Remote Procedure Call (RPC) Security Version 3
draft-williams-rpcsecgssv3-00.txt

Status of this Memo

This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as “work in progress.”

The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt.

The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html.

This Internet-Draft will expire on July 17, 2009.

Copyright Notice

Copyright (c) 2009 IETF Trust and the persons identified as the document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents in effect on the date of publication of this document (http://trustee.ietf.org/license-info). Please review these documents carefully, as they describe your rights and restrictions with respect to this document.

Abstract

This document specifies version 3 of the Remote Procedure Call (RPC) security protocol (RPCSEC_GSS). This protocol provides for: compound authentication of client hosts and users to server (constructed by generic composition), channel binding, security label assertions for multi-level and type enforcement, privilege assertions and identity assertions.

Table of Contents

1.  Introduction
1.1.  Motivation
1.2.  Applications of RPCSEC_GSSv3
1.3.  Conventions used in this document
2.  The RPCSEC_GSSv3 protocol
2.1.  Control messages
2.1.1.  New auth_stat values
2.1.2.  Create request
2.1.3.  Context handle destruction
2.1.4.  List request
2.1.5.  Extensibility
3.  Privileges and identity representation for NFSv4
4.  Security Considerations
5.  IANA Considerations
6.  Normative References
§  Author's Address

1.  Introduction

The original RPCSEC_GSS protocol [RFC2203] (Eisler, M., Chiu, A., and L. Ling, “RPCSEC_GSS Protocol Specification,” September 1997.) provided for authentication of RPC clients and servers to each other using the Generic Security Services Application Programming Interface (GSS-API) [RFC2743] (Linn, J., “Generic Security Service Application Program Interface Version 2, Update 1,” January 2000.). The second version of RPCSEC_GSS [add ref.] added support for channel binding [RFC5056] (Williams, N., “On the Use of Channel Bindings to Secure Channels,” November 2007.).

We find that GSS-API mechanisms are insufficient for communicating certain aspects of a client's identity and authority to a server. The GSS-API and its mechanisms certainly could be extended to address this shortcomming, but it seems be far simpler to address it at the application layer, namely, in this case, RPCSEC_GSS.

We therefore provide a new version of RPCSEC_GSS that allows for the following:

• compound authentication of the client host and user to the server (done by binding of two RPCSEC_GSS handles)
• channel binding (even though RPCSEC_GSSv2 provides this also; see below)
• client-side assertions of authority:
  • security labels (for multi-level, type enforcement, and other labeled security models) [add refs. for labeled security]
  • application-specific privileges
• client-side assertions of identity:
  • primary client/user identity
  • supplementary group memberships of the client/user, including support for specifying deltas to the membership list as seen on the server

Assertions of labels, privilege and identity are evaluated by the server, which may then map the asserted values to other values, all according to server-side policy.

We also add an option for enumerating server-side domains of interpretation (DOI), though this seems likely to be unnecessary.

RPCSEC_GSSv3 is patterned as follows:

• a client uses an existing RPCSEC_GSS context handle (of any RPCSEC_GSS version) to protect RPCSEC_GSSv3 exchanges (this will be termed the "parent" or "outer" handle)
• the server issues a "child" RPCSEC_GSSv3 handle, but the underlying GSS-API security context for the parent handle is used in all subsequent exchanges using the child handle (this works because the RPCSEC_GSS handle is included in the integrity protected RPCSEC_GSS auth/verifier header for all versions of RPCSEC_GSS)

This means that RPCSEC_GSSv3 depends on RPCSEC_GSS versions 1 and/or 2 for actual GSS-API security context establishment. This keeps the specification of RPCSEC_GSSv3 simple by avoiding the need to duplicate the core functionality of RPCSEC_GSS version 1.

