based replay cache to prevent the spurious re-execution of non-idempotent requests or modifying idempotent requests.¶
• The transport delivers data in the order it was sent. Ordered delivery simplifies detection of transmit errors, and simplifies the sending of arbitrary sized requests and responses via the record marking protocol [3].¶

Because efficient handling is required when sending large amounts of data, congestion control facilities are a significant concern.¶

• When NFSv4.1 is used over an IP-based network protocol, it is REQUIRED that the transport provide congestion control.¶
• When NFSv4.1 is used over a non-IP network protocol, it is RECOMMENDED that the transport provide congestion control.¶

To enhance the possibilities for interoperability, it is strongly recommended that NFSv4.1 client and server implementations support operation over the TCP transport protocol.¶

It is permissible for a connectionless transport to be used under NFSv4.1; however, reliable and in-order delivery of data combined with congestion control by the connectionless transport is REQUIRED. As a consequence, UDP by itself MUST NOT be used as an NFSv4.1 transport, although transports to be used for NFSv4.1 may be layered on UDP. NFSv4.1 assumes that a client transport address and server transport address used to send data over a transport together constitute a connection, even if the underlying transport eschews the concept of a connection.¶

5.7.2. Client and Server Transport Behavior

When a connection-oriented transport (e.g., TCP) is used, the client and server are normally expected to maintain the use of connections already established for a considerable length of time. This is for a number reasons:¶

• This will prevent the weakening of the transport's congestion control mechanisms by the confusion resulting from dispersing a single burst of network traffic into multiple connections.¶
• This will improve performance for the WAN environment by eliminating the need for connection setup handshakes. This is particularly so given that each new connection requires the re-establishment of a connection id, and setting up new connections.¶
• The NFSv4.1 callback model requires the client and server to maintain a client-created backchannel (see Section 7.3.1) for the server to use so that any dropping of the connection will interfere wit the use of the backchannel.¶

Although it is not fatal for a requester to retry without a disconnect between the request and retry, there are good reasons to avoid this practice. The retry does consume resources, especially with RDMA, where each request, retry or not, consumes a credit. Retries for no reason, especially retries sent shortly after the previous attempt, are a poor use of network bandwidth and defeat the purpose of a transport's inherent congestion control system.¶

There is no situation in which a replier is allowed to silently drop a request, whether the request is a retry or not. (The silent drop behavior of RPCSEC_GSS [4] is not within scope here since this behavior happens at the RPCSEC_GSS layer, which is at a lower layer in the request processing.) While the replier MAY disconnect the connection, if it does not do so, it is obligated to execute the request or return an appropriate error based on the contents of the reply cache (see Section 7.6.1).¶

When sending a reply, the replier MUST send the reply to the same full network address (e.g., if using an IP-based transport, the source port of the requester is part of the full network address) from which the requester sent the request. If using a connection-oriented transport, replies MUST be sent on the connection from which the request was received.¶

If a connection is dropped after the replier receives the request but before the replier sends the reply, the replier might have a pending reply. If a connection is established with the same source and destination full network address as the dropped connection, then the replier MUST NOT send the reply until the requester retries the request. The reason for this prohibition is that the requester MAY retry a request over a different connection (provided that connection is associated with the original request's session).¶

When using RDMA transports, there are other reasons for avoiding retries over the same connection:¶

• RDMA transports use "credits" to enforce flow control, where a credit is a right to a peer to transmit a message. If one peer were to retransmit a request (or reply), it would consume an additional credit. If the replier retransmitted a reply, it would certainly result in an RDMA connection loss, since the requester would typically only post a single receive buffer for each request. If the requester retransmitted a request, the additional credit consumed on the server might lead to RDMA connection failure unless the client accounted for it and decreased its available credit, leading to wasted resources.¶
• RDMA credits present a new issue to the reply cache in NFSv4.1. The reply cache may be used when a connection within a session is lost, such as after the client reconnects. Credit information is a dynamic property of the RDMA connection, and stale values must not be replayed from the cache. This implies that the reply cache contents must not be blindly used when replies are sent from it, and credit information appropriate to the channel must be refreshed by the RPC layer.¶

In addition, as described in Section 7.6.2, while a session is active, an NFSv4.1 requester that ceases to wait for an outstanding reply MUST take appropriate care to avoid that situation vitiating guarantees needed to maintain the exactly-once semantics needed for the successful operation of the session-based reply cache.¶

5.7.3. Ports

Historically, NFSv3 servers have listened over TCP port 2049. The registered port 2049 [44] for the NFS protocol should be the default configuration for NFSv4.1, although the port 20049 is used for NFSv4.1 layered on RPC-over-RDMA.¶

The use of a reserved port has been common for NFS implementations and it is expected that this will apply to NFSv4.1 as well. While the use of RPC binding protocols as described in RFC1833 [45] is a possibility, there is no requirement that servers provide support for such use.¶

In light of this, a client should avoid this sort of use unless it has good reason to expect such support to be present on the server, while accessing NFS services at the appropriate well-known port depending on the transport to be used.¶

6.2.1. Security Considerations for Cryptographic Algorithms in Kerberos V5

When deploying NFSv4.1, the strength of the security achieved depends on the existing Kerberos V5 infrastructure. The algorithms of Kerberos V5 are not directly exposed to or selectable by the client or server, so there is some due diligence required by the user of NFSv4.1 to ensure that security is acceptable where needed.¶

6.3.1. Put Filehandle Operations

The term "put filehandle operation" refers to PUTROOTFH, PUTPUBFH, PUTFH, and RESTOREFH. Each of the subsections herein describes how the server handles a subseries of operations that starts with a put filehandle operation.¶

6.3.2. LINK and RENAME

The LINK and RENAME operations use both the current and saved filehandles. Technically, the server MAY return NFS4ERR_WRONGSEC from LINK or RENAME if the security policy of the saved filehandle rejects the security flavor used in the COMPOUND request's credentials. If the server does so, then if there is no intersection between the security policies of saved and current filehandles, this means that it will be impossible for the client to perform the intended LINK or RENAME operation.¶

For example, suppose the client sends this COMPOUND request: SEQUENCE, PUTFH bFH, SAVEFH, PUTFH aFH, RENAME "c" "d", where filehandles bFH and aFH refer to different directories. Suppose no common security tuple exists between the security policies of aFH and bFH. If the client sends the request using credentials acceptable to bFH's security policy but not aFH's policy, then the PUTFH aFH operation will fail with NFS4ERR_WRONGSEC. After a SECINFO_NO_NAME request, the client sends SEQUENCE, PUTFH bFH, SAVEFH, PUTFH aFH, RENAME "c" "d", using credentials acceptable to aFH's security policy but not bFH's policy. The server returns NFS4ERR_WRONGSEC on the RENAME operation.¶

To prevent a client from an endless sequence of a request containing LINK or RENAME, followed by a request containing SECINFO_NO_NAME or SECINFO, the server MUST detect when the security policies of the current and saved filehandles have no mutually acceptable security tuple, and MUST NOT return NFS4ERR_WRONGSEC from LINK or RENAME in that situation. Instead the server MUST do one of two things:¶

• The server can return NFS4ERR_XDEV.¶
• The server can allow the security policy of the current filehandle to override that of the saved filehandle, and so return NFS4_OK.¶

7.2.1. SEQUENCE and CB_SEQUENCE

In NFSv4.1, when the SEQUENCE operation is present, it MUST be the first operation in the COMPOUND procedure. The primary purpose of SEQUENCE is to carry the session identifier. The session identifier associates all other operations in the COMPOUND procedure with a particular session. SEQUENCE also contains required information for maintaining EOS (see Section 7.6). Session-enabled NFSv4.1 COMPOUND requests thus have the form:¶

    +-----+--------------+-----------+------------+-----------+----
    | tag | minorversion | numops    |SEQUENCE op | op + args | ...
    |     |   (== 1)     | (limited) |  + args    |           |
    +-----+--------------+-----------+------------+-----------+----

and the replies have the form:¶

    +------------+-----+--------+-------------------------------+--//
    |last status | tag | numres |status + SEQUENCE op + results |  //
    +------------+-----+--------+-------------------------------+--//
            //-----------------------+----
            // status + op + results | ...
            //-----------------------+----

A CB_COMPOUND procedure request and reply has a similar form to COMPOUND, but instead of a SEQUENCE operation, there is a CB_SEQUENCE operation. CB_COMPOUND also has an additional field called "callback_ident", which is superfluous in NFSv4.1 and MUST be ignored by the client. CB_SEQUENCE has the same information as SEQUENCE, and also includes other information needed to resolve callback races (Section 7.6.3).¶

7.2.2. Client ID and Session Association

Each client ID (Section 5.5) can have zero or more active sessions. A client ID and associated session are required to perform file access in NFSv4.1. Each time a session is used (whether by a client sending a request to the server or the client replying to a callback request from the server), the state leased to its associated client ID is automatically renewed.¶

State (which can consist of share reservations, locks, delegations, and layouts (Section 3)) is tied to the client ID. Client state is not tied to any individual session. Successive state changing operations from a given state owner MAY go over different sessions, provided the session is associated with the same client ID. A callback MAY arrive over a different session than that of the request that originally acquired the state pertaining to the callback. For example, if session A is used to acquire a delegation, a request to recall the delegation MAY arrive over session B if both sessions are associated with the same client ID. Sections 7.8.1 and 7.8.2 discuss the security considerations around callbacks.¶

7.3.1. Association of Connections, Channels, and Sessions

Each channel is associated with zero or more transport connections (whether of the same transport protocol or different transport protocols). A connection can be associated with one channel or both channels of a session; the client and server negotiate whether a connection will carry traffic for one channel or both channels via the CREATE_SESSION (Section 22.36) and the BIND_CONN_TO_SESSION (Section 22.34) operations. When a session is created via CREATE_SESSION, the connection that transported the CREATE_SESSION request is automatically associated with the fore channel, and optionally the backchannel. If the client specifies no state protection (Section 22.35) when the session is created, then when SEQUENCE is transmitted on a different connection, the connection is automatically associated with the fore channel of the session specified in the SEQUENCE operation.¶

A connection's association with a session is not exclusive. A connection associated with the channel(s) of one session may be simultaneously associated with the channel(s) of other sessions including sessions associated with other client IDs.¶

It is permissible for connections of multiple transport types to be associated with the same channel. For example, both TCP and RDMA connections can be associated with the fore channel. In the event an RDMA and non-RDMA connection are associated with the same channel, it is desirable for the maximum number slots to be at least one more than the total number of RDMA credits (Section 7.6.1). This way, if all RDMA credits are used, the non-RDMA connection can have at least one outstanding request. If a server supports multiple transport types, it MUST allow a client to associate connections from each transport to a channel.¶

It is permissible for a connection of one type of transport to be associated with the fore channel, and a connection of a different type to be associated with the backchannel.¶

7.5.1. Verifying Claims of Matching Server Identity

When the server responds using two different connections that claim matching or partially matching eir_server_owner, eir_server_scope, and eir_clientid values, the client does not have to trust the servers' claims. The client may verify these claims before trunking traffic in the following ways:¶

• For session trunking, clients SHOULD reliably verify if connections between different network paths are in fact associated with the same NFSv4.1 server and usable on the same session, and servers MUST allow clients to perform reliable verification. When a client ID is created, the client SHOULD, unless client host authentication is in effect, specify that BIND_CONN_TO_SESSION is to be verified according to the SP4_SSV or SP4_MACH_CRED (Section 22.35) state protection options. For SP4_SSV, reliable verification depends on a shared secret (the SSV) that is established via the SET_SSV (see Section 22.47) operation.¶

  When a new connection is associated with the session (via the BIND_CONN_TO_SESSION operation, see Section 22.34), if the client specified SP4_SSV state protection for the BIND_CONN_TO_SESSION operation, the client MUST send the BIND_CONN_TO_SESSION with RPCSEC_GSS protection, using integrity or privacy, and an RPCSEC_GSS handle created with the GSS SSV mechanism (see Section 7.9).¶

  If the client mistakenly tries to associate a connection to a session of a wrong server, the server will either reject the attempt because it is not aware of the session identifier of the BIND_CONN_TO_SESSION arguments, or it will reject the attempt because the RPCSEC_GSS authentication fails. Even if the server mistakenly or maliciously accepts the connection association attempt, the RPCSEC_GSS verifier it computes in the response will not be verified by the client, so the client will know it cannot use the connection for trunking the specified session.¶

  If the client specified SP4_MACH_CRED state protection, the BIND_CONN_TO_SESSION operation will use RPCSEC_GSS integrity or privacy, using the same credential that was used when the client ID was created. Mutual authentication via RPCSEC_GSS assures the client that the connection is associated with the correct session of the correct server.¶

• For client ID trunking, the client has at least two options for verifying that the same client ID obtained from two different EXCHANGE_ID operations came from the same server. The first option is to use RPCSEC_GSS authentication when sending each EXCHANGE_ID operation. Each time an EXCHANGE_ID is sent with RPCSEC_GSS authentication, the client notes the principal name of the GSS target. If the EXCHANGE_ID results indicate that client ID trunking is possible, and the GSS targets' principal names are the same, the servers are the same and client ID trunking is allowed.¶

  The second option for verification is to use SP4_SSV protection. When the client sends EXCHANGE_ID, it specifies SP4_SSV protection. The first EXCHANGE_ID the client sends always has to be confirmed by a CREATE_SESSION call. The client then sends SET_SSV. Later, the client sends EXCHANGE_ID to a second destination network address different from the one the first EXCHANGE_ID was sent to. The client checks that each EXCHANGE_ID reply has the same eir_clientid, eir_server_owner.so_major_id, and eir_server_scope. If so, the client verifies the claim by sending a CREATE_SESSION operation to the second destination address, protected with RPCSEC_GSS integrity using an RPCSEC_GSS handle returned by the second EXCHANGE_ID. If the server accepts the CREATE_SESSION request, and if the client verifies the RPCSEC_GSS verifier and integrity codes, then the client has proof the second server knows the SSV, and thus the two servers are cooperating for the purposes of specifying server scope and client ID trunking.¶

7.6.1. Slot Identifiers and Reply Cache

The RPC layer provides a transaction ID (XID), which, while required to be unique, is not convenient for tracking requests for two reasons. First, the XID is only meaningful to the requester; it cannot be interpreted by the replier except to test for equality with previously sent requests. When consulting an RPC-based duplicate request cache, the opaqueness of the XID requires a computationally expensive look up (often via a hash that includes XID and source address). NFSv4.1 requests use a non-opaque slot ID, which is an index into a slot table, which is far more efficient. Second, because RPC requests can be executed by the replier in any order, there is no bound on the number of requests that may be outstanding at any time. To achieve perfect EOS, using ONC RPC would require storing all replies in the reply cache. XIDs are 32 bits; storing over four billion (2³²) replies in the reply cache is not practical. In practice, previous versions of NFS have chosen to store a fixed number of replies in the cache, and to use a least recently used (LRU) approach to replacing cache entries with new entries when the cache is full. In NFSv4.1, the number of outstanding requests is bounded by the size of the slot table, and a sequence ID per slot is used to tell the replier when it is safe to delete a cached reply.¶

In the NFSv4.1 reply cache, when the requester sends a new request, it selects a slot ID in the range 0..N, where N is the replier's current maximum slot ID granted to the requester on the session over which the request is to be sent. The value of N starts out as equal to ca_maxrequests - 1 (Section 22.36), but can be adjusted by the response to SEQUENCE or CB_SEQUENCE as described later in this section. The slot ID must be unused by any of the requests that the requester has already active on the session. "Unused" here means the requester has no outstanding request for that slot ID.¶

A slot contains a sequence ID and the cached reply corresponding to the request sent with that sequence ID. The sequence ID is a 32-bit unsigned value, and is therefore in the range 0..0xFFFFFFFF (2³² - 1). The first time a slot is used, the requester MUST specify a sequence ID of one (Section 22.36). Each time a slot is reused, the request MUST specify a sequence ID that is one greater than that of the previous request on the slot. If the previous sequence ID was 0xFFFFFFFF, then the next request for the slot MUST have the sequence ID set to zero (i.e., (2³² - 1) + 1 mod 2³²).¶

The sequence ID accompanies the slot ID in each request. It is for the critical check at the replier: it used to efficiently determine whether a request using a certain slot ID is a retransmit or a new, never-before-seen request. It is not feasible for the requester to assert that it is retransmitting to implement this, because for any given request the requester cannot know whether the replier has seen it unless the replier actually replies. Of course, if the requester has seen the reply, the requester would not retransmit.¶

The replier compares each received request's sequence ID with the last one previously received for that slot ID, to see if the new request is:¶

• A new request, in which the sequence ID is one greater than that previously seen in the slot (accounting for sequence wraparound). The replier proceeds to execute the new request, and the replier MUST increase the slot's sequence ID by one.¶
• A retransmitted request, in which the sequence ID is equal to that currently recorded in the slot. If the original request has executed to completion, the replier returns the cached reply. See Section 7.6.2 for direction on how the replier deals with retries of requests that are still in progress.¶
• A misordered retry, in which the sequence ID is less than (accounting for sequence wraparound) that previously seen in the slot. The replier MUST return NFS4ERR_SEQ_MISORDERED (as the result from SEQUENCE or CB_SEQUENCE).¶
• A misordered new request, in which the sequence ID is two or more than (accounting for sequence wraparound) that previously seen in the slot. Note that because the sequence ID MUST wrap around to zero once it reaches 0xFFFFFFFF, a misordered new request and a misordered retry cannot be distinguished. Thus, the replier MUST return NFS4ERR_SEQ_MISORDERED (as the result from SEQUENCE or CB_SEQUENCE).¶

Unlike the XID, the slot ID is always within a specific range; this has two implications. The first implication is that for a given session, the replier need only cache the results of a limited number of COMPOUND requests. The second implication derives from the first, which is that unlike XID-indexed reply caches (also known as duplicate request caches - DRCs), the slot ID-based reply cache cannot be overflowed. Through use of the sequence ID to identify retransmitted requests, the replier does not need to actually cache the request itself, reducing the storage requirements of the reply cache further. These facilities make it practical to maintain all the required entries for an effective reply cache.¶

The slot ID, sequence ID, and session ID therefore take over the traditional role of the XID and source network address in the replier's reply cache implementation. This approach is considerably more portable and completely robust -- it is not subject to the reassignment of ports as clients reconnect over IP networks. In addition, the RPC XID is not used in the reply cache, enhancing robustness of the cache in the face of any rapid reuse of XIDs by the requester. While the replier does not care about the XID for the purposes of reply cache management (but the replier MUST return the same XID that was in the request), nonetheless there are considerations for the XID in NFSv4.1 that are the same as all other previous versions of NFS. The RPC XID remains in each message and needs to be formulated in NFSv4.1 requests as in any other ONC RPC request. The reasons include:¶

• The RPC layer retains its existing semantics and implementation.¶
• The requester and replier must be able to interoperate at the RPC layer, prior to the NFSv4.1 decoding of the SEQUENCE or CB_SEQUENCE operation.¶
• If an operation is being used that does not start with SEQUENCE or CB_SEQUENCE (e.g., BIND_CONN_TO_SESSION), then the RPC XID is needed for correct operation to match the reply to the request.¶
• The SEQUENCE or CB_SEQUENCE operation may generate an error. If so, the embedded slot ID, sequence ID, and session ID (if present) in the request will not be in the reply, and the requester has only the XID to match the reply to the request.¶

Given that well-formulated XIDs continue to be required, this raises the question: why do SEQUENCE and CB_SEQUENCE replies have a session ID, slot ID, and sequence ID? Having the session ID in the reply means that the requester does not have to use the XID to look up the session ID, which would be necessary if the connection were associated with multiple sessions. Having the slot ID and sequence ID in the reply means that the requester does not have to use the XID to look up the slot ID and sequence ID. Furthermore, since the XID is only 32 bits, it is too small to guarantee the re-association of a reply with its request [46]; having session ID, slot ID, and sequence ID in the reply allows the client to validate that the reply in fact belongs to the matched request.¶

