2.4.0-pre3+zerocopy: weird messages

I get messages in syslog looking like:

Undo loss 192.147.174.183/59953 c2 l0 ss2/65535 p0 Undo loss 63.148.232.53/4423 c2 l0 ss2/2 p0 Undo loss 204.253.105.63/25 c2 l0 ss2/2 p0

This is on a dual UltraSPARC box with a Happy Meal network card. I don't think either Postfix or tinydns (the main load) are using sendfile().

The machine is stable so far, but it's been up for only 15 minutes.

--
Petru Paler, mailto:[email protected]
http://www.ppetru.net - ICQ: 41817235
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

These are normal, if they annoy you please change FASTRETRANS_DEBUG back to "1" in include/net/tcp.h

This is just an increased debugging setting compared to Linus's tree, the message you see is harmless.

Later,
David S. Miller
[email protected]
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

Ok. Should I keep reporting new syslog messages as they appear ? This machine has lots of traffic, both TCP (SMTP) and UDP (DNS). Since the last email I also got (minus the "Undo loss" ones, and I only included one of each message types, as they repeat):

Undo partial loss 193.230.129.57/33659 c1 l5 ss2/3 p5

udp v4 hw csum failure.

Disorder0 3 5 f0 s1 rr1

Undo Hoe 203.162.5.28/25 c8 l1 ss5/65535 p8

--
Petru Paler, mailto:[email protected]
http://www.ppetru.net - ICQ: 41817235
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

Not the "Undo ***" and "Disorder ***" ones".

But this one is curious:

Oh, I think I know why this happens. Can you add this patch, and next time the UDP bad csum message appears, tell me if it says "UDP packet with bad csum was fragmented." in the next line of your syslog messages? Thanks.

Later,
David S. Miller
[email protected]
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

Ok.

Sure, but I also need the actual patch :)

--
Petru Paler, mailto:[email protected]
http://www.ppetru.net - ICQ: 41817235
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

Duh, here it is :-)

@@ -855,6 +855,8 @@
 		if (!udp_check(uh, ulen, saddr, daddr, skb->csum))
 			return 0;
 		NETDEBUG(printk(KERN_DEBUG "udp v4 hw csum failure.\n"));
+		if (skb_shinfo(skb)->frag_list != NULL)
+			printk(KERN_DEBUG "UDP packet with bad csum was fragmented.\n");
 		skb->ip_summed = CHECKSUM_NONE;
 	}
 	if (skb->ip_summed != CHECKSUM_UNNECESSARY)

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

Jan 14 06:54:08 grey kernel: Undo loss 193.230.129.57/34342 c2 l0 ss2/2 p0
Jan 14 06:56:40 grey kernel: udp v4 hw csum failure.
Jan 14 06:57:05 grey kernel: Undo partial loss 193.230.129.57/34342 c1 l5 ss2/3 p5        

So no "UDP packet with bad csum was fragmented" line. This is the first one though, will let you know if the fragmented thing occurs.

--
Petru Paler, mailto:[email protected]
http://www.ppetru.net - ICQ: 41817235
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

Got more "udp v4 hw csum failure" messages but still no "UDP packet with bad csum was fragmented".

--
Petru Paler, mailto:[email protected]
http://www.ppetru.net - ICQ: 41817235
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

OK, last experiment :-) Add this patch, and watch to see if the UDP "InErrors" field in /proc/net/snmp has a non-zero value after letting it run for a while. Thanks.

@@ -2109,11 +2109,11 @@
 
 			skb = copy_skb;
 		}
-
+#if 0
 		/* This card is _fucking_ hot... */
 		skb->csum = (csum ^ 0xffff);
 		skb->ip_summed = CHECKSUM_HW;
-
+#endif
 		RXD(("len=%d csum=%4x]", len, csum));
 		skb->protocol = eth_type_trans(skb, dev);
 		netif_rx(skb);

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

Ok, will do.

In the mean time, the box locked up hard. The last message in syslog was:

Jan 14 10:14:45 grey kernel: Undo loss 194.67.160.18/3103 c2 l0 ss2/65535 p0              

I'm not sure when it died, and I also could not check the console (the server is on the other side of the planet for me :).

--
Petru Paler, mailto:[email protected]
http://www.ppetru.net - ICQ: 41817235
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

Ok, here it is:

grey:/proc/net# uptime
 12:14am  up 11:01,  1 user,  load average: 0.45, 0.47, 0.41
grey:/proc/net# cat snmp
Ip: Forwarding DefaultTTL InReceives InHdrErrors InAddrErrors ForwDatagrams InUnknownProtos InDiscards InDelivers OutRequests OutDiscards OutNoRoutes ReasmTimeout ReasmReqds ReasmOKs ReasmFails FragOKs FragFails FragCreates
Ip: 2 64 8340770 0 0 0 0 8 8290517 1359445 6 0 0 0 0 0 0 0 0
Icmp: InMsgs InErrors InDestUnreachs InTimeExcds InParmProbs InSrcQuenchs InRedirects InEchos InEchoReps InTimestamps InTimestampReps InAddrMasks InAddrMaskReps OutMsgs OutErrors OutDestUnreachs OutTimeExcds OutParmProbs OutSrcQuenchs OutRedirects OutEchos OutEchoReps OutTimestamps OutTimestampReps OutAddrMasks OutAddrMaskReps
Icmp: 7370 446 3846 772 0 39 0 2708 0 0 0 0 0 9661 0 6953 0 0 0 0 0 2708 0 0 0 0
Tcp: RtoAlgorithm RtoMin RtoMax MaxConn ActiveOpens PassiveOpens AttemptFails EstabResets CurrEstab InSegs OutSegs RetransSegs InErrs OutRsts
Tcp: 0 0 0 0 16025 0 128 0 20 825925 921717 44521 189 28300
Udp: InDatagrams NoPorts InErrors OutDatagrams
Udp: 7500511 6953 30 428614

--
Petru Paler, mailto:[email protected]
http://www.ppetru.net - ICQ: 41817235
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/