[v1] wifi: mac80211: guard drv_net_setup_tc() against unbound AP_VLAN sdata

[PATCH] wifi: mac80211: guard drv_net_setup_tc() against unbound AP_VLAN sdata

Posted by Ayushman Rout 1 month, 2 weeks ago

#syz test: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master

syzbot reports a NULL/invalid pointer dereference in
trace_event_raw_event_drv_net_setup_tc(), reached via
ieee80211_netdev_setup_tc() -> drv_net_setup_tc().

drv_net_setup_tc() calls get_bss_sdata(sdata) unconditionally. For an
NL80211_IFTYPE_AP_VLAN interface this does
container_of(sdata->bss, ...), but sdata->bss is only linked
opportunistically at interface-add time when a matching same-address
AP interface exists - it is not enforced, so an AP_VLAN interface can
be fully created and registered with sdata->bss left NULL.
container_of() on NULL yields a small invalid pointer rather than
NULL, which the trace_drv_net_setup_tc tracepoint then dereferences
to read the interface name.

Guard against an unbound AP_VLAN sdata before calling
get_bss_sdata(), matching the WARN_ON_ONCE(!bss) precondition already
used for this same relationship in sta_info.c. Also add
check_sdata_in_driver(), used by the neighboring
drv_net_fill_forward_path() but missing here.

The underlying gap in ieee80211_if_add() - AP_VLAN creation not
requiring a bound bss - is not fixed here; other get_bss_sdata()
callers may share the exposure.

Fixes: 61587f1556fe ("wifi: mac80211: add support for letting drivers register tc offload support")
Reported-by: syzbot+f1ba58d6b55abd13239e@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=f1ba58d6b55abd13239e
Signed-off-by: Ayushman Rout <ayushmanrout27@gmail.com>
---
 net/mac80211/driver-ops.h | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/net/mac80211/driver-ops.h b/net/mac80211/driver-ops.h
index f1c0b87fddd5..ecfdb51152f4 100644
--- a/net/mac80211/driver-ops.h
+++ b/net/mac80211/driver-ops.h
@@ -1702,7 +1702,23 @@ static inline int drv_net_setup_tc(struct ieee80211_local *local,
 
 	might_sleep();
 
+	/*
+	 * An AP_VLAN interface created without a matching, same-address
+	 * AP interface present never gets sdata->bss populated (see the
+	 * interface-add validation in iface.c, which links bss only
+	 * opportunistically and does not require it). Such an sdata is
+	 * not safe to pass through get_bss_sdata(): container_of() on a
+	 * NULL sdata->bss yields a small invalid pointer, which the
+	 * tracepoint below then dereferences to read the interface name,
+	 * causing a crash.
+	 */
+	if (sdata->vif.type == NL80211_IFTYPE_AP_VLAN && !sdata->bss)
+		return -EIO;
+
 	sdata = get_bss_sdata(sdata);
+	if (!check_sdata_in_driver(sdata))
+		return -EIO;
+
 	trace_drv_net_setup_tc(local, sdata, type);
 	if (local->ops->net_setup_tc)
 		ret = local->ops->net_setup_tc(&local->hw, &sdata->vif, dev,
-- 
2.54.0

Re: [PATCH] wifi: mac80211: guard drv_net_setup_tc() against unbound AP_VLAN sdata

Posted by syzbot 1 month, 2 weeks ago

> #syz test: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master

This crash does not have a reproducer. I cannot test it.

>
> syzbot reports a NULL/invalid pointer dereference in
> trace_event_raw_event_drv_net_setup_tc(), reached via
> ieee80211_netdev_setup_tc() -> drv_net_setup_tc().
>
> drv_net_setup_tc() calls get_bss_sdata(sdata) unconditionally. For an
> NL80211_IFTYPE_AP_VLAN interface this does
> container_of(sdata->bss, ...), but sdata->bss is only linked
> opportunistically at interface-add time when a matching same-address
> AP interface exists - it is not enforced, so an AP_VLAN interface can
> be fully created and registered with sdata->bss left NULL.
> container_of() on NULL yields a small invalid pointer rather than
> NULL, which the trace_drv_net_setup_tc tracepoint then dereferences
> to read the interface name.
>
> Guard against an unbound AP_VLAN sdata before calling
> get_bss_sdata(), matching the WARN_ON_ONCE(!bss) precondition already
> used for this same relationship in sta_info.c. Also add
> check_sdata_in_driver(), used by the neighboring
> drv_net_fill_forward_path() but missing here.
>
> The underlying gap in ieee80211_if_add() - AP_VLAN creation not
> requiring a bound bss - is not fixed here; other get_bss_sdata()
> callers may share the exposure.
>
> Fixes: 61587f1556fe ("wifi: mac80211: add support for letting drivers register tc offload support")
> Reported-by: syzbot+f1ba58d6b55abd13239e@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=f1ba58d6b55abd13239e
> Signed-off-by: Ayushman Rout <ayushmanrout27@gmail.com>
> ---
>  net/mac80211/driver-ops.h | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
>
> diff --git a/net/mac80211/driver-ops.h b/net/mac80211/driver-ops.h
> index f1c0b87fddd5..ecfdb51152f4 100644
> --- a/net/mac80211/driver-ops.h
> +++ b/net/mac80211/driver-ops.h
> @@ -1702,7 +1702,23 @@ static inline int drv_net_setup_tc(struct ieee80211_local *local,
>  
>  	might_sleep();
>  
> +	/*
> +	 * An AP_VLAN interface created without a matching, same-address
> +	 * AP interface present never gets sdata->bss populated (see the
> +	 * interface-add validation in iface.c, which links bss only
> +	 * opportunistically and does not require it). Such an sdata is
> +	 * not safe to pass through get_bss_sdata(): container_of() on a
> +	 * NULL sdata->bss yields a small invalid pointer, which the
> +	 * tracepoint below then dereferences to read the interface name,
> +	 * causing a crash.
> +	 */
> +	if (sdata->vif.type == NL80211_IFTYPE_AP_VLAN && !sdata->bss)
> +		return -EIO;
> +
>  	sdata = get_bss_sdata(sdata);
> +	if (!check_sdata_in_driver(sdata))
> +		return -EIO;
> +
>  	trace_drv_net_setup_tc(local, sdata, type);
>  	if (local->ops->net_setup_tc)
>  		ret = local->ops->net_setup_tc(&local->hw, &sdata->vif, dev,
> -- 
> 2.54.0
>