Internet Engineering Task Force (IETF)                          B. Kaduk
Request for Comments: 7546                                           MIT
Category: Informational                                         May 2015
ISSN: 2070-1721

    Structure of the Generic Security Service (GSS) Negotiation Loop

Abstract

   This document specifies the generic structure of the negotiation loop
   to establish a Generic Security Service (GSS) security context
   between initiator and acceptor.  The control flow of the loop is
   indicated for both parties, including error conditions, and
   indications are given for where application-specific behavior must be
   specified.

Status of This Memo

   This document is not an Internet Standards Track specification; it is
   published for informational purposes.

   This document is a product of the Internet Engineering Task Force
   (IETF).  It represents the consensus of the IETF community.  It has
   received public review and has been approved for publication by the
   Internet Engineering Steering Group (IESG).  Not all documents
   approved by the IESG are a candidate for any level of Internet
   Standard; see Section 2 of RFC 5741.

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at
   http://www.rfc-editor.org/info/rfc7546.

Copyright Notice

   Copyright (c) 2015 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
   (http://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.  Code Components extracted from this document must
   include Simplified BSD License text as described in Section 4.e of
   the Trust Legal Provisions and are provided without warranty as
   described in the Simplified BSD License.

Table of Contents

   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   2
   2.  Application Protocol Requirements . . . . . . . . . . . . . .   3
   3.  Loop Structure  . . . . . . . . . . . . . . . . . . . . . . .   4
     3.1.  Anonymous Initiators  . . . . . . . . . . . . . . . . . .   4
     3.2.  GSS_Init_sec_context  . . . . . . . . . . . . . . . . . .   5
     3.3.  Sending from Initiator to Acceptor  . . . . . . . . . . .   5   6
     3.4.  Acceptor Sanity Checking  . . . . . . . . . . . . . . . .   6
     3.5.  GSS_Accept_sec_context  . . . . . . . . . . . . . . . . .   7
     3.6.  Sending from Acceptor to Initiator  . . . . . . . . . . .   8
     3.7.  Initiator Input Validation  . . . . . . . . . . . . . . .   8
     3.8.  Continue the Loop . . . . . . . . . . . . . . . . . . . .   9
   4.  After Security Context Negotiation  . . . . . . . . . . . . .   9
     4.1.  Authorization Checks  . . . . . . . . . . . . . . . . . .  10
     4.2.  Using Partially Complete Security Contexts  . . . . . . .  10
     4.3.  Additional Context Tokens . . . . . . . . . . . . . . . .  10
   5.  Sample Code . . . . . . . . . . . . . . . . . . . . . . . . .  11  12
     5.1.  GSS Application Sample Code . . . . . . . . . . . . . . .  12
   6.  Security Considerations . . . . . . . . . . . . . . . . . . .  18
   7.  References  . . . . . . . . . . . . . . . . . . . . . . . . .  19
     7.1.  Normative References  . . . . . . . . . . . . . . . . . .  19
     7.2.  Informative References  . . . . . . . . . . . . . . . . .  20
   Acknowledgements  . . . . . . . . . . . . . . . . . . . . . . . .  21  22
   Author's Address  . . . . . . . . . . . . . . . . . . . . . . . .  21  22

1.  Introduction

   "Generic Security Service Application Program Interface Version 2,
   Update 1" [RFC2743] provides a generic interface for security
   services in the form of an abstraction layer over the underlying
   security mechanisms that an application may use.  A GSS initiator and
   acceptor exchange messages, called "tokens", until a security context
   is established.  Such a security context allows for each party to
   authenticate the other, the passing of confidential and/or integrity-
   protected messages between the initiator and acceptor, the generation
   of identical pseudorandom bit strings by both participants [RFC4401],
   and more.

   During context establishment, security context tokens are exchanged
   synchronously, one at a time; the initiator sends the first context
   token.  The number of tokens that must be exchanged between the
   initiator and acceptor in order to establish the security context is
   dependent on the underlying mechanism as well as the desired
   properties of the security context and is, in general, not known to
   the application.  Accordingly, the application's control flow must
   include a loop within which GSS security context tokens are
   exchanged, which
   exchanged; the loop terminates upon successful establishment of a
   security context or an error condition.  The GSS-API, together with
   its security mechanisms, specifies the format and encoding of the
   context tokens themselves, but the application protocol must specify
   the necessary framing for the application to determine what octet
   strings constitute GSS security context tokens and pass them into the
   GSS-API implementation as appropriate.

   The GSS-API C-bindings [RFC2744] provide some example code for such a
   negotiation loop, but this code does not specify the application's
   behavior on unexpected or error conditions.  As such, individual
   application protocol specifications have had to specify the structure
   of their GSS negotiation loops, including error handling, on a per-
   protocol basis (see [RFC4462], [RFC3645], [RFC5801], [RFC4752], and
   [RFC2203]).  This represents a substantial duplication of effort, and
   the various specifications go into different levels of detail and
   describe different possible error conditions.  Therefore, it is
   preferable to have the structure of the GSS negotiation loop,
   including error conditions and token passing, described in a single
   specification that can then be referred to from other documents in
   lieu of repeating the structure of the loop each time.  This document
   fills that role.

