Re: [PATCH] net: Fix error handling in netdev_register_kobject

[Eric Dumazet]

On Fri, Nov 7, 2025 at 12:01 AM Ma Ke <make24@xxxxxxxxxxx> wrote:
>
> After calling device_initialize(), the reference count of the device
> is set to 1. If device_add() fails or register_queue_kobjects() fails,
> the function returns without calling put_device() to release the
> initial reference, causing a memory leak of the device structure.
> Similarly, in netdev_unregister_kobject(), after calling device_del(),
> there is no call to put_device() to release the initial reference,
> leading to a memory leak. Add put_device() in the error paths of
> netdev_register_kobject() and after device_del() in
> netdev_unregister_kobject() to properly release the device references.
>
> Found by code review.
>
> Cc: stable@xxxxxxxxxxxxxxx
> Fixes: a1b3f594dc5f ("net: Expose all network devices in a namespaces in sysfs")
> Signed-off-by: Ma Ke <make24@xxxxxxxxxxx>
> ---
>  net/core/net-sysfs.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/net/core/net-sysfs.c b/net/core/net-sysfs.c
> index ca878525ad7c..d3895f26a0c8 100644
> --- a/net/core/net-sysfs.c
> +++ b/net/core/net-sysfs.c
> @@ -2327,6 +2327,7 @@ void netdev_unregister_kobject(struct net_device *ndev)
>         pm_runtime_set_memalloc_noio(dev, false);
>
>         device_del(dev);
> +       put_device(dev);

Please take a look at free_netdev()

>  }
>
>  /* Create sysfs entries for network device. */
> @@ -2357,7 +2358,7 @@ int netdev_register_kobject(struct net_device *ndev)
>
>         error = device_add(dev);
>         if (error)
> -               return error;
> +               goto out_put_device;
>
>         error = register_queue_kobjects(ndev);
>         if (error) {
> @@ -2367,6 +2368,10 @@ int netdev_register_kobject(struct net_device *ndev)
>
>         pm_runtime_set_memalloc_noio(dev, true);
>
> +       return 0;
> +
> +out_put_device:
> +       put_device(dev);
>         return error;

This seems bogus.

Was your report based on AI or some tooling ?

You would think that syzbot would have found an issue a long time ago...