bpf: arsh is not supported in 32 bit alu thus reject it
authorDaniel Borkmann <daniel@iogearbox.net>
Mon, 29 Jan 2018 01:48:57 +0000 (02:48 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 31 Jan 2018 11:55:57 +0000 (12:55 +0100)
[ upstream commit 7891a87efc7116590eaba57acc3c422487802c6f ]

The following snippet was throwing an 'unknown opcode cc' warning
in BPF interpreter:

  0: (18) r0 = 0x0
  2: (7b) *(u64 *)(r10 -16) = r0
  3: (cc) (u32) r0 s>>= (u32) r0
  4: (95) exit

Although a number of JITs do support BPF_ALU | BPF_ARSH | BPF_{K,X}
generation, not all of them do and interpreter does neither. We can
leave existing ones and implement it later in bpf-next for the
remaining ones, but reject this properly in verifier for the time
being.

Fixes: 17a5267067f3 ("bpf: verifier (add verifier core)")
Reported-by: syzbot+93c4904c5c70348a6890@syzkaller.appspotmail.com
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
kernel/bpf/verifier.c

index 19c44cf..8e0bf89 100644 (file)
@@ -1843,6 +1843,11 @@ static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
                        return -EINVAL;
                }
 
+               if (opcode == BPF_ARSH && BPF_CLASS(insn->code) != BPF_ALU64) {
+                       verbose("BPF_ARSH not supported for 32 bit ALU\n");
+                       return -EINVAL;
+               }
+
                if ((opcode == BPF_LSH || opcode == BPF_RSH ||
                     opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
                        int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;