kernfs: Use RCU to access kernfs_node::parent.

kernfs_rename_lock is used to obtain stable kernfs_node::{name|parent}
pointer. This is a preparation to access kernfs_node::parent under RCU
and ensure that the pointer remains stable under the RCU lifetime
guarantees.

For a complete path, as it is done in kernfs_path_from_node(), the
kernfs_rename_lock is still required in order to obtain a stable parent
relationship while computing the relevant node depth. This must not
change while the nodes are inspected in order to build the path.
If the kernfs user never moves the nodes (changes the parent) then the
kernfs_rename_lock is not required and the RCU guarantees are
sufficient. This "restriction" can be set with
KERNFS_ROOT_INVARIANT_PARENT. Otherwise the lock is required.

Rename kernfs_node::parent to kernfs_node::__parent to denote the RCU
access and use RCU accessor while accessing the node.
Make cgroup use KERNFS_ROOT_INVARIANT_PARENT since the parent here can
not change.

Acked-by: Tejun Heo <tj@kernel.org>
Cc: Yonghong Song <yonghong.song@linux.dev>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Link: https://lore.kernel.org/r/20250213145023.2820193-6-bigeasy@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

4 files changed, 106 insertions, 55 deletions

diff --git a/fs/kernfs/dir.c b/fs/kernfs/dir.c
index 43fbada67838..1d370c497e8a 100644
--- a/fs/kernfs/dir.c
+++ b/fs/kernfs/dir.c
@@ -17,7 +17,7 @@
 
 #include "kernfs-internal.h"
 
-static DEFINE_RWLOCK(kernfs_rename_lock);	/* kn->parent and ->name */
+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()
@@ -56,7 +56,7 @@ static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen)
 	if (!kn)
 		return strscpy(buf, "(null)", buflen);
 
-	return strscpy(buf, kn->parent ? kn->name : "/", buflen);
+	return strscpy(buf, rcu_access_pointer(kn->__parent) ? kn->name : "/", buflen);
 }
 
 /* kernfs_node_depth - compute depth from @from to @to */
@@ -64,9 +64,9 @@ static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
 {
 	size_t depth = 0;
 
-	while (to->parent && to != from) {
+	while (rcu_dereference(to->__parent) && to != from) {
 		depth++;
-		to = to->parent;
+		to = rcu_dereference(to->__parent);
 	}
 	return depth;
 }
@@ -84,18 +84,18 @@ static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
 	db = kernfs_depth(rb->kn, b);
 
 	while (da > db) {
-		a = a->parent;
+		a = rcu_dereference(a->__parent);
 		da--;
 	}
 	while (db > da) {
-		b = b->parent;
+		b = rcu_dereference(b->__parent);
 		db--;
 	}
 
 	/* worst case b and a will be the same at root */
 	while (b != a) {
-		b = b->parent;
-		a = a->parent;
+		b = rcu_dereference(b->__parent);
+		a = rcu_dereference(a->__parent);
 	}
 
 	return a;
@@ -168,8 +168,9 @@ static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
 
 	/* Calculate how many bytes we need for the rest */
 	for (i = depth_to - 1; i >= 0; i--) {
+
 		for (kn = kn_to, j = 0; j < i; j++)
-			kn = kn->parent;
+			kn = rcu_dereference(kn->__parent);
 
 		len += scnprintf(buf + len, buflen - len, "/%s", kn->name);
 	}
@@ -226,6 +227,7 @@ int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
 	unsigned long flags;
 	int ret;
 
+	guard(rcu)();
 	read_lock_irqsave(&kernfs_rename_lock, flags);
 	ret = kernfs_path_from_node_locked(to, from, buf, buflen);
 	read_unlock_irqrestore(&kernfs_rename_lock, flags);
@@ -295,7 +297,7 @@ struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
 	unsigned long flags;
 
 	read_lock_irqsave(&kernfs_rename_lock, flags);
-	parent = kn->parent;
+	parent = kernfs_parent(kn);
 	kernfs_get(parent);
 	read_unlock_irqrestore(&kernfs_rename_lock, flags);
 
@@ -360,8 +362,12 @@ static int kernfs_sd_compare(const struct kernfs_node *left,
  */
 static int kernfs_link_sibling(struct kernfs_node *kn)
 {
-	struct rb_node **node = &kn->parent->dir.children.rb_node;
 	struct rb_node *parent = NULL;
+	struct kernfs_node *kn_parent;
+	struct rb_node **node;
+
+	kn_parent = kernfs_parent(kn);
+	node = &kn_parent->dir.children.rb_node;
 
 	while (*node) {
 		struct kernfs_node *pos;
@@ -380,13 +386,13 @@ static int kernfs_link_sibling(struct kernfs_node *kn)
 
