Re: [PATCH] bpf: Reject BPF_MAP_TYPE_INSN_ARRAY in check_reg_const_str()

[Anton Protopopov <a.s.protopopov@gmail.com>]

On 26/01/07 07:40AM, Deepanshu Kartikey wrote:
> BPF_MAP_TYPE_INSN_ARRAY maps store instruction pointers in their
> ips array, not string data. The map_direct_value_addr callback for
> this map type returns the address of the ips array, which is not
> suitable for use as a constant string argument.
> 
> When a BPF program passes a pointer to an insn_array map value as
> ARG_PTR_TO_CONST_STR (e.g., to bpf_snprintf), the verifier's
> null-termination check in check_reg_const_str() operates on the
> wrong memory region, and at runtime bpf_bprintf_prepare() can read
> out of bounds searching for a null terminator.
> 
> Reject BPF_MAP_TYPE_INSN_ARRAY in check_reg_const_str() since this
> map type is not designed to hold string data.
> 
> Reported-by: syzbot+2c29addf92581b410079@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=2c29addf92581b410079
> Tested-by: syzbot+2c29addf92581b410079@syzkaller.appspotmail.com
> Fixes: 493d9e0d6083 ("bpf, x86: add support for indirect jumps")
> Signed-off-by: Deepanshu Kartikey <kartikey406@gmail.com>
> ---
>  kernel/bpf/verifier.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index f0ca69f888fa..3135643d5695 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -9609,6 +9609,11 @@ static int check_reg_const_str(struct bpf_verifier_env *env,
>  	if (reg->type != PTR_TO_MAP_VALUE)
>  		return -EINVAL;
>  
> +	if (map->map_type == BPF_MAP_TYPE_INSN_ARRAY) {
> +		verbose(env, "R%d points to insn_array map which cannot be used as const string\n", regno);
> +		return -EACCES;
> +	}
> +
>  	if (!bpf_map_is_rdonly(map)) {
>  		verbose(env, "R%d does not point to a readonly map'\n", regno);
>  		return -EACCES;
> -- 
> 2.43.0
> 

Acked-by: Anton Protopopov <a.s.protopopov@gmail.com>