On Thu, Aug 15, 2013 at 03:45:49PM +0200, Simon Wunderlich wrote:
Hey Linus,
I'm currently playing a little with this patchset ...
On Tue, Aug 13, 2013 at 10:23:58AM +0200, Linus Lüssing wrote:
[...]
- count = atomic_read(&tt_global_entry->orig_list_count);
[...]
- atomic_inc(&tt_global->orig_list_count);
[...]
atomic_set(&tt_global_entry->orig_list_count, 0);[...] @@ -830,12 +833,14 @@ struct batadv_tt_local_entry {
- struct batadv_tt_global_entry - translation table global entry data
- @common: general translation table data
- @orig_list: list of orig nodes announcing this non-mesh client
*/
- @orig_list_count: number of items in the orig_list
- @list_lock: lock protecting orig_list
- @roam_at: time at which TT_GLOBAL_ROAM was set
struct batadv_tt_global_entry { struct batadv_tt_common_entry common; struct hlist_head orig_list;
- atomic_t orig_list_count; spinlock_t list_lock; /* protects orig_list */ unsigned long roam_at;
You never decrease this orig_list_count, do you? But actually you should when a node leaves the group. What I've tried is:
Node 1 pings a multicast address -> no group member, nothing is sent Node 2 is added to the group -> multicast is now sent as unicast Node 3 is added to the group -> multicast is now sent as broadcast Node 2 is removed from the group -> multicast is still sent as broadcast (should be unicast!) Node 3 is removed from the group -> nothing is sent (as to be expected).
Could you please check the removal case?
Thanks, Simon
Hrm, stupid regression introduced with v8 while rebasing... Sorry and thanks for spotting it! Going to send v9 in a minute.