Re: [PATCH wpan] mac802154: llsec: reject secured frames with security level 0

From: Miquel Raynal

Date: Tue Nov 17 2026 - 04:35:29 EST


On 10/11/2026 at 09:03:23 GMT, Joas Antonio dos Santos via B4 Relay <devnull+joasantonio108.gmail.com@xxxxxxxxxx> wrote:

> From: Joas Antonio dos Santos <joasantonio108@xxxxxxxxx>
>
> mac802154_llsec_decrypt() handles every received frame that has the
> Security Enabled bit set. The security level is taken from the
> auxiliary security header of the frame and is not required to be
> non-zero. The only check that looks at level 0 rejects the frame only
> if level 0 is not allowed by the matching seclevel entry, the entry has
> device_override set and the device is not seclevel_exempt:
>
> if (!(seclevel.sec_levels & BIT(hdr.sec.level)) &&
> (hdr.sec.level == 0 && seclevel.device_override &&
> !dev->dev.seclevel_exempt)) {
>
> Otherwise the frame reaches llsec_do_decrypt(), which sends every level
> other than IEEE802154_SCF_SECLEVEL_ENC to llsec_do_decrypt_auth().
> There ieee802154_sechdr_authtag_len() returns 0 for level 0 and
>
> req = aead_request_alloc(llsec_tfm_by_len(key, authlen), GFP_ATOMIC);
>
> looks for a transform with an authentication tag size of 0. The key
> only has transforms for tag sizes 4, 8 and 16, so llsec_tfm_by_len()
> runs into its BUG().
>
> The frame is not authenticated at that point. A station in radio range
> can crash a node that has llsec enabled if it sends a frame that matches
> a configured key entry and a device table entry (source address, key id
> mode and key id are sent in the clear and can be observed), uses a frame
> type that has a seclevel entry and carries a frame counter that is not
> lower than the stored one. No key material is needed. The frame
> counter of that device is also updated before the BUG() is reached.
>
> A frame with the Security Enabled bit set and security level 0 carries
> no protection and cannot be processed by the incoming security
> procedure. Reject it before any key or device state is looked up. The
> transmit side already leaves such frames alone in
> mac802154_llsec_encrypt().

Reviewed-by: Miquel Raynal <miquel.raynal@xxxxxxxxxxx>