Sun, Jan 07, 2018 at 02:11:19PM CET, jhs@mojatatu.com wrote:
On 18-01-06 03:43 PM, Jiri Pirko wrote:
quoted
quoted
@@ -886,8 +887,13 @@ static int tcf_fill_node(struct net *net, struct sk_buff *skb,
tcm->tcm_family = AF_UNSPEC;
tcm->tcm__pad1 = 0;
tcm->tcm__pad2 = 0;
- tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
- tcm->tcm_parent = parent;
+ if (q) {
+ tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
+ tcm->tcm_parent = parent;
+ } else {
+ tcm->tcm_ifindex = 0; /* block index is stored in parent */
+ tcm->tcm_parent = block->index;
+ }
Please guys, please look at this reuse (also on clt side). I would like
you to double-check this reuse of existing API for balock_index carrying
purpose. I believe it's UAPI safe. But please, check it out carefully.
Should not break any ABI/UAPI AFAIK. Maybe go for a negative ifindex
(not sure if zero means something speacial to someone).
Like -1 means parent is block_index?
Why would 0 mean something special? Could you point to a code that
suggests it?
cheers,
jamal