> On Sep 8, 2019, at 1:09 PM, Jeremy Harris <jgh@xxxxxxxxxxx> wrote:
>
>> I have a note from 2017 in my code to the effect that
>> SSL_get_certificate() is broken in that it returns the last
>> cert loaded rather than the one passed out to the client
>> (on the server).
>
> Note that one might have both an EC and an RSA cert loaded.
One of the "CHANGES" entries for 1.0.1d reads:
*) Call OCSP Stapling callback after ciphersuite has been chosen, so
the right response is stapled. Also change SSL_get_certificate()
so it returns the certificate actually sent.
See http://rt.openssl.org/Ticket/Display.html?id=2836.
[Rob Stradling <rob.stradling@xxxxxxxxxx>]
Consequently 1.0.1d and later had the expected behaviour. However,
in commits this was updated:
dc144417571735c82853421a8845ef603d828a0b (1.0.2-beta1)
e5db9c3b67deb80e274f66e3832a9cfba931670c (also master, at the time 1.1.0-dev)
Minor enhancement to PR#2836 fix. Instead of modifying SSL_get_certificate
change the current certificate (in s->cert->key) to the one used and then
SSL_get_certificate and SSL_get_privatekey will automatically work.
The code for "change the current certificate" was:
diff --git a/ssl/t1_lib.c b/ssl/t1_lib.c
index d0764e8cd3..a438321a41 100644
--- a/ssl/t1_lib.c
+++ b/ssl/t1_lib.c
@@ -2355,6 +2355,18 @@ int ssl_check_clienthello_tlsext_late(SSL *s)
if ((s->tlsext_status_type != -1) && s->ctx && s->ctx->tlsext_status_cb)
{
int r;
+ CERT_PKEY *certpkey;
+ certpkey = ssl_get_server_send_pkey(s);
+ /* If no certificate can't return certificate status */
+ if (certpkey == NULL)
+ {
+ s->tlsext_status_expected = 0;
+ return 1;
+ }
+ /* Set current certificate to one we will use so
+ * SSL_get_certificate et al can pick it up.
+ */
+ s->cert->key = certpkey;
But it only runs if there's a "tlsext_status_cb" callback, which may
not cover all the expected use-cases. I think this merits a new
issue on Github. On servers the function should return the certificate
used, whether that happened via an SNI callback, or simply via choosing
a certificate for the negotiated algorithm...
>> Is this still the case? I can't track down any obvious fix in the
>> OpenSSL git.
>
> Nobody knows?
Perhaps I do now...
--
Viktor.
