On Fri, Jul 1, 2022 at 4:55 PM Siddh Raman Pant <[email protected]> wrote:
>
> ieee80211_scan_rx() tries to access scan_req->flags after a null check
> (see line 303 of mac80211/scan.c), but ___cfg80211_scan_done() uses
> kfree() on the scan_req (see line 991 of wireless/scan.c).
>
> This results in a UAF.
>
> ieee80211_scan_rx() is called inside a RCU read-critical section
> initiated by ieee80211_rx_napi() (see line 5043 of mac80211/rx.c).
>
> Thus, add an rcu_head to the scan_req struct so as to use kfree_rcu()
> instead of kfree() so that we don't free during the critical section.
>
> Bug report (3): https://syzkaller.appspot.com/bug?extid=f9acff9bf08a845f225d
> Reported-by: [email protected]
> Reported-by: [email protected]
> Reported-by: [email protected]
>
> Signed-off-by: Siddh Raman Pant <[email protected]>
> ---
> include/net/cfg80211.h | 2 ++
> net/wireless/scan.c | 2 +-
> 2 files changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/include/net/cfg80211.h b/include/net/cfg80211.h
> index 6d02e12e4702..ba4a49884de8 100644
> --- a/include/net/cfg80211.h
> +++ b/include/net/cfg80211.h
> @@ -2368,6 +2368,7 @@ struct cfg80211_scan_6ghz_params {
> * @n_6ghz_params: number of 6 GHz params
> * @scan_6ghz_params: 6 GHz params
> * @bssid: BSSID to scan for (most commonly, the wildcard BSSID)
> + * @rcu_head: (internal) RCU head to use for freeing
> */
> struct cfg80211_scan_request {
> struct cfg80211_ssid *ssids;
> @@ -2397,6 +2398,7 @@ struct cfg80211_scan_request {
> bool scan_6ghz;
> u32 n_6ghz_params;
> struct cfg80211_scan_6ghz_params *scan_6ghz_params;
> + struct rcu_head rcu_head;
>
> /* keep last */
> struct ieee80211_channel *channels[];
> diff --git a/net/wireless/scan.c b/net/wireless/scan.c
> index 6d82bd9eaf8c..638b2805222c 100644
> --- a/net/wireless/scan.c
> +++ b/net/wireless/scan.c
> @@ -988,7 +988,7 @@ void ___cfg80211_scan_done(struct cfg80211_registered_device *rdev,
> kfree(rdev->int_scan_req);
> rdev->int_scan_req = NULL;
>
> - kfree(rdev->scan_req);
> + kfree_rcu(rdev->scan_req, rcu_head);
> rdev->scan_req = NULL;
Note: this is slightly racy.
You are supposed to follow this order in this situation.
1) Clear the pointer
Then:
2) wait an rcu grace period (synchronize_rcu()) or use call_rcu()/kfree_rcu().
>
> if (!send_message)
> --
> 2.35.1
>
>
On Tue, 26 Jul 2022 15:37:16 +0530 Eric Dumazet <[email protected]> wrote:
> Note: this is slightly racy.
>
> You are supposed to follow this order in this situation.
>
> 1) Clear the pointer
>
> Then:
>
> 2) wait an rcu grace period (synchronize_rcu()) or use call_rcu()/kfree_rcu().
>
Noted. Due to rcu_dereference() used to get scan_req, null ptr dereference
cannot happen. That had completely missed my eyes, sorry for that.
I will send a v2.
Thanks,
Siddh