   The necessary requirements for correctly performing a GSS negotiation
   loop are essentially all included in [RFC2743], but they are
   scattered in many different places.  This document brings all the
   requirements together into one place for the convenience of
   implementors, even though the normative requirements remain in
   [RFC2743].  In a few places, this document notes additional behavior
   that is useful for applications but is not mandated by [RFC2743].

2.  Application Protocol Requirements

   Part of the purpose of this document is to guide the development of
   new application protocols using the GSS-API, as well as the
   development of new application software using such protocols.  The
   following list consists of features that are necessary or useful in
   such an application protocol:

   o  A  Protocols require a way to frame and identify security context
      negotiation tokens in during the GSS negotiation loop.

   o  Error tokens should generally also get special framing, as the
      recipient may have no other way to distinguish unexpected error
      context tokens from per-message tokens.

   o  If an error token has no special framing, a way to indicate error
      status  Protocols may benefit from one peer to the other is necessary, a generic means of indicating an error,
      possibly accompanied by a descriptive string. string, to be used if error
      tokens are not available or not usable due to constraints of the
      application protocol.

   o  A protocol may use the negotiated GSS security context for per-
      message operations; in such cases, the protocol will need a way to
      frame and identify those per-message tokens and the nature of
      their contents.  For example, a protocol message may be
      accompanied by the output of GSS_GetMIC() over that message; the
      protocol must identify the location and size of that Message
      Identity Code (MIC) token and indicate that it is a MIC token and
      to what cleartext it corresponds.

   o  Applications are responsible for authorization of the
      authenticated peer principal names that are supplied by the GSS-
      API.  Such names are mechanism specific and may come from a
      different portion of a federated identity scheme.  Application
      protocols may need to supply additional information to support the
      authorization of access to a given resource, such as the Secure
      Shell version 2 (SSHv2) "username" parameter.

3.  Loop Structure

   The loop is begun by the appropriately named initiator, which calls
   GSS_Init_sec_context() with an empty (zero-length) input_token and a
   fixed set of input flags containing the desired attributes for the
   security context.  The initiator should not change any of the input
   parameters to GSS_Init_sec_context() between calls to it during the
   loop, with the exception of the input_token parameter, which will
   contain a message from the acceptor after the initial call, and the
   input_context_handle, which must be the result returned in the
   output_context_handle of the previous call to GSS_Init_sec_context()
   (GSS_C_NO_CONTEXT for the first call).  (In the C bindings, there is
   only a single read/modify context_handle argument, so the same
   variable should be passed for each call in the loop.)  RFC 2743 only
   requires that the claimant_cred_handle argument remain constant over
   all calls in the loop, but the other non-excepted arguments should
   also remain fixed for reliable operation.

   The following subsections will describe the various steps of the
   loop, without special consideration to whether a call to
   GSS_Init_sec_context() or GSS_Accept_sec_context() is the first such
   call in the loop.

3.1.  Anonymous Initiators

   If the initiator is requesting anonymity by setting the anon_req_flag
   input to GSS_Init_sec_context(), then on non-error returns from
   GSS_Init_sec_context() (that is, when the major status is
   GSS_S_COMPLETE or GSS_S_CONTINUE_NEEDED) the initiator must verify
   that the output value of anon_state from GSS_Init_sec_context() is
   true before sending the security context token to the acceptor.
   Failing to perform this check could cause the initiator to lose
   anonymity.

3.2.  GSS_Init_sec_context

   The initiator calls GSS_Init_sec_context() using the
   input_context_handle for the current security context being
   established and its fixed set of input parameters and the input_token
   received from the acceptor (if this is not the first iteration of the
   loop).  The presence or absence of a nonempty output_token and the
   value of the major status code are the indicators for how to proceed:

   o  If the major status code is GSS_S_COMPLETE and the output_token is
      empty, then the context negotiation is fully complete and ready
      for use by the initiator with no further actions.

   o  If the major status code is GSS_S_COMPLETE and the output_token is
      nonempty, then the initiator's portion of the security context
      negotiation is complete but the acceptor's is not.  The initiator
      must send the output_token to the acceptor so that the acceptor
      can establish its half of the security context.

   o  If the major status code is GSS_S_CONTINUE_NEEDED and the
      output_token is nonempty, the context negotiation is incomplete.
      The initiator must send the output_token to the acceptor and await
      another input_token from the acceptor.

   o  If the major status code is GSS_S_CONTINUE_NEEDED and the
      output_token is empty, the mechanism has produced an output that
      is not compliant with [RFC2743].  However, there are some known
      implementations of certain mechanisms such as NT LAN Manager
      Security Support Provider [NTLMSSP] that do produce empty context
      negotiation tokens.  For maximum interoperability, applications
      should be prepared to accept such tokens and should transmit them
      to the acceptor if they are generated.

   o  If the major status code is any other value, the context
      negotiation has failed.  If the output_token is nonempty, it is an
      error token and the initiator should send it to the acceptor.  If
      the output_token is empty, then the initiator should indicate the
      failure to the acceptor if an appropriate application-protocol
      channel to do so is available.

