ovl: enable RCU'd ->get_acl()

Overlayfs does not cache ACL's (to avoid double caching).  Instead it just
calls the underlying filesystem's i_op->get_acl(), which will return the
cached value, if possible.

In rcu path walk, however, get_cached_acl_rcu() is employed to get the
value from the cache, which will fail on overlayfs resulting in dropping
out of rcu walk mode.  This can result in a big performance hit in certain
situations.

Fix by calling ->get_acl() with rcu=true in case of ACL_DONT_CACHE (which
indicates pass-through)

Reported-by: garyhuang <zjh.20052005@163.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

4 files changed, 23 insertions, 5 deletions

diff --git a/fs/overlayfs/inode.c b/fs/overlayfs/inode.c
index ea335d3e55cf..832b17589733 100644
--- a/fs/overlayfs/inode.c
+++ b/fs/overlayfs/inode.c
@@ -13,6 +13,7 @@
 #include <linux/fiemap.h>
 #include <linux/fileattr.h>
 #include <linux/security.h>
+#include <linux/namei.h>
 #include "overlayfs.h"
 
 
@@ -452,12 +453,12 @@ struct posix_acl *ovl_get_acl(struct inode *inode, int type, bool rcu)
 	const struct cred *old_cred;
 	struct posix_acl *acl;
 
-	if (rcu)
-		return ERR_PTR(-ECHILD);
-
 	if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
 		return NULL;
 
+	if (rcu)
+		return get_cached_acl_rcu(realinode, type);
+
 	old_cred = ovl_override_creds(inode->i_sb);
 	acl = get_acl(realinode, type);
 	revert_creds(old_cred);
diff --git a/fs/posix_acl.c b/fs/posix_acl.c
index 4f1ef826e040..f5c25f580dd9 100644
--- a/fs/posix_acl.c
+++ b/fs/posix_acl.c
@@ -22,6 +22,7 @@
 #include <linux/xattr.h>
 #include <linux/export.h>
 #include <linux/user_namespace.h>
+#include <linux/namei.h>
 
 static struct posix_acl **acl_by_type(struct inode *inode, int type)
 {
@@ -56,7 +57,17 @@ EXPORT_SYMBOL(get_cached_acl);
 
 struct posix_acl *get_cached_acl_rcu(struct inode *inode, int type)
 {
-	return rcu_dereference(*acl_by_type(inode, type));
+	struct posix_acl *acl = rcu_dereference(*acl_by_type(inode, type));
+
+	if (acl == ACL_DONT_CACHE) {
+		struct posix_acl *ret;
+
+		ret = inode->i_op->get_acl(inode, type, LOOKUP_RCU);
+		if (!IS_ERR(ret))
+			acl = ret;
+	}
+
+	return acl;
 }
 EXPORT_SYMBOL(get_cached_acl_rcu);
 
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 73376dfe28d0..c6e5bcbff0c0 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -581,6 +581,11 @@ static inline void mapping_allow_writable(struct address_space *mapping)
 
 struct posix_acl;
 #define ACL_NOT_CACHED ((void *)(-1))
+/*
+ * ACL_DONT_CACHE is for stacked filesystems, that rely on underlying fs to
+ * cache the ACL.  This also means that ->get_acl() can be called in RCU mode
+ * with the LOOKUP_RCU flag.
+ */
 #define ACL_DONT_CACHE ((void *)(-3))
 
 static inline struct posix_acl *
diff --git a/include/linux/posix_acl.h b/include/linux/posix_acl.h
index 307094ebb88c..b65c877d92b8 100644
--- a/include/linux/posix_acl.h
+++ b/include/linux/posix_acl.h
@@ -72,6 +72,8 @@ extern struct posix_acl *get_posix_acl(struct inode *, int);
 extern int set_posix_acl(struct user_namespace *, struct inode *, int,
 			 struct posix_acl *);
 
+struct posix_acl *get_cached_acl_rcu(struct inode *inode, int type);
+
 #ifdef CONFIG_FS_POSIX_ACL
 int posix_acl_chmod(struct user_namespace *, struct inode *, umode_t);
 extern int posix_acl_create(struct inode *, umode_t *, struct posix_acl **,
@@ -84,7 +86,6 @@ extern int simple_set_acl(struct user_namespace *, struct inode *,
 extern int simple_acl_create(struct inode *, struct inode *);
 
 struct posix_acl *get_cached_acl(struct inode *inode, int type);
-struct posix_acl *get_cached_acl_rcu(struct inode *inode, int type);
 void set_cached_acl(struct inode *inode, int type, struct posix_acl *acl);
 void forget_cached_acl(struct inode *inode, int type);
 void forget_all_cached_acls(struct inode *inode);