Re: [PATCH RFC] netdevsim: remove debugfs files before freeing net_device

On Tue, Jun 23, 2026 at 4:31 PM 'syzbot' via
syzkaller-upstream-moderation
<[email protected]> wrote:
>
> A KASAN slab-use-after-free was detected in debugfs_u32_get() when reading
> a debugfs file associated with a netdevsim port.
>
> BUG: KASAN: slab-use-after-free in debugfs_u32_get+0x6c/0x70
> Read of size 4 at addr ffff88810cd6db00
>
> Call Trace:
>  debugfs_u32_get+0x6c/0x70
>  simple_attr_read+0x227/0x490
>  debugfs_attr_read+0x76/0x130
>  vfs_read+0x213/0xa80
>
> The root cause is that the net_device (and its embedded netdevsim private
> data) is freed before its associated debugfs files are removed. When a user
> reads a debugfs file like rx_max_pending, debugfs_u32_get() dereferences a
> pointer to the freed memory. This happens because nsim_destroy() calls
> free_netdev() but does not remove the debugfs files created by
> nsim_ethtool_init() or nsim_bpf_init(). These files are only removed later
> when nsim_dev_port_debugfs_exit() is called. A similar issue exists in the
> error path of nsim_create().
>
> To fix this, swap the teardown order in __nsim_dev_port_del() and the
> __nsim_dev_port_add() error path so that nsim_dev_port_debugfs_exit() is
> called before nsim_destroy(). This ensures the debugfs directory is
> recursively removed before the net_device is freed. Additionally, remove
> explicit debugfs_remove() calls from various subsystem teardown functions
> (like nsim_psp_uninit, nsim_ipsec_teardown, etc.) since the parent
> directory is now removed earlier, which would otherwise lead to a
> use-after-free of the dentry itself. Finally, ensure the debugfs directory
> is removed before free_netdev() is called in the nsim_create() error path.
>
> Fixes: e05b2d141fef ("netdevsim: move netdev creation/destruction to dev probe")
> Assisted-by: Gemini:gemini-3.1-pro-preview Gemini:gemini-3-flash-preview syzbot
> Reported-by: [email protected]
> Closes: https://syzkaller.appspot.com/bug?extid=6c25f4750230faf70be9
> Link: https://syzkaller.appspot.com/ai_job?id=84a7c962-d10e-424d-a37a-bf27d4830a40
> To: "Andrew Lunn" <[email protected]>
> To: "David S. Miller" <[email protected]>
> To: "Eric Dumazet" <[email protected]>
> To: "Jakub Kicinski" <[email protected]>
> To: <[email protected]>
> To: "Paolo Abeni" <[email protected]>
> Cc: <[email protected]>
>
> ---
> diff --git a/drivers/net/netdevsim/dev.c b/drivers/net/netdevsim/dev.c
> index f00fc2f9e..beecc0c6e 100644
> --- a/drivers/net/netdevsim/dev.c
> +++ b/drivers/net/netdevsim/dev.c
> @@ -1524,7 +1524,9 @@ static int __nsim_dev_port_add(struct nsim_dev *nsim_dev, enum nsim_dev_port_typ
>         return 0;
>
>  err_nsim_destroy:
> +       nsim_dev_port_debugfs_exit(nsim_dev_port);
>         nsim_destroy(nsim_dev_port->ns);
> +       goto err_port_resource_unregister;

goto to goto looks very weird..
Maybe we can simplify it with an `if (nsim_dev_port->ns != NULL)`
check? Then `err_nsim_destroy` could be dropped.