3.3.  Sending from Initiator to Acceptor

   The establishment of a GSS security context between initiator and
   acceptor requires some communication channel by which to exchange the
   context negotiation tokens.  The nature of this channel is not
   specified by the GSS specification -- it could be a dedicated TCP
   channel, a UDP-based Remote Procedure Call (RPC) protocol, or any
   other sort of channel.  In many cases, the channel will be
   multiplexed with non-GSS application data; the application protocol
   must always provide some means by which the GSS context tokens can be
   identified (e.g., length and start location) and passed through to
   the mechanism accordingly.  The application protocol may also include
   a facility for indicating errors from one party to the other, which
   can be used to convey errors resulting from GSS-API calls when
   appropriate (such as when no error token was generated by the GSS-API
   implementation).  Note that GSS major and minor status codes are
   specified by language bindings, not the abstract API; sending a major
   status code and optionally the display form of the two error codes
   may be the best that can be done in this case.

   However, even the presence of a communication channel does not
   necessarily indicate that it is appropriate for the initiator to
   indicate such errors.  For example, if the acceptor is a stateless or
   near-stateless UDP server, there is probably no need for the
   initiator to explicitly indicate its failure to the acceptor.
   Conditions such as this can be treated in individual application
   protocol specifications.

   If a regular security context output_token is produced by the call to
   GSS_Init_sec_context(), the initiator must transmit this token to the
   acceptor over the application's communication channel.  If the call
   to GSS_Init_sec_context() returns an error token as output_token, it
   is recommended that the initiator transmit this token to the acceptor
   over the application's communication channel.

3.4.  Acceptor Sanity Checking

   The acceptor's half of the negotiation loop is triggered by the
   receipt of a context token from the initiator.  Before calling
   GSS_Accept_sec_context(), the acceptor may find it useful to perform
   some sanity checks on the state of the negotiation loop.

   If the acceptor receives a context token but was not expecting such a
   token (for example, if the acceptor's previous call to
   GSS_Accept_sec_context() returned GSS_S_COMPLETE), this is probably
   an error condition indicating that the initiator's state is invalid.
   See Section 4.3 for some exceptional cases.  It is likely appropriate
   for the acceptor to report this error condition to the initiator via
   the application's communication channel.

   If the acceptor is expecting a context token (e.g., if the previous
   call to GSS_Accept_sec_context() returned GSS_S_CONTINUE_NEEDED) but
   does not receive such a token within a reasonable amount of time
   after transmitting the previous output_token to the initiator, the
   acceptor should assume that the initiator's state is invalid
   (timeout) and fail the GSS negotiation.  Again, it is likely
   appropriate for the acceptor to report this error condition to the
   initiator via the application's communication channel.

3.5.  GSS_Accept_sec_context

   The GSS acceptor responds to the actions of an initiator; as such,
   there should always be a nonempty input_token to calls to
   GSS_Accept_sec_context().  The input_context_handle parameter will
   always be given as the output_context_handle from the previous call
   to GSS_Accept_sec_context() in a given negotiation loop, or
   GSS_C_NO_CONTEXT on the first call, but the acceptor_cred_handle and
   chan_bindings arguments should remain fixed over the course of a
   given GSS negotiation loop.  [RFC2743] only requires that the
   acceptor_cred_handle remain fixed throughout the loop, but the
   chan_bindings argument should also remain fixed for reliable
   operation.

   The GSS acceptor calls GSS_Accept_sec_context(), using the
   input_context_handle for the current security context being
   established and the input_token received from the initiator.  The
   presence or absence of a nonempty output_token and the value of the
   major status code are the indicators for how to proceed:

   o  If the major status code is GSS_S_COMPLETE and the output_token is
      empty, then the context negotiation is fully complete and ready
      for use by the acceptor with no further actions.

   o  If the major status code is GSS_S_COMPLETE and the output_token is
      nonempty, then the acceptor's portion of the security context
      negotiation is complete but the initiator's is not.  The acceptor
      must send the output_token to the initiator so that the initiator
      can establish its half of the security context.

   o  If the major status code is GSS_S_CONTINUE_NEEDED and the
      output_token is nonempty, the context negotiation is incomplete.
      The acceptor must send the output_token to the initiator and await
      another input_token from the initiator.

   o  If the major status code is GSS_S_CONTINUE_NEEDED and the
      output_token is empty, the mechanism has produced an output that
      is not compliant with [RFC2743].  However, there are some known
      implementations of certain mechanisms such as NTLMSSP [NTLMSSP]
      that do produce empty context negotiation tokens.  For maximum
      interoperability, applications should be prepared to accept such
      tokens and should transmit them to the initiator if they are
      generated.

   o  If the major status code is any other value, the context
      negotiation has failed.  If the output_token is nonempty, it is an
      error token and the acceptor should send it to the initiator.  If
      the output_token is empty, then the acceptor should indicate the
      failure to the initiator if an appropriate application-protocol
      channel to do so is available.

