Re: [asio-users] SSL: only first async_write data goes to channel, subsequent goes nowhere.

[Andrey K. <a.k...@ne...>]

And here is a programm attached.

On 2015-04-27 20:42, Andrey Kulikov wrote:
> Hello,
> 
> While developing a simple program, what send series of messages via
> SSL over network using boost::asio, I faced a problem:
> Only first message goes to channel, and others stays somewhere, while
> async_write() handler was called.
> 
> There is a small programm in attachment, illustrates the issue.
> In order to use, first start opnessl s_server, allowing NULL 
> encryption:
> 
> openssl s_server -WWW  -tls1  -accept 443  -state   -cert server.crt
> -key server.key -cipher "eNULL"
> 
> Then compile the porgramm like that:
> c++ -pedantic -ansi -std=c++11 -Wall -Werror ggdb3 -O0 -DDEBUG  -o
> aio_ssl aio_ssl.cpp -L/usr/lib -lstdc++ -lpthread -lboost_thread
> -lboost_system -lcrypto -lssl
> and run, pointing to listening OpenSSL s_server:
> ./aio_ssl 127.0.0.1 443
> 
> Issue reproduceы on Debian and on Windows x86. With boost 1.49, 1.57
> and 1.58 everyting looks the same.
> 
> Programm establish communication with SSL server, and send series of
> messages (sime pieces of text) here.
> All messages reported to be sent sucessfully (async_write handlers
> where called).
> But, Wireshark reports that only first string goes to channel (here
> NULL encryption can serve us - we may see what really goes inside
> SSL).
> 
> Could anyone enlighten me, ignoramus, what I'm doing wrong?
> 
> Diging little bit deeper, I found where it probably goes wrong:
> 
> /usr/include/boost/asio/ssl/detail/io.hpp:179
> 
>         case engine::want_output_and_retry:
>         case engine::want_output:
> 
>           // The engine wants some data to be written to the output. 
> However, we
>           // cannot allow more than one write operation at a time on 
> the
>           // underlying transport. The pending_write_ timer's expiry 
> is set to
>           // pos_infin if a write is in progress, and neg_infin 
> otherwise.
>           if (core_.pending_write_.expires_at() == 
> boost::posix_time::neg_infin)
>           {
>             // Prevent other write operations from being started.
>             
> core_.pending_write_.expires_at(boost::posix_time::pos_infin);
> 
>             // Start writing all the data to the underlying transport.
>             boost::asio::async_write(next_layer_,
>                 core_.engine_.get_output(core_.output_buffer_),
>                 BOOST_ASIO_MOVE_CAST(io_op)(*this));
>           }
>           else
>           {
>             // Wait until the current write operation completes.
>             
> core_.pending_write_.async_wait(BOOST_ASIO_MOVE_CAST(io_op)(*this));
>           }
> 
> First message goes to if{}, and all subsequens messages goes to 
> else{}.
> One write operation still active, and other postponed till... some 
> event.
> But it never happens, and data not being send at all.
> 
> Does anone have any idea how to fix this situation?
> 
> Best wishes,
> Andrey