Skip to contentSkip to conversation

Problems with ipv4 part of Kernels post-2.6.16

DISCUSSION Rogério Brito 3 participants 5 messages Sep 3, 2006 – Sep 4, 2006

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi, John, David and others developers.

I have been trying to keep up with the kernel releases (not only the -rc ones, but also some -mm's) and I noticed something quite strange:
with some post 2.6.16 kernels (say, 2.6.17), I can't access (from where I am) the site www.everymac.com, while I can access other sites.

This has made me quite curious, because just booting back with a 2.6.16.x kernel, I could access it, which, of course, led me to think this was a problem with the networking part of the kernel.

Well, to cut a long story short, yesterday I decided to learn how to use git, grabbed Linus's tree and started a bisection session.

After 12 recompilations, I found the following patch being the first suspect (sorry for the line wrapping, but I copied this one from one console to another, since I don't know how to generate it again with git):

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

rbrito@dumont:/usr/local/media/progs/linux/kernel/linux-git$ git bisect
good
7b4f4b5ebceab67ce440a61081a69f0265e17c2a is first bad commit
commit 7b4f4b5ebceab67ce440a61081a69f0265e17c2a
Author: John Heffner <[email protected]>
Date:   Sat Mar 25 01:34:07 2006 -0800

[TCP]: Set default max buffers from memory pool size

This patch sets the maximum TCP buffer sizes (available to automatic buffer tuning, not to setsockopt) based on the TCP memory pool size. The maximum sndbuf and rcvbuf each will be up to 4 MB, but no more than 1/128 of the memory pressure threshold.

:040000 040000 514849b6a38c5fb671f3aeae1c0108a0e8e897dc
3b912fd10db444b22262f995fac99f2851363531 M net
rbrito@dumont:/usr/local/media/progs/linux/kernel/linux-git$
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

I'm currently connected to my provider via cable modem and I'm assigned an IP address via DHCP (actually, I'm using a D-Link DI-524 router in between, but I already tested *without* the router in the middle and the problem remains). The contents of my /etc/sysctl.conf file are attached.

My userspace is Debian's testing/etch, regularly upgraded every day. My system is a Duron 1.1GHz, with 512MB of RAM and a KT133 (VIA) chipset, with the network card being a Realtek RTL8139.

I would like to point out that this has prevented me from using/testing other newer kernels. :-(

If anything else is required, please, don't hesitate to ask. I will try my best to use any patches that may seem relevant, until we can point out what may be happening.

Thanks, Rogério Brito.

P.S.: Please, I'm not (currently) subscribed to any mailing list. I'd appreciate if the CCs weren't trimmed.

Signature · 3 lines
Rogério Brito : [email protected] : http://www.ime.usp.br/~rbrito
Homepage of the algorithms package : http://algorithms.berlios.de
Homepage on freshmeat:  http://freshmeat.net/projects/algorithms/

#

# /etc/sysctl.conf - Configuration file for setting system variables
# See sysctl.conf (5) for information.
#

# Uncomment the following to stop low-level messages on console #kernel.printk = 4 4 1 7

##############################################################3 # Functions previously found in netbase
#
#net.ipv4.conf.default.forwarding=1 #net.ipv6.conf.default.forwarding=1

dev.rtc.max-user-freq=1024
net.ipv4.conf.all.accept_redirects=0
net.ipv4.conf.all.accept_source_route=0
net.ipv4.conf.all.log_martians=1
net.ipv4.conf.all.rp_filter=1
net.ipv4.conf.default.accept_redirects=0
net.ipv4.conf.default.accept_source_route=0
net.ipv4.conf.default.log_martians=1
net.ipv4.conf.default.rp_filter=1
net.ipv4.icmp_echo_ignore_broadcasts=1
net.ipv4.tcp_syncookies=1

#1 lore raw

Hi Rogério,

Quoted 6 lines · Rogério Brito

Hi, John, David and others developers.