The SEQUENCE (and CB_SEQUENCE) operation also carries a "highest_slotid" value, which carries additional requester slot usage information. The requester MUST always indicate the slot ID representing the outstanding request with the highest-numbered slot value. The requester should in all cases provide the most conservative value possible, although it can be increased somewhat above the actual instantaneous usage to maintain some minimum or optimal level. This provides a way for the requester to yield unused request slots back to the replier, which in turn can use the information to reallocate resources.¶

The replier responds with both a new target highest_slotid and an enforced highest_slotid, described as follows:¶

• The target highest_slotid is an indication to the requester of the highest_slotid the replier wishes the requester to be using. This permits the replier to withdraw (or add) resources from a requester that has been found to not be using them, in order to more fairly share resources among a varying level of demand from other requesters. The requester must always comply with the replier's value updates, since they indicate newly established hard limits on the requester's access to session resources. However, because of request pipelining, the requester may have active requests in flight reflecting prior values; therefore, the replier must not immediately require the requester to comply.¶

• The enforced highest_slotid indicates the highest slot ID the requester is permitted to use on a subsequent SEQUENCE or CB_SEQUENCE operation. The replier's enforced highest_slotid SHOULD be no less than the highest_slotid the requester indicated in the SEQUENCE or CB_SEQUENCE arguments.¶

  A requester can be intransigent with respect to lowering its highest_slotid argument to a Sequence operation, i.e. the requester continues to ignore the target highest_slotid in the response to a Sequence operation, and continues to set its highest_slotid argument to be higher than the target highest_slotid. This can be considered particularly egregious behavior when the replier knows there are no outstanding requests with slot IDs higher than its target highest_slotid. When faced with such intransigence, the replier is free to take more forceful action, and MAY reply with a new enforced highest_slotid that is less than its previous enforced highest_slotid. Thereafter, if the requester continues to send requests with a highest_slotid that is greater than the replier's new enforced highest_slotid, the server MAY return NFS4ERR_BAD_HIGH_SLOT, unless the slot ID in the request is greater than the new enforced highest_slotid and the request is a retry.¶

  The replier SHOULD retain the slots it wants to retire until the requester sends a request with a highest_slotid less than or equal to the replier's new enforced highest_slotid.¶

  The requester can also be intransigent with respect to sending non-retry requests that have a slot ID that exceeds the replier's highest_slotid. Once the replier has forcibly lowered the enforced highest_slotid, the requester is only allowed to send retries on slots that exceed the replier's highest_slotid. If a request is received with a slot ID that is higher than the new enforced highest_slotid, and the sequence ID is one higher than what is in the slot's reply cache, then the server can both retire the slot and return NFS4ERR_BADSLOT (however, the server MUST NOT do one and not the other). The reason it is safe to retire the slot is because by using the next sequence ID, the requester is indicating it has received the previous reply for the slot.¶

• The requester SHOULD use the lowest available slot when sending a new request. This way, the replier may be able to retire slot entries faster. However, where the replier is actively adjusting its granted highest_slotid, it will not be able to use only the receipt of the slot ID and highest_slotid in the request. Neither the slot ID nor the highest_slotid used in a request may reflect the replier's current idea of the requester's session limit, because the request may have been sent from the requester before the update was received. Therefore, in the downward adjustment case, the replier may have to retain a number of reply cache entries at least as large as the old value of maximum requests outstanding, until it can infer that the requester has seen a reply containing the new granted highest_slotid. The replier can infer that the requester has seen such a reply when it receives a new request with the same slot ID as the request replied to and the next higher sequence ID.¶

7.6.2. Retry and Replay of Reply

[DN: Need to clean up this section, once the working group has had time to discuss the issues involved, particularly the viability of modifying an existing MUST to say something else, even though the MUST in question has not been strictly adhered to.¶

A requester MUST NOT retry a request, unless the connection it used to send the request disconnects. The requester can then reconnect and re-send the request, or it can re-send the request over a different connection that is associated with the same session.¶

If the requester is a server wanting to re-send a callback operation over the backchannel of a session, the requester of course cannot reconnect because only the client can associate connections with the backchannel. The server can re-send the request over another connection that is bound to the same session's backchannel. If there is no such connection, the server MUST indicate that the session has no backchannel by setting the SEQ4_STATUS_CB_PATH_DOWN_SESSION flag bit in the response to the next SEQUENCE operation from the client. The client MUST then associate a connection with the session (or destroy the session).¶

Note that it is not fatal for a requester to retry without a disconnect between the request and retry. However, the retry does consume resources, especially with RDMA, where each request, retry or not, consumes a credit. Retries for no reason, especially retries sent shortly after the previous attempt, are a poor use of network bandwidth and defeat the purpose of a transport's inherent congestion control system.¶

A requester MUST wait for a reply to a request before using the slot for another request. If it does not wait for a reply, then the requester does not know what sequence ID to use for the slot on its next request. For example, suppose a requester sends a request with sequence ID 1, and does not wait for the response. The next time it uses the slot, it sends the new request with sequence ID 2. If the replier has not seen the request with sequence ID 1, then the replier is not expecting sequence ID 2, and rejects the requester's new request with NFS4ERR_SEQ_MISORDERED (as the result from SEQUENCE or CB_SEQUENCE).¶

RDMA fabrics do not guarantee that the memory handles (Steering Tags) within each RPC/RDMA "chunk" [27] are valid on a scope outside that of a single connection. Therefore, handles used by the direct operations become invalid after connection loss. The server must ensure that any RDMA operations that must be replayed from the reply cache use the newly provided handle(s) from the most recent request.¶

A retry might be sent while the original request is still in progress on the replier. The replier SHOULD deal with the issue by returning NFS4ERR_DELAY as the reply to SEQUENCE or CB_SEQUENCE operation, but implementations MAY return NFS4ERR_MISORDERED. Since errors from SEQUENCE and CB_SEQUENCE are never recorded in the reply cache, this approach allows the results of the execution of the original request to be properly recorded in the reply cache (assuming that the requester specified the reply to be cached).¶

7.6.3. Resolving Server Callback Races

It is possible for server callbacks to arrive at the client before the reply from related fore channel operations. For example, a client may have been granted a delegation to a file it has opened, but the reply to the OPEN (informing the client of the granting of the delegation) may be delayed in the network. If a conflicting operation arrives at the server, it will recall the delegation using the backchannel, which may be on a different transport connection, perhaps even a different network, or even a different session associated with the same client ID.¶

The presence of a session between the client and server alleviates this issue. When a session is in place, each client request is uniquely identified by its { session ID, slot ID, sequence ID } triple. By the rules under which slot entries (reply cache entries) are retired, the server has knowledge whether the client has "seen" each of the server's replies. The server can therefore provide sufficient information to the client to allow it to disambiguate between an erroneous or conflicting callback race condition.¶

For each client operation that might result in some sort of server callback, the server SHOULD "remember" the { session ID, slot ID, sequence ID } triple of the client request until the slot ID retirement rules allow the server to determine that the client has, in fact, seen the server's reply. Until the time the { session ID, slot ID, sequence ID } request triple can be retired, any recalls of the associated object MUST carry an array of these referring identifiers (in the CB_SEQUENCE operation's arguments), for the benefit of the client. After this time, it is not necessary for the server to provide this information in related callbacks, since it is certain that a race condition can no longer occur.¶

The CB_SEQUENCE operation that begins each server callback carries a list of "referring" { session ID, slot ID, sequence ID } triples. If the client finds the request corresponding to the referring session ID, slot ID, and sequence ID to be currently outstanding (i.e., the server's reply has not been seen by the client), it can determine that the callback has raced the reply, and act accordingly. If the client does not find the request corresponding to the referring triple to be outstanding (including the case of a session ID referring to a destroyed session), then there is no race with respect to this triple. The server SHOULD limit the referring triples to requests that refer to just those that apply to the objects referred to in the CB_COMPOUND procedure.¶

The client must not simply wait forever for the expected server reply to arrive before responding to the CB_COMPOUND that won the race, because it is possible that it will be delayed indefinitely. The client should assume the likely case that the reply will arrive within the average round-trip time for COMPOUND requests to the server, and wait that period of time. If that period of time expires, it can respond to the CB_COMPOUND with NFS4ERR_DELAY. There are other scenarios under which callbacks may race replies. Among them are pNFS layout recalls as described in Section 16.5.5.2.¶

7.6.4. COMPOUND and CB_COMPOUND Construction Issues

Very large requests and replies may pose both buffer management issues (especially with RDMA) and reply cache issues. When the session is created (Section 22.36), for each channel (fore and back), the client and server negotiate the maximum-sized request they will send or process (ca_maxrequestsize), the maximum-sized reply they will return or process (ca_maxresponsesize), and the maximum-sized reply they will store in the reply cache (ca_maxresponsesize_cached).¶

If a request exceeds ca_maxrequestsize, the reply will have the status NFS4ERR_REQ_TOO_BIG. A replier MAY return NFS4ERR_REQ_TOO_BIG as the status for the first operation (SEQUENCE or CB_SEQUENCE) in the request (which means that no operations in the request executed and that the state of the slot in the reply cache is unchanged), or it MAY opt to return it on a subsequent operation in the same COMPOUND or CB_COMPOUND request (which means that at least one operation did execute and that the state of the slot in the reply cache does change). The replier SHOULD set NFS4ERR_REQ_TOO_BIG on the operation that exceeds ca_maxrequestsize.¶

If a reply exceeds ca_maxresponsesize, the reply will have the status NFS4ERR_REP_TOO_BIG. A replier MAY return NFS4ERR_REP_TOO_BIG as the status for the first operation (SEQUENCE or CB_SEQUENCE) in the request, or it MAY opt to return it on a subsequent operation (in the same COMPOUND or CB_COMPOUND reply). A replier MAY return NFS4ERR_REP_TOO_BIG in the reply to SEQUENCE or CB_SEQUENCE, even if the response would still exceed ca_maxresponsesize.¶

If sa_cachethis or csa_cachethis is TRUE, then the replier MUST cache a reply except if an error is returned by the SEQUENCE or CB_SEQUENCE operation (see Section 7.6.1.2). If the reply exceeds ca_maxresponsesize_cached (and sa_cachethis or csa_cachethis is TRUE), then the server MUST return NFS4ERR_REP_TOO_BIG_TO_CACHE. Even if NFS4ERR_REP_TOO_BIG_TO_CACHE (or any other error for that matter) is returned on an operation other than the first operation (SEQUENCE or CB_SEQUENCE), then the reply MUST be cached if sa_cachethis or csa_cachethis is TRUE. For example, if a COMPOUND has eleven operations, including SEQUENCE, the fifth operation is a RENAME, and the tenth operation is a READ for one million bytes, the server may return NFS4ERR_REP_TOO_BIG_TO_CACHE on the tenth operation. Since the server executed several operations, especially the non-idempotent RENAME, the client's request to cache the reply needs to be honored in order for the correct operation of exactly once semantics. If the client retries the request, the server will have cached a reply that contains results for ten of the eleven requested operations, with the tenth operation having a status of NFS4ERR_REP_TOO_BIG_TO_CACHE.¶

A client needs to take care that, when sending operations that change the current filehandle (except for PUTFH, PUTPUBFH, PUTROOTFH, and RESTOREFH), it does not exceed the maximum reply buffer before the GETFH operation. Otherwise, the client will have to retry the operation that changed the current filehandle, in order to obtain the desired filehandle. For the OPEN operation (see Section 22.16), retry is not always available as an option. The following guidelines for the handling of filehandle-changing operations are advised:¶

• Within the same COMPOUND procedure, a client SHOULD send GETFH immediately after a current filehandle-changing operation. A client MUST send GETFH after a current filehandle-changing operation that is also non-idempotent (e.g., the OPEN operation), unless the operation is RESTOREFH. RESTOREFH is an exception, because even though it is non-idempotent, the filehandle RESTOREFH produced originated from an operation that is either idempotent (e.g., PUTFH, LOOKUP), or non-idempotent (e.g., OPEN, CREATE). If the origin is non-idempotent, then because the client MUST send GETFH after the origin operation, the client can recover if RESTOREFH returns an error.¶
• A server MAY return NFS4ERR_REP_TOO_BIG or NFS4ERR_REP_TOO_BIG_TO_CACHE (if sa_cachethis is TRUE) on a filehandle-changing operation if the reply would be too large on the next operation.¶
• A server SHOULD return NFS4ERR_REP_TOO_BIG or NFS4ERR_REP_TOO_BIG_TO_CACHE (if sa_cachethis is TRUE) on a filehandle-changing, non-idempotent operation if the reply would be too large on the next operation, especially if the operation is OPEN.¶
• A server MAY return NFS4ERR_UNSAFE_COMPOUND to a non-idempotent current filehandle-changing operation, if it looks at the next operation (in the same COMPOUND procedure) and finds it is not GETFH. The server SHOULD do this if it is unable to determine in advance whether the total response size would exceed ca_maxresponsesize_cached or ca_maxresponsesize.¶

7.6.5. Persistence

Since the reply cache is bounded, it is practical for the reply cache to persist across server restarts. The replier MUST persist the following information if it agreed to persist the session (when the session was created; see Section 22.36):¶

• The session ID.¶
• The slot table including the sequence ID and cached reply for each slot.¶

The above are sufficient for a replier to provide EOS semantics for any requests that were sent and executed before the server restarted. If the replier is a client, then there is no need for it to persist any more information, unless the client will be persisting all other state across client restart, in which case, the server will never see any NFSv4.1-level protocol manifestation of a client restart. If the replier is a server, with just the slot table and session ID persisting, any requests the client retries after the server restart will return the results that are cached in the reply cache, and any new requests (i.e., the sequence ID is one greater than the slot's sequence ID) MUST be rejected with NFS4ERR_DEADSESSION (returned by SEQUENCE). Such a session is considered dead. A server MAY re-animate a session after a server restart so that the session will accept new requests as well as retries. To re-animate a session, the server needs to persist additional information through server restart:¶

• The client ID. This is a prerequisite to let the client create more sessions associated with the same client ID as the re-animated session.¶
• The client ID's sequence ID that is used for creating sessions (see Sections 22.35 and 22.36). This is a prerequisite to let the client create more sessions.¶
• The principal that created the client ID. This allows the server to authenticate the client when it sends EXCHANGE_ID.¶
• The SSV, if SP4_SSV state protection was specified when the client ID was created (see Section 22.35). This lets the client create new sessions, and associate connections with the new and existing sessions.¶
• The properties of the client ID as defined in Section 22.35.¶

A persistent reply cache places certain demands on the server. The transition between the execution of the sequence of operations (starting with SEQUENCE) and the placement of its results in the persistent cache MUST be atomic. If a client retries a sequence of operations that was issued to the server, the only acceptable outcomes are an indication that the request is still being processed, the cached reply reflecting the completion of the request, or an indication that the client ID or session has been lost (indicating a catastrophic loss of the reply cache or a session that has been deleted because the client failed to use the session for an extended period of time).¶

The possibility exists of situations in which a server can fail and restart in the middle of a COMPOUND procedure that contains one or more non-idempotent or idempotent-but-modifying operations. If the server allows COMPOUND procedures to be continued after failure, it creates a even greater challenge for the execution of such requests and the atomic placement of results in the reply cache.¶

When a server providing a persistent reply cache does not continue a COMPOUND procedure that was interrupted by a server failure, the error NFS4ERR_DEADSESSION is returned.¶

7.7.1. RDMA Connection Resources

RDMA requires its consumers to register memory and post buffers of a specific size and number for receive operations.¶

Registration of memory can be a relatively high-overhead operation, since it requires pinning of buffers, assignment of attributes (e.g., readable/writable), and initialization of hardware translation. Preregistration is desirable to reduce overhead. These registrations are specific to hardware interfaces and even to RDMA connection endpoints; therefore, negotiation of their limits is desirable to manage resources effectively.¶

Following basic registration, these buffers must be posted by the RPC layer to handle receives. These buffers remain in use by the RPC/NFSv4.1 implementation; the size and number of them must be known to the remote peer in order to avoid RDMA errors that would cause a fatal error on the RDMA connection.¶

NFSv4.1 manages slots as resources on a per-session basis (see Section 7), while RDMA connections manage credits on a per-connection basis. This means that in order for a peer to send data over RDMA to a remote buffer, it has to have both an NFSv4.1 slot and an RDMA credit. If multiple RDMA connections are associated with a session, then if the total number of credits across all RDMA connections associated with the session is X, and the number of slots in the session is Y, then the maximum number of outstanding requests is the lesser of X and Y.¶

7.7.2. Flow Control

Previous versions of NFS do not provide flow control; instead, they rely on the windowing provided by transports like TCP to throttle requests. This does not work with RDMA, which provides no operation flow control and will terminate a connection in error when limits are exceeded. Limits such as maximum number of requests outstanding are therefore negotiated when a session is created (see the ca_maxrequests field in Section 22.36). These limits then provide the maxima within which each connection associated with the session's channel(s) must remain. RDMA connections are managed within these limits as described in Section 3.3 of [27]; if there are multiple RDMA connections, then the maximum number of requests for a channel will be divided among the RDMA connections. Put a different way, the onus is on the replier to ensure that the total number of RDMA credits across all connections associated with the replier's channel does exceed the channel's maximum number of outstanding requests.¶

The limits may also be modified dynamically at the replier's choosing by manipulating certain parameters present in each NFSv4.1 reply. In addition, the CB_RECALL_SLOT callback operation (see Section 24.8) can be sent by a server to a client to return RDMA credits to the server, thereby lowering the maximum number of requests a client can have outstanding to the server.¶

7.7.3. Padding

Header padding is requested by each peer at session initiation (see the ca_headerpadsize argument to CREATE_SESSION in Section 22.36), and subsequently used by the RPC RDMA layer, as described in [27]. Zero padding is permitted.¶

Padding leverages the useful property that RDMA preserve alignment of data, even when they are placed into anonymous (untagged) buffers. If requested, client inline writes will insert appropriate pad bytes within the request header to align the data payload on the specified boundary. The client is encouraged to add sufficient padding (up to the negotiated size) so that the "data" field of the WRITE operation is aligned. Most servers can make good use of such padding, which allows them to chain receive buffers in such a way that any data carried by client requests will be placed into appropriate buffers at the server, ready for file system processing. The receiver's RPC layer encounters no overhead from skipping over pad bytes, and the RDMA layer's high performance makes the insertion and transmission of padding on the sender a significant optimization. In this way, the need for servers to perform RDMA Read to satisfy all but the largest client writes is obviated. An added benefit is the reduction of message round trips on the network -- a potentially good trade, where latency is present.¶

The value to choose for padding is subject to a number of criteria. A primary source of variable-length data in the RPC header is the authentication information, the form of which is client-determined, possibly in response to server specification. The contents of COMPOUNDs, sizes of strings such as those passed to RENAME, etc. all go into the determination of a maximal NFSv4.1 request size and therefore minimal buffer size. The client must select its offered value carefully, so as to avoid overburdening the server, and vice versa. The benefit of an appropriate padding value is higher performance.¶

                 Sender gather:
     |RPC Request|Pad  bytes|Length| -> |User data...|
     \------+----------------------/      \
             \                             \
              \    Receiver scatter:        \-----------+- ...
         /-----+----------------\            \           \
         |RPC Request|Pad|Length|   ->  |FS buffer|->|FS buffer|->...

