On Mon, May 18, 2020 at 1:07 PM John Fastabend <john.fastabend@xxxxxxxxx> wrote:
>
> When we have pointer type that is known to be non-null we only follow
> the non-null branch. This adds tests to cover the map_value pointer
> returned from a map lookup. To force an error if both branches are
> followed we do an ALU op on R10.
>
> Signed-off-by: John Fastabend <john.fastabend@xxxxxxxxx>
> ---
LGTM. Here likelihood of someone comparing map value pointer against
non-zero scalar is much less likely, so I won't bother you to add test
for that.
Acked-by: Andrii Nakryiko <andriin@xxxxxx>
> .../testing/selftests/bpf/verifier/value_or_null.c | 19 +++++++++++++++++++
> 1 file changed, 19 insertions(+)
>
> diff --git a/tools/testing/selftests/bpf/verifier/value_or_null.c b/tools/testing/selftests/bpf/verifier/value_or_null.c
> index 860d4a7..3ecb70a 100644
> --- a/tools/testing/selftests/bpf/verifier/value_or_null.c
> +++ b/tools/testing/selftests/bpf/verifier/value_or_null.c
> @@ -150,3 +150,22 @@
> .result_unpriv = REJECT,
> .flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
> },
> +{
> + "map lookup and null branch prediction",
> + .insns = {
> + BPF_MOV64_IMM(BPF_REG_1, 10),
> + BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_1, -8),
> + BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
> + BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
> + BPF_LD_MAP_FD(BPF_REG_1, 0),
> + BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
> + BPF_MOV64_REG(BPF_REG_6, BPF_REG_0),
> + BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 2),
> + BPF_JMP_IMM(BPF_JNE, BPF_REG_6, 0, 1),
> + BPF_ALU64_IMM(BPF_ADD, BPF_REG_10, 10),
> + BPF_EXIT_INSN(),
> + },
> + .fixup_map_hash_8b = { 4 },
> + .prog_type = BPF_PROG_TYPE_SCHED_CLS,
> + .result = ACCEPT,
> +},
>
