bpf: fix incorrect sign extension in check_alu_op()

Distinguish between
BPF_ALU64|BPF_MOV|BPF_K (load 32-bit immediate, sign-extended to 64-bit)
and BPF_ALU|BPF_MOV|BPF_K (load 32-bit immediate, zero-padded to 64-bit);
only perform sign extension in the first case.

Starting with v4.14, this is exploitable by unprivileged users as long as
the unprivileged_bpf_disabled sysctl isn't set.

Debian assigned CVE-2017-16995 for this issue.

v3:
 - add CVE number (Ben Hutchings)

Fixes: 484611357c ("bpf: allow access into map value arrays")
Signed-off-by: Jann Horn <[email protected]>
Acked-by: Edward Cree <[email protected]>
Signed-off-by: Alexei Starovoitov <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>

kernel/bpf/verifier.c

@@ -2408,7 +2408,13 @@ static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
 			 * remember the value we stored into this reg
 			 */
 			regs[insn->dst_reg].type = SCALAR_VALUE;
-			__mark_reg_known(regs + insn->dst_reg, insn->imm);
+			if (BPF_CLASS(insn->code) == BPF_ALU64) {
+				__mark_reg_known(regs + insn->dst_reg,
+						 insn->imm);
+			} else {
+				__mark_reg_known(regs + insn->dst_reg,
+						 (u32)insn->imm);
+			}
 		}
 
 	} else if (opcode > BPF_END) {