 	/* add new node and rebalance the tree */
 	rb_link_node(&kn->rb, parent, node);
-	rb_insert_color(&kn->rb, &kn->parent->dir.children);
+	rb_insert_color(&kn->rb, &kn_parent->dir.children);
 
 	/* successfully added, account subdir number */
 	down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
 	if (kernfs_type(kn) == KERNFS_DIR)
-		kn->parent->dir.subdirs++;
-	kernfs_inc_rev(kn->parent);
+		kn_parent->dir.subdirs++;
+	kernfs_inc_rev(kn_parent);
 	up_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
 
 	return 0;
@@ -407,16 +413,19 @@ static int kernfs_link_sibling(struct kernfs_node *kn)
  */
 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
 {
+	struct kernfs_node *kn_parent;
+
 	if (RB_EMPTY_NODE(&kn->rb))
 		return false;
 
+	kn_parent = kernfs_parent(kn);
 	down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
 	if (kernfs_type(kn) == KERNFS_DIR)
-		kn->parent->dir.subdirs--;
-	kernfs_inc_rev(kn->parent);
+		kn_parent->dir.subdirs--;
+	kernfs_inc_rev(kn_parent);
 	up_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
 
-	rb_erase(&kn->rb, &kn->parent->dir.children);
+	rb_erase(&kn->rb, &kn_parent->dir.children);
 	RB_CLEAR_NODE(&kn->rb);
 	return true;
 }
@@ -562,7 +571,7 @@ void kernfs_put(struct kernfs_node *kn)
 	 * Moving/renaming is always done while holding reference.
 	 * kn->parent won't change beneath us.
 	 */
-	parent = kn->parent;
+	parent = kernfs_parent(kn);
 
 	WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
 		  "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
@@ -701,7 +710,7 @@ struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
 			       name, mode, uid, gid, flags);
 	if (kn) {
 		kernfs_get(parent);
-		kn->parent = parent;
+		rcu_assign_pointer(kn->__parent, parent);
 	}
 	return kn;
 }
@@ -769,13 +778,14 @@ err_unlock:
  */
 int kernfs_add_one(struct kernfs_node *kn)
 {
-	struct kernfs_node *parent = kn->parent;
-	struct kernfs_root *root = kernfs_root(parent);
+	struct kernfs_root *root = kernfs_root(kn);
 	struct kernfs_iattrs *ps_iattr;
+	struct kernfs_node *parent;
 	bool has_ns;
 	int ret;
 
 	down_write(&root->kernfs_rwsem);
+	parent = kernfs_parent(kn);
 
 	ret = -EINVAL;
 	has_ns = kernfs_ns_enabled(parent);
@@ -949,6 +959,11 @@ struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
 	return kn;
 }
 
+unsigned int kernfs_root_flags(struct kernfs_node *kn)
+{
+	return kernfs_root(kn)->flags;
+}
+
 /**
  * kernfs_create_root - create a new kernfs hierarchy
  * @scops: optional syscall operations for the hierarchy
@@ -1112,7 +1127,7 @@ struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
 static int kernfs_dop_revalidate(struct inode *dir, const struct qstr *name,
 				 struct dentry *dentry, unsigned int flags)
 {
-	struct kernfs_node *kn;
+	struct kernfs_node *kn, *parent;
 	struct kernfs_root *root;
 
 	if (flags & LOOKUP_RCU)
@@ -1163,8 +1178,9 @@ static int kernfs_dop_revalidate(struct inode *dir, const struct qstr *name,
 	if (!kernfs_active(kn))
 		goto out_bad;
 
+	parent = kernfs_parent(kn);
 	/* The kernfs node has been moved? */
-	if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
+	if (kernfs_dentry_node(dentry->d_parent) != parent)
 		goto out_bad;
 
 	/* The kernfs node has been renamed */
@@ -1172,7 +1188,7 @@ static int kernfs_dop_revalidate(struct inode *dir, const struct qstr *name,
 		goto out_bad;
 
 	/* The kernfs node has been moved to a different namespace */
-	if (kn->parent && kernfs_ns_enabled(kn->parent) &&
+	if (parent && kernfs_ns_enabled(parent) &&
 	    kernfs_info(dentry->d_sb)->ns != kn->ns)
 		goto out_bad;
 
@@ -1365,7 +1381,7 @@ static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
 		return kernfs_leftmost_descendant(rb_to_kn(rbn));
 
 	/* no sibling left, visit parent */
-	return pos->parent;
+	return kernfs_parent(pos);
 }
 
 static void kernfs_activate_one(struct kernfs_node *kn)
@@ -1377,7 +1393,7 @@ static void kernfs_activate_one(struct kernfs_node *kn)
 	if (kernfs_active(kn) || (kn->flags & (KERNFS_HIDDEN | KERNFS_REMOVING)))
 		return;
 
