Skip to content

Commit

Permalink
netfilter: ctnetlink: fix race between delete and timeout expiration
Browse files Browse the repository at this point in the history
Kerin Millar reported hardlockups while running `conntrackd -c'
in a busy firewall. That system (with several processors) was
acting as backup in a primary-backup setup.

After several tries, I found a race condition between the deletion
operation of ctnetlink and timeout expiration. This patch fixes
this problem.

Tested-by: Kerin Millar <kerframil@gmail.com>
Reported-by: Kerin Millar <kerframil@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
  • Loading branch information
ummakynes authored and davem330 committed Mar 17, 2012
1 parent c577923 commit a16a164
Showing 1 changed file with 12 additions and 11 deletions.
23 changes: 12 additions & 11 deletions net/netfilter/nf_conntrack_netlink.c
Expand Up @@ -943,20 +943,21 @@ ctnetlink_del_conntrack(struct sock *ctnl, struct sk_buff *skb,
}
}

if (nf_conntrack_event_report(IPCT_DESTROY, ct,
NETLINK_CB(skb).pid,
nlmsg_report(nlh)) < 0) {
if (del_timer(&ct->timeout)) {
if (nf_conntrack_event_report(IPCT_DESTROY, ct,
NETLINK_CB(skb).pid,
nlmsg_report(nlh)) < 0) {
nf_ct_delete_from_lists(ct);
/* we failed to report the event, try later */
nf_ct_insert_dying_list(ct);
nf_ct_put(ct);
return 0;
}
/* death_by_timeout would report the event again */
set_bit(IPS_DYING_BIT, &ct->status);
nf_ct_delete_from_lists(ct);
/* we failed to report the event, try later */
nf_ct_insert_dying_list(ct);
nf_ct_put(ct);
return 0;
}

/* death_by_timeout would report the event again */
set_bit(IPS_DYING_BIT, &ct->status);

nf_ct_kill(ct);
nf_ct_put(ct);

return 0;
Expand Down

0 comments on commit a16a164

Please sign in to comment.