Re: Re: Re: [PATCH] infiniband/core: Fix a use after free in cm_work_handler

[lyl2019@mail.ustc.edu.cn]

> -----原始邮件-----
> 发件人: "Leon Romanovsky" <leon@kernel.org>
> 发送时间: 2021-03-11 19:05:03 (星期四)
> 收件人: lyl2019@mail.ustc.edu.cn
> 抄送: dledford@redhat.com, jgg@ziepe.ca, linux-rdma@vger.kernel.org, linux-kernel@vger.kernel.org
> 主题: Re: Re: [PATCH] infiniband/core: Fix a use after free in cm_work_handler
> 
> On Thu, Mar 11, 2021 at 06:29:19PM +0800, lyl2019@mail.ustc.edu.cn wrote:
> > In the implementation of destory_cm_id(), it restores cm_id_priv by
> > "cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);".
> >
> > And the last line of destory_cm_id() calls "(void)iwcm_deref_id(cm_id_priv);"
> > to free the cm_id_priv.
> 
> It is not enough to see double call to iwcm_deref_id() because it is
> protected with refcount to claim use-after-free. Did you hit the BUG_ON()
> for the second call to iwcm_deref_id()?
> 
> And please don't do top-posting.
> 
> Thanks
> 
> >
> >
> > > -----原始邮件-----
> > > 发件人: "Leon Romanovsky" <leon@kernel.org>
> > > 发送时间: 2021-03-11 17:22:03 (星期四)
> > > 收件人: "Lv Yunlong" <lyl2019@mail.ustc.edu.cn>
> > > 抄送: dledford@redhat.com, jgg@ziepe.ca, linux-rdma@vger.kernel.org, linux-kernel@vger.kernel.org
> > > 主题: Re: [PATCH] infiniband/core: Fix a use after free in cm_work_handler
> > >
> > > On Wed, Mar 10, 2021 at 06:21:53PM -0800, Lv Yunlong wrote:
> > > > In cm_work_handler, it calls destory_cm_id() to release
> > > > the initial reference of cm_id_priv taken by iw_create_cm_id()
> > > > and free the cm_id_priv. After destory_cm_id(), iwcm_deref_id
> > > > (cm_id_priv) will be called and cause a use after free.
> > > >
> > > > Fixes: 59c68ac31e15a ("iw_cm: free cm_id resources on the last deref")
> > > > Signed-off-by: Lv Yunlong <lyl2019@mail.ustc.edu.cn>
> > > > ---
> > > > drivers/infiniband/core/iwcm.c | 4 +++-
> > > > 1 file changed, 3 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/infiniband/core/iwcm.c b/drivers/infiniband/core/iwcm.c
> > > > index da8adadf4755..cb6b4ac45e21 100644
> > > > --- a/drivers/infiniband/core/iwcm.c
> > > > +++ b/drivers/infiniband/core/iwcm.c
> > > > @@ -1035,8 +1035,10 @@ static void cm_work_handler(struct work_struct *_work)
> > > >
> > > > 		if (!test_bit(IWCM_F_DROP_EVENTS, &cm_id_priv->flags)) {
> > > > 			ret = process_event(cm_id_priv, &levent);
> > > > -			if (ret)
> > > > +			if (ret) {
> > > > 				destroy_cm_id(&cm_id_priv->id);
> > > > +				return;
> > >
> > > The destroy_cm_id() is called to free ->id and not cm_id_priv. This "return"
> > > leaks cm_id_priv now and what "a use after free" do you see?
> > >
> > > > +			}
> > > > 		} else
> > > > 			pr_debug("dropping event %d\n", levent.event);
> > > > 		if (iwcm_deref_id(cm_id_priv))
> > > > --
> > > > 2.25.1
> > > >
> > > >

I'm not familiar with debug the kernel, sorry.This problem was
 reported by my code analyzer and reviewed by myself.

But i think as long as destroy_cm_id() is called, iwcm_deref_id() will be called twice.
Then "BUG_ON(atomic_read(&cm_id_priv->refcount)==0);" in iwcm_deref_id() will be triggered.

Is it not a true problem?