-	WARN_ON_ONCE(kn->parent && RB_EMPTY_NODE(&kn->rb));
+	WARN_ON_ONCE(rcu_access_pointer(kn->__parent) && RB_EMPTY_NODE(&kn->rb));
 	WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
 
 	atomic_sub(KN_DEACTIVATED_BIAS, &kn->active);
@@ -1447,7 +1463,7 @@ void kernfs_show(struct kernfs_node *kn, bool show)
 
 static void __kernfs_remove(struct kernfs_node *kn)
 {
-	struct kernfs_node *pos;
+	struct kernfs_node *pos, *parent;
 
 	/* Short-circuit if non-root @kn has already finished removal. */
 	if (!kn)
@@ -1459,7 +1475,7 @@ static void __kernfs_remove(struct kernfs_node *kn)
 	 * This is for kernfs_remove_self() which plays with active ref
 	 * after removal.
 	 */
-	if (kn->parent && RB_EMPTY_NODE(&kn->rb))
+	if (kernfs_parent(kn) && RB_EMPTY_NODE(&kn->rb))
 		return;
 
 	pr_debug("kernfs %s: removing\n", kn->name);
@@ -1485,14 +1501,14 @@ static void __kernfs_remove(struct kernfs_node *kn)
 		kernfs_get(pos);
 
 		kernfs_drain(pos);
-
+		parent = kernfs_parent(pos);
 		/*
 		 * kernfs_unlink_sibling() succeeds once per node.  Use it
 		 * to decide who's responsible for cleanups.
 		 */
-		if (!pos->parent || kernfs_unlink_sibling(pos)) {
+		if (!parent || kernfs_unlink_sibling(pos)) {
 			struct kernfs_iattrs *ps_iattr =
-				pos->parent ? pos->parent->iattr : NULL;
+				parent ? parent->iattr : NULL;
 
 			/* update timestamps on the parent */
 			down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
@@ -1722,7 +1738,7 @@ int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
 	int error;
 
 	/* can't move or rename root */
-	if (!kn->parent)
+	if (!rcu_access_pointer(kn->__parent))
 		return -EINVAL;
 
 	root = kernfs_root(kn);
@@ -1733,8 +1749,15 @@ int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
 	    (new_parent->flags & KERNFS_EMPTY_DIR))
 		goto out;
 
+	old_parent = kernfs_parent(kn);
+	if (root->flags & KERNFS_ROOT_INVARIANT_PARENT) {
+		error = -EINVAL;
+		if (WARN_ON_ONCE(old_parent != new_parent))
+			goto out;
+	}
+
 	error = 0;
-	if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
+	if ((old_parent == new_parent) && (kn->ns == new_ns) &&
 	    (strcmp(kn->name, new_name) == 0))
 		goto out;	/* nothing to rename */
 
@@ -1761,8 +1784,8 @@ int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
 	/* rename_lock protects ->parent and ->name accessors */
 	write_lock_irq(&kernfs_rename_lock);
 
-	old_parent = kn->parent;
-	kn->parent = new_parent;
+	old_parent = kernfs_parent(kn);
+	rcu_assign_pointer(kn->__parent, new_parent);
 
 	kn->ns = new_ns;
 	if (new_name) {
@@ -1795,7 +1818,8 @@ static struct kernfs_node *kernfs_dir_pos(const void *ns,
 {
 	if (pos) {
 		int valid = kernfs_active(pos) &&
-			pos->parent == parent && hash == pos->hash;
+			rcu_access_pointer(pos->__parent) == parent &&
+			hash == pos->hash;
 		kernfs_put(pos);
 		if (!valid)
 			pos = NULL;
diff --git a/fs/kernfs/kernfs-internal.h b/fs/kernfs/kernfs-internal.h
index b42ee6547cdc..c43bee18b79f 100644
--- a/fs/kernfs/kernfs-internal.h
+++ b/fs/kernfs/kernfs-internal.h
@@ -19,6 +19,8 @@
 #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;
@@ -64,11 +66,14 @@ struct kernfs_root {
  *
  * Return: the kernfs_root @kn belongs to.
  */
-static inline struct kernfs_root *kernfs_root(struct kernfs_node *kn)
+static inline struct kernfs_root *kernfs_root(const struct kernfs_node *kn)
 {
+	const struct kernfs_node *knp;
 	/* if parent exists, it's always a dir; otherwise, @sd is a dir */
-	if (kn->parent)
-		kn = kn->parent;
+	guard(rcu)();
+	knp = rcu_dereference(kn->__parent);
+	if (knp)
+		kn = knp;
 	return kn->dir.root;
 }
 
