Re: [V9fs-developer] [PATCH] p9_parse_header() validate PDU length

From: Dominique Martinet
Date: Wed Jul 18 2018 - 01:09:13 EST


Tomas Bortoli wrote on Thu, Jul 12, 2018:
> + Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>
> On 07/12/2018 01:43 PM, Dominique Martinet wrote:
> > Tomas Bortoli wrote on Thu, Jul 12, 2018:
> >> This patch adds checks to the p9_parse_header() function to
> >> verify that the length found within the header coincides with the actual
> >> length of the PDU. Furthermore, it checks that the length stays within the
> >> acceptable range. To do this the patch brings the actual length of the PDU
> >> from the different transport layers (rdma and virtio). For TCP (trans_fd.c)
> >> the length is not know before, so we get it from the header but we check it
> >> anyway that it's within the valid range.
>
> Still for TCP it you could read "garbage" pre-allocated memory but I
> don't know how much it is a risk, it might be a good idea to zero it
> post allocation (I mean pdu->sdata). Allocated at:
>
> https://github.com/torvalds/linux/blob/master/net/9p/client.c#L236
>
> > Just a note on transports here, I totally had forgotten about trans_xen
> > when we discussed this earlier as it is fairly new, but it looks like it
> > sets the length in the fcall properly so it should work without any
> > change.
> >
> > I however cannot test trans=xen, so if someone could either point me to
> > how to set that up (I couldn't find any decent documentation) or do some
> > very basic tests that would be great.
>
> >> Signed-off-by: Tomas Bortoli <tomasbortoli@xxxxxxxxx>
> >> Reported-by: syzbot+65c6b72f284a39d416b4@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Looks good to me, as the rdma/virtio part come from my suggestion:
> > Signed-off-by: Dominique Martinet <dominique.martinet@xxxxxx>
>
> True
> >
> >> diff --git a/net/9p/trans_rdma.c b/net/9p/trans_rdma.c
> >> index 3d414acb7015..002badbcc9c0 100644
> >> --- a/net/9p/trans_rdma.c
> >> +++ b/net/9p/trans_rdma.c
> >> @@ -319,7 +319,7 @@ recv_done(struct ib_cq *cq, struct ib_wc *wc)
> >>
> >> if (wc->status != IB_WC_SUCCESS)
> >> goto err_out;
> >> -
> >> + c->rc->size = wc->byte_len;
> > (nitpick, I'd keep the empty line here. If you don't mind I'll add it
> > back in my tree; this doesn't warrant a v2)
> >
>
> Sure,
>
> Tomas
>