Re: [PATCH iproute2-next 0/2] bridge: add cfm command

[Abdul Wasey <w453y.me@gmail.com>]

On Tue, Oct 06, 2026 at 10:42:49AM +0000, Ido Schimmel wrote:
> Please elaborate on the motivation: What is your interest in CFM? Are
> you using it in production? Does anyone? Who needs this in iproute2?
> What is missing from Microchip's cfm tool?
>
> Looking at the kernel log, since CFM was merged six years ago, no new
> features were added and the only bug fixes are for issues found by
> static checkers and LLMs with no reviews from the original authors.
> There are also no selftests.
>
> If nobody cares about CFM or uses it, then I would prefer to start
> deprecating it instead of accumulating more code that needs to be
> maintained and that nobody is going to use.

Hi Ido,

Thanks for asking directly. Short answer: I don't run CFM in
production and I don't know anyone who does.

How I got here: I work on BFD (FRR's bfdd and an XDP BFD data plane)
and was looking at how other liveness protocols are handled on Linux.
CFM in the bridge had no iproute2 support, so I wrote it. I didn't
check first whether anyone needed it, and I should have.

Testing it, I found the bridge CFM does not keep its own timing at the
short intervals: with a 3.3 ms CCM interval it sends every ~5 ms and
declares loss after about 8 intervals instead of 3.5, because the
timers are delayed works in jiffies. The ccm-tx period also overflows
a u32 above 4294 s, and interval "none" with CC enabled rearms the work
with no delay. I think that fits what you describe: anyone running
the short intervals would have hit the first one.

Microchip's cfm tool covers configuration and show; what it does not
have is being packaged anywhere and a monitor for events. That alone
is not a strong reason to carry 1300 lines in iproute2.

So I'm fine with dropping this series. Horatiu, do you know of anyone
using bridge CFM? If nobody does, I'm also fine with deprecating it,
and I can help with that if you want.

The same goes for my net-next series "net: bridge: cfm: notify
userspace on CFM config changes", which only makes sense if CFM
stays; I'll leave it until this is settled.

Thanks,
Abdul Wasey