@@ -97,6 +102,27 @@ struct kernfs_super_info {
 };
 #define kernfs_info(SB) ((struct kernfs_super_info *)(SB->s_fs_info))
 
+static inline bool kernfs_root_is_locked(const struct kernfs_node *kn)
+{
+	return lockdep_is_held(&kernfs_root(kn)->kernfs_rwsem);
+}
+
+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.
+	 * 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) ||
+				     !atomic_read(&kn->count));
+}
+
 static inline struct kernfs_node *kernfs_dentry_node(struct dentry *dentry)
 {
 	if (d_really_is_negative(dentry))
diff --git a/fs/kernfs/mount.c b/fs/kernfs/mount.c
index 4a0ff08d589c..2252b16e6ef0 100644
--- a/fs/kernfs/mount.c
+++ b/fs/kernfs/mount.c
@@ -148,7 +148,7 @@ static struct dentry *kernfs_get_parent_dentry(struct dentry *child)
 	struct kernfs_root *root = kernfs_root(kn);
 
 	guard(rwsem_read)(&root->kernfs_rwsem);
-	return d_obtain_alias(kernfs_get_inode(child->d_sb, kn->parent));
+	return d_obtain_alias(kernfs_get_inode(child->d_sb, kernfs_parent(kn)));
 }
 
 static const struct export_operations kernfs_export_ops = {
@@ -188,10 +188,10 @@ static struct kernfs_node *find_next_ancestor(struct kernfs_node *child,
 		return NULL;
 	}
 
-	while (child->parent != parent) {
-		if (!child->parent)
+	while (kernfs_parent(child) != parent) {
+		child = kernfs_parent(child);
+		if (!child)
 			return NULL;
-		child = child->parent;
 	}
 
 	return child;
@@ -216,7 +216,7 @@ struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
 	dentry = dget(sb->s_root);
 
 	/* Check if this is the root kernfs_node */
-	if (!kn->parent)
+	if (!rcu_access_pointer(kn->__parent))
 		return dentry;
 
 	root = kernfs_root(kn);
diff --git a/fs/kernfs/symlink.c b/fs/kernfs/symlink.c
index 45371a70caa7..05c62ca93c53 100644
--- a/fs/kernfs/symlink.c
+++ b/fs/kernfs/symlink.c
@@ -62,10 +62,10 @@ static int kernfs_get_target_path(struct kernfs_node *parent,
 
 	/* go up to the root, stop at the base */
 	base = parent;
-	while (base->parent) {
-		kn = target->parent;
-		while (kn->parent && base != kn)
-			kn = kn->parent;
+	while (kernfs_parent(base)) {
+		kn = kernfs_parent(target);
+		while (kernfs_parent(kn) && base != kn)
+			kn = kernfs_parent(kn);
 
 		if (base == kn)
 			break;
@@ -75,14 +75,14 @@ static int kernfs_get_target_path(struct kernfs_node *parent,
 
 		strcpy(s, "../");
 		s += 3;
-		base = base->parent;
+		base = kernfs_parent(base);
 	}
 
 	/* determine end of target string for reverse fillup */
 	kn = target;
-	while (kn->parent && kn != base) {
+	while (kernfs_parent(kn) && kn != base) {
 		len += strlen(kn->name) + 1;
-		kn = kn->parent;
+		kn = kernfs_parent(kn);
 	}
 
 	/* check limits */
@@ -94,7 +94,7 @@ static int kernfs_get_target_path(struct kernfs_node *parent,
 
 	/* reverse fillup of target string from target to base */
 	kn = target;
-	while (kn->parent && kn != base) {
+	while (kernfs_parent(kn) && kn != base) {
 		int slen = strlen(kn->name);
 
 		len -= slen;
@@ -102,7 +102,7 @@ static int kernfs_get_target_path(struct kernfs_node *parent,
 		if (len)
 			s[--len] = '/';
 
-		kn = kn->parent;
+		kn = kernfs_parent(kn);
 	}
 
 	return 0;
@@ -111,12 +111,13 @@ static int kernfs_get_target_path(struct kernfs_node *parent,
 static int kernfs_getlink(struct inode *inode, char *path)
 {
 	struct kernfs_node *kn = inode->i_private;
-	struct kernfs_node *parent = kn->parent;
+	struct kernfs_node *parent;
 	struct kernfs_node *target = kn->symlink.target_kn;
-	struct kernfs_root *root = kernfs_root(parent);
+	struct kernfs_root *root = kernfs_root(kn);
 	int error;
 
 	down_read(&root->kernfs_rwsem);
+	parent = kernfs_parent(kn);
 	error = kernfs_get_target_path(parent, target, path);
 	up_read(&root->kernfs_rwsem);