netfilter: conntrack: fix rmmod double-free race
authorFlorian Westphal <fw@strlen.de>
Tue, 14 Feb 2023 16:23:59 +0000 (17:23 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 11 Mar 2023 12:55:24 +0000 (13:55 +0100)
commit1fd3c69f6511cc0785d5f5f828a5c095d74df69f
tree063dafb46f9f01b1f8f7469c690ca670b61f5bd6
parent4f25d1dff80535f088b8f8568dd731fb098e29b4
netfilter: conntrack: fix rmmod double-free race

[ Upstream commit e6d57e9ff0aec323717ee36fc9ea34ad89217151 ]

nf_conntrack_hash_check_insert() callers free the ct entry directly, via
nf_conntrack_free.

This isn't safe anymore because
nf_conntrack_hash_check_insert() might place the entry into the conntrack
table and then delteted the entry again because it found that a conntrack
extension has been removed at the same time.

In this case, the just-added entry is removed again and an error is
returned to the caller.

Problem is that another cpu might have picked up this entry and
incremented its reference count.

This results in a use-after-free/double-free, once by the other cpu and
once by the caller of nf_conntrack_hash_check_insert().

Fix this by making nf_conntrack_hash_check_insert() not fail anymore
after the insertion, just like before the 'Fixes' commit.

This is safe because a racing nf_ct_iterate() has to wait for us
to release the conntrack hash spinlocks.

While at it, make the function return -EAGAIN in the rmmod (genid
changed) case, this makes nfnetlink replay the command (suggested
by Pablo Neira).

Fixes: c56716c69ce1 ("netfilter: extensions: introduce extension genid count")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
net/netfilter/nf_conntrack_bpf.c
net/netfilter/nf_conntrack_core.c
net/netfilter/nf_conntrack_netlink.c