[RFC PATCH 15/25] reducing rcu_read_lock() scopes in dput and friends, step 4

[Al Viro <viro@zeniv.linux.org.uk>]

Shrink rcu_read_lock() scope around lock_for_kill() into
lock_for_kill() itself.  Note that spin_unlock() can be
transposed with rcu_read_unlock() - RCU read-side critical
area remains the same and even if the structure containing
the spinlock gets freed from RCU callback run immediately
after spin_unlock(), we are OK - it won't run until after
the last access to spinlock done by spin_unlock().

Inside lock_for_kill() the scope can be shrunk down to the
area where we drop and regain ->d_lock - rcu_read_lock() scope
is needed only to bridge the ->d_lock scopes together.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/dcache.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/fs/dcache.c b/fs/dcache.c
index bcaf17840939..4222d89692a8 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -763,6 +763,7 @@ static bool lock_for_kill(struct dentry *dentry)
 	if (!inode || likely(spin_trylock(&inode->i_lock)))
 		return true;
 
+	rcu_read_lock();
 	do {
 		spin_unlock(&dentry->d_lock);
 		spin_lock(&inode->i_lock);
@@ -772,6 +773,7 @@ static bool lock_for_kill(struct dentry *dentry)
 		spin_unlock(&inode->i_lock);
 		inode = dentry->d_inode;
 	} while (inode);
+	rcu_read_unlock();
 	if (likely(!dentry->d_lockref.count))
 		return true;
 	if (inode)
@@ -781,13 +783,10 @@ static bool lock_for_kill(struct dentry *dentry)
 
 static struct dentry *dentry_kill(struct dentry *dentry)
 {
-	rcu_read_lock();
 	if (unlikely(!lock_for_kill(dentry))) {
 		spin_unlock(&dentry->d_lock);
-		rcu_read_unlock();
 		return NULL;
 	}
-	rcu_read_unlock();
 	return __dentry_kill(dentry);
 }
 
-- 
2.47.3