Re: general protection fault in sctp_sched_prio_sched

[Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>]

Hi,

On Tue, Jun 18, 2019 at 04:04:01PM +0800, Hillf Danton wrote:
> 
> Hello Marcelo
> 
> On Mon, 17 Jun 2019 22:43:38 +0800 Marcelo Ricardo Leitner wrote:
> > On Mon, Jun 17, 2019 at 10:49:13AM -0300, Marcelo Ricardo Leitner wrote:
> > > Hi,
> > >
> > > On Sun, Jun 16, 2019 at 11:38:03PM +0800, Hillf Danton wrote:
> > > >
> > > > Hello Syzbot
> > > >
> > > > On Sat, 15 Jun 2019 16:36:06 -0700 (PDT) syzbot wrote:
> > > > > Hello,
> > > > >
> > > > > syzbot found the following crash on:
> > > > >
> > > ...
> > > > Check prio_head and bail out if it is not valid.
> > > >
> > > > Thanks
> > > > Hillf
> > > > ----->8---
> > > > ---
> > > > net/sctp/stream_sched_prio.c | 2 ++
> > > > 1 file changed, 2 insertions(+)
> > > >
> > > > diff --git a/net/sctp/stream_sched_prio.c b/net/sctp/stream_sched_prio.c
> > > > index 2245083..db25a43 100644
> > > > --- a/net/sctp/stream_sched_prio.c
> > > > +++ b/net/sctp/stream_sched_prio.c
> > > > @@ -135,6 +135,8 @@ static void sctp_sched_prio_sched(struct sctp_stream *stream,
> > > > 	struct sctp_stream_priorities *prio, *prio_head;
> > > >
> > > > 	prio_head = soute->prio_head;
> > > > +	if (!prio_head)
> > > > +		return;
> > > >
> > > > 	/* Nothing to do if already scheduled */
> > > > 	if (!list_empty(&soute->prio_list))
> > > > --
> > >
> > > Thanks but this is not a good fix for this. It will cause the stream
> > > to never be scheduled.
> > >
> Thanks very much for the light you are casting.
> 
> > > The problem happens because of the fault injection that happened a bit
> > > before the crash, in here:
> > >
> > > int sctp_stream_init_ext(struct sctp_stream *stream, __u16 sid)
> > > {
> > >         struct sctp_stream_out_ext *soute;
> > >
> > >         soute = kzalloc(sizeof(*soute), GFP_KERNEL);
> > >         if (!soute)
> > >                 return -ENOMEM;
> > >         SCTP_SO(stream, sid)->ext = soute;  <---- [A]
> > >
> > >         return sctp_sched_init_sid(stream, sid, GFP_KERNEL);
> > >                       ^^^^^^^^^^^^---- [B] failed
> > > }
> > >
> Eagle eye.
> 
> > > This causes the 1st sendmsg to bail out with the error. When the 2nd
> > > one gets in, it will:
> > >
> > > sctp_sendmsg_to_asoc()
> > > {
> > > ...
> > >         if (unlikely(!SCTP_SO(&asoc->stream, sinfo->sinfo_stream)->ext)) {
> > >                                                                  ^^^^^--- [C]
> > >                 err = sctp_stream_init_ext(&asoc->stream, sinfo->sinfo_stream);
> > >                 if (err)
> > >                         goto err;
> > >         }
> > >
> > > [A] leaves ext initialized, despite the failed in [B]. Then in [C], it
> > > will not try to initialize again.
> > >
> Fairly concise.
> 
> > > We need to either uninitialize ->ext as error handling for [B], or
> > > improve the check on [C].
> > 
> > The former one, please. This should be enough (untested):
> > 
> > diff --git a/net/sctp/stream.c b/net/sctp/stream.c
> > index 93ed07877337..25946604af85 100644
> > --- a/net/sctp/stream.c
> > +++ b/net/sctp/stream.c
> > @@ -153,13 +153,20 @@ int sctp_stream_init(struct sctp_stream *stream, __u1=
> > 6 outcnt, __u16 incnt,
> >  int sctp_stream_init_ext(struct sctp_stream *stream, __u16 sid)
> >  {
> >  	struct sctp_stream_out_ext *soute;
> > +	int ret;
> > 
> >  	soute = kzalloc(sizeof(*soute), GFP_KERNEL);
> >  	if (!soute)
> >  		return -ENOMEM;
> >  	SCTP_SO(stream, sid)->ext = soute;
> > 
> > -	return sctp_sched_init_sid(stream, sid, GFP_KERNEL);
> > +	ret = sctp_sched_init_sid(stream, sid, GFP_KERNEL);
> > +	if (ret) {
> > +		kfree(SCTP_SO(stream, sid)->ext);
> > +		SCTP_SO(stream, sid)->ext = NULL;

[D]

> > +	}
> > +
> > +	return ret;
> >  }
> > 
> Definitely nice.
> 
> >  void sctp_stream_free(struct sctp_stream *stream)
> > 
> Hmmm, ->ext will be valid, provided it is loaded with a valid slab in
> sctp_stream_init_ext() regardless of whether sid is successfully
> initialised, until it is released, for instance, in sctp_stream_free(),
> and based on that assumption, it looks hardly likely that ->ext has a
> chance to create a gfp in sctp_sched_prio_sched().

I'm not sure I follow you. Anyway, with the patch above, after calling
sctp_stream_init_ext() ->ext will be either completely valid, or it
will not be present at all as it is seting ->ext to NULL if sid
initialization ended up failing.

  Marcelo