
jms1 at jms1
Feb 17, 2009, 4:03 AM
Post #3 of 11
(3284 views)
Permalink
|
|
Re: netqmail 1.06 + ucspi-tls problems (sslserver segfault)
[In reply to]
|
|
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 2009-02-13, at 1703, Brad Faler wrote: > > I am attempting to get SSL/TLS support up and running using ucspi-ssl > 0.70 and the ucspi-tls patch, and while mail seems to be flowing I am > getting segfaults similiar to the one below off of sslserver > (presumably?) at the end of every connection (both SMTP and POP3). > If I > modify the smtp or pop daemon to run under tcpserver rather than > sslserver the problem goes away, but I would very much like to be able > to support SSL/TLS on this system when it goes live.. i'm also using sslserver 0.70, and i'm not having any problems with it. it allows users to connect to port 465, do their SSL negotiations, and then start the SMTP conversation. i haven't had any problems with it, and i've been using it for a few years now. i'm also using a patch which adds support for TLS into qmail-smtpd. the patch is the one many people have been using for years, most recently updated by dr. hoffman, and included as part of my combined patch. i've added a few configuration items (controlled by environment variables) so the administrator can require, allow, or deny TLS for a particular service (or client IP, if you change the variables in the tcpserver access control file.) setting up support for SSL, and setting up support for TLS, are two totally different things. it sounds like you have combined the ideas in your head, but you really need to consider each thing separately. the SSL and STARTTLS mechanisms work totally differently, and are handled by totally different programs. > from syslog (the segfault in question): > > sandbox kernel: [1107554.043727] sslserver[26877]: segfault at > 5f4c5353 > eip b7e1b34a esp bfd0c838 error 4 look at the value 5f4c5353. this is not a memory location one would normally expect to see in use, at least not on any system for which i've ever written and debugged code. it IS, however, four bytes' worth of ASCII values, namely "SSL_" (assuming you're using a little-endian processor, such as the intel x86 series or one of its clones.) a segfault happens when a process tries to access a memory address for which the kernel hasn't mapped any physical memory. on most systems, a process's memory usage would need to be on the order of 1.5GB in order for such an address to actually have any memory mapped into it. my guess is that either the patch you're using is broken, or there was a failed hunk when you applied it, you didn't notice it (or ignored it) and the resulting code (with most of the patch in there) just happened to compile without any errors. > @400000004995e1d12d5cdc94 sslserver: end 26877 status 11 the "status 11" tells you that it segfaulted. read "man 7 signal", i suspect you will find that on your system, SIGSEGV is number 11. - ---------------------------------------------------------------- | John M. Simpson --- KG4ZOW --- Programmer At Large | | http://www.jms1.net/ <jms1 [at] jms1> | - ---------------------------------------------------------------- | http://video.google.com/videoplay?docid=-1656880303867390173 | - ---------------------------------------------------------------- -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.8 (Darwin) iEYEARECAAYFAkmap3QACgkQj42MmpAUrRqiBACcCPHqn+JSfRFXCv36JpoW6/IF o48AoNPKNYqfxlc/8NMfhs1us2isq5WX =qVoM -----END PGP SIGNATURE-----
|