I have been trying to keep up with the kernel releases (not only the -rc ones, but also some -mm's) and I noticed something quite strange:
with some post 2.6.16 kernels (say, 2.6.17), I can't access (from where I am) the site www.everymac.com, while I can access other sites.

I believe I read on LKML last month that there's a problem on this site with window scaling. There's a patch floating around to allow per destination window clamping. I believe that you can workaround the problem by disabling TCP window scaling :

echo 0 >/proc/sys/net/ipv4/tcp_window_scaling

Hoping this helps,
Willy

Quoted 86 lines

This has made me quite curious, because just booting back with a 2.6.16.x kernel, I could access it, which, of course, led me to think this was a problem with the networking part of the kernel.

Well, to cut a long story short, yesterday I decided to learn how to use git, grabbed Linus's tree and started a bisection session.

After 12 recompilations, I found the following patch being the first suspect (sorry for the line wrapping, but I copied this one from one console to another, since I don't know how to generate it again with git):

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

rbrito@dumont:/usr/local/media/progs/linux/kernel/linux-git$ git bisect
good
7b4f4b5ebceab67ce440a61081a69f0265e17c2a is first bad commit
commit 7b4f4b5ebceab67ce440a61081a69f0265e17c2a
Author: John Heffner <[email protected]>
Date:   Sat Mar 25 01:34:07 2006 -0800

[TCP]: Set default max buffers from memory pool size

This patch sets the maximum TCP buffer sizes (available to automatic buffer tuning, not to setsockopt) based on the TCP memory pool size. The maximum sndbuf and rcvbuf each will be up to 4 MB, but no more than 1/128 of the memory pressure threshold.

:040000 040000 514849b6a38c5fb671f3aeae1c0108a0e8e897dc
3b912fd10db444b22262f995fac99f2851363531 M net
rbrito@dumont:/usr/local/media/progs/linux/kernel/linux-git$
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

I'm currently connected to my provider via cable modem and I'm assigned an IP address via DHCP (actually, I'm using a D-Link DI-524 router in between, but I already tested *without* the router in the middle and the problem remains). The contents of my /etc/sysctl.conf file are attached.

My userspace is Debian's testing/etch, regularly upgraded every day. My system is a Duron 1.1GHz, with 512MB of RAM and a KT133 (VIA) chipset, with the network card being a Realtek RTL8139.

I would like to point out that this has prevented me from using/testing other newer kernels. :-(

If anything else is required, please, don't hesitate to ask. I will try my best to use any patches that may seem relevant, until we can point out what may be happening.

Thanks, Rogério Brito.

P.S.: Please, I'm not (currently) subscribed to any mailing list. I'd appreciate if the CCs weren't trimmed.

Signature · 29 lines
Rogério Brito : [email protected] : http://www.ime.usp.br/~rbrito
Homepage of the algorithms package : http://algorithms.berlios.de
Homepage on freshmeat:  http://freshmeat.net/projects/algorithms/

#
# /etc/sysctl.conf - Configuration file for setting system variables
# See sysctl.conf (5) for information.
#

# Uncomment the following to stop low-level messages on console
#kernel.printk = 4 4 1 7

##############################################################3
# Functions previously found in netbase
#
#net.ipv4.conf.default.forwarding=1
#net.ipv6.conf.default.forwarding=1

dev.rtc.max-user-freq=1024
net.ipv4.conf.all.accept_redirects=0
net.ipv4.conf.all.accept_source_route=0
net.ipv4.conf.all.log_martians=1
net.ipv4.conf.all.rp_filter=1
net.ipv4.conf.default.accept_redirects=0
net.ipv4.conf.default.accept_source_route=0
net.ipv4.conf.default.log_martians=1
net.ipv4.conf.default.rp_filter=1
net.ipv4.icmp_echo_ignore_broadcasts=1
net.ipv4.tcp_syncookies=1
Signature · 1 line
VGER BF report: H 1.33782e-13
#2 lore raw
Quoted 15 lines · Willy Tarreau

Hi Rogério,

Quoted 6 lines · Rogério Brito

Hi, John, David and others developers.

