Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Mon, Dec 15, 2008 at 03:35:00PM +0100, Lars Ellenberg wrote: > On Mon, Dec 15, 2008 at 03:14:28PM +0100, Thomas Reinhold wrote: > > Thanks for your quick answer. > > > > This is the complete error message (I shoud have posted that in the first > > place, sorry): > > > > > > Dec 11 21:47:56 bftest-54 kernel: FAILED > > Dec 11 21:47:56 bftest-54 kernel: status = 1, message = 00, host = 0, driver = 08 > > Dec 11 21:47:56 bftest-54 kernel: <6>sd: Current: sense key: Illegal Request > > Dec 11 21:47:56 bftest-54 kernel: Additional sense: Invalid command operation code > > well, that is from the scsi layer. > and yes, that is exactly the decoded scsi_execute_req return value > 0x8000002 ;) > > > Dec 11 21:47:56 bftest-54 kernel: drbd0: local disk flush failed with > > status 134217730 > > > > What I forgot to mention is that DRBD runs on top of a dm device (dm-crypt), > > which in turn runs on top of the raid array. Could that cause the problem? > > I was not aware that redhat has patched its 2.6.18 dm-crypt to > (pretend to?) support flushed/barriers. correction. apparently device mapper in 2.6.9 up to and including 2.6.22 (kernel.org; cannot speak for vendor kernels) do always just pass on flushes and barriers, and let the lower level targets/drivers handle it. or not. 2.6.23 and later do reject barriers with EOPNOTSUPP right in dm_req, and since flushes became empty barriers in 2.6.24, flushes are now no longer supported on devicemapper, either. strange move. > but that should be irrelevant in this context. > > apparendly your lower level driver (what driver module is in use?) does > not support flush requests, or something in the device mapper stack > converts the flush request into something that the driver does not > understand. > > we may consider to handle any return value != 0 in the same way in drbd. > meanwhile, the correct fix is: > > > I have disabled disk flushing for now, and the error obiously has gone. -- : Lars Ellenberg : LINBIT | Your Way to High Availability : DRBD/HA support and consulting http://www.linbit.com DRBD® and LINBIT® are registered trademarks of LINBIT, Austria. __ please don't Cc me, but send to list -- I'm subscribed