2019-06-25 08:31:04

by Yauheni Kaliuta

[permalink] [raw]
Subject: bpf: test_verifier: sanitation: alu with different scalars

Hi!

I'm wondering, how the sanitaion tests (#903 5.2-rc6 for example)
are supposed to work on BE arches:

{
"sanitation: alu with different scalars 1",
.insns = {
BPF_MOV64_IMM(BPF_REG_0, 1),
BPF_LD_MAP_FD(BPF_REG_ARG1, 0),
BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_FP),
BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG2, -16),
BPF_ST_MEM(BPF_DW, BPF_REG_FP, -16, 0),
BPF_EMIT_CALL(BPF_FUNC_map_lookup_elem),
BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 1),
BPF_EXIT_INSN(),
BPF_LDX_MEM(BPF_B, BPF_REG_1, BPF_REG_0, 0),

reads one byte 0 on BE and 28 on LE (from ->index) since

struct test_val {
unsigned int index;
int foo[MAX_ENTRIES];
};

struct test_val value = {
.index = (6 + 1) * sizeof(int),
.foo[6] = 0xabcdef12,
};

BPF_JMP_IMM(BPF_JEQ, BPF_REG_1, 0, 3),

So different branches are taken depending of the endianness.

BPF_MOV64_IMM(BPF_REG_2, 0),
BPF_MOV64_IMM(BPF_REG_3, 0x100000),
BPF_JMP_A(2),
BPF_MOV64_IMM(BPF_REG_2, 42),
BPF_MOV64_IMM(BPF_REG_3, 0x100001),
BPF_ALU64_REG(BPF_ADD, BPF_REG_2, BPF_REG_3),
BPF_MOV64_REG(BPF_REG_0, BPF_REG_2),
BPF_EXIT_INSN(),
},
.fixup_map_array_48b = { 1 },
.result = ACCEPT,
.retval = 0x100000,
},



--
WBR,
Yauheni Kaliuta


2019-06-25 12:01:52

by Daniel Borkmann

[permalink] [raw]
Subject: Re: bpf: test_verifier: sanitation: alu with different scalars

On 06/25/2019 10:29 AM, Yauheni Kaliuta wrote:
> Hi!
>
> I'm wondering, how the sanitaion tests (#903 5.2-rc6 for example)
> are supposed to work on BE arches:
>
> {
> "sanitation: alu with different scalars 1",
> .insns = {
> BPF_MOV64_IMM(BPF_REG_0, 1),
> BPF_LD_MAP_FD(BPF_REG_ARG1, 0),
> BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_FP),
> BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG2, -16),
> BPF_ST_MEM(BPF_DW, BPF_REG_FP, -16, 0),
> BPF_EMIT_CALL(BPF_FUNC_map_lookup_elem),
> BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 1),
> BPF_EXIT_INSN(),
> BPF_LDX_MEM(BPF_B, BPF_REG_1, BPF_REG_0, 0),
>
> reads one byte 0 on BE and 28 on LE (from ->index) since
>
> struct test_val {
> unsigned int index;
> int foo[MAX_ENTRIES];
> };
>
> struct test_val value = {
> .index = (6 + 1) * sizeof(int),
> .foo[6] = 0xabcdef12,
> };
>
> BPF_JMP_IMM(BPF_JEQ, BPF_REG_1, 0, 3),
>
> So different branches are taken depending of the endianness.
>
> BPF_MOV64_IMM(BPF_REG_2, 0),
> BPF_MOV64_IMM(BPF_REG_3, 0x100000),
> BPF_JMP_A(2),
> BPF_MOV64_IMM(BPF_REG_2, 42),
> BPF_MOV64_IMM(BPF_REG_3, 0x100001),
> BPF_ALU64_REG(BPF_ADD, BPF_REG_2, BPF_REG_3),
> BPF_MOV64_REG(BPF_REG_0, BPF_REG_2),
> BPF_EXIT_INSN(),
> },
> .fixup_map_array_48b = { 1 },
> .result = ACCEPT,
> .retval = 0x100000,
> },

Let me get my hands on a s390x box later today and get back to you.

Thanks,
Daniel

2019-07-26 11:27:57

by Yauheni Kaliuta

[permalink] [raw]
Subject: Re: bpf: test_verifier: sanitation: alu with different scalars

Hi, Daniel,

On Tue, Jun 25, 2019 at 12:39 PM Daniel Borkmann <[email protected]> wrote:
>
> On 06/25/2019 10:29 AM, Yauheni Kaliuta wrote:
> > Hi!
> >
> > I'm wondering, how the sanitaion tests (#903 5.2-rc6 for example)
> > are supposed to work on BE arches:
> >
> > {
> > "sanitation: alu with different scalars 1",
> > .insns = {
> > BPF_MOV64_IMM(BPF_REG_0, 1),
> > BPF_LD_MAP_FD(BPF_REG_ARG1, 0),
> > BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_FP),
> > BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG2, -16),
> > BPF_ST_MEM(BPF_DW, BPF_REG_FP, -16, 0),
> > BPF_EMIT_CALL(BPF_FUNC_map_lookup_elem),
> > BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 1),
> > BPF_EXIT_INSN(),
> > BPF_LDX_MEM(BPF_B, BPF_REG_1, BPF_REG_0, 0),
> >
> > reads one byte 0 on BE and 28 on LE (from ->index) since
> >
> > struct test_val {
> > unsigned int index;
> > int foo[MAX_ENTRIES];
> > };
> >
> > struct test_val value = {
> > .index = (6 + 1) * sizeof(int),
> > .foo[6] = 0xabcdef12,
> > };
> >
> > BPF_JMP_IMM(BPF_JEQ, BPF_REG_1, 0, 3),
> >
> > So different branches are taken depending of the endianness.
> >
> > BPF_MOV64_IMM(BPF_REG_2, 0),
> > BPF_MOV64_IMM(BPF_REG_3, 0x100000),
> > BPF_JMP_A(2),
> > BPF_MOV64_IMM(BPF_REG_2, 42),
> > BPF_MOV64_IMM(BPF_REG_3, 0x100001),
> > BPF_ALU64_REG(BPF_ADD, BPF_REG_2, BPF_REG_3),
> > BPF_MOV64_REG(BPF_REG_0, BPF_REG_2),
> > BPF_EXIT_INSN(),
> > },
> > .fixup_map_array_48b = { 1 },
> > .result = ACCEPT,
> > .retval = 0x100000,
> > },
>
> Let me get my hands on a s390x box later today and get back to you.

Any progress with that?

--
WBR, Yauheni