nf_reinject() and ip_queue -- hmm...

[Alexander Demenshin]

Hello folks,

	So I've found another problem... So far only description - will dig
	in later...
	
	In case of (I assume clean tables):
	
		iptables -A INPUT -d 127.1.1.1 -j QUEUE
		iptables -A OUTPUT -d 127.1.1.1 -j QUEUE
		iptables -t mangle -A OUTPUT -d 127.1.1.1 -j QUEUE
	
	Then:
	
		ping -c1 127.1.1.1
	
	I get OOPS. Without QUEUE in mangle table it is OK,
	without QUEUE in filter table it is also OK.
	
	(OOPS is in net/ipv4/ip_output:output_maybe_reroute(), called from nf_reinject()).
	
	Problem exists in test2-ac2 thru test4. Even though system _looks_ usable after
	that, it is in fact unstable (so better to reboot - I had very bad experience).
	
	Moreover... Concerning recent change to netfilter.c:
	

====	
diff -u --recursive --new-file v2.4.0-test3/linux/net/core/netfilter.c linux/net/core/netfilter.c
--- v2.4.0-test3/linux/net/core/netfilter.c     Wed Apr 26 16:34:09 2000
+++ linux/net/core/netfilter.c  Tue Jul 11 19:02:37 2000
@@ -512,7 +512,6 @@
                                     info->indev, info->outdev, &elem,
                                     info->okfn);
        }
-       br_read_unlock_bh(BR_NETPROTO_LOCK);
 
        switch (verdict) {
        case NF_ACCEPT:
@@ -527,6 +526,7 @@
                kfree_skb(skb);
                break;
        }
+       br_read_unlock_bh(BR_NETPROTO_LOCK);
 
        /* Release those devices we held, or Alexey will kill me. */
        if (info->indev) dev_put(info->indev);


====
	When it is in effect, I've something more terrible (additionally
	to OOPS) - freezing of kernel with some backtrack (too many
	lines), sometimes after this SysRq doesn't work too.
	
	I've some suspects, but... I need time, of course :)
	
	Good luck!
	
	(sometimes I am going to hate Linux - too many reboots, but I had
	no one so far on my Win2K :))
	
/Al