Prior to commit b3f55bdda8df, the networking core doesn't wire an in-place
actions list the when the low level driver is called to offload the flow,
but all low level drivers do that (call tcf_exts_to_list()) in their
offloading "add" logic.
Now, the in-place list is set in the core which goes over the list in a loop,
but also by the hw driver when their offloading code is invoked indirectly:
cls_xxx add flow -> tc_setup_cb_call -> tc_exts_setup_cb_egdev_call ->
hw driver
which messes up the core list instance upon driver return. Fix that by avoiding
in-place list on the net core code that deals with adding flows.
Fixes: b3f55bdda8df ('net: sched: introduce per-egress action device callbacks')
Signed-off-by: Or Gerlitz <ogerl...@mellanox.com>
Acked-by: Jiri Pirko <j...@mellanox.com>
---
net/sched/cls_api.c | 7 +++----
1 file changed, 3 insertions(+), 4 deletions(-)
diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c
index cdfdc24..0e96cda 100644
--- a/net/sched/cls_api.c
+++ b/net/sched/cls_api.c
@@ -1184,14 +1184,13 @@ static int tc_exts_setup_cb_egdev_call(struct tcf_exts
*exts,
#ifdef CONFIG_NET_CLS_ACT
const struct tc_action *a;
struct net_device *dev;
- LIST_HEAD(actions);
- int ret;
+ int i, ret;
if (!tcf_exts_has_actions(exts))
return 0;
- tcf_exts_to_list(exts, &actions);
- list_for_each_entry(a, &actions, list) {
+ for (i = 0; i < exts->nr_actions; i++) {
+ a = exts->actions[i];
if (!a->ops->get_dev)
continue;
dev = a->ops->get_dev(a);
--
2.3.7