I have been trying to keep up with the kernel releases (not only the -rc ones, but also some -mm's) and I noticed something quite strange:
with some post 2.6.16 kernels (say, 2.6.17), I can't access (from where I am) the site www.everymac.com, while I can access other sites.

I believe I read on LKML last month that there's a problem on this site with window scaling. There's a patch floating around to allow per destination window clamping. I believe that you can workaround the problem by disabling TCP window scaling :

Quoted 4 lines

echo 0 >/proc/sys/net/ipv4/tcp_window_scaling

Hoping this helps,
Willy

The above does help while hitting certain websites from behind my corporate proxy (while others are okay); same websites can be accessed without any issue from my home ISP connection.

I logged an internal ticket for this, will check whether there's been any update as of recently; both happens with recent 2.6.18-rc and with FC5-latest kernel.

Quoted 96 lines
Quoted 75 lines

This has made me quite curious, because just booting back with a 2.6.16.x kernel, I could access it, which, of course, led me to think this was a problem with the networking part of the kernel.

Well, to cut a long story short, yesterday I decided to learn how to use git, grabbed Linus's tree and started a bisection session.

After 12 recompilations, I found the following patch being the first suspect (sorry for the line wrapping, but I copied this one from one console to another, since I don't know how to generate it again with git):

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

rbrito@dumont:/usr/local/media/progs/linux/kernel/linux-git$ git bisect
good
7b4f4b5ebceab67ce440a61081a69f0265e17c2a is first bad commit
commit 7b4f4b5ebceab67ce440a61081a69f0265e17c2a
Author: John Heffner <[email protected]>
Date:   Sat Mar 25 01:34:07 2006 -0800

[TCP]: Set default max buffers from memory pool size

This patch sets the maximum TCP buffer sizes (available to automatic buffer tuning, not to setsockopt) based on the TCP memory pool size. The maximum sndbuf and rcvbuf each will be up to 4 MB, but no more than 1/128 of the memory pressure threshold.

:040000 040000 514849b6a38c5fb671f3aeae1c0108a0e8e897dc
3b912fd10db444b22262f995fac99f2851363531 M net
rbrito@dumont:/usr/local/media/progs/linux/kernel/linux-git$
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

I'm currently connected to my provider via cable modem and I'm assigned an IP address via DHCP (actually, I'm using a D-Link DI-524 router in between, but I already tested *without* the router in the middle and the problem remains). The contents of my /etc/sysctl.conf file are attached.

My userspace is Debian's testing/etch, regularly upgraded every day. My system is a Duron 1.1GHz, with 512MB of RAM and a KT133 (VIA) chipset, with the network card being a Realtek RTL8139.

I would like to point out that this has prevented me from using/testing other newer kernels. :-(

If anything else is required, please, don't hesitate to ask. I will try my best to use any patches that may seem relevant, until we can point out what may be happening.

Thanks, Rogério Brito.

P.S.: Please, I'm not (currently) subscribed to any mailing list. I'd appreciate if the CCs weren't trimmed.
--
Rogério Brito : [email protected] : http://www.ime.usp.br/~rbrito
Homepage of the algorithms package : http://algorithms.berlios.de
Homepage on freshmeat: http://freshmeat.net/projects/algorithms/

#

# /etc/sysctl.conf - Configuration file for setting system variables
# See sysctl.conf (5) for information.
#

# Uncomment the following to stop low-level messages on console #kernel.printk = 4 4 1 7

##############################################################3 # Functions previously found in netbase
#
#net.ipv4.conf.default.forwarding=1 #net.ipv6.conf.default.forwarding=1