3.6.  Sending from Acceptor to Initiator

   The mechanism for sending the context token from acceptor to
   initiator will depend on the nature of the communication channel
   between the two parties.  For a synchronous bidirectional channel, it
   can be just another piece of data sent over the link, but for a
   stateless UDP RPC acceptor, the token will probably end up being sent
   as an RPC output parameter.  Application protocol specifications will
   need to specify the nature of this behavior.

   If the application protocol has the initiator driving the
   application's control flow, it is particularly helpful for the
   acceptor to indicate a failure to the initiator, as mentioned in some
   of the above cases conditional on "an appropriate application-
   protocol channel to do so".

   If a regular security context output_token is produced by the call to
   GSS_Accept_sec_context(), the acceptor must transmit this token to
   the initiator over the application's communication channel.  If the
   call to GSS_Accept_sec_context() returns an error token as
   output_token, it is recommended that the acceptor transmit this token
   to the initiator over the application's communication channel.

3.7.  Initiator Input Validation

   The initiator's half of the negotiation loop is triggered (after the
   first call) by receipt of a context token from the acceptor.  Before
   calling GSS_Init_sec_context(), the initiator may find it useful to
   perform some sanity checks on the state of the negotiation loop.

   If the initiator receives a context token but was not expecting such
   a token (for example, if the initiator's previous call to
   GSS_Init_sec_context() returned GSS_S_COMPLETE), this is probably an
   error condition indicating that the acceptor's state is invalid.  See
   Section 4.3 for some exceptional cases.  It may be appropriate for
   the initiator to report this error condition to the acceptor via the
   application's communication channel.

   If the initiator is expecting a context token (that is, the previous
   call to GSS_Init_sec_context() returned GSS_S_CONTINUE_NEEDED) but
   does not receive such a token within a reasonable amount of time
   after transmitting the previous output_token to the acceptor, the
   initiator should assume that the acceptor's state is invalid and fail
   the GSS negotiation.  Again, it may be appropriate for the initiator
   to report this error condition to the acceptor via the application's
   communication channel.

3.8.  Continue the Loop

   If the loop is in neither a success nor a failure condition, then the
   loop must continue.  Control flow returns to Section 3.2.

4.  After Security Context Negotiation

   Once a party has completed its half of the security context and
   fulfilled its obligations to the other party, the context is
   complete, but it is not necessarily ready and appropriate for use.
   In particular, the security context flags may not be appropriate for
   the given application's use.  In some cases, the context may be ready
   for use before the negotiation is complete, see Section 4.2.

   The initiator specifies, as part of its fixed set of inputs to
   GSS_Init_sec_context(), values for all defined request flag booleans,
   among them: deleg_req_flag, mutual_req_flag, replay_det_req_flag,
   sequence_req_flag, conf_req_flag, and integ_req_flag.  Upon
   completion of the security context negotiation, the initiator must
   verify that the values of the deleg_state, mutual_state,
   replay_det_state, sequence_state, conf_avail, and integ_avail (and
   any additional flags added by extensions) from the last call to
   GSS_Init_sec_context() correspond to the requested flags.  If a flag
   was requested but is not available and that feature is necessary for
   the application protocol, the initiator must destroy the security
   context and not use the security context for application traffic.

   Application protocol specifications citing this document should
   indicate which context flags are required for their application
   protocol.

   The acceptor receives as output the following booleans: deleg_state,
   mutual_state, replay_det_state, sequence_state, anon_state,
   trans_state, conf_avail, and integ_avail, and any additional flags
   added by extensions to the GSS-API.  The acceptor must verify that
   any flags necessary for the application protocol are set.  If a
   necessary flag is not set, the acceptor must destroy the security
   context and not use the security context for application traffic.

4.1.  Authorization Checks

   The acceptor receives as one of the outputs of
   GSS_Accept_sec_context() the name of the initiator that has
   authenticated during the security context negotiation.  Applications
   need to implement authorization checks on this received name
   ('client_name' in the sample code) before providing access to
   restricted resources.  In particular, security context negotiation
   can be successful when the client is anonymous or is from a different
   identity realm than the acceptor, depending on the details of the
   mechanism used by the GSS-API to establish the security context.
   Acceptor applications can check which target name was used by the
   initiator, but the details are out of scope for this document.  See
   Sections 2.2.6 and 1.1.5 of [RFC2743].  Additional information can be
   available in GSS-API Naming Extensions [RFC6680].

4.2.  Using Partially Complete Security Contexts

   For mechanism/flag combinations that require multiple token
   exchanges, the GSS-API specification [RFC2743] provides a
   prot_ready_state output value from GSS_Init_sec_context() and
   GSS_Accept_sec_context(), which indicates when per-message operations
   are available.  However, many mechanism implementations do not
   provide this functionality and the analysis of the security
   consequences of its use is rather complicated, so it is not expected
   to be useful in most application protocols.

   In particular, mutual authentication, replay protection, and other
   services (if requested) are services that will be active when
   GSS_S_COMPLETE is returned but that are not necessarily active before
   the security context is fully established.

