I am pretty much complete creating a patch which changes how dnsmasq is started by moving the command line parameters into a conf file. This new file is placed into the same directory and the lease file. The test for the command line arguments now checks the contents of the conf-file and there is no longer any tests for the command line parameters which are now two. The first command line parameter is (naturally) --conf-file=<filename>. The second parameter adds new functionality and is --conf-dir=<directory>. This directory is placed into the same directory as the conf-file and the lease file. The name of this directory is "<net-name>.d". This was added to make testing/debugging of new dnsmasq options easier since it no longer requires rebuilding the binaries. This is also useful for adding log-dhcp and/or log-queries for a specific network. Now the option question. I can submit the patch assuming the my previously submitted patch to add --interface to the command line has been applied or I can assume that it has not been applied. In either case, the new code adds a interface=<dev-name> to the conf-file. All development and testing was done with 0.10.2 libvirt src.rpm on Fedora 17. The patch will be submitted based on git.

On 10/22/2012 11:24 AM, Gene Czarcinski wrote: > On 10/22/2012 09:26 AM, Gene Czarcinski wrote: >> I am pretty much complete creating a patch which changes how dnsmasq >> is started by moving the command line parameters into a conf file. >> This new file is placed into the same directory and the lease file. >> >> The test for the command line arguments now checks the contents of >> the conf-file and there is no longer any tests for the command line >> parameters which are now two. >> >> The first command line parameter is (naturally) --conf-file=<filename>. >> >> The second parameter adds new functionality and is >> --conf-dir=<directory>. This directory is placed into the same >> directory as the conf-file and the lease file. The name of this >> directory is "<net-name>.d". This was added to make testing/debugging >> of new dnsmasq options easier since it no longer requires rebuilding >> the binaries. This is also useful for adding log-dhcp and/or >> log-queries for a specific network. >> >> Now the option question. I can submit the patch assuming the my >> previously submitted patch to add --interface to the command line has >> been applied or I can assume that it has not been applied. In either >> case, the new code adds a interface=<dev-name> to the conf-file. >> >> All development and testing was done with 0.10.2 libvirt src.rpm on >> Fedora 17. >> >> The patch will be submitted based on git. >> >> > I have checked and the patch applies clean to the v0.10.2-maint branch > but has problems with the top level. Is providing the patch against > the v0.10.2-maint branch adequate or do you want it reworked (does not > look like a big deal) to the top level? It's much simpler if it applies to the head of master. Try doing this: 1) "git log" and grab the commit ID 2) git checkout master 3) git pull 4) git checkout -b newbranch (or whatever you want to call it) 5) git cherry-pick ${commit-id} This may give you a clean merge (at which point you can just "git send-email -1") or it may give some conflicts. These conflicts will be marked in the source file with: <<<<<<<<< code on current branch ========= conflicting code from cherry-picked patch do a hand merge of the differences, then run "git commit". You should now have a properly merged commit - do "make check && make syntax-check" then git send-email -1.

On 10/22/2012 09:26 AM, Gene Czarcinski wrote: > I am pretty much complete creating a patch which changes how dnsmasq > is started by moving the command line parameters into a conf file. > This new file is placed into the same directory and the lease file. > > The test for the command line arguments now checks the contents of the > conf-file and there is no longer any tests for the command line > parameters which are now two. This will also change the requirements of whether or not a dnsmasq restart or "refresh" (SIGHUP) is required when certain pieces of the network definition are changed with virNetworkUpdate(), but that can be submitted as a separate patch. (basically, things that used to require a change on the commandline, but now only require a change in the conf file, will no longer require a restart).

> > The first command line parameter is (naturally) --conf-file=<filename>. > > The second parameter adds new functionality and is > --conf-dir=<directory>. This directory is placed into the same > directory as the conf-file and the lease file. The name of this > directory is "<net-name>.d". This was added to make testing/debugging > of new dnsmasq options easier since it no longer requires rebuilding > the binaries. This is also useful for adding log-dhcp and/or > log-queries for a specific network. > > Now the option question. I can submit the patch assuming the my > previously submitted patch to add --interface to the command line has > been applied or I can assume that it has not been applied. In either > case, the new code adds a interface=<dev-name> to the conf-file. Submit it assuming that the patch *has* been applied. I think we do need to make that change (adding "--interface"), but it needs to be explicitly visible in a separate patch (and we need to do some testing before pushing it).

On 10/22/2012 03:59 PM, Gene Czarcinski wrote: > Oops. But then I am not sure how to do it. The new code changes > "everything." > > If you want, I can go back and remove it ... make yet-another-patch on > top of the one one to way to the list which removes interface= from > the conf-file and then submit a third patch which puts it back in. I think we're crossing wires. This is what patches I think should be sent: 1) a patch to add --interface to the commandline 2) a patch to switch from using the "long commandline" to using a conf file (which will still put the equivalent of --interface=xxx into the conf file). Isn't that what you already have?

On 10/22/2012 03:17 PM, Laine Stump wrote: > On 10/22/2012 09:26 AM, Gene Czarcinski wrote: >> I am pretty much complete creating a patch which changes how dnsmasq >> is started by moving the command line parameters into a conf file. >> This new file is placed into the same directory and the lease file. >> >> The test for the command line arguments now checks the contents of the >> conf-file and there is no longer any tests for the command line >> parameters which are now two. > This will also change the requirements of whether or not a dnsmasq > restart or "refresh" (SIGHUP) is required when certain pieces of the > network definition are changed with virNetworkUpdate(), but that can be > submitted as a separate patch. (basically, things that used to require a > change on the commandline, but now only require a change in the conf > file, will no longer require a restart). SIGHUP will not do it ... I check into it. By design, configuration files are not revisited once started. Some other files are but not the configuration files. It will take a restart whether is is on the command line or in a conf-file. I believe I understand why it is done this way ... just too much could change ... better to start off clean with a restart (that is from the dnsmasq perspective). > >> >> The first command line parameter is (naturally) --conf-file=<filename>. >> >> The second parameter adds new functionality and is >> --conf-dir=<directory>. This directory is placed into the same >> directory as the conf-file and the lease file. The name of this >> directory is "<net-name>.d". This was added to make testing/debugging >> of new dnsmasq options easier since it no longer requires rebuilding >> the binaries. This is also useful for adding log-dhcp and/or >> log-queries for a specific network. >> >> Now the option question. I can submit the patch assuming the my >> previously submitted patch to add --interface to the command line has >> been applied or I can assume that it has not been applied. In either >> case, the new code adds a interface=<dev-name> to the conf-file. > Submit it assuming that the patch *has* been applied. I think we do need > to make that change (adding "--interface"), but it needs to be > explicitly visible in a separate patch (and we need to do some testing > before pushing it). Oops. But then I am not sure how to do it. The new code changes "everything." If you want, I can go back and remove it ... make yet-another-patch on top of the one one to way to the list which removes interface= from the conf-file and then submit a third patch which puts it back in. I am not sure I understand the reluctance about putting the --interface patch in. This problem exists in using any dnsmasq => 2.61. I believe I have seen it happen although I did not realize it at the time. Simon says that there is logging of the event in dnsmasq's rfc3315.c but I have not gone back to prove that the problem exists if --interface is not specified. I am not talking about adding this to older versions of libvirt ... just the current one and the version that is on Fedora 17 since dnsmasq was so quickly updated to 2.63. OK, over to you ... what do you want me to do?