1.1.  Motivation

The initial motivation for RPCSEC_GSSv3 is to add support for labeled security. Several alternatives to revising RPCSEC_GSS were considered:

a.

application-level protocol extensions, such as new operations for the Network File System version 4 (NFSv4) protocol [RFC3530] (Shepler, S., Callaghan, B., Robinson, D., Thurlow, R., Beame, C., Eisler, M., and D. Noveck, “Network File System (NFS) version 4 Protocol,” April 2003.);

b.

a stackable GSS-API pseudo-mechanism that could be composed with concrete GSS-API mechanisms to provide both, authentication and protected security label assertions;

c.

per-GSS-API mechanism extensions for transporting security label assertions;

Alternative (c) is not sufficiently general. One possible benefit of (c) might be the ability to have per-{user, label} credentials, though that might be difficult to manage (and, anyways, can be emulated with regular GSS-API mechanisms through principal naming conventions), whereas with the other approaches there is a single credential per-user that can be used to assert multiple security labels.

Alternative (a) is not general either, though for the purpose of the NFSv4 community it would suffice. However, a solution at the RPCSEC_GSS or GSS-API layers does, or arguably should, fit more naturally into most, if not all, NFSv4 implementations.

Alternative (b) is certainly general enough. In fact, it is more general than the RPCSEC_GSSv3 solution in that it could be used in non-RPC protocols that support the use of the GSS-API. However, the RPCSEC_GSSv3 approach is attractively simple. For example, to pursue (b) would likely entail having to specify a framework for mechanism composition, as well as GSS-API interfaces to access assertions that would typically be very platform-specific. (The KITTEN WG has explored stackable pseudo-mechanisms in the past, but that work is currently stagnant.) It is possible that stackable pseudo-mechanisms may materialize in the future; such mechanisms would be usable through all versions of RPCSEC_GSS so far.

As we considered these alternatives we also realized that we needed other features that could all be packed into a single solution. For example, the assertion of security label is conceptually equivalent, protocol-wise, to assertions of privilege and identity.

Additionally, assertions need to be verified, and in this case the one party that can verify an assertion is the client host, which can authenticate to the server using its own credentials. Yet we want to continue authenticating users as well. This calls for compound authentication.

Finally, because the design of RPCSEC_GSSv3 relies on RPCSEC_GSSv1 (though v2 can also be used) to do the actual GSS-API security context establishment, we add support for channel binding so that implementors who have implemented RPCSEC_GSSv1 but not version 2 can still provide channel binding without having to implement version 2. Channel binding is accomplished in a more simple manner in v3 also.

1.2.  Applications of RPCSEC_GSSv3

The common uses of RPCSEC_GSSv3, particularly for NFSv4, are expected to be:

a.

labeled security: client-side process label assertion [+ privilege assertion] + compound client host & user authentication;

b.

compound client host & user authentication [+ privilege assertion];

c.

client-side process credentials assertion [+ privilege assertion] as a replacement for AUTH_SYS that is more secure than AUTH_SYS while not requiring per-user credentials

1.3.  Conventions used in this document

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119] (Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels,” March 1997.).

2.  The RPCSEC_GSSv3 protocol

The following External Data Representation (XDR) [RFC4506] (Eisler, M., “XDR: External Data Representation Standard,” May 2006.) definitions are used:

	#define RPCSEC_GSS_VERS_3 3

	enum rpc_gss3_proc_t {
		RPCSEC_GSS3_DATA = 0,
		RPCSEC_GSS3_LIST = 1,
		RPCSEC_GSS3_CREATE = 2,
		RPCSEC_GSS3_DESTROY = 3
	};

	struct {
		rpc_gss3_proc_t		gss_proc;
		unsigned int		seq_num;
		rpc_gss_service_t	service;
		unsigned int		handle_version;
		opaque			handle<>;
	} rpc_gss_cred_vers_3_t;

	struct {
		int	type;
		bool	critical;
		opaque	data<>;
	} rpc_gss3_extension;

	struct {
		unsigned int	version;
		opaque		handle<>;
		opaque		nonce<>;
		opaque		mic<>;
	} rpc_gss3_gss_binding;

	typedef opaque rpc_gss3_chan_binding<>;

	typedef opaque rpc_gss3_doi<>;
	struct {
		rpc_gss3_doi	doi;
		opaque		label<>;
	} rpc_gss3_label;

	typedef opaque rpc_gss3_privs<>;

	typedef opaque rpc_gss3_name<>;

