Re: [PATCH v4] btrfs: retry verity reads for not-uptodate Merkle folios

[David Sterba]

On Wed, Jul 22, 2026 at 10:54:35AM +0800, Yichong Chen wrote:
> btrfs_read_merkle_tree_page() can find a folio in the mapping that is not
> uptodate.  After taking the folio lock, the current code treats that state
> as a read error and returns -EIO.
> 
> That can make a previous transient read failure sticky.  If the failed read
> left a not-uptodate folio in the mapping, later callers find that folio and
> fail instead of retrying the read.
> 
> Keep the existing page-cache insertion and locking order, but retry the
> Merkle item read when a not-uptodate folio is found in the mapping.  Also
> unlock the folio when read_key_bytes() fails so that a later caller can
> lock it and retry the read.
> 
> Fixes: 06ed09351b67 ("btrfs: convert btrfs_read_merkle_tree_page() to use a folio")
> Reviewed-by: Boris Burkov <boris@xxxxxx>
> Signed-off-by: Yichong Chen <chenyichong@xxxxxxxxxxxxx>
> ---
> v4:
> - Add a comment explaining the locked uptodate recheck.
> - Add Boris' Reviewed-by.
> 
> v3:
> - Keep the existing filemap_add_folio() and read ordering.
> - Retry the Merkle item read when a not-uptodate folio is found, as
>   suggested by Boris.
> - Unlock the folio on read_key_bytes() failure so later callers can retry.
> 
> v2:
> - Avoid calling filemap_remove_folio(), which is not exported.
> - Add the folio to the page cache only after read_key_bytes() succeeds.
> ---
>  fs/btrfs/verity.c | 16 +++++++++++-----
>  1 file changed, 11 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/btrfs/verity.c b/fs/btrfs/verity.c
> index 983365a73541..1133a56c0568 100644
> --- a/fs/btrfs/verity.c
> +++ b/fs/btrfs/verity.c
> @@ -720,14 +720,18 @@ static struct page *btrfs_read_merkle_tree_page(struct inode *inode,
>  			goto out;
>  
>  		folio_lock(folio);
> -		/* If it's not uptodate after we have the lock, we got a read error. */
> -		if (!folio_test_uptodate(folio)) {
> +		/* Folio was truncated from mapping. */
> +		if (!folio->mapping) {
>  			folio_unlock(folio);
>  			folio_put(folio);
> -			return ERR_PTR(-EIO);
> +			goto again;
>  		}
> -		folio_unlock(folio);
> -		goto out;
> +		/* Another reader may have filled the folio while we waited. */
> +		if (folio_test_uptodate(folio)) {
> +			folio_unlock(folio);
> +			goto out;
> +		}
> +		goto read_folio;

With that new label the main loop becomes a less readable, previously
there was just again and out. From a distance the loop likes a big while
from again: to read_key_bytes, so this may be a good followup cleanup
(if the end result is reasonable).

>  	}
>  
>  	folio = filemap_alloc_folio(mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS),
> @@ -744,6 +748,7 @@ static struct page *btrfs_read_merkle_tree_page(struct inode *inode,
>  		return ERR_PTR(ret);
>  	}
>  
> +read_folio:
>  	/*
>  	 * Merkle item keys are indexed from byte 0 in the merkle tree.
>  	 * They have the form:
> @@ -753,6 +758,7 @@ static struct page *btrfs_read_merkle_tree_page(struct inode *inode,
>  	ret = read_key_bytes(BTRFS_I(inode), BTRFS_VERITY_MERKLE_ITEM_KEY, off,
>  			     folio_address(folio), PAGE_SIZE, folio);
>  	if (ret < 0) {
> +		folio_unlock(folio);
>  		folio_put(folio);
>  		return ERR_PTR(ret);
>  	}