Quoted 11 lines

dev.rtc.max-user-freq=1024
net.ipv4.conf.all.accept_redirects=0
net.ipv4.conf.all.accept_source_route=0
net.ipv4.conf.all.log_martians=1
net.ipv4.conf.all.rp_filter=1
net.ipv4.conf.default.accept_redirects=0
net.ipv4.conf.default.accept_source_route=0
net.ipv4.conf.default.log_martians=1
net.ipv4.conf.default.rp_filter=1
net.ipv4.icmp_echo_ignore_broadcasts=1
net.ipv4.tcp_syncookies=1

--
VGER BF report: H 1.33782e-13
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Signature · 11 lines
--alessandro

"What's the name of the word for the precise moment when you
realize that you've actually forgotten how it felt to make love
to somebody you really liked a long time ago?"
"There isn't one."
"Oh. I thought maybe there was."
    (The Sandman, dialogue between Delirium and Dream)

-- 
VGER BF report: H 0.0642894
#3 lore raw

Sorry - reply-all got a bounce, didn't check the bad netdev address...

---------- Forwarded message ----------
From: Alessandro Suardi <[email protected]>
Date: Sep 4, 2006 12:07 AM
Subject: Re: Problems with ipv4 part of Kernels post-2.6.16
To: Willy Tarreau <[email protected]>
Cc: Rogério Brito <[email protected]>, [email protected], [email protected], John Heffner <[email protected]>, "David S. Miller" <[email protected]>

Quoted 15 lines · Willy Tarreau

Hi Rogério,

Quoted 6 lines · Rogério Brito

Hi, John, David and others developers.

I have been trying to keep up with the kernel releases (not only the -rc ones, but also some -mm's) and I noticed something quite strange:
with some post 2.6.16 kernels (say, 2.6.17), I can't access (from where I am) the site www.everymac.com, while I can access other sites.

I believe I read on LKML last month that there's a problem on this site with window scaling. There's a patch floating around to allow per destination window clamping. I believe that you can workaround the problem by disabling TCP window scaling :

Quoted 4 lines

echo 0 >/proc/sys/net/ipv4/tcp_window_scaling

Hoping this helps,
Willy

The above does help while hitting certain websites from behind my corporate proxy (while others are okay); same websites can be accessed without any issue from my home ISP connection.

I logged an internal ticket for this, will check whether there's been any update as of recently; both happens with recent 2.6.18-rc and with FC5-latest kernel.

Quoted 75 lines
Quoted 75 lines

This has made me quite curious, because just booting back with a 2.6.16.x kernel, I could access it, which, of course, led me to think this was a problem with the networking part of the kernel.

Well, to cut a long story short, yesterday I decided to learn how to use git, grabbed Linus's tree and started a bisection session.

After 12 recompilations, I found the following patch being the first suspect (sorry for the line wrapping, but I copied this one from one console to another, since I don't know how to generate it again with git):

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

rbrito@dumont:/usr/local/media/progs/linux/kernel/linux-git$ git bisect
good
7b4f4b5ebceab67ce440a61081a69f0265e17c2a is first bad commit
commit 7b4f4b5ebceab67ce440a61081a69f0265e17c2a
Author: John Heffner <[email protected]>
Date:   Sat Mar 25 01:34:07 2006 -0800

[TCP]: Set default max buffers from memory pool size

This patch sets the maximum TCP buffer sizes (available to automatic buffer tuning, not to setsockopt) based on the TCP memory pool size. The maximum sndbuf and rcvbuf each will be up to 4 MB, but no more than 1/128 of the memory pressure threshold.

:040000 040000 514849b6a38c5fb671f3aeae1c0108a0e8e897dc
3b912fd10db444b22262f995fac99f2851363531 M net
rbrito@dumont:/usr/local/media/progs/linux/kernel/linux-git$
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

I'm currently connected to my provider via cable modem and I'm assigned an IP address via DHCP (actually, I'm using a D-Link DI-524 router in between, but I already tested *without* the router in the middle and the problem remains). The contents of my /etc/sysctl.conf file are attached.

