Re: [PATCH] xfs: cache open zone in inode->i_private

[hch <hch@lst.de>]

On Thu, Oct 16, 2025 at 05:17:31AM +0000, Hans Holmberg wrote:
> > +	 * reused for GC in the meantime.  Skip it in that case.
> > +	 */
> > +	rcu_read_lock();
> > +	oz = VFS_I(ip)->i_private;
> > +	if (oz && (oz->oz_is_gc || !atomic_inc_not_zero(&oz->oz_ref)))
> > +		oz = NULL;
> 
> We still won't steal any open zones for gc(except at mount time), so
> the keep the assert and comment below in stead?

True.  When writing this I thought that a zone pointed to could be
reused as the GC zone, and it could.  But the open zone would not
have the oz_is_gc set in that case, but instead we're protected by
it being marked as full inside the oz still.

I'll bring back the old assert and will write extensive comments
explaining all this for v2.