Kernel 2.6.14 and old versions of traceroute

Kernel 2.6.14 and old versions of traceroute

[David R]

I've noticed that old versions of traceroute no longer work properly
with the latest kernel. 2.6.13.4 is OK. I've done a bit of strace and am
posting the differences here. These are from a 64 bit kernel using
traceroute 0.6.2 as shipped with most versions of SuSE. I have confirmed
that the same problem is present in a 32 bit kernel on a different
machine. A later traceroute 1.4a12 works properly.

2.6.13.4

poll([{fd=6, events=POLLERR, revents=POLLERR}, {fd=3, events=POLLERR},
{fd=4, events=POLLERR}, {fd=5, events=POLLERR}], 4, 1) = 1
recvmsg(6, {msg_name(16)={sa_family=AF_INET, sin_port=htons(33443),
sin_addr=inet_addr("69.10.132.115")}, msg_iov(0)=[], msg_controllen=80,
{cmsg_len=32, cmsg_level=SOL_SOCKET, cmsg_type=0x1d /* SCM_??? */, ...},
msg_flags=MSG_ERRQUEUE}, MSG_ERRQUEUE) = 0


2.6.14
poll([{fd=3, events=POLLERR, revents=POLLERR}, {fd=4, events=POLLERR},
{fd=5, events=POLLERR}, {fd=6, events=POLLERR}, {fd=7, events=POLLERR},
{fd=8, events=POLLERR}], 6, 1) = 1
recvmsg(3, 0x7fffffa30960, MSG_ERRQUEUE) = -1 EAGAIN (Resource
temporarily unavailable)
....
....
poll([{fd=3, events=POLLERR, revents=POLLERR}, {fd=4, events=POLLERR,
revents=POLLERR}, {fd=5, events=POLLERR}, {fd=6, events=POLLERR}, {fd=7,
events=POLLERR}, {fd=8, events=POLLERR}, {fd=9, events=POLLERR}], 7, 1) = 2
recvmsg(3, 0x7fffffa30960, MSG_ERRQUEUE) = -1 EAGAIN (Resource
temporarily unavailable)
recvmsg(4, 0x7fffffa30960, MSG_ERRQUEUE) = -1 EFAULT (Bad address)

I'm up for more diagnostics if necessary.

Cheers
David

Re: Kernel 2.6.14 and old versions of traceroute

[Chris Boot]

David R wrote:
> I've noticed that old versions of traceroute no longer work properly
> with the latest kernel. 2.6.13.4 is OK. I've done a bit of strace and am
> posting the differences here. These are from a 64 bit kernel using
> traceroute 0.6.2 as shipped with most versions of SuSE. I have confirmed
> that the same problem is present in a 32 bit kernel on a different
> machine. A later traceroute 1.4a12 works properly.
> 
> 2.6.13.4
> 
> poll([{fd=6, events=POLLERR, revents=POLLERR}, {fd=3, events=POLLERR},
> {fd=4, events=POLLERR}, {fd=5, events=POLLERR}], 4, 1) = 1
> recvmsg(6, {msg_name(16)={sa_family=AF_INET, sin_port=htons(33443),
> sin_addr=inet_addr("69.10.132.115")}, msg_iov(0)=[], msg_controllen=80,
> {cmsg_len=32, cmsg_level=SOL_SOCKET, cmsg_type=0x1d /* SCM_??? */, ...},
> msg_flags=MSG_ERRQUEUE}, MSG_ERRQUEUE) = 0
> 
> 
> 2.6.14
> poll([{fd=3, events=POLLERR, revents=POLLERR}, {fd=4, events=POLLERR},
> {fd=5, events=POLLERR}, {fd=6, events=POLLERR}, {fd=7, events=POLLERR},
> {fd=8, events=POLLERR}], 6, 1) = 1
> recvmsg(3, 0x7fffffa30960, MSG_ERRQUEUE) = -1 EAGAIN (Resource
> temporarily unavailable)
> ....
> ....
> poll([{fd=3, events=POLLERR, revents=POLLERR}, {fd=4, events=POLLERR,
> revents=POLLERR}, {fd=5, events=POLLERR}, {fd=6, events=POLLERR}, {fd=7,
> events=POLLERR}, {fd=8, events=POLLERR}, {fd=9, events=POLLERR}], 7, 1) = 2
> recvmsg(3, 0x7fffffa30960, MSG_ERRQUEUE) = -1 EAGAIN (Resource
> temporarily unavailable)
> recvmsg(4, 0x7fffffa30960, MSG_ERRQUEUE) = -1 EFAULT (Bad address)
> 
> I'm up for more diagnostics if necessary.
> 
> Cheers
> David

Smells suspiciously similar to the BIND trouble that's been reported 
here in the last few days:

http://lkml.org/lkml/2005/10/29/247
http://lkml.org/lkml/2005/10/30/32

You might like to look into it on the netdev list.

HTH,
Chris

-- 
Chris Boot
bootc@bootc.net
http://www.bootc.net/

Re: Kernel 2.6.14 and old versions of traceroute

[Steffen Moser]

Hi David,

* On Mon, Oct 31, 2005 at 07:30 PM (+0000), David R wrote:

> I've noticed that old versions of traceroute no longer work properly
> with the latest kernel. 2.6.13.4 is OK. I've done a bit of strace and am
> posting the differences here. These are from a 64 bit kernel using
> traceroute 0.6.2 as shipped with most versions of SuSE. 

I am experiencing exactly the same problem with traceroute-0.6.2 
running on SuSE 10.0 together with kernel 2.6.14. The whole thing
happens on a single core dual Opteron machine. I've tested this
kernel version on that machine only, yet.

I've also tried the latest traceroute version (1.0.2) from

  ftp://ftp.lst.de/pub/people/okir/traceroute/

and experienced the same behaviour.

Olaf Kirch has just sent me a patch against 2.6.14. It has also 
been discussed in NETDEV. 

This fixed it for me:

--- a/net/core/datagram.c       2005-11-01 11:38:31.000000000 +0100
+++ b/net/core/datagram.c       2005-11-01 11:38:45.000000000 +0100
@@ -213,6 +213,10 @@
 {
        int i, err, fraglen, end = 0;
        struct sk_buff *next = skb_shinfo(skb)->frag_list;
+
+       if (!len)
+               return 0;
+
 next_skb:
        fraglen = skb_headlen(skb);
        i = -1;

Bye,
Steffen