CIFS regression in 2.6.27.8
Subsystems (2)
I just mounted a network drive (Windows Server) via CIFS under 2.6.27.8, and the sizes reported by "df" are completely incorrect:
root@exodus:~# df -h /mnt/fsv1 Filesystem Size Used Avail Use% Mounted on //fsv1/Users 446G -222G 667G - /mnt/fsv1 root@exodus:~# df /mnt/fsv1 Filesystem 1K-blocks Used Available Use% Mounted on //fsv1/Users 467403140 -231816588 699219728 - /mnt/fsv1 root@exodus:~# mount | grep fsv1 //fsv1/Users on /mnt/fsv1 type cifs (rw,mand) root@exodus:~#
Does anyone else see this sort of behavior with 2.6.27.8? This worked fine in 2.6.27.6 (we skipped 2.6.27.7)...
Ideas?
Signature · 2 lines
View this message in context: http://www.nabble.com/CIFS-regression-in-2.6.27.8-tp20900699p20900699.html Sent from the linux-kernel mailing list archive at Nabble.com.
Hi Dennis,
Quoted 13 lines · Dennis Hardy
I just mounted a network drive (Windows Server) via CIFS under 2.6.27.8, and the sizes reported by "df" are completely incorrect:
root@exodus:~# df -h /mnt/fsv1 Filesystem Size Used Avail Use% Mounted on //fsv1/Users 446G -222G 667G - /mnt/fsv1 root@exodus:~# df /mnt/fsv1 Filesystem 1K-blocks Used Available Use% Mounted on //fsv1/Users 467403140 -231816588 699219728 - /mnt/fsv1 root@exodus:~# mount | grep fsv1 //fsv1/Users on /mnt/fsv1 type cifs (rw,mand) root@exodus:~#
Quoted 4 lines
Does anyone else see this sort of behavior with 2.6.27.8? This worked fine in 2.6.27.6 (we skipped 2.6.27.7)...
Ideas?
Well, lets cc the relevant people so your report isn't lost in the noise.
Quoted 16 lines · Pekka Enberg
Hi Dennis,
Quoted 13 lines · Dennis Hardy
I just mounted a network drive (Windows Server) via CIFS under 2.6.27.8, and the sizes reported by "df" are completely incorrect:
root@exodus:~# df -h /mnt/fsv1 Filesystem Size Used Avail Use% Mounted on //fsv1/Users 446G -222G 667G - /mnt/fsv1 root@exodus:~# df /mnt/fsv1 Filesystem 1K-blocks Used Available Use% Mounted on //fsv1/Users 467403140 -231816588 699219728 - /mnt/fsv1 root@exodus:~# mount | grep fsv1 //fsv1/Users on /mnt/fsv1 type cifs (rw,mand) root@exodus:~#
Quoted 6 lines
Quoted 4 lines
Does anyone else see this sort of behavior with 2.6.27.8? This worked fine in 2.6.27.6 (we skipped 2.6.27.7)...
Ideas?
Well, lets cc the relevant people so your report isn't lost in the noise.
FWIW, it seems to be related to this one: http://lkml.org/lkml/2008/12/8/7
Thanks,
Rafael
Quoted 5 lines
FWIW, it seems to be related to this one: http://lkml.org/lkml/2008/12/8/7
Thanks,
Rafael
Please note that in my case, I'm not seeing any samba log errors or any problems with CPU utilization... File read access via the CIFS mount seems to work fine, it is just that "df" is reporting completely incorrect information for the CIFS mount.
Signature · 2 lines
View this message in context: http://www.nabble.com/CIFS-regression-in-2.6.27.8-tp20900699p20903107.html Sent from the linux-kernel mailing list archive at Nabble.com.
Quoted 3 lines · Rafael J. Wysocki
Quoted 3 lines · Dennis Hardy
Quoted 2 lines
I just mounted a network drive (Windows Server) via CIFS under 2.6.27.8, and the sizes reported by "df" are completely incorrect:
...
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Does anyone else see this sort of behavior with 2.6.27.8? This worked fine in 2.6.27.6 (we skipped 2.6.27.7)...
There were several cifs updates in .8 but AFAIU none in .7.
Quoted 1 line
FWIW, it seems to be related to this one: http://lkml.org/lkml/2008/12/8/7
Holger's report is about the userland SMB/CIFS server, while Denis's is about kernelspace client code. They look like separate issues to me.
Signature · 3 lines
Stefan Richter -=====-==--- ==-- -=--- http://arcgraph.de/sr/
Yes - they look distinct
Steve French
Senior Software Engineer
Linux Technology Center - IBM Austin
phone: 512-838-2294
email: sfrench at-sign us dot ibm dot com
Stefan Richter <[email protected]>
12/08/2008 02:58 PM
To
"Rafael J. Wysocki" <[email protected]>
cc
Pekka Enberg <[email protected]>, Dennis Hardy <[email protected]>, [email protected], [email protected], Steven French/Austin/IBM@IBMUS, Greg KH <[email protected]>, [email protected], Holger Hoffstaette <[email protected]> Subject
Re: CIFS regression in 2.6.27.8
Quoted 1 line · Rafael J. Wysocki
Quoted 1 line
On Mon, Dec 8, 2008 at 8:18 PM, Dennis Hardy <[email protected]>
Quoted 1 line · wrote:
Quoted 1 line
Quoted 1 line
I just mounted a network drive (Windows Server) via CIFS under
2.6.27.8, and
Quoted 1 line
Quoted 1 line
Quoted 1 line
the sizes reported by "df" are completely incorrect:
...
