Hi Alan, I have a implementation of EAP-TEAP from EAP-FAST which is very similar to what we have in the 3.2x where I am overriding the tls decision "TEAP-Idenity-Ordinal" (gives me its machine/user) +#define TEAP_MACHINE_CERT "1" +#define TEAP_USER_CERT "2" + #ifdef ENABLE_OPENSSL_VERSION_CHECK typedef struct libssl_defect { uint64_t high; @@ -1651,6 +1654,7 @@ static CONF_PARSER tls_server_config[] = { #endif { "ca_path_reload_interval", FR_CONF_OFFSET(PW_TYPE_INTEGER, fr_tls_server_conf_t, ca_path_reload_interval), "0" }, { "allow_expired_crl", FR_CONF_OFFSET(PW_TYPE_BOOLEAN, fr_tls_server_conf_t, allow_expired_crl), NULL }, + { "allow_cert_failure", FR_CONF_OFFSET(PW_TYPE_BOOLEAN, fr_tls_server_conf_t, allow_cert_failure), "no" }, { "check_cert_cn", FR_CONF_OFFSET(PW_TYPE_STRING, fr_tls_server_conf_t, check_cert_cn), NULL }, { "cipher_list", FR_CONF_OFFSET(PW_TYPE_STRING, fr_tls_server_conf_t, cipher_list), NULL }, { "cipher_server_preference", FR_CONF_OFFSET(PW_TYPE_BOOLEAN, fr_tls_server_conf_t, cipher_server_preference), NULL }, @@ -3145,6 +3149,23 @@ int cbtls_verify(int ok, X509_STORE_CTX *ctx) X509_STORE_CTX_set_error( ctx, 0 ); } + if (!my_ok && (conf->allow_cert_failure)) { + VALUE_PAIR *teap_ordinal_vp = NULL; + DICT_ATTR const *teap_attr = dict_attrbyname("TEAP-Identity-Ordinal"); + if (teap_attr) { + teap_ordinal_vp = fr_pair_find_by_da(request->packet->vps, teap_attr, TAG_ANY); + // If TEAP User Certificate -> override the decision to success + if (strcmp(teap_ordinal_vp->vp_strvalue, TEAP_USER_CERT) == 0) { + DEBUG("TEAP context detected - Overriding User Cert Failure!"); + my_ok = 1; + X509_STORE_CTX_set_error( ctx, 0 ); + } + } else { + // TODO: Remove this later + DEBUG("Extracting teap_ordinal something went wrong!"); + } + } I know this is deviating from the RFC but I was able to override the TLS failure if this helps Thanks, Suriya On Tue, Dec 9, 2025 at 12:40 PM Jan Kříž <jan.kriz1867@gmail.com> wrote:
The situation of *skipping* user authentication is very different from allowing *failed* user authentication. TEAP explicitly allows the server to say certain kinds of authentication are allowed (but not required), or required.
Oh I see, I misunderstood it completely. Thank you for pointing this out.
The server allows this today, with a manual / complex configuration. It should be easier in a few weeks when I push some tweaks back. What it doesn't support is "user tried to authenticate, and got a reject. We still allow the overall TEAP authentication to succeed". The TEAP specs don't envision this workflow. I would be surprised if anything supported it.
This is exactly what I tried to set up in FreeRADIUS some time ago, but I must have misconfigured something and if you say the configuration for this is complex, I most likely just did it wrong. Looking forward to those patches and thank you very much for clarifying this to me.
On Tue, Dec 9, 2025 at 8:57 PM Alan DeKok <alan.dekok@inkbridge.io> wrote:
On Dec 9, 2025, at 2:15 PM, Jan Kříž <jan.kriz1867@gmail.com> wrote:
To give a bit more context on why I went down this rabbit hole: I was researching options for handling "partial failures", specifically the "chicken and egg" scenario where a user certificate is missing on a freshly provisioned device.
The situation of *skipping* user authentication is very different from
allowing *failed* user authentication.
TEAP explicitly allows the server to say certain kinds of
authentication are allowed (but not required), or required.
That led me to a "Wires and Wi-Fi" blog post describing how TEAP is based on EAP-FASTv2 and how EAP chaining allows the server to make decisions based on combined states, including "User Failed and Machine Succeeded" (
https://www.wiresandwi.fi/blog/windows-network-authentication-sequence).
According to that explanation, Windows can perform machine-only authentication during startup, even when the user certificate does not exist yet, and the RADIUS server can return an Access-Accept with a restricted authorization profile. This gives the device enough limited network access to reach the PKI and enroll the missing user certificate.
Yes, that's explicitly allowed. FreeRADIUS supports this with a bit of manual configuration. I'm working on patches to make that easier. They should be available in January.
I also found a Cisco ISE guide showing this workflow in practice, with explicit rules for "Machine authenticated, User not" that grant reduced access, distinct from the full "User and Machine" success state. ( https://www.ise-support.com/2020/05/29/using-teap-for-eap-chaining/).
Based on these examples, it appears that several commercial platforms intentionally treat the session as successful overall (sending Access-Accept), even when one inner method fails,
Again, you have to double / triple check the difference between "not doing user authentication", and "did user authentication, but it failed".
Many people confuse the two, which isn't good.
Given that this "partial success with restricted access" behavior seems to have become a de facto standard method for handling similar scenarios, I wanted to ask whether FreeRADIUS would consider supporting this mode?. Or is the project's stance that FreeRADIUS should follow the RFC strictly until the specification explicitly allows this behavior?
The server allows this today, with a manual / complex configuration. It should be easier in a few weeks when I push some tweaks back.
What it doesn't support is "user tried to authenticate, and got a reject. We still allow the overall TEAP authentication to succeed". The TEAP specs don't envision this workflow. I would be surprised if anything supported it.
Alan DeKok.
- List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html