Re: [PATCH 02/11] readahead: push invalidate_lock out of page_cache_ra_unbounded

[Matthew Wilcox <willy@infradead.org>]

On Mon, Feb 02, 2026 at 07:06:31AM +0100, Christoph Hellwig wrote:
> +++ b/fs/f2fs/file.c
> @@ -4418,7 +4418,9 @@ static int redirty_blocks(struct inode *inode, pgoff_t page_idx, int len)
>  	pgoff_t redirty_idx = page_idx;
>  	int page_len = 0, ret = 0;
>  
> +	filemap_invalidate_lock_shared(mapping);
>  	page_cache_ra_unbounded(&ractl, len, 0);
> +	filemap_invalidate_unlock_shared(mapping);

Why is f2fs calling page_cache_ra_unbounded() here?  The documentation
literally says not to call it:

 * This function is for filesystems to call when they want to start
 * readahead beyond a file's stated i_size.  This is almost certainly
 * not the function you want to call.  Use page_cache_async_readahead()
 * or page_cache_sync_readahead() instead.

(in this case, f2fs doesn't have a folio, so page_cache_async_ra() is
probably the right function to call).  But what's the point in writing
documentation when people don't read it?

> @@ -228,9 +229,10 @@ void page_cache_ra_unbounded(struct readahead_control *ractl,
>  	 */
>  	unsigned int nofs = memalloc_nofs_save();
>  
> +	lockdep_assert_held_read(&mapping->invalidate_lock);

Hm, why are we asserting that it's not write-locked?  For the
purposes of this function, I'd think we want to just
lockdep_assert_held()?

In the tree I'm looking at, there are also calls to
page_cache_ra_unbounded() in fs/ext4/verity.c and fs/f2fs/verity.c
which probably need the lock taken too?