Re: [PATCH net-next v2 1/2] ipv6: update NUD_FAILED neighbors from NA messages

[Ido Schimmel]

On Wed, Sep 16, 2026 at 05:43:11PM +0000, Lawrence Lee wrote:
> On Wed Sep 16, 2026 at 12:02 PM UTC, Ido Schimmel wrote:
> > I'm not sure why you added these lowpan changes to the patch. They are
> > not described in the commit message. Given that lowpan_ndisc_update()
> > already has a TODO comment about only handling overrides, I suggest to
> > ignore it. If needed, it can be modified in the future by someone who
> > can explain the use case and test the change.
> >
> 
> I added the lowpan changes after my local Sashiko review run identified 
> an issue where a FAILED neighbor can retain an outdated private short 
> address if it's moved to STALE by a non-override NA. Happy to drop all 
> lowpan-related changes or update comments/commit message to reflect the 
> changes, please let me know your preference.

And if the entry is INCOMPLETE and we get a non-override NA?
lowpan_ndisc_update() will ignore it. It's a pre-existing issue and
there is a TODO comment about this:

"/* react on overrides only. TODO check if this is really right. */"

You can mention that it's a pre-existing issue under the '---'. That
way, hopefully, Sashiko will not flag it and it won't clutter the git
log.

[...]

> Was originally unsure if I should modify the existing logical structure.  
> Thanks for the suggestion, will implement this. Is it appropriate to 
> credit you with a commit tag?

No. I will add my R-b tag when appropriate.

[...]

> > What is the reason for this change? It's also not explained in the
> > commit message and I suspect it's not needed.
> 
> This was added in response to another Sashiko local review finding.  
> Let's say we have some FAILED neighbor T with NTF_ROUTER set. If we get 
> an NA from source address S with target address T and with the Router 
> bit clear, existing kernel code will cleanup routes with gateway S, but 
> IMO we should clean routes with gateway T instead since that is the 
> neighbor which was updated by the NA. I can either update 
> comments/commit message to reflect this or remove the change entirely, 
> please let me know your preference.

And if the neighbour was in a different state than FAILED? It seems like
another pre-existing issue. I guess that in practice, most of the time,
the source address of the NA is also the target address in the payload.
You can mention this under the '---'.