> But as per the behavior of the Edge triggered epoll, > the epoll_wait will raise event only if it receives a data > after putting this in the read event of epoll every time. > Does this affect the behavior of SSL_read / SSL_write. If any > one implements the server using epoll ET can you please let me > know that there is any special case that needs to taken care > for SSL.
It works just fine. There are a few special cases, but they all come down to this: Any time you either make forward progress or have an indication that forward progress is possible, re-attempt *all* pending operations (read or write), not just the one you think should now be possible. The ugliest possible case is probably this one: 1) You call SSL_write, but negotiation is not finished (or re-negotiation is in progress). You get a 'want read' indication. Forward progress is not possible until some data is read. 2) The negotiation data arrives on the socket. 3) You call SSL_read and get a 'want read' indication because no application data is available, however, this processes the negotiation data. 4) You don't retry the SSL_write, because you got a 'want read' indication the last time you tried it and don't realize the 'SSL_read' made forward progress. An SSL_read call that returns 'want read', sadly, may have made forward progress. DS ______________________________________________________________________ OpenSSL Project http://www.openssl.org User Support Mailing List openssl-users@openssl.org Automated List Manager [EMAIL PROTECTED]
