Initial credentials

Software that performs tasks such as logging users into a computer when they type their Kerberos password needs to get initial credentials (usually ticket granting tickets) from Kerberos. Such software shares some behavior with the kinit program.

Whenever a program grants access to a resource (such as a local login session on a desktop computer) based on a user successfully getting initial Kerberos credentials, it must verify those credentials against a secure shared secret (e.g., a host keytab) to ensure that the user credentials actually originate from a legitimate KDC. Failure to perform this verification is a critical vulnerability, because a malicious user can execute the “Zanarotti attack”: the user constructs a fake response that appears to come from the legitimate KDC, but whose contents come from an attacker-controlled KDC.

Some applications read a Kerberos password over the network (ideally over a secure channel), which they then verify against the KDC. While this technique may be the only practical way to integrate Kerberos into some existing legacy systems, its use is contrary to the original design goals of Kerberos.

The function krb5_get_init_creds_password() will get initial credentials for a client using a password. An application that needs to verify the credentials can call krb5_verify_init_creds(). Here is an example of code to obtain and verify TGT credentials, given strings princname and password for the client principal name and password:

krb5_error_code ret;
krb5_creds creds;
krb5_principal client_princ = NULL;

memset(&creds, 0, sizeof(creds));
ret = krb5_parse_name(context, princname, &client_princ);
if (ret)
    goto cleanup;
ret = krb5_get_init_creds_password(context, &creds, client_princ,
                                   password, NULL, NULL, 0, NULL, NULL);
if (ret)
    goto cleanup;
ret = krb5_verify_init_creds(context, &creds, NULL, NULL, NULL, NULL);

cleanup:
krb5_free_principal(context, client_princ);
krb5_free_cred_contents(context, &creds);
return ret;

Options for get_init_creds

The function krb5_get_init_creds_password() takes an options parameter (which can be a null pointer). Use the function krb5_get_init_creds_opt_alloc() to allocate an options structure, and krb5_get_init_creds_opt_free() to free it. For example:

krb5_error_code ret;
krb5_get_init_creds_opt *opt = NULL;
krb5_creds creds;

memset(&creds, 0, sizeof(creds));
ret = krb5_get_init_creds_opt_alloc(context, &opt);
if (ret)
    goto cleanup;
krb5_get_init_creds_opt_set_tkt_life(opt, 24 * 60 * 60);
ret = krb5_get_init_creds_password(context, &creds, client_princ,
                                   password, NULL, NULL, 0, NULL, opt);
if (ret)
    goto cleanup;

cleanup:
krb5_get_init_creds_opt_free(context, opt);
krb5_free_cred_contents(context, &creds);
return ret;

Getting anonymous credentials

As of release 1.8, it is possible to obtain fully anonymous or partially anonymous (realm-exposed) credentials, if the KDC supports it. The MIT KDC supports issuing fully anonymous credentials as of release 1.8 if configured appropriately (see Anonymous PKINIT), but does not support issuing realm-exposed anonymous credentials at this time.

To obtain fully anonymous credentials, call krb5_get_init_creds_opt_set_anonymous() on the options structure to set the anonymous flag, and specify a client principal with the KDC’s realm and a single empty data component (the principal obtained by parsing @realmname). Authentication will take place using anonymous PKINIT; if successful, the client principal of the resulting tickets will be WELLKNOWN/ANONYMOUS@WELLKNOWN:ANONYMOUS. Here is an example:

krb5_get_init_creds_opt_set_anonymous(opt, 1);
ret = krb5_build_principal(context, &client_princ, strlen(myrealm),
                           myrealm, "", (char *)NULL);
if (ret)
    goto cleanup;
ret = krb5_get_init_creds_password(context, &creds, client_princ,
                                   password, NULL, NULL, 0, NULL, opt);
if (ret)
    goto cleanup;

To obtain realm-exposed anonymous credentials, set the anonymous flag on the options structure as above, but specify a normal client principal in order to prove membership in the realm. Authentication will take place as it normally does; if successful, the client principal of the resulting tickets will be WELLKNOWN/ANONYMOUS@realmname.

User interaction

Authenticating a user usually requires the entry of secret information, such as a password. A password can be supplied directly to krb5_get_init_creds_password() via the password parameter, or the application can supply prompter and/or responder callbacks instead. If callbacks are used, the user can also be queried for other secret information such as a PIN, informed of impending password expiration, or prompted to change a password which has expired.

Prompter callback

A prompter callback can be specified via the prompter and data parameters to krb5_get_init_creds_password(). The prompter will be invoked each time the krb5 library has a question to ask or information to present. When the prompter callback is invoked, the banner argument (if not null) is intended to be displayed to the user, and the questions to be answered are specified in the prompts array. Each prompt contains a text question in the prompt field, a hidden bit to indicate whether the answer should be hidden from display, and a storage area for the answer in the reply field. The callback should fill in each question’s reply->data with the answer, up to a maximum number of reply->length bytes, and then reset reply->length to the length of the answer.

A prompter callback can call krb5_get_prompt_types() to get an array of type constants corresponding to the prompts, to get programmatic information about the semantic meaning of the questions. krb5_get_prompt_types() may return a null pointer if no prompt type information is available.

