Re: [PATCH] of: overlay: fix device_node leak when attaching node to changeset fails

From: Rob Herring (Arm)

Date: Thu Nov 19 2026 - 15:12:24 EST



On Sun, 25 Oct 2026 20:02:37 +0200, Selby Mashiki wrote:
> In add_changeset_node(), when creating a new target child node via
> __of_node_dup(), tchild is initialized with a reference count of 1. If
> of_changeset_attach_node() subsequently fails, add_changeset_node()
> returns the error without calling of_node_put(tchild), leaking the
> allocated device_node.
>
> Furthermore, because OF_OVERLAY is set on tchild before
> of_changeset_attach_node() is called, calling of_node_put(tchild) when
> tchild was never added to the changeset (and thus never had
> OF_OVERLAY_FREE_CSET set by __of_changeset_entry_destroy()) would cause
> of_node_release() to log a premature refcount zero error and skip
> freeing the node.
>
> Move of_node_set_flag(tchild, OF_OVERLAY) after
> of_changeset_attach_node() succeeds and call of_node_put(tchild) on the
> failure path, matching of_changeset_create_node().
>
> Fixes: 7c528e457d53 ("of: overlay: add missing of_node_put() after add new node to changeset")
> Fixes: 144552c78692 ("of: overlay: add tests to validate kfrees from overlay removal")

These don't look right to me, so I dropped them.

> Signed-off-by: Selby Mashiki <mashikiselby@xxxxxxxxx>
> ---
> drivers/of/overlay.c | 8 +++++---
> 1 file changed, 5 insertions(+), 3 deletions(-)
>

Applied, thanks!