	struct {
		rpc_gss3_name	*username;
		rpc_gss3_name	*groups<>;
		rpc_gss3_name   *groups_add<>;
		rpc_gss3_name   *groups_remove<>;
	} rpc_gss3_id;

	enum rpc_gss3_assertion_type {
		LABEL = 0,
		PRIVS = 1,
		IDENTITY = 2
	};

	union switch (rpc_gss3_assertion_type atype) {
	case LABEL:
		rpc_gss3_label	label;
	case PRIVILEGES:
		rpc_gss3_privs	privs;
	case IDENTITY:
		rpc_gss3_id	id;
	default:
		opaque		ext<>;
	} rpc_gss3_assertion_u;

	struct rpc_gss3_assertion {
		bool			critical;
		rpc_gss3_assertion_u	assertion;
	};

	struct {
		rpc_gss3_gss_binding	*compound_binding;
		rpc_gss3_chan_binding	*chan_binding_mic;
		rpc_gss3_assertion	assertions<>;
		rpc_gss3_extension	extensions<>;
	} rpc_gss3_create_args;

	struct {
		opaque			handle<>;
		rpc_gss3_chan_binding	*chan_binding_mic;
		rpc_gss3_assertion	granted_assertions<>;
		rpc_gss3_assertion	server_assertions<>;
		rpc_gss3_extension	extensions<>;
	} rpc_gss3_create_res;

	enum rpc_gss3_list_item {
		DOI = 0,
		PRIV = 1,
		PRIV_GROUP = 2
	};

	struct {
		rpc_gss3_list_item	list_what<>;
	} rpc_gss3_list_args;

	union switch (rpc_gss3_list_item itype) {
	case DOI:
		rpc_gss3_doi	dois<>;
	case PRIV:
		string		privs;
	case PRIV_GROUP:
		string		priv_groups;
	default:
		opaque		ext<>;
	} rpc_gss3_list_item_u;

	typedef rpc_gss3_list_item_u rpc_gss3_list_res<>;

The rpc_gss_cred_vers_3_t type is used in much the same way that rpc_gss_cred_vers_1_t is used in RPCSEC_GSSv1, that is: as the arm of the rpc_gss_cred_t discriminated union corresponding to version 3 (RPCSEC_GSS_VERS_3). It differs from rpc_gss_cred_vers_1_t in that: a) the values for gss_proc corresponding to control messages are different, b) the presence of a field indicating the version of RPCSEC_GSS used to established the context handle used, if any.

RPC data messages using RPCSEC_GSSv3 context handles differ from RPCSEC_GSSv1 only in that the version number used MUST be '3' instead of '1' and, as described above, in that there is one more field in the RPCSEC_GSS header to name the version of RPCSEC_GSS used to establish the context handle used to protect this message. All other protocol elements from RPCSEC_GSSv1-protected RPC data messages MUST remain the same in v3 as in v1.

RPCSEC_GSSv3 control messages are the same as RPCSEC_GSSv3 data messages, but with a gss_proc value that indicates a control message is contained in the data payload.

2.1.  Control messages

There are two RPCSEC_GSSv3 control messages: RPCSEC_GSS3_CREATE and RPCSEC_GSS3_LIST.

The client MUST use one of the following security services to protect any RPCSEC_GSSv3 control message:

• rpc_gss_svc_channel_prot (see RPCSEC_GSSv2)
• rpc_gss_svc_integrity
• rpc_gss_svc_privacy

Specifically the client MUST NOT use rpc_gss_svc_none.

2.1.1.  New auth_stat values

RPCSEC_GSSv3 requires the addition of several values to the auth_stat enumerated type definition:

	enum auth_stat {
		...
		/*
		 * RPCSEC_GSS errors
		 */
		RPCSEC_GSS3_COMPOUND_PROBEM = <TBD>,
		RPCSEC_GSS3_LABEL_PROBLEM = <TBD>,
		RPCSEC_GSS3_IDENTITY_PROBLEM = <TBD>
		RPCSEC_GSS3_UNKNOWN_ASSERTION = <TBD>
		RPCSEC_GSS3_UNKNOWN_EXTENSION = <TBD>
		RPCSEC_GSS3_UNKNOWN_MESSAGE = <TBD>
	};

2.1.2.  Create request

The RPCSEC_GSS3_CREATE call message consists of inputs to bind into a new RPCSEC_GSSv3 handle. The context handle used to protect the RPCSEC_GSS3_CREATE call message is termed the "parent" (or "outer") handle. The reply to this message consists of either an error or a new RPCSEC_GSSv3 handle, termed the "child" handle.

