[PATCH bpf 0/2] bpf: clear zext_dst of dead insns

[PATCH bpf 0/2] bpf: clear zext_dst of dead insns

[Ilya Leoshkevich]

Fix the "verifier bug. zext_dst is set, but no reg is defined" failure
in the "access skb fields ok" test on s390.

Patch 1 is the fix, patch 2 adds a test.

v1: https://lore.kernel.org/bpf/20210812111220.181824-1-iii@linux.ibm.com/
v1 -> v2: Rebase to bpf branch, add Fixes:, add a test (Daniel).

Ilya Leoshkevich (2):
  bpf: clear zext_dst of dead insns
  selftests: bpf: test that dead ldx_w insns are accepted

 kernel/bpf/verifier.c                            |  1 +
 tools/testing/selftests/bpf/verifier/dead_code.c | 13 +++++++++++++
 2 files changed, 14 insertions(+)

-- 
2.31.1

[PATCH bpf 1/2] bpf: clear zext_dst of dead insns

[Ilya Leoshkevich]

"access skb fields ok" verifier test fails on s390 with the "verifier
bug. zext_dst is set, but no reg is defined" message. The first insns
of the test prog are:

   0:	61 01 00 00 00 00 00 00 	ldxw %r0,[%r1+0]
   8:	35 00 00 01 00 00 00 00 	jge %r0,0,1
  10:	61 01 00 08 00 00 00 00 	ldxw %r0,[%r1+8]

and the 3rd one is dead (this does not look intentional to me, but this
is a separate topic).

sanitize_dead_code() converts dead insns into "ja -1", but keeps
zext_dst. When opt_subreg_zext_lo32_rnd_hi32() tries to parse such
an insn, it sees this discrepancy and bails. This problem can be seen
only with JITs whose bpf_jit_needs_zext() returns true.

Fix by clearning dead insns' zext_dst.

The commits that contributed to this problem are:

1. commit 5aa5bd14c5f8 ("bpf: add initial suite for selftests"), which
   introduced the test with the dead code.
2. commit 5327ed3d44b7 ("bpf: verifier: mark verified-insn with
   sub-register zext flag"), which introduced the zext_dst flag.
3. commit 83a2881903f3 ("bpf: Account for BPF_FETCH in
   insn_has_def32()"), which introduced the sanity check.
4. commit 9183671af6db ("bpf: Fix leakage under speculation on
   mispredicted branches"), which bisect points to.

It's best to fix this on stable branches that contain the second one,
since that's the point where the inconsistency was introduced.

Fixes: 5327ed3d44b7 ("bpf: verifier: mark verified-insn with sub-register zext flag")
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
---
 kernel/bpf/verifier.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index f9bda5476ea5..381d3d6f24bc 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -11663,6 +11663,7 @@ static void sanitize_dead_code(struct bpf_verifier_env *env)
 		if (aux_data[i].seen)
 			continue;
 		memcpy(insn + i, &trap, sizeof(trap));
+		aux_data[i].zext_dst = false;
 	}
 }
 
-- 
2.31.1

[PATCH bpf 2/2] selftests: bpf: test that dead ldx_w insns are accepted

[Ilya Leoshkevich]

Prevent regressions related to zero-extension metadata handling during
dead code sanitization.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
---
 tools/testing/selftests/bpf/verifier/dead_code.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/tools/testing/selftests/bpf/verifier/dead_code.c b/tools/testing/selftests/bpf/verifier/dead_code.c
index 2c8935b3e65d..c642138b7fc2 100644
--- a/tools/testing/selftests/bpf/verifier/dead_code.c
+++ b/tools/testing/selftests/bpf/verifier/dead_code.c
@@ -159,3 +159,16 @@
 	.result = ACCEPT,
 	.retval = 2,
 },
+{
+	"dead code: zero extension",
+	.insns = {
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_JMP_IMM(BPF_JGE, BPF_REG_0, 0, 1),
+	BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_10, 0),
+	BPF_EXIT_INSN(),
+	},
+	.errstr_unpriv = "invalid read from stack R10 off=0 size=4",
+	.result_unpriv = REJECT,
+	.result = ACCEPT,
+	.retval = 0,
+},
-- 
2.31.1

Re: [PATCH bpf 2/2] selftests: bpf: test that dead ldx_w insns are accepted

[Ilya Leoshkevich]

On Thu, 2021-08-12 at 16:05 +0200, Ilya Leoshkevich wrote:
> Prevent regressions related to zero-extension metadata handling during
> dead code sanitization.
> 
> Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
> ---
>  tools/testing/selftests/bpf/verifier/dead_code.c | 13 +++++++++++++
>  1 file changed, 13 insertions(+)
> 
> diff --git a/tools/testing/selftests/bpf/verifier/dead_code.c
> b/tools/testing/selftests/bpf/verifier/dead_code.c
> index 2c8935b3e65d..c642138b7fc2 100644
> --- a/tools/testing/selftests/bpf/verifier/dead_code.c
> +++ b/tools/testing/selftests/bpf/verifier/dead_code.c
> @@ -159,3 +159,16 @@
>         .result = ACCEPT,
>         .retval = 2,
>  },
> +{
> +       "dead code: zero extension",
> +       .insns = {
> +       BPF_MOV64_IMM(BPF_REG_0, 0),
> +       BPF_JMP_IMM(BPF_JGE, BPF_REG_0, 0, 1),
> +       BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_10, 0),
> +       BPF_EXIT_INSN(),
> +       },
> +       .errstr_unpriv = "invalid read from stack R10 off=0 size=4",
> +       .result_unpriv = REJECT,
> +       .result = ACCEPT,
> +       .retval = 0,
> +},

Please disregard this patch: the test does not fail in absence of the
fix. What rather fails is:

	BPF_MOV64_IMM(BPF_REG_0, 0),
	BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_0, -4),
	BPF_JMP_IMM(BPF_JGE, BPF_REG_0, 0, 1),
	BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_10, -4),
	BPF_EXIT_INSN(),

The difference is that here the dead ldx_w is actually safe. I will
send a v3 shortly (I also realized I forgot to tag this series with
v2).