In the above case, the server may recycle unused buffers to the next posted receive if unused by the actual received request, or may pass the now-complete buffers by reference for normal write processing. For a server that can make use of it, this removes any need for data copies of incoming data, without resorting to complicated end-to-end buffer advertisement and management. This includes most kernel-based and integrated server designs, among many others. The client may perform similar optimizations, if desired.¶

7.7.4. Dual RDMA and Non-RDMA Transports

Some RDMA transports (e.g., RFC 5040 [8]) permit a "streaming" (non-RDMA) phase, where ordinary traffic might flow before "stepping up" to RDMA mode, commencing RDMA traffic. Some RDMA transports start connections always in RDMA mode. NFSv4.1 allows, but does not assume, a streaming phase before RDMA mode. When a connection is associated with a session, the client and server negotiate whether the connection is used in RDMA or non-RDMA mode (see Sections 22.36 and 22.34).¶

7.8.1. Session Callback Security

Via session/connection association, NFSv4.1 improves security over that provided by NFSv4.0 for the backchannel. The connection is client-initiated (see Section 22.34) and subject to the same firewall and routing checks as the fore channel. At the client's option (see Section 22.35), connection association is fully authenticated before being activated (see Section 22.34). Traffic from the server over the backchannel is authenticated exactly as the client specifies (see Section 7.8.2).¶

7.8.2. Backchannel RPC Security

When the NFSv4.1 client establishes the backchannel, it informs the server of the security flavors and principals to use when sending requests. If the security flavor is RPCSEC_GSS, the client expresses the principal in the form of an established RPCSEC_GSS context. The server is free to use any of the flavor/principal combinations the client offers, but it MUST NOT use combinations not offered. This way, the client need not provide a target GSS principal for the backchannel as it did with NFSv4.0, nor does the server have to implement an RPCSEC_GSS initiator as it did with NFSv4.0 [39].¶

The CREATE_SESSION (Section 22.36) and BACKCHANNEL_CTL (Section 22.33) operations allow the client to specify flavor/principal combinations.¶

Also note that the SP4_SSV state protection mode (see Sections 22.35 and 7.8.3) has the side benefit of providing SSV-derived RPCSEC_GSS contexts (Section 7.9).¶

7.8.3. Protection from Unauthorized State Changes

As described to this point in the specification, the state model of NFSv4.1 is vulnerable to an attacker that sends a SEQUENCE operation with a forged session ID and with a slot ID that it expects the legitimate client to use next. When the legitimate client uses the slot ID with the same sequence number, the server returns the attacker's result from the reply cache, which disrupts the legitimate client and thus denies service to it. Similarly, an attacker could send a CREATE_SESSION with a forged client ID to create a new session associated with the client ID. The attacker could send requests using the new session that change locking state, such as LOCKU operations to release locks the legitimate client has acquired. Setting a security policy on the file that requires RPCSEC_GSS credentials when manipulating the file's state is one potential work around, but has the disadvantage of preventing a legitimate client from releasing state when RPCSEC_GSS is required to do so, but a GSS context cannot be obtained (possibly because the user has logged off the client).¶

NFSv4.1 provides three options to a client for state protection, which are specified when a client creates a client ID via EXCHANGE_ID (Section 22.35).¶

The first (SP4_NONE) is to simply waive state protection, except for that provided by client host authentication.¶

The other two options (SP4_MACH_CRED and SP4_SSV) share several traits:¶

• An RPCSEC_GSS-based credential is used to authenticate client ID and session maintenance operations, including creating and destroying a session, associating a connection with the session, and destroying the client ID.¶
• Because RPCSEC_GSS is used to authenticate client ID and session maintenance, the attacker cannot associate a rogue connection with a legitimate session, or associate a rogue session with a legitimate client ID in order to maliciously alter the client ID's lock state via CLOSE, LOCKU, DELEGRETURN, LAYOUTRETURN, etc.¶
• In cases where the server's security policies on a portion of its namespace require RPCSEC_GSS authentication, a client may have to use an RPCSEC_GSS credential to remove per-file state (e.g., LOCKU, CLOSE, etc.). The server may require that the principal that removes the state match certain criteria (e.g., the principal might have to be the same as the one that acquired the state). However, the client might not have an RPCSEC_GSS context for such a principal, and might not be able to create such a context (perhaps because the user has logged off). When the client establishes SP4_MACH_CRED or SP4_SSV protection, it can specify a list of operations that the server MUST allow using the machine credential (if SP4_MACH_CRED is used) or the SSV credential (if SP4_SSV is used).¶

The SP4_MACH_CRED state protection option uses a machine credential where the principal that creates the client ID MUST also be the principal that performs client ID and session maintenance operations. The security of the machine credential state protection approach depends entirely on safeguarding the per-machine credential. Assuming a proper safeguard using the per-machine credential for operations like CREATE_SESSION, BIND_CONN_TO_SESSION, DESTROY_SESSION, and DESTROY_CLIENTID will prevent an attacker from associating a rogue connection with a session, or associating a rogue session with a client ID.¶

There are at least three scenarios for the SP4_MACH_CRED option:¶

1. The system administrator configures a unique, permanent per-machine credential for one of the mandated GSS mechanisms (e.g., if Kerberos V5 is used, a "keytab" containing a principal derived from a client host name could be used).¶
2. The client is used by a single user, and so the client ID and its sessions are used by just that user. If the user's credential expires, then session and client ID maintenance cannot occur, but since the client has a single user, only that user is inconvenienced.¶
3. The physical client has multiple users, but the client implementation has a unique client ID for each user. This is effectively the same as the second scenario, but a disadvantage is that each user needs to be allocated at least one session each, so the approach suffers from lack of economy.¶

The SP4_SSV protection option uses the SSV (Section 2.5), via RPCSEC_GSS and the SSV GSS mechanism (Section 7.9), to protect state from attack. The SP4_SSV protection option is intended for the situation comprised of a client that has multiple active users and a system administrator who wants to avoid the burden of installing a permanent machine credential on each client. The SSV is established and updated on the server via SET_SSV (see Section 22.47). To prevent eavesdropping, a client SHOULD send SET_SSV via RPCSEC_GSS with the privacy service or use tls encryption on the connection making the request. Several aspects of the SSV make it intractable for an attacker to guess the SSV, and thus associate rogue connections with a session, and rogue sessions with a client ID:¶

• The arguments to and results of SET_SSV include digests of the old and new SSV, respectively.¶
• Because the initial value of the SSV is zero, therefore known, the client that opts for SP4_SSV protection and opts to apply SP4_SSV protection to BIND_CONN_TO_SESSION and CREATE_SESSION MUST send at least one SET_SSV operation before the first BIND_CONN_TO_SESSION operation or before the second CREATE_SESSION operation on a client ID. If it does not, the SSV mechanism will not generate tokens (Section 7.9). A client SHOULD send SET_SSV as soon as a session is created.¶
• A SET_SSV request does not replace the SSV with the argument to SET_SSV. Instead, the current SSV on the server is logically exclusive ORed (XORed) with the argument to SET_SSV. Each time a new principal uses a client ID for the first time, the client SHOULD send a SET_SSV with that principal's RPCSEC_GSS credentials, with RPCSEC_GSS service set to RPC_GSS_SVC_PRIVACY.¶

Here are the types of attacks that can be attempted by an attacker named Eve on a victim named Bob, and how SP4_SSV protection foils each attack:¶

• Suppose Eve is the first user to log into a legitimate client. Eve's use of an NFSv4.1 file system will cause the legitimate client to create a client ID with SP4_SSV protection, specifying that the BIND_CONN_TO_SESSION operation MUST use the SSV credential. Eve's use of the file system also causes an SSV to be created. The SET_SSV operation that creates the SSV will be protected by the RPCSEC_GSS context created by the legitimate client, which uses Eve's GSS principal and credentials. Eve can eavesdrop on the network while her RPCSEC_GSS context is created and the SET_SSV using her context is sent. Even if the legitimate client sends the SET_SSV with RPC_GSS_SVC_PRIVACY, because Eve knows her own credentials, she can decrypt the SSV. Eve can compute an RPCSEC_GSS credential that BIND_CONN_TO_SESSION will accept, and so associate a new connection with the legitimate session. Eve can change the slot ID and sequence state of a legitimate session, and/or the SSV state, in such a way that when Bob accesses the server via the same legitimate client, the legitimate client will be unable to use the session.¶

  The client's only recourse is to create a new client ID for Bob to use, and establish a new SSV for the client ID. The client will be unable to delete the old client ID, and will let the lease on the old client ID expire.¶

  Once the legitimate client establishes an SSV over the new session using Bob's RPCSEC_GSS context, Eve can use the new session via the legitimate client, but she cannot disrupt Bob. Moreover, because the client SHOULD have modified the SSV due to Eve using the new session, Bob cannot get revenge on Eve by associating a rogue connection with the session.¶

  The question is how did the legitimate client detect that Eve has hijacked the old session? When the client detects that a new principal, Bob, wants to use the session, it SHOULD have sent a SET_SSV, which leads to the following sub-scenarios:¶

  • Let us suppose that from the rogue connection, Eve sent a SET_SSV with the same slot ID and sequence ID that the legitimate client later uses. The server will assume the SET_SSV sent with Bob's credentials is a retry, and return to the legitimate client the reply it sent Eve. However, unless Eve can correctly guess the SSV the legitimate client will use, the digest verification checks in the SET_SSV response will fail. That is an indication to the client that the session has apparently been hijacked.¶

  • Alternatively, Eve sent a SET_SSV with a different slot ID than the legitimate client uses for its SET_SSV. Then the digest verification of the SET_SSV sent with Bob's credentials fails on the server, and the error returned to the client makes it apparent that the session has been hijacked.¶

  • Alternatively, Eve sent an operation other than SET_SSV, but with the same slot ID and sequence that the legitimate client uses for its SET_SSV. The server returns to the legitimate client the response it sent Eve. The client sees that the response is not at all what it expects. The client assumes either session hijacking or a server bug, and either way destroys the old session.¶

• Eve associates a rogue connection with the session as above, and then destroys the session. Again, Bob goes to use the server from the legitimate client, which sends a SET_SSV using Bob's credentials. The client receives an error that indicates that the session does not exist. When the client tries to create a new session, this will fail because the SSV it has does not match that which the server has, and now the client knows the session was hijacked. The legitimate client establishes a new client ID.¶

• If Eve creates a connection before the legitimate client establishes an SSV, because the initial value of the SSV is zero and therefore known, Eve can send a SET_SSV that will pass the digest verification check. However, because the new connection has not been associated with the session, the SET_SSV is rejected for that reason.¶

In summary, an attacker's disruption of state when SP4_SSV protection is in use is limited to the formative period of a client ID, its first session, and the establishment of the SSV. Once a non-malicious user uses the client ID, the client quickly detects any hijack and rectifies the situation. Once a non-malicious user successfully modifies the SSV, the attacker cannot use NFSv4.1 operations to disrupt the non-malicious user.¶

Note that neither the SP4_MACH_CRED nor SP4_SSV protection approaches prevent hijacking of a transport connection that has previously been associated with a session. If the goal of a counter-threat strategy is to prevent connection hijacking, the use of IPsec or TLS is RECOMMENDED.¶

If a connection hijack occurs, the hijacker could in theory change locking state and negatively impact the service to legitimate clients. However, if the server is configured to require the use of RPCSEC_GSS with integrity or privacy on the affected file objects, and if EXCHGID4_FLAG_BIND_PRINC_STATEID capability (Section 22.35) is in force, this will thwart unauthorized attempts to change locking state.¶

7.11.1. Obligations of the Server

The server has the primary obligation to monitor the state of backchannel resources that the client has created for the server (RPCSEC_GSS contexts and backchannel connections). If these resources vanish, the server takes action as specified in Section 7.13.2.¶

7.11.2. Obligations of the Client

The client SHOULD honor the following obligations in order to utilize the session:¶

• Keep a necessary session from going idle on the server. A client that requires a session but nonetheless is not sending operations risks having the session be destroyed by the server. This is because sessions consume resources, and resource limitations may force the server to cull an inactive session. A server MAY consider a session to be inactive if the client has not used the session before the session inactivity timer (Section 7.12) has expired.¶
• Destroy the session when not needed. If a client has multiple sessions, one of which has no requests waiting for replies, and has been idle for some period of time, it SHOULD destroy the session.¶
• Maintain GSS contexts and RPCSEC_GSS handles for the backchannel. If the client requires the server to use the RPCSEC_GSS security flavor for callbacks, then it needs to be sure the RPCSEC_GSS handles and/or their GSS contexts that are handed to the server via BACKCHANNEL_CTL or CREATE_SESSION are unexpired.¶
• Preserve a connection for a backchannel. The server requires a backchannel in order to gracefully recall recallable state or notify the client of certain events. Note that if the connection is not being used for the fore channel, there is no way for the client to tell if the connection is still alive (e.g., the server restarted without sending a disconnect). The onus is on the server, not the client, to determine if the backchannel's connection is alive, and to indicate in the response to a SEQUENCE operation when the last connection associated with a session's backchannel has disconnected.¶

7.11.3. Steps the Client Takes to Establish a Session

If the client does not have a client ID, the client sends EXCHANGE_ID to establish a client ID. If it opts for SP4_MACH_CRED or SP4_SSV protection, in the spo_must_enforce list of operations, it SHOULD at minimum specify CREATE_SESSION, DESTROY_SESSION, BIND_CONN_TO_SESSION, BACKCHANNEL_CTL, and DESTROY_CLIENTID. If it opts for SP4_SSV protection, the client needs to ask for SSV-based RPCSEC_GSS handles.¶

The client uses the client ID to send a CREATE_SESSION on a connection to the server. The results of CREATE_SESSION indicate whether or not the server will persist the session reply cache through a server that has restarted, and the client notes this for future reference.¶

If the client specified SP4_SSV state protection when the client ID was created, then it SHOULD send SET_SSV in the first COMPOUND after the session is created. Each time a new principal goes to use the client ID, it SHOULD send a SET_SSV again.¶

If the client wants to use delegations, layouts, directory notifications, or any other state that requires a backchannel, then it needs to add a connection to the backchannel if CREATE_SESSION did not already do so. The client creates a connection, and calls BIND_CONN_TO_SESSION to associate the connection with the session and the session's backchannel. If CREATE_SESSION did not already do so, the client MUST tell the server what security is required in order for the client to accept callbacks. The client does this via BACKCHANNEL_CTL. If the client selected SP4_MACH_CRED or SP4_SSV protection when it called EXCHANGE_ID, then the client SHOULD specify that the backchannel use RPCSEC_GSS contexts for security.¶

If the client wants to use additional connections for the backchannel, then it needs to call BIND_CONN_TO_SESSION on each connection it wants to use with the session. If the client wants to use additional connections for the fore channel, then it needs to call BIND_CONN_TO_SESSION if it specified SP4_SSV or SP4_MACH_CRED state protection when the client ID was created.¶

At this point, the session has reached steady state.¶

7.13.1. Events Requiring Client Action

The following events require client action to recover.¶

7.13.2. Events Requiring Server Action

The following events require server action to recover.¶

10.8.1. Definitions of REQUIRED Attributes

10.8.2. Definitions of Uncategorized Recommended Attributes

The definitions of most of the Recommended attributes follow. Collections that share a common category are defined in other sections.¶

10.11.1. Attribute 56: dir_notif_delay

The dir_notif_delay attribute is the minimum number of seconds the server will delay before notifying the client of a change to the directory's attributes.¶

10.11.2. Attribute 57: dirent_notif_delay

The dirent_notif_delay attribute is the minimum number of seconds the server will delay before notifying the client of a change to a file object that has an entry in the directory.¶

10.12.1. Attribute 62: fs_layout_type

The fs_layout_type attribute (see Section 8.3.13) applies to a file system and indicates what layout types are supported by the file system. When the client encounters a new fsid, the client SHOULD obtain the value for the fs_layout_type attribute associated with the new file system. This attribute is used by the client to determine if the layout types supported by the server match any of the client's supported layout types.¶

10.12.2. Attribute 66: layout_alignment

When a client holds layouts on files of a file system, the layout_alignment attribute indicates the preferred alignment for I/O to files on that file system. Where possible, the client should send READ and WRITE operations with offsets that are whole multiples of the layout_alignment attribute.¶

10.12.3. Attribute 65: layout_blksize

When a client holds layouts on files of a file system, the layout_blksize attribute indicates the preferred block size for I/O to files on that file system. Where possible, the client should send READ operations with a count argument that is a whole multiple of layout_blksize, and WRITE operations with a data argument of size that is a whole multiple of layout_blksize.¶

10.12.4. Attribute 63: layout_hint

The layout_hint attribute (see Section 8.3.19) may be set on newly created files to influence the metadata server's choice for the file's layout. If possible, this attribute is one of those set in the initial attributes within the OPEN operation. The metadata server may choose to ignore this attribute. The layout_hint attribute is a subset of the layout structure returned by LAYOUTGET. For example, instead of specifying particular devices, this would be used to suggest the stripe width of a file. The server implementation determines which fields within the layout will be used.¶

10.12.5. Attribute 64: layout_type

This attribute lists the layout type(s) available for a file. The value returned by the server is for informational purposes only. The client will use the LAYOUTGET operation to obtain the information needed in order to perform I/O, for example, the specific device information for the file and its layout.¶

10.12.6. Attribute 68: mdsthreshold

This attribute is a server-provided hint used to communicate to the client when it is more efficient to send READ and WRITE operations to the metadata server or the data server. The two types of thresholds described are file size thresholds and I/O size thresholds. If a file's size is smaller than the file size threshold, data accesses SHOULD be sent to the metadata server. If an I/O request has a length that is below the I/O size threshold, the I/O SHOULD be sent to the metadata server. Each threshold type is specified separately for read and write.¶

The server MAY provide both types of thresholds for a file. If both file size and I/O size are provided, the client SHOULD reach or exceed both thresholds before sending its read or write requests to the data server. Alternatively, if only one of the specified thresholds is reached or exceeded, the I/O requests are sent to the metadata server.¶

For each threshold type, a value of zero indicates no READ or WRITE should be sent to the metadata server, while a value of all ones indicates that all READs or WRITEs should be sent to the metadata server.¶

The attribute is available on a per-filehandle basis. If the current filehandle refers to a non-pNFS file or directory, the metadata server should return an attribute that is representative of the filehandle's file system. It is suggested that this attribute is queried as part of the OPEN operation. Due to dynamic system changes, the client should not assume that the attribute will remain constant for any specific time period; thus, it should be periodically refreshed.¶

10.13.1. Attribute 69: retention_get

If retention is enabled for the associated file, this attribute's value represents the retention begin time of the file object. This attribute's value is only readable with the GETATTR operation and MUST NOT be modified by the SETATTR operation (Section 10.5). The value of the attribute consists of:¶

const RET4_DURATION_INFINITE    = 0xffffffffffffffff;
struct retention_get4 {
        uint64_t        rg_duration;
        nfstime4        rg_begin_time<1>;
};

The field rg_duration is the duration in seconds indicating how long the file will be retained once retention is enabled. The field rg_begin_time is an array of up to one absolute time value. If the array is zero length, no beginning retention time has been established, and retention is not enabled. If rg_duration is equal to RET4_DURATION_INFINITE, the file, once retention is enabled, will be retained for an infinite duration.¶

If (as soon as) rg_duration is zero, then rg_begin_time will be of zero length, and again, retention is not (no longer) enabled.¶

10.13.2. Attribute 70: retention_set

This attribute is used to set the retention duration and optionally enable retention for the associated file object. This attribute is only modifiable via the SETATTR operation and MUST NOT be retrieved by the GETATTR operation (Section 10.5). This attribute corresponds to retention_get. The value of the attribute consists of:¶

struct retention_set4 {
        bool            rs_enable;
        uint64_t        rs_duration<1>;
};

If the client sets rs_enable to TRUE, then it is enabling retention on the file object with the begin time of retention starting from the server's current time and date. The duration of the retention can also be provided if the rs_duration array is of length one. The duration is the time in seconds from the begin time of retention, and if set to RET4_DURATION_INFINITE, the file is to be retained forever. If retention is enabled, with no duration specified in either this SETATTR or a previous SETATTR, the duration defaults to zero seconds. The server MAY restrict the enabling of retention or the duration of retention on the basis of the ACE4_WRITE_RETENTION ACL permission. The enabling of retention MUST NOT prevent the enabling of event-based retention or the modification of the retention_hold attribute.¶

The following rules apply to both the retention_set and retentevt_set attributes.¶

