Re: Re: Re: [Drbd-dev] [PATCH] drbd: Fix a use after free in get_initial_state

[lyl2019@mail.ustc.edu.cn]

> -----原始邮件-----
> 发件人: lyl2019@mail.ustc.edu.cn
> 发送时间: 2021-04-01 23:13:58 (星期四)
> 收件人: "Christoph Böhmwalder" <christoph.boehmwalder@linbit.com>
> 抄送: philipp.reisner@linbit.com, lars.ellenberg@linbit.com, axboe@kernel.dk, linux-block@vger.kernel.org, linux-kernel@vger.kernel.org, drbd-dev@lists.linbit.com
> 主题: Re: Re: [Drbd-dev] [PATCH] drbd: Fix a use after free in get_initial_state
> 
> 
> 
> 
> > -----原始邮件-----
> > 发件人: "Christoph Böhmwalder" <christoph.boehmwalder@linbit.com>
> > 发送时间: 2021-04-01 21:01:20 (星期四)
> > 收件人: "Lv Yunlong" <lyl2019@mail.ustc.edu.cn>
> > 抄送: philipp.reisner@linbit.com, lars.ellenberg@linbit.com, axboe@kernel.dk, linux-block@vger.kernel.org, linux-kernel@vger.kernel.org, drbd-dev@lists.linbit.com
> > 主题: Re: [Drbd-dev] [PATCH] drbd: Fix a use after free in get_initial_state
> > 
> > On 4/1/21 1:57 PM, Lv Yunlong wrote:
> > > In get_initial_state, it calls notify_initial_state_done(skb,..) if
> > > cb->args[5]==1. I see that if genlmsg_put() failed in
> > > notify_initial_state_done(), the skb will be freed by nlmsg_free(skb).
> > > Then get_initial_state will goto out and the freed skb will be used by
> > > return value skb->len.
> > > 
> > > My patch lets skb_len = skb->len and return the skb_len to avoid the uaf.
> > > 
> > > Fixes: a29728463b254 ("drbd: Backport the "events2" command")
> > > Signed-off-by: Lv Yunlong <lyl2019@mail.ustc.edu.cn>
> > > ---
> > >   drivers/block/drbd/drbd_nl.c | 3 ++-
> > >   1 file changed, 2 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/block/drbd/drbd_nl.c b/drivers/block/drbd/drbd_nl.c
> > > index bf7de4c7b96c..474f84675d0a 100644
> > > --- a/drivers/block/drbd/drbd_nl.c
> > > +++ b/drivers/block/drbd/drbd_nl.c
> > > @@ -4905,6 +4905,7 @@ static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
> > >   	struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0];
> > >   	unsigned int seq = cb->args[2];
> > >   	unsigned int n;
> > > +	unsigned int skb_len = skb->len;
> > >   	enum drbd_notification_type flags = 0;
> > >   
> > >   	/* There is no need for taking notification_mutex here: it doesn't
> > > @@ -4915,7 +4916,7 @@ static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
> > >   	cb->args[5]--;
> > >   	if (cb->args[5] == 1) {
> > >   		notify_initial_state_done(skb, seq);
> > > -		goto out;
> > > +		return skb_len;
> > >   	}
> > >   	n = cb->args[4]++;
> > >   	if (cb->args[4] < cb->args[3])
> > > 
> > 
> > Thanks for the patch!
> > 
> > I think the problem goes even further: skb can also be freed in the 
> > notify_*_state_change -> notify_*_state calls below.
> > 
> > Also, at the point where we save skb->len into skb_len, skb is not 
> > initialized yet. Maybe it makes more sense to not return a length in the 
> > first place here, but an error code instead.
> > 
> > -- 
> > Christoph Böhmwalder
> > LINBIT | Keeping the Digital World Running
> > DRBD HA —  Disaster Recovery — Software defined Storage
> 
> Ok, I see.
> I found that drbd_adm_get_initial_state() has called the get_initial_state(),
> and return -ENOMEM if it calls remember_old_state() failed.
> 
> So, i think that means if get_initial_state() failed on the notify_initial_state_done(),
> it should return -ENOMEM too.
> 
> I will submit the PATCH v2 to fix the first place. The fixes of the further problem is 
> hard for me.
> 
> Thanks.

I found that notify_initial_state_done() uses err = -EMSGSIZE, so the first place should
return -EMSGSIZE not -ENOMEM. Sorry.