Re: [PATCH] xfs: defer lazytime timestamp updates to inodegc during eviction

[Jan Kara]

On Wed 25-03-26 07:38:22, Christoph Hellwig wrote:
> On Tue, Mar 24, 2026 at 04:14:23PM +0100, Jan Kara wrote:
> > Isn't this problem already fixed by Christoph's patch
> > https://lore.kernel.org/all/20260317134409.1691317-1-hch@xxxxxx ?
> 
> It should.  My question to Ted what we're even trying to do in this
> context is also still outstanding.

Let me see. One reason is that some filesystems don't cache inodes at VFS
level (such as XFS but there are others which just always return 1 from
their ->drop_inode). For these we'd have to come up with some other place
where they write the outstanding time modification before evicting the
inode from memory.

Second effect is that without I_DIRTY_TIME writeback in iput()
__inode_lru_list_add() in iput_final() will not add these inodes to the lru
list and they won't be added there after flush worker did the dirty time
writeback either so they'd be sitting in memory pretty much until unmount
(or until somebody uses them again).

								Honza
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR