Re: net/dccp: dccp_create_openreq_child freed held lock

[Dmitry Vyukov]

On Wed, Mar 1, 2017 at 4:35 PM, Arnaldo Carvalho de Melo
<acme@xxxxxxxxxx> wrote:
> Em Wed, Mar 01, 2017 at 10:38:54AM +0100, Dmitry Vyukov escreveu:
>> Hello,
>>
>> I've got the following report while running syzkaller fuzzer on
>> 86292b33d4b79ee03e2f43ea0381ef85f077c760:
>>
>>
>> It seems that dccp_create_openreq_child needs to unlock the sock if
>> dccp_feat_activate_values fails.
>
> Yeah, can you please use the patch below, that mimics the error paths in
> sk_clone_new(), from where I think even the comment about it being a raw
> copy came, but the bh_unlock_sock() didn't?
>
> - Arnaldo
>
> diff --git a/net/dccp/minisocks.c b/net/dccp/minisocks.c
> index 53eddf99e4f6..d20d948a98ed 100644
> --- a/net/dccp/minisocks.c
> +++ b/net/dccp/minisocks.c
> @@ -122,6 +122,7 @@ struct sock *dccp_create_openreq_child(const struct sock *sk,
>                         /* It is still raw copy of parent, so invalidate
>                          * destructor and make plain sk_free() */
>                         newsk->sk_destruct = NULL;
> +                       bh_unlock_sock(newsk);
>                         sk_free(newsk);
>                         return NULL;
>                 }


Applied the patch on bots. Will report if it happens again.