Hi Sasha, Could you post you sources somewhere?
If not, you could look at https://github.com/ellzey/libevhtp/blob/master/evhtp.c from libevhtp project. On Fri, Jan 25, 2013 at 6:47 AM, Sasha Limanov <szlima...@gmail.com> wrote: > Greetings all, > > First I apologize for my previous failure to finish typing the message > before clicking send. > > I have read through Diwaker's and Nick's past email relating to > bufferevent_openssl, but can't find the answer I'm looking for - if > it's been answered before, please point me to it. > > What is the proper way to shutdown a bufferevent_openssl filter? > Specifically in the case where my app is a server, and the client > suddenly looses connectivity, thus no proper ssl shutdown handshake. > > I have followed the lazy shutdown recommendation at > http://www.wangafu.net/~nickm/libevent-book/Ref6a_advanced_bufferevents.html > but I am still having trouble. > > Specifically, I get a SIGSEGV after the bufferevent ssl filter has > been freed, and the SSL *ctx freed also. The bufferevent_free() call > is made from within the event callback that signals BEV_EVENT_ERROR. > The bufferevent filter itself is created with > BUFFEREVENT_SSL_ACCEPTING and without BEV_OPT_DEFER_CALLBACKS > (although enabling this option doesn't help avoid the SEGV). > > #0 SSL_get_error (s=0x8071758, i=0) at ssl_lib.c:2366 > #1 0x00134085 in do_read (bev_ssl=0x8073290, n_to_read=<value > optimized out>) at bufferevent_openssl.c:598 > #2 0x00134701 in consider_reading (bev_ssl=0x8073290) at > bufferevent_openssl.c:783 > #3 0x00134982 in be_openssl_enable (bev=0x8073290, events=2) at > bufferevent_openssl.c:1147 > #4 0x0014d1b1 in bufferevent_unsuspend_read (bufev=0x8073290, > what=16) at bufferevent.c:85 > #5 0x001508a8 in be_filter_destruct (bev=0x80739f8) at bufferevent_filter.c:239 > #6 0x0014de91 in _bufferevent_decref_and_unlock (bufev=0x80739f8) at > bufferevent.c:622 > #7 0x0015031e in be_filter_eventcb (underlying=0x8073290, what=32, > _me=0x80739f8) at bufferevent_filter.c:467 > #8 0x0014e986 in _bufferevent_run_eventcb (bufev=0x8073290, what=32) > at bufferevent.c:267 > #9 0x001328de in be_openssl_eventcb (bev_base=0x80701b0, what=17, > ctx=0x8073290) at bufferevent_openssl.c:926 > #10 0x0014e986 in _bufferevent_run_eventcb (bufev=0x80701b0, what=17) > at bufferevent.c:267 > #11 0x0014f397 in bufferevent_readcb (fd=16, event=2, arg=0x80701b0) > at bufferevent_sock.c:195 > #12 0x001465e8 in event_process_active_single_queue (base=0x805df58, > flags=0) at event.c:1350 > #13 event_process_active (base=0x805df58, flags=0) at event.c:1420 > #14 event_base_loop (base=0x805df58, flags=0) at event.c:1621 > #15 0x00147216 in event_base_dispatch (event_base=0x805df58) at event.c:1450 > #16 0x0804c188 in main (argc=3, argv=0xbffff5c4) at main.c:155 > > It seems that the event dispatcher loop does not acknowledge that the > bev has been freed. Is it not correct to bufferevent_free() from > within the event callback? > > The event callback I use does the following: > > static void event_error_cb(struct bufferevent *buffev, short what, void *_arg) > { > if (what == BEV_EVENT_ERROR) { > struct bufferevent *underlying = bufferevent_get_underlying(bev); > SSL *ssl= bufferevent_openssl_get_ssl(bev); > int state = SSL_get_shutdown(ssl); > > if (state == 0) SSL_set_shutdown(ssl, SSL_RECEIVED_SHUTDOWN); > SSL_shutdown(ssl); > bufferevent_free(bev); > SSL_free(ssl); > > // this frees the stack of filters below "underlying" > cleanup_ev_stack(underlying); > > // when this cb handler returns, all the bufferevents > // have been freed > } > } > > Thank you > Sasha > *********************************************************************** > To unsubscribe, send an e-mail to majord...@freehaven.net with > unsubscribe libevent-users in the body. -- Azat Khuzhin
