On Tue, Jun 15, 2021 at 7:12 PM Greg KH <greg@xxxxxxxxx> wrote:
>
> On Tue, Jun 15, 2021 at 06:24:17PM +0800, Dongliang Mu wrote:
> > On Tue, Jun 15, 2021 at 6:10 PM Dongliang Mu <mudongliangabcd@xxxxxxxxx> wrote:
> > >
> > > On Tue, Jun 15, 2021 at 5:44 PM Greg KH <greg@xxxxxxxxx> wrote:
> > > >
> > > > On Tue, Jun 15, 2021 at 03:56:32PM +0800, Dongliang Mu wrote:
> > > > > On Tue, Jun 15, 2021 at 3:38 PM Greg KH <greg@xxxxxxxxx> wrote:
> > > > > >
> > > > > > On Mon, Jun 14, 2021 at 11:37:12PM +0800, Dongliang Mu wrote:
> > > > > > > The commit 46a8b29c6306 ("net: usb: fix memory leak in smsc75xx_bind")
> > > > > > > fails to clean up the work scheduled in smsc75xx_reset->
> > > > > > > smsc75xx_set_multicast, which leads to use-after-free if the work is
> > > > > > > scheduled to start after the deallocation. In addition, this patch also
> > > > > > > removes one dangling pointer - dev->data[0].
> > > > > > >
> > > > > > > This patch calls cancel_work_sync to cancel the schedule work and set
> > > > > > > the dangling pointer to NULL.
> > > > > > >
> > > > > > > Fixes: 46a8b29c6306 ("net: usb: fix memory leak in smsc75xx_bind")
> > > > > > > Signed-off-by: Dongliang Mu <mudongliangabcd@xxxxxxxxx>
> > > > > > > ---
> > > > > > > drivers/net/usb/smsc75xx.c | 3 +++
> > > > > > > 1 file changed, 3 insertions(+)
> > > > > > >
> > > > > > > diff --git a/drivers/net/usb/smsc75xx.c b/drivers/net/usb/smsc75xx.c
> > > > > > > index b286993da67c..f81740fcc8d5 100644
> > > > > > > --- a/drivers/net/usb/smsc75xx.c
> > > > > > > +++ b/drivers/net/usb/smsc75xx.c
> > > > > > > @@ -1504,7 +1504,10 @@ static int smsc75xx_bind(struct usbnet *dev, struct usb_interface *intf)
> > > > > > > return 0;
> > > > > > >
> > > > > > > err:
> > > > > > > + cancel_work_sync(&pdata->set_multicast);
> > > > > > > kfree(pdata);
> > > > > > > + pdata = NULL;
> > > > > >
> > > > > > Why do you have to set pdata to NULL afterward?
> > > > > >
> > > > >
> > > > > It does not have to. pdata will be useless when the function exits. I
> > > > > just referred to the implementation of smsc75xx_unbind.
> > > >
> > > > It's wrong there too :)
> > >
> > > /: I will fix such two sites in the v2 patch.
> >
> > Hi gregkh,
> >
> > If the schedule_work is not invoked, can I call
> > ``cancel_work_sync(&pdata->set_multicast)''?
>
> Why can you not call this then?
I don't know the internal of schedule_work and cancel_work_sync, so I
ask this question to confirm my patch does not introduce any new
issues.
>
> Did you try it and see?
Yes, I thought up a method and tested it in my local workspace.
First, I reproduced the memory leak in smsc75xx_bind [1] since the PoC
triggered an error before schedule_work.
Then, I merged two patches, and run the PoC. The result showed that my
patch does not trigger any new issues even the schedule_work is not
called.
[1] https://syzkaller.appspot.com/bug?id=c978ec308a1b89089a17ff48183d70b4c840dfb0
>
> thanks,
>
> greg k-h
