Re: postfix-current broken on amd64 platform
On Mon, 2012-11-26 at 07:05:59 +1100, Peter Jeremy wrote: ... [postfix dies with a Protocol not supported when built in a jail without an IPv6 address] ... I've just bumped into this exact situation with mail/postfix28 and suspect that earlier postfix ports have the same issue. The above fix works on postfix28 and I would request that it be added to that port's patch list. Since this is a workaround for a FreeBSD-specific issue, I don't believe it's reasonable to expect Wietse to patch old postfix variants to work around it. OK, mail/postfix2[6-8] are now patched. Let me know if you have any problems. -- Sahil Tandon pgpJmhcIpv61k.pgp Description: PGP signature
Re: postfix-current broken on amd64 platform
Resurrecting an old thread because I've just run into this problem...
[postfix dies with a Protocol not supported when built in a jail
without an IPv6 address]
On 2011-Nov-17 15:41:12 -0500, Wietse Venema wie...@porcupine.org wrote:
A more proactive workaround would be to treat protocol not supported
as a non-fatal error, just like address family not supported is.
Please let me know if this works in the build cluster. It will log
a warning but that should be justfied, because the system does have
an abnormal configuration.
Wietse
*** src/util/inet_proto.c- Tue Jan 8 15:36:13 2008
--- src/util/inet_proto.c Thu Nov 17 08:49:31 2011
***
*** 219,225
pf-dns_atype_list = make_unsigned_vector(3, T_A, T_, 0);
pf-sa_family_list = make_uchar_vector(3, AF_INET, AF_INET6, 0);
break;
! } else if (errno == EAFNOSUPPORT) {
msg_warn(%s: IPv6 support is disabled: %m, context);
msg_warn(%s: configuring for IPv4 support only, context);
/* FALLTHROUGH */
--- 219,225
pf-dns_atype_list = make_unsigned_vector(3, T_A, T_, 0);
pf-sa_family_list = make_uchar_vector(3, AF_INET, AF_INET6, 0);
break;
! } else if (errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
msg_warn(%s: IPv6 support is disabled: %m, context);
msg_warn(%s: configuring for IPv4 support only, context);
/* FALLTHROUGH */
I've just bumped into this exact situation with mail/postfix28 and
suspect that earlier postfix ports have the same issue. The above fix
works on postfix28 and I would request that it be added to that port's
patch list. Since this is a workaround for a FreeBSD-specific issue,
I don't believe it's reasonable to expect Wietse to patch old postfix
variants to work around it.
--
Peter Jeremy
pgpDuvvvy2IWG.pgp
Description: PGP signature
Re: postfix-current broken on amd64 platform
Peter Jeremy: I've just bumped into this exact situation with mail/postfix28 and suspect that earlier postfix ports have the same issue. The above fix works on postfix28 and I would request that it be added to that port's patch list. Since this is a workaround for a FreeBSD-specific issue, I don't believe it's reasonable to expect Wietse to patch old postfix variants to work around it. Indeed. I rewrote this module for Postfix 2.9. If it deals with this problem, then I won't take further action. Wietse ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On Sun, 20 Nov 2011 01:36:20 -0500 Sahil Tandon articulated: This port has been updated to Postfix 2.9 Snapshot 2019. Thanks Sahil. :) -- Jerry ♔ Disclaimer: off-list followups get on-list replies or get ignored. Please do not ignore the Reply-To header. __ ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Sahil Tandon píše v so 19. 11. 2011 v 15:48 -0500: Wietse has made the change upstream and I plan to commit that in a little bit. Just running it through my tinderbox if folks don't mind the wait (a few hours). Not at all, and thanks both! -- -- Pav Lucistnik p...@oook.cz p...@freebsd.org ... the obese drugged penguin used by Linux. -- Scott Long signature.asc Description: This is a digitally signed message part
Re: postfix-current broken on amd64 platform
Wietse Venema píše v čt 17. 11. 2011 v 15:41 -0500:
Pav Lucistnik:
Wietse Venema p??e v ?t 17. 11. 2011 v 13:23 -0500:
Actually, a Postfix built-in default setting changed on 20110918.
It now enables IPv6 unless this is turned off in a configuration
file. I can fix that at compile time, and thereby not trigger
the error on build systems with unexpected IPv6 configurations.
This is interesting.
The build jails are configured to have only IPv4 address on lo0,
but the host have both IPv4 and IPv6 configured on its lo0.
Changing the jail configuration is possible but if a reasonable
workaround can be made in postfix-current port I'd prefer not to touch
pointyhat configuration (unexpected consequences and all that...)
I can competely fix this specific error instance with a one-line
config file change, but I don't want to end up in the same boat
again when some other program needs to be run during build time.
A more proactive workaround would be to treat protocol not supported
as a non-fatal error, just like address family not supported is.
Please let me know if this works in the build cluster. It will log
a warning but that should be justfied, because the system does have
an abnormal configuration.
Yes, this works on the build cluster.
Sahil, can I commit it to the port?
