[LU-17295] avoiding VFS exclusive locking on shared parent directory in ->atomic_open()

• Status: Open
• Project: Lustre
• Type: Improvement
• Priority: Minor
• Reporter: Qian Yingjin
• Assignee: Qian Yingjin
• Resolution: Unresolved
• Severity: 3

Description

This is an experimental feature to avoid the VFS exclusive locking on parent directory in ->atomic_open during file creations.

It converted the exclusive semaphore mutex into a shared lock to avoid waiting for the heavy creation RPC to finished 

Comments

[Qian Yingjin]

Found a possible bug in Lustre:

 

        static int ll_atomic_open()
        /* VFS has locked the inode before calling this */
        ll_set_inode_lock_owner(dir); 

As the dir maybe lock in shared mode, we may wrongly thought that the @dir is locked in exclusive mode:

 

        if (open_flag & O_CREAT)
		inode_lock(dir->d_inode);
	else
		inode_lock_shared(dir->d_inode);
	error = lookup_open(nd, &path, file, op, got_write);
	if (open_flag & O_CREAT)
		inode_unlock(dir->d_inode);
	else
		inode_unlock_shared(dir->d_inode);

 

in vvp_inode_ops(),

we may unlock the inode wrongly...

 

case COIO_INODE_UNLOCK:
        if (ll_get_inode_lock_owner(inode) == current)
               ll_inode_unlock(inode); 

 

 

 

[Gerrit Updater]

"Qian Yingjin <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/53169
Subject: LU-17295 llite: avoid VFS exclusive locking in atomic_open
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: a5d02885d5eaacaefc10c7962056587b74199daf