[PATCH 11 of 11] SSL: automatic rotation of session ticket keys

Sergey Kandaurov pluknet at nginx.com
Wed Oct 12 13:57:07 UTC 2022


> On 9 Oct 2022, at 08:59, Maxim Dounin <mdounin at mdounin.ru> wrote:
> 
> Hello!
> 
> On Sat, Oct 01, 2022 at 11:58:20AM +0300, Maxim Dounin wrote:
> 
>> On Thu, Sep 29, 2022 at 08:00:03PM +0400, Sergey Kandaurov wrote:
>> 
>>>> On 28 Sep 2022, at 22:37, Maxim Dounin <mdounin at mdounin.ru> wrote:
>>>> 
>>>> On Mon, Sep 26, 2022 at 02:17:18PM +0400, Sergey Kandaurov wrote:
>> 
>> [...]
>> 
>>>>> And by the way, while reviewing this patch, I noticed that
>>>>> OpenSSL doesn't allow a client to gracefully renew TLSv1.2 session
>>>>> when the client receives a new session ticket in resumed sessions.
>>>>> In practice, it is visible when client resumes a not yet expired
>>>>> session encrypted with not a fresh ticket key (after rotation),
>>>>> which results in sending a new session ticket.
>>>>> See ssl_update_cache() for the !s->hit condition.
>>>>> In the opposite, BoringSSL always allows to renew TLSv1.2 sessions.
>>>> 
>>>> You mean on the client side?  Yes, it looks like 
>>>> ngx_ssl_new_client_session() won't be called for such a new 
>>>> session ticket, and updated ticket will be never saved.  This 
>>>> might need to be worked around.
>>> 
>>> Yes, I mean the client side.
>>> 
>>>> 
>>>> This should be safe with the key rotation logic introduced in this 
>>>> patch though, given that the previous key is preserved till the 
>>>> last ticket encrypted with it is expected to expire.
>>>> 
>>>> One of the possible solutions might be to avoid re-encryption of 
>>>> tickets with the new key, as the old key is anyway expected to be 
>>>> available till the session expires.
>>> 
>>> I don't think it's worth the effort.  If I got you right, and
>>> as far as I understand, re-encrypting the ticket essentially
>>> means sending a fresh session (renewal).
>> 
>> Well, not really.  Re-encryption of a ticket does not imply 
>> session renewal.  Further, doing so implies security risk: if we 
>> renew a session during re-encryption, this makes it possible to 
>> create essentially infinite sessions.  And, for example, if a 
>> session used a client certificate, this effectively means that 
>> this certificate will never expire and cannot be revoked.
>> 
>> With TLSv1.2, OpenSSL follows this logic: session expiration time 
>> is set when a session is created, and ticket re-encryption only 
>> re-encrypts the session, but doesn't change session expiration.  
>> As such, any certificate validation which happens during session 
>> creation needs to be redone once session timeout expires - and 
>> this makes it possible to implement certificate revocation.
>> 
>> On the other hand, as implemented for TLSv1.3 at least in OpenSSL 
>> it seems not to be the case.  Every ticket sent to the client 
>> actually creates a new session with _updated_ expiration time.  As 
>> such, it is possible to create a session authenticated with a client 
>> certificate, and use this session indefinitely, even after the 
>> certificate will expire and/or will be revoked.
>> 
>> This seems to be a security issue in OpenSSL.
>> 
>> BoringSSL seems to behave similarly with TLSv1.3, that is, it 
>> updates session expiration time, making it possible to use an 
>> authenticated session for much longer than session timeout 
>> configured.  But BoringSSL also has session's auth_timeout, which 
>> prevents indefinite use of the session.  The auth_timeout value is 
>> hardcoded to 7 days (SSL_DEFAULT_SESSION_AUTH_TIMEOUT), and does 
>> not seem to be adjustable (only with SSL_SESSION_set_timeout(), 
>> which is documented to be a function for writing tests).
>> 
>> I would rather say it is also a security issue in BoringSSL, 
>> though it's slightly better than in OpenSSL due to the 7 days 
>> limit.
> 
> For the record:
> 
> https://github.com/openssl/openssl/issues/19341
> 
> Note that with the automatic ticket key rotation this issue with 
> TLSv1.3 sessions becomes slightly worse in a typical configuration 
> (with ssl_session_cache in shared memory, but without 
> ssl_session_ticket_key explicitly set and/or ssl_session_tickets 
> switched off).  Notably, configuration reload is no longer enough 
> to invalidate all tickets, since ticket keys are now preserved in 
> shared memory.
> 
> For example, consider that a CRL file is updated with new 
> revocations, and nginx configuration is reloaded.  New revocations 
> will be loaded by nginx and will appear to work with new sessions, 
> but can be easily bypassed by maintaining a previously established 
> TLSv1.3 session.  Previously, it was possible to bypass 
> revocations in such scenario only if ticket keys were explicitly 
> set or if session tickets were switched off and sessions were 
> cached in shared memory.
> 
> Given that we do not enable TLSv1.3 by default, we probably can 
> ignore this and wait for appropriate fixes from the affected 
> libraries.  On the other hand, it might be a good idea to 
> introduce a workaround, especially if we want to enable TLSv1.3 by 
> default in the foreseeable future.
> 
> The following patch seems to be simple enough and forces session 
> timeouts for TLSv1.3 sessions to be as configured, for both 
> OpenSSL and BoringSSL:

Although somewhat tricky, I like the approach.
Nitpicking comments inline.

