Re: [PATCH] net: mac80211: fix hard-coding of length check on skb->len in ieee80211_scan_rx()

From: Du Cheng <ducheng2@gmail.com>
To: Shuah Khan <skhan@linuxfoundation.org>
Cc: Thomas Pedersen <thomas@adapt-ip.com>,
	Johannes Berg <johannes@sipsolutions.net>,
	linux-wireless@vger.kernel.org,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	syzbot+405843667e93b9790fc1@syzkaller.appspotmail.com
Subject: Re: [PATCH] net: mac80211: fix hard-coding of length check on skb->len in ieee80211_scan_rx()
Date: Wed, 12 May 2021 08:55:10 +0800	[thread overview]
Message-ID: <YJsnbhkOLSPqFMLM@nuc> (raw)
In-Reply-To: <e52ff0a3-7b64-037d-7825-9cbbcafa21b6@linuxfoundation.org>

Le Tue, May 11, 2021 at 04:29:07PM -0600, Shuah Khan a écrit :
> On 5/9/21 10:16 PM, Du Cheng wrote:
> > Replace hard-coding with compile-time constants for header length
> > check on skb->len. This skb->len will be checked again further down the
> > callstack in cfg80211_inform_bss_frame_data() in net/wireless/scan.c
> > (which has a proper length check with WARN_ON()). If the kernel is
> > configure to panic_on_warn(), the insuffient check of skb->len in
> > ieee80211_scan_rx() causes kernel crash in
> > cfg80211_inform_bss_frame_data().
> > 
> 
> Where is this WARN_ON? I didn't see it cfg80211_inform_bss_frame_data()
> 
> Please add more information on why the values of min_hdr_len in this
> patch make sense for each of these cases.

Hi Shuah,

The WARN_ON() is located here:
linux/net/wireless/scan.c: 2331
```
    if (ieee80211_is_s1g_beacon(mgmt->frame_control)) {
        ext = (void *) mgmt;
        min_hdr_len = offsetof(struct ieee80211_ext, u.s1g_beacon);
        if (ieee80211_is_s1g_short_beacon(mgmt->frame_control))
            min_hdr_len = offsetof(struct ieee80211_ext,
                           u.s1g_short_beacon.variable);
    }

    if (WARN_ON(len < min_hdr_len)) // <- warn_on line
        return NULL;

```
the min_hdr_len that I added in was following the setup of min_hdr_len before that
WARN_ON(len < min_hdr_len)

> 
> > Bug reported by syzbot:
> > https://syzkaller.appspot.com/bug?id=183869c2f25b1c8692e381d8fcd69771a99221cc
> > 
> > Reported-by: syzbot+405843667e93b9790fc1@syzkaller.appspotmail.com
> > Signed-off-by: Du Cheng <ducheng2@gmail.com>
> > ---
> > 
> > This patch has passed syzbot testing.
> > 
> >   net/mac80211/scan.c | 18 +++++++++++++-----
> >   1 file changed, 13 insertions(+), 5 deletions(-)
> > 
> > diff --git a/net/mac80211/scan.c b/net/mac80211/scan.c
> > index d4cc9ac2d703..562eda13e802 100644
> > --- a/net/mac80211/scan.c
> > +++ b/net/mac80211/scan.c
> > @@ -251,13 +251,21 @@ void ieee80211_scan_rx(struct ieee80211_local *local, struct sk_buff *skb)
> >   	struct ieee80211_mgmt *mgmt = (void *)skb->data;
> >   	struct ieee80211_bss *bss;
> >   	struct ieee80211_channel *channel;
> > +	size_t min_hdr_len = offsetof(struct ieee80211_mgmt, u.probe_resp.variable);
> > +
> > +	if (!ieee80211_is_probe_resp(mgmt->frame_control) &&
> > +			!ieee80211_is_beacon(mgmt->frame_control) &&
> > +			!ieee80211_is_s1g_beacon(mgmt->frame_control))
> > +		return;
> 
> Is the above check necessary? This doesn't look right. This skips
> the ieee80211_is_s1g_beacon() all together.

the original check only has the first two conditions (ieee80211_is_probe_resp()
and ieee80211_is_beacon()), but they were placed after condition of
ieee80211_is_s1g_beacon() not being met. Since I moved these checks at the
above, _before_ the if(ieee80211_is_s1g_beacon()), hence I joined
ieee80211_is_s1g_beacon() with the two orginal conditions.
> 
> 
>    	if (ieee80211_is_s1g_beacon(mgmt->frame_control)) {
> > -		if (skb->len < 15)
> > -			return;
> 
> Why not compare the header offset you expect here instead of 15 and
> return?

In fact, I do not understand where 15 (and the 24 shortly after) comes from.
They were there more than 10 years ago, but the more recent guard code in
cfg80211_inform_single_bss_frame_data() on the same header length seems more
correct, therefore I followed examples and copied the calculation from there,
for consistency.
> 
> > -	} else if (skb->len < 24 ||
> 
> Can you explain why it makes sense to remove < 24 check?

I replaced 24 with 
`size_t min_hdr_len = offsetof(struct ieee80211_mgmt, u.probe_resp.variable);`
which was found in cfg80211_inform_single_bss_frame_data(), for conditions:

	ieee80211_is_probe_resp(mgmt->frame_control)
or
	ieee80211_is_beacon(mgmt->frame_control)

> 
> > -		 (!ieee80211_is_probe_resp(mgmt->frame_control) &&
> > -		  !ieee80211_is_beacon(mgmt->frame_control)))
> > +		if (ieee80211_is_s1g_short_beacon(mgmt->frame_control))
> > +			min_hdr_len = offsetof(struct ieee80211_ext, u.s1g_short_beacon.variable);
> > +		else
> > +			min_hdr_len = offsetof(struct ieee80211_ext, u.s1g_beacon);
> > +	}
> > +
> > +	if (skb->len < min_hdr_len)
> >   		return;
> >   	sdata1 = rcu_dereference(local->scan_sdata);
> > 
> 
> thanks,
> -- Shuah

Best regards,
Du Cheng