Messages in this thread | | | Date | Wed, 3 Mar 2010 10:30:15 -0500 | From | Ed Cashin <> | Subject | Re: aoeblk_make_request() chokes on zero-length barriers |
| |
On Wed, Mar 03, 2010 at 01:11:12PM +0900, Tejun Heo wrote: > Hello, Ed L. Cashin. > > This is reported on openSUSE 11.2 by Lars Tauber on bnc581271[1] but > upstream code seems to have the same problem. If a bio doesn't have > bi_io_vec, aoeblk_make_request() triggers BUG() but in recent kernels > the block layer will happily send down zero-length barrier bios w/o > bi_io_vec set.
It looks like the opensuse 11.x kernel doesn't have the change that made aoe return -EOPNOTSUPP for barriers.
It's commit 18d8217bc441630c3c5ec7416c5a65c69e8a0979, appearing in v2.6.32-rc1.
I have most of a barrier implementation for aoe done but got sidetracked. I'd welcome any volunteers who would like to do testing. Please email me directly.
-- Ed Cashin <ecashin@coraid.com>
| |