Wietse
*** src/util/inet_proto.c-Tue Jan 8 15:36:13 2008
--- src/util/inet_proto.c Thu Nov 17 08:49:31 2011
***
*** 219,225
pf-dns_atype_list = make_unsigned_vector(3, T_A, T_, 0);
pf-sa_family_list = make_uchar_vector(3, AF_INET, AF_INET6, 0);
break;
! } else if (errno == EAFNOSUPPORT) {
msg_warn(%s: IPv6 support is disabled: %m, context);
msg_warn(%s: configuring for IPv4 support only, context);
/* FALLTHROUGH */
--- 219,225
pf-dns_atype_list = make_unsigned_vector(3, T_A, T_, 0);
pf-sa_family_list = make_uchar_vector(3, AF_INET, AF_INET6, 0);
break;
! } else if (errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
msg_warn(%s: IPv6 support is disabled: %m, context);
msg_warn(%s: configuring for IPv4 support only, context);
/* FALLTHROUGH */
--
--
Pav Lucistnik p...@oook.cz
p...@freebsd.org
Why does the Earth have colors?
signature.asc
Description: This is a digitally signed message part
Re: postfix-current broken on amd64 platform
On Sat, 2011-11-19 at 19:42:02 +0100, Pav Lucistnik wrote: Wietse Venema píše v čt 17. 11. 2011 v 15:41 -0500: Pav Lucistnik: Wietse Venema p??e v ?t 17. 11. 2011 v 13:23 -0500: Actually, a Postfix built-in default setting changed on 20110918. It now enables IPv6 unless this is turned off in a configuration file. I can fix that at compile time, and thereby not trigger the error on build systems with unexpected IPv6 configurations. This is interesting. The build jails are configured to have only IPv4 address on lo0, but the host have both IPv4 and IPv6 configured on its lo0. Changing the jail configuration is possible but if a reasonable workaround can be made in postfix-current port I'd prefer not to touch pointyhat configuration (unexpected consequences and all that...) I can competely fix this specific error instance with a one-line config file change, but I don't want to end up in the same boat again when some other program needs to be run during build time. A more proactive workaround would be to treat protocol not supported as a non-fatal error, just like address family not supported is. Please let me know if this works in the build cluster. It will log a warning but that should be justfied, because the system does have an abnormal configuration. Yes, this works on the build cluster. Sahil, can I commit it to the port? Wietse has made the change upstream and I plan to commit that in a little bit. Just running it through my tinderbox if folks don't mind the wait (a few hours). -- Sahil Tandon pgprZl2FD4UW3.pgp Description: PGP signature
Re: postfix-current broken on amd64 platform
This port has been updated to Postfix 2.9 Snapshot 2019. -- Sahil Tandon pgpVTSiYHETTH.pgp Description: PGP signature
Re: postfix-current broken on amd64 platform
On Fri, 18 Nov 2011 03:48:49 -0800 per...@pluto.rain.com articulated: Pav Lucistnik p...@freebsd.org wrote: The build jails are configured to have only IPv4 address on lo0, but the host have both IPv4 and IPv6 configured on its lo0. Even disregarding RFC3513, is an IPv6-enabled kernel without an IPv6 address on lo0 a realistic configuration for a real FreeBSD system? If not, I'd think it worthwhile to make pointyhat more realistic. (Either way, it seems unobjectionable to improve the robustness of postfix, making it more liberal in what it accepts.) Changing the jail configuration is possible but if a reasonable workaround can be made in postfix-current port I'd prefer not to touch pointyhat configuration (unexpected consequences and all that...) It may be a bit late in the 9.0 release cycle to be messing with the pointyhat configuration, but an adjustment might be considered after 9.0-RELEASE is done. postfix-current -- 2.9.20111012,4 has been unmarked BROKEN on amd64 and builds without incident. I hope the actual current version, ie postfix-2.9-2017.tar.gz will be in the ports system soon. I would like to thank Sahil Tandon sa...@freebsd.org for his efforts in getting this valuable port problem corrected. Interestingly enough, and I cannot prove it to anyone's satisfaction, but I did mention to a colleague that if I posted on the ports forum regrading this problem and could get Wietse involved, the problem would be corrected within 24 hours. -- Jerry ♔ Disclaimer: off-list followups get on-list replies or get ignored. Please do not ignore the Reply-To header. __ ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On 18/11/2011 11:48, per...@pluto.rain.com wrote: Pav Lucistnikp...@freebsd.org wrote: The build jails are configured to have only IPv4 address on lo0, but the host have both IPv4 and IPv6 configured on its lo0. Even disregarding RFC3513, is an IPv6-enabled kernel without an IPv6 address on lo0 a realistic configuration for a real FreeBSD system? If not, I'd think it worthwhile to make pointyhat more realistic. (Either way, it seems unobjectionable to improve the robustness of postfix, making it more liberal in what it accepts.) FreeBSD jails only expose the IP addresses they are configured to use. So if you have an IPv6 enabled host (which is the default) with jails that are configured with only IPv4 addresses, then yes, it is very realistic. Furthermore, by default, jails don't have any v4 addresses assigned to the loopback interface - 127.0.0.1 and inaddr_loopback are automagically aliased to the primary v4 IP assigned to the jail. Jase. ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Doug Barton: On 11/16/2011 05:03, Wietse Venema wrote: I tend to believe that network-less build environments are not representative for the environment where an Internet MTA would run, but hey, what do I know. I think you're right that having a network makes the MTA more useful. :) However given the prevalence and popularity of pre-packaged software in the Unix world it's probably worth thinking about what parts of the build/configuration process need to be done at build time, and what parts should be done at run time. Not only do I work on the FreeBSD project generally, I also maintain a package building system that we use at $JOB to install packages on the various remote hosts. Having those packages do the right thing even if the package builder and the remote systems don't have identical configurations is a huge advantage. Yup, hence my request for the immensely helpful information about your build environment. Obviously, yours differs from mine; addressing this in a responsible manner requires that I can reproduce conditions that are currently unspecified (no network is too vague). Wietse ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Jase Thew: [ Charset ISO-8859-1 unsupported, converting... ] On 16/11/2011 23:30, Sahil Tandon wrote: On Wed, 2011-11-16 at 13:16:34 -0500, Wietse Venema wrote: [SNIP] It would be immensely helpful if run-time details could be made available including line of source code, and any system configuration details that are necessary to reproduce the condition. +1 I hope someone with access to the cluster machine can provide this information. As much as I've tried to mimic the cluster conditions myself, I am unable to reproduce the fatal socket error on any FreeBSD version. I've just managed to reproduce the fatal socket error locally. It can occur when IPv6 is enabled, but you don't have any IPv6 addresses configured on any interfaces. (Yes, having an IPv6 enabled interface with no addresses assigned is non-RFC3513 compliant, but it can and does occur). The cause (at least locally) is the socket call on line 214 in util/inet_proto.c which fails with errno 43 (EPROTONOSUPPORT - Protocol not supported). Thanks, this is very helpful. That code has not changed in 6 years, so we are looking at an incompatible change in build environment. Would it be better to make the build environment RFC-compliant? It is easy enough for me to work around a non-standard socket() result (*), but the same bug may trip up other packages in the future, especially packages that pay attention to errors. Sloppy programmers should never be rewarded. Wietse (*) Add a test for EPROTONOSUPPORT, to the existing test for EAFNOSUPPORT. ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On Thu, 17 Nov 2011 08:54:02 -0500 (EST) Wietse Venema articulated: Sloppy programmers should never be rewarded. I couldn't agree more fully. Unfortunately, they all too often are. -- Jerry ♔ Disclaimer: off-list followups get on-list replies or get ignored. Please do not ignore the Reply-To header. __ It is not a good omen when goldfish commit suicide. ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Wietse Venema: Jase Thew: It can occur when IPv6 is enabled, but you don't have any IPv6 addresses configured on any interfaces. (Yes, having an IPv6 enabled interface with no addresses assigned is non-RFC3513 compliant, but it can and does occur). ... Thanks, this is very helpful. That code has not changed in 6 years, so we are looking at an incompatible change in build environment. Actually, a Postfix built-in default setting changed on 20110918. It now enables IPv6 unless this is turned off in a configuration file. I can fix that at compile time, and thereby not trigger the error on build systems with unexpected IPv6 configurations. Wietse ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Wietse Venema píše v čt 17. 11. 2011 v 13:23 -0500: Wietse Venema: Jase Thew: It can occur when IPv6 is enabled, but you don't have any IPv6 addresses configured on any interfaces. (Yes, having an IPv6 enabled interface with no addresses assigned is non-RFC3513 compliant, but it can and does occur). ... Thanks, this is very helpful. That code has not changed in 6 years, so we are looking at an incompatible change in build environment. Actually, a Postfix built-in default setting changed on 20110918. It now enables IPv6 unless this is turned off in a configuration file. I can fix that at compile time, and thereby not trigger the error on build systems with unexpected IPv6 configurations. This is interesting. The build jails are configured to have only IPv4 address on lo0, but the host have both IPv4 and IPv6 configured on its lo0. Changing the jail configuration is possible but if a reasonable workaround can be made in postfix-current port I'd prefer not to touch pointyhat configuration (unexpected consequences and all that...) -- -- Pav Lucistnik p...@oook.cz p...@freebsd.org MIPS: Meaningless Information Provided by Salesmen signature.asc Description: This is a digitally signed message part
Re: postfix-current broken on amd64 platform
Pav Lucistnik p...@freebsd.org wrote: The build jails are configured to have only IPv4 address on lo0, but the host have both IPv4 and IPv6 configured on its lo0. Even disregarding RFC3513, is an IPv6-enabled kernel without an IPv6 address on lo0 a realistic configuration for a real FreeBSD system? If not, I'd think it worthwhile to make pointyhat more realistic. (Either way, it seems unobjectionable to improve the robustness of postfix, making it more liberal in what it accepts.) Changing the jail configuration is possible but if a reasonable workaround can be made in postfix-current port I'd prefer not to touch pointyhat configuration (unexpected consequences and all that...) It may be a bit late in the 9.0 release cycle to be messing with the pointyhat configuration, but an adjustment might be considered after 9.0-RELEASE is done. ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On 2011/11/16 02:56, Sahil Tandon wrote: On Tue, 2011-11-15 at 17:55:57 +0100, Pav Lucistnik wrote: Jase Thew píše v út 15. 11. 2011 v 16:31 +: What networking/DNS configuration is Pointyhat lacking (or have sufficiently different to break the socket code inside of postconf)? It is a purposefully no-networking sandbox jail. What networking activity postconf wants to run? Wietse, in a post[1] on the Postfix mailing list, lends further credence to a suspicion that this issue is particular to pointyhat: Postconf opens a socket to determine the mynetworks value (it determines the local interfaces and their netmasks). I have heard about bizarre errors on FreeBSD (jail) systems where the user-land library was out of sync with kernel-land, resulting in data structure mis-matches and system calls returning nonsensical results. [1] http://archives.neohapsis.com/archives/postfix/2011-11/0385.html That's entirely possible. Is it a clever idea to hardcode local interfaces on build machine into a package that will then be redistributed to other machines? Sounds like postfix will have to do without official packages on FreeBSD from now on. -- Pav Lucistnik p...@oook.cz p...@freebsd.org ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On Wed, Nov 16, 2011 at 10:28:03AM +0100, Pav Lucistnik wrote: Is it a clever idea to hardcode local interfaces on build machine into a package that will then be redistributed to other machines? Am I misunderstanding something, or is this a local configuration script that customizes the installation for the target machine in a similar way the mai/qmail port runs post-install 'config/config-fast' scripts to identify the system hostname (which then populates the 'me' file, for example)? -- Glen Barber pgpLdYXoIPYiY.pgp Description: PGP signature
Re: postfix-current broken on amd64 platform
On Wed, 16 Nov 2011 10:28:03 +0100 Pav Lucistnik articulated: On 2011/11/16 02:56, Sahil Tandon wrote: On Tue, 2011-11-15 at 17:55:57 +0100, Pav Lucistnik wrote: Jase Thew píše v út 15. 11. 2011 v 16:31 +: What networking/DNS configuration is Pointyhat lacking (or have sufficiently different to break the socket code inside of postconf)? It is a purposefully no-networking sandbox jail. What networking activity postconf wants to run? Wietse, in a post[1] on the Postfix mailing list, lends further credence to a suspicion that this issue is particular to pointyhat: Postconf opens a socket to determine the mynetworks value (it determines the local interfaces and their netmasks). I have heard about bizarre errors on FreeBSD (jail) systems where the user-land library was out of sync with kernel-land, resulting in data structure mis-matches and system calls returning nonsensical results. [1] http://archives.neohapsis.com/archives/postfix/2011-11/0385.html That's entirely possible. Is it a clever idea to hardcode local interfaces on build machine into a package that will then be redistributed to other machines? Sounds like postfix will have to do without official packages on FreeBSD from now on. I would find that unacceptable. If the problem is with FreeBSD as Wietse has indicated, then the problem should be rectified on the FreeBSD side of the equation. In any case, since Postfix does apparently build on FreeBSD with the exception of pointyhat, perhaps the BROKEN tag should be removed from the port and just a warning message displayed that pointyhat cannot build/install the port successfully but that the end users mileage may vary. At the very least, it would be a more honest approach to the problem. -- Jerry ♔ Disclaimer: off-list followups get on-list replies or get ignored. Please do not ignore the Reply-To header. __ ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Hi! Postconf opens a socket to determine the mynetworks value (it determines the local interfaces and their netmasks). [quote http://archives.neohapsis.com/archives/postfix/2011-11/0385.html] I have heard about bizarre errors on FreeBSD (jail) systems where the user-land library was out of sync with kernel-land, resulting in data structure mis-matches and system calls returning nonsensical results. Is it a clever idea to hardcode local interfaces on build machine into a package that will then be redistributed to other machines? No. Sounds like postfix will have to do without official packages on FreeBSD from now on. I would find that unacceptable. If the problem is with FreeBSD as Wietse has indicated, then the problem should be rectified on the FreeBSD side of the equation. Well, the problem is on both sides. On FreeBSD, one can have systems without any interface -- and postfix assumes that every system it is built on has some kind of interface. So, maybe it's time to change that assumption for the postfix built process, if possible ? -- p...@opsec.eu+49 171 3101372 9 years to go ! ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Kurt Jaeger: Hi! Postconf opens a socket to determine the mynetworks value (it determines the local interfaces and their netmasks). [quote http://archives.neohapsis.com/archives/postfix/2011-11/0385.html] I have heard about bizarre errors on FreeBSD (jail) systems where the user-land library was out of sync with kernel-land, resulting in data structure mis-matches and system calls returning nonsensical results. Is it a clever idea to hardcode local interfaces on build machine into a package that will then be redistributed to other machines? No. Postfix does none of that. Can someone please explain what socket call is failing, and what the reasons for that might be? I tend to believe that network-less build environments are not representative for the environment where an Internet MTA would run, but hey, what do I know. Wietse ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On 2011-11-16 10:28, Pav Lucistnik wrote:
On 2011/11/16 02:56, Sahil Tandon wrote:
On Tue, 2011-11-15 at 17:55:57 +0100, Pav Lucistnik wrote:
Jase Thew píše v út 15. 11. 2011 v 16:31 +:
What networking/DNS configuration is Pointyhat lacking (or have
sufficiently different to break the socket code inside of postconf)?
It is a purposefully no-networking sandbox jail. What networking
activity postconf wants to run?
Wietse, in a post[1] on the Postfix mailing list, lends further credence
to a suspicion that this issue is particular to pointyhat:
Postconf opens a socket to determine the mynetworks value (it
determines the local interfaces and their netmasks).
I have heard about bizarre errors on FreeBSD (jail) systems where the
user-land library was out of sync with kernel-land, resulting in data
structure mis-matches and system calls returning nonsensical results.
[1] http://archives.neohapsis.com/archives/postfix/2011-11/0385.html
That's entirely possible.
Is it a clever idea to hardcode local interfaces on build machine into a
package that will then be redistributed to other machines?
Sounds like postfix will have to do without official packages on FreeBSD from
now on.
What about the attached diff until we found the exact issue.
Unbreak the port, and use MANUAL_PACKAGE_BUILD instead.
Additional bump to latest postfix-current.
Since postfix-current should be current and not months behind I see no issue if
we
do not provide packages which are not updated regularly enough to reflect
current.
--
Regards,
olli
Index: Makefile
===
RCS file: /home/pcvs/ports/mail/postfix-current/Makefile,v
retrieving revision 1.263
diff -u -r1.263 Makefile
--- Makefile26 Oct 2011 22:44:50 - 1.263
+++ Makefile16 Nov 2011 13:11:03 -
@@ -6,7 +6,7 @@
#
PORTNAME= postfix
-DISTVERSION= 2.9-20111012
+DISTVERSION= 2.9-2013
PORTEPOCH= 4
CATEGORIES=mail ipv6
MASTER_SITES= ftp://ftp.porcupine.org/mirrors/postfix-release/experimental/ \
@@ -25,6 +25,7 @@
postfix-current-2.[0-8]* postfix-current-base-2.[0-8]* \
sendmail-8.* sendmail+*-8.* smail-3.* zmailer-2.*
+MANUAL_PACKAGE_BUILD= build OK in tinderbox, broken on pointyhat (issue not
identified)
.if !defined(BATCH) !defined(PACKAGE_BUILDING)
IS_INTERACTIVE=yes
.endif
@@ -92,10 +93,6 @@
HTML1= body_checks.5.html bounce.5.html postfix-power.png \
scache.8.html tlsmgr.8.html
-.if ${ARCH} == amd64
-BROKEN=fails during installation
-.endif
-
.if !defined(DEBUG)
MAKEFILEFLAGS+=DEBUG=
.endif
Index: distinfo
===
RCS file: /home/pcvs/ports/mail/postfix-current/distinfo,v
retrieving revision 1.196
diff -u -r1.196 distinfo
--- distinfo18 Oct 2011 01:50:41 - 1.196
+++ distinfo16 Nov 2011 13:11:03 -
@@ -1,2 +1,2 @@
-SHA256 (postfix/postfix-2.9-20111012.tar.gz) =
fb3d55c8a16c687bcee7e7a26a84c69e8765cc19ba7ab47620e7ac6b19efd3be
-SIZE (postfix/postfix-2.9-20111012.tar.gz) = 3673596
+SHA256 (postfix/postfix-2.9-2013.tar.gz) =
c8f493b9206b4ac52fdfdae72fee214d22d29206d353a6110188305019d906e3
+SIZE (postfix/postfix-2.9-2013.tar.gz) = 3685450
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Hi! Postconf opens a socket to determine the mynetworks value (it determines the local interfaces and their netmasks). [...] Postfix does none of that. The fbsd postfix-current port during the post-install script does something which fails on a jail-based build-cluster where the jail itself has no network connectivity. See this message for some analysis: http://lists.freebsd.org/pipermail/freebsd-ports/2011-November/071419.html And this message explains it: http://lists.freebsd.org/pipermail/freebsd-ports/2011-November/071421.html It says: -- It is a purposefully no-networking sandbox jail. What networking activity postconf wants to run? -- Can someone please explain what socket call is failing, and what the reasons for that might be? The reason is that postconf is called during post-install and fails because the build jail is without any interface. I tend to believe that network-less build environments are not representative for the environment where an Internet MTA would run, but hey, what do I know. The build environment seems to be network-less. It's not the environment where the package will run. -- p...@opsec.eu+49 171 3101372 9 years to go ! ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On 2011/11/16 14:18, olli hauer wrote: On 2011-11-16 10:28, Pav Lucistnik wrote: On 2011/11/16 02:56, Sahil Tandon wrote: On Tue, 2011-11-15 at 17:55:57 +0100, Pav Lucistnik wrote: Jase Thew píše v út 15. 11. 