Re: ZFS data error without reasons

On Wednesday 25 March 2009 09:41:39 kevin wrote:
> Mark Powell wrote:
> > Kevin,
> > Did you fix your ZFS CRC errors?
> > I responded to your thread, but no-one got back to me.
> > I'm gonna start another thread later.
> > This time I re-made the zpool in 8 compatible with 7. Once the errors 
> > started showing up in 8 I moved back to 7, on the same hardware, to 
> > perform the scrub to prove the problem is with 8. The 1st scrub in 7 
> > found some errors, but of course it would if 8 had messed up the data. 
> > Removed the few unimportant bad files (all were in snapshots).
> > Just performing the 2nd scrub in 7 now. If this comes back with no 
> > errors, then we have stronger proof that there is some wrong, which 
> > seems quite intermittent, in 8 that randomly writes bad data.
> > Cheers.
> >
> Yes，I can fix some ZFS CRC errors,and sometimes i can recover all error 
> files.Before i run "zpool import backup" to mount the zpool on a usb 
> hard disk, "zpool status" return no errors. When i copy files to the usb 
> hard disk,soon I can get lots of file errors.After a reboot,if i run 
> scrub,i can fix many errors. I just think copy files between two zpools, 
> one is on local hard disk and the other one is on a usb hard disk, may 
> easily reproduce the bug.



I have not been folloing the entire thread - but I can reproduce ZFS CRC corruption on the
current kernel just by unpluging a USB disk drive  - The is no errors on the disks - revert to
and old kernel 
 FreeBSD w2fzz0vc03.aah-go-on.com 8.0-CURRENT FreeBSD 8.0-CURRENT #1 r189454M: Fri Mar  6 18:46:25 GMT 2009     root_at_w2fzz0vc03.aah-go-on.com:/usr/obj/usr/src/sys/GENERIC  amd64

the problem can be solved - The weird thing is that it will give CRC errros (and permenent errors) in blocks that has not been touched (or at least I think so) 
I suspect that It may have to do with the USB DMA bounce buffer as an example see the message file included