My userspace is Debian's testing/etch, regularly upgraded every day. My system is a Duron 1.1GHz, with 512MB of RAM and a KT133 (VIA) chipset, with the network card being a Realtek RTL8139.

I would like to point out that this has prevented me from using/testing other newer kernels. :-(

If anything else is required, please, don't hesitate to ask. I will try my best to use any patches that may seem relevant, until we can point out what may be happening.

Thanks, Rogério Brito.

P.S.: Please, I'm not (currently) subscribed to any mailing list. I'd appreciate if the CCs weren't trimmed.
--
Rogério Brito : [email protected] : http://www.ime.usp.br/~rbrito
Homepage of the algorithms package : http://algorithms.berlios.de
Homepage on freshmeat: http://freshmeat.net/projects/algorithms/

#

# /etc/sysctl.conf - Configuration file for setting system variables
# See sysctl.conf (5) for information.
#

# Uncomment the following to stop low-level messages on console #kernel.printk = 4 4 1 7

##############################################################3 # Functions previously found in netbase
#
#net.ipv4.conf.default.forwarding=1 #net.ipv6.conf.default.forwarding=1

Quoted 11 lines
Quoted 11 lines

dev.rtc.max-user-freq=1024
net.ipv4.conf.all.accept_redirects=0
net.ipv4.conf.all.accept_source_route=0
net.ipv4.conf.all.log_martians=1
net.ipv4.conf.all.rp_filter=1
net.ipv4.conf.default.accept_redirects=0
net.ipv4.conf.default.accept_source_route=0
net.ipv4.conf.default.log_martians=1
net.ipv4.conf.default.rp_filter=1
net.ipv4.icmp_echo_ignore_broadcasts=1
net.ipv4.tcp_syncookies=1

--alessandro

"What's the name of the word for the precise moment when you realize that you've actually forgotten how it felt to make love to somebody you really liked a long time ago?"
"There isn't one."
"Oh. I thought maybe there was."
(The Sandman, dialogue between Delirium and Dream)

Signature · 11 lines
--alessandro

"What's the name of the word for the precise moment when you
realize that you've actually forgotten how it felt to make love
to somebody you really liked a long time ago?"
"There isn't one."
"Oh. I thought maybe there was."
    (The Sandman, dialogue between Delirium and Dream)

-- 
VGER BF report: H 3.88436e-05
#4 lore raw
In reply to Alessandro Suardi

Hi, Willy, Alessandro and others,

Quoted 7 lines · Alessandro Suardi
Quoted 6 lines · Willy Tarreau

Hi Rogério,

Quoted 2 lines · Rogério Brito

I can't access (from where I am) the site www.everymac.com, while I can access other sites.

Quoted 6 lines
Quoted 6 lines

I believe I read on LKML last month that there's a problem on this site with window scaling. There's a patch floating around to allow per destination window clamping. I believe that you can workaround the problem by disabling TCP window scaling :

echo 0 >/proc/sys/net/ipv4/tcp_window_scaling

I still have not had the opportunity of disabling TCP window scaling, since I'm running a quite intensive process right now under a kernel that doesn't exhibit the problem.

Quoted 3 lines

The above does help while hitting certain websites from behind my corporate proxy (while others are okay); same websites can be accessed without any issue from my home ISP connection.

Well, it's "nice" to see that I am not the only one hitting this problem. :-(

Quoted 3 lines

I logged an internal ticket for this, will check whether there's been any update as of recently; both happens with recent 2.6.18-rc and with FC5-latest kernel.

One thing that I might test (when I have some time) is to just reverse the patch and see if the problem persists, but, of course, the net developers would know much better what should be done.

Thanks for the answers, Rogério Brito.

Signature · 3 lines
Rogério Brito : [email protected] : http://www.ime.usp.br/~rbrito
Homepage of the algorithms package : http://algorithms.berlios.de
Homepage on freshmeat:  http://freshmeat.net/projects/algorithms/