Remote Procedure Call (RPC) Security Version 3

The original RPCSEC_GSS protocol provided for authentication of RPC clients and servers to each other using the Generic Security Services Application Programming Interface (GSS-API) . The second version of RPCSEC_GSS added support for channel bindings . Existing GSS-API mechanisms are insufficient for communicating certain aspects of authority to a server. The GSS-API and its mechanisms certainly could be extended to address this shortcoming. However, here it is addressed at the application layer, i.e. in RPCSEC_GSS. A major motivation for RPCSEC_GSSv3 is to add support for multi-level (labeled) security and server-side copy for NFSv4. Multi-Level Security (MLS) is a traditional model where subjects are given a security level (Unclassified, Secret, Top Secret, etc.) and objects are given security labels that mandate the access of the subject to the object (see and ). Labeled NFS (see Section 8 of ) uses the MLS subject label provided by the client via the RPCSEC_GSSv3 layer to enforce MAC access to objects owned by the server to enable server guest mode. RPCSEC_GSSv3 label assertions provide the means to achieve full mode labeled NFS. A traditional inter-server file copy entails the user gaining access to a file on the source, reading it, and writing it to a file on the destination. In secure NFSv4 inter-server server-side copy (see Section 3.4.1 of ), the user first secures access to both source and destination files, and then uses NFSv4.2 defined RPCSEC_GSSv3 structured privileges to authorize the destination to copy the file from the source on behalf of the user. Multi-principal assertions can be used to address shared cache poisoning attacks on the client cache by a user. As described in Section 7 of , multi-user machines with a single cache manager can fetch and cache data on a users' behalf, and re-display it for another user from the cache without re-fetching the data from the server. The initial data acquisition is authenticated by the first user's credentials, and if only that user's credentials are used, it may be possible for a malicious user or users to "poison" the cache for other users by introducing bogus data into the cache. Another use of the multi-principal assertion is the secure conveyance of privilege information for processes running with more (or even with less) privilege than the user normally would be accorded.

We therefore describe RPCSEC_GSS version 3 (RPCSEC_GSSv3). RPCSEC_GSSv3 is the same as RPCSEC_GSSv2, except that the following assertions of authority have been added. Security labels for multi-level security type enforcement, and other labeled security models (See ). Application-specific structured privileges. For an example see server-side copy . Multi-principal authentication of the client host and user to the server done by binding two RPCSEC_GSS handles. Simplified channel binding. Assertions of labels and privileges are evaluated by the server, which may then map the asserted values to other values, all according to server-side policy. See . An option for enumerating server supported label format specifiers (LFS) is provided. See for detail.

This document contains the External Data Representation (XDR) () definitions for the RPCSEC_GSSv3 protocol. The XDR description is provided in this document in a way that makes it simple for the reader to extract into ready to compile form. The reader can feed this document in the following shell script to produce the machine readable XDR description of RPCSEC_GSSv3: <CODE BEGINS>

#!/bin/sh grep "^ *///" | sed 's?^ */// ??' | sed 's?^ *///$??' <CODE ENDS> I.e. if the above script is stored in a file called "extract.sh", and this document is in a file called "spec.txt", then the reader can do: <CODE BEGINS>

rpcsec_gss_v3.x ]]> <CODE ENDS> The effect of the script is to remove leading white space from each line, plus a sentinel sequence of "///".

RPCSEC_GSS version 3 (RPCSEC_GSSv3) is very similar to RPCSEC_GSS version 2 (RPCSEC_GSSv2). The differences are the addition of support for assertions and channel bindings are supported via a different mechanism. The entire RPCSEC_GSSv3 protocol is not presented here. Only the differences between it and RPCSEC_GSSv2 are shown. The use of RPCSEC_GSSv3 is structured as follows: A client uses an existing RPCSEC_GSSv3 context handle established in the usual manner (See Section 5.2 ) to protect RPCSEC_GSSv3 exchanges, this will be termed the "parent" handle. The server issues a "child" RPCSEC_GSSv3 handle in the RPCSEC_GSS_CREATE response which uses the underlying GSS-API security context of the parent handle in all subsequent exchanges that uses the child handle. An RPCSEC_GSSv3 child handle MUST NOT be used as the parent handle in an RPCSEC_GSS3_CREATE control message.

