Bug 7435

Summary: tg3 module malfunction
Product: Networking Reporter: leonardo LeOS buffa (leos)
Component: IPV4Assignee: Stephen Hemminger (stephen)
Status: REJECTED WILL_NOT_FIX    
Severity: normal    
Priority: P2    
Hardware: i386   
OS: Linux   
Kernel Version: 2.6.18.1 Subsystem:
Regression: --- Bisected commit-id:

Description leonardo LeOS buffa 2006-10-30 08:08:22 UTC 
Most recent kernel where this bug did not occur: 2.6.15.7

Distribution: slackware 11

Hardware Environment: Ethernet controller: Broadcom Corporation NetXtreme
BCM5751F Fast Ethernet PCI Express (rev 21)

Software Environment: firefox 1.5.0.7 - lynx 2.8.5rel.5 - squid 2.5.STABLE12p1
on a OpenBSD 3.9 box (the client doesn't matter)

Problem Description: strange malfuncition trying to connect to a http proxy
(squid) for normal navigation, problems browsing local http server
the client hangs and doesn't get anything, here a couple of example:
---
client:
leos@NetWalker:~$ telnet 10.0.10.1 80
Trying 10.0.10.1...
Connected to 10.0.10.1.
Escape character is '^]'.
GET

proxy server:
# tail -f /var/www/logs/access.log
192.168.1.2 - - [26/Oct/2006:15:40:41 +0200] "GET" 200 1320


client:
leos@NetWalker:~$ lynx www.pippo.com 
HTTP request sent; waiting for response.

proxy server:
# tail -f /var/squid/logs/access.log|grep 192.168.1.2
1161870222.546      5 192.168.1.2 TCP_MEM_HIT/200 3314 GET
http://www.pippo.com/ - NONE/- text/html
---

Steps to reproduce: kernel >= 2.6.16 - try to navigate using squid proxy
Comment 1 Stephen Hemminger 2006-10-30 08:40:40 UTC 
This is almost certainly a problem on the other site due to TCP window scaling.
If there is a non-conformant firewall or gateway in the path, the TCP window
scaling in later kernels will cause a hung connection.

This is not a linux problem.  See:
http://lwn.net/Articles/92727/

You can workaround the problem by either turning window scaling off globally
or by putting a reduced congestion window limit on for the particular route.
Comment 2 Paolo Ciarrocchi 2006-10-30 12:18:16 UTC 
Bug 7435 should be marked as duplicated of 7435, since the submitter posted the
same report two times by mistake (confirmed by the submitter)
Comment 3 Paolo Ciarrocchi 2006-10-30 12:19:17 UTC 
Sorry...

Bug 7434 should be marked as duplicated of 7435, since the submitter posted the
same report two times by mistake (confirmed by the submitter)
Comment 4 Stephen Hemminger 2006-10-30 12:48:35 UTC 
*** Bug 7434 has been marked as a duplicate of this bug. ***