Re: [Linux-kernel-mentees] [PATCH] net: bluetooth: Fix null pointer deref in hci_phy_link_complete_evt

[Greg KH <gregkh@linuxfoundation.org>]

On Sun, Aug 30, 2020 at 05:56:23PM +0530, Anmol Karn wrote:
> On Sun, Aug 30, 2020 at 11:19:17AM +0200, Greg KH wrote:
> > On Sat, Aug 29, 2020 at 10:27:12PM +0530, Anmol Karn wrote:
> > > Fix null pointer deref in hci_phy_link_complete_evt, there was no 
> > > checking there for the hcon->amp_mgr->l2cap_conn->hconn, and also 
> > > in hci_cmd_work, for hdev->sent_cmd.
> > > 
> > > To fix this issue Add pointer checking in hci_cmd_work and
> > > hci_phy_link_complete_evt.
> > > [Linux-next-20200827]
> > > 
> > > This patch corrected some mistakes from previous patch.
> > > 
> > > Reported-by: syzbot+0bef568258653cff272f@syzkaller.appspotmail.com
> > > Link: https://syzkaller.appspot.com/bug?id=0d93140da5a82305a66a136af99b088b75177b99
> > > Signed-off-by: Anmol Karn <anmol.karan123@gmail.com>
> > > ---
> > >  net/bluetooth/hci_core.c  | 5 ++++-
> > >  net/bluetooth/hci_event.c | 4 ++++
> > >  2 files changed, 8 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
> > > index 68bfe57b6625..996efd654e7a 100644
> > > --- a/net/bluetooth/hci_core.c
> > > +++ b/net/bluetooth/hci_core.c
> > > @@ -4922,7 +4922,10 @@ static void hci_cmd_work(struct work_struct *work)
> > >  
> > >  		kfree_skb(hdev->sent_cmd);
> > >  
> > > -		hdev->sent_cmd = skb_clone(skb, GFP_KERNEL);
> > > +		if (hdev->sent_cmd) {
> > > +			hdev->sent_cmd = skb_clone(skb, GFP_KERNEL);
> > > +		}
> > 
> > How can sent_cmd be NULL here?  Are you sure something previous to this
> > shouldn't be fixed instead?
> 
> Sir, sent_cmd was freed before this condition check, thats why i checked it,

But it can not be NULL at that point in time, as nothing set it to NULL,
correct?

> i think i should check it before the free of hdev->sent_cmd like,
> 
> if (hdev->sent_cmd)
> 	kfree_skb(hdev->sent_cmd);

No, that's not needed.

What is the problem with these lines that you are trying to solve?

> > > +
> > >  		if (hdev->sent_cmd) {
> > >  			if (hci_req_status_pend(hdev))
> > >  				hci_dev_set_flag(hdev, HCI_CMD_PENDING);
> > > diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c
> > > index 4b7fc430793c..1e7d9bee9111 100644
> > > --- a/net/bluetooth/hci_event.c
> > > +++ b/net/bluetooth/hci_event.c
> > > @@ -4941,6 +4941,10 @@ static void hci_phy_link_complete_evt(struct hci_dev *hdev,
> > >  		hci_dev_unlock(hdev);
> > >  		return;
> > >  	}
> > > +	if (!(hcon->amp_mgr->l2cap_conn->hcon)) {
> > > +		hci_dev_unlock(hdev);
> > > +		return;
> > > +	}
> > 
> > How can this be triggered?
> 
> syzbot showed that this line is accessed irrespective of the null value it contains, so  added a 
> pointer check for that.

But does hcon->amp_mgr->l2cap_conn->hcon become NULL here?

thanks,

greg k-h