Re: [PATCH 6/8] btrfs: don't allocate file extent tree for non regular files

[David Sterba <dsterba@suse.cz>]

On Thu, May 09, 2024 at 09:41:50AM +0100, Filipe Manana wrote:
> On Thu, May 9, 2024 at 1:39 AM Qu Wenruo <quwenruo.btrfs@gmx.com> wrote:
> > 在 2024/5/8 21:47, fdmanana@kernel.org 写道:
> > > From: Filipe Manana <fdmanana@suse.com>
> > >
> > > When not using the NO_HOLES feature we always allocate an io tree for an
> > > inode's file_extent_tree.
> >
> > I'm wondering can we discard non-NO_HOLES support directly so that we
> > can get rid of the file_extent_tree completely?
> >
> > Initially I'm wondering why NO_HOLES makes a difference, especially as
> > NO_HOLES can be set halfway, thus we can still have explicit hole extents.
> >
> > But it turns out that the whole file_extent_tree() is only to handle
> > non-NO_HOLES case so that we always have a hole for the whole file range
> > until i_size.
> >
> > Considering NO_HOLES is considered safe at 4.0 kernel, can we start
> > deprecating non-NO_HOLES?
> 
> NO_HOLES is a mkfs default for only 2 or 3 years IIRC. It's not that long.
> 
> And how do you know everyone is already using NO_HOLES?
> Last week I saw a user filesystem that doesn't use it.
> 
> And probably there are many more.

I think so. Soft deprecation can start but we still have to keep the
code around and given that many filesystems do not change features since
mkfs the time ("mkfs and forget") there are many created before the 5.15
defaults change. So, estimmated hard deprecation time is when the kernel
5.15 will be out of LTS support upstream or distros we care about. Rough
estimate is 10 years from now.