2011 v 16:31 +: What networking/DNS configuration is Pointyhat lacking (or have sufficiently different to break the socket code inside of postconf)? It is a purposefully no-networking sandbox jail. What networking activity postconf wants to run? Wietse, in a post[1] on the Postfix mailing list, lends further credence to a suspicion that this issue is particular to pointyhat: Postconf opens a socket to determine the mynetworks value (it determines the local interfaces and their netmasks). I have heard about bizarre errors on FreeBSD (jail) systems where the user-land library was out of sync with kernel-land, resulting in data structure mis-matches and system calls returning nonsensical results. [1] http://archives.neohapsis.com/archives/postfix/2011-11/0385.html That's entirely possible. Is it a clever idea to hardcode local interfaces on build machine into a package that will then be redistributed to other machines? Sounds like postfix will have to do without official packages on FreeBSD from now on. What about the attached diff until we found the exact issue. Unbreak the port, and use MANUAL_PACKAGE_BUILD instead. MANUAL_PACKAGE_BUILD is fine with me. -- Pav Lucistnik p...@oook.cz p...@freebsd.org ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Kurt Jaeger: Hi! Postconf opens a socket to determine the mynetworks value (it determines the local interfaces and their netmasks). [...] Postfix does none of that. The fbsd postfix-current port during the post-install script does something which fails on a jail-based build-cluster where the jail itself has no network connectivity. Would it be possible to give the jail a loopback interface? I don't mind if FreeBSD improves its build environment such that it breaks the build after 12 years. If adding a loopback interface fixes this, then we can all move on to more interesting things. Wietse ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On 16 Nov 2011 16:00, Wietse Venema wie...@porcupine.org wrote: Kurt Jaeger: Hi! Postconf opens a socket to determine the mynetworks value (it determines the local interfaces and their netmasks). [...] Postfix does none of that. The fbsd postfix-current port during the post-install script does something which fails on a jail-based build-cluster where the jail itself has no network connectivity. Would it be possible to give the jail a loopback interface? I don't mind if FreeBSD improves its build environment such that it breaks the build after 12 years. If adding a loopback interface fixes this, then we can all move on to more interesting things. Wietse Clamav uses loopback on its testing phase, and works fine on the build cluster. Chris ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On Nov 16, 2011, at 11:24 AM, Chris Rees utis...@gmail.com wrote: On 16 Nov 2011 16:00, Wietse Venema wie...@porcupine.org wrote: Kurt Jaeger: Hi! Postconf opens a socket to determine the mynetworks value (it determines the local interfaces and their netmasks). [...] Postfix does none of that. The fbsd postfix-current port during the post-install script does something which fails on a jail-based build-cluster where the jail itself has no network connectivity. Would it be possible to give the jail a loopback interface? I don't mind if FreeBSD improves its build environment such that it breaks the build after 12 years. If adding a loopback interface fixes this, then we can all move on to more interesting things. Wietse Clamav uses loopback on its testing phase, and works fine on the build cluster. What irks me is that all other Postfix versions (that also call postconf) have no problem building/installing on the cluster.___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Wietse: Would it be possible to give the jail a loopback interface? I don't mind if FreeBSD improves its build environment such that it breaks the build after 12 years. If adding a loopback interface fixes this, then we can all move on to more interesting things. On Nov 16, 2011, at 11:24 AM, Chris Rees utis...@gmail.com wrote: Clamav uses loopback on its testing phase, and works fine on the build cluster. Sahil Tandon: What irks me is that all other Postfix versions (that also call postconf) have no problem building/installing on the cluster. It would be immensely helpful if run-time details could be made available including line of source code, and any system configuration details that are necessary to reproduce the condition. If this turns out to be highly-classified information, then we can communicate off-list. My PGP key is linked off http://www.porcupine.org/wietse/. Wietse ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On Wed, 2011-11-16 at 13:16:34 -0500, Wietse Venema wrote: Wietse: Would it be possible to give the jail a loopback interface? I don't mind if FreeBSD improves its build environment such that it breaks the build after 12 years. If adding a loopback interface fixes this, then we can all move on to more interesting things. On Nov 16, 2011, at 11:24 AM, Chris Rees utis...