4.3.  Additional Context Tokens

   Under some conditions, a context token will be received by a party to
   a security context negotiation after that party has completed the
   negotiation (i.e., after GSS_Init_sec_context() or
   GSS_Accept_sec_context() has returned GSS_S_COMPLETE).  Such tokens
   must be passed to GSS_Process_context_token() for processing.  It may
   not always be necessary for a mechanism implementation to generate an
   error token on the initiator side or for an initiator application to
   transmit that token to the acceptor; such decisions are out of scope
   for this document.  Both peers should always be prepared to process
   such tokens and application protocols should provide means by which
   they can be transmitted.

   Such tokens can be security context deletion tokens, emitted when the
   remote party called GSS_Delete_sec_context() with a non-null
   output_context_token parameter, or error tokens, emitted when the
   remote party experiences an error processing the last token in a
   security context negotiation exchange.  Errors experienced when
   processing tokens earlier in the negotiation would be transmitted as
   normal security context tokens and processed by
   GSS_Init_sec_context() or GSS_Accept_sec_context(), as appropriate.
   With the GSS-API version 2, it is not recommended to use security
   context deletion tokens, so error tokens are expected to be the most
   common form of additional context token for new application
   protocols.

   GSS_Process_context_token() may indicate an error in its major_status
   field if an error is encountered locally during token processing, processing or
   it may
   to indicate that an error was encountered on the peer and conveyed in
   an error token.  See [Err4151] of [RFC2743].  Regardless of the
   major_status output of GSS_Process_context_token(),
   GSS_Inquire_context() should be used after processing the extra token
   token, to query the status of the security context and whether it can
   supply the features necessary for the application protocol.

   At present, all tokens that should be handled by
   GSS_Process_context_token() will lead to the security context being
   effectively unusable.  Future extensions to the GSS-API may allow for
   applications to continue to function after a call to
   GSS_Process_context_token(), and it is expected that the outputs of
   GSS_Inquire_context() will indicate whether it is safe to do so.
   However, since there are no such extensions at present (error tokens
   and deletion tokens both result in the security context being
   essentially unusable), there is no guidance to give to applications
   regarding this possibility at this time.

   Even if GSS_Process_context_token() processes an error or deletion
   token that renders the context essentially unusable, the resources
   associated with the context must eventually be freed with a call to
   GSS_Delete_sec_context(), just as would be needed if
   GSS_Init_sec_context() or GSS_Accept_sec_context() had returned an
   error while processing an input context token and the
   input_context_handle was not GSS_C_NO_CONTEXT.  RFC 2743 has some
   text that is slightly ambiguous in this regard, but the best practice
   is to always call GSS_Delete_sec_context().

5.  Sample Code

   This section gives sample code for the GSS negotiation loop, both for
   a regular application and for an application where the initiator
   wishes to remain anonymous.  Since the code for the two cases is very
   similar, the anonymous-specific additions are wrapped in a
   conditional check; that check and the conditional code may be ignored
   if anonymous processing is not needed.

   Since the communication channel between the initiator and acceptor is
   a matter for individual application protocols, it is inherently
   unspecified at the GSS-API level, which can lead to examples that are
   less satisfying than may be desired.  For example, the sample code in
   [RFC2744] uses an unspecified send_token_to_peer() routine.  Fully
   correct and general code to frame and transmit tokens requires a
   substantial amount of error checking and would detract from the core
   purpose of this document, so we only present the function signature
   for one example of what such functions might be and leave some
   comments in the otherwise empty function bodies.

   This sample code is written in C, using the GSS-API C-bindings
   [RFC2744].  It uses the macro GSS_ERROR() to help unpack the various
   sorts of information that can be stored in the major status field;
   supplementary information does not necessarily indicate an error.
   Applications written in other languages will need to exercise care
   that checks against the major status value are written correctly.

   This sample code should be compilable as a standalone program, linked
   against a GSS-API library.  In addition to supplying implementations
   for the token transmission/receipt routines, in order for the program
   to successfully run when linked against most GSS-API libraries, the
   initiator will need to specify an explicit target name for the
   acceptor, which must match the credentials available to the acceptor.
   A skeleton for how this may be done is provided, using a dummy name.

   This sample code assumes use of v2 of the GSS-API.  Applications
   wishing to remain compatible with v1 of the GSS-API may need to
   perform additional checks in some locations.

