Fix nonce-count generation in Curl_auth_create_digest_http_message()

[mkhon]

• on the first invocation: keep security context returned by
  InitializeSecurityContext()
• on subsequent invocations: use MakeSignature() instead of
  InitializeSecurityContext() to generate HTTP digest response

Fixes #870

[mention-bot]

@mkhon, thanks for your PR! By analyzing the history of the files in this pull request, we identified @bagder, @captain-caveman2k and @yangtse to be potential reviewers.

[bagder]

(The test 1903 failure is probably unrelated as it seems to do this intermittently.)

[MarcelRaad]

Looks good and works for me.

[jay]

This looks good however why is the have_context variable needed? If context is set then you have context, and if it's not then you don't right? It seems to me we'd want to del context in cleanup even if have_context was not set, but when would it happen that there is context and it's not set?

edit: nm I see what you did, typically for that I'd expect context to be a pointer ie CtxtHandle *context. I have started some changes for that and a test.

[jay]

I've just landed this with a test and modified your code slightly.

I got rid of have_context by using a pointer to the http context that is NULL when there is no context. Also I renamed it as http_context so it is not accidentally confused in the future with the SASL/md5 version of context of the function in the same file. Further, I moved the MakeSignature block to before a new context is created so that if MakeSignature fails we delete the context and fall back on creating a new context. Your code is otherwise the same. So basically it is this now:

if(http_context) {
do makesignature, if it fails verbose warn and delete http_context and set to null
}
if(!http_context) {
create new http_context
}

Thanks!

[mkhon]

@jay I am not sure this Curl_safefree() is ok:

resp = malloc(output_token_len + 1);
if(!resp) {
free(output_token);

Curl_safefree(digest->http_context);

return CURLE_OUT_OF_MEMORY;

}

I think if the context is ok it should be kept for the subsequent invocations.

[jay]

I think if the context is ok it should be kept for the subsequent invocations.

Freeing the context there isn't needed but not for the reason you think. When CURLE_OUT_OF_MEMORY the connection is cleaned up, including digests, so that cleans up the valid context. I removed it in af5fbb1.

[bagder]

This seems to have caused #1276

[jay]

This seems to have caused #1276

Fixed in f4739f6