All uses of a child context handle MUST use the GSS-API security context associated with the parent context handle of the RPCSEC_GSS3_CREATE request that produced the child context handle. The child context, however, has its own sequence number space and window, distinct from that of the parent.

As described in the introduction, the RPCSEC_GSS3_CREATE call message binds one or more items of several kinds into a new RPCSEC_GSSv3 context handle:

• another RPCSEC_GSS (version 1, 2, or 3) context handle
• a channel binding
• authorization assertions (label, privileges)
• identity assertions

Servers MUST either ignore, reject or apply policy to the authorization and identity assertions. Policies should take into account the identity of the client and/or user as authenticated via the GSS-API. Server implementation and policy MAY result in labels, privileges and identities being mapped to concepts and values that are local to the server.

2.1.2.1.  Compound authentication

RPCSEC_GSSv3 allows for compound authentication of client hosts and users to servers. This is done by using an integrity protected RPCSEC_GSSv3 message of RPCSEC_GSS3_CREATE type which includes a reference to the context handle to bind, a nonce and a MIC of that nonce using the GSS-API security context associated with the named context handle. We'll term the two context handles "parent" (or "outer") and "inner," and the resulting context handle the "child" handle, where the outer context handle is the context handle providing integrity protection to the RPCSEC_GSS3_CREATE message, and the inner context handle is the one referenced via the compound_binding field of the RPCSEC_GSS3_CREATE arguments structure (rpc_gss3_create_args).

All uses of a child context handle that is bound to an inner context MUST be treated as speaking for the initiator principal (as modified by any assertions in the RPCSEC_GSS3_CREATE message) of the inner context handle's GSS-API security context.

This feature is needed, for example, when a client wishes to use authority assertions that the server may only grant if a user and a client are authenticated together to the server. Thus a server may refuse to grant requested authority to a user acting alone (e.g., via an unprivileged user-space program), but may grant requested authority to a client acting on behalf of a user if the server trusts the client.

It is assumed that an unprivileged user-space program would not have access to client host credentials needed to establish a GSS-API security context authenticating the client to the server, therefore an unprivileged user-space program could not create an RPCSEC_GSSv3 RPCSEC_GSS3_CREATE message that successfully binds a client and a user security context.

Clients using RPCSEC_GSS context binding MUST use, as the outer context handle, an RPCSEC_GSS context handle that corresponds to a GSS-API security context that authenticates the client host, and for the inner context handle it SHOULD use a context handle to authenticates a user. The reverse (outer handle authenticates user, inner authenticates client) MUST NOT be used. Other compounds might eventually make sense.

An RPCSEC_GSSv3 context handle that is bound to another RPCSEC_GSS context MUST be treated by servers as authenticating the GSS-API initiator principal authenticated by the inner context handle's GSS-API security context. This principal may be mapped to a server-side notion of user or principal as modified by any identity assertions by the client in the same RPCSEC_GSS3_CREATE request that the server accepts.

2.1.2.2.  Channel binding

RPCSEC_GSSv3 provides a different way to do channel binding that RPCSEC_GSSv2. Specifically: a) RPCSEC_GSSv3 builds on RPCSEC_GSSv1 by reusing existing, established context handles rather than providing a different RPC security flavor for establishing context handles, b) channel bindings data are not hashed because the community now agrees that it is the secure channel's responsibility to produce channel bindings data of manageable size. (a) is useful in keeping RPCSEC_GSSv3 simple in general, not just for channel binding. (b) is useful in keeping RPCSEC_GSSv3 simple specifically for channel binding.

