invalid inode numbers ? after drive moved from Linux 2.6.32 to Linux 3.16.0 then back again

Nigel Tamplin ntamplin at codefaber.co.uk
Wed Jun 8 07:31:54 CDT 2016


On 08/06/16 12:49, Carlos Maiolino wrote:
> On Wed, Jun 08, 2016 at 12:16:10PM +0100, Nigel Tamplin wrote:
>> Hello,
>>
>> I have a drive with a partition containing an XFS file system.
>> This drive normally resides in a Linux 2.6.32 (Debian 6) server, which is
>> also where the file system was created many months ago.
>> Last week I removed this drive from it's usual 2.6.32 server and for a few
>> hours it was attached to an alternative Linux 3.16.0 (Debian 8) server,
>> where the XFS file system was mounted and during this time I created
>> /modified / deleted some files.
>> Later the drive was moved back to the normal Linux 2.6.32.
>>
>
> Linux 3.16 has inode64 mount option enabled by default, which will cause inodes
> to be spread along the whole disk, causing some inodes (those allocated on disk
> blocks beyond 1TB offset) to have 64bit numbers.
>
> 2.6.32 doesn't use inode64 option by default, and so, files with inodes > 2^32
> will not be accessible by the kernel 2.6.32, but, you can use `mount -o inode64`
> mount option in the old system and then be able to access these inodes again.
>

Excellent explanation Carlos.
This exactly explains what had happened, the partition being moved 
between systems is 2TB in size.
As you suggested, remounting on the older systems with option inode64 
resolved it.

Thank you.



More information about the xfs mailing list