Text-based applications can use a built-in text prompter implementation by supplying krb5_prompter_posix() as the prompter parameter and a null pointer as the data parameter. For example:

ret = krb5_get_init_creds_password(context, &creds, client_princ,
                                   NULL, krb5_prompter_posix, NULL, 0,
                                   NULL, NULL);

Responder callback

A responder callback can be specified through the init_creds options using the krb5_get_init_creds_opt_set_responder() function. Responder callbacks can present a more sophisticated user interface for authentication secrets. The responder callback is usually invoked only once per authentication, with a list of questions produced by all of the allowed preauthentication mechanisms.

When the responder callback is invoked, the rctx argument can be accessed to obtain the list of questions and to answer them. The krb5_responder_list_questions() function retrieves an array of question types. For each question type, the krb5_responder_get_challenge() function retrieves additional information about the question, if applicable, and the krb5_responder_set_answer() function sets the answer.

Responder question types, challenges, and answers are UTF-8 strings. The question type is a well-known string; the meaning of the challenge and answer depend on the question type. If an application does not understand a question type, it cannot interpret the challenge or provide an answer. Failing to answer a question typically results in the prompter callback being used as a fallback.

Password question

The KRB5_RESPONDER_QUESTION_PASSWORD (or "password") question type requests the user’s password. This question does not have a challenge, and the response is simply the password string.

One-time password question

The KRB5_RESPONDER_QUESTION_OTP (or "otp") question type requests a choice among one-time password tokens and the PIN and value for the chosen token. The challenge and answer are JSON-encoded strings, but an application can use convenience functions to avoid doing any JSON processing itself.

The krb5_responder_otp_get_challenge() function decodes the challenge into a krb5_responder_otp_challenge structure. The krb5_responder_otp_set_answer() function selects one of the token information elements from the challenge and supplies the value and pin for that token.

PKINIT password or PIN question

The KRB5_RESPONDER_QUESTION_PKINIT (or "pkinit") question type requests PINs for hardware devices and/or passwords for encrypted credentials which are stored on disk, potentially also supplying information about the state of the hardware devices. The challenge and answer are JSON-encoded strings, but an application can use convenience functions to avoid doing any JSON processing itself.

The krb5_responder_pkinit_get_challenge() function decodes the challenges into a krb5_responder_pkinit_challenge structure. The krb5_responder_pkinit_set_answer() function can be used to supply the PIN or password for a particular client credential, and can be called multiple times.

Example

Here is an example of using a responder callback:

static krb5_error_code
my_responder(krb5_context context, void *data,
             krb5_responder_context rctx)
{
    krb5_error_code ret;
    krb5_responder_otp_challenge *chl;

    if (krb5_responder_get_challenge(context, rctx,
                                     KRB5_RESPONDER_QUESTION_PASSWORD)) {
        ret = krb5_responder_set_answer(context, rctx,
                                        KRB5_RESPONDER_QUESTION_PASSWORD,
                                        "open sesame");
        if (ret)
            return ret;
    }
    ret = krb5_responder_otp_get_challenge(context, rctx, &chl);
    if (ret == 0 && chl != NULL) {
        ret = krb5_responder_otp_set_answer(context, rctx, 0, "1234",
                                            NULL);
        krb5_responder_otp_challenge_free(context, rctx, chl);
        if (ret)
            return ret;
    }
    return 0;
}

static krb5_error_code
get_creds(krb5_context context, krb5_principal client_princ)
{
    krb5_error_code ret;
    krb5_get_init_creds_opt *opt = NULL;
    krb5_creds creds;

    memset(&creds, 0, sizeof(creds));
    ret = krb5_get_init_creds_opt_alloc(context, &opt);
    if (ret)
        goto cleanup;
    ret = krb5_get_init_creds_opt_set_responder(context, opt, my_responder,
                                                NULL);
    if (ret)
        goto cleanup;
    ret = krb5_get_init_creds_password(context, &creds, client_princ,
                                       NULL, NULL, NULL, 0, NULL, opt);

cleanup:
    krb5_get_init_creds_opt_free(context, opt);
    krb5_free_cred_contents(context, &creds);
    return ret;
}

Verifying initial credentials

Use the function krb5_verify_init_creds() to verify initial credentials. It takes an options structure (which can be a null pointer). Use krb5_verify_init_creds_opt_init() to initialize the caller-allocated options structure, and krb5_verify_init_creds_opt_set_ap_req_nofail() to set the “nofail” option. For example:

krb5_verify_init_creds_opt vopt;

krb5_verify_init_creds_opt_init(&vopt);
krb5_verify_init_creds_opt_set_ap_req_nofail(&vopt, 1);
ret = krb5_verify_init_creds(context, &creds, NULL, NULL, NULL, &vopt);

The confusingly named “nofail” option, when set, means that the verification must actually succeed in order for krb5_verify_init_creds() to indicate success. The default state of this option (cleared) means that if there is no key material available to verify the user credentials, the verification will succeed anyway. (The default can be changed by a configuration file setting.)

This accommodates a use case where a large number of unkeyed shared desktop workstations need to allow users to log in using Kerberos. The security risks from this practice are mitigated by the absence of valuable state on the shared workstations—any valuable resources that the users would access reside on networked servers.