First implemented and tested working is decode_https_rr_alpn().
Also we use the svk params in a case statement to decipher the
hexstream better.
The hexstream ($line) has now no blanks anymore. They seem to be
arbitrary.
Variables need to be declared in get_https_rrecord() .
- quote vars (hoping it'll resolve the Mac runner issue)
- make sure CNAMEs are properly parsed
- end get_https_rrecord() earlier when there's no record but DNS binaries are "HTTPS record aware"
- while loop was redundant
- better comments
Elsewhere:
make sure get_https_rrecord is called with a trailing dot for the NODE
as there is an inexplicable difference between a real Mac
which passes the run and the one in github
-"DNS_HTTPS_rrecord","testssl.sh/81.169.235.32","443","OK","81.169.235.32","",""
+"DNS_HTTPS_rrecord","testssl.sh/81.169.235.32","443","OK","1 . alpn='h2'","",""
The first line comes from the runner
This is a fresh start for #2484 as the PR wasn't ready yet for 3.2 by the time it was released. And it continues #2866
which was kind of messed up by accident.
The info for the HTTPS RR shows up in the very beginning, i.e. in `service_detection()`. All keys are listed now in bold, values in a regular font.
`get_https_rrecord()` was introduced by copying and modifying `get_caa_rr_record()`.
There's a similar obstacle as with CAA RRs: older binaries show the resource records binary encoded. Thus a new set of global vars is introduced HAS_*_HTTPS which check whether the binaries support decoding the RR directly. As of now raw decoding doesn't work completely.
Todo:
- Add logic in QUIC
- if RR is detected and not QUIC is possible
- add time for QUIC detection when RR is retrieved
- show full HTTPS RR record, at least when having a new DNS client
- coninue with raw decoding, if possible (otherwise problematic for MacOS)
- shorten the comments in `get_https_rrecord()`
- man page
- when ASSUME_HTTP is set and no services was detected: this needs to be handled
- The placement of the output should be reconsidered and/or cached when multiple IPs belong to a FQDN
... which warns also via file output when not recommended command
line options are used.
This function named issue_cmdline_warnings() is being called in
lets roll after all fileout() functions has been initialized.
It needs to make use of fileout_insert_warning() though because
otherwise the JSON output is not correct.
Besides the previoulsy introduced warning when scanning IP addresses,
warnings of usage of '--fast' and '--ssl-native' will end up also
in a file now which gives ther tools using the machine readable
output to detect bad scan conditions.
Also warnings when scanning the most known IPv4 addresses
from Cloudflare, Google and Quad9, are avoided.
* move message when scanning IP address to the very beginning, inside parse_cmd_line()
* improve message
* just check whether there are no chars a-zA-Z
* move [[ $caa_node =~ '.'$ ]] || caa_node+="." into the while loop
- Skip CAA lookup entirely when NODE is an IP address; show
"not checked (IP address scan)" instead of spuriously querying
IP octets as domain labels and reporting "not offered"
- Force FQDN (trailing dot) on the initial caa_node before the
walk loop so dig does not apply the resolv.conf search domain
to the first query, which could return a false result
- Add a visible warning in the scan header when scanning by IP
address, noting that trust/CAA and other domain-specific checks
may be unreliable and the user should rescan with the hostname
went through a couple of pcap files and determined ja3 + ja4 sums.
- Android 15/16 are the same (previously ja3 taken instead of ja4 and wrong host. One has to use chrome !)
- Edge 101/Chrome 101 are the same (will be deprated next time)
- surprisingly Java 17.0.3 and 21.0.6 were the same.
- Added: Ja3/ja4 for old Apple Mail and Thunderbird
As mentioned in the comment: For Androids ja3 is is not unique, probably because of GREASE.
One can add two handshakes after another and they are different. ja4 seems more consistent here.
This should be kept in mind for all clients "supplying some grease"
The trailing error messages were swapped in the paragraphs /
description for MAX_SOCKET_FAIL + MAX_OSSL_FAIL .
This fixes the confusion for 3.3dev , see #3028 .
This fixes#3003 .
The conversion to proper UTF-8 should have taken place by just using
`-nameopt RFC2253`, see manpage openssl-namedisplay-options(1ssl).
As @dcooper16 suggested removing esc_msb should help. This may look
counterintuitive but works.
This commit changes the way that TLS 1.3 ciphers are identified by the OpenSSL names. To the degree possible, rather than checking for prefixes that have historically been used in various versions of OpenSSL and LibreSSL, the cipher name being checked against the known list of TLS 1.3 cipher suites that $OPENSSL supports.
In the few places in which the cipher suite name to be checked may not be supported by $OPENSSL, a check for the prefix "TLS_" is also used.
This commit fixes a few places where new signature schemes were not added:
1) It adds ECDSA with the Brainpool curves for TLS 1.3 (0x081a, 0x081b, and 0x081c) to get_server_certificate(), certificate_transparency(), and prepare_tls_clienthello().
2) It adds rsa_pss_pss (0x0809, 0x080a, 0x080b) to certificate_transparency().
3) It adds the signature schemes for EdDSA (0x0807, 0x0808) and ML-DSA (0x0904, 0x0905, 0x0906) to certificate_transparency().
When $OPENSSL s_client supports the "-sigalgs" option, get_server_certificate() uses $OPENSSL rather than tls_sockets() to obtain the server's certificate, but only for certificates with RSA and ECDSA public keys.
With OpenSSL 3.5 and newer the list command can be used to get a list of supported TLS signature algorithms.
With this commit, if OpenSSL 3.5 or newer is being used, the list of supported TLS signature algorithms is obtained and get_server_certificate() uses $OPENSSL s_client rather than tls_sockets() whenever $OPENSSL supports the relevant signature scheme.
In addition to making the code a bit faster, this may be helpful if a server has a certificate with an SM2 public key and it only supports curveSM2 for key exchange, since tls_sockets() can not decrypt server responses if curveSM2 is used.
It seems that OpenSSL 4.0.0 allows for the possibility that a server's response to the status request extension may include more than one OCSP response (presumably one for each certificate in the certification path).
As a result, the line indicating that the server does not provide status information was changed from "OCSP response: no response sent" to "OCSP responses: no responses sent". If a response was included, "OCSP responses:" is followed by an indication of the number of responses included.
This commit addresses the change from "response" to "responses".
I do not know of any servers that provide more than one OCSP response, so I have not tried to make any changes to handle more than one response.
Dirk Wetter