@gmail.com wrote: Clamav uses loopback on its testing phase, and works fine on the build cluster. Sahil Tandon: What irks me is that all other Postfix versions (that also call postconf) have no problem building/installing on the cluster. It would be immensely helpful if run-time details could be made available including line of source code, and any system configuration details that are necessary to reproduce the condition. +1 I hope someone with access to the cluster machine can provide this information. As much as I've tried to mimic the cluster conditions myself, I am unable to reproduce the fatal socket error on any FreeBSD version. -- Sahil Tandon ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On 16/11/2011 23:30, Sahil Tandon wrote: On Wed, 2011-11-16 at 13:16:34 -0500, Wietse Venema wrote: [SNIP] It would be immensely helpful if run-time details could be made available including line of source code, and any system configuration details that are necessary to reproduce the condition. +1 I hope someone with access to the cluster machine can provide this information. As much as I've tried to mimic the cluster conditions myself, I am unable to reproduce the fatal socket error on any FreeBSD version. I've just managed to reproduce the fatal socket error locally. It can occur when IPv6 is enabled, but you don't have any IPv6 addresses configured on any interfaces. (Yes, having an IPv6 enabled interface with no addresses assigned is non-RFC3513 compliant, but it can and does occur). The cause (at least locally) is the socket call on line 214 in util/inet_proto.c which fails with errno 43 (EPROTONOSUPPORT - Protocol not supported). Regards, Jase. ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On 11/16/2011 05:03, Wietse Venema wrote: I tend to believe that network-less build environments are not representative for the environment where an Internet MTA would run, but hey, what do I know. I think you're right that having a network makes the MTA more useful. :) However given the prevalence and popularity of pre-packaged software in the Unix world it's probably worth thinking about what parts of the build/configuration process need to be done at build time, and what parts should be done at run time. Not only do I work on the FreeBSD project generally, I also maintain a package building system that we use at $JOB to install packages on the various remote hosts. Having those packages do the right thing even if the package builder and the remote systems don't have identical configurations is a huge advantage. Doug -- We could put the whole Internet into a book. Too practical. Breadth of IT experience, and depth of knowledge in the DNS. Yours for the right price. :) http://SupersetSolutions.com/ ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
It was marked broken on a particular architecture, hence the genesis of this
thread. I appreciate that diagnosis is difficult; perhaps Olli's suggestion is
helpful in isolating the issue. I do not know how else to troubleshoot since
these pointyhat errors are not - AFAIK - reproducible by others, on either i386
or amd64 platforms.
On Nov 15, 2011, at 4:45 AM, Pav Lucistnik p...@freebsd.org wrote:
1) The problem is not amd64 specific
2) No point unmarking BROKEN, it currently fails on i386 pointyhat nodes
too
3) Diagnosis is hard because the postfix-install shell script prints no
useful progress messages
Example failure log:
http://pointyhat.freebsd.org/errorlogs/i386-errorlogs/e.8.2003071512/postfix-current-2.9.20111012,4.log
Sahil Tandon píše v po 14. 11. 2011 v 21:24 -0500:
[ pav@ and those who tested mail/postifx-current on amd64 added to Cc: ]
On Mon, 2011-11-14 at 08:37:13 -0500, Jerry wrote:
The postfix-current port is still marked as broken:
.if ${ARCH} == amd64
BROKEN= fails during installation
.endif
Since all previous releases of Postfix worked on FreeBSD, and since
Postfix is/was developed on FreeBSD, I was wondering what the problem
is with this release. Is there a possibility that this phenomena might
be rectified in the near future?
Thanks for your report, Jerry. You are correct that FreeBSD is the main
development platform for Postfix. It seems that pointyhat's amd64
machine throws an error during the install phase. Pav noticed this and
marked the port BROKEN; however, neither I nor a few others I've
enlisted can reproduce the error. Would you mind removing the
conditional that marks this port BROKEN, try to build/install in your
amd64 environment, and report the results?
Pav, if nobody else can reproduce the error seen on pointyhat, can
portmgr look into whether there is something quirky with the amd64
pointyhat machine?
--
--
Pav Lucistnik p...@oook.cz
p...@freebsd.org
Two sausages are in a frying pan. One says, Geez, it's hot in here
isn't it? And the other one says, Aah! A talking sausage!
