[Ocfs2-devel] [PATCH V2] ocfs2: kill EBUSY from dlmfs_evict_inode

[Joseph Qi]

On 6/8/22 1:12 AM, Junxiao Bi wrote:
> When unlink a dlmfs, first it will invoke dlmfs_unlink(), and then invoke
> dlmfs_evict_inode(), user_dlm_destroy_lock() is invoked in both places,
> the second one from dlmfs_evict_inode() will get EBUSY error because
> USER_LOCK_IN_TEARDOWN is already set in lockres. This doesn't affect
> any function, just the error log is annoying.
> 
> Signed-off-by: Junxiao Bi <junxiao.bi at oracle.com>

Reviewed-by: Joseph Qi <joseph.qi at linux.alibaba.com>
> ---
> V2 <- V1:
>  - s/anonying/annoying in patch log.
>  - s/destroyed/teardown in dlmfs_evict_inode()
> 
>  fs/ocfs2/dlmfs/dlmfs.c | 14 +++++++++++---
>  1 file changed, 11 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/ocfs2/dlmfs/dlmfs.c b/fs/ocfs2/dlmfs/dlmfs.c
> index e360543ad7e7..8b2020f92b5f 100644
> --- a/fs/ocfs2/dlmfs/dlmfs.c
> +++ b/fs/ocfs2/dlmfs/dlmfs.c
> @@ -296,17 +296,25 @@ static void dlmfs_evict_inode(struct inode *inode)
>  {
>  	int status;
>  	struct dlmfs_inode_private *ip;
> +	struct user_lock_res *lockres;
> +	int teardown;
>  
>  	clear_inode(inode);
>  
>  	mlog(0, "inode %lu\n", inode->i_ino);
>  
>  	ip = DLMFS_I(inode);
> +	lockres = &ip->ip_lockres;
>  
>  	if (S_ISREG(inode->i_mode)) {
> -		status = user_dlm_destroy_lock(&ip->ip_lockres);
> -		if (status < 0)
> -			mlog_errno(status);
> +		spin_lock(&lockres->l_lock);
> +		teardown = !!(lockres->l_flags & USER_LOCK_IN_TEARDOWN);
> +		spin_unlock(&lockres->l_lock);
> +		if (!teardown) {
> +			status = user_dlm_destroy_lock(lockres);
> +			if (status < 0)
> +				mlog_errno(status);
> +		}
>  		iput(ip->ip_parent);
>  		goto clear_fields;
>  	}