Bugzilla – Bug 137095
Cifs modules freezes the system
Last modified: 2007-02-01 13:14:47 UTC
I apologize for not having followed bugzilla but I was really busy at work. Here it is the original report: I use the cifs module (instead of smbfs) to access samba shares. If the connection is fast, over 6Mbyte/sec on a 100Mbit LAN, sometimes I get a total freeze. It happens mainly when connecting to a Debian Stable Pc, with others it usually get many stalls, but nothing else happens. When freezing, the machine stop answering ping too. The only solution is rebooting. Booting with nmi_watchdog=1 does not change the situation. How do I attach the hwinfo information?
Please execute `hwinfo > config.txt'. This file can be attched here with the attachment feature (see below, the direct link for this bug is https://bugzilla.novell.com/attachment.cgi?bugid=137095&action=enter). Please also attach 500 lines of your syslog in /var/log/messages with the same method. Thanks.
One note: Please always choose the correct component and version. openSUSE is the category with bugs within the Wiki at opensuse.org only. Changing component.
Created attachment 59926 [details] hwinfo information
Created attachment 59928 [details] Requested syslog These are not 500 lines but it covers the moment from mounting the share to the crash.
Created attachment 59929 [details] 500 lines of /var/log/messages These are my last 500 lines of /var/log/messages
Normally, if such a thing happens, the kernel oupses or at least outputs some useful information. Please check your syslog for such errors or reproduce the problem and look what it produces. If there is a hang without any logged information (rather unlikely) we will have little chance to fix this.
Sorry, but I have to close this bug until the required information can be provided. In that case, please reopen this report.
Note: this is my first bug-report/edit ever. Please have patiance with me if i dont do everything correct at once. I have had this problem to i think. The problem is not that my computer freezes, the problem is that the amount of log messages generated from the cifsd, user space program completely overwhelms my computer. However, killing cifsd will give me back complete control of my computer but leaves the mounts in a io-error type of state which seames unrecoverable. "uname -a" says: Linux mg 2.6.13-15.8-default #1 Tue Feb 7 11:07:24 UTC 2006 x86_64 x86_64 x86_64 GNU/Linux The log messages looks normal up to messages that looks like (this is the messages that floods the log and is basically all that i see.) ---------------------- Mar 15 09:54:48 mg kernel: CIFS VFS: No task to wake, unknown frame rcvd! Mar 15 09:54:48 mg kernel: Received Data is: : dump of 37 bytes of data at 0xffff81002ff0fe80 Mar 15 09:54:48 mg kernel: Mar 15 09:54:48 mg kernel: 3a000000 424d53ff 00000032 c0418000 . . . : ÿ S M B 2 . . . . . A À Mar 15 09:54:48 mg kernel: 00000000 00000000 00000000 43600001 . . . . . . . . . . . . . . ` C Mar 15 09:54:48 mg kernel: 06cd0064 0000020a d . Í . . ---------------------- The bug is generally trigged when using emacs/auctex with preview and source specials. The operations involved when saving-and-compiling touches about 10 files at once and this is when the bug shows up. Now, since i have a solution i thought i'd stop here. The sollution for me was to download the latest kernel and make a diff in the linux-xxx/fs/cifs directory. Since the functions called by file.c, connect.c and inode.c have changed somewhat since the kernel i use "2.6.13-15.8-default" i had to edit them slightly. 1) Change kzalloc( to kcalloc(1, in connect.c 2) Change filemap_write_and_wait to filemap_fdatawrite followed by filemap_fdatawait in inode.c and file.c This solves the problem i had. The operations that triggered the bug now seems to do something that generates a single log-message exactly the same as shown above and the recovers. However, the save-and-compile does not, when the bug is triggered, complete its operation so some of the index-files are left empty. This is however a minor problem. I guess what i'm asking is that you upgrade the cifs part of the suse kernel to the content of the latest kernel release so that i wont have to each time the kernel is updated.
Reassigning to the kernel maintainers.
Jeremy, can you help?
For CIFS problems, it's usually best to involve Steve French directly.
Whoops, mid-air collision with Olaf. Steve will need to see this reproduced on a 10.1 kernel.
So what you are telling me is that i should try and compile the 10.1 kernel on my 10.0 system to make sure the behaviour is unacceptable in 10.1 too?
STill in NEEDINFO for 7months. Markings as WONTFIX.