pfSense » pfSense Packages

https://github.com/pfsense/FreeBSD-src/pull/37

see also https://www.freebsd.org/cgi/man.cgi?query=multicast&apropos=0&sektion=4&manpath=FreeBSD+12.1-RELEASE&arch=default&format=html

do i need to open another request for the pimd part?

xavier Lemaire wrote:

do i need to open another request for the pimd part?

Please do it

Renato, I am running an early pimdv3 here. Not for release yet, however working properly .... (maintainer is working on the code , I am doing some testing). However after updating from 13/10 snapshot to todays snapshot, it is not working any more. Not sure what causes that, can be the multicast of some unicast related changes. Do not know. Something has changes in past few days code. I will inform the maintainer about the change in no of vifs and ask his opinion.

Louis van Breda wrote:

Renato, I am running an early pimdv3 here. Not for release yet, however working properly .... (maintainer is working on the code , I am doing some testing). However after updating from 13/10 snapshot to todays snapshot, it is not working any more. Not sure what causes that, can be the multicast of some unicast related changes. Do not know. Something has changes in past few days code. I will inform the maintainer about the change in no of vifs and ask his opinion.

Hi Louis,

We applied the change on FreeBSD-src repository changing #define MAXVIFS 32 to 64. Did it break pimd? Xavier mentioned a new PR would be submitted for pimd part

Almost sure that MAXVIFS 64 does very likely! break the system!

Yesterday, I spend the whole day! , together with the pimd maintainer!, trying to find out why the upcoming pimd release does work on the 20201013-0650 snapshot, but not on the actual release. We did not manage to get pimd running on the actual build! The conclusion is almost sure that there is a problem inside the actual FreeBSD pfSense is using. Given the change that both pfSense snapshots provide a freebsd-version of three times 12.2 stable (huuuu ..... I am only aware of 12.2 release), there is a significant change that the problem is related to MAXVIFS 64. Note that even 12.2-release (we used for testing), is not 100% stable. We had kernel dumps!

Also note that 16/10 I allready noticed that a problem had arised since/after the 13/10 release!

Louis van Breda wrote:

Almost sure that MAXVIFS 64 does very likely! break the system!

Yesterday, I spend the whole day! , together with the pimd maintainer!, trying to find out why the upcoming pimd release does work on the 20201013-0650 snapshot, but not on the actual release. We did not manage to get pimd running on the actual build! The conclusion is almost sure that there is a problem inside the actual FreeBSD pfSense is using. Given the change that both pfSense snapshots provide a freebsd-version of three times 12.2 stable (huuuu ..... I am only aware of 12.2 release), there is a significant change that the problem is related to MAXVIFS 64. Note that even 12.2-release (we used for testing), is not 100% stable. We had kernel dumps!

Also note that 16/10 I allready noticed that a problem had arised since/after the 13/10 release!

Should we restore MAXVIFS to 32? Then we can check what has changed on the repository at that date.

If you are testing and seeing problems on stock FreeBSD, it would be nice to open a ticket at https://bugs.freebsd.org with all details about how to reproduce. If you can test it on a kernel with all debug stuff enabled, and then submit the kernel dump to the ticket, it would make it much easier for kernel developers do debug. If you do so, please link the FreeBSD ticket here so we can keep tracking it

Renato,

Restore to MAXVIFS 32 is a good idea for two reasons:
- it would allow me to test the actual package behavior without that patch, becoming even more sure that that patch is causing the trouble
- pimd (version 3beta) and other things would probably work for the moment.

We could change back to MAXVIFS 64 on a later moment, when the underlying issue is exactly identified and hopefully fixed.

Louis
PS I have been testing on pfSense and it is just not working!, there was no kernel dump / crash. We analyzed the problem using logging's and dumps (and could not find strange things, apart from the kernel refusing to add MFC-entrys). "Troblobit" has tested on 12.2-final with MaxVIFS32 which did work in his test VM.

Renato, you only have to rebuild the OS. There is no neat to do something with the pfSense pimd package since those are based on the old pimd v2 (which is not working in conjunction with FreeBSD 12, due to api changes, at all!). I am using packages based on the latest pimd v3 beta repository and a modified version of the pfSense pimd-package.

Renato, that ^fixed the problem :) I did create a FreeBSD bug report (251047)

I fixed the problem. It was not in the OS. I did compile the PIMD-package in the wrong way in relation MAXVIFS64. So you can revert MAXVIFS back to 64.

What I absolutely do not like, is that the package has to be build for a specific number of MAXVIFS. With as consequence that the package build is (in that regard) target computer depending. IMHO the package should determine the MAXVIFS number at runtime, however it is (historically) not developed that way :( (I can not easily change that and apart from that, I am not the pimd maintainer)

Louis van Breda wrote:

I fixed the problem. It was not in the OS. I did compile the PIMD-package in the wrong way in relation MAXVIFS64. So you can revert MAXVIFS back to 64.

What I absolutely do not like, is that the package has to be build for a specific number of MAXVIFS. With as consequence that the package build is (in that regard) target computer depending. IMHO the package should determine the MAXVIFS number at runtime, however it is (historically) not developed that way :( (I can not easily change that and apart from that, I am not the pimd maintainer)

Unfortunately now we are on code freeze in preparation for 2.5.0 to be released and it's not possible now to apply such change because it would require us to rebuild all package build jails. It will need to wait 2.5.0 to be out and then be added to 2.5.1

OK, I understand. I will rebuild for 32 MAXVIFS. 32 VIFS are plenty for me, but others would perhaps like to have more.

Personally I would prefer a pimd-package with adapt itself to the actual number of maxvifs, but that is of course a little bit tricky change and I do not think troglobit is going to do that, the more since his priority is linux.

For info if you have to change the maxvifs number for the upcomming pimd package from the default 32, you have to build the package with a special config option:
./configure --localstatedir=/var --with-max-vifs=64 (using git)
CONFIGURE_ARGS+= --localstatedir=/var --with-max-vifs=64 (using ports)

Louis van Breda wrote:

OK, I understand. I will rebuild for 32 MAXVIFS. 32 VIFS are plenty for me, but others would perhaps like to have more.

Personally I would prefer a pimd-package with adapt itself to the actual number of maxvifs, but that is of course a little bit tricky change and I do not think troglobit is going to do that, the more since his priority is linux.

For info if you have to change the maxvifs number for the upcomming pimd package from the default 32, you have to build the package with a special config option:
./configure --localstatedir=/var --with-max-vifs=64 (using git)
CONFIGURE_ARGS+= --localstatedir=/var --with-max-vifs=64 (using ports)

It would be nice if pimd build system detect system's MAXVIFS define and use it as default when no value is specified