5.1.  GSS Application Sample Code

#include <unistd.h>
#include <err.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <gssapi/gssapi.h>

/*
 * This helper is used only on buffers that we allocate ourselves (e.g.,
 * from receive_token()).  Buffers allocated by GSS routines must use
 * gss_release_buffer().
 */
static void
release_buffer(gss_buffer_t buf)
{
    free(buf->value);
    buf->value = NULL;
    buf->length = 0;
}

/*
 * Helper to send a token on the specified fd. file descriptor.
 *
 * If errors are encountered, this routine must not directly cause
 * termination of the process because compliant GSS applications
 * must release resources allocated by the GSS library before
 * exiting.
 *
 * Returns 0 on success, nonzero on failure.
 */
static int
send_token(int fd, gss_buffer_t token)
{
    /*
     * Supply token framing and transmission code here.
     *
     * It is advisable for the application protocol to specify the
     * length of the token being transmitted unless the underlying
     * transit does so implicitly.
     *
     * In addition to checking for error returns from whichever
     * syscall(s) are used to send data, applications should have
     * a loop to handle EINTR returns.
     */
    return 1;
}

/*
 * Helper to receive a token on the specified fd. file descriptor.
 *
 * If errors are encountered, this routine must not directly cause
 * termination of the process because compliant GSS applications
 * must release resources allocated by the GSS library before
 * exiting.
 *
 * Returns 0 on success, nonzero on failure.
 */
static int
receive_token(int fd, gss_buffer_t token)
{
    /*
     * Supply token framing and transmission code here.
     *
     * In addition to checking for error returns from whichever
     * syscall(s) are used to receive data, applications should have
     * a loop to handle EINTR returns.
     *
     * This routine is assumed to allocate memory for the local copy
     * of the received token, which must be freed with release_buffer().
     */
    return 1;
}

static void
do_initiator(int readfd, int writefd, int anon)
{
    int initiator_established = 0, ret;
    gss_ctx_id_t ctx = GSS_C_NO_CONTEXT;
    OM_uint32 major, minor, req_flags, ret_flags;
    gss_buffer_desc input_token = GSS_C_EMPTY_BUFFER;
    gss_buffer_desc output_token = GSS_C_EMPTY_BUFFER;
    gss_buffer_desc name_buf = GSS_C_EMPTY_BUFFER;
    gss_name_t target_name = GSS_C_NO_NAME;

    /* Applications should set target_name to a real value. */
    name_buf.value = "<service>@<hostname.domain>";
    name_buf.length = strlen(name_buf.value);
    major = gss_import_name(&minor, &name_buf,
                            GSS_C_NT_HOSTBASED_SERVICE, &target_name);
    if (GSS_ERROR(major)) {
        warnx(1, "Could not import name\n");
        goto cleanup;
    }

    /* Mutual authentication will require a token from acceptor to
     * initiator and thus a second call to gss_init_sec_context(). */
    req_flags = GSS_C_MUTUAL_FLAG | GSS_C_CONF_FLAG | GSS_C_INTEG_FLAG;
    if (anon)
        req_flags |= GSS_C_ANON_FLAG;

    while (!initiator_established) {
        /* The initiator_cred_handle, mech_type, time_req,
         * input_chan_bindings, actual_mech_type, and time_rec
         * parameters are not needed in many cases.  We pass
         * GSS_C_NO_CREDENTIAL, GSS_C_NO_OID, 0, NULL, NULL, and NULL
         * for them, respectively. */
        major = gss_init_sec_context(&minor, GSS_C_NO_CREDENTIAL, &ctx,
                                     target_name, GSS_C_NO_OID,
                                     req_flags, 0, NULL, &input_token,
                                     NULL, &output_token, &ret_flags,
                                     NULL);
        /* This was allocated by receive_token() and is no longer
         * needed.  Free it now to avoid leaks if the loop continues. */
        release_buffer(&input_token);
        if (anon) {
            /* Initiators that wish to remain anonymous must check
             * whether their request has been honored before sending
             * each token. */
            if (!(ret_flags & GSS_C_ANON_FLAG)) {
                warnx("Anonymous requested but not available\n");
                goto cleanup;
            }
        }
        /* Always send a token if we are expecting another input token
         * (GSS_S_CONTINUE_NEEDED is set) or if it is nonempty. */
        if ((major & GSS_S_CONTINUE_NEEDED) ||
            output_token.length > 0) {
            ret = send_token(writefd, &output_token);
            if (ret != 0)
                goto cleanup;
        }
        /* Check for errors after sending the token so that we will send
         * error tokens. */
        if (GSS_ERROR(major)) {
            warnx("gss_init_sec_context() error major 0x%x\n", major);
            goto cleanup;
        }
        /* Free the output token's storage; we don't need it anymore.
         * gss_release_buffer() is safe to call on the output buffer
         * from gss_int_sec_context(), even if there is no storage
         * associated with that buffer. */
        (void)gss_release_buffer(&minor, &output_token);

        if (major & GSS_S_CONTINUE_NEEDED) {
            ret = receive_token(readfd, &input_token);
            if (ret != 0)
                goto cleanup;
        } else if (major == GSS_S_COMPLETE) {
            initiator_established = 1;
        } else {
            /* This situation is forbidden by RFC 2743.  Bail out. */
            warnx("major not complete or continue but not error\n");
            goto cleanup;
        }
    }   /* while (!initiator_established) */
    if ((ret_flags & req_flags) != req_flags) {
        warnx("Negotiated context does not support requested flags\n");
        goto cleanup;
    }
    printf("Initiator's context negotiation successful\n");
cleanup:
    /* We are required to release storage for nonzero-length output
     * tokens.  gss_release_buffer() zeros the length, so we
     * will not attempt to release the same buffer twice. */
    if (output_token.length > 0)
        (void)gss_release_buffer(&minor, &output_token);
    /* Do not request a context deletion token; pass NULL. */
    (void)gss_delete_sec_context(&minor, &ctx, NULL);
    (void)gss_release_name(&minor, &target_name);
}

