Immediate is incorrectly cast to u32 before being spilled, losing sign information. The range information is incorrect after load again. Fix immediate spill by remove the cast. The second patch add a test case for this.
Signed-off-by: Hao Sun sunhao.th@gmail.com --- Hao Sun (2): bpf: Fix check_stack_write_fixed_off() to correctly spill imm selftests/bpf: Add test for immediate spilled to stack
kernel/bpf/verifier.c | 2 +- tools/testing/selftests/bpf/verifier/bpf_st_mem.c | 32 +++++++++++++++++++++++ 2 files changed, 33 insertions(+), 1 deletion(-) --- base-commit: 399f6185a1c02f39bcadb8749bc2d9d48685816f change-id: 20231026-fix-check-stack-write-c40996694dfa
Best regards,
In check_stack_write_fixed_off(), imm value is cast to u32 before being spilled to the stack. Therefore, the sign information is lost, and the range information is incorrect when load from the stack again.
For the following prog: 0: r2 = r10 1: *(u64*)(r2 -40) = -44 2: r0 = *(u64*)(r2 - 40) 3: if r0 s<= 0xa goto +2 4: r0 = 1 5: exit 6: r0 = 0 7: exit
The verifier gives: func#0 @0 0: R1=ctx(off=0,imm=0) R10=fp0 0: (bf) r2 = r10 ; R2_w=fp0 R10=fp0 1: (7a) *(u64 *)(r2 -40) = -44 ; R2_w=fp0 fp-40_w=4294967252 2: (79) r0 = *(u64 *)(r2 -40) ; R0_w=4294967252 R2_w=fp0 fp-40_w=4294967252 3: (c5) if r0 s< 0xa goto pc+2 mark_precise: frame0: last_idx 3 first_idx 0 subseq_idx -1 mark_precise: frame0: regs=r0 stack= before 2: (79) r0 = *(u64 *)(r2 -40) 3: R0_w=4294967252 4: (b7) r0 = 1 ; R0_w=1 5: (95) exit verification time 7971 usec stack depth 40 processed 6 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
So remove the incorrect cast, since imm field is declared as s32, and __mark_reg_known() takes u64, so imm would be correctly sign extended by compiler.
Signed-off-by: Hao Sun sunhao.th@gmail.com --- kernel/bpf/verifier.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 857d76694517..44af69ce1301 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -4674,7 +4674,7 @@ static int check_stack_write_fixed_off(struct bpf_verifier_env *env, insn->imm != 0 && env->bpf_capable) { struct bpf_reg_state fake_reg = {};
- __mark_reg_known(&fake_reg, (u32)insn->imm); + __mark_reg_known(&fake_reg, insn->imm); fake_reg.type = SCALAR_VALUE; save_register_state(state, spi, &fake_reg, size); } else if (reg && is_spillable_regtype(reg->type)) {
On Thu, Oct 26, 2023 at 05:13:10PM +0200, Hao Sun wrote:
In check_stack_write_fixed_off(), imm value is cast to u32 before being spilled to the stack. Therefore, the sign information is lost, and the range information is incorrect when load from the stack again.
For the following prog: 0: r2 = r10 1: *(u64*)(r2 -40) = -44 2: r0 = *(u64*)(r2 - 40) 3: if r0 s<= 0xa goto +2 4: r0 = 1 5: exit 6: r0 = 0 7: exit
The verifier gives: func#0 @0 0: R1=ctx(off=0,imm=0) R10=fp0 0: (bf) r2 = r10 ; R2_w=fp0 R10=fp0 1: (7a) *(u64 *)(r2 -40) = -44 ; R2_w=fp0 fp-40_w=4294967252 2: (79) r0 = *(u64 *)(r2 -40) ; R0_w=4294967252 R2_w=fp0 fp-40_w=4294967252 3: (c5) if r0 s< 0xa goto pc+2 mark_precise: frame0: last_idx 3 first_idx 0 subseq_idx -1 mark_precise: frame0: regs=r0 stack= before 2: (79) r0 = *(u64 *)(r2 -40) 3: R0_w=4294967252 4: (b7) r0 = 1 ; R0_w=1 5: (95) exit verification time 7971 usec stack depth 40 processed 6 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
So remove the incorrect cast, since imm field is declared as s32, and __mark_reg_known() takes u64, so imm would be correctly sign extended by compiler.
Signed-off-by: Hao Sun sunhao.th@gmail.com
Acked-by: Shung-Hsi Yu shung-hsi.yu@suse.com
The acked-by applies to future version of the patchset as well.