• As long as retention is not enabled, the client is permitted to decrease the duration.¶
• The duration can always be set to an equal or higher value, even if retention is enabled. Note that once retention is enabled, the actual duration (as returned by the retention_get or retentevt_get attributes; see Section 10.13.1 or Section 10.13.3) is constantly counting down to zero (one unit per second), unless the duration was set to RET4_DURATION_INFINITE. Thus, it will not be possible for the client to precisely extend the duration on a file that has retention enabled.¶
• While retention is enabled, attempts to disable retention or decrease the retention's duration MUST fail with the error NFS4ERR_INVAL.¶
• If the principal attempting to change retention_set or retentevt_set does not have ACE4_WRITE_RETENTION permissions, the attempt MUST fail with NFS4ERR_ACCESS.¶

10.13.3. Attribute 71: retentevt_get

Gets the event-based retention duration, and if enabled, the event-based retention begin time of the file object. This attribute is like retention_get, but refers to event-based retention. The event that triggers event-based retention is not defined by the NFSv4.1 specification.¶

10.13.4. Attribute 72: retentevt_set

Sets the event-based retention duration, and optionally enables event-based retention on the file object. This attribute corresponds to retentevt_get and is like retention_set, but refers to event-based retention. When event-based retention is set, the file MUST be retained even if non-event-based retention has been set, and the duration of non-event-based retention has been reached. Conversely, when non-event-based retention has been set, the file MUST be retained even if event-based retention has been set, and the duration of event-based retention has been reached. The server MAY restrict the enabling of event-based retention or the duration of event-based retention on the basis of the ACE4_WRITE_RETENTION ACL permission. The enabling of event-based retention MUST NOT prevent the enabling of non-event-based retention or the modification of the retention_hold attribute.¶

10.13.5. Attribute 73: retention_hold

Gets or sets administrative retention holds, one hold per bit position.¶

This attribute allows one to 64 administrative holds, one hold per bit on the attribute. If retention_hold is not zero, then the file MUST NOT be deleted, renamed, or modified, even if the duration on enabled event or non-event-based retention has been reached. The server MAY restrict the modification of retention_hold on the basis of the ACE4_WRITE_RETENTION_HOLD ACL permission. The enabling of administration retention holds does not prevent the enabling of event-based or non-event-based retention.¶

If the principal attempting to change retention_hold does not have ACE4_WRITE_RETENTION_HOLD permissions, the attempt MUST fail with NFS4ERR_ACCESS.¶

14.2.1. Delegation Recovery

There are three situations that delegation recovery needs to deal with:¶

• client restart¶
• server restart¶
• network partition (full or backchannel-only)¶

In the event the client restarts, the failure to renew the lease will result in the revocation of byte-range locks and share reservations. Delegations, however, may be treated a bit differently.¶

There will be situations in which delegations will need to be re-established after a client restarts. The reason for this is that the client may have file data stored locally and this data was associated with the previously held delegations. The client will need to re-establish the appropriate file state on the server.¶

To allow for this type of client recovery, the server MAY extend the period for delegation recovery beyond the typical lease expiration period. This implies that requests from other clients that conflict with these delegations will need to wait. Because the normal recall process may require significant time for the client to flush changed state to the server, other clients need be prepared for delays that occur because of a conflicting delegation. This longer interval would increase the window for clients to restart and consult stable storage so that the delegations can be reclaimed. For OPEN delegations, such delegations are reclaimed using OPEN with a claim type of CLAIM_DELEGATE_PREV or CLAIM_DELEG_PREV_FH (see Sections 14.5 and 22.16 for discussion of OPEN delegation and the details of OPEN, respectively).¶

A server MAY support claim types of CLAIM_DELEGATE_PREV and CLAIM_DELEG_PREV_FH, and if it does, it MUST NOT remove delegations upon a CREATE_SESSION that confirm a client ID created by EXCHANGE_ID. Instead, the server MUST, for a period of time no less than that of the value of the lease_time attribute, maintain the client's delegations to allow time for the client to send CLAIM_DELEGATE_PREV and/or CLAIM_DELEG_PREV_FH requests. The server that supports CLAIM_DELEGATE_PREV and/or CLAIM_DELEG_PREV_FH MUST support the DELEGPURGE operation.¶

When the server restarts, delegations are reclaimed (using the OPEN operation with CLAIM_PREVIOUS) in a similar fashion to byte-range locks and share reservations. However, there is a slight semantic difference. In the normal case, if the server decides that a delegation should not be granted, it performs the requested action (e.g., OPEN) without granting any delegation. For reclaim, the server grants the delegation but a special designation is applied so that the client treats the delegation as having been granted but recalled by the server. Because of this, the client has the duty to write all modified state to the server and then return the delegation. This process of handling delegation reclaim reconciles three principles of the NFSv4.1 protocol:¶

• Upon reclaim, a client reporting resources assigned to it by an earlier server instance must be granted those resources.¶
• The server has unquestionable authority to determine whether delegations are to be granted and, once granted, whether they are to be continued.¶
• The use of callbacks should not be depended upon until the client has proven its ability to receive them.¶

When a client needs to reclaim a delegation and there is no associated open, the client may use the CLAIM_PREVIOUS variant of the WANT_DELEGATION operation. However, since the server is not required to support this operation, an alternative is to reclaim via a dummy OPEN together with the delegation using an OPEN of type CLAIM_PREVIOUS. The dummy open file can be released using a CLOSE to re-establish the original state to be reclaimed, a delegation without an associated open.¶

When a client has more than a single open associated with a delegation, state for those additional opens can be established using OPEN operations of type CLAIM_DELEGATE_CUR. When these are used to establish opens associated with reclaimed delegations, the server MUST allow them when made within the grace period.¶

When a network partition occurs, delegations are subject to freeing by the server when the lease renewal period expires. This is similar to the behavior for locks and share reservations. For delegations, however, the server may extend the period in which conflicting requests are held off. Eventually, the occurrence of a conflicting request from another client will cause revocation of the delegation. A loss of the backchannel (e.g., by later network configuration change) will have the same effect. A recall request will fail and revocation of the delegation will result.¶

A client normally finds out about revocation of a delegation when it uses a stateid associated with a delegation and receives one of the errors NFS4ERR_EXPIRED, NFS4ERR_ADMIN_REVOKED, or NFS4ERR_DELEG_REVOKED. It also may find out about delegation revocation after a client restart when it attempts to reclaim a delegation and receives that same error. Note that in the case of a revoked OPEN_DELEGATE_WRITE delegation, there are issues because data may have been modified by the client whose delegation is revoked and separately by other clients. See Section 14.5.1 for a discussion of such issues. Note also that when delegations are revoked, information about the revoked delegation will be written by the server to stable storage (as described in Section 12.4.3). This is done to deal with the case in which a server restarts after revoking a delegation but before the client holding the revoked delegation is notified about the revocation.¶

14.4.1. Open Delegation and Data Caching

An OPEN delegation allows much of the message overhead associated with the opening and closing files to be eliminated. An open when an OPEN delegation is in effect does not require that a validation message be sent to the server. The continued endurance of the "OPEN_DELEGATE_READ delegation" provides a guarantee that no OPEN for OPEN4_SHARE_ACCESS_WRITE/OPEN4_SHARE_ACCESS_BOTH, and thus no write, has occurred. Similarly, when closing a file opened for OPEN4_SHARE_ACCESS_WRITE/OPEN4_SHARE_ACCESS_BOTH and if an OPEN_DELEGATE_WRITE delegation is in effect, the data written does not have to be written to the server until the OPEN delegation is recalled. The continued endurance of the OPEN delegation provides a guarantee that no open, and thus no READ or WRITE, has been done by another client.¶

For the purposes of OPEN delegation, READs and WRITEs done without an OPEN are treated as the functional equivalents of a corresponding type of OPEN. Although a client SHOULD NOT use special stateids when an open exists, delegation handling on the server can use the client ID associated with the current session to determine if the operation has been done by the holder of the delegation (in which case, no recall is necessary) or by another client (in which case, the delegation must be recalled and I/O not proceed until the delegation is returned or revoked).¶

With delegations, a client is able to avoid writing data to the server when the CLOSE of a file is serviced. The file close system call is the usual point at which the client is notified of a lack of stable storage for the modified file data generated by the application. At the close, file data is written to the server and, through normal accounting, the server is able to determine if the available file system space for the data has been exceeded (i.e., the server returns NFS4ERR_NOSPC or NFS4ERR_DQUOT). This accounting includes quotas. The introduction of delegations requires that an alternative method be in place for the same type of communication to occur between client and server.¶

In the delegation response, the server provides either the limit of the size of the file or the number of modified blocks and associated block size. The server must ensure that the client will be able to write modified data to the server of a size equal to that provided in the original delegation. The server must make this assurance for all outstanding delegations. Therefore, the server must be careful in its management of available space for new or modified data, taking into account available file system space and any applicable quotas. The server can recall delegations as a result of managing the available file system space. The client should abide by the server's state space limits for delegations. If the client exceeds the stated limits for the delegation, the server's behavior is undefined.¶

Based on server conditions, quotas, or available file system space, the server may grant OPEN_DELEGATE_WRITE delegations with very restrictive space limitations. The limitations may be defined in a way that will always force modified data to be flushed to the server on close.¶

With respect to authentication, flushing modified data to the server after a CLOSE has occurred may be problematic. For example, the user of the application may have logged off the client, and unexpired authentication credentials may not be present. In this case, the client may need to take special care to ensure that local unexpired credentials will in fact be available. This may be accomplished by tracking the expiration time of credentials and flushing data well in advance of their expiration or by making private copies of credentials to assure their availability when needed.¶

14.4.3. Handling of CB_GETATTR

The server needs to employ special handling for a GETATTR where the target is a file that has an OPEN_DELEGATE_WRITE delegation in effect. The reason for this is that the client holding the OPEN_DELEGATE_WRITE delegation may have modified the data, and the server needs to reflect this change to the second client that submitted the GETATTR. Therefore, the client holding the OPEN_DELEGATE_WRITE delegation needs to be interrogated. The server will use the CB_GETATTR operation. The only attributes that the server can reliably query via CB_GETATTR are size and change.¶

Since CB_GETATTR is being used to satisfy another client's GETATTR request, the server only needs to know if the client holding the delegation has a modified version of the file. If the client's copy of the delegated file is not modified (data or size), the server can satisfy the second client's GETATTR request from the attributes stored locally at the server. If the file is modified, the server only needs to know about this modified state. If the server determines that the file is currently modified, it will respond to the second client's GETATTR as if the file had been modified locally at the server.¶

Since the form of the change attribute is determined by the server and is opaque to the client, the client and server need to agree on a method of communicating the modified state of the file. For the size attribute, the client will report its current view of the file size. For the change attribute, the handling is more involved.¶

For the client, the following steps will be taken when receiving an OPEN_DELEGATE_WRITE delegation:¶

• The value of the change attribute will be obtained from the server and cached. Let this value be represented by c.¶
• The client will create a value greater than c that will be used for communicating that modified data is held at the client. Let this value be represented by d.¶
• When the client is queried via CB_GETATTR for the change attribute, it checks to see if it holds modified data. If the file is modified, the value d is returned for the change attribute value. If this file is not currently modified, the client returns the value c for the change attribute.¶

For simplicity of implementation, the client MAY for each CB_GETATTR return the same value d. This is true even if, between successive CB_GETATTR operations, the client again modifies the file's data or metadata in its cache. The client can return the same value because the only requirement is that the client be able to indicate to the server that the client holds modified data. Therefore, the value of d may always be c + 1.¶

While the change attribute is opaque to the client in the sense that it has no idea what units of time, if any, the server is counting change with, it is not opaque in that the client has to treat it as an unsigned integer, and the server has to be able to see the results of the client's changes to that integer. Therefore, the server MUST encode the change attribute in network order when sending it to the client. The client MUST decode it from network order to its native order when receiving it, and the client MUST encode it in network order when sending it to the server. For this reason, change is defined as an unsigned integer rather than an opaque array of bytes.¶

For the server, the following steps will be taken when providing an OPEN_DELEGATE_WRITE delegation:¶

• Upon providing an OPEN_DELEGATE_WRITE delegation, the server will cache a copy of the change attribute in the data structure it uses to record the delegation. Let this value be represented by sc.¶
• When a second client sends a GETATTR operation on the same file to the server, the server obtains the change attribute from the first client. Let this value be cc.¶
• If the value cc is equal to sc, the file is not modified and the server returns the current values for change, time_metadata, and time_modify (for example) to the second client.¶
• If the value cc is NOT equal to sc, the file is currently modified at the first client and most likely will be modified at the server at a future time. The server then uses its current time to construct attribute values for time_metadata and time_modify. A new value of sc, which we will call nsc, is computed by the server, such that nsc >= sc + 1. The server then returns the constructed time_metadata, time_modify, and nsc values to the requester. The server replaces sc in the delegation record with nsc. To prevent the possibility of time_modify, time_metadata, and change from appearing to go backward (which would happen if the client holding the delegation fails to write its modified data to the server before the delegation is revoked or returned), the server SHOULD update the file's metadata record with the constructed attribute values. For reasons of reasonable performance, committing the constructed attribute values to stable storage is OPTIONAL.¶

As discussed earlier in this section, the client MAY return the same cc value on subsequent CB_GETATTR calls, even if the file was modified in the client's cache yet again between successive CB_GETATTR calls. Therefore, the server must assume that the file has been modified yet again, and MUST take care to ensure that the new nsc it constructs and returns is greater than the previous nsc it returned. An example implementation's delegation record would satisfy this mandate by including a boolean field (let us call it "modified") that is set to FALSE when the delegation is granted, and an sc value set at the time of grant to the change attribute value. The modified field would be set to TRUE the first time cc != sc, and would stay TRUE until the delegation is returned or revoked. The processing for constructing nsc, time_modify, and time_metadata would use this pseudo code:¶

    if (!modified) {
        do CB_GETATTR for change and size;

        if (cc != sc)
            modified = TRUE;
    } else {
        do CB_GETATTR for size;
    }

    if (modified) {
        sc = sc + 1;
        time_modify = time_metadata = current_time;
        update sc, time_modify, time_metadata into file's metadata;
    }

This would return to the client (that sent GETATTR) the attributes it requested, but make sure size comes from what CB_GETATTR returned. The server would not update the file's metadata with the client's modified size.¶

In the case that the file attribute size is different than the server's current value, the server treats this as a modification regardless of the value of the change attribute retrieved via CB_GETATTR and responds to the second client as in the last step.¶

This methodology resolves issues of clock differences between client and server and other scenarios where the use of CB_GETATTR break down.¶

It should be noted that the server is under no obligation to use CB_GETATTR, and therefore the server MAY simply recall the delegation to avoid its use.¶

14.4.7. Delegations via WANT_DELEGATION

In addition to providing delegations as part of the reply to OPEN operations, servers MAY provide delegations separate from open, via the OPTIONAL WANT_DELEGATION operation. This allows delegations to be obtained in advance of an OPEN that might benefit from them, for objects that are not a valid target of OPEN, or to deal with cases in which a delegation has been recalled and the client wants to make an attempt to re-establish it if the absence of use by other clients allows that.¶

The WANT_DELEGATION operation may be performed on any type of file object other than a directory.¶

When a delegation is obtained using WANT_DELEGATION, any open files for the same filehandle held by that client are to be treated as subordinate to the delegation, just as if they had been created using an OPEN of type CLAIM_DELEGATE_CUR. They are otherwise unchanged as to seqid, access and deny modes, and the relationship with byte-range locks. Similarly, because existing byte-range locks are subordinate to an open, those byte-range locks also become indirectly subordinate to that new delegation.¶

The WANT_DELEGATION operation provides for delivery of delegations via callbacks, when the delegations are not immediately available. When a requested delegation is available, it is delivered to the client via a CB_PUSH_DELEG operation. When this happens, open files for the same filehandle become subordinate to the new delegation at the point at which the delegation is delivered, just as if they had been created using an OPEN of type CLAIM_DELEGATE_CUR. Similarly, this occurs for existing byte-range locks subordinate to an open.¶

14.5.1. Revocation Recovery for Write Open Delegation

Revocation recovery for an OPEN_DELEGATE_WRITE delegation poses the special issue of modified data in the client cache while the file is not open. In this situation, any client that does not flush modified data to the server on each close must ensure that the user receives appropriate notification of the failure as a result of the revocation. Since such situations may require human action to correct problems, notification schemes in which the appropriate user or administrator is notified may be necessary. Logging and console messages are typical examples.¶

If there is modified data on the client, it must not be flushed normally to the server. A client may attempt to provide a copy of the file data as modified during the delegation under a different name in the file system namespace to ease recovery. Note that when the client can determine that the file has not been modified by any other client, or when the client has a complete cached copy of the file in question, such a saved copy of the client's view of the file may be of particular value for recovery. In another case, recovery using a copy of the file based partially on the client's cached data and partially on the server's copy as modified by other clients will be anything but straightforward, so clients may avoid saving file contents in these situations or specially mark the results to warn users of possible problems.¶

Saving of such modified data in delegation revocation situations may be limited to files of a certain size or might be used only when sufficient disk space is available within the target file system. Such saving may also be restricted to situations when the client has sufficient buffering resources to keep the cached copy available until it is properly stored to the target file system.¶

14.8.1. Name Caching

The results of LOOKUP and READDIR operations may be cached to avoid the cost of subsequent LOOKUP operations. Just as in the case of attribute caching, inconsistencies may arise among the various client caches. To mitigate the effects of these inconsistencies and given the context of typical file system APIs, an upper time boundary is maintained for how long a client name cache entry can be kept without verifying that the entry has not been made invalid by a directory change operation performed by another client.¶

When a client is not making changes to a directory for which there exist name cache entries, the client needs to periodically fetch attributes for that directory to ensure that it is not being modified. After determining that no modification has occurred, the expiration time for the associated name cache entries may be updated to be the current time plus the name cache staleness bound.¶

When a client is making changes to a given directory, it needs to determine whether there have been changes made to the directory by other clients. It does this by using the change attribute as reported before and after the directory operation in the associated change_info4 value returned for the operation. The server is able to communicate to the client whether the change_info4 data is provided atomically with respect to the directory operation. If the change values are provided atomically, the client has a basis for determining, given proper care, whether other clients are modifying the directory in question.¶

The simplest way to enable the client to make this determination is for the client to serialize all changes made to a specific directory. When this is done, and the server provides before and after values of the change attribute atomically, the client can simply compare the after value of the change attribute from one operation on a directory with the before value on the subsequent operation modifying that directory. When these are equal, the client is assured that no other client is modifying the directory in question.¶

When such serialization is not used, and there may be multiple simultaneous outstanding operations modifying a single directory sent from a single client, making this sort of determination can be more complicated. If two such operations complete in a different order than they were actually performed, that might give an appearance consistent with modification being made by another client. Where this appears to happen, the client needs to await the completion of all such modifications that were started previously, to see if the outstanding before and after change numbers can be sorted into a chain such that the before value of one change number matches the after value of a previous one, in a chain consistent with this client being the only one modifying the directory.¶

In either of these cases, the client is able to determine whether the directory is being modified by another client. If the comparison indicates that the directory was updated by another client, the name cache associated with the modified directory is purged from the client. If the comparison indicates no modification, the name cache can be updated on the client to reflect the directory operation and the associated timeout can be extended. The post-operation change value needs to be saved as the basis for future change_info4 comparisons.¶

As demonstrated by the scenario above, name caching requires that the client revalidate name cache data by inspecting the change attribute of a directory at the point when the name cache item was cached. This requires that the server update the change attribute for directories when the contents of the corresponding directory is modified. For a client to use the change_info4 information appropriately and correctly, the server must report the pre- and post-operation change attribute values atomically. When the server is unable to report the before and after values atomically with respect to the directory operation, the server must indicate that fact in the change_info4 return value. When the information is not atomically reported, the client should not assume that other clients have not changed the directory.¶

14.8.2. Directory Caching

The results of READDIR operations may be used to avoid subsequent READDIR operations. Just as in the cases of attribute and name caching, inconsistencies may arise among the various client caches. To mitigate the effects of these inconsistencies, and given the context of typical file system APIs, the following rules should be followed:¶

