VFS: Impose ordering on accesses of d_inode and d_flags
Impose ordering on accesses of d_inode and d_flags to avoid the need to do
this:
if (!dentry->d_inode || d_is_negative(dentry)) {
when this:
if (d_is_negative(dentry)) {
should suffice.
This check is especially problematic if a dentry can have its type field set
to something other than DENTRY_MISS_TYPE when d_inode is NULL (as in
unionmount).
What we really need to do is stick a write barrier between setting d_inode and
setting d_flags and a read barrier between reading d_flags and reading
d_inode.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
diff --git a/fs/dcache.c b/fs/dcache.c
index d99736a..656ce52 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -269,6 +269,41 @@
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)
+{
+ 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;
+}
+
static void dentry_free(struct dentry *dentry)
{
WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
@@ -311,7 +346,7 @@
{
struct inode *inode = dentry->d_inode;
if (inode) {
- dentry->d_inode = NULL;
+ __d_clear_type_and_inode(dentry);
hlist_del_init(&dentry->d_u.d_alias);
spin_unlock(&dentry->d_lock);
spin_unlock(&inode->i_lock);
@@ -335,8 +370,7 @@
__releases(dentry->d_inode->i_lock)
{
struct inode *inode = dentry->d_inode;
- __d_clear_type(dentry);
- dentry->d_inode = NULL;
+ __d_clear_type_and_inode(dentry);
hlist_del_init(&dentry->d_u.d_alias);
dentry_rcuwalk_barrier(dentry);
spin_unlock(&dentry->d_lock);
@@ -1715,11 +1749,9 @@
unsigned add_flags = d_flags_for_inode(inode);
spin_lock(&dentry->d_lock);
- dentry->d_flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
- dentry->d_flags |= add_flags;
if (inode)
hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
- dentry->d_inode = inode;
+ __d_set_inode_and_type(dentry, inode, add_flags);
dentry_rcuwalk_barrier(dentry);
spin_unlock(&dentry->d_lock);
fsnotify_d_instantiate(dentry, inode);
@@ -1937,8 +1969,7 @@
add_flags |= DCACHE_DISCONNECTED;
spin_lock(&tmp->d_lock);
- tmp->d_inode = inode;
- tmp->d_flags |= add_flags;
+ __d_set_inode_and_type(tmp, inode, add_flags);
hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
hlist_bl_lock(&tmp->d_sb->s_anon);
hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);