On Wed, 2007-06-06 at 16:07 +0800, Daniel wrote:

I think there is deadlock in tproxy. Two CPUs will wait for each other, ip_conntrack_lock and ip_nat_lock will never be unlocked. To reproduce this deadlock, we need ftp traffic(or other traffic which yields expect) and iptable_tproxy loaded.

Plz see call trace below:

CPU0 call trace:

run_timer_softirq <-- invoked by timer death_by_timeout <-- write_lock_bh(&ip_conntrack_lock) clean_from_lists remove_expectations unexpect_related __unexpect_related ip_conntrack_expect_put ip_conntrack_expect_destroyed <-- hooked by nat_reservation patch ####### #ifdef CONFIG_IP_NF_NAT_NRES ip_conntrack_expect_destroyed = &ip_nat_reserved_cleanup_expect; #endif ####### ip_nat_reserved_cleanup_expect ip_nat_reserved_unregister_all <-- write_lock_bh(&ip_nat_lock) ##failed## [waiting...]

CPU1 call trace:

ip_tproxy_fn ip_tproxy_setup_nat ip_tproxy_setup_nat_bidir <-- write_lock_bh(&ip_nat_lock) ip_nat_setup_info ip_conntrack_alter_reply <-- write_lock_bh(&ip_conntrack_lock) ##failed## [waiting...]

I see, we don't usually use conntrack helpers and tproxy together, as the functionality of the helper is performed by our proxies. I'll see what I can do. -- Bazsi