• Cached READDIR information for a directory that is not obtained in a single READDIR operation must always be a consistent snapshot of directory contents. This is determined by using a GETATTR before the first READDIR and after the last READDIR that contributes to the cache.¶
• An upper time boundary is maintained to indicate the length of time a directory cache entry is considered valid before the client must revalidate the cached information.¶

The revalidation technique parallels that discussed in the case of name caching. When the client is not changing the directory in question, checking the change attribute of the directory with GETATTR is adequate. The lifetime of the cache entry can be extended at these checkpoints. When a client is modifying the directory, the client needs to use the change_info4 data to determine whether there are other clients modifying the directory. If it is determined that no other client modifications are occurring, the client may update its directory cache to reflect its own changes.¶

As demonstrated previously, directory caching requires that the client revalidate directory cache data by inspecting the change attribute of a directory at the point when the directory was cached. This requires that the server update the change attribute for directories when the contents of the corresponding directory is modified. For a client to use the change_info4 information appropriately and correctly, the server must report the pre- and post-operation change attribute values atomically. When the server is unable to report the before and after values atomically with respect to the directory operation, the server must indicate that fact in the change_info4 return value. When the information is not atomically reported, the client should not assume that other clients have not changed the directory.¶

14.9.2. Directory Delegation Design

NFSv4.1 introduces the GET_DIR_DELEGATION (Section 22.39) operation to allow the client to ask for a directory delegation. The delegation covers directory attributes and all entries in the directory. If either of these change, the delegation will be recalled synchronously. The operation causing the recall will have to wait before the recall is complete. Any changes to directory entry attributes will not cause the delegation to be recalled.¶

In addition to asking for delegations, a client can also ask for notifications for certain events. These events include changes to the directory's attributes and/or its contents. If a client asks for notification for a certain event, the server will notify the client when that event occurs. This will not result in the delegation being recalled for that client. The notifications are asynchronous and provide a way of avoiding recalls in situations where a directory is changing enough that the pure recall model may not be effective while trying to allow the client to get substantial benefit. In the absence of notifications, once the delegation is recalled the client has to refresh its directory cache; this might not be very efficient for very large directories.¶

The delegation is read-only and the client may not make changes to the directory other than by performing NFSv4.1 operations that modify the directory or the associated file attributes so that the server has knowledge of these changes. In order to keep the client's namespace synchronized with that of the server, the server will notify the delegation-holding client (assuming it has requested notifications) of the changes made as a result of that client's directory-modifying operations. This is to avoid any need for that client to send subsequent GETATTR or READDIR operations to the server. If a single client is holding the delegation and that client makes any changes to the directory (i.e., the changes are made via operations sent on a session associated with the client ID holding the delegation), the delegation will not be recalled. Multiple clients may hold a delegation on the same directory, but if any such client modifies the directory, the server MUST recall the delegation from the other clients, unless those clients have made provisions to be notified of that sort of modification.¶

Delegations can be recalled by the server at any time. Normally, the server will recall the delegation when the directory changes in a way that is not covered by the notification, or when the directory changes and notifications have not been requested. If another client removes the directory for which a delegation has been granted, the server will recall the delegation.¶

15.1.1. Terminology Related to Trunking

It is particularly important to clarify the distinction between trunking detection and trunking discovery. The definitions we present are applicable to all minor versions of NFSv4, but we will focus on how these terms apply to NFS version 4.1.¶

• Trunking detection refers to ways of deciding whether two specific network addresses are connected to the same NFSv4 server. The means available to make this determination depends on the protocol version, and, in some cases, on the client implementation.¶

  In the case of NFS version 4.1 and later minor versions, the means of trunking detection are as described in this document and are available to every client. Two network addresses connected to the same server can always be used together to access a particular server but cannot necessarily be used together to access a single session. See below for definitions of the terms "server-trunkable" and "session-trunkable".¶

• Trunking discovery is a process by which a client using one network address can obtain other addresses that are connected to the same server. Typically, it builds on a trunking detection facility by providing one or more methods by which candidate addresses are made available to the client, who can then use trunking detection to appropriately filter them.¶

  Despite the support for trunking detection, there was no description of trunking discovery provided in RFC 5661 [65], making it necessary to provide those means in this document.¶

The combination of a server network address and a particular connection type to be used by a connection is referred to as a "server endpoint". Although using different connection types may result in different ports being used, the use of different ports by multiple connections to the same network address in such cases is not the essence of the distinction between the two endpoints used. This is in contrast to the case of port-specific endpoints, in which the explicit specification of port numbers within network addresses is used to allow a single server node to support multiple NFS servers.¶

Two network addresses connected to the same server are said to be server-trunkable. Two such addresses support the use of client ID trunking, as described in Section 7.5.¶

Two network addresses connected to the same server such that those addresses can be used to support a single common session are referred to as session-trunkable. Note that two addresses may be server-trunkable without being session-trunkable, and that, when two connections of different connection types are made to the same network address and are based on a single file system location entry, they are always session-trunkable, independent of the connection type, as specified by Section 7.5, since their derivation from the same file system location entry, together with the identity of their network addresses, assures that both connections are to the same server and will return server-owner information, allowing session trunking to be used.¶

15.1.2. Terminology Related to File System Location

Regarding the terminology that relates to the construction of multi-server namespaces out of a set of local per-server namespaces:¶

• Each server has a set of exported file systems that may be accessed by NFSv4 clients. Typically, this is done by assigning each file system a name within the pseudo-fs associated with the server, although the pseudo-fs may be dispensed with if there is only a single exported file system. Each such file system is part of the server's local namespace, and can be considered as a file system instance within a larger multi-server namespace.¶
• The set of all exported file systems for a given server constitutes that server's local namespace.¶
• In some cases, a server will have a namespace more extensive than its local namespace by using features associated with attributes that provide file system location information. These features, which allow construction of a multi-server namespace, are all described in individual sections below and include referrals (Section 15.5.6), migration (Section 15.5.5), and replication (Section 15.5.4).¶
• A file system present in a server's pseudo-fs may have multiple file system instances on different servers associated with it. All such instances are considered replicas of one another. Whether such replicas can be used simultaneously is discussed in Section 15.11.1, while the level of coordination between them (important when switching between them) is discussed in Sections 15.11.2 through 15.11.8 below.¶
• When a file system is present in a server's pseudo-fs, but there is no corresponding local file system, it is said to be "absent". In such cases, all associated instances will be accessed on other servers.¶

Regarding the terminology that relates to attributes used in trunking discovery and other multi-server namespace features:¶

• File system location attributes include the fs_locations and fs_locations_info attributes.¶

• File system location entries provide the individual file system locations within the file system location attributes. Each such entry specifies a server, in the form of a hostname or an address, and an fs name, which designates the location of the file system within the server's local namespace. A file system location entry designates a set of server endpoints to which the client may establish connections. There may be multiple endpoints because a hostname may map to multiple network addresses and because multiple connection types may be used to communicate with a single network address. However, except where explicit port numbers are used to designate a set of servers within a single server node, all such endpoints MUST designate a way of connecting to a single server. The exact form of the location entry varies with the particular file system location attribute used, as described in Section 15.2.¶

  The network addresses used in file system location entries typically appear without port number indications and are used to designate a server at one of the standard ports for NFS access, e.g., 2049 for TCP or 20049 for use with RPC-over-RDMA. Port numbers may be used in file system location entries to designate servers (typically user-level ones) accessed using other port numbers. In the case where network addresses indicate trunking relationships, the use of an explicit port number is inappropriate since trunking is a relationship between network addresses. See Section 15.5.2 for details.¶

• File system location elements are derived from location entries, and each describes a particular network access path consisting of a network address and a location within the server's local namespace. Such location elements need not appear within a file system location attribute, but the existence of each location element derives from a corresponding location entry. When a location entry specifies an IP address, there is only a single corresponding location element. File system location entries that contain a hostname are resolved using DNS, and may result in one or more location elements. All location elements consist of a location address that includes the IP address of an interface to a server and an fs name, which is the location of the file system within the server's local namespace. The fs name can be empty if the server has no pseudo-fs and only a single exported file system at the root filehandle.¶
• Two file system location elements are said to be server-trunkable if they specify the same fs name and the location addresses are such that the location addresses are server-trunkable. When the corresponding network paths are used, the client will always be able to use client ID trunking, but will only be able to use session trunking if the paths are also session-trunkable.¶
• Two file system location elements are said to be session-trunkable if they specify the same fs name and the location addresses are such that the location addresses are session-trunkable. When the corresponding network paths are used, the client will be able to able to use either client ID trunking or session trunking.¶

Discussion of the term "replica" is complicated by the fact that the term was used in RFC 5661 [65] with a meaning different from that used in this document. In short, in [65] each replica is identified by a single network access path, while in the current document, a set of network access paths that have server-trunkable network addresses and the same root-relative file system pathname is considered to be a single replica with multiple network access paths.¶

Each set of server-trunkable location elements defines a set of available network access paths to a particular file system. When there are multiple such file systems, each of which containing the same data, these file systems are considered replicas of one another. Logically, such replication is symmetric, since the fs currently in use and an alternate fs are replicas of each other. Often, in other documents, the term "replica" is not applied to the fs currently in use, despite the fact that the replication relation is inherently symmetric.¶

15.4.1. GETATTR within an Absent File System

As mentioned above, an exception is made for GETATTR in that attributes may be obtained for a filehandle within an absent file system. This exception only applies if the attribute mask contains at least one attribute bit that indicates the client is interested in a result regarding an absent file system: fs_locations, fs_locations_info, or fs_status. If none of these attributes is requested, GETATTR will result in an NFS4ERR_MOVED error.¶

When a GETATTR is done on an absent file system, the set of supported attributes is very limited. Many attributes, including those that are normally REQUIRED, will not be available on an absent file system. In addition to the attributes mentioned above (fs_locations, fs_locations_info, fs_status), the following attributes SHOULD be available on absent file systems. In the case of Recommended attributes, they should be available at least to the same degree that they are available on present file systems.¶

change_policy:

This attribute is useful for absent file systems and can be helpful in summarizing to the client when any of the location-related attributes change.¶

fsid:

This attribute should be provided so that the client can determine file system boundaries, including, in particular, the boundary between present and absent file systems. This value must be different from any other fsid on the current server and need have no particular relationship to fsids on any particular destination to which the client might be directed.¶

mounted_on_fileid:

For objects at the top of an absent file system, this attribute needs to be available. Since the fileid is within the present parent file system, there should be no need to reference the absent file system to provide this information.¶

Other attributes SHOULD NOT be made available for absent file systems, even when it is possible to provide them. The server should not assume that more information is always better and should avoid gratuitously providing additional information.¶

When a GETATTR operation includes a bit mask for one of the attributes fs_locations, fs_locations_info, or fs_status, but where the bit mask includes attributes that are not supported, GETATTR will not return an error, but will return the mask of the actual attributes supported with the results.¶

Handling of VERIFY/NVERIFY is similar to GETATTR in that if the attribute mask does not include fs_locations, fs_locations_info, or fs_status, the error NFS4ERR_MOVED will result. It differs in that any appearance in the attribute mask of an attribute not supported for an absent file system (and note that this will include some normally REQUIRED attributes) will also cause an NFS4ERR_MOVED result.¶

15.4.2. READDIR and Absent File Systems

A READDIR performed when the current filehandle is within an absent file system will result in an NFS4ERR_MOVED error, since, unlike the case of GETATTR, no such exception is made for READDIR.¶

Attributes for an absent file system may be fetched via a READDIR for a directory in a present file system, when that directory contains the root directories of one or more absent file systems. In this case, the handling is as follows:¶

• If the attribute set requested includes one of the attributes fs_locations, fs_locations_info, or fs_status, then fetching of attributes proceeds normally and no NFS4ERR_MOVED indication is returned, even when the rdattr_error attribute is requested.¶
• If the attribute set requested does not include one of the attributes fs_locations, fs_locations_info, or fs_status, then if the rdattr_error attribute is requested, each directory entry for the root of an absent file system will report NFS4ERR_MOVED as the value of the rdattr_error attribute.¶
• If the attribute set requested does not include any of the attributes fs_locations, fs_locations_info, fs_status, or rdattr_error, then the occurrence of the root of an absent file system within the directory will result in the READDIR failing with an NFS4ERR_MOVED error.¶
• The unavailability of an attribute because of a file system's absence, even one that is ordinarily REQUIRED, does not result in any error indication. The set of attributes returned for the root directory of the absent file system in that case is simply restricted to those actually available.¶

15.5.1. Combining Multiple Uses in a Single Attribute

A file system location attribute will sometimes contain information relating to the location of multiple replicas, which may be used in different ways:¶

• File system location entries that relate to the file system instance currently in use provide trunking information, allowing the client to find additional network addresses by which the instance may be accessed.¶
• File system location entries that provide information about replicas to which access is to be transferred.¶
• Other file system location entries that relate to replicas that are available to use in the event that access to the current replica becomes unsatisfactory.¶

In order to simplify client handling and to allow the best choice of replicas to access, the server should adhere to the following guidelines:¶

• All file system location entries that relate to a single file system instance should be adjacent.¶
• File system location entries that relate to the instance currently in use should appear first.¶
• File system location entries that relate to replica(s) to which migration is occurring should appear before replicas that are available for later use if the current replica should become inaccessible.¶

15.5.2. File System Location Attributes and Trunking

Trunking is the use of multiple connections between a client and server in order to increase the speed of data transfer. A client may determine the set of network addresses to use to access a given file system in a number of ways:¶

• When the name of the server is known to the client, it may use DNS to obtain a set of network addresses to use in accessing the server.¶
• The client may fetch the file system location attribute for the file system. This will provide either the name of the server (which can be turned into a set of network addresses using DNS) or a set of server-trunkable location entries. Using the latter alternative, the server can provide addresses it regards as desirable to use to access the file system in question. Although these entries can contain port numbers, these port numbers are not used in determining trunking relationships. Once the candidate addresses have been determined and EXCHANGE_ID done to the proper server, only the value of the so_major_id field returned by the servers in question determines whether a trunking relationship actually exists.¶

When the client fetches a location attribute for a file system, it should be noted that the client may encounter multiple entries for a number of reasons, such that when it determines trunking information, it may need to bypass addresses not trunkable with one already known.¶

The server can provide location entries that include either names or network addresses. It might use the latter form because of DNS-related security concerns or because the set of addresses to be used might require active management by the server.¶

Location entries used to discover candidate addresses for use in trunking are subject to change, as discussed in Section 15.5.7 below. The client may respond to such changes by using additional addresses once they are verified or by ceasing to use existing ones. The server can force the client to cease using an address by returning NFS4ERR_MOVED when that address is used to access a file system. This allows a transfer of client access that is similar to migration, although the same file system instance is accessed throughout.¶

15.5.3. File System Location Attributes and Connection Type Selection

Because of the need to support multiple types of connections, clients face the issue of determining the proper connection type to use when establishing a connection to a given server network address. In some cases, this issue can be addressed through the use of the connection "step-up" facility described in Section 22.36. However, because there are cases in which that facility is not available, the client may have to choose a connection type with no possibility of changing it within the scope of a single connection.¶

The two file system location attributes differ as to the information made available in this regard. The fs_locations attribute provides no information to support connection type selection. As a result, clients supporting multiple connection types would need to attempt to establish connections using multiple connection types until the one preferred by the client is successfully established.¶

The fs_locations_info attribute includes the FSLI4TF_RDMA flag, which is convenient for a client wishing to use RDMA. When this flag is set, it indicates that RPC-over-RDMA support is available using the specified location entry. A client can establish a TCP connection and then convert that connection to use RDMA by using the step-up facility.¶

Irrespective of the particular attribute used, when there is no indication that a step-up operation can be performed, a client supporting RDMA operation can establish a new RDMA connection, and it can be bound to the session already established by the TCP connection, allowing the TCP connection to be dropped and the session converted to further use in RDMA mode, if the server supports that.¶

15.5.4. File System Replication

The fs_locations and fs_locations_info attributes provide alternative file system locations, to be used to access data in place of or in addition to the current file system instance. On first access to a file system, the client should obtain the set of alternate locations by interrogating the fs_locations or fs_locations_info attribute, with the latter being preferred.¶

In the event that the occurrence of server failures, communications problems, or other difficulties make continued access to the current file system impossible or otherwise impractical, the client can use the alternate locations as a way to get continued access to its data.¶

The alternate locations may be physical replicas of the (typically read-only) file system data supplemented by possible asynchronous propagation of updates. Alternatively, they may provide for the use of various forms of server clustering in which multiple servers provide alternate ways of accessing the same physical file system. How the difference between replicas affects file system transitions can be represented within the fs_locations and fs_locations_info attributes, and how the client deals with file system transition issues will be discussed in detail in later sections.¶

Although the location attributes provide some information about the nature of the inter-replica transition, many aspects of the semantics of possible asynchronous updates are not currently described by the protocol, which makes it necessary for clients using replication to switch among replicas undergoing change to familiarize themselves with the semantics of the update approach used. Due to this lack of specificity, many applications may find the use of migration more appropriate because a server can propagate all updates made before an established point in time to the new replica as part of the migration event.¶

15.5.5. File System Migration

When a file system is present and becomes inaccessible using the current access path, the NFSv4.1 protocol provides a means by which clients can be given the opportunity to have continued access to their data. This may involve using a different access path to the existing replica or providing a path to a different replica. The new access path or the location of the new replica is specified by a file system location attribute. The ensuing migration of access includes the ability to retain locks across the transition. Depending on circumstances, this can involve:¶

• The continued use of the existing clientid when accessing the current replica using a new access path.¶
• Use of lock reclaim, taking advantage of a per-fs grace period.¶
• Use of Transparent State Migration.¶

Typically, a client will be accessing the file system in question, get an NFS4ERR_MOVED error, and then use a file system location attribute to determine the new access path for the data. When fs_locations_info is used, additional information will be available that will define the nature of the client's handling of the transition to a new server.¶

In most instances, servers will choose to migrate all clients using a particular file system to a successor replica at the same time to avoid cases in which different clients are updating different replicas. However, migration of an individual client can be helpful in providing load balancing, as long as the replicas in question are such that they represent the same data as described in Section 15.11.8.¶

• In the case in which there is no transition between replicas (i.e., only a change in access path), there are no special difficulties in using of this mechanism to effect load balancing.¶
• In the case in which the two replicas are sufficiently coordinated as to allow a single client coherent, simultaneous access to both, there is, in general, no obstacle to the use of migration of particular clients to effect load balancing. Generally, such simultaneous use involves cooperation between servers to ensure that locks granted on two coordinated replicas cannot conflict and can remain effective when transferred to a common replica.¶
• In the case in which a large set of clients is accessing a file system in a read-only fashion, it can be helpful to migrate all clients with writable access simultaneously, while using load balancing on the set of read-only copies, as long as the rules in Section 15.11.8, which are designed to prevent data reversion, are followed.¶

In other cases, the client might not have sufficient guarantees of data similarity or coherence to function properly (e.g., the data in the two replicas is similar but not identical), and the possibility that different clients are updating different replicas can exacerbate the difficulties, making the use of load balancing in such situations a perilous enterprise.¶

The protocol does not specify how the file system will be moved between servers or how updates to multiple replicas will be coordinated. It is anticipated that a number of different server-to-server coordination mechanisms might be used, with the choice left to the server implementer. The NFSv4.1 protocol specifies the method used to communicate the migration event between client and server.¶

In the case of various forms of server clustering, the new location may be another server providing access to the same physical file system. The client's responsibilities in dealing with this transition will depend on whether a switch between replicas has occurred and the means the server has chosen to provide continuity of locking state. These issues will be discussed in detail below.¶

Although a single successor location is typical, multiple locations may be provided. When multiple locations are provided, the client will typically use the first one provided. If that is inaccessible for some reason, later ones can be used. In such cases, the client might consider the transition to the new replica to be a migration event, even though some of the servers involved might not be aware of the use of the server that was inaccessible. In such a case, a client might lose access to locking state as a result of the access transfer.¶

