See #1258
To do:
* more robustness. At least the success value from the response need to be retrieved and checked via starttls_io().
* double check the pre-handshake before the OID whether it's correct for every case
* documentation
* inline help
It seems to work though against db.debian.org
In run_http_header() the GET command is first sent over TLS using a background process, and then if that does not hang, it is sent again in the foreground. Similarly, service_detection() runs the command in the background.
This commit changes determine_optimal_proto() to follow the example of run_http_header() as protection against the possibility of the HTTP query stalling.
This commit fixes determine_optimal_proto() and run_server_defaults() so that a "Local problem" is reported if a $URL_PATH is specified, the server is TLS 1.3-only, and $OPENSSL does not support -enable_pha (and the server does not offer client authentication as part of the initial TLS handshake).
Checking for client authentication with TLS 1.3 requires post-handshake authentication, which does not appear to be supported by LibreSSL. This commit improves the check for client authentication when testing a TLS 1.3 server using LibreSSL by having determine_optimal_proto() first test for connectivity with TLS 1.3 without checking for client authentication and then performing a separate check for client authentication using a non-TLS 1.3 protocol.
This commit only affects the flow of the program if a $URL_PATH is specified, the server supports TLS 1.3, and $OPENSSL supports TLS 1.3 but not -enable_pha.
testss.sh may still provide incorrect information about client authentication if a $URL_PATH is provided, the server is TLS 1.3-only, and LibreSSL is used.
Based on initial testing, this commit improves the check for client authentication in the case that the server only requests client authentication for specific URLs. However, it does not work correctly if the server supports TLS 1.3 and $OPENSSL is a version of LibreSSL that supports TLS 1.3 in s_client. The problem is that LibreSSL does not support post-handshake authentication with TLS 1.3
As noted in #1709, some servers will only request client authentication if a specific URL is requested. This commit modifies the check for client authentication, in the case that a $URL_PATH is provided, by having testssl.sh perform a GET request on the URL provided on the command line.
As discussed in #2079, this commit adds "</dev/null" to calls "$OPENSSL s_client" that are intended to test whether a certain option is supported. This is done to prevent hanging in the case that the option is supported, a TLS server happens to be listening on the port to which s_client tries to connect, and the connection is successful.
This PR also adds a new helper function, sclient_supported(), which is called from locally_supported() and run_prototest_openssl(). The helper function makes use of the already defined variables $HAS_SSL2, $HAS_SSL3, and $HAS_TLS13 in order to avoid calling "$OPENSSL s_client" when it has already been determined whether $OPENSSL supports the specified protocol.
... for all forward + non-mdns lookups.
This might help to avoid supplying domain names to local hosts (they would need
to have a valid certificate for the short DNS name then).
Fixes#2077
* address my comments
* add json fields HTTP_headerTime + HTTP_headerAge if they exists
* output HTTP_AGE if it was detected
* do stripping of line feeds closer to where variables were set
This commit fixes two issues with the headers in the structured JSON output. First, if run_cipher_match() is performed the header this is used is "pretest" rather than "singleCipher". Second, the headers for "serverPreferences" and "fs" are swapped.
Shellcheck complains on line 2234 that keyopts is referenced but not assigned and there is no explanation in the code (or in the documentation) why "$keyopts" is there. This commit adds a comment so that "$keyopts" isn't deleted as part of a code cleanup.
This commit adds several variables to reset_hostdepended_vars() that are not currently being reset for each host being tested, but that should be reset.
testssl.sh currently calls "git log --format='%h %ci' -1 2>/dev/null" three times. This commits changes testssl.sh to make this call just once and then use Bash string manipulation to extract the necessary information from the result.
Sending the entire log to /dev/null is a rather expensive way of checking whether
we are inside a git working tree. Use `git rev-parse --is-inside-work-tree` instead.
Dirk Wetter
dependabot[bot]
Emmanuel Bouthenot
David Cooper
Dan Church
Dirk Wetter
Dr. Matthias St. Pierre