> 
> # HG changeset patch
> # User Maxim Dounin <mdounin at mdounin.ru>
> # Date 1665286021 -10800
> #      Sun Oct 09 06:27:01 2022 +0300
> # Node ID c0ec4df7ccbb95b7f2c2842f40012082991bed52
> # Parent  37a4ac7ba1c5a003ab85f73d77767058af4eae30
> SSL: workaround for session timeout handling with TLSv1.3.
> 
> OpenSSL with TLSv1.3 updates the session creation time on session
> resumption and keeps the session timeout unmodified, making it possible
> to maintain the session forever, bypassing client certificate expiration
> and revocation.  To make sure session timeouts are actually used, we
> now update the session creation time and reduce the session timeout
> accordingly.
> 
> BoringSSL with TLSv1.3 ignores configured session timeouts and uses a
> hardcoded timeout instead, 7 days.  So we update session timeout to
> the configured value as soon as a session is created.
> 
> diff --git a/src/event/ngx_event_openssl.c b/src/event/ngx_event_openssl.c
> --- a/src/event/ngx_event_openssl.c
> +++ b/src/event/ngx_event_openssl.c
> @@ -1086,6 +1086,53 @@ ngx_ssl_info_callback(const ngx_ssl_conn
> 
> #endif
> 
> +#ifdef TLS1_3_VERSION
> +
> +    if ((where & SSL_CB_ACCEPT_LOOP) == SSL_CB_ACCEPT_LOOP
> +        && SSL_version(ssl_conn) == TLS1_3_VERSION)
> +    {
> +        time_t        now, time, timeout, conf_timeout;
> +        SSL_SESSION  *sess;

Not sure if this should use ngx_ssl_session_t.  But, given the macro
is intended for external consumption, SSL_SESSION should be fine.

> +
> +        /*
> +         * OpenSSL with TLSv1.3 updates the session creation time on
> +         * session resumption and keeps the session timeout unmodified,
> +         * making it possible to maintain the session forever, bypassing
> +         * client certificate expiration and revocation.  To make sure
> +         * session timeouts are actually used, we now update the session
> +         * creation time and reduce the session timeout accordingly.
> +         *
> +         * BoringSSL with TLSv1.3 ignores configured session timeouts
> +         * and uses a hardcoded timeout instead, 7 days.  So we update
> +         * session timeout to the configured value as soon as a session
> +         * is created.
> +         */
> +
> +        c = ngx_ssl_get_connection((ngx_ssl_conn_t *) ssl_conn);
> +        sess = SSL_get0_session(ssl_conn);
> +
> +        if (!c->ssl->session_timeout_set && sess) {
> +            c->ssl->session_timeout_set = 1;
> +
> +            now = ngx_time();
> +            time = SSL_SESSION_get_time(sess);
> +            timeout = SSL_SESSION_get_timeout(sess);
> +            conf_timeout = SSL_CTX_get_timeout(c->ssl->session_ctx);
> +
> +            timeout = ngx_min(timeout, conf_timeout);
> +
> +            if (now - time >= timeout) {
> +                SSL_SESSION_set1_id_context(sess, (unsigned char *) "", 0);

Why not u_char? If this is to strictly follow the declaration, without
typedef's, then I wonder why const is omitted in casting 2nd parameter.
Anyway, this passes compilation on known CI platforms.
For the record, BoringSSL moved long ago to uint8_t in its headers,
so is incompatible with CHAR_BIT > 8 (though, POSIX mandates 8).

I've been pondering if it's worth to call SSL_SESSION_set1_id_context()
in order to (obfuscatedly) cancel the going to expire sessions.
At least, OpenSSL handles this on theirself for sessions that've already
expired ("now > time + timeout"), rejecting such session, so the only
viable condition seems to be is when the session is going to be expired
("now == time + timeout").
For the record, invalidating session context (or even session removal)
this way doesn't prevent from reusing the session in this connection,
since the info callback is called too late, after the ticket has been
successfully decrypted (with a check for timeout and session context)
and session restored.  Such session will be rejected only the next time.
It may have sense though to call it still to handle the going to expire
sessions that pass the session timeout check on server, see sess_timedout()
in OpenSSL sources, it has slightly different condition to reject sessions.
In this case the control goes to the "if (now - time >= timeout) {"
condition, where we need to take the action, as otherwise OpenSSL
will update the session creation time, such that the session will
continue to be resumable for another "timeout" seconds.
There should be no difference between invalidating context and setting
intentionally old values for session time and timeout, both should
work to make it stop from being resumable.  So, condition could be
collapsed to update session time and timeout values in both cases.
But zero session timeout doesn't seem to pass i2d_SSL_SESSION checks
in OpenSSL while constructing new session ticket, so invalidating
session context looks like the only solution.

> +
> +            } else {
> +                SSL_SESSION_set_time(sess, now);
> +                SSL_SESSION_set_timeout(sess, timeout - (now - time));
> +            }
> +        }
> +    }
> +
> +#endif
> +
>     if ((where & SSL_CB_ACCEPT_LOOP) == SSL_CB_ACCEPT_LOOP) {
>         c = ngx_ssl_get_connection((ngx_ssl_conn_t *) ssl_conn);
> 
> diff --git a/src/event/ngx_event_openssl.h b/src/event/ngx_event_openssl.h
> --- a/src/event/ngx_event_openssl.h
> +++ b/src/event/ngx_event_openssl.h
> @@ -114,6 +114,7 @@ struct ngx_ssl_connection_s {
>     unsigned                    no_send_shutdown:1;
>     unsigned                    shutdown_without_free:1;
>     unsigned                    handshake_buffer_set:1;
> +    unsigned                    session_timeout_set:1;
>     unsigned                    try_early_data:1;
>     unsigned                    in_early:1;
>     unsigned                    in_ocsp:1;
> 

Looks good.

-- 
Sergey Kandaurov



More information about the nginx-devel mailing list