Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:4 as permitted sender) client-ip=2620:137:e000::3:4;
MIME-Version: 1.0
References: <CACkBjsZ-M=1Yj2PQZM7JN4=9rnDLP36fVO35o9fuAvAMKe=9Nw@mail.gmail.com>
In-Reply-To: <CACkBjsZ-M=1Yj2PQZM7JN4=9rnDLP36fVO35o9fuAvAMKe=9Nw@mail.gmail.com>
From:   Andrii Nakryiko <andrii.nakryiko@gmail.com>
Date:   Tue, 28 Nov 2023 21:46:14 -0800
Message-ID: <CAEf4BzYkc-4kk6gVJPY50txLV_5keNkOruJREKMbew7+Qp71YA@mail.gmail.com>
Subject: Re: [Bug Report] bpf: reg invariant voilation after JSET
To:     Hao Sun <sunhao.th@gmail.com>
Cc:     Andrii Nakryiko <andrii@kernel.org>,
        Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        John Fastabend <john.fastabend@gmail.com>,
        Martin KaFai Lau <martin.lau@linux.dev>,
        Song Liu <song@kernel.org>,
        Yonghong Song <yonghong.song@linux.dev>,
        KP Singh <kpsingh@kernel.org>,
        Stanislav Fomichev <sdf@google.com>,
        Jiri Olsa <jolsa@kernel.org>, bpf <bpf@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Tue, Nov 21, 2023 at 4:57=E2=80=AFAM Hao Sun <sunhao.th@gmail.com> wrote=
:
>
> Hi,
>
> The following program (reduced) breaks reg invariant:
>
> C Repro: https://pastebin.com/raw/FmM9q9D4
>
> -------- Verifier Log --------
> func#0 @0
> 0: R1=3Dctx() R10=3Dfp0
> 0: (18) r8 =3D 0x3d                     ; R8_w=3D61
> 2: (85) call bpf_ktime_get_ns#5       ; R0_w=3Dscalar()
> 3: (ce) if w8 s< w0 goto pc+1         ; R0_w=3Dscalar(smax32=3D61) R8_w=
=3D61
> 4: (95) exit
>
> from 3 to 5: R0_w=3Dscalar(smin=3D0x800000000000003e,smax=3D0x7fffffff7ff=
fffff,umin=3Dsmin32=3Dumin32=3D62,umax=3D0xffffffff7fffffff,umax32=3D0x7fff=
ffff,var_off=3D(0x0;
> 0xffffffff7fffffff)) R8_w=3D61 R10=3Dfp0
> 5: R0_w=3Dscalar(smin=3D0x800000000000003e,smax=3D0x7fffffff7fffffff,umin=
=3Dsmin32=3Dumin32=3D62,umax=3D0xffffffff7fffffff,umax32=3D0x7fffffff,var_o=
ff=3D(0x0;
> 0xffffffff7fffffff)) R8_w=3D61 R10=3Dfp0
> 5: (45) if r0 & 0xfffffff7 goto pc+2
> REG INVARIANTS VIOLATION (false_reg1): range bounds violation
> u64=3D[0x3e, 0x8] s64=3D[0x3e, 0x8] u32=3D[0x3e, 0x8] s32=3D[0x3e, 0x8]
> var_off=3D(0x0, 0x8)
> 5: R0_w=3Dscalar(var_off=3D(0x0; 0x8))
> 6: (dd) if r0 s<=3D r8 goto pc+1
> REG INVARIANTS VIOLATION (false_reg1): range bounds violation
> u64=3D[0x0, 0x8] s64=3D[0x3e, 0x8] u32=3D[0x0, 0x8] s32=3D[0x0, 0x8]
> var_off=3D(0x0, 0x8)
> 6: R0_w=3Dscalar(var_off=3D(0x0; 0x8)) R8_w=3D61
> 7: (bc) w1 =3D w0                       ; R0=3Dscalar(var_off=3D(0x0; 0x8=
))
> R1=3Dscalar(smin=3Dsmin32=3D0,smax=3Dumax=3Dsmax32=3Dumax32=3D8,var_off=
=3D(0x0; 0x8))
> 8: (95) exit
>
> from 6 to 8: safe
>
> from 5 to 8: safe
> processed 10 insns (limit 1000000) max_states_per_insn 0 total_states
> 1 peak_states 1 mark_read 1
>
> The tnum after #5 is correct, but the ranges are incorrect, which seems a=
 bug in
> reg_bounds_sync().  Thoughts?
>

It would be great if in addition to reporting the bug and repro
program, you could also try to analyse why this is happening and
suggest fixes in the verifier.

As I mentioned in another email, when we see REG INVARIANTS VIOLATION,
verifier reverts to conservative unknown scalar register state. We
should try to avoid this pessimistic outcome, but generally speaking
it should not be a critical bug.

> Best
> Hao Sun