When an alternate location is designated as the target for migration, it must designate the same data (with metadata being the same to the degree indicated by the fs_locations_info attribute). Where file systems are writable, a change made on the original file system must be visible on all migration targets. Where a file system is not writable but represents a read-only copy (possibly periodically updated) of a writable file system, similar requirements apply to the propagation of updates. Any change visible in the original file system must already be effected on all migration targets, to avoid any possibility that a client, in effecting a transition to the migration target, will see any reversion in file system state.¶

15.5.6. Referrals

Referrals allow the server to associate a file system namespace entry located on one server with a file system located on another server. When this includes the use of pure referrals, servers are provided a way of placing a file system in a location within the namespace essentially without respect to its physical location on a particular server. This allows a single server or a set of servers to present a multi-server namespace that encompasses file systems located on a wider range of servers. Some likely uses of this facility include establishment of site-wide or organization-wide namespaces, with the eventual possibility of combining such together into a truly global namespace, such as the one provided by AFS (the Andrew File System) [64].¶

Referrals occur when a client determines, upon first referencing a position in the current namespace, that it is part of a new file system and that the file system is absent. When this occurs, typically upon receiving the error NFS4ERR_MOVED, the actual location or locations of the file system can be determined by fetching a locations attribute.¶

The file system location attribute may designate a single file system location or multiple file system locations, to be selected based on the needs of the client. The server, in the fs_locations_info attribute, may specify priorities to be associated with various file system location choices. The server may assign different priorities to different locations as reported to individual clients, in order to adapt to client physical location or to effect load balancing. When both read-only and read-write file systems are present, some of the read-only locations might not be absolutely up-to-date (as they would have to be in the case of replication and migration). Servers may also specify file system locations that include client-substituted variables so that different clients are referred to different file systems (with different data contents) based on client attributes such as CPU architecture.¶

If the fs_locations_info attribute lists multiple possible targets, the relationships among them may be important to the client in selecting which one to use. The same rules specified in Section 15.5.5 below regarding multiple migration targets apply to these multiple replicas as well. For example, the client might prefer a writable target on a server that has additional writable replicas to which it subsequently might switch. Note that, as distinguished from the case of replication, there is no need to deal with the case of propagation of updates made by the current client, since the current client has not accessed the file system in question.¶

Use of multi-server namespaces is enabled by NFSv4.1 but is not required. The use of multi-server namespaces and their scope will depend on the applications used and system administration preferences.¶

Multi-server namespaces can be established by a single server providing a large set of pure referrals to all of the included file systems. Alternatively, a single multi-server namespace may be administratively segmented with separate referral file systems (on separate servers) for each separately administered portion of the namespace. The top-level referral file system or any segment may use replicated referral file systems for higher availability.¶

Generally, multi-server namespaces are for the most part uniform, in that the same data made available to one client at a given location in the namespace is made available to all clients at that namespace location. However, there are facilities provided that allow different clients to be directed to different sets of data, for reasons such as enabling adaptation to such client characteristics as CPU architecture. These facilities are described in Section 15.17.3.¶

Note that it is possible, when providing a uniform namespace, to provide different location entries to different clients in order to provide each client with a copy of the data physically closest to it or otherwise optimize access (e.g., provide load balancing).¶

15.5.7. Changes in a File System Location Attribute

Although clients will typically fetch a file system location attribute when first accessing a file system and when NFS4ERR_MOVED is returned, a client can choose to fetch the attribute periodically, in which case, the value fetched may change over time.¶

For clients not prepared to access multiple replicas simultaneously (see Section 15.11.1), the handling of the various cases of location change are as follows:¶

• Changes in the list of replicas or in the network addresses associated with replicas do not require immediate action. The client will typically update its list of replicas to reflect the new information.¶
• Additions to the list of network addresses for the current file system instance need not be acted on promptly. However, to prepare for a subsequent migration event, the client can choose to take note of the new address and then use it whenever it needs to switch access to a new replica.¶
• Deletions from the list of network addresses for the current file system instance do not require the client to immediately cease use of existing access paths, although new connections are not to be established on addresses that have been deleted. However, clients can choose to act on such deletions by preparing for an eventual shift in access, which becomes unavoidable as soon as the server returns NFS4ERR_MOVED to indicate that a particular network access path is not usable to access the current file system.¶

For clients that are prepared to access several replicas simultaneously, the following additional cases need to be addressed. As in the cases discussed above, changes in the set of replicas need not be acted upon promptly, although the client has the option of adjusting its access even in the absence of difficulties that would lead to the selection of a new replica.¶

• When a new replica is added, which may be accessed simultaneously with one currently in use, the client is free to use the new replica immediately.¶
• When a replica currently in use is deleted from the list, the client need not cease using it immediately. However, since the server may subsequently force such use to cease (by returning NFS4ERR_MOVED), clients might decide to limit the need for later state transfer. For example, new opens might be done on other replicas, rather than on one not present in the list.¶

15.11.1. File System Transitions and Simultaneous Access

The fs_locations_info attribute (described in Section 15.17) may indicate that two replicas may be used simultaneously, although some situations in which such simultaneous access is permitted are more appropriately described as instances of trunking (see Section 15.5.4.1). Although situations in which multiple replicas may be accessed simultaneously are somewhat similar to those in which a single replica is accessed by multiple network addresses, there are important differences since locking state is not shared among multiple replicas.¶

Because of this difference in state handling, many clients will not have the ability to take advantage of the fact that such replicas represent the same data. Such clients will not be prepared to use multiple replicas simultaneously but will access each file system using only a single replica, although the replica selected might make multiple server-trunkable addresses available.¶

Clients who are prepared to use multiple replicas simultaneously can divide opens among replicas however they choose. Once that choice is made, any subsequent transitions will treat the set of locking state associated with each replica as a single entity.¶

For example, if one of the replicas become unavailable, access will be transferred to a different replica, which is also capable of simultaneous access with the one still in use.¶

When there is no such replica, the transition may be to the replica already in use. At this point, the client has a choice between merging the locking state for the two replicas under the aegis of the sole replica in use or treating these separately until another replica capable of simultaneous access presents itself.¶

15.11.2. Filehandles and File System Transitions

There are a number of ways in which filehandles can be handled across a file system transition. These can be divided into two broad classes depending upon whether the two file systems across which the transition happens share sufficient state to effect some sort of continuity of file system handling.¶

When there is no such cooperation in filehandle assignment, the two file systems are reported as being in different handle classes. In this case, all filehandles are assumed to expire as part of the file system transition. Note that this behavior does not depend on the fh_expire_type attribute and supersedes the specification of the FH4_VOL_MIGRATION bit, which only affects behavior when fs_locations_info is not available.¶

When there is cooperation in filehandle assignment, the two file systems are reported as being in the same handle classes. In this case, persistent filehandles remain valid after the file system transition, while volatile filehandles (excluding those that are only volatile due to the FH4_VOL_MIGRATION bit) are subject to expiration on the target server.¶

15.11.3. Fileids and File System Transitions

In NFSv4.0, the issue of continuity of fileids in the event of a file system transition was not addressed. The general expectation had been that in situations in which the two file system instances are created by a single vendor using some sort of file system image copy, fileids would be consistent across the transition, while in the analogous multi-vendor transitions they would not. This poses difficulties, especially for the client without special knowledge of the transition mechanisms adopted by the server. Note that although fileid is not a REQUIRED attribute, many servers support fileids and many clients provide APIs that depend on fileids.¶

It is important to note that while clients themselves may have no trouble with a fileid changing as a result of a file system transition event, applications do typically have access to the fileid (e.g., via stat). The result is that an application may work perfectly well if there is no file system instance transition or if any such transition is among instances created by a single vendor, yet be unable to deal with the situation in which a multi-vendor transition occurs at the wrong time.¶

Providing the same fileids in a multi-vendor (multiple server vendors) environment has generally been held to be quite difficult. While there is work to be done, it needs to be pointed out that this difficulty is partly self-imposed. Servers have typically identified fileid with inode number, i.e. with a quantity used to find the file in question. This identification poses special difficulties for migration of a file system between vendors where assigning the same index to a given file may not be possible. Note here that a fileid is not required to be useful to find the file in question, only that it is unique within the given file system. Servers prepared to accept a fileid as a single piece of metadata and store it apart from the value used to index the file information can relatively easily maintain a fileid value across a migration event, allowing a truly transparent migration event.¶

In any case, where servers can provide continuity of fileids, they should, and the client should be able to find out that such continuity is available and take appropriate action. Information about the continuity (or lack thereof) of fileids across a file system transition is represented by specifying whether the file systems in question are of the same fileid class.¶

Note that when consistent fileids do not exist across a transition (either because there is no continuity of fileids or because fileid is not a supported attribute on one of instances involved), and there are no reliable filehandles across a transition event (either because there is no filehandle continuity or because the filehandles are volatile), the client is in a position where it cannot verify that files it was accessing before the transition are the same objects. It is forced to assume that no object has been renamed, and, unless there are guarantees that provide this (e.g., the file system is read-only), problems for applications may occur. Therefore, use of such configurations should be limited to situations where the problems that this may cause can be tolerated.¶

15.11.4. Fsids and File System Transitions

Since fsids are generally only unique on a per-server basis, it is likely that they will change during a file system transition. Clients should not make the fsids received from the server visible to applications since they may not be globally unique, and because they may change during a file system transition event. Applications are best served if they are isolated from such transitions to the extent possible.¶

Although normally a single source file system will transition to a single target file system, there is a provision for splitting a single source file system into multiple target file systems, by specifying the FSLI4F_MULTI_FS flag.¶

15.11.5. The Change Attribute and File System Transitions

Since the change attribute is defined as a server-specific one, change attributes fetched from one server are normally presumed to be invalid on another server. Such a presumption is troublesome since it would invalidate all cached change attributes, requiring refetching. Even more disruptive, the absence of any assured continuity for the change attribute means that even if the same value is retrieved on refetch, no conclusions can be drawn as to whether the object in question has changed. The identical change attribute could be merely an artifact of a modified file with a different change attribute construction algorithm, with that new algorithm just happening to result in an identical change value.¶

When the two file systems have consistent change attribute formats, and this fact is communicated to the client by reporting in the same change class, the client may assume a continuity of change attribute construction and handle this situation just as it would be handled without any file system transition.¶

15.11.6. Write Verifiers and File System Transitions

In a file system transition, the two file systems might be cooperating in the handling of unstably written data. Clients can determine if this is the case by seeing if the two file systems belong to the same write-verifier class. When this is the case, write verifiers returned from one system may be compared to those returned by the other and superfluous writes can be avoided.¶

When two file systems belong to different write-verifier classes, any verifier generated by one must not be compared to one provided by the other. Instead, the two verifiers should be treated as not equal even when the values are identical.¶

15.11.7. READDIR Cookies and Verifiers and File System Transitions

In a file system transition, the two file systems might be consistent in their handling of READDIR cookies and verifiers. Clients can determine if this is the case by seeing if the two file systems belong to the same readdir class. When this is the case, readdir class, READDIR cookies, and verifiers from one system will be recognized by the other, and READDIR operations started on one server can be validly continued on the other simply by presenting the cookie and verifier returned by a READDIR operation done on the first file system to the second.¶

When two file systems belong to different readdir classes, any READDIR cookie and verifier generated by one is not valid on the second and must not be presented to that server by the client. The client should act as if the verifier were rejected.¶

15.11.8. File System Data and File System Transitions

When multiple replicas exist and are used simultaneously or in succession by a client, applications using them will normally expect that they contain either the same data or data that is consistent with the normal sorts of changes that are made by other clients updating the data of the file system (with metadata being the same to the degree indicated by the fs_locations_info attribute). However, when multiple file systems are presented as replicas of one another, the precise relationship between the data of one and the data of another is not, as a general matter, specified by the NFSv4.1 protocol. It is quite possible to present as replicas file systems where the data of those file systems is sufficiently different that some applications have problems dealing with the transition between replicas. The namespace will typically be constructed so that applications can choose an appropriate level of support, so that in one position in the namespace, a varied set of replicas might be listed, while in another, only those that are up-to-date would be considered replicas. The protocol does define three special cases of the relationship among replicas to be specified by the server and relied upon by clients:¶

• When multiple replicas exist and are used simultaneously by a client (see the FSLIB4_CLSIMUL definition within fs_locations_info), they must designate the same data. Where file systems are writable, a change made on one instance must be visible on all instances at the same time, regardless of whether the interrogated instance is the one on which the modification was done. This allows a client to use these replicas simultaneously without any special adaptation to the fact that there are multiple replicas, beyond adapting to the fact that locks obtained on one replica are maintained separately (i.e., under a different client ID). In this case, locks (whether share reservations or byte-range locks) and delegations obtained on one replica are immediately reflected on all replicas, in the sense that access from all other servers is prevented regardless of the replica used. However, because the servers are not required to treat two associated client IDs as representing the same client, it is best to access each file using only a single client ID.¶
• When one replica is designated as the successor instance to another existing instance after the return of NFS4ERR_MOVED (i.e., the case of migration), the client may depend on the fact that all changes written to stable storage on the original instance are written to stable storage of the successor (uncommitted writes are dealt with in Section 15.11.6 above).¶
• Where a file system is not writable but represents a read-only copy (possibly periodically updated) of a writable file system, clients have similar requirements with regard to the propagation of updates. They may need a guarantee that any change visible on the original file system instance must be immediately visible on any replica before the client transitions access to that replica, in order to avoid any possibility that a client, in effecting a transition to a replica, will see any reversion in file system state. The specific means of this guarantee varies based on the value of the fss_type field that is reported as part of the fs_status attribute (see Section 15.18). Since these file systems are presumed to be unsuitable for simultaneous use, there is no specification of how locking is handled; in general, locks obtained on one file system will be separate from those on others. Since these are expected to be read-only file systems, this is not likely to pose an issue for clients or applications.¶

When none of these special situations applies, there is no basis within the protocol for the client to make assumptions about the contents of a replica file system or its relationship to previous file system instances. Thus, switching between nominally identical read-write file systems would not be possible because either the client does not use the fs_locations_info attribute, or the server does not support it.¶

15.11.9. Lock State and File System Transitions

While accessing a file system, clients obtain locks enforced by the server, which may prevent actions by other clients that are inconsistent with those locks.¶

When access is transferred between replicas, clients need to be assured that the actions disallowed by holding these locks cannot have occurred during the transition. This can be ensured by the methods below. Unless at least one of these is implemented, clients will not be assured of continuity of lock possession across a migration event:¶

• Providing the client an opportunity to re-obtain his locks via a per-fs grace period on the destination server, denying all clients using the destination file system the opportunity to obtain new locks that conflict with those held by the transferred client as long as that client has not completed its per-fs grace period. Because the lock reclaim mechanism was originally defined to support server reboot, it implicitly assumes that filehandles will, upon reclaim, be the same as those at open. In the case of migration, this requires that source and destination servers use the same filehandles, as evidenced by using the same server scope (see Section 7.4) or by showing this agreement using fs_locations_info (see Section 15.11.2 above).¶

  Note that such a grace period can be implemented without interfering with the ability of non-transferred clients to obtain new locks while it is going on. As long as the destination server is aware of the transferred locks, it can distinguish requests to obtain new locks that contrast with existing locks from those that do not, allowing it to treat such client requests without reference to the ongoing grace period.¶

• Locking state can be transferred as part of the transition by providing Transparent State Migration as described in Section 15.12.¶

Of these, Transparent State Migration provides the smoother experience for clients in that there is no need to go through a reclaim process before new locks can be obtained; however, it requires a greater degree of inter-server coordination. In general, the servers taking part in migration are free to provide either facility. However, when the filehandles can differ across the migration event, Transparent State Migration is the only available means of providing the needed functionality.¶

It should be noted that these two methods are not mutually exclusive and that a server might well provide both. In particular, if there is some circumstance preventing a specific lock from being transferred transparently, the destination server can allow it to be reclaimed by implementing a per-fs grace period for the migrated file system.¶

15.12.1. Transparent State Migration and pNFS

When pNFS is involved, the protocol is capable of supporting:¶

• Migration of the Metadata Server (MDS), leaving the Data Servers (DSs) in place.¶
• Migration of the file system as a whole, including the MDS and associated DSs.¶
• Replacement of one DS by another.¶
• Migration of a pNFS file system to one in which pNFS is not used.¶
• Migration of a file system not using pNFS to one in which layouts are available.¶

Note that migration, per se, is only involved in the transfer of the MDS function. Although the servicing of a layout may be transferred from one data server to another, this not done using the file system location attributes. The MDS can effect such transfers by recalling or revoking existing layouts and granting new ones on a different data server.¶

Migration of the MDS function is directly supported by Transparent State Migration. Layout state will normally be transparently transferred, just as other state is. As a result, Transparent State Migration provides a framework in which, given appropriate inter-MDS data transfer, one MDS can be substituted for another.¶

Migration of the file system function as a whole can be accomplished by recalling all layouts as part of the initial phase of the migration process. As a result, I/O will be done through the MDS during the migration process, and new layouts can be granted once the client is interacting with the new MDS. An MDS can also effect this sort of transition by revoking all layouts as part of Transparent State Migration, as long as the client is notified about the loss of locking state.¶

In order to allow migration to a file system on which pNFS is not supported, clients need to be prepared for a situation in which layouts are not available or supported on the destination file system and so direct I/O requests to the destination server, rather than depending on layouts being available.¶

Replacement of one DS by another is not addressed by migration as such but can be effected by an MDS recalling layouts for the DS to be replaced and issuing new ones to be served by the successor DS.¶

Migration may transfer a file system from a server that does not support pNFS to one that does. In order to properly adapt to this situation, clients that support pNFS, but function adequately in its absence, should check for pNFS support when a file system is migrated and be prepared to use pNFS when support is available on the destination.¶

15.13.1. Client Transition Notifications

When there is a change in the network access path that a client is to use to access a file system, there are a number of related status indications with which clients need to deal:¶

• If an attempt is made to use or return a filehandle within a file system that is no longer accessible at the address previously used to access it, the error NFS4ERR_MOVED is returned.¶

  Exceptions are made to allow such filehandles to be used when interrogating a file system location attribute. This enables a client to determine a new replica's location or a new network access path.¶

  This condition continues on subsequent attempts to access the file system in question. The only way the client can avoid the error is to cease accessing the file system in question at its old server location and access it instead using a different address at which it is now available.¶

• Whenever a client sends a SEQUENCE operation to a server that generated state held on that client and associated with a file system no longer accessible on that server, the response will contain the status bit SEQ4_STATUS_LEASE_MOVED, indicating that there has been a lease migration.¶

  This condition continues until the client acknowledges the notification by fetching a file system location attribute for the file system whose network access path is being changed. When there are multiple such file systems, a location attribute for each such file system needs to be fetched. The location attribute for all migrated file systems needs to be fetched in order to clear the condition. Even after the condition is cleared, the client needs to respond by using the location information to access the file system at its new location to ensure that leases are not needlessly expired.¶

Unlike NFSv4.0, in which the corresponding conditions are both errors and thus mutually exclusive, in NFSv4.1 the client can, and often will, receive both indications on the same request. As a result, implementations need to address the question of how to coordinate the necessary recovery actions when both indications arrive in the response to the same request. It should be noted that when processing an NFSv4 COMPOUND, the server will normally decide whether SEQ4_STATUS_LEASE_MOVED is to be set before it determines which file system will be referenced or whether NFS4ERR_MOVED is to be returned.¶

Since these indications are not mutually exclusive in NFSv4.1, the following combinations are possible results when a COMPOUND is issued:¶

• The COMPOUND status is NFS4ERR_MOVED, and SEQ4_STATUS_LEASE_MOVED is asserted.¶

  In this case, transition recovery is required. While it is possible that migration discovery is needed in addition, it is likely that only the accessed file system has transitioned. In any case, because addressing NFS4ERR_MOVED is necessary to allow the rejected requests to be processed on the target, dealing with it will typically have priority over migration discovery.¶