FWIW I think we'd also need the same treatment for the (BPF_ALU | BPF_MOV | BPF_K) case in check_alu_op().
kernel/bpf/verifier.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 857d76694517..44af69ce1301 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -4674,7 +4674,7 @@ static int check_stack_write_fixed_off(struct bpf_verifier_env *env, insn->imm != 0 && env->bpf_capable) { struct bpf_reg_state fake_reg = {};
__mark_reg_known(&fake_reg, (u32)insn->imm);
fake_reg.type = SCALAR_VALUE; save_register_state(state, spi, &fake_reg, size); } else if (reg && is_spillable_regtype(reg->type)) {__mark_reg_known(&fake_reg, insn->imm);
-- 2.34.1
On Fri, Oct 27, 2023 at 03:14:10PM +0800, Shung-Hsi Yu wrote:
On Thu, Oct 26, 2023 at 05:13:10PM +0200, Hao Sun wrote:
In check_stack_write_fixed_off(), imm value is cast to u32 before being spilled to the stack. Therefore, the sign information is lost, and the range information is incorrect when load from the stack again.
For the following prog: 0: r2 = r10 1: *(u64*)(r2 -40) = -44 2: r0 = *(u64*)(r2 - 40) 3: if r0 s<= 0xa goto +2 4: r0 = 1 5: exit 6: r0 = 0 7: exit
The verifier gives: func#0 @0 0: R1=ctx(off=0,imm=0) R10=fp0 0: (bf) r2 = r10 ; R2_w=fp0 R10=fp0 1: (7a) *(u64 *)(r2 -40) = -44 ; R2_w=fp0 fp-40_w=4294967252 2: (79) r0 = *(u64 *)(r2 -40) ; R0_w=4294967252 R2_w=fp0 fp-40_w=4294967252 3: (c5) if r0 s< 0xa goto pc+2 mark_precise: frame0: last_idx 3 first_idx 0 subseq_idx -1 mark_precise: frame0: regs=r0 stack= before 2: (79) r0 = *(u64 *)(r2 -40) 3: R0_w=4294967252 4: (b7) r0 = 1 ; R0_w=1 5: (95) exit verification time 7971 usec stack depth 40 processed 6 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
So remove the incorrect cast, since imm field is declared as s32, and __mark_reg_known() takes u64, so imm would be correctly sign extended by compiler.
Signed-off-by: Hao Sun sunhao.th@gmail.com
Acked-by: Shung-Hsi Yu shung-hsi.yu@suse.com
The acked-by applies to future version of the patchset as well.
Oh and since this is a fix it would be great to have the fixes tag[1] to specify when the bug was introduced
Fixes: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction")
Add Cc tag for stable[2] so stable kernels pick up the fix as well
Cc: stable@vger.kernel.org
And ideally specify that the patch should be applied to the bpf tree rather than bpf-next[3] (though the BPF maintainers has the final say on which tree this patch should be applied).
I'd owe you a big thank as well since this helps with our internal process at my company. So thank you in advance!
1: https://docs.kernel.org/process/submitting-patches.html#describe-your-change... 2: https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html#opti... 3: https://docs.kernel.org/bpf/bpf_devel_QA.html#q-how-do-the-changes-make-thei...
FWIW I think we'd also need the same treatment for the (BPF_ALU | BPF_MOV | BPF_K) case in check_alu_op().
kernel/bpf/verifier.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 857d76694517..44af69ce1301 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -4674,7 +4674,7 @@ static int check_stack_write_fixed_off(struct bpf_verifier_env *env, insn->imm != 0 && env->bpf_capable) { struct bpf_reg_state fake_reg = {};
__mark_reg_known(&fake_reg, (u32)insn->imm);
fake_reg.type = SCALAR_VALUE; save_register_state(state, spi, &fake_reg, size); } else if (reg && is_spillable_regtype(reg->type)) {__mark_reg_known(&fake_reg, insn->imm);
-- 2.34.1
On Fri, Oct 27, 2023 at 9:44 AM Shung-Hsi Yu shung-hsi.yu@suse.com wrote:
On Fri, Oct 27, 2023 at 03:14:10PM +0800, Shung-Hsi Yu wrote:
On Thu, Oct 26, 2023 at 05:13:10PM +0200, Hao Sun wrote:
In check_stack_write_fixed_off(), imm value is cast to u32 before being spilled to the stack. Therefore, the sign information is lost, and the range information is incorrect when load from the stack again.
