Re: [PATCH v16 15/21] xfs: add fs-verity support

[Christoph Hellwig <hch@infradead.org>]

On Tue, Sep 22, 2026 at 10:56:09AM +0200, Andrey Albershteyn wrote:
> On 2026-09-22 00:22:52, Christoph Hellwig wrote:
> > On Fri, Sep 18, 2026 at 01:15:27PM +0200, Andrey Albershteyn wrote:
> > > +	xfs_ilock(ip, XFS_ILOCK_EXCL);
> > > +	xfs_trans_ijoin(tp, ip, 0);
> > > +
> > > +	truncate_inode_pages(VFS_I(ip)->i_mapping, XFS_ISIZE(ip));
> > 
> > We can't call truncate_inode_pages with the ilock held.
> > 
> > I also don't see what this is trying to protect to start with.
> > 
> 
> I added this to remove stale pages created post-EOF, so, writeback
> doesn't try to flush them latter (but now I see that writeback will
> fail to map them to extents anyway). Also, if EOF gets extended
> these pages would be exposed as they are up-to-date which would be
> unexpected garbage.

Yeah, but what is the ilock trying to protect against?  ilock is
the innermost sleeping lock and protects modifications to the
inode fields and the extent mapping.   Higher level synchronization
is done using the iolock and the mmaplock (which is already taken
here).