On 2025/6/27 22:19, Tetsuo Handa wrote:
> When commit d3556babd7fa ("ocfs2: fix d_splice_alias() return code
> checking") was merged into v3.18-rc3, d_splice_alias() was returning
> one of a valid dentry, NULL or an ERR_PTR.
>
> When commit b5ae6b15bd73 ("merge d_materialise_unique() into
> d_splice_alias()") was merged into v3.19-rc1, d_splice_alias() started
> returning -ELOOP as one of ERR_PTR values.
>
> Now, when syzkaller mounts a crafted ocfs2 filesystem image that hits
> d_splice_alias() == -ELOOP case from ocfs2_lookup(), ocfs2_lookup() fails
> to handle -ELOOP case and generic_shutdown_super() hits "VFS: Busy inodes
> after unmount" message.
>
> Instead of calling ocfs2_dentry_attach_lock() or ocfs2_dentry_attach_gen()
> when d_splice_alias() returned an ERR_PTR value, change ocfs2_lookup() to
> bail out immediately.
>
> Also, ocfs2_lookup() needs to call dupt() when ocfs2_dentry_attach_lock()
> returned an ERR_PTR value.
>
> Reported-by: syzbot <syzbot+1134d3a5b062e9665a7a@xxxxxxxxxxxxxxxxxxxxxxxxx>
> Closes: https://syzkaller.appspot.com/bug?extid=1134d3a5b062e9665a7a
> Suggested-by: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
> Signed-off-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
> ---
> fs/ocfs2/namei.c | 9 +++++----
> 1 file changed, 5 insertions(+), 4 deletions(-)
>
> diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
> index 99278c8f0e24..f75fd19974bc 100644
> --- a/fs/ocfs2/namei.c
> +++ b/fs/ocfs2/namei.c
> @@ -142,6 +142,8 @@ static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
>
> bail_add:
> ret = d_splice_alias(inode, dentry);
> + if (IS_ERR(ret))
> + goto bail_unlock;
>
> if (inode) {
> /*
> @@ -154,13 +156,12 @@ static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
> * NOTE: This dentry already has ->d_op set from
> * ocfs2_get_parent() and ocfs2_get_dentry()
> */
> - if (!IS_ERR_OR_NULL(ret))
> - dentry = ret;
> -
I'd like change this to:
if (ret)
dentry = ret;
Instead of using "ret ? ret : dentry" below calling ocfs2_dentry_attach_lock().
> - status = ocfs2_dentry_attach_lock(dentry, inode,
> + status = ocfs2_dentry_attach_lock(ret ? ret : dentry, inode,
> OCFS2_I(dir)->ip_blkno);
> if (status) {
> mlog_errno(status);
> + if (ret)
> + dput(ret);
> ret = ERR_PTR(status);
> goto bail_unlock;
The "goto" here can be eliminated since it has no real effect.
Thanks,
Joseph
> }
