Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:8 as permitted sender) client-ip=2620:137:e000::3:8;
MIME-Version: 1.0
References: <VI1PR0702MB3840EB26EF2A1A35BFEA53BFD9D5A@VI1PR0702MB3840.eurprd07.prod.outlook.com>
 <20231018154205.GT800259@ZenIV> <VI1PR0702MB3840F2D594B9681BF2E0CD81D9D4A@VI1PR0702MB3840.eurprd07.prod.outlook.com>
 <20231019050250.GV800259@ZenIV> <20231019061427.GW800259@ZenIV> <20231019063925.GX800259@ZenIV>
In-Reply-To: <20231019063925.GX800259@ZenIV>
From:   Eric Dumazet <edumazet@google.com>
Date:   Thu, 19 Oct 2023 09:39:45 +0200
Message-ID: <CANn89iJre=VQ6J=UuD0d2J5t=kXr2b9Dk9b=SwzPX1CM+ph60A@mail.gmail.com>
Subject: Re: AW: [PATCH] amd64: Fix csum_partial_copy_generic()
To:     Al Viro <viro@zeniv.linux.org.uk>
Cc:     gus Gusenleitner Klaus <gus@keba.com>,
        Al Viro <viro@ftp.linux.org.uk>,
        Thomas Gleixner <tglx@linutronix.de>,
        lkml <linux-kernel@vger.kernel.org>,
        Ingo Molnar <mingo@redhat.com>, "bp@alien8.de" <bp@alien8.de>,
        "dave.hansen@linux.intel.com" <dave.hansen@linux.intel.com>,
        "x86@kernel.org" <x86@kernel.org>,
        "David S. Miller" <davem@davemloft.net>,
        "dsahern@kernel.org" <dsahern@kernel.org>,
        "kuba@kernel.org" <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Thu, Oct 19, 2023 at 8:39=E2=80=AFAM Al Viro <viro@zeniv.linux.org.uk> w=
rote:
>
> On Thu, Oct 19, 2023 at 07:14:27AM +0100, Al Viro wrote:
> > On Thu, Oct 19, 2023 at 06:02:50AM +0100, Al Viro wrote:
> > > On Thu, Oct 19, 2023 at 04:44:04AM +0000, gus Gusenleitner Klaus wrot=
e:
> > > > > On Wed, Oct 18, 2023 at 06:18:05AM +0000, gus Gusenleitner Klaus =
wrote:
> > > > > > The checksum calculation is wrong in case of an source buffer
> > > > > > containing zero bytes only. The expected return value is 0, the
> > > > > > actual return value is 0xfffffff.
> > > > >
> > > > > Expected where?  The actual checksum is defined modulo 0xffff, so
> > > > > 0 and 0xffffffff represent the same final value.
> > > > >
> > > > > The only twist is that in some situations we internally use 0 for
> > > > > "not calculated yet".
> > > > >
> > > > > > This problem occurs when a ICMP echo reply is sent that has set
> > > > > > zero identifier, sequence number and data.
> > > > >
> > > > > What problem?  Could you please either point to specific RFC or
> > > > > show that packets are rejected by some existing system, or...?
> > > >
> > > > Here's our situation:
> > > > Our device gets pinged by a third party manufacturer robot controll=
er.
> > > > We have updated the kernel in our device to 5.15 from 4.9, the robo=
t
> > > > controller is kept unchanged. At 4.9, our device's ping reply is ac=
cepted
> > > > by the robot controller, at 5.15 it's not.
> > > >
> > > > Wireshark shows a bad checksum warning:
> > > >  'Checksum: 0x0000 incorrect, should be 0xffff'
> > > >
> > >
> > > Lovely.  I think I see what's going on, give me a few to think about =
it...
> >
> > The real source of trouble was switching csum_and_copy_{to,from}_user()
> > to reporting faults as 0.  And yes, it's broken.  Bugger...
>
> I really hate the idea of bringing back the old horrors and splitting
> _nocheck and _user variants ;-/  Especially since we don't care (and
> never had, really) where in the EFAULT case had the damn thing faulted
> and what csum had it managed to accumulate prior to that point.
>
> The only callers are csum_and_copy_..._iter() and they discard
> the entire iovec segment if fault happens; all users of
> csum_and_copy_from_iter() are actually discarding everything in
> that case (reverting iterator to the point where it had been
> prior to the call).
>
> And they are all thread-synchronous.  Hell, it's tempting to steal
> a thread flag, clear it before the call of those suckers, set it in
> exception handlers in those and check in csum_and_copy_..._iter()
> after the call...  Let me see how ugly something of that sort would
> be...

I wonder if the csum_and_copy_...() helpers are really needed in modern day=
s,
with much bigger cpu caches.

Maybe we could remove them and use more standard copy + standard
checksum over kernel buffers.