On Sat, Jan 2, 2016 at 12:03 AM, Dmitry Vyukov [off-list ref] wrote:
On Sat, Jan 2, 2016 at 1:34 AM, Cong Wang [off-list ref] wrote:
quoted
llcp_sock_getname() checks llcp_sock->dev to make sure
llcp_sock is already connected or bound, however, we could
be in the middle of llcp_sock_bind() where llcp_sock->dev
is bound and llcp_sock->service_name_len is set,
but llcp_sock->service_name is not, in this case we would
lead to copy some bytes from a NULL pointer.
We should just check if sk->sk_state is still closed since
both connect() and bind() will update this state at the end.
Hi Cong,
This is still racy. If you want to play lock-free then you also need
proper memory barriers. Stores to sk_state need to be
smp_store_release, while the load needs to be smp_load_acquire.
Otherwise getname still can see partially initialized socket.
Right... Or just lock sock perhaps. I will update my patch.
Thanks!