2021-05-25 15:42:01

by David Howells

Subject: [PATCH] afs: Fix the nlink handling of dir-over-dir rename

Fix rename of one directory over another such that the nlink on the deleted
directory is cleared to 0 rather than being decremented to 1.

This was causing the generic/035 xfstest to fail.

Fixes: e49c7b2f6de7 ("afs: Build an abstraction around an "operation" concept")
Signed-off-by: David Howells <[email protected]>
cc: [email protected]
---

fs/afs/dir.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/fs/afs/dir.c b/fs/afs/dir.c
index 9fbe5a5ec9bd..78719f2f567e 100644
--- a/fs/afs/dir.c
+++ b/fs/afs/dir.c
@@ -1919,7 +1919,9 @@ static void afs_rename_edit_dir(struct afs_operation *op)
new_inode = d_inode(new_dentry);
if (new_inode) {
spin_lock(&new_inode->i_lock);
- if (new_inode->i_nlink > 0)
+ if (S_ISDIR(new_inode->i_mode))
+ clear_nlink(new_inode);
+ else if (new_inode->i_nlink > 0)
drop_nlink(new_inode);
spin_unlock(&new_inode->i_lock);
}

2021-05-25 18:08:27

by Marc Dionne

Subject: Re: [PATCH] afs: Fix the nlink handling of dir-over-dir rename

On Tue, May 25, 2021 at 8:57 AM David Howells <[email protected]> wrote:
>
> Fix rename of one directory over another such that the nlink on the deleted
> directory is cleared to 0 rather than being decremented to 1.
>
> This was causing the generic/035 xfstest to fail.
>
> Fixes: e49c7b2f6de7 ("afs: Build an abstraction around an "operation" concept")
> Signed-off-by: David Howells <[email protected]>
> cc: [email protected]
> ---
>
> fs/afs/dir.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/fs/afs/dir.c b/fs/afs/dir.c
> index 9fbe5a5ec9bd..78719f2f567e 100644
> --- a/fs/afs/dir.c
> +++ b/fs/afs/dir.c
> @@ -1919,7 +1919,9 @@ static void afs_rename_edit_dir(struct afs_operation *op)
> new_inode = d_inode(new_dentry);
> if (new_inode) {
> spin_lock(&new_inode->i_lock);
> - if (new_inode->i_nlink > 0)
> + if (S_ISDIR(new_inode->i_mode))
> + clear_nlink(new_inode);
> + else if (new_inode->i_nlink > 0)
> drop_nlink(new_inode);
> spin_unlock(&new_inode->i_lock);
> }

Reviewed-by: Marc Dionne <[email protected]>

Marc