Allow setting preferred source address in routes per interface via config file

[christf]

this replaces #15 and is directed towards the unicast branch.

With this patch, we can set preferred source address stanzas in routes for each interface using a configuration like this:
interface babel-vpn-1374 type tunnel link-quality true update-interval 300 pref-src aaaa:bbbb:cccc:dddd:eeee::fffff

[jech]

Agreed in principle. Please provide a clean patch series, preferably just a single patch, and preferably against rfc6126bis.

[christf]

the patch towards rfc6126bis was raised via mailing list.

[christf]

with the current branching model I am getting confused with the PRs.

[jech]

The first and last patches are no longer relevant. The second patch (the manpage typo) I've applied.

As to the main patch, it has a bug: use_prefsrc should be set to CONFIG_YES, otherwise the MERGE macro won't work.

Rather than using an extra boolean in the config, I recommend simply setting the prefix to all-zeroes if it is not set, and checking whether it's all zeroes before passing it to the kernel. This avoids having the same information in two places, which risks having inconsistent data.

[christf]

reworked to include review comments.

[jech]

Sorry for the delay, but I was waiting for @killianlufau to submit #31.

I haven't looked at the code in #31 yet, but I rather like the idea of attaching source addresses to install filters rather than to the interface configuration. @christf, I'd like to hear your opinion.

[christf]

On Sun, Mar 31, 2019 at 02:57:15PM -0700, Juliusz Chroboczek wrote: Sorry for the delay, but I was waiting for @killianlufau to submit #31. I haven't looked at the code in #31 yet, but I rather like the idea of attaching source addresses to install filters rather than to the interface configuration. @christf, I'd like to hear your opinion.

It seems like using filters can be more flexible. So yes, this approach is interesting. Why didn't we think of that before?

…

-- () ascii ribbon campaign - against html e-mail /\ against proprietary attachments

[jech]

@christf I'm hoping to find the time to review #31 next week. If you beat me to it, I'd be grateful if you could comment on whether I should apply #31 as it is, or whether you have changes to suggest.

[jech]

Killian,

• your first patch does not compile, please fix that.
• please keep the plen parameter to find_table, unused parameters are okay, carrying a prefix without the associated plen is not;

In the second patch:

• please change the patch summary to "Add install filter parameter pref_src" or something;
• in change_route, please split the conditional into two: first m = install_filter(...) then if(m < INFINITY);
• on this subject — should the check for m < INFINITY be here or in kernel_route? I don't mind either way, but I'm wondering.
• please add documentation to the manual page.

Other than that, it looks good to me.

[rotanid]

@jech this PR is not by kilian, but #31 is

[jech]

@rotanid, right, this was about #31. Since @killianlufau is progressing nicely, I'm closing this request in favour of his. (Thanks for your work, @christf.)