Re: [PATCH v2] btrfs: send: fix missing hole punching during incremental send

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

在 2026/10/22 08:11, fdmanana@kernel.org 写道:
> From: Filipe Manana <fdmanana@suse.com>
> 
> When using the no-holes feature (enabled by default since btrfs-progs
> v5.15) and doing an incremental send, we can miss issuing a hole punching
> command when the following conditions happen:
> 
> 1) A hole is punched in a tail section of a file extent item that happens
>     to be the last item in a leaf (for example the extent covers the file
>     range [1M, 3M) and a hole is punched in the range [2M, 3M)). Between
>     the end of this file extent item's range and the i_size, there are
>     other file extent items;
> 
> 2) The next leaf was not modified in the send snapshot, remaining shared
>     with the parent snapshot.
> 
> In this scenario we are unable to detect a hole was punched and never
> issue a hole punch command because with no-holes there we don't have an
> exlicit file extent item to mark a hole.
> 
> Fix this by teaching maybe_send_hole() to detect such case.
> A test case for fstests will follow soon.
> 
> Reported-by: Tristan Elvers <tristanelvers@gmail.com>
> Link: https://lore.kernel.org/linux-btrfs/CAFoLKK+aErV4p6ZJPnVeH5SHQ5++0f=8ZgxPWe-E-CkeKBs8PA@mail.gmail.com/
> Fixes: 16e7549f045d ("Btrfs: incompatible format change to remove hole extents")
> Signed-off-by: Filipe Manana <fdmanana@suse.com>
> ---
> 
> V2: Address a couple issues pointed by sashiko, simplify a few things.
> 
>   fs/btrfs/send.c | 62 +++++++++++++++++++++++++++++++++++++++++++++----
>   1 file changed, 58 insertions(+), 4 deletions(-)
> 
> diff --git a/fs/btrfs/send.c b/fs/btrfs/send.c
> index c523bf950c89..cec1aae5ce14 100644
> --- a/fs/btrfs/send.c
> +++ b/fs/btrfs/send.c
> @@ -6371,6 +6371,8 @@ static int range_is_hole_in_parent(struct send_ctx *sctx,
>   static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
>   			   struct btrfs_key *key)
>   {
> +	BTRFS_PATH_AUTO_FREE(next_path);
> +	u64 next_offset;
>   	int ret = 0;
>   
>   	if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
> @@ -6397,14 +6399,66 @@ static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
>   		ret = range_is_hole_in_parent(sctx,
>   					      sctx->cur_inode_last_extent,
>   					      key->offset);
> -		if (ret < 0)
> -			return ret;
> -		else if (ret == 0)
> +		if (ret == 0)
>   			ret = send_hole(sctx, key->offset);
> -		else
> +		else if (ret == 1)
>   			ret = 0;
> +
> +		if (ret < 0)
> +			return ret;
>   	}
>   	sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
> +
> +	/*
> +	 * If we are the last file extent item in a leaf, we may have had a tail
> +	 * section that got a hole punched, and if there is no change in the
> +	 * next leaf, we must find out if we have such a hole and punch it.
> +	 */
> +	if (path->slots[0] != btrfs_header_nritems(path->nodes[0]) - 1)
> +		return ret;

Just a nitpick, do we need to check NO_HOLES features? IIRC without 
NO_HOLES, we should always have a file extent covering i_size, thus 
without NO_HOLES, we should not need to check for the next leaf?

Thanks,
Qu

> +
> +	next_path = alloc_path_for_send();
> +	if (!next_path)
> +		return -ENOMEM;
> +
> +	ret = btrfs_search_slot(NULL, sctx->send_root, key, next_path, 0, 0);
> +	if (ret < 0)
> +		return ret;
> +	if (WARN_ON(ret != 0))
> +		return -EUCLEAN;
> +
> +	ret = btrfs_next_leaf(sctx->send_root, next_path);
> +	if (ret < 0) {
> +		return ret;
> +	} else if (ret > 0) {
> +		/*
> +		 * No next leaf, must punch hole to i_size and set the path back
> +		 * to where it was before.
> +		 */
> +		next_offset = sctx->cur_inode_size;
> +	} else {
> +		struct btrfs_key next_key;
> +
> +		btrfs_item_key_to_cpu(next_path->nodes[0], &next_key,
> +				      next_path->slots[0]);
> +		if (next_key.objectid != key->objectid ||
> +		    next_key.type != BTRFS_EXTENT_DATA_KEY)
> +			next_offset = sctx->cur_inode_size;
> +		else
> +			next_offset = next_key.offset;
> +	}
> +
> +	btrfs_free_path(next_path);
> +	next_path = NULL;
> +
> +	ret = range_is_hole_in_parent(sctx, sctx->cur_inode_last_extent, next_offset);
> +	if (ret == 0)
> +		ret = send_hole(sctx, next_offset);
> +	else if (ret == 1)
> +		ret = 0;
> +
> +	sctx->cur_inode_last_extent = next_offset;
> +
>   	return ret;
>   }
>