• The COMPOUND status is NFS4ERR_MOVED, and SEQ4_STATUS_LEASE_MOVED is clear.¶

  In this case, transition recovery is also required. It is clear that migration discovery is not needed to find file systems that have been migrated other than the one returning NFS4ERR_MOVED. Cases in which this result can arise include a referral or a migration for which there is no associated locking state. This can also arise in cases in which an access path transition other than migration occurs within the same server. In such a case, there is no need to set SEQ4_STATUS_LEASE_MOVED, since the lease remains associated with the current server even though the access path has changed.¶

• The COMPOUND status is not NFS4ERR_MOVED, and SEQ4_STATUS_LEASE_MOVED is asserted.¶

  In this case, no transition recovery activity is required on the file system(s) accessed by the request. However, to prevent avoidable lease expiration, migration discovery needs to be done.¶

• The COMPOUND status is not NFS4ERR_MOVED, and SEQ4_STATUS_LEASE_MOVED is clear.¶

  In this case, neither transition-related activity nor migration discovery is required.¶

Note that the specified actions only need to be taken if they are not already going on. For example, when NFS4ERR_MOVED is received while accessing a file system for which transition recovery is already occurring, the client merely waits for that recovery to be completed, while the receipt of the SEQ4_STATUS_LEASE_MOVED indication only needs to initiate migration discovery for a server if such discovery is not already underway for that server.¶

The fact that a lease-migrated condition does not result in an error in NFSv4.1 has a number of important consequences. In addition to the fact that the two indications are not mutually exclusive, as discussed above, there are number of issues that are important in considering implementation of migration discovery, as discussed in Section 15.13.2.¶

Because SEQ4_STATUS_LEASE_MOVED is not an error condition, it is possible for file systems whose access paths have not changed to be successfully accessed on a given server even though recovery is necessary for other file systems on the same server. As a result, access can take place while:¶

• The migration discovery process is happening for that server.¶
• The transition recovery process is happening for other file systems connected to that server.¶

15.13.2. Performing Migration Discovery

Migration discovery can be performed in the same context as transition recovery, allowing recovery for each migrated file system to be invoked as it is discovered. Alternatively, it may be done in a separate migration discovery thread, allowing migration discovery to be done in parallel with one or more instances of transition recovery.¶

In either case, because the lease-migrated indication does not result in an error, other access to file systems on the server can proceed normally, with the possibility that further such indications will be received, raising the issue of how such indications are to be dealt with. In general:¶

• No action needs to be taken for such indications received by any threads performing migration discovery, since continuation of that work will address the issue.¶
• In other cases in which migration discovery is currently being performed, nothing further needs to be done to respond to such lease migration indications, as long as one can be certain that the migration discovery process would deal with those indications. See below for details.¶
• For such indications received in all other contexts, the appropriate response is to initiate or otherwise provide for the execution of migration discovery for file systems associated with the server IP address returning the indication.¶

This leaves a potential difficulty in situations in which the migration discovery process is near to completion but is still operating. One should not ignore a SEQ4_STATUS_LEASE_MOVED indication if the migration discovery process is not able to respond to the discovery of additional migrating file systems without additional aid. A further complexity relevant in addressing such situations is that a lease-migrated indication may reflect the server's state at the time the SEQUENCE operation was processed, which may be different from that in effect at the time the response is received. Because new migration events may occur at any time, and because a SEQ4_STATUS_LEASE_MOVED indication may reflect the situation in effect a considerable time before the indication is received, special care needs to be taken to ensure that SEQ4_STATUS_LEASE_MOVED indications are not inappropriately ignored.¶

A useful approach to this issue involves the use of separate externally-visible migration discovery states for each server. Separate values could represent the various possible states for the migration discovery process for a server:¶

• Non-operation, in which migration discovery is not being performed.¶
• Normal operation, in which there is an ongoing scan for migrated file systems.¶
• Completion/verification of migration discovery processing, in which the possible completion of migration discovery processing needs to be verified.¶

Given that framework, migration discovery processing would proceed as follows:¶

• While in the normal-operation state, the thread performing discovery would fetch, for successive file systems known to the client on the server being worked on, a file system location attribute plus the fs_status attribute.¶
• If the fs_status attribute indicates that the file system is a migrated one (i.e., fss_absent is true, and fss_type != STATUS4_REFERRAL), then a migrated file system has been found. In this situation, it is likely that the fetch of the file system location attribute has cleared one of the file systems contributing to the lease-migrated indication.¶
• In cases in which that happened, the thread cannot know whether the lease-migrated indication has been cleared, and so it enters the completion/verification state and proceeds to issue a COMPOUND to see if the SEQ4_STATUS_LEASE_MOVED indication has been cleared.¶
• When the discovery process is in the completion/verification state, if other requests get a lease-migrated indication, they note that it was received. Later, the existence of such indications is used when the request completes, as described below.¶

When the request used in the completion/verification state completes:¶

• If a lease-migrated indication is returned, the discovery continues normally. Note that this is so even if all file systems have been traversed, since new migrations could have occurred while the process was going on.¶
• Otherwise, if there is any record that other requests saw a lease-migrated indication while the request was occurring, that record is cleared, and the verification request is retried. The discovery process remains in the completion/verification state.¶
• If there have been no lease-migrated indications, the work of migration discovery is considered completed, and it enters the non-operating state. Once it enters this state, subsequent lease-migrated indications will trigger a new migration discovery process.¶

It should be noted that the process described above is not guaranteed to terminate, as a long series of new migration events might continually delay the clearing of the SEQ4_STATUS_LEASE_MOVED indication. To prevent unnecessary lease expiration, it is appropriate for clients to use the discovery of migrations to effect lease renewal immediately, rather than waiting for the clearing of the SEQ4_STATUS_LEASE_MOVED indication when the complete set of migrations is available.¶

Lease discovery needs to be provided as described above. This ensures that the client discovers file system migrations soon enough to renew its leases on each destination server before they expire. Non-renewal of leases can lead to loss of locking state. While the consequences of such loss can be ameliorated through implementations of courtesy locks, servers are under no obligation to do so, and a conflicting lock request may mean that a lock is revoked unexpectedly. Clients should be aware of this possibility.¶

15.13.3. Overview of Client Response to NFS4ERR_MOVED

This section outlines a way in which a client that receives NFS4ERR_MOVED can effect transition recovery by using a new server or server endpoint if one is available. As part of that process, it will determine:¶

• Whether the NFS4ERR_MOVED indicates migration has occurred, or whether it indicates another sort of file system access transition as discussed in Section 15.10 above.¶
• In the case of migration, whether Transparent State Migration has occurred.¶
• Whether any state has been lost during the process of Transparent State Migration.¶
• Whether sessions have been transferred as part of Transparent State Migration.¶

During the first phase of this process, the client proceeds to examine file system location entries to find the initial network address it will use to continue access to the file system or its replacement. For each location entry that the client examines, the process consists of five steps:¶

1. Performing an EXCHANGE_ID directed at the location address. This operation is used to register the client owner (in the form of a client_owner4) with the server, to obtain a client ID to be used subsequently to communicate with it, to obtain that client ID's confirmation status, and to determine server_owner4 and scope for the purpose of determining if the entry is trunkable with the address previously being used to access the file system (i.e., that it represents another network access path to the same file system and can share locking state with it).¶
2. Making an initial determination of whether migration has occurred. The initial determination will be based on whether the EXCHANGE_ID results indicate that the current location element is server-trunkable with that used to access the file system when access was terminated by receiving NFS4ERR_MOVED. If it is, then migration has not occurred. In that case, the transition is dealt with, at least initially, as one involving continued access to the same file system on the same server through a new network address.¶
3. Obtaining access to existing session state or creating new sessions. How this is done depends on the initial determination of whether migration has occurred and can be done as described in Section 15.13.4 below in the case of migration or as described in Section 15.13.5 below in the case of a network address transfer without migration.¶
4. Verifying the trunking relationship assumed in step 2 as discussed in Section 7.5.1. Although this step will generally confirm the initial determination, it is possible for verification to invalidate the initial determination of network address shift (without migration) and instead determine that migration had occurred. There is no need to redo step 3 above, since it will be possible to continue use of the session established already.¶
5. Obtaining access to existing locking state and/or re-obtaining it. How this is done depends on the final determination of whether migration has occurred and can be done as described below in Section 15.13.4 in the case of migration or as described in Section 15.13.5 in the case of a network address transfer without migration.¶

Once the initial address has been determined, clients are free to apply an abbreviated process to find additional addresses trunkable with it (clients may seek session-trunkable or server-trunkable addresses depending on whether they support client ID trunking). During this later phase of the process, further location entries are examined using the abbreviated procedure specified below:¶

A:

Before the EXCHANGE_ID, the fs name of the location entry is examined, and if it does not match that currently being used, the entry is ignored. Otherwise, one proceeds as specified by step 1 above.¶

B:

In the case that the network address is session-trunkable with one used previously, a BIND_CONN_TO_SESSION is used to access that session using the new network address. Otherwise, or if the bind operation fails, a CREATE_SESSION is done.¶

C:

The verification procedure referred to in step 4 above is used. However, if it fails, the entry is ignored and the next available entry is used.¶

15.13.4. Obtaining Access to Sessions and State after Migration

In the event that migration has occurred, migration recovery will involve determining whether Transparent State Migration has occurred. This decision is made based on the client ID returned by the EXCHANGE_ID and the reported confirmation status.¶

• If the client ID is an unconfirmed client ID not previously known to the client, then Transparent State Migration has not occurred.¶
• If the client ID is a confirmed client ID previously known to the client, then any transferred state would have been merged with an existing client ID representing the client to the destination server. In this state merger case, Transparent State Migration might or might not have occurred, and a determination as to whether it has occurred is deferred until sessions are established and the client is ready to begin state recovery.¶
• If the client ID is a confirmed client ID not previously known to the client, then the client can conclude that the client ID was transferred as part of Transparent State Migration. In this transferred client ID case, Transparent State Migration has occurred, although some state might have been lost.¶

Once the client ID has been obtained, it is necessary to obtain access to sessions to continue communication with the new server. In any of the cases in which Transparent State Migration has occurred, it is possible that a session was transferred as well. To deal with that possibility, clients can, after doing the EXCHANGE_ID, issue a BIND_CONN_TO_SESSION to connect the transferred session to a connection to the new server. If that fails, it is an indication that the session was not transferred and that a new session needs to be created to take its place.¶

In some situations, it is possible for a BIND_CONN_TO_SESSION to succeed without session migration having occurred. If state merger has taken place, then the associated client ID may have already had a set of existing sessions, with it being possible that the session ID of a given session is the same as one that might have been migrated. In that event, a BIND_CONN_TO_SESSION might succeed, even though there could have been no migration of the session with that session ID. In such cases, the client will receive sequence errors when the slot sequence values used are not appropriate on the new session. When this occurs, the client can create a new a session and cease using the existing one.¶

Once the client has determined the initial migration status, and determined that there was a shift to a new server, it needs to re-establish its locking state, if possible. To enable this to happen without loss of the guarantees normally provided by locking, the destination server needs to implement a per-fs grace period in all cases in which lock state was lost, including those in which Transparent State Migration was not implemented. Each client for which there was a transfer of locking state to the new server will have the duration of the grace period to reclaim its locks, from the time its locks were transferred.¶

Clients need to deal with the following cases:¶

• In the state merger case, it is possible that the server has not attempted Transparent State Migration, in which case state may have been lost without it being reflected in the SEQ4_STATUS bits. To determine whether this has happened, the client can use TEST_STATEID to check whether the stateids created on the source server are still accessible on the destination server. Once a single stateid is found to have been successfully transferred, the client can conclude that Transparent State Migration was begun, and any failure to transport all of the stateids will be reflected in the SEQ4_STATUS bits. Otherwise, Transparent State Migration has not occurred.¶
• In a case in which Transparent State Migration has not occurred, the client can use the per-fs grace period provided by the destination server to reclaim locks that were held on the source server.¶
• In a case in which Transparent State Migration has occurred, and no lock state was lost (as shown by SEQ4_STATUS flags), no lock reclaim is necessary.¶
• In a case in which Transparent State Migration has occurred, and some lock state was lost (as shown by SEQ4_STATUS flags), existing stateids need to be checked for validity using TEST_STATEID, and reclaim used to re-establish any that were not transferred.¶

For all of the cases above, RECLAIM_COMPLETE with an rca_one_fs value of TRUE needs to be done before normal use of the file system, including obtaining new locks for the file system. This applies even if no locks were lost and there was no need for any to be reclaimed.¶

15.13.5. Obtaining Access to Sessions and State after Network Address Transfer

The case in which there is a transfer to a new network address without migration is similar to that described in Section 15.13.4 above in that there is a need to obtain access to needed sessions and locking state. However, the details are simpler and will vary depending on the type of trunking between the address receiving NFS4ERR_MOVED and that to which the transfer is to be made.¶

To make a session available for use, a BIND_CONN_TO_SESSION should be used to obtain access to the session previously in use. Only if this fails, should a CREATE_SESSION be done. While this procedure mirrors that in Section 15.13.4 above, there is an important difference in that preservation of the session is not purely optional but depends on the type of trunking.¶

Access to appropriate locking state will generally need no actions beyond access to the session. However, the SEQ4_STATUS bits need to be checked for lost locking state, including the need to reclaim locks after a server reboot, since there is always a possibility of locking state being lost.¶

15.14.1. Server Responsibilities in Effecting State Reclaim after Migration

In this case, the destination server needs no knowledge of the locks held on the source server. It relies on the clients to accurately report (via reclaim operations) the locks previously held, and does not allow new locks to be granted on migrated file systems until the grace period expires. Disallowing of new locks applies to all clients accessing these file systems, while grace period expiration occurs for each migrated client independently.¶

During this grace period, clients have the opportunity to use reclaim operations to obtain locks for file system objects within the migrated file system, in the same way that they do when recovering from server restart, and the servers typically rely on clients to accurately report their locks, although they have the option of subjecting these requests to verification. If the clients only reclaim locks held on the source server, no conflict can arise. Once the client has reclaimed its locks, it indicates the completion of lock reclamation by performing a RECLAIM_COMPLETE specifying rca_one_fs as TRUE.¶

While it is not necessary for source and destination servers to cooperate to transfer information about locks, implementations are well advised to consider transferring the following useful information:¶

• If information about the set of clients that have locking state for the transferred file system is made available, the destination server will be able to terminate the grace period once all such clients have reclaimed their locks, allowing normal locking activity to resume earlier than it would have otherwise.¶
• Locking summary information for individual clients (at various possible levels of detail) can detect some instances in which clients do not accurately represent the locks held on the source server.¶

15.14.2. Server Responsibilities in Effecting Transparent State Migration

The basic responsibility of the source server in effecting Transparent State Migration is to make available to the destination server a description of each piece of locking state associated with the file system being migrated. In addition to client id string and verifier, the source server needs to provide for each stateid:¶

• The stateid including the current sequence value.¶
• The associated client ID.¶
• The handle of the associated file.¶
• The type of the lock, such as open, byte-range lock, delegation, or layout.¶
• For locks such as opens and byte-range locks, there will be information about the owner(s) of the lock.¶
• For recallable/revocable lock types, the current recall status needs to be included.¶
• For each lock type, there will be associated type-specific information. For opens, this will include share and deny mode while for byte-range locks and layouts, there will be a type and a byte-range.¶

Such information will most probably be organized by client id string on the destination server so that it can be used to provide appropriate context to each client when it makes itself known to the client. Issues connected with a client impersonating another by presenting another client's client id string can be addressed using NFSv4.1 state protection features, as described in Section 25.¶

A further server responsibility concerns locks that are revoked or otherwise lost during the process of file system migration. Because locks that appear to be lost during the process of migration will be reclaimed by the client, the servers have to take steps to ensure that locks revoked soon before or soon after migration are not inadvertently allowed to be reclaimed in situations in which the continuity of lock possession cannot be assured.¶

• For locks lost on the source but whose loss has not yet been acknowledged by the client (by using FREE_STATEID), the destination must be aware of this loss so that it can deny a request to reclaim them.¶
• For locks lost on the destination after the state transfer but before the client's RECLAIM_COMPLETE is done, the destination server should note these and not allow them to be reclaimed.¶

An additional responsibility of the cooperating servers concerns situations in which a stateid cannot be transferred transparently because it conflicts with an existing stateid held by the client and associated with a different file system. In this case, there are two valid choices:¶

• Treat the transfer, as in NFSv4.0, as one without Transparent State Migration. In this case, conflicting locks cannot be granted until the client does a RECLAIM_COMPLETE, after reclaiming the locks it had, with the exception of reclaims denied because they were attempts to reclaim locks that had been lost.¶
• Implement Transparent State Migration, except for the lock with the conflicting stateid. In this case, the client will be aware of a lost lock (through the SEQ4_STATUS flags) and be allowed to reclaim it.¶

When transferring state between the source and destination, the issues discussed in Section 7.2 of [67] must still be attended to. In this case, the use of NFS4ERR_DELAY may still be necessary in NFSv4.1, as it was in NFSv4.0, to prevent locking state changing while it is being transferred. See Section 19.1.1.3 for information about appropriate client retry approaches in the event that NFS4ERR_DELAY is returned.¶

There are a number of important differences in the NFS4.1 context:¶

• The absence of RELEASE_LOCKOWNER means that the one case in which an operation could not be deferred by use of NFS4ERR_DELAY no longer exists.¶
• Sequencing of operations is no longer done using owner-based operation sequences numbers. Instead, sequencing is session- based.¶

As a result, when sessions are not transferred, the techniques discussed in Section 7.2 of [67] are adequate and will not be further discussed.¶

15.14.3. Server Responsibilities in Effecting Session Transfer

The basic responsibility of the source server in effecting session transfer is to make available to the destination server a description of the current state of each slot with the session, including the following:¶

• The last sequence value received for that slot.¶
• Whether there is cached reply data for the last request executed and, if so, the cached reply.¶

When sessions are transferred, there are a number of issues that pose challenges in terms of making the transferred state unmodifiable during the period it is gathered up and transferred to the destination server:¶

• A single session may be used to access multiple file systems, not all of which are being transferred.¶
• Requests made on a session may, even if rejected, affect the state of the session by advancing the sequence number associated with the slot used.¶

As a result, when the file system state might otherwise be considered unmodifiable, the client might have any number of in-flight requests, each of which is capable of changing session state, which may be of a number of types:¶

1. Those requests that were processed on the migrating file system before migration began.¶
2. Those requests that received the error NFS4ERR_DELAY because the file system being accessed was in the process of being migrated.¶
3. Those requests that received the error NFS4ERR_MOVED because the file system being accessed had been migrated.¶
4. Those requests that accessed the migrating file system in order to obtain location or status information.¶
5. Those requests that did not reference the migrating file system.¶

It should be noted that the history of any particular slot is likely to include a number of these request classes. In the case in which a session that is migrated is used by file systems other than the one migrated, requests of class 5 may be common and may be the last request processed for many slots.¶

Since session state can change even after the locking state has been fixed as part of the migration process, the session state known to the client could be different from that on the destination server, which necessarily reflects the session state on the source server at an earlier time. In deciding how to deal with this situation, it is helpful to distinguish between two sorts of behavioral consequences of the choice of initial sequence ID values:¶

• The error NFS4ERR_SEQ_MISORDERED is returned when the sequence ID in a request is neither equal to the last one seen for the current slot nor the next greater one.¶

  In view of the difficulty of arriving at a mutually acceptable value for the correct last sequence value at the point of migration, it may be necessary for the server to show some degree of forbearance when the sequence ID is one that would be considered unacceptable if session migration were not involved.¶

• Returning the cached reply for a previously executed request when the sequence ID in the request matches the last value recorded for the slot.¶