>  err_port_debugfs_exit:
>         nsim_dev_port_debugfs_exit(nsim_dev_port);
>  err_port_resource_unregister:
> @@ -1544,8 +1546,8 @@ static void __nsim_dev_port_del(struct nsim_dev_port *nsim_dev_port)
>         list_del(&nsim_dev_port->list);
>         if (nsim_dev_port_is_vf(nsim_dev_port))
>                 devl_rate_leaf_destroy(&nsim_dev_port->devlink_port);
> -       nsim_destroy(nsim_dev_port->ns);
>         nsim_dev_port_debugfs_exit(nsim_dev_port);
> +       nsim_destroy(nsim_dev_port->ns);
>         if (nsim_dev_port_is_pf(nsim_dev_port))
>                 devl_port_resources_unregister(devlink_port);
>         devl_port_unregister(devlink_port);
> diff --git a/drivers/net/netdevsim/ipsec.c b/drivers/net/netdevsim/ipsec.c
> index 36a1be492..0bc9f42a5 100644
> --- a/drivers/net/netdevsim/ipsec.c
> +++ b/drivers/net/netdevsim/ipsec.c
> @@ -292,5 +292,4 @@ void nsim_ipsec_teardown(struct netdevsim *ns)
>         if (ipsec->count)
>                 netdev_err(ns->netdev, "tearing down IPsec offload with %d SAs left\n",
>                            ipsec->count);
> -       debugfs_remove_recursive(ipsec->pfile);
>  }
> diff --git a/drivers/net/netdevsim/netdev.c b/drivers/net/netdevsim/netdev.c
> index a75076891..6918679ca 100644
> --- a/drivers/net/netdevsim/netdev.c
> +++ b/drivers/net/netdevsim/netdev.c
> @@ -1164,6 +1164,8 @@ struct netdevsim *nsim_create(struct nsim_dev *nsim_dev,
>         return ns;
>
>  err_free_netdev:
> +       debugfs_remove_recursive(nsim_dev_port->ddir);
> +       nsim_dev_port->ddir = NULL;
>         free_netdev(dev);
>         return ERR_PTR(err);
>  }
> @@ -1174,10 +1176,6 @@ void nsim_destroy(struct netdevsim *ns)
>         struct netdevsim *peer;
>         u16 vid;
>
> -       debugfs_remove(ns->vlan_dfs);
> -       debugfs_remove(ns->qr_dfs);
> -       debugfs_remove(ns->pp_dfs);

Now that vlan_dfs, qr_dfs, pp_dfs are not used for debugfs_remove, do
we actually still need those fields?

> -
>         if (ns->nb.notifier_call)
>                 unregister_netdevice_notifier_dev_net(ns->netdev, &ns->nb,
>                                                       &ns->nn);
> diff --git a/drivers/net/netdevsim/psp.c b/drivers/net/netdevsim/psp.c
> index 6936ecb81..2f27a2114 100644
> --- a/drivers/net/netdevsim/psp.c
> +++ b/drivers/net/netdevsim/psp.c
> @@ -227,7 +227,6 @@ static void __nsim_psp_uninit(struct netdevsim *ns, bool teardown)
>
>  void nsim_psp_uninit(struct netdevsim *ns)
>  {
> -       debugfs_remove(ns->psp.rereg);
>         mutex_destroy(&ns->psp.rereg_lock);
>         __nsim_psp_uninit(ns, true);
>  }
> diff --git a/drivers/net/netdevsim/udp_tunnels.c b/drivers/net/netdevsim/udp_tunnels.c
> index 89fff76e5..a69b15f12 100644
> --- a/drivers/net/netdevsim/udp_tunnels.c
> +++ b/drivers/net/netdevsim/udp_tunnels.c
> @@ -188,9 +188,6 @@ int nsim_udp_tunnels_info_create(struct nsim_dev *nsim_dev,
>
>  void nsim_udp_tunnels_info_destroy(struct net_device *dev)
>  {
> -       struct netdevsim *ns = netdev_priv(dev);
> -
> -       debugfs_remove_recursive(ns->udp_ports.ddir);
>         kfree(dev->udp_tunnel_nic_info);
>         dev->udp_tunnel_nic_info = NULL;
>  }
>
>
> base-commit: 8cd9520d35a6c38db6567e97dd93b1f11f185dc6
> --
> This is an AI-generated patch subject to moderation.
> Reply with '#syz upstream' to Sign-off the patch as a human author
> and send it to the upstream kernel mailing lists.
> Reply with '#syz reject' to reject it ('#syz unreject' to undo).
>
> See https://goo.gle/syzbot-ai-patches for information about AI-generated patches.
> You can comment on the patch as usual, syzbot will try to address
> the comments and send a new version of the patch if necessary.
> syzbot engineers can be reached at [email protected].
>