For the following prog: 0: r2 = r10 1: *(u64*)(r2 -40) = -44 2: r0 = *(u64*)(r2 - 40) 3: if r0 s<= 0xa goto +2 4: r0 = 1 5: exit 6: r0 = 0 7: exit
The verifier gives: func#0 @0 0: R1=ctx(off=0,imm=0) R10=fp0 0: (bf) r2 = r10 ; R2_w=fp0 R10=fp0 1: (7a) *(u64 *)(r2 -40) = -44 ; R2_w=fp0 fp-40_w=4294967252 2: (79) r0 = *(u64 *)(r2 -40) ; R0_w=4294967252 R2_w=fp0 fp-40_w=4294967252 3: (c5) if r0 s< 0xa goto pc+2 mark_precise: frame0: last_idx 3 first_idx 0 subseq_idx -1 mark_precise: frame0: regs=r0 stack= before 2: (79) r0 = *(u64 *)(r2 -40) 3: R0_w=4294967252 4: (b7) r0 = 1 ; R0_w=1 5: (95) exit verification time 7971 usec stack depth 40 processed 6 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
So remove the incorrect cast, since imm field is declared as s32, and __mark_reg_known() takes u64, so imm would be correctly sign extended by compiler.
Signed-off-by: Hao Sun sunhao.th@gmail.com
Acked-by: Shung-Hsi Yu shung-hsi.yu@suse.com
The acked-by applies to future version of the patchset as well.
(BPF_ALU | BPF_MOV | BPF_K) is handled correctly in the current code, i.e., no cast in BPF_ALU64 so that the sign is extended, and the cast in BPF_ALU correctly zero extend the reg.
Oh and since this is a fix it would be great to have the fixes tag[1] to specify when the bug was introduced
Fixes: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction")
Noted, thanks.
Add Cc tag for stable[2] so stable kernels pick up the fix as well
Cc: stable@vger.kernel.org
And ideally specify that the patch should be applied to the bpf tree rather than bpf-next[3] (though the BPF maintainers has the final say on which tree this patch should be applied).
I'd owe you a big thank as well since this helps with our internal process at my company. So thank you in advance!
1: https://docs.kernel.org/process/submitting-patches.html#describe-your-change... 2: https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html#opti... 3: https://docs.kernel.org/bpf/bpf_devel_QA.html#q-how-do-the-changes-make-thei...
FWIW I think we'd also need the same treatment for the (BPF_ALU | BPF_MOV | BPF_K) case in check_alu_op().
kernel/bpf/verifier.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 857d76694517..44af69ce1301 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -4674,7 +4674,7 @@ static int check_stack_write_fixed_off(struct bpf_verifier_env *env, insn->imm != 0 && env->bpf_capable) { struct bpf_reg_state fake_reg = {};
__mark_reg_known(&fake_reg, (u32)insn->imm);
} else if (reg && is_spillable_regtype(reg->type)) {__mark_reg_known(&fake_reg, insn->imm); fake_reg.type = SCALAR_VALUE; save_register_state(state, spi, &fake_reg, size);
-- 2.34.1
On Fri, Oct 27, 2023 at 09:51:58AM +0200, Hao Sun wrote:
On Fri, Oct 27, 2023 at 9:44 AM Shung-Hsi Yu shung-hsi.yu@suse.com wrote:
On Fri, Oct 27, 2023 at 03:14:10PM +0800, Shung-Hsi Yu wrote:
On Thu, Oct 26, 2023 at 05:13:10PM +0200, Hao Sun wrote:
In check_stack_write_fixed_off(), imm value is cast to u32 before being spilled to the stack. Therefore, the sign information is lost, and the range information is incorrect when load from the stack again.
For the following prog: 0: r2 = r10 1: *(u64*)(r2 -40) = -44 2: r0 = *(u64*)(r2 - 40) 3: if r0 s<= 0xa goto +2 4: r0 = 1 5: exit 6: r0 = 0 7: exit
The verifier gives: func#0 @0 0: R1=ctx(off=0,imm=0) R10=fp0 0: (bf) r2 = r10 ; R2_w=fp0 R10=fp0 1: (7a) *(u64 *)(r2 -40) = -44 ; R2_w=fp0 fp-40_w=4294967252 2: (79) r0 = *(u64 *)(r2 -40) ; R0_w=4294967252 R2_w=fp0 fp-40_w=4294967252 3: (c5) if r0 s< 0xa goto pc+2 mark_precise: frame0: last_idx 3 first_idx 0 subseq_idx -1 mark_precise: frame0: regs=r0 stack= before 2: (79) r0 = *(u64 *)(r2 -40) 3: R0_w=4294967252 4: (b7) r0 = 1 ; R0_w=1 5: (95) exit verification time 7971 usec stack depth 40 processed 6 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
So remove the incorrect cast, since imm field is declared as s32, and __mark_reg_known() takes u64, so imm would be correctly sign extended by compiler.
Signed-off-by: Hao Sun sunhao.th@gmail.com
Acked-by: Shung-Hsi Yu shung-hsi.yu@suse.com
The acked-by applies to future version of the patchset as well.