  In the cases in which an error is returned and there is no possibility of any non-idempotent operation having been executed, it may not be necessary to adhere to this as strictly as might be proper if session migration were not involved. For example, the fact that the error NFS4ERR_DELAY was returned may not assist the client in any material way, while the fact that NFS4ERR_MOVED was returned by the source server may not be relevant when the request was reissued and directed to the destination server.¶

An important issue is that the specification needs to take note of all potential COMPOUNDs, even if they might be unlikely in practice. For example, a COMPOUND is allowed to access multiple file systems and might perform non-idempotent operations in some of them before accessing a file system being migrated. Also, a COMPOUND may return considerable data in the response before being rejected with NFS4ERR_DELAY or NFS4ERR_MOVED, and may in addition be marked as sa_cachethis. However, note that if the client and server adhere to rules in Section 19.1.1.3, there is no possibility of non-idempotent operations being spuriously reissued after receiving NFS4ERR_DELAY response.¶

To address these issues, a destination server MAY do any of the following when implementing session transfer:¶

• Avoid enforcing any sequencing semantics for a particular slot until the client has established the starting sequence for that slot on the destination server.¶
• For each slot, avoid returning a cached reply returning NFS4ERR_DELAY or NFS4ERR_MOVED until the client has established the starting sequence for that slot on the destination server.¶
• Until the client has established the starting sequence for a particular slot on the destination server, avoid reporting NFS4ERR_SEQ_MISORDERED or returning a cached reply that contains either NFS4ERR_DELAY or NFS4ERR_MOVED and consists solely of a series of operations where the response is NFS4_OK until the final error.¶

Because of the considerations mentioned above, including the rules for the handling of NFS4ERR_DELAY included in Section 19.1.1.3, the destination server can respond appropriately to SEQUENCE operations received from the client by adopting the three policies listed below:¶

• Not responding with NFS4ERR_SEQ_MISORDERED for the initial request on a slot within a transferred session because the destination server cannot be aware of requests made by the client after the server handoff but before the client became aware of the shift. In cases in which NFS4ERR_SEQ_MISORDERED would normally have been reported, the request is to be processed normally as a new request.¶
• Replying as it would for a retry whenever the sequence matches that transferred by the source server, even though this would not provide retry handling for requests issued after the server handoff, under the assumption that, when such requests are issued, they will never be responded to in a state-changing fashion, making retry support for them unnecessary.¶
• Once a non-retry SEQUENCE is received for a given slot, using that as the basis for further sequence checking, with no further reference to the sequence value transferred by the source server.¶

15.15.1. Referral Example (LOOKUP)

Let us suppose that the following COMPOUND is sent in an environment in which /this/is/the/path is absent from the target server. This may be for a number of reasons. It may be that the file system has moved, or it may be that the target server is functioning mainly, or solely, to refer clients to the servers on which various file systems are located.¶

• PUTROOTFH¶
• LOOKUP "this"¶
• LOOKUP "is"¶
• LOOKUP "the"¶
• LOOKUP "path"¶
• GETFH¶
• GETATTR (fsid, fileid, size, time_modify)¶

Under the given circumstances, the following will be the result.¶

• PUTROOTFH --> NFS_OK. The current fh is now the root of the pseudo-fs.¶
• LOOKUP "this" --> NFS_OK. The current fh is for /this and is within the pseudo-fs.¶
• LOOKUP "is" --> NFS_OK. The current fh is for /this/is and is within the pseudo-fs.¶
• LOOKUP "the" --> NFS_OK. The current fh is for /this/is/the and is within the pseudo-fs.¶
• LOOKUP "path" --> NFS_OK. The current fh is for /this/is/the/path and is within a new, absent file system, but ... the client will never see the value of that fh.¶
• GETFH --> NFS4ERR_MOVED. Fails because current fh is in an absent file system at the start of the operation, and the specification makes no exception for GETFH.¶
• GETATTR (fsid, fileid, size, time_modify). Not executed because the failure of the GETFH stops processing of the COMPOUND.¶

Given the failure of the GETFH, the client has the job of determining the root of the absent file system and where to find that file system, i.e., the server and path relative to that server's root fh. Note that in this example, the client did not obtain filehandles and attribute information (e.g., fsid) for the intermediate directories, so that it would not be sure where the absent file system starts. It could be the case, for example, that /this/is/the is the root of the moved file system and that the reason that the look up of "path" succeeded is that the file system was not absent on that operation but was moved between the last LOOKUP and the GETFH (since COMPOUND is not atomic). Even if we had the fsids for all of the intermediate directories, we could have no way of knowing that /this/is/the/path was the root of a new file system, since we don't yet have its fsid.¶

In order to get the necessary information, let us re-send the chain of LOOKUPs with GETFHs and GETATTRs to at least get the fsids so we can be sure where the appropriate file system boundaries are. The client could choose to get fs_locations_info at the same time but in most cases the client will have a good guess as to where file system boundaries are (because of where NFS4ERR_MOVED was, and was not, received) making fetching of fs_locations_info unnecessary.¶

OP01:

PUTROOTFH --> NFS_OK¶

• Current fh is root of pseudo-fs.¶

OP02:

GETATTR(fsid) --> NFS_OK¶

• Just for completeness. Normally, clients will know the fsid of the pseudo-fs as soon as they establish communication with a server.¶

OP03:

LOOKUP "this" --> NFS_OK¶

OP04:

GETATTR(fsid) --> NFS_OK¶

• Get current fsid to see where file system boundaries are. The fsid will be that for the pseudo-fs in this example, so no boundary.¶

OP05:

GETFH --> NFS_OK¶

• Current fh is for /this and is within pseudo-fs.¶

OP06:

LOOKUP "is" --> NFS_OK¶

• Current fh is for /this/is and is within pseudo-fs.¶

OP07:

GETATTR(fsid) --> NFS_OK¶

• Get current fsid to see where file system boundaries are. The fsid will be that for the pseudo-fs in this example, so no boundary.¶

OP08:

GETFH --> NFS_OK¶

• Current fh is for /this/is and is within pseudo-fs.¶

OP09:

LOOKUP "the" --> NFS_OK¶

• Current fh is for /this/is/the and is within pseudo-fs.¶

OP10:

GETATTR(fsid) --> NFS_OK¶

• Get current fsid to see where file system boundaries are. The fsid will be that for the pseudo-fs in this example, so no boundary.¶

OP11:

GETFH --> NFS_OK¶

• Current fh is for /this/is/the and is within pseudo-fs.¶

OP12:

LOOKUP "path" --> NFS_OK¶

• Current fh is for /this/is/the/path and is within a new, absent file system, but ...¶
• The client will never see the value of that fh.¶

OP13:

GETATTR(fsid, fs_locations_info) --> NFS_OK¶

• We are getting the fsid to know where the file system boundaries are. In this operation, the fsid will be different than that of the parent directory (which in turn was retrieved in OP10). Note that the fsid we are given will not necessarily be preserved at the new location. That fsid might be different, and in fact the fsid we have for this file system might be a valid fsid of a different file system on that new server.¶
• In this particular case, we are pretty sure anyway that what has moved is /this/is/the/path rather than /this/is/the since we have the fsid of the latter and it is that of the pseudo-fs, which presumably cannot move. However, in other examples, we might not have this kind of information to rely on (e.g., /this/is/the might be a non-pseudo file system separate from /this/is/the/path), so we need to have other reliable source information on the boundary of the file system that is moved. If, for example, the file system /this/is had moved, we would have a case of migration rather than referral, and once the boundaries of the migrated file system was clear we could fetch fs_locations_info.¶
• We are fetching fs_locations_info because the fact that we got an NFS4ERR_MOVED at this point means that it is most likely that this is a referral and we need the destination. Even if it is the case that /this/is/the is a file system that has migrated, we will still need the location information for that file system.¶

OP14:

GETFH --> NFS4ERR_MOVED¶

• Fails because current fh is in an absent file system at the start of the operation, and the specification makes no exception for GETFH. Note that this means the server will never send the client a filehandle from within an absent file system.¶

Given the above, the client knows where the root of the absent file system is (/this/is/the/path) by noting where the change of fsid occurred (between "the" and "path"). The fs_locations_info attribute also gives the client the actual location of the absent file system, so that the referral can proceed. The server gives the client the bare minimum of information about the absent file system so that there will be very little scope for problems of conflict between information sent by the referring server and information of the file system's home. No filehandles and very few attributes are present on the referring server, and the client can treat those it receives as transient information with the function of enabling the referral.¶

15.15.2. Referral Example (READDIR)

Another context in which a client may encounter referrals is when it does a READDIR on a directory in which some of the sub-directories are the roots of absent file systems.¶

Suppose such a directory is read as follows:¶

• PUTROOTFH¶
• LOOKUP "this"¶
• LOOKUP "is"¶
• LOOKUP "the"¶
• READDIR (fsid, size, time_modify, mounted_on_fileid)¶

In this case, because rdattr_error is not requested, fs_locations_info is not requested, and some of the attributes cannot be provided, the result will be an NFS4ERR_MOVED error on the READDIR, with the detailed results as follows:¶

• PUTROOTFH --> NFS_OK. The current fh is at the root of the pseudo-fs.¶
• LOOKUP "this" --> NFS_OK. The current fh is for /this and is within the pseudo-fs.¶
• LOOKUP "is" --> NFS_OK. The current fh is for /this/is and is within the pseudo-fs.¶
• LOOKUP "the" --> NFS_OK. The current fh is for /this/is/the and is within the pseudo-fs.¶
• READDIR (fsid, size, time_modify, mounted_on_fileid) --> NFS4ERR_MOVED. Note that the same error would have been returned if /this/is/the had migrated, but it is returned because the directory contains the root of an absent file system.¶

So now suppose that we re-send with rdattr_error:¶

• PUTROOTFH¶
• LOOKUP "this"¶
• LOOKUP "is"¶
• LOOKUP "the"¶
• READDIR (rdattr_error, fsid, size, time_modify, mounted_on_fileid)¶

The results will be:¶

• PUTROOTFH --> NFS_OK. The current fh is at the root of the pseudo-fs.¶
• LOOKUP "this" --> NFS_OK. The current fh is for /this and is within the pseudo-fs.¶
• LOOKUP "is" --> NFS_OK. The current fh is for /this/is and is within the pseudo-fs.¶
• LOOKUP "the" --> NFS_OK. The current fh is for /this/is/the and is within the pseudo-fs.¶
• READDIR (rdattr_error, fsid, size, time_modify, mounted_on_fileid) --> NFS_OK. The attributes for directory entry with the component named "path" will only contain rdattr_error with the value NFS4ERR_MOVED, together with an fsid value and a value for mounted_on_fileid.¶

Suppose we do another READDIR to get fs_locations_info (although we could have used a GETATTR directly, as in Section 15.15.1).¶

• PUTROOTFH¶
• LOOKUP "this"¶
• LOOKUP "is"¶
• LOOKUP "the"¶
• READDIR (rdattr_error, fs_locations_info, mounted_on_fileid, fsid, size, time_modify)¶

The results would be:¶

• PUTROOTFH --> NFS_OK. The current fh is at the root of the pseudo-fs.¶
• LOOKUP "this" --> NFS_OK. The current fh is for /this and is within the pseudo-fs.¶
• LOOKUP "is" --> NFS_OK. The current fh is for /this/is and is within the pseudo-fs.¶
• LOOKUP "the" --> NFS_OK. The current fh is for /this/is/the and is within the pseudo-fs.¶
• READDIR (rdattr_error, fs_locations_info, mounted_on_fileid, fsid, size, time_modify) --> NFS_OK. The attributes will be as shown below.¶

The attributes for the directory entry with the component named "path" will only contain:¶

• rdattr_error (value: NFS_OK)¶
• fs_locations_info¶
• mounted_on_fileid (value: unique fileid within referring file system)¶
• fsid (value: unique value within referring server)¶

The attributes for entry "path" will not contain size or time_modify because these attributes are not available within an absent file system.¶

15.17.1. The fs_locations_server4 Structure

The fs_locations_server4 structure consists of the following items in addition to the fls_server field, which specifies a network address or set of addresses to be used to access the specified file system. Note that both of these items (i.e., fls_currency and fls_info) specify attributes of the file system replica and should not be different when there are multiple fs_locations_server4 structures, each specifying a network path to the chosen replica, for the same replica.¶

When these values are different in two fs_locations_server4 structures, a client has no basis for choosing one over the other and is best off simply ignoring both entries, whether these entries apply to migration replication or referral. When there are more than two such entries, majority voting can be used to exclude a single erroneous entry from consideration. In the case in which trunking information is provided for a replica currently being accessed, the additional trunked addresses can be ignored while access continues on the address currently being used, even if the entry corresponding to that path might be considered invalid.¶

• An indication of how up-to-date the file system is (fls_currency) in seconds. This value is relative to the master copy. A negative value indicates that the server is unable to give any reasonably useful value here. A value of zero indicates that the file system is the actual writable data or a reliably coherent and fully up-to-date copy. Positive values indicate how out-of-date this copy can normally be before it is considered for update. Such a value is not a guarantee that such updates will always be performed on the required schedule but instead serves as a hint about how far the copy of the data would be expected to be behind the most up-to-date copy.¶
• A counted array of one-byte values (fls_info) containing information about the particular file system instance. This data includes general flags, transport capability flags, file system equivalence class information, and selection priority information. The encoding will be discussed below.¶
• The server string (fls_server). For the case of the replica currently being accessed (via GETATTR), a zero-length string MAY be used to indicate the current address being used for the RPC call. The fls_server field can also be an IPv4 or IPv6 address, formatted the same way as an IPv4 or IPv6 address in the "server" field of the fs_location4 data type (see Section 15.16).¶

With the exception of the transport-flag field (at offset FSLI4BX_TFLAGS with the fls_info array), all of this data defined in this specification applies to the replica specified by the entry, rather than the specific network path used to access it. The classification of data in extensions to this data is discussed below.¶

Data within the fls_info array is in the form of 8-bit data items with constants giving the offsets within the array of various values describing this particular file system instance. This style of definition was chosen, in preference to explicit XDR structure definitions for these values, for a number of reasons.¶

• The kinds of data in the fls_info array, representing flags, file system classes, and priorities among sets of file systems representing the same data, are such that 8 bits provide a quite acceptable range of values. Even where there might be more than 256 such file system instances, having more than 256 distinct classes or priorities is unlikely.¶
• Explicit definition of the various specific data items within XDR would limit expandability in that any extension within would require yet another attribute, leading to specification and implementation clumsiness. In the context of the NFSv4 extension model in effect at the time fs_locations_info was designed (i.e., that which is described in RFC 5661 [65]), this would necessitate a new minor version to effect any Standards Track extension to the data in fls_info.¶

The set of fls_info data is subject to expansion in a future minor version or in a Standards Track RFC within the context of a single minor version. The server SHOULD NOT send and the client MUST NOT use indices within the fls_info array or flag bits that are not defined in Standards Track RFCs.¶

In light of the new extension model defined in RFC 8178 [32] and the fact that the individual items within fls_info are not explicitly referenced in the XDR, the following practices should be followed when extending or otherwise changing the structure of the data returned in fls_info within the scope of a single minor version:¶

• All extensions need to be described by Standards Track documents. There is no need for such documents to be marked as updating RFC 5661 [65] or this document.¶
• It needs to be made clear whether the information in any added data items applies to the replica specified by the entry or to the specific network paths specified in the entry.¶
• There needs to be a reliable way defined to determine whether the server is aware of the extension. This may be based on the length field of the fls_info array, but it is more flexible to provide fs-scope or server-scope attributes to indicate what extensions are provided.¶

This encoding scheme can be adapted to the specification of multi-byte numeric values, even though none are currently defined. If extensions are made via Standards Track RFCs, multi-byte quantities will be encoded as a range of bytes with a range of indices, with the byte interpreted in big-endian byte order. Further, any such index assignments will be constrained by the need for the relevant quantities not to cross XDR word boundaries.¶

The fls_info array currently contains:¶

• Two 8-bit flag fields, one devoted to general file-system characteristics and a second reserved for transport-related capabilities.¶
• Six 8-bit class values that define various file system equivalence classes as explained below.¶
• Four 8-bit priority values that govern file system selection as explained below.¶

The general file system characteristics flag (at byte index FSLI4BX_GFLAGS) has the following bits defined within it:¶

• FSLI4GF_WRITABLE indicates that this file system target is writable, allowing it to be selected by clients that may need to write on this file system. When the current file system instance is writable and is defined as of the same simultaneous use class (as specified by the value at index FSLI4BX_CLSIMUL) to which the client was previously writing, then it must incorporate within its data any committed write made on the source file system instance. See Section 15.11.6, which discusses the write-verifier class. While there is no harm in not setting this flag for a file system that turns out to be writable, turning the flag on for a read-only file system can cause problems for clients that select a migration or replication target based on the flag and then find themselves unable to write.¶
• FSLI4GF_CUR_REQ indicates that this replica is the one on which the request is being made. Only a single server entry may have this flag set and, in the case of a referral, no entry will have it set. Note that this flag might be set even if the request was made on a network access path different from any of those specified in the current entry.¶
• FSLI4GF_ABSENT indicates that this entry corresponds to an absent file system replica. It can only be set if FSLI4GF_CUR_REQ is set. When both such bits are set, it indicates that a file system instance is not usable but that the information in the entry can be used to determine the sorts of continuity available when switching from this replica to other possible replicas. Since this bit can only be true if FSLI4GF_CUR_REQ is true, the value could be determined using the fs_status attribute, but the information is also made available here for the convenience of the client. An entry with this bit, since it represents a true file system (albeit absent), does not appear in the event of a referral, but only when a file system has been accessed at this location and has subsequently been migrated.¶

• FSLI4GF_GOING indicates that a replica, while still available, should not be used further. The client, if using it, should make an orderly transfer to another file system instance as expeditiously as possible. It is expected that file systems going out of service will be announced as FSLI4GF_GOING some time before the actual loss of service. It is also expected that the fli_valid_for value will be sufficiently small to allow clients to detect and act on scheduled events, while large enough that the cost of the requests to fetch the fs_locations_info values will not be excessive. Values on the order of ten minutes seem reasonable.¶

  When this flag is seen as part of a transition into a new file system, a client might choose to transfer immediately to another replica, or it may reference the current file system and only transition when a migration event occurs. Similarly, when this flag appears as a replica in the referral, clients would likely avoid being referred to this instance whenever there is another choice.¶

  This flag, like the other items within fls_info, applies to the replica rather than to a particular path to that replica. When it appears, a transition to a new replica, rather than to a different path to the same replica, is indicated.¶

• FSLI4GF_SPLIT indicates that when a transition occurs from the current file system instance to this one, the replacement may consist of multiple file systems. In this case, the client has to be prepared for the possibility that objects on the same file system before migration will be on different ones after. Note that FSLI4GF_SPLIT is not incompatible with the file systems belonging to the same fileid class since, if one has a set of fileids that are unique within a file system, each subset assigned to a smaller file system after migration would not have any conflicts internal to that file system.¶

  A client, in the case of a split file system, will interrogate existing files with which it has continuing connection (it is free to simply forget cached filehandles). If the client remembers the directory filehandle associated with each open file, it may proceed upward using LOOKUPP to find the new file system boundaries. Note that in the event of a referral, there will not be any such files and so these actions will not be performed. Instead, a reference to a portion of the original file system now split off into other file systems will encounter an fsid change and possibly a further referral.¶

  Once the client recognizes that one file system has been split into two, it can prevent the disruption of running applications by presenting the two file systems as a single one until a convenient point to recognize the transition, such as a restart. This would require a mapping from the server's fsids to fsids as seen by the client, but this is already necessary for other reasons. As noted above, existing fileids within the two descendant file systems will not conflict. Providing non-conflicting fileids for newly created files on the split file systems is the responsibility of the server (or servers working in concert). The server can encode filehandles such that filehandles generated before the split event can be discerned from those generated after the split, allowing the server to determine when the need for emulating two file systems as one is over.¶

  Although it is possible for this flag to be present in the event of referral, it would generally be of little interest to the client, since the client is not expected to have information regarding the current contents of the absent file system.¶