Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <CACkBjsaRPHUpfST=3-FTWbQycHsNgvWyVFpCLkJEwcrLhSZfnQ@mail.gmail.com>
In-Reply-To: <CACkBjsaRPHUpfST=3-FTWbQycHsNgvWyVFpCLkJEwcrLhSZfnQ@mail.gmail.com>
From:   Alexei Starovoitov <alexei.starovoitov@gmail.com>
Date:   Mon, 7 Nov 2022 12:51:59 -0800
Message-ID: <CAADnVQLev0xX3wHm-dgD-eJcV0vkuvZ0Tm=X_P68qcUabKw8eA@mail.gmail.com>
Subject: Re: BUG in BPF verifier, 10 insns costs 2 mins+ to load
To:     Hao Sun <sunhao.th@gmail.com>
Cc:     Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        bpf <bpf@vger.kernel.org>, Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        John Fastabend <john.fastabend@gmail.com>,
        Andrii Nakryiko <andrii@kernel.org>,
        Martin KaFai Lau <martin.lau@linux.dev>, ong@kernel.org,
        Yonghong Song <yhs@fb.com>, KP Singh <kpsingh@kernel.org>,
        Stanislav Fomichev <sdf@google.com>,
        Hao Luo <haoluo@google.com>, Jiri Olsa <jolsa@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Sun, Nov 6, 2022 at 8:26 PM Hao Sun <sunhao.th@gmail.com> wrote:
>
> Hi,
>
> I've just written a BPF verifier fuzzer, targeting logic bugs in the
> BPF verifier.
> The following is an abnormal case it generated. The case only contains 10
> BPF instructions but costs more than 2 mins to load on :

with full verbose verifier logging, right?
That is expected for any prog that is going to hit the 1M insn limit.

>     HEAD commit: f0c4d9fc9cc9 Linux 6.1-rc4
>     git tree: upstream
>     kernel config: https://pastebin.com/raw/SBxaikiG
>     C reproducer: https://pastebin.com/raw/HsDXdraZ
>     verifier log: https://pastebin.com/raw/sNmSsVxs
>
> Ideally, the verifier should exit quickly in this case, since R2=42
> always holds.
> The behaviour of the verifier does not make sense to me, seems it lost
> the range information of R2.
>
> Please point out if I missed anything, the C reproducer in the link
> (https://pastebin.com/raw/HsDXdraZ)
> essentially loads the following case into `test_verifier.c`:
> {
> "BVF verifier test",
> .insns = {
> BPF_MOV64_IMM(BPF_REG_1, 42),
> BPF_MOV64_IMM(BPF_REG_2, 0),
> BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 1, 0, 1),
> BPF_EXIT_INSN(),
> BPF_ALU64_REG(BPF_ADD, BPF_REG_2, BPF_REG_1),
> BPF_ALU32_IMM(BPF_DIV, BPF_REG_2, 1),
> BPF_ALU64_IMM(BPF_SUB, BPF_REG_1, 108),
> BPF_JMP32_REG(BPF_JGT, BPF_REG_1, BPF_REG_2, -3),
> BPF_MOV64_REG(BPF_REG_0, BPF_REG_2),
> BPF_EXIT_INSN(),
> },
> .prog_type = BPF_PROG_TYPE_XDP,
> },
>
> The verifier's log is more then 4M, but essentially is:
>     0: R1=ctx(off=0,imm=0) R10=fp0
>     0: (b7) r1 = 42                       ; R1_w=P42
>     1: (b7) r2 = 0                        ; R2_w=P0
>     2: (85) call pc+1
>     caller:
>      R10=fp0
>     callee:
>      frame1: R1_w=P42 R2_w=P0 R10=fp0
>     4: (57) r2 &= -52                     ; frame1: R2_w=P0
>     5: (0f) r2 += r1                      ; frame1: R1_w=P42 R2_w=P42
>     6: (34) w2 /= 1                       ; frame1:
> R2_w=Pscalar(umax=4294967295,var_off=(0x0; 0xffffffff))
>     7: (17) r1 -= 108                     ; frame1: R1_w=P-66
>     8: (2e) if w1 > w2 goto pc-3 6: frame1: R1_w=P-66
> R2_w=Pscalar(umax=4294967229,var_off=(0x0; 0xffffffff)) R10=fp0
>     6: (34) w2 /= 1                       ; frame1:
> R2_w=Pscalar(umax=4294967295,var_off=(0x0; 0xffffffff))
>     7: (17) r1 -= 108                     ; frame1: R1_w=P-174
>     8: (2e) if w1 > w2 goto pc-3 6: frame1: R1_w=P-174
> R2_w=Pscalar(umax=4294967121,var_off=(0x0; 0xffffffff)) R10=fp0
>     6: (34) w2 /= 1                       ; frame1:
> R2_w=Pscalar(umax=4294967295,var_off=(0x0; 0xffffffff))
>     7: (17) r1 -= 108                     ; frame1: R1=P-282
>     8: (2e) if w1 > w2 goto pc-3 6: frame1: R1=P-282
> R2=Pscalar(umax=4294967013,var_off=(0x0; 0xffffffff)) R10=fp0
>     ...
>     6: (34) w2 /= 1                       ; frame1:
> R2_w=Pscalar(umax=4294967295,var_off=(0x0; 0xffffffff))
>     7: (17) r1 -= 108                     ; frame1: R1_w=P-6342690
>     8: (2e) if w1 > w2 goto pc-3 6: frame1: R1_w=P-6342690
> R2_w=Pscalar(umax=4288624605,var_off=(0x0; 0xffffffff)) R10=fp0
>     6: (34) w2 /= 1                       ; frame1:
> R2_w=Pscalar(umax=4294967295,var_off=(0x0; 0xffffffff))
>     7: (17) r1 -= 108                     ; frame1: R1_w=P-6342798
>     8: (2e) if w1 > w2 goto pc-3          ; frame1: R1_w=P-6342798
> R2_w=Pscalar(umin=4288624498,umax=4294967295,var_off=(0xff800000;
> 0x7fffff),s32_min=-6342798,s32_max=-1)
>     9: (bf) r0 = r2                       ; frame1:
> R0_w=Pscalar(id=58730,umin=4288624498,umax=4294967295,var_off=(0xff800000;
> 0x7fffff),s32_min=-6342798,s32_max=-1)
> R2_w=Pscalar(id=58730,umin=4288624498,umax=4294967295,var_off=(0xff800000;
> 0x7fffff),s32_min=-6342798,s32_max=-1)
>     10: (95) exit
>     returning from callee:
>      frame1: R0_w=Pscalar(id=58730,umin=4288624498,umax=4294967295,var_off=(0xff800000;
> 0x7fffff),s32_min=-6342798,s32_max=-1) R1_w=P-6342798
> R2_w=Pscalar(id=58730,umin=4288624498,umax=4294967295,var_off=(0xff800000;
> 0x7fffff),s32_min=-6342798,s32_max=-1) R10=fp0
>     to caller at 3:
>      R0_w=Pscalar(id=58730,umin=4288624498,umax=429496