netdev - Re: [PATCH bpf RESEND 2/4] bpf: Remove size check for sk in bpf_skb_is_valid

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAEf4BzaJMfCXf_uUgyuWBddyd3qrV7SgpVy-hicuOn87FigMSg@mail.gmail.com>
Date:   Fri, 4 Nov 2022 15:43:44 -0700
From:   Andrii Nakryiko <andrii.nakryiko@...il.com>
To:     Alexei Starovoitov <alexei.starovoitov@...il.com>
Cc:     "Russell King (Oracle)" <linux@...linux.org.uk>,
        Yang Jihong <yangjihong1@...wei.com>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Andrii Nakryiko <andrii@...nel.org>,
        Martin KaFai Lau <martin.lau@...ux.dev>,
        Song Liu <song@...nel.org>, Yonghong Song <yhs@...com>,
        John Fastabend <john.fastabend@...il.com>,
        KP Singh <kpsingh@...nel.org>,
        Stanislav Fomichev <sdf@...gle.com>,
        Hao Luo <haoluo@...gle.com>, Jiri Olsa <jolsa@...nel.org>,
        Shubham Bansal <illusionist.neo@...il.com>,
        "David S. Miller" <davem@...emloft.net>,
        Eric Dumazet <edumazet@...gle.com>,
        Jakub Kicinski <kuba@...nel.org>,
        Paolo Abeni <pabeni@...hat.com>,
        Mykola Lysenko <mykolal@...com>, Shuah Khan <shuah@...nel.org>,
        Benjamin Tissoires <benjamin.tissoires@...hat.com>,
        Kumar Kartikeya Dwivedi <memxor@...il.com>,
        Delyan Kratunov <delyank@...com>,
        Artem Savkov <asavkov@...hat.com>, bpf <bpf@...r.kernel.org>,
        linux-arm-kernel <linux-arm-kernel@...ts.infradead.org>,
        LKML <linux-kernel@...r.kernel.org>,
        Network Development <netdev@...r.kernel.org>,
        "open list:KERNEL SELFTEST FRAMEWORK" 
        <linux-kselftest@...r.kernel.org>
Subject: Re: [PATCH bpf RESEND 2/4] bpf: Remove size check for sk in
 bpf_skb_is_valid_access for 32-bit architecture

On Thu, Nov 3, 2022 at 11:15 AM Alexei Starovoitov
<alexei.starovoitov@...il.com> wrote:
>
> On Thu, Nov 3, 2022 at 4:23 AM Russell King (Oracle)
> <linux@...linux.org.uk> wrote:
> >
> > On Thu, Nov 03, 2022 at 05:21:16PM +0800, Yang Jihong wrote:
> > > The error code -EACCES is returned when bpf prog is tested in 32-bit environment,
> > > This is because bpf_object__relocate modifies the instruction to change memory
> > > size to 4 bytes, as shown in the following messages:
> > >
> > > libbpf: prog 'kfunc_call_test1': relo #2: matching candidate #0 <byte_off> [18342] struct __sk_buff.sk (0:30:0 @ offset 168)
> > > libbpf: prog 'kfunc_call_test1': relo #2: patched insn #1 (LDX/ST/STX) off 168 -> 168
> > > libbpf: prog 'kfunc_call_test1': relo #2: patched insn #1 (LDX/ST/STX) mem_sz 8 -> 4
> > >
> > > As a result, the bpf_skb_is_valid_access check fails. For 32-bit architecture,
> > > unnecessary checks need to be deleted.
> >
> > Isn't the purpose of this check to ensure that the entire pointer is
> > written, and BPF can't write half of it?
> >
> >
> > >       case offsetof(struct __sk_buff, sk):
> > > -             if (type == BPF_WRITE || size != sizeof(__u64))
> > > -                     return false;
> >
> > Wouldn't "(size != sizeof(struct bpf_sock *) && size != sizeof(__u64))"
> > be more appropriate here, so 32-bit can only write the 32-bit pointer
> > or the full 64-bit value, and 64-bit can only write the 64-bit pointer?
> > Or is there a reason not to? bpf folk?
>
> You're correct. The patch is completely wrong.
> The bug is elsewhere.

So I looked at this a bit (and replied to the old version of this
patch). What happens in the kernel is that we expect 64-bit load but
rewrite it to 32-bit load on 32-bit architectures (because we just use
sizeof(struct sk_buff, sk) which is 4 bytes on 32-bit arch.

The problem here is that libbpf adjusts such pointer accesses from
8-byte read to 4-byte reads for preserve_access_index (because libbpf
sees that pointer is really 4 byte long), which is what we actually
want in the general case. Here the assumption was made before CO-RE
that __sk_buff is a stable (and fake) UAPI and the correct BPF program
will access sk as a 64-bit pointer because BPF-side pointers always
appear as 64-bit.

But from a correctness standpoint I think it should be fine to enable
both 32- and 64-bit loads for such pointers in __sk_buff for 32-bit
host arch. This will work well with CO-RE and will be correctly
rewritten to 32-bit or 64-bit accesses, depending on host
architecture.

We should still reject 32-bit load on 64-bit host arch, though.