The functionality of RPCSEC_GSSv2 is fully supported by RPCSEC_GSSv3 with the exception of the RPCSEC_GSS_BIND_CHANNEL operation which is deprecated (see ).

An initiator that supports version 3 of RPCSEC_GSS simply issues an RPCSEC_GSS request with the rgc_version field set to RPCSEC_GSS_VERS_3. If the target does not recognize RPCSEC_GSS_VERS_3, the target will return an RPC error per Section 5.1 of . The initiator MUST NOT attempt to use an RPCSEC_GSS handle returned by version 3 of a target with version 1 or version 2 of the same target. The initiator MUST NOT attempt to use an RPCSEC_GSS handle returned by version 1 or version 2 of a target with version 3 of the same target.

A new reply verifier is needed for RPCSEC_GSSv3 because of a situation that arises from the use of the same GSS context by child and parent handles. Because the RPCSEC_GSSv3 child handle uses the same GSS context as the parent handle, a child and parent RPCSEC_GSSv3 handle could have the same RPCSEC_GSS sequence numbers. Since the reply verifier of previous versions of RPCSEC_GSS computes a MIC on just the sequence number, this provides opportunities for man in the middle attacks. This issue is addressed in RPCSEC_GSS version 3 by computing the verifier using the exact same input as is used to compute the request verifier, except for the mtype is changed from CALL to REPLY. The new reply verifier computes a MIC over the following RPC reply header data:

unsigned int xid; msg_type mtype; /* set to REPLY */ unsigned int rpcvers; unsigned int prog; unsigned int vers; unsigned int proc; opaque_auth cred; /* captures the RPCSEC_GSS handle */

; /* context handle */ /// }; /// /// const RPCSEC_GSS_VERS_1 = 1; /// const RPCSEC_GSS_VERS_2 = 2; /// const RPCSEC_GSS_VERS_3 = 3; /* new */ /// /// union rpc_gss_cred_t switch (unsigned int rgc_version) { /// case RPCSEC_GSS_VERS_1: /// case RPCSEC_GSS_VERS_2: /// case RPCSEC_GSS_VERS_3: /* new */ /// rpc_gss_cred_vers_1_t rgc_cred_v1; /// }; /// ]]> <CODE ENDS> As seen above, the RPCSEC_GSSv3 credential has the same format as the RPCSEC_GSSv1 and RPCSEC_GSSv2 credential. Setting the rgc_version field to 3 indicates that the initiator and target support the new RPCSEC_GSSv3 control procedures.

RPCSEC_GSSv3 provides a channel binding assertion that replaces the RPCSEC_GSSv2 RPCSEC_GSS_BIND_CHANNEL operation. RPCSEC_GSS_BIND_CHANNEL MUST NOT be used on RPCSEC_GSS version 3 handles.

RPCSEC_GSSv3 requires the addition of several values to the auth_stat enumerated type definition. The use of these new auth_stat values is explained throughout this document.

enum auth_stat { ... /* * RPCSEC_GSSv3 errors */ RPCSEC_GSS_INNER_CREDPROBLEM = 15, RPCSEC_GSS_LABEL_PROBLEM = 16, RPCSEC_GSS_PRIVILEGE_PROBLEM = 17, RPCSEC_GSS_UNKNOWN_MESSAGE = 18 };

