Bug 68072

Thank you for this very useful report. The fact that 2.4.9-XX work and 2.4.18
doesn't is valueable. The changes to the pppoe code between those versions is
rather small and I'll just revert to the 2.4.9 version.

I've seen something similar to this problem.  It's not exactly the
same, but close enough that I think it's related.  In my case, certain
web pages are consistently unviewable - download progress stops part
of the way through these pages.  I spent some time trying to track
things down.  I've attached a couple tcpdump logs showing that certain
packets have consistently incorrect checksums (checksum is wrong on
all retransmits).  The tcpdump logs are from the firewall machine,
which does the pppoe.  torch.int-domain is the main machine, where the
web browser is running.  Behavior is the same if I run the web browser
on the firewall.  The problem started when I switched to a new
firewall machine, which runs RH 7.3.  The old firewall ran 7.0, so
both hardware and software changed, making it hard to narrow down the
problem.  I can investigate further if someone wants to point me in a
particular direction.

Logs generated using the following command:

# tcpdump -l -s0 -v -i eth0 host 162.129.49.63

Created attachment 64146 [details]
tcpdump log of web connection stall condition

The problem that boultonj reported may not be the same.  I saw a problem with
Mozilla on Linux where it *consistently* hung on certain web pages.  I solved
this problem by going into the Mozilla HTTP Networking preferences and turning
off HTTP 1.1, and unchecking 'Enable Pipelining' and 'Enable Persistent
Connections'.  I assumed that this was a Mozilla problem.

The problem I'm reporting in this bug report is not consistent.  The same page
may work fine, hang, or get a name resolution error each time it is reloaded. 
Since it's dependent on the kernel version, I'm pretty sure it's a Linux bug.

More info:  After checking the pppoe forum over at roaring penguin, I decided to
check my syslog when the problem occurred.  Lo and behold, I get the following
set of messages when the Mozilla connection hangs:


--- snip /var/log/messages ---
Jul  9 00:13:46 localhost pppoe[16910]: Inactivity timeout... something wicked
happened
Jul  9 00:13:46 localhost pppoe[16910]: Sent PADT
Jul  9 00:13:46 localhost pppd[16909]: Modem hangup
Jul  9 00:13:46 localhost pppd[16909]: Connection terminated.
Jul  9 00:13:46 localhost pppd[16909]: Connect time 8.4 minutes.
Jul  9 00:13:46 localhost pppd[16909]: Sent 654179 bytes, received 2061398 bytes.
Jul  9 00:13:46 localhost /etc/hotplug/net.agent: NET unregister event not supported
Jul  9 00:13:46 localhost pppd[16909]: Exit.
Jul  9 00:13:46 localhost adsl-connect: ADSL connection lost; attempting
re-connection.
Jul  9 00:13:51 localhost pppd[17029]: pppd 2.4.1 started by root, uid 0
Jul  9 00:13:51 localhost pppd[17029]: Using interface ppp0
Jul  9 00:13:51 localhost pppd[17029]: Connect: ppp0 <--> /dev/pts/2
Jul  9 00:13:51 localhost /etc/hotplug/net.agent: assuming ppp0 is already up
Jul  9 00:13:51 localhost pppoe[17030]: PPP session is 10512
Jul  9 00:13:52 localhost pppd[17029]: local  IP address 64.123.14.128
Jul  9 00:13:52 localhost pppd[17029]: remote IP address 64.123.15.254
Jul  9 00:13:52 localhost pppd[17029]: primary   DNS address 151.164.20.201
Jul  9 00:13:52 localhost pppd[17029]: secondary DNS address 151.164.11.201
Jul  9 00:14:31 localhost pppoe[17030]: Inactivity timeout... something wicked
happened
Jul  9 00:14:31 localhost pppoe[17030]: Sent PADT
Jul  9 00:14:31 localhost pppd[17029]: Modem hangup
Jul  9 00:14:31 localhost pppd[17029]: Connection terminated.
Jul  9 00:14:31 localhost pppd[17029]: Connect time 0.7 minutes.
Jul  9 00:14:31 localhost pppd[17029]: Sent 1474 bytes, received 14473 bytes.
Jul  9 00:14:31 localhost /etc/hotplug/net.agent: NET unregister event not supported
Jul  9 00:14:31 localhost pppd[17029]: Exit.
Jul  9 00:14:31 localhost adsl-connect: ADSL connection lost; attempting
re-connection.
Jul  9 00:14:36 localhost pppd[17107]: pppd 2.4.1 started by root, uid 0
Jul  9 00:14:36 localhost pppd[17107]: Using interface ppp0
Jul  9 00:14:36 localhost pppd[17107]: Connect: ppp0 <--> /dev/pts/2
Jul  9 00:14:37 localhost /etc/hotplug/net.agent: assuming ppp0 is already up
Jul  9 00:14:37 localhost pppoe[17108]: PPP session is 10523
Jul  9 00:14:38 localhost pppd[17107]: local  IP address 64.217.72.252
Jul  9 00:14:38 localhost pppd[17107]: remote IP address 64.217.73.254
Jul  9 00:14:38 localhost pppd[17107]: primary   DNS address 151.164.20.201
Jul  9 00:14:38 localhost pppd[17107]: secondary DNS address 151.164.11.201
--- end snip ---

The log posted by wpwood on 7-9 is definitely a different problem from
the one I'm having.  I don't believe that my problem is a Mozilla
issue, as suggested in the message from 7-8, although when I switch
firewall machines again (soon, I hope) I'm going to try changing the
Mozilla settings as suggested to see if it makes life tolerable.

Anyway, since it looks like my problem is different, I'll probably open
up a new bug for it.

More info -- I downloaded and built the kernel 2.4.19-rc1 code from kernel.org.
 After this, the problem went away.  I have no problem with running a custom
version of the kernel, but I assume that there are other PPPoX users out there
seeing the same problem.

So, with the 2.4.19-rc1 kernel, no more hangs, and no "ADSL connection lost"
errors in the syslog.

From the /var/log/messages snippet above I see that was the same problem I was 
experiencing. After some digging I found that the 
/etc/sysconfig/network-scripts/ifcfg-ppp0 contained PPPOE_TIMEOUT=20 and 
LCP_INTERVAL=20, now I've read somewhere that the timeout should be about 4 
times the LCP interval so I changed it to 80, the problem went away.

This seems to be fixed in Red Hat 8.0

Thanks for the bug report. However, Red Hat no longer maintains this version of
the product. Please upgrade to the latest version and open a new bug if the problem
persists.

The Fedora Legacy project (http://fedoralegacy.org/) maintains some older releases, 
and if you believe this bug is interesting to them, please report the problem in
the bug tracker at: http://bugzilla.fedora.us/