Quoted 1 line
Quoted 1 line
Quoted 1 line
Does anyone else see this sort of behavior with 2.6.27.8? This worked
fine
Quoted 1 line
Quoted 1 line
Quoted 1 line
in 2.6.27.6 (we skipped 2.6.27.7)...
There were several cifs updates in .8 but AFAIU none in .7.
Quoted 1 line
FWIW, it seems to be related to this one:
http://lkml.org/lkml/2008/12/8/7
Holger's report is about the userland SMB/CIFS server, while Denis's is about kernelspace client code. They look like separate issues to me.
Signature · 3 lines
Stefan Richter -=====-==--- ==-- -=--- http://arcgraph.de/sr/
Quoted 17 lines · On Mon, 8 Dec 2008 21:12:48 +0100 "Rafael J. Wysocki" <[email protected]> wrote:
Quoted 16 lines · Pekka Enberg
Hi Dennis,
Quoted 13 lines · Dennis Hardy
I just mounted a network drive (Windows Server) via CIFS under 2.6.27.8, and the sizes reported by "df" are completely incorrect:
root@exodus:~# df -h /mnt/fsv1 Filesystem Size Used Avail Use% Mounted on //fsv1/Users 446G -222G 667G - /mnt/fsv1 root@exodus:~# df /mnt/fsv1 Filesystem 1K-blocks Used Available Use% Mounted on //fsv1/Users 467403140 -231816588 699219728 - /mnt/fsv1 root@exodus:~# mount | grep fsv1 //fsv1/Users on /mnt/fsv1 type cifs (rw,mand) root@exodus:~#
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Does anyone else see this sort of behavior with 2.6.27.8? This worked fine in 2.6.27.6 (we skipped 2.6.27.7)...
I tried it on both i386 and x86_64 and had no problems. Is this client non-x86 by any chance?
Quoted 3 lines
I tried it on both i386 and
x86_64and had no problems. Is this client non-x86 by any chance?
Curious...my kernel is CONFIG_X86_32=y and CONFIG_X86=y, CONFIG_MK6=y.
Perhaps this could be an issue with the Windows Server side (Windows Home Server, latest Windows SPs, etc.) Maybe MSFT introduced some "improvement" in a recent update that caused an incompatibility of some sort. I can try mounting some other Windows machines via CIFS from the 2.6.27.8 end and see what happens, plus boot back to 2.6.27.6 and compare.
Signature · 2 lines
View this message in context: http://www.nabble.com/CIFS-regression-in-2.6.27.8-tp20900699p20924383.html Sent from the linux-kernel mailing list archive at Nabble.com.
Quoted 4 lines · On Tue, 9 Dec 2008 13:47:56 -0800 (PST) Dennis Hardy <[email protected]>…
Quoted 3 lines
I tried it on both i386 and
x86_64and had no problems. Is this client non-x86 by any chance?
Quoted 8 lines
Curious...my kernel is
CONFIG_X86_32=yandCONFIG_X86=y,CONFIG_MK6=y.Perhaps this could be an issue with the Windows Server side (Windows Home Server, latest Windows SPs, etc.) Maybe MSFT introduced some "improvement" in a recent update that caused an incompatibility of some sort. I can try mounting some other Windows machines via CIFS from the 2.6.27.8 end and see what happens, plus boot back to 2.6.27.6 and compare.
There is a problem with CIFS in 2.6.27.8, but it mainly manifests itself as a kthread that doesn't shut down when it should (regression due to a missing couple of prerequisite patches). I've got a patch for 2.6.27.9 that should fix that problem (assuming there is a .9)
The fixes that went into 2.6.27.8 were mostly patches to fix races on mount and unmount and some fixes for socket handling. The problem you've described seems very consistent and localized to an area that wasn't patched in .8.
I could easily be wrong, but the problem you've described seems more likely to be a server issue. If you can get binary network captures between the client and server like Steve F. requested, then we should be able to confirm it either way...
Signature · 1 line
Jeff Layton <[email protected]>
Quoted 3 lines
I could easily be wrong, but the problem you've described seems more likely to be a server issue.
Gentlemen, I am heading out to get an "L" (for loser) tattooed to my forehead. MSFT pushed some updates for the Windows Server previously, and when we switched to 2.6.27.8 we rebooted everything, including the Windows Server, which activated the updates. MSFT broke volume size reporting -- even from other Windows machines! We should have checked this before posting.
Please accept my deepest apologies for this -- Linux is, as always, working great.
Signature · 2 lines
View this message in context: http://www.nabble.com/CIFS-regression-in-2.6.27.8-tp20900699p20942054.html Sent from the linux-kernel mailing list archive at Nabble.com.