(BPF_ALU | BPF_MOV | BPF_K) is handled correctly in the current code, i.e., no cast in BPF_ALU64 so that the sign is extended, and the cast in BPF_ALU correctly zero extend the reg.
My mistake, you're right. Thank you for the explanation.
Oh and since this is a fix it would be great to have the fixes tag[1] to specify when the bug was introduced
Fixes: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction")
Noted, thanks.
Add Cc tag for stable[2] so stable kernels pick up the fix as well
Cc: stable@vger.kernel.org
And ideally specify that the patch should be applied to the bpf tree rather than bpf-next[3] (though the BPF maintainers has the final say on which tree this patch should be applied).
I'd owe you a big thank as well since this helps with our internal process at my company. So thank you in advance!
1: https://docs.kernel.org/process/submitting-patches.html#describe-your-change... 2: https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html#opti... 3: https://docs.kernel.org/bpf/bpf_devel_QA.html#q-how-do-the-changes-make-thei...
FWIW I think we'd also need the same treatment for the (BPF_ALU | BPF_MOV | BPF_K) case in check_alu_op().
^ This statement is incorrect as Hao has explained above.
kernel/bpf/verifier.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 857d76694517..44af69ce1301 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -4674,7 +4674,7 @@ static int check_stack_write_fixed_off(struct bpf_verifier_env *env, insn->imm != 0 && env->bpf_capable) { struct bpf_reg_state fake_reg = {};
__mark_reg_known(&fake_reg, (u32)insn->imm);
} else if (reg && is_spillable_regtype(reg->type)) {__mark_reg_known(&fake_reg, insn->imm); fake_reg.type = SCALAR_VALUE; save_register_state(state, spi, &fake_reg, size);
-- 2.34.1
Add a test to check if the verifier correctly reason about the sign of an immediate spilled to stack by BPF_ST instruction.
Signed-off-by: Hao Sun sunhao.th@gmail.com --- tools/testing/selftests/bpf/verifier/bpf_st_mem.c | 32 +++++++++++++++++++++++ 1 file changed, 32 insertions(+)
diff --git a/tools/testing/selftests/bpf/verifier/bpf_st_mem.c b/tools/testing/selftests/bpf/verifier/bpf_st_mem.c index 3af2501082b2..0ba23807c46c 100644 --- a/tools/testing/selftests/bpf/verifier/bpf_st_mem.c +++ b/tools/testing/selftests/bpf/verifier/bpf_st_mem.c @@ -65,3 +65,35 @@ .expected_attach_type = BPF_SK_LOOKUP, .runs = -1, }, +{ + "BPF_ST_MEM stack imm sign", + /* Check if verifier correctly reasons about sign of an + * immediate spilled to stack by BPF_ST instruction. + * + * fp[-8] = -44; + * r0 = fp[-8]; + * if r0 s< 0 goto ret0; + * r0 = -1; + * exit; + * ret0: + * r0 = 0; + * exit; + */ + .insns = { + BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, -44), + BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_10, -8), + BPF_JMP_IMM(BPF_JSLT, BPF_REG_0, 0, 2), + BPF_MOV64_IMM(BPF_REG_0, -1), + BPF_EXIT_INSN(), + BPF_MOV64_IMM(BPF_REG_0, 0), + BPF_EXIT_INSN(), + }, + /* Use prog type that requires return value in range [0, 1] */ + .prog_type = BPF_PROG_TYPE_SK_LOOKUP, + .expected_attach_type = BPF_SK_LOOKUP, + .result = VERBOSE_ACCEPT, + .runs = -1, + .errstr = "0: (7a) *(u64 *)(r10 -8) = -44 ; R10=fp0 fp-8_w=-44\ + 2: (c5) if r0 s< 0x0 goto pc+2\ + 2: R0_w=-44", +},
On Thu, 2023-10-26 at 17:13 +0200, Hao Sun wrote:
Immediate is incorrectly cast to u32 before being spilled, losing sign information. The range information is incorrect after load again. Fix immediate spill by remove the cast. The second patch add a test case for this.
Signed-off-by: Hao Sun sunhao.th@gmail.com
Thank you for finding and fixing this issue.
Acked-by: Eduard Zingerman eddyz87@gmail.com
Hao Sun (2): bpf: Fix check_stack_write_fixed_off() to correctly spill imm selftests/bpf: Add test for immediate spilled to stack
kernel/bpf/verifier.c | 2 +- tools/testing/selftests/bpf/verifier/bpf_st_mem.c | 32 +++++++++++++++++++++++ 2 files changed, 33 insertions(+), 1 deletion(-)
base-commit: 399f6185a1c02f39bcadb8749bc2d9d48685816f change-id: 20231026-fix-check-stack-write-c40996694dfa
Best regards,
linux-kselftest-mirror@lists.linaro.org