-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
I wrote:
| Correcting my previous post, | | |>... one can see an EAGAIN on the -r socket of stunnel. | | | That's not true, the EAGAIN is on the accepted file descriptor. | | The problem is not solved, though ... | | Martin.
The problem was that between stunnel 3.22 and stunnel 3.26 the listening socket was set to non-blocking mode during the accept (which is ok). But at least on Solaris the session socket returned from accept inherits the non-blocking mode from the listening socket. On Linux the socket does not inherit the mode, resulting in a blocking socket.
The stunnel code seems not to handle non-blocking session sockets (EAGAIN), resulting in the problem on Solaris.
Please find below a patch restoring blocking mode for the session socket on all platforms. This solves my problem described this morning, but I am not 100% sure about any side effects.
Brian, please let me know if / when the patch will make it into a new stunnel-3 version.
=== cut here === - --- stunnel-3.26/stunnel.c Sat Aug 30 06:34:57 2003 +++ stunnel-3.26-mkn/stunnel.c Wed Sep 8 15:27:41 2004 @@ -264,6 +264,19 @@ ~ log(LOG_DEBUG, "%s accepted FD=%d from %s:%d", options.servname, s, ~ inet_ntoa(addr.sin_addr), ntohs(addr.sin_port)); ~ leave_critical_section(CRIT_NTOA); + +/* + * Martin.Kneissl@atosorigin.com: On Solaris the new socket s inherits + * the O_NONBLOCK from ls. This would cause EAGAIN in the read/writes of the + * socket s, later. + * EAGAIN is not handled here, so better have the socket in its original + * mode... + */ +#ifndef USE_WIN32 + if (old_val >= 0) + fcntl(s, F_SETFL, old_val); +#endif + ~ if(options.clients>=max_clients) { ~ log(LOG_WARNING, "Connection rejected: too many clients (>=%d)", ~ max_clients); === cut here ===
With kind regards,
Martin Kneissl
- -- Atos Worldline GmbH CRM and Telco
Pascalstrasse 19 52076 Aachen Germany
Phone: +49 (0) 2408 148 173 Fax: +49 (0) 2408 148 204 mailto:martin.kneissl@atosorigin.com www.atosworldline.de