On Tue, May 5, 2020 at 3:27 PM Michal Kubecek <mkube...@suse.cz> wrote:
>
> On Tue, May 05, 2020 at 02:58:19PM -0700, Cong Wang wrote:
> > syzbot managed to trigger a recursive NETDEV_FEAT_CHANGE event
> > between bonding master and slave. I managed to find a reproducer
> > for this:
> >
> > ip li set bond0 up
> > ifenslave bond0 eth0
> > brctl addbr br0
> > ethtool -K eth0 lro off
> > brctl addif br0 bond0
> > ip li set br0 up
> >
> > When a NETDEV_FEAT_CHANGE event is triggered on a bonding slave,
> > it captures this and calls bond_compute_features() to fixup its
> > master's and other slaves' features. However, when syncing with
> > its lower devices by netdev_sync_lower_features() this event is
> > triggered again on slaves, so it goes back and forth recursively
> > until the kernel stack is exhausted.
> >
> > It is unnecessary to trigger it for a second time, because when
> > we update the features from top down, we rely on each
> > dev->netdev_ops->ndo_fix_features() to do the job, each stacked
> > device should implement it. NETDEV_FEAT_CHANGE event is necessary
> > when we update from bottom up, like in existing stacked device
> > implementations.
> >
> > Just calling __netdev_update_features() is sufficient to fix this
> > issue.
> >
> > Fixes: fd867d51f889 ("net/core: generic support for disabling netdev
> > features down stack")
> > Reported-by: syzbot+e73ceacfd8560cc8a...@syzkaller.appspotmail.com
> > Reported-by: syzbot+c2fb6f9ddcea95ba4...@syzkaller.appspotmail.com
> > Cc: Jarod Wilson <ja...@redhat.com>
> > Cc: Josh Poimboeuf <jpoim...@redhat.com>
> > Cc: Jay Vosburgh <j.vosbu...@gmail.com>
> > Cc: Jann Horn <ja...@google.com>
> > Signed-off-by: Cong Wang <xiyou.wangc...@gmail.com>
> > ---
> > net/core/dev.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/net/core/dev.c b/net/core/dev.c
> > index 522288177bbd..ece50ae346c3 100644
> > --- a/net/core/dev.c
> > +++ b/net/core/dev.c
> > @@ -8907,7 +8907,7 @@ static void netdev_sync_lower_features(struct
> > net_device *upper,
> > netdev_dbg(upper, "Disabling feature %pNF on lower
> > dev %s.\n",
> > &feature, lower->name);
> > lower->wanted_features &= ~feature;
> > - netdev_update_features(lower);
> > + __netdev_update_features(lower);
> >
> > if (unlikely(lower->features & feature))
> > netdev_WARN(upper, "failed to disable %pNF on
> > %s!\n",
>
> Wouldn't this mean that when we disable LRO on a bond manually with
> "ethtool -K", LRO will be also disabled on its slaves but no netlink
> notification for them would be sent to userspace?
What netlink notification are you talking about?
When we change features from top down, ->ndo_fix_features()
does the work, in bonding case, it is bond_fix_features().
I see no netlink notification either in bond_compute_features()
or bond_fix_features().
Thanks.
