sock_map: Fix a potential use-after-free in sock_map_close()
authorCong Wang <cong.wang@bytedance.com>
Thu, 8 Apr 2021 03:05:56 +0000 (20:05 -0700)
committerDaniel Borkmann <daniel@iogearbox.net>
Mon, 12 Apr 2021 15:35:26 +0000 (17:35 +0200)
commitaadb2bb83ff789de63b48b4edeab7329423a50d3
tree07700d24aae0e305a821a073a1662967172377df
parent51e0158a54321a48d260e95998393934bb0de52c
sock_map: Fix a potential use-after-free in sock_map_close()

The last refcnt of the psock can be gone right after
sock_map_remove_links(), so sk_psock_stop() could trigger a UAF.
The reason why I placed sk_psock_stop() there is to avoid RCU read
critical section, and more importantly, some callee of
sock_map_remove_links() is supposed to be called with RCU read lock,
we can not simply get rid of RCU read lock here. Therefore, the only
choice we have is to grab an additional refcnt with sk_psock_get()
and put it back after sk_psock_stop().

Fixes: 799aa7f98d53 ("skmsg: Avoid lock_sock() in sk_psock_backlog()")
Reported-by: syzbot+7b6548ae483d6f4c64ae@syzkaller.appspotmail.com
Signed-off-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/bpf/20210408030556.45134-1-xiyou.wangcong@gmail.com
net/core/sock_map.c