If receiving an OGM from a neighbor other than the currently selected and if it has the same TQ then we are supposed to switch if this neighbor provides a more symmetric link than the currently selected one.
However this symmetry check currently is broken if the interface of the neighbor we received the OGM from and the one of the currently selected neighbor differ: We are currently trying to determine the symmetry of the link towards the selected router via the link we received the OGM from instead of just checking via the link towards the currently selected router.
This leads to way more route switches than necessary and can lead to permanent route flapping in many common multi interface setups.
This patch fixes this issue by using the right interface for this symmetry check.
Signed-off-by: Linus Lüssing linus.luessing@web.de --- This fixes the route flapping observed in ticket 163 (but unfortunately doesn't fix the main issue of this ticket, the starving routes, yet).
bat_iv_ogm.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/bat_iv_ogm.c b/bat_iv_ogm.c index 7f0adad..2c6b560 100644 --- a/bat_iv_ogm.c +++ b/bat_iv_ogm.c @@ -743,7 +743,8 @@ batadv_iv_ogm_orig_update(struct batadv_priv *bat_priv, if (router && (neigh_node->tq_avg == router->tq_avg)) { orig_node_tmp = router->orig_node; spin_lock_bh(&orig_node_tmp->ogm_cnt_lock); - sum_orig = orig_node_tmp->bcast_own_sum[if_incoming->if_num]; + sum_orig = orig_node_tmp-> + bcast_own_sum[router->if_incoming->if_num]; spin_unlock_bh(&orig_node_tmp->ogm_cnt_lock);
orig_node_tmp = neigh_node->orig_node;
On Sat, Sep 15, 2012 at 11:51:46 +0200, Linus Lüssing wrote:
If receiving an OGM from a neighbor other than the currently selected and if it has the same TQ then we are supposed to switch if this neighbor provides a more symmetric link than the currently selected one.
However this symmetry check currently is broken if the interface of the neighbor we received the OGM from and the one of the currently selected neighbor differ: We are currently trying to determine the symmetry of the link towards the selected router via the link we received the OGM from instead of just checking via the link towards the currently selected router.
This leads to way more route switches than necessary and can lead to permanent route flapping in many common multi interface setups.
This patch fixes this issue by using the right interface for this symmetry check.
Signed-off-by: Linus Lüssing linus.luessing@web.de
This fixes the route flapping observed in ticket 163 (but unfortunately doesn't fix the main issue of this ticket, the starving routes, yet).
bat_iv_ogm.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/bat_iv_ogm.c b/bat_iv_ogm.c index 7f0adad..2c6b560 100644 --- a/bat_iv_ogm.c +++ b/bat_iv_ogm.c @@ -743,7 +743,8 @@ batadv_iv_ogm_orig_update(struct batadv_priv *bat_priv, if (router && (neigh_node->tq_avg == router->tq_avg)) { orig_node_tmp = router->orig_node; spin_lock_bh(&orig_node_tmp->ogm_cnt_lock);
sum_orig = orig_node_tmp->bcast_own_sum[if_incoming->if_num];
sum_orig = orig_node_tmp->
bcast_own_sum[router->if_incoming->if_num];
what about: int if_num; ... if_num = router->if_incoming->if_num; sum_orig = orig_node_tmp->bcast_own_sum[if_num];
?
If receiving an OGM from a neighbor other than the currently selected and if it has the same TQ then we are supposed to switch if this neighbor provides a more symmetric link than the currently selected one.
However this symmetry check currently is broken if the interface of the neighbor we received the OGM from and the one of the currently selected neighbor differ: We are currently trying to determine the symmetry of the link towards the selected router via the link we received the OGM from instead of just checking via the link towards the currently selected router.
This leads to way more route switches than necessary and can lead to permanent route flapping in many common multi interface setups.
This patch fixes this issue by using the right interface for this symmetry check.
Signed-off-by: Linus Lüssing linus.luessing@web.de --- bat_iv_ogm.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/bat_iv_ogm.c b/bat_iv_ogm.c index 7f0adad..a92a663 100644 --- a/bat_iv_ogm.c +++ b/bat_iv_ogm.c @@ -657,6 +657,7 @@ batadv_iv_ogm_orig_update(struct batadv_priv *bat_priv, struct batadv_neigh_node *router = NULL; struct batadv_orig_node *orig_node_tmp; struct hlist_node *node; + int if_num; uint8_t sum_orig, sum_neigh; uint8_t *neigh_addr; uint8_t tq_avg; @@ -743,7 +744,8 @@ batadv_iv_ogm_orig_update(struct batadv_priv *bat_priv, if (router && (neigh_node->tq_avg == router->tq_avg)) { orig_node_tmp = router->orig_node; spin_lock_bh(&orig_node_tmp->ogm_cnt_lock); - sum_orig = orig_node_tmp->bcast_own_sum[if_incoming->if_num]; + if_num = router->if_incoming->if_num; + sum_orig = orig_node_tmp->bcast_own_sum[if_num]; spin_unlock_bh(&orig_node_tmp->ogm_cnt_lock);
orig_node_tmp = neigh_node->orig_node;
On Tuesday, September 18, 2012 09:01:08 Linus Lüssing wrote:
@@ -743,7 +744,8 @@ batadv_iv_ogm_orig_update(struct batadv_priv *bat_priv, if (router && (neigh_node->tq_avg == router->tq_avg)) { orig_node_tmp = router->orig_node; spin_lock_bh(&orig_node_tmp->ogm_cnt_lock);
sum_orig =
orig_node_tmp->bcast_own_sum[if_incoming->if_num]; + if_num = router->if_incoming->if_num;
sum_orig = orig_node_tmp->bcast_own_sum[if_num]; spin_unlock_bh(&orig_node_tmp->ogm_cnt_lock); orig_node_tmp = neigh_node->orig_node;
Good catch! May I suggest another modifcation to make this section easier to understand ? It won't change any behavior it just makes it easier to see where if_num is coming from.
