scp transfer causes system to freeze

[Charles Majola]

I have just installed Dapper Beta on a HP 6120 with PV HT 3.0 GHz. When I try to copy a file from our server the entire system freezes and nothing responds, the only way to use it again is to press the reset button. This happens on both -386 and -686 kernels and this is a fresh install. I tried debugging methods, there doesn't seem to be any kernel oops or anything.

The command used :
#> sudo scp charles@10.0.0.2:/home/backup/data-files.tar.gz .

as soon as this runs, the system freezes.

I'll be happy to help debug this

[Charles Majola (chmj)]

correction: PV should be P4 with HyperThreading

[Charles Majola (chmj)]

dmesg output

[Charles Majola (chmj)]

lspci -vv output

[Soren Hansen (soren)]

Is it only scp that does this? And does it only happen when connection to that server?

Try doing it in a console and when it crashes, do an Alt-SysRq-p. That should give you a backtrace. Try copying that backtrace here.

[Carthik Sharma (carthik)]

Changing to "Needs info" since some information has been requested from the reporter.

[Ahmed Kamal (kim0)]

Happens to me too, except that my system doesnt freeze, but raher the networking subsystem becomes totally broken. Here is what happens:

1- I try to scp some tgz file to a server
2- It copies around 2% of the 250MB file (Zimbra collaboration suite)
3- It stops copying
4- I kill scp
5- Now whatever I do (ifconfig, dhclient, knetworkmanager ...) I cannot get back on the network. I get IPs, but I can't reach any other machine!! It was too weird.
6- Finally networking worked, only when I rebooted
7- BTW, I repeated the same situation again before I started doubting dapper and made the scp from Suse10

This seems really critical bug
One last thing my driver is sky2, on a toshiba A105-S361 laptop

[brentdurksen (brentdurksen)]

I can confirm this bug (or similar) on Toshiba M50-MX5 laptop running Dapper with latest updates installed. For me, my networking fails and my computer eventually hangs when I try to copy a large file to a remote computer using scp, or the sshfs FUSE filesystem, or Samba (to an XP box). The only keystroke that gets any response is Alt+SysRq+B.

[Matt Zimmerman (mdz)]

Please get a trace from the crash by following https://wiki.ubuntu.com/DebuggingSystemCrash

brentdurksen, what kind of network device do you have? Charles seems to have:

0000:40:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM5751 Gigabit Ethernet PCI Express (rev 01)

[brentdurksen (brentdurksen)]

I can no longer reproduce this bug. I don't know if it was a recent update that fixed the problem, or something more random, but this issue is no longer affecting my machine. For the record, my ethernet controller is a Marvell Technology Group Ltd. 88E8036 Fast Ethernet Controller (rev 10)

[Ben Collins (ben-collins)]

More than likely, this was a sky2 issue, and if it isn't fixed yet, it will be in the next upload.

[Charles Majola (chmj)]

I still have this bug, however, instead of freezing the whole system now, it just shuts down my networking, and I have to reboot to get it running again. I'm using 686 2.6.15-22.34 . I don't know if this will help but the last entry on dmesg before the crash is "ibm_acpi: ec object not found". I'm gonna turn acpi off on next reboot and retest.

[Charles Majola (chmj)]

reopening

[Charles Majola (chmj)]

dmesg output with "noacpi"

[Charles Majola (chmj)]

Screenshot for the kernel oops, this showed one when I tried to transfer file from console, only happened ONCE, otherwise the system becomes totaly unresponsive.

[Ahmed Kamal (kim0)]

Hi,
    I just faced this bug again, but not with scp, but rather with smb:/ copying from konqueror. Again, the network card became totally dead, I could not do any network related activity except after rebooting.

Note: Just noticed 2.6.15-23 is being installed on my system now. Not sure if it has this fixed or not.

[pogets (pogets)]

dmesg output after eth0 stops working during cp from NFS-mounted to local directory - system trace obtained with Alt+SysRq+1, Alt+SysRq+t

[pogets (pogets)]

In addition to above, my computer is also a toshiba laptop with Marvell 88E8036 Fast Ethernet Controller. eth0 stops working in other situations, too, but file transfer from nfs-mounted to local directory is very reproducible.

[pogets (pogets)]

I just wanted to mention that the network freeze upon nsf-transfer as described above still occurs with the latest kernel of today (2.6.15-23.39). In the pre-release days of Ubuntu 5.10, my LAN card was not recognised at all, and I had to build a custom kernel with the sk98lin driver, but this got fixed for the final release. Is anybody working on this bug in Dapper right now? If so, let me know if I can be of any help - in spite of having no programing and only limited linux experience.

[Rodrigo Novo (rodarvus)]

Searching for related messages on linux-kernel@, I found the following message (part of a larger thread):

http://marc.theaimsgroup.com/?l=linux-kernel&m=114849250712330&w=2

which leads to the following patch, applied to 2.6.17-rc5:

commit a1433ac4ab46fb23ae77804c207a1f710a7b12f1
Author: Stephen Hemminger <email address hidden>
Date: Mon May 22 12:03:42 2006 -0700

    [PATCH] sky2: fix jumbo packet support

    The truncate threshold calculation to prevent receiver from getting stuck
    was incorrect, and it didn't take into account the upper limit on bits
    in the register so the jumbo packet support was broken.

    Signed-off-by: Stephen Hemminger <email address hidden>
    Signed-off-by: Jeff Garzik <email address hidden>

Could this be related to the problem some people are having with the sky2 driver here?

(hope I'm not steping into someone elses toes here :) )