Channel binding is accomplished as follows. The client prefixes the channel bindings data octet string with the channel type as described in [RFC5056] (Williams, N., “On the Use of Channel Bindings to Secure Channels,” November 2007.), then the client calls GSS_GetMIC() to get a MIC of resulting octet string, using the outer RPCSEC_GSS context handle's GSS-API security context. The MIC is then placed in the chan_binding_mic field of RPCSEC_GSS3_CREATE arguments (rpc_gss3_create_args).

If the chan_binding_mic field of the arguments of a RPCSEC_GSS3_CREATE control message is set, then the server MUST verify the client's channel binding MIC if the server supports this feature. If channel binding verification succeeds then the server MUST generate a new MIC of the same channel bindings and place it in the chan_binding_mic field of the RPCSEC_GSS3_CREATE results. If channel binding verification fails or the server doesn't support channel binding then the server MUST indicate this in its reply by not including a chan_binding_mic value (chan_binding_mic is an optional field).

The client MUST verify the result's chan_binding_mic value, if the server included it, by calling GSS_VerifyMIC() with the given MIC and the channel bindings data (including the channel type prefix). If client-side channel binding verification fails then the client MUST call RPCSEC_GSS3_DESTROY. If the client requested channel binding but the server did not include a chan_binding_mic field in the results, then the client MAY continue to use the resulting context handle as though channel binding had never been requested, otherwise (if the client really wanted channel binding) it MUST call RPCSEC_GSS3_DESTROY.

As per-RPCSEC_GSSv2:

• "Once a successful [channel binding] procedure has been performed on an [RPCSEC_GSSv3] context handle, the initiator's implementation may map application requests for rpc_gss_svc_none and rpc_gss_svc_integrity to rpc_gss_svc_channel_prot credentials. And if the secure channel has privacy enabled, requests for rpc_gss_svc_privacy can also be mapped to rpc_gss_svc_channel_prot."
• ...

Any RPCSEC_GSSv3 context handle that has been bound to a secure channel in this way SHOULD be used only with the rpc_gss_svc_channel_prot, and SHOULD NOT be used with rpc_gss_svc_none nor rpc_gss_svc_integrity -- if the secure channel does not provide privacy protection then the client MAY use rpc_gss_svc_privacy where privacy protection is needed or desired.

2.1.2.3.  Label assertions

RPCSEC_GSSv3 clients MAY assert a security label in some DOI by binding this assertion into an RPCSEC_GSSv3 context handle. This is done by including an assertion of type rpc_gss3_label in the 'assertions' field (discriminant: 'LABEL') of the RPCSEC_GSS3_CREATE arguments to the desired DOI and label.

Label encoding is specific to each DOI and not described herein. DOI encoding is <TBD> [fill in... Solaris uses integers to name DOIs, and there is an IANA registry of DOIs as 32-bit integers, and IPsec (whence the IANA registry) and CALIPSO use 32-bit integers for DOIs as well. So a 32-bit unsinged integer seems to be the way to go. Add references... -Nico]

If a label itself requires privacy protection (i.e., that the user can assert that label is a secret) then the client MUST use the rpc_gss_svc_privacy protection service for the RPCSEC_GSS3_CREATE request or, if the parent handle is bound to a secure channel that provides privacy protection, rpc_gss_svc_channel_prot.

If a client wants to ensure that the server understands the asserted label then it MUST set the 'critical' field of the label assertion to TRUE, otherwise it MUST set it to FALSE.

Servers that don't support labeling MUST ignore non-critical label assertions. Servers that don't support the requested DOI MUST either ignore non-critical label assertions or map them to a suitable label in a supported DOI. Servers that don't support labeling or don't support the requested DOI MUST return an error if the label request is critical. Servers that support labeling in the requested DOI MAY map the requested label to different label as a result of server-side policy evaluation.

2.1.2.4.  Privilege assertions

Privilege assertions are similar to label assertions, except that there is no DOI, and the privileges supported are specified by the RPC application.

Privileges are encoded US-ASCII strings containing comma-separated privilege names, as well as up to one privilege group name and zero or more exclusions, where each exclusion is a privilege name or privilege group name prefixed with an exclamation point. Two special privilege group names are defined hereing: "all" (which represents all possible privileges) and "basic" (which represents privileges normally granted to all users).

