On 24/09/2024 07:17, Christoph Hellwig wrote:
On Mon, Sep 23, 2024 at 01:33:12PM +0100, John Garry wrote:
As a first step by not making it worse, and that not only means not
spreading the rtextent stuff further,
I assume that refactoring rtextent into "big alloc unit" is spreading
(rtextent stuff), right? If so, what other solution? CoW?
Well, if you look at the force align series you'd agree that it
spreads the thing out into the btree allocator. Or do I misread it?
Yes, there are more changes than just refactoring "big alloc unit"
stuff. There are btree allocator changes.
About those btree allocator changes, strictly speaking there are just a
couple of changes to provide forcealign support - the rest are prep
patches. And those forcealign changes build on pre-existing allocator
features, like extent alignment and length specifiers.
but more importantly not introducing
additional complexities by requiring to be able to write over the
written/unwritten boundaries created by either rtextentsize > 1 or
the forcealign stuff if you actually want atomic writes.
The very original solution required a single mapping and in written state
for atomic writes. Reverting to that would save a lot of hassle in the
kernel. It just means that the user needs to manually pre-zero.
What atomic I/O sizes do your users require? Would they fit into
a large sector size now supported by XFS (i.e. 32k for now).
It could be used, but then we have 16KB filesystem block size, which
some just may not want. And we just don't want 16KB sector size, but I
don't think that we require that if we use RWF_ATOMIC.
