On Fri, 11 Jul 2025 08:41:28 +0200 Johannes Berg wrote: > > Some good news and some bad news. Bad news is that this missed today's > > PR. Good news is that we shipped a bug to Linus which is likely to > > break nl80211 users: > > https://lore.kernel.org/all/20250711001121.3649033-1-kuba@xxxxxxxxxx/ > > so I'm gonna send a second PR tomorrow, with the netlink fix and I'll > > include your fixes in it. I suppose you may want to wait with the fast > > forward until then. > > Now that I typed this I guess it may be two pieces of bad news.. > > No worries, and thanks for the heads-up, I can wait. I actually really > hope this was the last pull request for the current -rc cycle anyway, > but of course now that I said it someone's going to come out of the > woodwork with a fix ;) > > Re the size problem, nl80211 can indeed create and fill really big SKBs > if userspace gives a big buffer, which it often does. Looking at it now, > I'm not sure why we set the rcvbuf to 8KiB in iw? That seems ... odd. It > looks like I originally did that to _increase_ it, but that seems > completely off, and my first fix didn't even do anything. Then we fixed > the call but ... > > Looks like the problem is entirely self-made there, but I guess we still > have to live with stupid userspace, sorry about that. I can take a > closer look at it all and try to fix things there, if you think it's > worth it? Probably not, unless it's hurting in another way. The well maintained userspace is usually not what we have to worry about the most with netlink. I was holding off replying until the dust has fully settled but it sounds like Linus is still hitting some hard to pin down regression. I don't _think_ it's us any more, so I reckon you're safe to fast forward.
