Re: [SECURITY] Bypass of Page Cache Protection in IPsec ESP TX Path via Incorrect skb_cow_data() Semantic

[Willy Tarreau <w@1wt.eu>]

On Sun, May 10, 2026 at 09:18:12AM -0700, Linus Torvalds wrote:
> On Sun, 10 May 2026 at 08:22, yuri08 <nvt031@gmail.com> wrote:
> >
> > In net/core/skbuff.c (Function __skb_cow_data): The core issue lies here:
> 
> Is this all just AI hallucination? There is no such function with
> double underscores, and the other "quoted" code seems to also be of
> the kind "almost, but not actually match reality".

Pretty disturbing indeed! It looks like a context with too strong
quantization where "tailbits" became "tailroom" and some other parts
were poorly remmebered when dumped.

@yuri08 please revisit entirely your report to fix it or trash it if
not valid. Please don't send unvalidated crap out of AI tools, it's
just wasting precious maintainers' time that would be better spent
doing something useful.

Please also recheck everything you may have previously shared, and
retract your claims if you spot anything dubious. I'm seeing at least
one patch about UAF on algif_aead shared with maintainers and LKML,
that we don't know anymore if we should trust it or not.

thanks,
Willy