There are two new RPCSEC_GSSv3 control procedures: RPCSEC_GSS_CREATE, RPCSEC_GSS_LIST. The RPCSEC_GSS_CREATE procedure binds any combination of assertions: multi-principal authentication, labels, structured privileges, or channel bindings to a new RPCSEC_GSSv3 context returned in the rgss3_create_res rcr_handle field. The RPCSEC_GSS_LIST procedure queries the target for supported assertions. RPCSEC_GSS version 3 control messages are similar to the RPCSEC_GSS version 1 and version 2 RPCSEC_GSS_DESTROY control message (see section 5.4 ) in that the sequence number in the request must be valid, and the header checksum in the verifier must be valid. As in RPCSEC_GSS version 1 and version 2, the RPCSEC_GSSv version 3 control messages may contain call data following the verifier in the body of the NULLPROC procedure. In other words, they look a lot like an RPCSEC_GSS data message with the header procedure set to NULLPROC. The client MUST use one of the following security services to protect the RPCSEC_GSS_CREATE or RPCSEC_GSS_LIST control message: rpc_gss_svc_integrity rpc_gss_svc_privacy Specifically the client MUST NOT use rpc_gss_svc_none. RPCSEC_GSS_LIST can also use rpc_gss_svc_channel_prot (see RPCSEC_GSSv2) if the request is sent using an RPCSEC_GSSv3 child handle with channel bindings enabled as described in .

; /// }; /// /// struct rgss3_create_res { /// opaque rcr_handle<>; /// rgss3_gss_mp_auth *rcr_mp_auth; /// rgss3_chan_binding *rcr_chan_bind_mic; /// rgss3_assertion_u rcr_assertions<>; /// }; /// /// enum rgss3_assertion_type { /// LABEL = 0, /// PRIVS = 1 /// }; /// /// union rgss3_assertion_u /// switch (rgss3_assertion_type atype) { /// case LABEL: /// rgss3_label rau_label; /// case PRIVS: /// rgss3_privs rau_privs; /// default: /// opaque rau_ext<>; /// }; /// ]]> <CODE ENDS> The call data for an RPCSEC_GSS_CREATE request consists of an rgss3_create_args which binds one or more items of several kinds to the returned rcr_handle RPCSEC_GSSv3 context handle called the "child" handle: Multi-principal authentication: another RPCSEC_GSS context handle A channel binding Authorization assertions: labels and or privileges The reply to this message consists of either an error or an rgss3_create_res structure. As noted in and successful rgss3_assertions are enumerated in rcr_assertions, and are REQUIRED be enumerated in the same order as they appeared in the rca_assertions argument. Upon successful RPCSEC_GSS_CREATE, both the client and the server SHOULD associate the resultant child rcr_handle context handle with the parent context handle in their GSS context caches so as to be able to reference the parent context given the child context handle. RPCSEC_GSSv3 child handles MUST be destroyed upon the destruction of the associated parent handle. Server implementation and policy MAY result in labels, privileges, and identities being mapped to concepts and values that are local to the server. Server policies should take into account the identity of the client and/or user as authenticated via the GSS-API.

