Re: [PATCH] bpf: fix divides by zero

[Alexei Starovoitov <alexei.starovoitov@gmail.com>]

On Fri, Jan 12, 2018 at 05:33:26PM -0800, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> Divides by zero are not nice, lets avoid them if possible.
> 
> Also do_div() seems not needed when dealing with 32bit operands,
> but this seems a minor detail.
> 
> Fixes: bd4cf0ed331a ("net: filter: rework/optimize internal BPF interpreter's instruction set")
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Reported-by: syzbot <syzkaller@googlegroups.com>
> ---
>  kernel/bpf/core.c |    4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> index
> 51ec2dda7f08c6c90af084589bb6d80662c77d12..7949e8b8f94e9cc196e0449214493
> ccce61b0903 100644
> --- a/kernel/bpf/core.c
> +++ b/kernel/bpf/core.c
> @@ -956,7 +956,7 @@ static unsigned int ___bpf_prog_run(u64 *regs,
> const struct bpf_insn *insn,
>  		DST = tmp;
>  		CONT;
>  	ALU_MOD_X:
> -		if (unlikely(SRC == 0))
> +		if (unlikely((u32)SRC == 0))

wow.
Acked-by: Alexei Starovoitov <ast@kernel.org>
we likely need to fix all JITs as well.
At least x64, arm64, sparc have the same bug.

Long term it's probably better to move all such checks out of JITs
and interpreter into the verifier and patch div/mod with
additional 'if src == 0'. This way we can do any type of
error reporting and/or aborting execution.