Re: [PATCH bpf-next] bpf: Emit explicit NULL pointer checks for PROBE_LDX instructions.

On Wed, Feb 03, 2021 at 12:56:39AM +0000, Song Liu wrote:

quoted

On Feb 1, 2021, at 9:38 PM, Alexei Starovoitov [off-list ref] wrote:

From: Alexei Starovoitov <ast@kernel.org>

PTR_TO_BTF_ID registers contain either kernel pointer or NULL.
Emit the NULL check explicitly by JIT instead of going into
do_user_addr_fault() on NULL deference.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
---
arch/x86/net/bpf_jit_comp.c | 19 +++++++++++++++++++
1 file changed, 19 insertions(+)

diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
index b7a2911bda77..a3dc3bd154ac 100644
--- a/arch/x86/net/bpf_jit_comp.c
+++ b/arch/x86/net/bpf_jit_comp.c

@@ -930,6 +930,7 @@ static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image,

		u32 dst_reg = insn->dst_reg;
		u32 src_reg = insn->src_reg;
		u8 b2 = 0, b3 = 0;
+		u8 *start_of_ldx;
		s64 jmp_offset;
		u8 jmp_cond;
		u8 *func;

@@ -1278,12 +1279,30 @@ st:			if (is_imm8(insn->off))

		case BPF_LDX | BPF_PROBE_MEM | BPF_W:
		case BPF_LDX | BPF_MEM | BPF_DW:
		case BPF_LDX | BPF_PROBE_MEM | BPF_DW:
+			if (BPF_MODE(insn->code) == BPF_PROBE_MEM) {
+				/* test src_reg, src_reg */
+				maybe_emit_mod(&prog, src_reg, src_reg, true); /* always 1 byte */
+				EMIT2(0x85, add_2reg(0xC0, src_reg, src_reg));
+				/* jne start_of_ldx */
+				EMIT2(X86_JNE, 0);
+				/* xor dst_reg, dst_reg */
+				emit_mov_imm32(&prog, false, dst_reg, 0);
+				/* jmp byte_after_ldx */
+				EMIT2(0xEB, 0);
+
+				/* populate jmp_offset for JNE above */
+				temp[4] = prog - temp - 5 /* sizeof(test + jne) */;

IIUC, this case only happens for i == 1 in the loop? If so, can we use temp[5(?)] 
instead of start_of_ldx?

I don't understand the question, but let me try anyway :)
temp is a buffer for single instruction.
prog=temp; for every loop iteration (not only i == 1)
temp[4] is second byte in JNE instruction as the comment says.
temp[5] is a byte after JNE. It's a first byte of XOR.
That XOR is variable length instruction. 
Hence while emitting JNE we don't know the target offset in JNE and just use 0.
So temp[4] assignment populates with actual offset, since now we know the size
of XOR.