Re: [PATCH RFC net-next] inet: add ip_retry_random_port sysctl to reduce sequential port retries

On Monday, February 9th, 2026 at 12:57 PM, Fernando Fernandez Mancera [off-list ref] wrote:

On 2/6/26 6:09 PM, Eric Dumazet wrote:

quoted

On Fri, Feb 6, 2026 at 5:28 PM Fernando Fernandez Mancera
fmancera@suse.de wrote:

quoted

It makes sense. I have tested this approach and we got a more even
distribution of source ports when having thousands of reserved ports. No
difference at all when not using reserved ports.

Please, you can find the distribution graph with the current algorithm
[1] and with the random step algorithm [2].

While I understand that this approach is introducing a call to
get_random_u32_below() on every connect, I am wondering if it makes
sense to replace the existing algorithm with this variant. What do you
think?

I would ask RFC 6056 experts like Fernando Gont what they think.

Note that if we use random at each connect(), we defeat one of the principles
of ephemeral port selection : try very hard to avoid 4-tuple collision.

Right. I will reach out to him and get his opinion. I have plenty of
time before net-next open again. I am also collecting some metrics
regarding the 4-tuple collision frequency.

We have had this problem in AWS for a long time. The patch works on our system. What is needed for it to be included in the next Linux release?

Please bring this to the stable versions.

quoted

quoted

Please, notice the implementation below. I plan to send an official v1
once net-next is open. In addition, I am rewriting the commit message as
I find the current one confusing.

[1] https://0xffsoftware.com/port_graph_current_alg.html

[2] https://0xffsoftware.com/port_graph_random_step_alg.html