kernfs: switch global kernfs_rename_lock to per-fs lock

The kernfs implementation has big lock granularity(kernfs_rename_lock) so
every kernfs-based(e.g., sysfs, cgroup) fs are able to compete the lock.

This patch switches the global kernfs_rename_lock to per-fs lock, which
put the rwlock into kernfs_root.

Signed-off-by: Jinliang Zheng <[email protected]>
Acked-by: Tejun Heo <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Greg Kroah-Hartman <[email protected]>

Author: fatherMatrix

fs/kernfs/dir.c

@@ -17,7 +17,6 @@
 
 #include "kernfs-internal.h"
 
-DEFINE_RWLOCK(kernfs_rename_lock);	/* kn->parent and ->name */
 /*
  * Don't use rename_lock to piggy back on pr_cont_buf. We don't want to
  * call pr_cont() while holding rename_lock. Because sometimes pr_cont()
@@ -228,7 +227,7 @@ int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
 	if (to) {
 		root = kernfs_root(to);
 		if (!(root->flags & KERNFS_ROOT_INVARIANT_PARENT)) {
-			guard(read_lock_irqsave)(&kernfs_rename_lock);
+			guard(read_lock_irqsave)(&root->kernfs_rename_lock);
 			return kernfs_path_from_node_locked(to, from, buf, buflen);
 		}
 	}
@@ -295,12 +294,14 @@ void pr_cont_kernfs_path(struct kernfs_node *kn)
 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
 {
 	struct kernfs_node *parent;
+	struct kernfs_root *root;
 	unsigned long flags;
 
-	read_lock_irqsave(&kernfs_rename_lock, flags);
+	root = kernfs_root(kn);
+	read_lock_irqsave(&root->kernfs_rename_lock, flags);
 	parent = kernfs_parent(kn);
 	kernfs_get(parent);
-	read_unlock_irqrestore(&kernfs_rename_lock, flags);
+	read_unlock_irqrestore(&root->kernfs_rename_lock, flags);
 
 	return parent;
 }
@@ -993,6 +994,7 @@ struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
 	init_rwsem(&root->kernfs_iattr_rwsem);
 	init_rwsem(&root->kernfs_supers_rwsem);
 	INIT_LIST_HEAD(&root->supers);
+	rwlock_init(&root->kernfs_rename_lock);
 
 	/*
 	 * On 64bit ino setups, id is ino.  On 32bit, low 32bits are ino.
@@ -1789,15 +1791,15 @@ int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
 	/* rename_lock protects ->parent accessors */
 	if (old_parent != new_parent) {
 		kernfs_get(new_parent);
-		write_lock_irq(&kernfs_rename_lock);
+		write_lock_irq(&root->kernfs_rename_lock);
 
 		rcu_assign_pointer(kn->__parent, new_parent);
 
 		kn->ns = new_ns;
 		if (new_name)
 			rcu_assign_pointer(kn->name, new_name);
 
-		write_unlock_irq(&kernfs_rename_lock);
+		write_unlock_irq(&root->kernfs_rename_lock);
 		kernfs_put(old_parent);
 	} else {
 		/* name assignment is RCU protected, parent is the same */

fs/kernfs/kernfs-internal.h

@@ -19,8 +19,6 @@
 #include <linux/kernfs.h>
 #include <linux/fs_context.h>
 
-extern rwlock_t kernfs_rename_lock;
-
 struct kernfs_iattrs {
 	kuid_t			ia_uid;
 	kgid_t			ia_gid;
@@ -53,6 +51,9 @@ struct kernfs_root {
 	struct rw_semaphore	kernfs_iattr_rwsem;
 	struct rw_semaphore	kernfs_supers_rwsem;
 
+	/* kn->parent and kn->name */
+	rwlock_t		kernfs_rename_lock;
+
 	struct rcu_head		rcu;
 };
 
@@ -108,6 +109,11 @@ static inline bool kernfs_root_is_locked(const struct kernfs_node *kn)
 	return lockdep_is_held(&kernfs_root(kn)->kernfs_rwsem);
 }
 
+static inline bool kernfs_rename_is_locked(const struct kernfs_node *kn)
+{
+	return lockdep_is_held(&kernfs_root(kn)->kernfs_rename_lock);
+}
+
 static inline const char *kernfs_rcu_name(const struct kernfs_node *kn)
 {
 	return rcu_dereference_check(kn->name, kernfs_root_is_locked(kn));
@@ -118,14 +124,15 @@ static inline struct kernfs_node *kernfs_parent(const struct kernfs_node *kn)
 	/*
 	 * The kernfs_node::__parent remains valid within a RCU section. The kn
 	 * can be reparented (and renamed) which changes the entry. This can be
-	 * avoided by locking kernfs_root::kernfs_rwsem or kernfs_rename_lock.
+	 * avoided by locking kernfs_root::kernfs_rwsem or
+	 * kernfs_root::kernfs_rename_lock.
 	 * Both locks can be used to obtain a reference on __parent. Once the
 	 * reference count reaches 0 then the node is about to be freed
 	 * and can not be renamed (or become a different parent) anymore.
 	 */
 	return rcu_dereference_check(kn->__parent,
 				     kernfs_root_is_locked(kn) ||
-				     lockdep_is_held(&kernfs_rename_lock) ||
+				     kernfs_rename_is_locked(kn) ||
 				     !atomic_read(&kn->count));
 }