On Wed, 31 Jan 2018 11:42:16 +0100
Daniel Borkmann <dan...@iogearbox.net> wrote:
> On 01/31/2018 08:24 AM, Jesper Dangaard Brouer wrote:
> > On Wed, 31 Jan 2018 01:31:11 +0100
> > Daniel Borkmann <dan...@iogearbox.net> wrote:
> >
> >> syzkaller was able to generate the following XDP program ...
> >>
> >> (18) r0 = 0x0
> >> (61) r5 = *(u32 *)(r1 +12)
> >> (04) (u32) r0 += (u32) 0
> >> (95) exit
> >>
> >> ... and trigger a NULL pointer dereference in ___bpf_prog_run()
> >> via bpf_prog_test_run_xdp() where this was attempted to run.
> >>
> >> Reason is that recent xdp_rxq_info addition to XDP programs
> >> updated all drivers, but not bpf_prog_test_run_xdp(), where
> >> xdp_buff is set up. Thus when context rewriter does the deref
> >> on the netdev it's NULL at runtime. Fix it by using xdp_rxq
> >> from loopback dev. netif_get_rx_queue() helper can also be
> >> reused in various other locations later on.
> >>
> >> Fixes: 02dd3291b2f0 ("bpf: finally expose xdp_rxq_info to XDP
> >> bpf-programs")
> >> Reported-by: syzbot+1eb094057b338eb1f...@syzkaller.appspotmail.com
> >> Signed-off-by: Daniel Borkmann <dan...@iogearbox.net>
> >> Cc: Jesper Dangaard Brouer <bro...@redhat.com>
> >> ---
> >> [ Note: Needs to wait till Linus pulled Dave's net-next so
> >> the affected commit lands in bpf tree. ]
> >>
> >> include/linux/netdevice.h | 6 ++++++
> >> net/bpf/test_run.c | 4 ++++
> >> tools/testing/selftests/bpf/test_verifier.c | 14 ++++++++++++++
> >> 3 files changed, 24 insertions(+)
> >>
> >> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> >> index cd46d3d..9630f4e 100644
> >> --- a/include/linux/netdevice.h
> >> +++ b/include/linux/netdevice.h
> >> @@ -3228,6 +3228,12 @@ static inline int
> >> netif_set_real_num_rx_queues(struct net_device *dev,
> >> }
> >> #endif
> >>
> >> +static inline struct netdev_rx_queue *
> >> +netif_get_rx_queue(struct net_device *dev, unsigned int rxq)
> >
> > This function name is very close to (net/core/dev.c):
> >
> > static struct netdev_rx_queue *
> > netif_get_rxqueue(struct sk_buff *skb)
> >
> >
> >> +{
> >> + return dev->_rx + rxq;
> >> +}
> >
> > I know above is correct, is just annoys my eyes as I find it more
> > natural to write as dev->_rx[rxq]. I'm not saying you should change
> > it, because it does preserve the access style elsewhere (and I also
> > kept this style in some of my changes).
> >
> > You mention netif_get_rx_queue() helper can also be reused in various
> > other locations later on, but there is not boundary checks (on this
>
> From a quick grep, few more could use this (didn't add this into
> the fix here, though, but can be later as cleanup):
Definitely save for a later cleanup, given the merge window timing.
> $ git grep -n "dev->_rx +"
> net/core/dev.c:3644: rxqueue = dev->_rx + rxq_index;
> net/core/dev.c:3784: struct netdev_rx_queue *rxqueue = dev->_rx +
> rxq_index;
> net/core/net-sysfs.c:940: struct netdev_rx_queue *queue = dev->_rx +
> index;
>
> > dynamically allocated array of size dev->num_rx_queues). We could add
> > a comment saying caller is responsible for boundary checks, or prefix
> > the function name with "__" to indicate this indirectly.
>
> Okay, I don't mind changing the name, would __netif_get_rx_queue()
> be a better one?
Yes, agreed.
(BTW - naming of function netif_get_rxqueue (in dev.c) is in principle
wrong it should have been netif_get_rx_queue I guess, but such nitpick
cleanup should also be postponed or ignored ;-))
--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
LinkedIn: http://www.linkedin.com/in/brouer
