OpenSSL 1.0.2* doesn’t preserve digests for SNI #4554
Comments
|
Do you all call Edit: Changing their order will probably upset |
Yes, we are. This is originally reported to Apache Traffic Server. ( apache/trafficserver#2482 ) If you need details of how we call, please refer below. |
|
Ah okay. If my theory is right, I believe you'll find the OpenSSL bug no longer reproduces if you turn |
|
Right. I don't see this issue after I turned off |
1ce95f1 was incomplete and did not handle the case when SSL_set_SSL_CTX was called from the cert_cb callback rather than the SNI callback. The consequence is any server using OpenSSL 1.0.2 and the cert_cb callback for SNI only ever signs a weak digest, SHA-1, even when connecting to clients which use secure ones. Fix this and add regression tests for both this and the original issue. Fixes openssl#4554.
1ce95f1 was incomplete and did not handle the case when SSL_set_SSL_CTX was called from the cert_cb callback rather than the SNI callback. The consequence is any server using OpenSSL 1.0.2 and the cert_cb callback for SNI only ever signs a weak digest, SHA-1, even when connecting to clients which use secure ones. Fix this and add regression tests for both this and the original issue. Fixes #4554. Reviewed-by: Emilia Käsper <emilia@openssl.org> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from #4577)
|
This is fixed in 1.0.2m. Thanks! |
SHA1 is always used as digest of signature algorithm, when
SSL_set_SSL_CTX()is called for SNI.We only see this issue with OpenSSL 1.0.2*. We don’t see this issue with 1.0.1* or 1.1.0*.
This was originally fixed by 4e05aed for 1.0.1* and 14e14bf for master. The change for master was cherry-picked to 1.0.2* (1ce95f1) but it looks not enough.
IMO, it should copy
CERT_PKEYs like the fix for 1.0.1* or negotiate signature algorithm from copiedpeer_sigalgsagain.The text was updated successfully, but these errors were encountered: