This commit modifies testssl.sh so that TLS 1.3 handshakes that use post-quantum algorithms for key exchange can be decrypted, if $OPENSSL supports the algorithms.
This commit adds support for server certificates that have ML-DSA pubilc keys. It also adds supports for certificates that are signed with ML-DSA or SLH-DSA. The ML-DSA code points for the signature_algorithms extension are taken from https://datatracker.ietf.org/doc/draft-tls-westerbaan-mldsa/ and are the ones used by OpenSSL 3.5.0.
In OpenSSL 3.5.0 the `list` command can be used to obtain a list of supported groups for TLS. The commit makes use of this command when $OPENSSL is OpenSSL 3.5.0 or later. This should be faster than testing curves one at a time.
This commit fixes a check where the Boolean variables $HAS_UDS and $HAS_UDS2 are checked for whether they are empty rather than for whether they are true.
Decryption is TLS 1.3 handshakes is very slow if the response is encrypted using ChaCha20 and the $OPENSSL enc command does not support ChaCha20. This commit mitigates that problem by using $OPENSSL2 for ChaCha20 decryption if such decryption is needed and $OPENSSL does not support it.
This commit also changes testssl.sh to make use of $OPENSSL2 for AES-GCM decryption, when $OPENSSL2 supports it, but $OPENSSL does not. However, this change is not as important. Implementing AES-GCM in Bash using $OPENSSL for AES ECB operations isn't nearly as slow as fully implementing ChaCha20 in Bash.
As a quick hack this PR enables *basically* the IPv6 proxy which results that testssl.sh
will use an IPv6 proxy when
* the binary supports that
* the binary is used an not tls_sockets()
* there's no A record but an AAAA record of the proxy or an IPv6 address as proxy address was specified.
The latter should guarantee that it doesn't break anything.
However tls_sockets() still uses IPv4 for the connection to the proxy.
See #1105
On MacOS `run_http_header()` hiccuped when in any place of the web site unprintable chars were returned, see https://github.com/testssl/testssl.sh/issues/2708#issuecomment-2738347784 .
This PR fixes that by moving the sanitization to a separate function and run it earlier before any processing of the returned content (header plus body) takes place.
Output was:
```
'HTTP Status Code awk: towc: multibyte conversion failure on: '� disabilitato");
input record number 36, file /tmp/testssl.FHu8E0/AAA.BBB.CCC.DDD.http_header.txt
source line number 1
'wk: towc: multibyte conversion failure on: '� disabilitato");
input record number 36, file /tmp/testssl.FHu8E0/AAA.BBB.CCC.DDD.http_header.txt
source line number 1
200 OK
```
When parsing the intermediate cerificates there was a parser problem when using
LibreSSL and not OpenSSL.
This PR fixes the issue #2705 and addresses the problem by using the -nameopt multiline command
of LibreSSL/OpenSSL .
The current code for setting $host_header in check_revocation_ocsp() will not work for LibreSSL 3.* or for upcoming versions of OpenSSL (version 4 or greater). The check will also not work correctly if $OPENSSL2 is used and $OPENSSL2 is not /usr/bin/openssl. This commit fixes these issues.
Previously in 4f1a91f92e there was a
double header sent to the server to check whether the certificate
was revoked.
This PR addresses that and fixes#2667 .
Before version 3.0.0, OpenSSL versions includes letters (e.g., 1.1.1w). Such version numbers confuse [[ ${OSSL_VER//./} -ge 210 ]]. Since this check is only needed for LibreSSL, this commit adds a conditional so that the check is only performed for LibreSSL.
As `--phone-out` sometimes doesn't work with our binary we switch transparently/automagically
to the vendor support openssl binary -- if available.
This fixes at least #2516 where the issue has been explained/debugged in detail.
See also #2667 and #1275.
According to https://openssl-library.org/roadmap/index.html, OpenSSL 4 is scheduled to be released in April 2026. This commit modifies code that checks $OPENSSL version numbers so that versions 4 and above will be recognized.
This commit modifies testssl.sh so that run_server_defaults() prints the server's supported TLS extensions sorted by extension number rather than listing them in the order in which they were found.
In order to simplify the sorting of the extensions, this commit changes $TLS_EXTENSIONS from a string to an array. In February 2017 comments were added (925e1061b2) saying that it would be $TLS_EXTENSIONS were an array. So, this commit addresses those comments. However, it is possible that the reason for those comments no longer apply.
With the current master branch of OpenSSL (3.5.0-dev), the output that is provided by s_client has changed in the case of a cipher suite that uses an ephemeral key. Rather than preceding the ephemeral key information with "Server Temp Key: ", it is now preceded by either "Peer Temp Key:" or "Negotiated TLS1.3 group:". This commit modifies the lines that extract ephemeral key information from OpenSSL responses to accept any of these strings.
In order to avoid delays due to lookups of the hostname "invalid." we
just avoid to use "invalid." whenever possible. :-)
Therefore we just do a test before whether `$OPENSSL s_client 2>&1 </dev/null`
does a connect, except when a WSL system is discovered. If that succeeds
we omit the part `-connect invalid.` to check whether the curve is supported.
In some quick testing this in fact improved the startup time.
This seemed to work under Linux with several openssl and one LibreSSL binary.
More testing would be required, especially e.g. under WSL / WSL2.
Also in `sclient_supported()` the `$OPENSSL s_client` statement was changed
in a similar fashion. That worked so far but would need to be observed more
closely.
In order to tell openssl binaries better apart the short banner below the
hash tag signs contain now also the date. That is the short version of the
build date unless it is not supplied which is the case of opensuse. Then
the name contains the date and it's taken from there.
The start and end banner lines have the same length now.
"sieve" was added in a comment and the sequence where sieve appears in
a pattern was trying to match other occurences (i.e. after nntp)
While testing the banners it appeared under Linux that a) the vendor
supplied openssl sometimes hangs during startup when determining the
supported curves using -connect b) a pattern was missing to detect
whether the curve was not supported which falsely labeled all supplied curves
as supported when using /usr/bin/openssl . The pattern for the latter
was added (b). For a) there needs to be a follow up PR to avoid the
long delays.
Three changes:
- added grep for "EV TLS" in addition to "EV SSL", as some issuers are
using this. This grep link actually picks-up most EV policies.
- Added policy detection for 2.23.140.1.1. This is from CA Browser
Forum https://cabforum.org/resources/object-registry/ extended-validation(1).
- Added policy detection for 1.3.6.1.4.1.38064.1.3.1.4 , which is SSL.com's EV policy.