; /* inner handle */ /// opaque rgmp_rpcheader_mic<>; /// }; /// ]]> <CODE ENDS> RPCSEC_GSSv3 clients MAY assert a multi-principal authentication of the RPC client host principal and a user principal. This feature is needed, for example, when an RPC client host wishes to use authority assertions that the server may only grant if a user and an RPC client host 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), or to an RPC client host acting alone (e.g. when an RPC client host is acting on behalf of a user) but may grant requested authority to an RPC client host acting on behalf of a user if the server identifies the user and trusts the RPC client host. It is assumed that an unprivileged user-space program would not have access to RPC client host credentials needed to establish a GSS-API security context authenticating the RPC client host to the server, therefore an unprivileged user-space program could not create an RPCSEC_GSSv3 RPCSEC_GSS_CREATE message that successfully binds an RPC client host and a user security context. In addition to the parent handle, the multi-principal authentication call data has an RPCSEC_GSS version 3 handle referenced via the rgmp_handle field termed the "inner" handle. Clients using RPCSEC_GSSv3 multi-principal authentication MUST use an RPCSEC_GSSv3 context handle that corresponds to a GSS-API security context that authenticates the RPC client host for the parent handle. The inner context handle it SHOULD use a context handle to authenticate a user. The reverse (parent handle authenticates user, inner authenticates an RPC client host) MUST NOT be used. Other multi-principal parent and inner context handle uses might eventually make sense, but would need to be introduced in a new revision of the RPCSEC_GSS protocol. The child context handle returned by a successful multi-principal assertion binds the inner RPCSEC_GSSv3 context handle to the parent RPCSEC_GSS context and 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. Multi-principal binding is done by including an assertion of type rgss3_gss_mp_auth in the RPCSEC_GSS_CREATE rgss3_create_args call data. The inner context handle is placed in the rgmp_handle field. A MIC of the RPC call header up to and including the credential is computed using the GSS-API security context associated with the inner context handle is placed in rgmp_rpcheader_mic field. The target verifies the multi-principal authentication by first confirming that the parent context used is an RPC client host context, and then verifies the rgmp_rpcheader_mic using the GSS-API security context associated with the rgmp_handle field. On a successful verification, the rgss3_gss_mp_auth field in the rgss3_create_res reply MUST be filled in with the inner RPCSEC_GSSv3 context handle as the rgmp_handle, and a MIC computed over the RPC reply header (see section ) using the GSS-API security context associated with the inner handle. On failure, the rgss3_gss_mp_auth field is not sent (rgss3_gss_mp_auth is an optional field). A MSG_DENIED reply to the RPCSEC_GSS_CREATE call is formulated as usual. As described in Section 5.3.3.3 of the server maintains a list of contexts for the clients that are currently in session with it. When a client request comes in, there may not be a context corresponding to its handle. When this occurs on an RPCSEC_GSS3_CREATE request processing of the parent handle, the server rejects the request with a reply status of MSG_DENIED with the reject_stat of AUTH_ERROR and with an auth_stat value of RPCSEC_GSS_CREDPROBLEM. A new value, RPCSEC_GSS_INNER_CREDPROBLEM, has been added to the auth_stat type. With a multi-pricipal authorization request, the server must also have a context corresponding to the inner context handle. When the server does not have a context handle corresponding to the inner context handle of a multi-pricipal authorization request, the server sends a reply status of MSG_DENIED with the reject_stat of AUTH_ERROR and with an auth_stat value of RPCSEC_GSS_INNER_CREDPROBLEM. When processing the multi-principal authentication request, if the GSS_VerifyMIC() call on the rgmp_rpcheader_mic fails to return GSS_S_COMPLETE, the server sends a reply status of MSG_DENIED with the reject_stat of AUTH_ERROR and with an auth_stat value of RPCSEC_GSS_INNER_CREDPROBLEM.

; /// ]]> <CODE ENDS> RPCSEC_GSSv3 provides a different way to do channel binding than RPCSEC_GSSv2. Specifically: RPCSEC_GSSv3 builds on RPCSEC_GSSv1 by reusing existing, established context handles rather than providing a different RPC security flavor for establishing context handles, channel bindings data are not hashed because there is now general agreement 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 , then the client calls GSS_GetMIC() to get a MIC of resulting octet string, using the parent RPCSEC_GSSv3 context handle's GSS-API security context. The MIC is then placed in the rca_chan_bind_mic field of RPCSEC_GSS_CREATE arguments (rgss3_create_args). If the rca_chan_bind_mic field of the arguments of a RPCSEC_GSS_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 rcr_chan_bind_mic field of the RPCSEC_GSS_CREATE rgss3_create_res 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 rgss3_chan_binding value in rgss3_create_res (rgss3_chan_binding is an optional field). The client MUST verify the result's rcr_chan_bind_mic value 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_GSS_DESTROY. If the client requested channel binding but the server did not include an rcr_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. If the client considers channel binding critical, it MUST call RPCSEC_GSS_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 child 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.

; /// }; /// /// struct rgss3_lfs { /// unsigned int rlf_lfs_id; /// unsigned int rlf_pi_id; /// }; /// ]]> <CODE ENDS> The client discovers which labels the server supports via the RPCSEC_GSS_LIST control message. Asserting a server supported label via RPCSEC_GSS_CREATE enables server guest mode labels. Full mode is enabled when an RPCSEC_GSS_CREATE label assertion is combined with asserting the same label with the NFSv4.2 sec_label attribute. Label encoding is specified to mirror the NFSv4.2 sec_label attribute described in Section 12.2.2 of . The label format specifier (LFS) is an identifier used by the client to establish the syntactic format of the security label and the semantic meaning of its components. The policy identifier (PI) is an optional part of the definition of an LFS which allows for clients and server to identify specific security policies. The opaque label field of rgss3_label is dependent on the MAC model to interpret and enforce. 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_GSS_CREATE request. RPCSEC_GSSv3 clients MAY assert a server security label in some LSF by binding a label assertion to the RPCSEC_GSSv3 context handle. This is done by including an assertion of type rgss3_label in the RPCSEC_GSS_CREATE rgss3_create_args rca_assertions call data. Servers that support labeling in the requested LFS MAY map the requested label to different label as a result of server-side policy evaluation. The labels that are accepted by the target and bound to the RPCSEC_GSSv3 context MUST be enumerated in the rcr_assertions field of the rgss3_create_res RPCSEC_GSS_CREATE reply. Servers that do not support labeling or that do not support the requested LFS reject the label assertion with a reply status of MSG_DENIED, a reject_status of AUTH_ERROR, and an auth_stat of RPCSEC_GSS_LABEL_PROBLEM.

; /* human readable */ /// opaque rp_privilege<>; /// }; ]]> <CODE ENDS> A structured privilege is an RPC application defined privilege. RPCSEC_GSSv3 clients MAY assert a structured privilege by binding the privilege to the RPCSEC_GSSv3 context handle. This is done by including an assertion of type rgss3_privs in the RPCSEC_GSS_CREATE rgss3_create_args rca_assertions call data. Encoding, server verification and any policies for structured privileges are described by the RPC application definition. A successful structured privilege assertion MUST be enumerated in the rcr_assertions field of the rgss3_create_res RPCSEC_GSS_CREATE reply. If a server receives a structured privilege assertion that it does not recognize the assertion is rejected with a reply status of MSG_DENIED, a reject_status of AUTH_ERROR, and an auth_stat of RPCSEC_GSS_UNKNOWN_MESSAGE. If a server receives a structured privilege assertion that it fails to verify according to the requirements of the RPC application defined behavior, the assertion is rejected with a reply status of MSG_DENIED, a reject_status of AUTH_ERROR, and an auth_stat of RPCSEC_GSS_PRIVILEGE_PROBLEM. Section 3.4.1.2. "Inter-Server Copy with RPCSEC_GSSv3" of shows an example of structured privilege definition and use.

; /// }; /// /// union rgss3_list_item_u /// switch (rgss3_list_item itype) { /// case LABEL: /// rgss3_label rli_labels<>; /// case PRIVS: /// rgss3_privs rli_privs<>; /// }; /// /// typedef rgss3_list_item_u rgss3_list_res<>; /// ]]> <CODE ENDS> The call data for an RPCSEC_GSS_LIST request consists of a list of integers (rla_list_what) indicating what assertions to be listed, and the reply consists of an error or the requested list. The result of requesting a list of rgss3_list_item LABEL is a list of LFSs supported by the server. The client can then use the LFS list to assert labels via the RPCSEC_GSS_CREATE label assertions. See .

Assertion types may be added in the future by adding arms to the 'rgss3_assertion_u' union. Other assertion types are described elsewhere and include: 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.