Re: Assigning to a boost::asio::ssl::context while asynchronous operations are using it

From: Dmitrij V (dmxvlx_at_[hidden])
Date: 2019-09-30 02:47:39

Dear Martijn Otto,
Please, if it is not hard for you: make PR with documentation the
behaviour for the context.

--
regards
2019-09-29 22:11 GMT+05:00, Martijn Otto via Boost-users
<boost-users_at_[hidden]>:
> On Sat, 2019-09-28 at 07:31 +0500, Dmitrij V via Boost-users wrote:
>> Martijn Otto wrote:
>> > I could make a pull-request to implement these changes, if so
>> > desired.
>>
>> +1, changes + documentation on its, please :)
>>
>> > Does this have any change of getting merged?
>>
>> Oh, that is not in my authority, but please, send the PR into
>> https://github.com/chriskohlhoff/asio too
>>
>> --
>> regards
>> _______________________________________________
>> Boost-users mailing list
>> Boost-users_at_[hidden]
>> https://lists.boost.org/mailman/listinfo.cgi/boost-users
>
> After reading through the source code some more, it seems that no
> changes are actually necessary, since my initial code was already safe.
> I'll explain why.
>
> The ssl::stream constructor, which does get a reference to the
> ssl::context and extracts the underlying SSL_CTX but does not keep this
> in a member. Instead, it forwards it to its core_ member, which is an
> instance of detail::stream_core. detail::stream_core then forwards it
> to its engine_ member, which is an instance of detail::engine.
>
> detail::engine then uses it to call SSL_new. The documentation for
> SSL_new does not explicitly mention it, but looking through the code I
> can see it call SSL_CTX_up_ref(), which increases the reference count
> on the underlying SSL_CTX, which means that even if the ssl_context
> calls SSL_CTX_free(), the context will remain valid if there is an
> ssl::stream using it.
>
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> https://lists.boost.org/mailman/listinfo.cgi/boost-users
>