netfilter: sync with packet rx also after removing queue entries

We need to sync packet rx again after flushing the queue entries.
Otherwise, the following race could happen:

cpu1: nf_unregister_hook(H) called, H unliked from lists, calls
synchronize_net() to wait for packet rx completion.

Problem is that while no new nf_queue_entry structs that use H can be
allocated, another CPU might receive a verdict from userspace just before
cpu1 calls nf_queue_nf_hook_drop to remove this entry:

cpu2: receive verdict from userspace, lock queue
cpu2: unlink nf_queue_entry struct E, which references H, from queue list
cpu1: calls nf_queue_nf_hook_drop, blocks on queue spinlock
cpu2: unlock queue
cpu1: nf_queue_nf_hook_drop drops affected queue entries
cpu2: call nf_reinject for E
cpu1: kfree(H)
cpu2: potential use-after-free for H

Cc: Eric W. Biederman <ebiederm@xmission.com>
Fixes: 085db2c04557 ("netfilter: Per network namespace netfilter hooks.")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

diff --git a/net/netfilter/core.c b/net/netfilter/core.c
index 8e47f8113495739082572d269797e00160251751..21a085686dc1b543439f33e448531bc64a273684 100644 (file)
--- a/net/netfilter/core.c
+++ b/net/netfilter/core.c
@@ -152,6 +152,8 @@ void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 #endif
        synchronize_net();
        nf_queue_nf_hook_drop(net, &entry->ops);
+       /* other cpu might still process nfqueue verdict that used reg */
+       synchronize_net();
        kfree(entry);
 }
 EXPORT_SYMBOL(nf_unregister_net_hook);