Jo, > > No, you will have to upgrade the Petr Rehor's milter helper > > to 1.1.3. It is promised to be out in few days. This upgrade is > > needed anyway to make it work again under FreeBSD and OpenBSD, > > so it was not worth providing special compatibility measures. > > Eh? This isn't mentioned in the release notes.
You mean the *BSD issue? It has nothing to do with amavisd-new, it is (was?) strictly a amavisd-milter issue. > Also, amavisd 1.4.3 is running just fine under FreeBSD with the 1.1.2 > milter right now. It's only that the headers aren't being added. > What is it that is believed not to work with FreeBSD? The last time I checked, it was crashing on FreeBSD because MAXAMABUF was increased in some version upgrade and is now exceeding per-thread stack size. It may have been fixed in ports by now, don't know. Petr said the fix was commited in CVS some time ago, but there was not yet a new release since then. I didn't investigate, so my information may be out of date by now. Below is a part of discussion on the issue: ======== I came across this same problem as reported by Joe Tylczak: > k4OGnG0I026718: HEADER: To: [EMAIL PROTECTED] > k4OGnG0I026718: HEADER: Subject: test 3 > k4OGnG0I026718: END OF HEADERS > k4OGnG0I026718: body chunk: 5 > Program received signal SIGSEGV, Segmentation fault. > mlfi_eom (ctx=0x0) at mlfi.c:647 > 647 struct mlfiCtx *mlfi = MLFICTX(ctx); although in my case it was on FreeBSD 5.4. Reducing MAXAMABUF to 48k just shifts the problem elsewhere, 20k seems low enough to get it running. The cause seems to be the same, namely the thread stack size defaults to 64kB, unless overridden by an explicit call to pthread_attr_setstacksize. There are no global settings and ulimit -s has no effect on thread stack size (unlike on Linux). So the increase of MAXAMABUF to 64k (used to allocate two such buffers) is a direct reason why amavisd-milter crashes on FreeBSD 5.4 and on OpenBSD 3.8. Apparently NetBSD and FreeBSD 6.x use larger default values for thread stack size. The solution seems awkward, as thread initialization is under control of milter library. Seems a solution like the one used for milter-greylist is needed, or give up a need for such large buffers allocated on a stack (allocating them via malloc on a heap would probably be another solution). See: http://lists.freebsd.org/pipermail/freebsd-ports-bugs/2004-May/033728.html http://mail.zope.org/pipermail/zope/2002-October/124478.html Mark ------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys -- and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV _______________________________________________ AMaViS-user mailing list AMaViS-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/amavis-user AMaViS-FAQ:http://www.amavis.org/amavis-faq.php3 AMaViS-HowTos:http://www.amavis.org/howto/
