Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs

Pull d_inode/d_flags race fix from Al Viro.

I love this fix.  Not only does it fix the race in the dentry type
handling, it entirely gets rid of the nasty and subtle memory ordering
rules for d_type and d_inode, and replaces them with the basic dentry
locking rules (sequence numbers under RCU, d_lock elsewhere).

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  use ->d_seq to get coherency between ->d_inode and ->d_flags

2 files changed, 6 insertions, 18 deletions

diff --git a/fs/dcache.c b/fs/dcache.c
index 92d5140de851..2398f9f94337 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -269,9 +269,6 @@ static inline int dname_external(const struct dentry *dentry)
 	return dentry->d_name.name != dentry->d_iname;
 }
 
-/*
- * Make sure other CPUs see the inode attached before the type is set.
- */
 static inline void __d_set_inode_and_type(struct dentry *dentry,
 					  struct inode *inode,
 					  unsigned type_flags)
@@ -279,28 +276,18 @@ static inline void __d_set_inode_and_type(struct dentry *dentry,
 	unsigned flags;
 
 	dentry->d_inode = inode;
-	smp_wmb();
 	flags = READ_ONCE(dentry->d_flags);
 	flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
 	flags |= type_flags;
 	WRITE_ONCE(dentry->d_flags, flags);
 }
 
-/*
- * Ideally, we want to make sure that other CPUs see the flags cleared before
- * the inode is detached, but this is really a violation of RCU principles
- * since the ordering suggests we should always set inode before flags.
- *
- * We should instead replace or discard the entire dentry - but that sucks
- * performancewise on mass deletion/rename.
- */
 static inline void __d_clear_type_and_inode(struct dentry *dentry)
 {
 	unsigned flags = READ_ONCE(dentry->d_flags);
 
 	flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
 	WRITE_ONCE(dentry->d_flags, flags);
-	smp_wmb();
 	dentry->d_inode = NULL;
 }
 
@@ -370,9 +357,11 @@ static void dentry_unlink_inode(struct dentry * dentry)
 	__releases(dentry->d_inode->i_lock)
 {
 	struct inode *inode = dentry->d_inode;
+
+	raw_write_seqcount_begin(&dentry->d_seq);
 	__d_clear_type_and_inode(dentry);
 	hlist_del_init(&dentry->d_u.d_alias);
-	dentry_rcuwalk_invalidate(dentry);
+	raw_write_seqcount_end(&dentry->d_seq);
 	spin_unlock(&dentry->d_lock);
 	spin_unlock(&inode->i_lock);
 	if (!inode->i_nlink)
@@ -1758,8 +1747,9 @@ static void __d_instantiate(struct dentry *dentry, struct inode *inode)
 	spin_lock(&dentry->d_lock);
 	if (inode)
 		hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
+	raw_write_seqcount_begin(&dentry->d_seq);
 	__d_set_inode_and_type(dentry, inode, add_flags);
-	dentry_rcuwalk_invalidate(dentry);
+	raw_write_seqcount_end(&dentry->d_seq);
 	spin_unlock(&dentry->d_lock);
 	fsnotify_d_instantiate(dentry, inode);
 }
diff --git a/include/linux/dcache.h b/include/linux/dcache.h
index 7781ce110503..c4b5f4b3f8f8 100644
--- a/include/linux/dcache.h
+++ b/include/linux/dcache.h
@@ -409,9 +409,7 @@ static inline bool d_mountpoint(const struct dentry *dentry)
  */
 static inline unsigned __d_entry_type(const struct dentry *dentry)
 {
-	unsigned type = READ_ONCE(dentry->d_flags);
-	smp_rmb();
-	return type & DCACHE_ENTRY_TYPE;
+	return dentry->d_flags & DCACHE_ENTRY_TYPE;
 }
 
 static inline bool d_is_miss(const struct dentry *dentry)