/*
 * Perform authorization checks on the initiator's GSS name object.
 *
 * Returns 0 on success (the initiator is authorized) and nonzero
 * when the initiator is not authorized.
 */
static int
check_authz(gss_name_t client_name)
{
    /*
     * Supply authorization checking code here.
     *
     * Options include bitwise comparison of the exported name against
     * a local database and introspection against name attributes.
     */
    return 0;
}

static void
do_acceptor(int readfd, int writefd)
{
    int acceptor_established = 0, ret;
    gss_ctx_id_t ctx = GSS_C_NO_CONTEXT;
    OM_uint32 major, minor, ret_flags;
    gss_buffer_desc input_token = GSS_C_EMPTY_BUFFER;
    gss_buffer_desc output_token = GSS_C_EMPTY_BUFFER;
    gss_name_t client_name;
    major = GSS_S_CONTINUE_NEEDED;

    while (!acceptor_established) {
        if (major & GSS_S_CONTINUE_NEEDED) {
            ret = receive_token(readfd, &input_token);
            if (ret != 0)
                goto cleanup;
        } else if (major == GSS_S_COMPLETE) {
            acceptor_established = 1;
            break;
        } else {
            /* This situation is forbidden by RFC 2743.  Bail out. */
            warnx("major not complete or continue but not error\n");
            goto cleanup;
        }
        /* We can use the default behavior or do not need the returned
         * information for the parameters acceptor_cred_handle,
         * input_chan_bindings, mech_type, time_rec, and
         * delegated_cred_handle, and pass the values
         * GSS_C_NO_CREDENTIAL, NULL, NULL, NULL, and NULL,
         * respectively.  In some cases the src_name will not be
         * needed, but most likely it will be needed for some
         * authorization or logging functionality. */
        major = gss_accept_sec_context(&minor, &ctx,
                                       GSS_C_NO_CREDENTIAL,
                                       &input_token, NULL,
                                       &client_name, NULL,
                                       &output_token, &ret_flags, NULL,
                                       NULL);
        /* This was allocated by receive_token() and is no longer
         * needed.  Free it now to avoid leaks if the loop continues. */
        release_buffer(&input_token);
        /* Always send a token if we are expecting another input token
         * (GSS_S_CONTINUE_NEEDED is set) or if it is nonempty. */
        if ((major & GSS_S_CONTINUE_NEEDED) ||
            output_token.length > 0) {
            ret = send_token(writefd, &output_token);
            if (ret != 0)
                goto cleanup;
        }
        /* Check for errors after sending the token so that we will send
         * error tokens. */
        if (GSS_ERROR(major)) {
            warnx("gss_accept_sec_context() error major 0x%x\n", major);
            goto cleanup;
        }
        /* Free the output token's storage; we don't need it anymore.
         * gss_release_buffer() is safe to call on the output buffer
         * from gss_accept_sec_context(), even if there is no storage
         * associated with that buffer. */
        (void)gss_release_buffer(&minor, &output_token);
    }   /* while (!acceptor_established) */
    if (!(ret_flags & GSS_C_INTEG_FLAG)) {
        warnx("Negotiated context does not support integrity\n");
        goto cleanup;
    }
    printf("Acceptor's context negotiation successful\n");
    ret = check_authz(client_name);
    if (ret != 0)
        printf("Client is not authorized; rejecting access\n");
cleanup:
    release_buffer(&input_token);
    /* We are required to release storage for nonzero-length output
     * tokens.  gss_release_buffer() zeros the length, so we
     * will not attempt to release the same buffer twice. */
    if (output_token.length > 0)
        (void)gss_release_buffer(&minor, &output_token);
    /* Do not request a context deletion token, pass NULL. */
    (void)gss_delete_sec_context(&minor, &ctx, NULL);
    (void)gss_release_name(&minor, &client_name);
}

int
main(void)
{
    pid_t pid;
    int fd1 = -1, fd2 = -1;

    /* Create fds file descriptors for reading/writing here. */
    pid = fork();
    if (pid == 0)
        do_initiator(fd1, fd2, 0);
    else if (pid > 0)
        do_acceptor(fd2, fd1);
    else
        err(1, "fork() failed\n");
    exit(0);
}

6.  Security Considerations

   This document provides a (reasonably) concise description and example
   for correct construction of the GSS-API security context negotiation
   loop.  Since everything relating to the construction and use of a GSS
   security context is security related, there are security-relevant
   considerations throughout the document.  It is useful to call out a
   few things in this section, though.

   The GSS-API uses a request-and-check model for features.  An
   application using the GSS-API requests certain features
   (confidentiality (e.g.,
   confidentiality protection for messages or anonymity), but such a
   request does not require the GSS implementation to provide that
   feature.  The application must check the returned flags to verify
   whether a requested feature is present; if the feature was non-
   optional for the application, the application must generate an error.
   Phrased differently, the GSS-API will not generate an error if it is
   unable to satisfy the features requested by the application.

