Re: [PATCH net] net/sched: Avoid quadratic handle scan in qdisc_alloc_handle

On Tue, 15 Sep 2026 07:50:54 -0400 Jamal Hadi Salim wrote:

On Mon, Sep 14, 2026 at 10:11 PM Jakub Kicinski [off-list ref] wrote:

quoted

On Fri, 11 Sep 2026 10:31:46 -0300 Victor Nogueira wrote:  

quoted

This is a less intrusive fix meant for backporting; a cleaner approach that
uses a per-device IDA over all qdisc handles is planned for net-next.  

I don't think this is net-worthy in the first place.
We have a ton of code under rtnl_lock.
We cannot provide any protection from malicious netns admin from
overloading the machine.  

That is the one million dollar question.  We labelled this as net for
this exact reason. i.e malicious container could overload the whole
machine.
But it borders on "hardening" (improves performance), so net-next as a
target sounds reasonable. It was a coin toss.

IMO any attack from containers / user ns is hardening. If it leads
to a crash we take it via net _because it's a crash_ not because
user ns can trigger it.
 

Since we have a few similar "grey" issues in our pending queue - so
where's the line for net/net-next?

TBH the line shifts depending on how bad the AI flood is.
I tried to document some rules but it only lead to bikesheding.
So my recommendation is to write the final fix, without worrying
about keeping it simple. And assume the maintainers will redirect
to the other tree based on their judgement when applying.

With all the AI kiddies these days we routinely apply to a different
tree than tagged.

quoted

Nothing against the patch itself, but if you plan to do something
else in net-next let's just go with that from the start..  

We could resend the same patch for net-next, the alternative "proper
fix" is more intrusive. See attached.

I'm not sure we have to do this, given Victor's patch already
drops the time 50x ?

Another option: Is this worth fixing?

Right, I think it is but IDK if it's worth maintaining state for.
IOW Victor's patch would be enough for my taste.