oplock_break: client failure in break ? / files locked forever(?)

Syzop syz at dds.nl
Wed Jun 20 20:41:37 GMT 2001


Martin Bahlinger wrote:

> Exactly, Netscape also. And sometimes a simple mouseclick in
> Windows-Explorer let it freeze.

[long part]
Some additional info (maybe not very helpfull, but just to let you
all know that I'm trying to trace the problem and exclude some
possibilities)...

The following applications were affected:
- some uninstall app
- some geography app
- netscape
- excel
- another german uninstall app

Also not very frequently..
10:54, 10:55, 14:17, (next day), 10:46, 09:43, 10:53, (next day), 08:07, 08:34.
Blablabla :).

At the dataserver:
- some .xls files
- netscape cache files and other netscape stuff.

Some other additional info:
- It's a self compiled version of samba 2.2.x with vfs, utmp, and some other stuff enabled.
- The dataserver uses a VFS module for logging, but the appserver does not, so this could not be the problem
- I guess the kernel version is also not the problem, because kernel oplocks = no.
- Clients: there are ~20 pcs with NT4 sp4, ~25 pcs with NT4 sp6, ~4 pcs with w2k, no w9x.
- The problem was reported both with w2k and NT
- Even if one user uses a file, that file can be 'locked forever' too.
- IIRC the 'smbstatus' did not gave me any additional info (the output of locked files
  was the same before the problem as after).
- before every oplock_break error there's a 'receive_smb timed out after 30 seconds.'
- at the appserver the clients don't have write permission, but also here the files get 'locked'
- it looks to me like not only files which are accessed many times are affected, but also
  files who are accessed only once (or maybe twice).

I've seen some stuff about 'oplock break wait time', do I need that with NT 4 sp4?

Also, something else... this whole thing should never happen...
I mean... a client who has only _read_ permissions should never be able to 'lock' a file like this, right?
[/long part]

Today (Wed Jun 20 22:23) I have enabled -d 10 logging, and I'll hopefully see some oplock_break
errors tomorrow. After the first few errors I really have to restart samba because otherwise the
users are getting crazy, but I think one or two cases of a such 'locked file' are enough.
Do I need to do something else except debug level 10 logging?

Cya,

    Syzop.








More information about the samba-technical mailing list