Message-ID: <CAM0EoMnvjitf-+YFt-qsFHXOnZ4gW3mnXBzMT_-Z6M_XSvWbhQ@mail.gmail.com>
Date: Wed, 5 Nov 2025 10:09:37 -0500
From: Jamal Hadi Salim <jhs@...atatu.com>
To: Simon Horman <horms@...nel.org>
Cc: Ranganath V N <vnranganath.20@...il.com>, davem@...emloft.net, 
	david.hunter.linux@...il.com, edumazet@...gle.com, jiri@...nulli.us, 
	khalid@...nel.org, kuba@...nel.org, linux-kernel@...r.kernel.org, 
	netdev@...r.kernel.org, pabeni@...hat.com, skhan@...uxfoundation.org, 
	syzbot+0c85cae3350b7d486aee@...kaller.appspotmail.com, 
	xiyou.wangcong@...il.com
Subject: Re: [PATCH v2 0/2] net: sched: act_ife: initialize struct tc_ife to
 fix KMSAN kernel-infoleak

On Wed, Nov 5, 2025 at 7:59 AM Simon Horman <horms@...nel.org> wrote:
>
> On Wed, Nov 05, 2025 at 03:33:58PM +0530, Ranganath V N wrote:
> > On 11/4/25 19:38, Simon Horman wrote:
> > > On Sat, Nov 01, 2025 at 06:04:46PM +0530, Ranganath V N wrote:
> > >> Fix a KMSAN kernel-infoleak detected  by the syzbot .
> > >>
> > >> [net?] KMSAN: kernel-infoleak in __skb_datagram_iter
> > >>
> > >> In tcf_ife_dump(), the variable 'opt' was partially initialized using a
> > >> designatied initializer. While the padding bytes are reamined
> > >> uninitialized. nla_put() copies the entire structure into a
> > >> netlink message, these uninitialized bytes leaked to userspace.
> > >>
> > >> Initialize the structure with memset before assigning its fields
> > >> to ensure all members and padding are cleared prior to beign copied.
> > >
> > > Perhaps not important, but this seems to only describe patch 1/2.
> > >
> > >>
> > >> Signed-off-by: Ranganath V N <vnranganath.20@...il.com>
> > >
> > > Sorry for not looking more carefully at v1.
> > >
> > > The presence of this padding seems pretty subtle to me.
> > > And while I agree that your change fixes the problem described.
> > > I wonder if it would be better to make things more obvious
> > > by adding a 2-byte pad member to the structures involved.
> >
> > Thanks for the input.
> >
> > One question — even though adding a 2-byte `pad` field silences KMSAN,
> > would that approach be reliable across all architectures?
> > Since the actual amount and placement of padding can vary depending on
> > structure alignment and compiler behavior, I’m wondering if this would only
> > silence the report on certain builds rather than fixing the root cause.
> >
> > The current memset-based initialization explicitly clears all bytes in the
> > structure (including any compiler-inserted padding), which seems safer and
> > more consistent across architectures.
> >
> > Also, adding a new member — even a padding field — could potentially alter
> > the structure size or layout as seen from user space. That might
> > unintentionally affect existing user-space expectations.
> >
> > Do you think relying on a manual pad field is good enough?
>
> I think these are the right questions to ask.
>
> My thinking is that structures will be padded to a multiple
> of either 4 or 8 bytes, depending on the architecture.
>
> And my observation is that that the unpadded length of both of the structures
> in question are 22 bytes. And that on x86_64 they are padded to 24 bytes.
> Which is divisible by both 4 and 8. So I assume this will be consistent
> for all architectures. If so, I think this would address the questions you
> raised.
>
> I do, however, agree that your current memset-based approach is safer
> in the sense that it carries a lower risk of breaking things because
> it has fewer assumptions (that we have thought of so far).

+1
My view is lets fix the immediate leak issue with the memset, and a
subsequent patch can add the padding if necessary.

cheers,
jamal

Powered by blists - more mailing lists