openssl: add support for CN / hostname verification #980
Conversation
This adds the standard OpenSSL api to get a pointer to the SSL struct's X509_VERIFY_PARAM. We need this for the OpenSSL api to set the peer hostname introduced in the next patch.
This lets the user code set the mbedtls hostname using the standard OpenSSL X509_VERIFY_PARAM_set1_host() API semantics. The API takes an X509_VERIFY_PARAM pointer. We use the fact that is a composed member of the SSL struct to derive the SSL pointer. The X509_VERIFY_PARAM_set1_host() is unusual in that it can accept a NUL terminated C string as usual, or a nonterminated pointer + length. This implementation converts the latter to the former if given, before using it. This is enough for user code to get the openssl wrapper to make mbedtls confirm the CN on the peer cert belongs to the hostname used to reach it, by doing, eg X509_VERIFY_PARAM_set1_host(SSL_get0_param(myssl), myhostname, 0);
This defines the OpenSSL X509_CHECK_FLAG_...s and the set/clear accessors. Since none of them are supported, the set / clear accessor currently always does nothing and returns error. This call is often part of the generic openssl user code to set up certificate verification. This patch allows it to compile for ESP32 and decide at runtime what to do about unsupported flags.
|
Thanks for sending this, @lws-team . Sorry noone has gotten back to you earlier. Changes LGTM. I've asked the maintainer of the openssl compatibility layer to take a look as well. |
|
I have a bunch more fixes on the wrapper too that are necessary for proper operation with nonblocking... basically we use the wrapper inside lws for mbedTLS compatibility. What is "LGTM"? |
Great. FWIW, there are some similar changes (openssl non-blocking support) pending internally. Unsure of their ETA.
Looks Good To Me. :) |
|
Well, then I'll leave the rest until we see what the author has updated himself. |
|
... any eta on my patch getting looked at by "the maintainer of the openssl compatibility layer"? |
|
Sorry for the (repeated) extended silence. I'm not sure what happened about this, but I'm going to chase it up again. |
|
As I said there were a bunch of problems with the esp-idf wrapper (some of them serious data corruption). I appreciate the work done on it though, it helped us a lot and beers are owed to the original author. I can see it's kind of a hopeless / pointless open-ended task for him trying to cover every openssl api variation. Since this is not an effective upstream for it (well... it isn't, is it), and so there's no way to cooperate to improve it, I forked a copy of it into our tree a while back and gave up trying to produce nice patches on it; where possible I add stuff using native mbedtls apis directly. That is a shame. |
This adds the standard OpenSSL api to get a pointer to the SSL struct's X509_VERIFY_PARAM. We need this for the OpenSSL api to set the peer hostname introduced in the next patch. Part of #980
This lets the user code set the mbedtls hostname using the standard OpenSSL X509_VERIFY_PARAM_set1_host() API semantics. The API takes an X509_VERIFY_PARAM pointer. We use the fact that is a composed member of the SSL struct to derive the SSL pointer. The X509_VERIFY_PARAM_set1_host() is unusual in that it can accept a NUL terminated C string as usual, or a nonterminated pointer + length. This implementation converts the latter to the former if given, before using it. This is enough for user code to get the openssl wrapper to make mbedtls confirm the CN on the peer cert belongs to the hostname used to reach it, by doing, eg X509_VERIFY_PARAM_set1_host(SSL_get0_param(myssl), myhostname, 0); Merges #980
This defines the OpenSSL X509_CHECK_FLAG_...s and the set/clear accessors. Since none of them are supported, the set / clear accessor currently always does nothing and returns error. This call is often part of the generic openssl user code to set up certificate verification. This patch allows it to compile for ESP32 and decide at runtime what to do about unsupported flags. Merges #980
|
Hi @lws-team , After the long delay we ended up merging these with some minor whitespace/formatting cleanup. I appreciate that this experience probably doesn't make you want to send any more PRs against the openssl wrapper code. For what it's worth if you change your mind then I'll push hard to get any future PRs looked at quickly without this kind of long delay. Angus |
This series of three patches adds support to the openssl wrapper for OpenSSL APIs related to setting the expected hostname in the CN field of the peer cert, using mbedtls
mbedtls_ssl_set_hostname(). That cannot be done at the moment because the necessary mbedtls ssl ctx is private to the wrapper. This makes it available to the user using only additional related OpenSSL apis.