On 2017/8/4 下午2:46, Stefan Priebe - Profihost AG wrote:

Hello,

under very high load i'm getting the following error on bcache devices
since upgrading from opensuse 42.2 to 42.3.

David Chinner and others pointed me that this must be an bcache problem
not an XFS one.

Is this an error or problem you know and just a patch is missing in
opensuse?

XFS (bcache1): metadata I/O error: block 0x575b6d70
("xfs_trans_read_buf_map") error 4 numblks 16
XFS (bcache1): xfs_imap_to_bp: xfs_trans_read_buf() returned error -4.
XFS (bcache1): xfs_do_force_shutdown(0x8) called from line 3519 of file
fs/xfs/xfs_inode.c. Return address = 0xffffffff993441ff
XFS (bcache1): Corruption of in-memory data detected. Shutting down
filesystem
XFS (bcache1): Please umount the filesystem and rectify the problem(s)
XFS (bcache1): xfs_log_force: error -5 returned.
XFS (bcache1): xfs_log_force: error -5 returned.
XFS (bcache1): xfs_log_force: error -5 returned.

Hi Stefan,

In Leap 42.3 there is a patch prevents bcache from recovering I/O error
on cached device for non-writethrough mode. This is to avoid recovering
staled data from backing device. And this patch is not upstream yet (I
am working on a more proper version for upstream at mean time).

So it is possible that bcache encounter I/O error on cache device for
writeback mode, then xfs code receives -EIO (-5) in the above kernel log.

I see on bugzilla.opensuse.org you mentioned after reverting this patch,
no error reported any more. Let's see what happens inside, and for
upstream version patch, there is sysfs option to disable/enable the I/O
recovery behavior.

Thanks.

-- 
Coly Li