Re: [Bug 201823] New: inconsistent behaviour of xfs_db, xfs_repair and xfs_scrub

[Dave Chinner <david@fromorbit.com>]

On Sat, Dec 01, 2018 at 08:16:13AM +0000, bugzilla-daemon@bugzilla.kernel.org wrote:
> https://bugzilla.kernel.org/show_bug.cgi?id=201823
> 
>             Bug ID: 201823
>            Summary: inconsistent behaviour of xfs_db, xfs_repair and
>                     xfs_scrub
>            Product: File System
>            Version: 2.5
>     Kernel Version: 4.19.5
>           Hardware: All
>                 OS: Linux
>               Tree: Mainline
>             Status: NEW
>           Severity: normal
>           Priority: P1
>          Component: XFS
>           Assignee: filesystem_xfs@kernel-bugs.kernel.org
>           Reporter: matthias@bodenbinder.de
>         Regression: No
> 
> Hi,
> 
> I am using xfs on Manjaro with kernel 4.19 and latest xfsprogs 4.19.0-1.
> 
> I checked my root filesystem with xfs_db for fragmentation and got metadata
> corruption messages:
> 
> 5# xfs_db -c frag -r /dev/sde1
> Metadata corruption detected at 0x558ed74133f3, xfs_inode block 0x7faac0/0x8000
> Metadata corruption detected at 0x558ed74133f3, xfs_inode block
> 0x20cb300/0x8000
> Metadata corruption detected at 0x558ed74133f3, xfs_inode block
> 0x2341340/0x8000
> aktuell 976616, ideal 970547, Unterteilungsfaktor 0,62%
> Note, this number is largely meaningless.
> Files on this filesystem average 1,01 extents per file

That is expected if you are running xfs_db on a mounted filesystem.
What is on disk at any given instant is not consistent. The combined
in-memory state, the journal and what is on-disk is consistent, but
you cannot see all that extra in-memory state from xfs_db. If the
system crashes, journal replay is what brings it back to being
consistent on disk. IOWs, this is expected behaviour.

> But xfs_repiar is not mentioning anything and hence not repairing anything.

xfs_repair won't run on a mounted filesystem.

> Also, xfs_scrub is not complaining:
> 
> 8# xfs_scrub -v /    
> EXPERIMENTAL xfs_scrub program in use! Use at your own risk!
> Phase 1: Find filesystem geometry.
> /: using 8 threads to scrub.
> Phase 2: Check internal metadata.
> Phase 3: Scan all inodes.
> Phase 5: Check directory tree.
> Phase 7: Check summary counters.
> 41,6GiB data used;  1,3M inodes used.
> 41,5GiB data found; 1,3M inodes found.
> 1,3M inodes counted; 1,3M inodes checked.

Scrub runs online, so sees the in-memory state of the filesystem
along with what is on disk, and hence it doesn't find any
corruptions because there are none.

IOWs, there are no problems and everything is behaving as expected.

-Dave.
-- 
Dave Chinner
david@fromorbit.com