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On Tue, 15 Nov 2011 09:31:49 -0500 Sahil Tandon articulated: It was marked broken on a particular architecture, hence the genesis of this thread. I appreciate that diagnosis is difficult; perhaps Olli's suggestion is helpful in isolating the issue. I do not know how else to troubleshoot since these pointyhat errors are not - AFAIK - reproducible by others, on either i386 or amd64 platforms. It would seem to me, and please correct me if my logic is faulty here, that if the problem is ONLY reproducible on pointyhat and not on other systems that the problem would therefore reside with pointyhat. Sahil, I know from time to time that you post on the Postfix forum. Have you tried contacting Wietse, aka Mr Grumpy personally and asking him for his take on the problem? -- Jerry ♔ Disclaimer: off-list followups get on-list replies or get ignored. Please do not ignore the Reply-To header. __ If you don't know the difference between a 'burro' and a 'burrow', then you don't know your ass from a hole in the ground. ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On 15/11/2011 14:31, Sahil Tandon wrote: It was marked broken on a particular architecture, hence the genesis of this thread. I appreciate that diagnosis is difficult; perhaps Olli's suggestion is helpful in isolating the issue. I do not know how else to troubleshoot since these pointyhat errors are not - AFAIK - reproducible by others, on either i386 or amd64 platforms. Hi all, The install is failing during a call to $wrksrc/conf/post-install. Taken from the pointyhat logs : ./postconf -d) |egrep -v '^(myhostname|mydomain|mynetworks) ' ../../conf/main.cf.default ./postconf: fatal: socket: Protocol not supported Postconf is dying with a fatal socket error. This binary is used in various places, including $wrksrc/conf/post-install. There are various tests in this post-install script that will exit with 1 should executing postconf fail for any reason. Therefore, fixing the reason for postconf throwing a fatal error will fix the build/install. What networking/DNS configuration is Pointyhat lacking (or have sufficiently different to break the socket code inside of postconf)? Regards, Jase Thew. ___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
Jase Thew píše v út 15. 11. 2011 v 16:31 +: What networking/DNS configuration is Pointyhat lacking (or have sufficiently different to break the socket code inside of postconf)? It is a purposefully no-networking sandbox jail. What networking activity postconf wants to run? -- -- Pav Lucistnik p...@oook.cz p...@freebsd.org You have acquired a scroll entitled 'irk gleknow mizk' (n). This is an IBM Manual scroll. You are permanently confused. signature.asc Description: This is a digitally signed message part
Re: postfix-current broken on amd64 platform
On Nov 15, 2011, at 11:55 AM, Pav Lucistnik p...@freebsd.org wrote: Jase Thew píše v út 15. 11. 2011 v 16:31 +: What networking/DNS configuration is Pointyhat lacking (or have sufficiently different to break the socket code inside of postconf)? It is a purposefully no-networking sandbox jail. I have an idea; unless someone else figures it out before then, I'll take a look in a few hours when I'm back in front of a computer. Thanks all for the help.___ freebsd-ports@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-ports To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On Tue, 2011-11-15 at 17:55:57 +0100, Pav Lucistnik wrote: Jase Thew píše v út 15. 11. 2011 v 16:31 +: What networking/DNS configuration is Pointyhat lacking (or have sufficiently different to break the socket code inside of postconf)? It is a purposefully no-networking sandbox jail. What networking activity postconf wants to run? Wietse, in a post[1] on the Postfix mailing list, lends further credence to a suspicion that this issue is particular to pointyhat: Postconf opens a socket to determine the mynetworks value (it determines the local interfaces and their netmasks). I have heard about bizarre errors on FreeBSD (jail) systems where the user-land library was out of sync with kernel-land, resulting in data structure mis-matches and system calls returning nonsensical results. [1] http://archives.neohapsis.com/archives/postfix/2011-11/0385.html -- Sahil Tandon pgptIjjuej4zI.pgp Description: PGP signature
Re: postfix-current broken on amd64 platform
[ pav@ and those who tested mail/postifx-current on amd64 added to Cc: ]
On Mon, 2011-11-14 at 08:37:13 -0500, Jerry wrote:
The postfix-current port is still marked as broken:
.if ${ARCH} == amd64
BROKEN= fails during installation
.endif
Since all previous releases of Postfix worked on FreeBSD, and since
Postfix is/was developed on FreeBSD, I was wondering what the problem
is with this release. Is there a possibility that this phenomena might
be rectified in the near future?
Thanks for your report, Jerry. You are correct that FreeBSD is the main
development platform for Postfix. It seems that pointyhat's amd64
machine throws an error during the install phase. Pav noticed this and
marked the port BROKEN; however, neither I nor a few others I've
enlisted can reproduce the error. Would you mind removing the
conditional that marks this port BROKEN, try to build/install in your
amd64 environment, and report the results?
Pav, if nobody else can reproduce the error seen on pointyhat, can
portmgr look into whether there is something quirky with the amd64
pointyhat machine?
--
Sahil Tandon
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org
Re: postfix-current broken on amd64 platform
On Mon, Nov 14, 2011 at 09:24:30PM -0500, Sahil Tandon wrote:
[ pav@ and those who tested mail/postifx-current on amd64 added to Cc: ]
On Mon, 2011-11-14 at 08:37:13 -0500, Jerry wrote:
The postfix-current port is still marked as broken:
.if ${ARCH} == amd64
BROKEN= fails during installation
.endif
Since all previous releases of Postfix worked on FreeBSD, and since
Postfix is/was developed on FreeBSD, I was wondering what the problem
is with this release. Is there a possibility that this phenomena might
be rectified in the near future?
Thanks for your report, Jerry. You are correct that FreeBSD is the main
development platform for Postfix. It seems that pointyhat's amd64
machine throws an error during the install phase. Pav noticed this and
marked the port BROKEN; however, neither I nor a few others I've
enlisted can reproduce the error. Would you mind removing the
conditional that marks this port BROKEN, try to build/install in your
amd64 environment, and report the results?
Pav, if nobody else can reproduce the error seen on pointyhat, can
portmgr look into whether there is something quirky with the amd64
pointyhat machine?
I don't have the successful build logs anymore, so I've just re-queued a
build of postfix-current on 9-STABLE (-RC2) amd64 in my tinderbox if
the results are of interest to those that can investigate any potential
pointyhat anomalies.
Regards,
Glen
--
Glen Barber | g...@freebsd.org
FreeBSD Documentation Project
pgpY4wWj429FM.pgp
Description: PGP signature
