On Wed, May 11, 2005 at 01:57:12PM +0200, Leon Vismer wrote:
> Hi Dave
>
> I did the following:
>
> # parted /dev/sdc
> parted> mklabel gpt
> parted> mkpart primary 0 2384080
> parted> quit
>
> I saw the following in tail /var/log/messages
>
> May 11 13:37:40 mail kernel: program parted is using a deprecated SCSI ioctl,
> please convert it to SG_IO
> May 11 13:37:50 mail kernel: sym1:4:0:phase change 2-7 16@01b75f60 resid=10.
>
> # xfs_check /dev/sdc1 (gives the following)
>
> bad sb magic # 0xb0b25aab in ag 30
> bad sb version # 0x9d8b in ag 30
> bad agf magic # 0x9be1c556 in ag 30
> bad agf version # 0xf2c55d8a in ag 30
> bad agi magic # 0xe0467f8a in ag 30
> bad agi version # 0xb4f92355 in ag 30
.....
Looks like binary file data in AG 30 and 31. Given that mkfs
probably gave you 32 AGs in your filesystem, these 2 AGs are the
only ones that lie totally above the 2TiB filesystem offset.
From this, I'm almost certain that accesses to blocks > 2TiB are
wrapping back to block zero. Can you rebuild a kernel yourself
so we are certain that it is built with CONFIG_LBD=y?
Hmmm - just a random though - the SCSI protocol is limited to
addressing 2^32 sectors, which is 2TiB on a 512 byte sector
size device. You're trying to address a 2.4TiB SCSI device -
what sector size is your RAID controller using? If you
set it to 1k or 2k and use the mkfs option "-s size=xxxx"
do you see this same problem?
Cheers,
Dave.
--
Dave Chinner
R&D Software Engineer
SGI Australian Software Group
|