RPC applications that wish to use this facility must define the set of known privileges, and must specify which privileges are in the "basic" privilege group. For example, NFSv4 might specify privileges for reading, writing, chowning, linking, etcetera.

2.1.2.5.  Identity assertions

Identity assertions can be used either to modify the set of groups assigned on the server-side to a given user (authenticated by the GSS-API) or to implement an AUTH_SYS-like [RFC1831] (Srinivasan, R., “RPC: Remote Procedure Call Protocol Specification Version 2,” August 1995.). In the latter case the client specifies at least a user-name and possibly groups that it thinks the user belongs to.

Clients may set a username, a group list, and/or lists of groups to be added or removed from the group list that the server would normally use for the given user.

The server MUST decide whether to accept identity assertions by applying local policy. Such policies is not described herein. Example policies:

• "always accept identity assertions"
• "always accept identity assertions where the identities are understood"
• "accept identity assertions ... only from trusted clients" (where the identity of the client is taken from the initiator principal of the outer context handle's GSS-API security context, or from the network address of the client...)
• "accept identity assertions ... only from trusted clients where IPsec policy protects this application's packet flows between the clients and this server"
• "accept only removals of groups from a user's group membership list as determined by the server"
• "never accept identity assertions"
• etcetera

Clients may mark an identity assertion as being critical, in which case the server MUST respond with an error if the server does not accept the identity assertion as-is.

The representation of users and groups is not given here, but is left to the application. It is expected that RPCSEC_GSSv3 identity assertions in the context of the NFSv4 application would consist of NFSv4 user and group representations as used on the wire in NFSv4 access control lists (ACLs).

2.1.2.6.  Server assertions

Servers MAY inform clients of assertions were granted by setting the 'granted_assertions' field of the RPCSEC_GSS3_CREATE reply.

The protocol provides a field ('server_assertions') for servers to make assertions about themselves. At this time there is not much use for this field, though servers MAY assert a single security label, indicating that all contents on the server is at that label. The client MUST, of course, either evaluate or ignore any server-side assertions.

2.1.3.  Context handle destruction

The RPCSEC_GSS3_DESTROY procedure is the same as for RPCSEC_GSSv1, but with the version 3 header.

2.1.4.  List request

The RPCSEC_GSS3_LIST call message consists of a single integer indicating what should be listed, and the reply consists of an error or the requested list.

The client may list DOIs, privilege names, or privilege group names.

The result is an opaque octet string containing a list of DOIs [encoding <TBD>] or a US-ASCII string containing a comma-separated list of privilege names or privilege group names.

2.1.5.  Extensibility

Assertion types may be added in the future by adding arms to the 'rpc_gss3_assertion_u' union. Every assertion has a 'critical' flag that can be used to indicate criticality.

New fields may be added through the 'extensions' typed hole. All such extensions have a 'critical' flag.

New message types may be added.

Clients receiving unknown critical server assertions MUST destroy the established RPCSEC_GSSv3 context handle. Servers receiving unknown critical client assertions or unknown RPCSEC_GSS_v3 extensions MUST return an error.

There is no IANA or other registry for RPCSEC_GSSv3 extensions. All extensions MUST be done by IETF Protocol Action.

3.  Privileges and identity representation for NFSv4

The representation of users and groups for use in identity assertions in RPCSEC_GSSv3 SHALL be the same as the user and group representations used by NFSv4 for access control list subjects on the wire, cast as an octet string ("opaque").

The following privileges are defined for use with the NFSv4 protocol:

file_chown

Generally allows the caller to change a file's owner regardless of who owns the file.

file_chown_self

Generally allows the caller to change the owner of a file it owns.

file_dac_execute

Generally allows the caller to read any file for execution.

file_dac_read

Generally allows the caller to read any file or directory.

file_dac_search

Generally allows the caller to search any directory.

file_dac_write

Generally allows the caller to write to any file (or create/delete/link objects in directories).

file_link_any

Generally allows the caller to create hardlinks to files not owned by the caller.

file_owner

Generally allows the caller to modify the access, modification and other timestamps of a filesystem object, as well as its permissions and ACL.

file_setid

Generally allows the caller to set the set-user-ID and set-group-ID bits of a file.

file_downgrade_sl

Generally allows the caller to downgrade the security label of a filesystem object.

file_update_sl

Generally allows the caller to upgrade the security label of a filesystem object.

