Re: [PATCH 6/7] bpf: Add instructions for atomic_cmpxchg and friends

[Brendan Jackman <jackmanb@google.com>]

Looks like I half-baked some last-minute changes before sending this
out. If you disable the JIT, test_verifier will fail.

On Mon, Nov 23, 2020 at 05:32:01PM +0000, Brendan Jackman wrote:
> diff --git a/include/linux/filter.h b/include/linux/filter.h
...
> +#define BPF_ATOMIC_SET(SIZE, DST, SRC, OFF)			\
> +	((struct bpf_insn) {					\
> +		.code  = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC,	\
> +		.dst_reg = DST,					\
> +		.src_reg = SRC,					\
> +		.off   = OFF,					\
> +		.imm   = BPF_SET })
> +

Should be deleted, and in the tools/ copy too. There's a test case in
the later commit that should be changed to assert that instructions like
this fail to verify.

> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 14f5053daf22..2e611d3695bf 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -3602,10 +3602,14 @@ static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_i
>  {
>  	struct bpf_reg_state *regs = cur_regs(env);
>  	int err;
> +	int load_reg;
>  
>  	switch (insn->imm) {
>  	case BPF_ADD:
>  	case BPF_ADD | BPF_FETCH:
> +	case BPF_SET:

Should be deleted.

> +	case BPF_SET | BPF_FETCH:
> +	case BPF_CMPSET | BPF_FETCH: /* CMPSET without FETCH is not supported */
>  		break;
>  	default:
>  		verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn->imm);

I'll fold the fix into the next revision along with the comments
generated by this initial version.