[net] mac80211: fix potential double free on mesh join

Commit Message

Linus Lüssing March 10, 2022, 6:35 p.m. UTC

From: Linus Lüssing <[e-mail]>

While commit 6a01afcf8468 ("mac80211: mesh: Free ie data when leaving
mesh") fixed a memory leak on mesh leave / teardown it introduced a
potential memory corruption caused by a double free when rejoining the
mesh:

  ieee80211_leave_mesh()
  -> kfree(sdata->u.mesh.ie);
  ...
  ieee80211_join_mesh()
  -> copy_mesh_setup()
     -> old_ie = ifmsh->ie;
     -> kfree(old_ie);

This double free / kernel panics can be reproduced by using wpa_supplicant
with an encrypted mesh (if set up without encryption via "iw" then
ifmsh->ie is always NULL, which avoids this issue). And then calling:

  $ iw dev mesh0 mesh leave
  $ iw dev mesh0 mesh join my-mesh

Note that typically these commands are not used / working when using
wpa_supplicant. And it seems that wpa_supplicant or wpa_cli are going
through a NETDEV_DOWN/NETDEV_UP cycle between a mesh leave and mesh join
where the NETDEV_UP resets the mesh.ie to NULL via a memcpy of
default_mesh_setup in cfg80211_netdev_notifier_call, which then avoids
the memory corruption, too.

The issue was first observed in an application which was not using
wpa_supplicant but "Senf" instead, which implements its own calls to
nl80211.

Fixing the issue by removing the kfree()'ing of the mesh IE in the mesh
join function and leaving it solely up to the mesh leave to free the
mesh IE.

Link: https://gitlab.fit.fraunhofer.de/wiback/senf
Fixes: 6a01afcf8468 ("mac80211: mesh: Free ie data when leaving mesh")
Reported-by: Matthias Kretschmer <[e-mail]>
Signed-off-by: Linus Lüssing <[e-mail]>
---
 net/mac80211/cfg.c | 3 ---
 1 file changed, 3 deletions(-)

Comments

Kretschmer, Mathias March 10, 2022, 7:02 p.m. UTC | #1

Yep, that's correct. We do usually not go through a NETDEV_DOWN/NETDEV_UP cycle,
when, for example, reassigning channels.

Will test your patch and report back.

BR,

Mathias

On Thu, 2022-03-10 at 19:35 +0100, Linus Lüssing wrote:
> From: Linus Lüssing <[e-mail]>
> 
> While commit 6a01afcf8468 ("mac80211: mesh: Free ie data when leaving
> mesh") fixed a memory leak on mesh leave / teardown it introduced a
> potential memory corruption caused by a double free when rejoining the
> mesh:
> 
>   ieee80211_leave_mesh()
>   -> kfree(sdata->u.mesh.ie);
>   ...
>   ieee80211_join_mesh()
>   -> copy_mesh_setup()
>      -> old_ie = ifmsh->ie;
>      -> kfree(old_ie);
> 
> This double free / kernel panics can be reproduced by using wpa_supplicant
> with an encrypted mesh (if set up without encryption via "iw" then
> ifmsh->ie is always NULL, which avoids this issue). And then calling:
> 
>   $ iw dev mesh0 mesh leave
>   $ iw dev mesh0 mesh join my-mesh
> 
> Note that typically these commands are not used / working when using
> wpa_supplicant. And it seems that wpa_supplicant or wpa_cli are going
> through a NETDEV_DOWN/NETDEV_UP cycle between a mesh leave and mesh join
> where the NETDEV_UP resets the mesh.ie to NULL via a memcpy of
> default_mesh_setup in cfg80211_netdev_notifier_call, which then avoids
> the memory corruption, too.
> 
> The issue was first observed in an application which was not using
> wpa_supplicant but "Senf" instead, which implements its own calls to
> nl80211.
> 
> Fixing the issue by removing the kfree()'ing of the mesh IE in the mesh
> join function and leaving it solely up to the mesh leave to free the
> mesh IE.
> 
> Link: https://gitlab.fit.fraunhofer.de/wiback/senf
> Fixes: 6a01afcf8468 ("mac80211: mesh: Free ie data when leaving mesh")
> Reported-by: Matthias Kretschmer <[e-mail]>
> Signed-off-by: Linus Lüssing <[e-mail]>
> ---
>  net/mac80211/cfg.c | 3 ---
>  1 file changed, 3 deletions(-)
> 
> diff --git a/net/mac80211/cfg.c b/net/mac80211/cfg.c
> index 87a208089caf..58ff57dc669c 100644
> --- a/net/mac80211/cfg.c
> +++ b/net/mac80211/cfg.c
> @@ -2148,14 +2148,12 @@ static int copy_mesh_setup(struct ieee80211_if_mesh
> *ifmsh,
>                 const struct mesh_setup *setup)
>  {
>         u8 *new_ie;
> -       const u8 *old_ie;
>         struct ieee80211_sub_if_data *sdata = container_of(ifmsh,
>                                         struct ieee80211_sub_if_data, u.mesh);
>         int i;
>  
>         /* allocate information elements */
>         new_ie = NULL;
> -       old_ie = ifmsh->ie;
>  
>         if (setup->ie_len) {
>                 new_ie = kmemdup(setup->ie, setup->ie_len,
> @@ -2165,7 +2163,6 @@ static int copy_mesh_setup(struct ieee80211_if_mesh
> *ifmsh,
>         }
>         ifmsh->ie_len = setup->ie_len;
>         ifmsh->ie = new_ie;
> -       kfree(old_ie);
>  
>         /* now copy the rest of the setup parameters */
>         ifmsh->mesh_id_len = setup->mesh_id_len;

Kretschmer, Mathias March 11, 2022, 7:31 a.m. UTC | #2

Tested-by: Mathias Kretschmer <[e-mail]>

Patch

diff --git a/net/mac80211/cfg.c b/net/mac80211/cfg.c
index 87a208089caf..58ff57dc669c 100644
--- a/net/mac80211/cfg.c
+++ b/net/mac80211/cfg.c
@@ -2148,14 +2148,12 @@  static int copy_mesh_setup(struct ieee80211_if_mesh *ifmsh,
 		const struct mesh_setup *setup)
 {
 	u8 *new_ie;
-	const u8 *old_ie;
 	struct ieee80211_sub_if_data *sdata = container_of(ifmsh,
 					struct ieee80211_sub_if_data, u.mesh);
 	int i;
 
 	/* allocate information elements */
 	new_ie = NULL;
-	old_ie = ifmsh->ie;
 
 	if (setup->ie_len) {
 		new_ie = kmemdup(setup->ie, setup->ie_len,
@@ -2165,7 +2163,6 @@  static int copy_mesh_setup(struct ieee80211_if_mesh *ifmsh,
 	}
 	ifmsh->ie_len = setup->ie_len;
 	ifmsh->ie = new_ie;
-	kfree(old_ie);
 
 	/* now copy the rest of the setup parameters */
 	ifmsh->mesh_id_len = setup->mesh_id_len;