Eric Sandeen wrote:
Jeremy Katz wrote:
Yes, but it's a hack that was added at the explicit request of the XFS
developers with a "this will fix the problems". And having to
constantly change it because they don't want to have the same semantics
as all of the other filesystems which are even pseudo-supported is a
waste of time.
Jeremy, I do apologize for the previous workaround which ... didn't work.
But I have to take issue with the "semantics" you mentioned. Nothing in
the linux kernel guarantees that the block device address space will be
coherent with the filesystem address space - so what grub is trying to
do here (write through the filesystem, read back via the block device
with fs still mounted) is fundamentally broken.
ext2 seems less prone to problems, I'm not sure why. But there is no
guarantee or mechanism in the kernel to make what grub is doing
bulletproof. (last I looked there were lots of wishful "syncs" in the
grub code, along with comments that did not inspire confidence!)
So, if as Eric says, the previous workaround doesn't work, can we have
the remount workaround (that works) instead?
Thanks
James Pearson
