Re: [GIT PULL] BPF changes for 7.2

[Jiayuan Chen <jiayuan.chen@linux.dev>]

On 6/17/26 4:37 PM, Linus Torvalds wrote:
> On Mon, 15 Jun 2026 at 19:09, Alexei Starovoitov
> <alexei.starovoitov@gmail.com> wrote:
>> There are two conflicts:
> Actually, since I ended up doing things a bit out of order and did the
> networking pull first, there were three conflicts, and while the two
> you mentioned were trivial, the one to net/core/filter.c was a bit
> different. Both networking and bpf had done "make sure to preserve the
> sg.copy" bits, but done it quite differently.
>
> And I think bpf did some things better, and the networking tree did
> other things better, so my resolution is actually a mix of the two.
>
> bpf had nicely named helper functions.
>
> And the networking version used __assign_bit() instead of duplicating that
>
>      if (set)
>            set_bit
>      else
>            clear_bit
>
> pattern.
>
> So my resolution is to take a bit from column A, and a bit from column B.
>
> And the networking tree had a case of clearing the sg.copy fields (in
> bpf_msg_pull_data()) that the bpf tree didn't have at all - but I took
> the nicer helper function from the bpf side.
>
> Did I get everything right? It _looks_ right to me, and better than
> either side on their own. But mistakes happen. Please check out my
> resolution and send me fixes from whatever I may have messed up.
>
> It doesn't help that I did this on my laptop while traveling - I hate
> not having my big monitors and better test build environment, but I
> tried to be careful, and it builds for me in my limited test
> environment.
>
> Please give it a good once-over - I've cc'd everybody involved on both
> sides of the conflicting changes.
>
>                  Linus


Thanks, Linus. I was just trying to figure out how to send a patch to
avoid this merge conflict.

The resolution looks correct to me.
One tiny thing: the call to sk_msg_set_elem_copy() in sk_msg_sg_move() has a
trailing tab — it doesn't matter at all, and I'll clean it up next time I
touch that code.

Sorry for the inconvenience.