No, please do not add this kind of complexity into Makefile. Parsing an XML file by using grep or awk is a big regression even if it is useful. Use a stylesheet instead.

Looks good to me. Please go ahead to commit.

FYI, r271910 includes a fix for the original problem which Xin worked on at r151023. It may be worth merging it, too.

• Can you please consider to reuse nd6_ns_input() in some way to send unsolicited NAs? nd6_announce() is a specific version of it and most of code can be shared. nd6_announce() lacks ifa checks other than anycast (e.g. tentative, deprecated, duplicated addresses must be filtered out) and R flag handling, and nd6_na_output_fib() instead of nd6_na_output() has to be used based on an interface's FIB for consistency. nd6_announce_delay() should use retrans timer (ND_IFINFO(ifa->ifa_ifp)->retrans) to separate NAs instead of hard-coded value (1000). And in practice, a down->up event sends an invalid NA because DAD of link-local address must be performed before sending an NA (i.e. nd6_announce() has to wait until a link-local address is ready to use).

Sending NAs for all of the AF_INET6 addresses unconditionally with ND_NA_FLAG_OVERRIDE flag is incorrect (anycast address should not have one, for example. See 7.2.6 in RFC 4861). Unsolicited NAs are certainly useful for L2 address change, but I am skeptical about if it works well in up/down events. This is because all of AF_INET6 addresses are marked as tentative when an interface is becoming down and duplicate address detection is always performed again when becoming up. So this behavior will conflict with NAs sent by the linkup event.

Looks good to me. Thank you for your work!

I already made my question clear but I did not get the answer yet. Please explain the goal. The proposed files in the git repo contain changes for both design and build infrastructure. Given that the goal is to replace the existing htdocs/ tree with it completely, the latter (build infrastructure) part cannot be imported because we are using XML extensively and we do not intend to change it. Although there may be pros and cons on both sides and some people agree that a python-based template engine is easy and useful, it is a separate discussion. What we need is a modern design in terms of appearance and user interface, not a new doc management infrastructure. The existing files in the htdocs/ directory are not garbage, and converting them into another format solves nothing.

It is true that this LOR is driver-specific but calling SIOCSIFMTU after acquiring a lock in lagg ioctl is not always safe. Change of lladdr suffers from the same situation and it was solved by using an asynchronous task queue to update addresses on each port. What do you think about piggybacking an MTU change to the queue by extending struct lagg_llq to a more generic one which makes it possible to handle per-port properties?

Please be more specific about what kind of feedback is expected. Which is the goal of this proposal, a new design, a new directory structure and build infrastructure, or the both of them?

This fix is correct. I will commit it asap.

Probably I overlooked something, but does new ip6_output() work with a loopback from/to link-local addresses? I think scope checks will fail when lo0 is passed as the outgoing interface. I will take a look into the patch more closely in this weekend...

The patch looks good to me.

In D3301#78885, @shahark_mellanox.com wrote:

I've tested the patch and it works

I do not think this fix is correct. lagg_port_destroy() should enqueue an lladdr change into sc_lladdr_task when the primary port is removed. Removing the llq_primary == 0 condition causes an unnecessary lladdr update on the primary interface.

Update the patch to support AF_UNSPEC and fix getnameinfo(3) manual page
based on a reviewer comment.