Hi all, Using Samba 2.2.3a on IRIX 6.5.5f we've hit an awkward problem. Kernel oplocks are supported (yay!), but, it turns out, not on IRIX efs filesystems, only XFS. We've had to discover this the hard way; within minutes of upgrading to 2.2.3a we had users coming to us reporting inability to open certain (Lotus Approach) database files. The error message:- [2002/04/09 13:11:33, 0] smbd/oplock_irix.c:(170) set_file_oplock: Unable to get kernel oplock on file LOGIN.APR, dev = cc, inode = 48354, file_id = 4l. Error was Invalid argument lead us to the 'fcntl' call on line 168 of smbd/oplock_irix.c and a "quick" trip through fcntl(2) reveals that EINVAL is returned if the oplock is being requested on a file that's on a non-XFS filesystem. Anyone else seen this? Is there any chance of 'kernel oplocks' being a share level (as opposed to global) paramter? Or could the oplock_irix code handle EINVAL as a special case? Thanks, Mac Assistant Systems Adminstrator @nibsc.ac.uk dmccann@nibsc.ac.uk Work: +44 1707 654753 x285 Everything else: +44 7956 237670 (anytime)
Bob Crandell
2002-Apr-09 08:30 UTC
[Samba] kernel oplocks fail (unprettily) on IRIX (with efs)
I've seen this on some reiserfs systems. Are kernel oplocks *only* supported with XFS? Mac (dmccann@nibsc.ac.uk) wrote*:> >Hi all, > > Using Samba 2.2.3a on IRIX 6.5.5f we've hit an awkward problem. > >Kernel oplocks are supported (yay!), but, it turns out, not on IRIX efs >filesystems, only XFS. > >We've had to discover this the hard way; within minutes of upgrading to >2.2.3a we had users coming to us reporting inability to open certain >(Lotus Approach) database files. > >The error message:- > >[2002/04/09 13:11:33, 0] smbd/oplock_irix.c:(170) > set_file_oplock: Unable to get kernel oplock on file LOGIN.APR, dev = cc, inode= 48354, file_id = 4l. Error was Invalid argument> >lead us to the 'fcntl' call on line 168 of smbd/oplock_irix.c and a >"quick" trip through fcntl(2) reveals that EINVAL is returned if the >oplock is being requested on a file that's on a non-XFS filesystem. > > >Anyone else seen this? Is there any chance of 'kernel oplocks' being a >share level (as opposed to global) paramter? Or could the oplock_irix >code handle EINVAL as a special case? > > Thanks, > > Mac > Assistant Systems Adminstrator @nibsc.ac.uk > dmccann@nibsc.ac.uk > Work: +44 1707 654753 x285 Everything else: +44 7956 237670 (anytime) > >