[What about NFSv3? The representation of privs would be the same for v3 as for v4, though there'd be no privs for dealing with labels (file_downgrade_sl and file_update_sl). And the representation of users/groups would NFSv3's representation thereof. But should we bother to specify this? -Nico]

[Also, this is derived from Solaris' notion of privileges. We should look at how well this scheme relates to other operating systems as NFSv4 clients and servers. -Nico]

The contents of the 'basic' privilege set is not defined herein. Note that 'file_link_any' and 'file_chown_self' may be present in the server's notion of the basic privilege set.

The NFSv4-specific privileges may be limited by the server in ways not specified above. For example, the server may deny access for certain operations that would normally be granted given the granted assertion of a given privilege (e.g., "no one may write to files owned by such and such user"), or the server may require that all privileges be asserted (and granted, of course) in order to allow certain operations (e.g., "all privileges are required in order to write to files owned by such and such user, not just file_dac_write").

4.  Security Considerations

This entire document deals with security issues.

The RPCSEC_GSSv3 protocol allows for client-side assertions of data that is relevant to server-side authorization decisions. These assertions must be evaludated by the server in the context of whether the client and/or user are authenticated, whether compound authentication was used, whether the client is trusted, what ranges of assertions are allowed for the client and the user (separately or together), and any relevant server-side policy.

The security semantics of assertions carried by RPCSEC_GSSv3 are application protocol-specific.

RPCSEC_GSSv3 supports a notion of critical assertions (and extensions), but there's no need for peers to tell each other what assertions were granted, or what they were mapped to.

Note that RPSEC_GSSv3 is not a complete solution for labeling: it conveys the labels of actors, but not the labels of objects. RPC application protocols may require extending in order to carry object label information.

The RPCSEC_GSSv3 protocol also provides for a replacement of the old AUTH_SYS RPC authentication flavor. AUTH_SYS relies on "privileged port numbers" for "authentication," and was quite limited in what assertions it supported and incompatible with NFSv4 representations of identity. To replace AUTH_SYS with RPCSEC_GSSv3 simply use a GSS-API mechanism to authenticate the client (but not the user) and let the client assert the user's identity. This is more secure than AUTH_SYS in that at least the client can be strongly authenticated using GSS-API mechanisms, and it is more functional than AUTH_SYS in that identity representations are defined by the application layer.

It is possible that a GSS-API mechanism that does not provide any security services could be created so as to make it possible to replace AUTH_SYS with RPCSEC_GSSv3 while retaining the same privileged port semantics. Such a mechanism is out of scope for this document and would have its own security considerations.

There may be interactions with NFSv4's callback security scheme and NFSv4.1's GSS-API "SSV" mechanisms. Specifically, the NFSv4 callback scheme requires that the server initiate GSS-API security contexts, which does not work well in practice, and in the context of client-side processes running as the same user but with different privileges and security labels the NFSv4 callback security scheme seems particularly unlikely to work well. NFSv4.1 has the server use an existing, client-initiated RPCSEC_GSS context handle to protect server-initiated callback RPCs. The NFSv4.1 callback security scheme lacks all the problems of the NFSv4 scheme, however, it is important that the server pick an appropriate RPCSEC_GSS context handle to protect any callbacks. Specifically, it is important that the server use RPCSEC_GSS context handles which authenticate the client to protect any callbacks relating to server state initiated by RPCs protected by RPCSEC_GSSv3 contexts. [Add text about interaction with GSS-SSV...]

[Anything else?]

5.  IANA Considerations

There are no IANA considerations in this document.

TBDs in this document will be assigned by the ONC RPC registrart (which is not IANA, as I recall).

6. Normative References

Author's Address