   In many cases, it is convenient for GSS acceptors to accept security
   contexts using multiple acceptor names (such as by using the default
   credential set, as happens when GSS_C_NO_CREDENTIAL is passed to
   GSS_Accept_sec_context()).  This allows acceptors to use any
   credentials to which it has they have access for accepting security
   contexts, which may not be the desired behavior for a given
   application.  (For example, the Solid-State Hybrid Drive (SSHD) Secure Shell daemon (sshd) may wish
   to accept only using GSS_C_NT_HOSTBASED credentials of the form
   host@<hostname>, and not nfs@<hostname>.)  Acceptor applications can
   check which target name was used by the initiator, but the details
   are out of scope for this document.  See Sections 2.2.6 and 1.1.5 of
   [RFC2743]

   The C sample code uses the macro GSS_ERROR() to assess the return
   value of gss_init_sec_context() and gss_accept_sec_context().  This
   is done to indicate where checks are needed in writing code for other
   languages and what the nature of those checks might be.  The C code
   could be made simpler by omitting that macro.  In applications
   expecting to receive protected octet streams, this macro should not
   be used on the result of per-message operations, as it omits checking
   for supplementary status values such as GSS_S_DUPLICATE_TOKEN,
   GSS_S_OLD_TOKEN, etc.  Use of the GSS_ERROR() macro on the results of
   GSS-API per-message operations has resulted in security
   vulnerabilities in existing software.

   The security considerations from RFCs 2743 and 2744 remain applicable
   to consumers of this document.

7.  References

7.1.  Normative References
   [RFC2743]  Linn, J., "Generic Security Service Application Program
              Interface Version 2, Update 1", RFC 2743, DOI 10.17487/
              RFC2743, January 2000,
              <http://www.rfc-editor.org/info/rfc2743>.

   [RFC2744]  Wray, J., "Generic Security Service API Version 2 :
              C-bindings", RFC 2744, DOI 10.17487/RFC2744, January 2000,
              <http://www.rfc-editor.org/info/rfc2744>.

7.2.  Informative References

   [Err4151]  RFC Errata, Errata ID 4151, RFC 2743.

   [NTLMSSP]  Microsoft Corporation, "[MS-NLMP]: NT LAN Manager (NTLM)
              Authentication Protocol", May 2014,
              <https://msdn.microsoft.com/en-us/library/cc236621.aspx>.

   [RFC2203]  Eisler, M., Chiu, A., and L. Ling, "RPCSEC_GSS Protocol
              Specification", RFC 2203, DOI 10.17487/RFC2203, September
              1997, <http://www.rfc-editor.org/info/rfc2203>.

   [RFC3645]  Kwan, S., Garg, P., Gilroy, J., Esibov, L., Westhead, J.,
              and R. Hall, "Generic Security Service Algorithm for
              Secret Key Transaction Authentication for DNS (GSS-TSIG)",
              RFC 3645, DOI 10.17487/RFC3645, October 2003,
              <http://www.rfc-editor.org/info/rfc3645>.

   [RFC4401]  Williams, N., "A Pseudo-Random Function (PRF) API
              Extension for the Generic Security Service Application
              Program Interface (GSS-API)", RFC 4401, DOI 10.17487/
              RFC4401, February 2006,
              <http://www.rfc-editor.org/info/rfc4401>.

   [RFC4462]  Hutzelman, J., Salowey, J., Galbraith, J., and V. Welch,
              "Generic Security Service Application Program Interface
              (GSS-API) Authentication and Key Exchange for the Secure
              Shell (SSH) Protocol", RFC 4462, DOI 10.17487/RFC4462, May
              2006, <http://www.rfc-editor.org/info/rfc4462>.

   [RFC4752]  Melnikov, A., Ed., "The Kerberos V5 ("GSSAPI") Simple
              Authentication and Security Layer (SASL) Mechanism", RFC
              4752, DOI 10.17487/RFC4752, November 2006,
              <http://www.rfc-editor.org/info/rfc4752>.

   [RFC5801]  Josefsson, S. and N. Williams, "Using Generic Security
              Service Application Program Interface (GSS-API) Mechanisms
              in Simple Authentication and Security Layer (SASL): The
              GS2 Mechanism Family", RFC 5801, DOI 10.17487/RFC5801,
              July 2010, <http://www.rfc-editor.org/info/rfc5801>.

   [RFC6680]  Williams, N., Johansson, L., Hartman, S., and S.
              Josefsson, "Generic Security Service Application
              Programming Interface (GSS-API) Naming Extensions", RFC
              6680, DOI 10.17487/RFC6680, August 2012,
              <http://www.rfc-editor.org/info/rfc6680>.

Acknowledgements

   Thanks to Nico Williams and Jeff Hutzleman for prompting me to write
   this document.

Author's Address

   Benjamin Kaduk
   MIT Kerberos Consortium

   EMail: kaduk@mit.edu