Cheers, Marek
From: Linus Lüssing linus.luessing@web.de
If receiving an OGM from a neighbor other than the currently selected and if it has the same TQ then we are supposed to switch if this neighbor provides a more symmetric link than the currently selected one.
However this symmetry check currently is broken if the interface of the neighbor we received the OGM from and the one of the currently selected neighbor differ: We are currently trying to determine the symmetry of the link towards the selected router via the link we received the OGM from instead of just checking via the link towards the currently selected router.
This leads to way more route switches than necessary and can lead to permanent route flapping in many common multi interface setups.
This patch fixes this issue by using the right interface for this symmetry check.
Signed-off-by: Linus Lüssing linus.luessing@web.de --- bat_iv_ogm.c | 7 +++++-- 1 files changed, 5 insertions(+), 2 deletions(-)
diff --git a/bat_iv_ogm.c b/bat_iv_ogm.c index 7f0adad..75403a4 100644 --- a/bat_iv_ogm.c +++ b/bat_iv_ogm.c @@ -657,6 +657,7 @@ batadv_iv_ogm_orig_update(struct batadv_priv *bat_priv, struct batadv_neigh_node *router = NULL; struct batadv_orig_node *orig_node_tmp; struct hlist_node *node; + int if_num; uint8_t sum_orig, sum_neigh; uint8_t *neigh_addr; uint8_t tq_avg; @@ -743,12 +744,14 @@ batadv_iv_ogm_orig_update(struct batadv_priv *bat_priv, if (router && (neigh_node->tq_avg == router->tq_avg)) { orig_node_tmp = router->orig_node; spin_lock_bh(&orig_node_tmp->ogm_cnt_lock); - sum_orig = orig_node_tmp->bcast_own_sum[if_incoming->if_num]; + if_num = router->if_incoming->if_num; + sum_orig = orig_node_tmp->bcast_own_sum[if_num]; spin_unlock_bh(&orig_node_tmp->ogm_cnt_lock);
orig_node_tmp = neigh_node->orig_node; spin_lock_bh(&orig_node_tmp->ogm_cnt_lock); - sum_neigh = orig_node_tmp->bcast_own_sum[if_incoming->if_num]; + if_num = neigh_node->if_incoming->if_num; + sum_neigh = orig_node_tmp->bcast_own_sum[if_num]; spin_unlock_bh(&orig_node_tmp->ogm_cnt_lock);
if (sum_orig >= sum_neigh)
Yes, looks good. Makes the symmetry patch look more, ehm, "symmetric".
Gesendet: Sonntag, 23. September 2012 um 11:16 Uhr Von: "Marek Lindner" lindner_marek@yahoo.de An: b.a.t.m.a.n@lists.open-mesh.org Cc: "Linus Lüssing" linus.luessing@web.de Betreff: [PATCHv3] batman-adv: Fix symmetry check / route flapping in multi interface setups
From: Linus Lüssing linus.luessing@web.de
If receiving an OGM from a neighbor other than the currently selected and if it has the same TQ then we are supposed to switch if this neighbor provides a more symmetric link than the currently selected one.
However this symmetry check currently is broken if the interface of the neighbor we received the OGM from and the one of the currently selected neighbor differ: We are currently trying to determine the symmetry of the link towards the selected router via the link we received the OGM from instead of just checking via the link towards the currently selected router.
This leads to way more route switches than necessary and can lead to permanent route flapping in many common multi interface setups.
This patch fixes this issue by using the right interface for this symmetry check.
Signed-off-by: Linus Lüssing linus.luessing@web.de
bat_iv_ogm.c | 7 +++++-- 1 files changed, 5 insertions(+), 2 deletions(-)
diff --git a/bat_iv_ogm.c b/bat_iv_ogm.c index 7f0adad..75403a4 100644 --- a/bat_iv_ogm.c +++ b/bat_iv_ogm.c @@ -657,6 +657,7 @@ batadv_iv_ogm_orig_update(struct batadv_priv *bat_priv, struct batadv_neigh_node *router = NULL; struct batadv_orig_node *orig_node_tmp; struct hlist_node *node;
- int if_num; uint8_t sum_orig, sum_neigh; uint8_t *neigh_addr; uint8_t tq_avg;
@@ -743,12 +744,14 @@ batadv_iv_ogm_orig_update(struct batadv_priv *bat_priv, if (router && (neigh_node->tq_avg == router->tq_avg)) { orig_node_tmp = router->orig_node; spin_lock_bh(&orig_node_tmp->ogm_cnt_lock);
sum_orig = orig_node_tmp->bcast_own_sum[if_incoming->if_num];
if_num = router->if_incoming->if_num;
sum_orig = orig_node_tmp->bcast_own_sum[if_num];
spin_unlock_bh(&orig_node_tmp->ogm_cnt_lock);
orig_node_tmp = neigh_node->orig_node; spin_lock_bh(&orig_node_tmp->ogm_cnt_lock);
sum_neigh = orig_node_tmp->bcast_own_sum[if_incoming->if_num];
if_num = neigh_node->if_incoming->if_num;
sum_neigh = orig_node_tmp->bcast_own_sum[if_num];
spin_unlock_bh(&orig_node_tmp->ogm_cnt_lock);
if (sum_orig >= sum_neigh)
-- 1.7.9.1
On Sunday, September 23, 2012 18:44:54 Linus Lüssing wrote:
Yes, looks good. Makes the symmetry patch look more, ehm, "symmetric".
Ok, merged the patch in revision 6c9d9ee.
Thanks, Marek
b.a.t.m.a.n@lists.open-mesh.org