On Aug 2, 6:28am, Martin K. Petersen wrote:
> Subject: Re: xfs with LVM
> >>>>> "Steve" == Steve Lord <lord@xxxxxxx> writes:
> Steve> Martin, are you out there to respond to this?
> Yes, I'm here. Been busy all day, though, so I haven't poked at it
> yet. Will have a look later today or tomorrow...
> >> Very minor problem. LVM complains about xfs_reapair doing a
> >> unknown ioctl:
> >> Jul 31 23:00:04 localhost kernel: lvm -- lvm_blk_ioctl: unknown
> >> command 25636
> >> I traced this to some code in the sim library that using the ioctl
> >> DIOCGETVOLDEV
This is a hangover from the port from IRIX (have a look in
/usr/include/sys/dkio.h on a recent IRIX) ... there is no such
ioctl on Linux. In the reworked libsim, I have commented this
sort of thing out using a '#ifdef HAVE_VOLUME_MANAGER' ... see
libxfs/init.c. The tools which use the sim_init code that have
been converted (logprint and db) will no longer issue this bogus
ioctl, whereas the tools which have not yet been converted (i.e.
mkfs, repair) still go through the code in sim/src/sim.random.c
... these will be fixed in the not too distant future.
I'm guessing Martin may be working on a way to do the equivalent
of this IRIX ioctl in Linux/LVM (?)... on IRIX the ioctl tells
the sim_init which data/log/realtime subvolumes are associated
with a given XLV/XVM device.