On Thu, 21 Mar 2019 11:26:56 -0700, Cong Wang wrote:
> On Thu, Mar 21, 2019 at 11:22 AM Jakub Kicinski wrote:
> > On Thu, 21 Mar 2019 11:05:21 -0700, Cong Wang wrote:
> > > On Wed, Mar 20, 2019 at 11:12 AM John Hurley wrote:
> > > >
> > > > A new mirred action is created by the tcf_mirred_init function. This
> > > > contains a list head struct which is inserted into a global list on
> > > > successful creation of a new action. However, after a creation, it is
> > > > still possible to error out if the egress device does not exist. This
> > > > calls the act_mirr cleanup function via __tcf_idr_release and
> > > > __tcf_action_put. This cleanup function tries to delete the list entry
> > > > which is as yet uninitialised, leading to a NULL pointer exception.
> > >
> > > Hmm, good catch but can this be just fixed by initializing it before
> > > taking the netdevice refcnt? Like this:
> > >
> > > @@ -163,6 +163,9 @@ static int tcf_mirred_init(struct net *net, struct
> > > nlattr *nla,
> > > m->tcf_action = parm->action;
> > > m->tcfm_eaction = parm->eaction;
> > >
> > > + if (ret == ACT_P_CREATED)
> > > + INIT_LIST_HEAD(&m->tcfm_list);
> > > +
> > > if (parm->ifindex) {
> > > dev = dev_get_by_index(net, parm->ifindex);
> > > if (!dev) {
> > >
> > > which is also much simpler.
> >
> > That's the initial way John fixed it, but I asked him to go back to the
> > previous way this code was written.
> >
> > I think having the parameters validated before any allocations happen
> > is less error prone, especially with the strange way actions get the
> > release call even when init failed. It's just a more reliable code
> > patter for actions' init callback.
>
> The point is a simpler version is better for -net and -stable,
> and for review too.
>
> You can always refactor it for net-next if you feel necessary.
I personally prefer to push a slightly bigger fix (30 lines of change
is hardly large after all) if it means we can have the same code in
net-next as in LTS kernels (4.19 in this case). Especially for
relatively stable code.
But yeah, I was on the fence too, so given your preference we can go
back to init head.
