batman-adv: Fix broadcast queue limit on a removed interface
PATCH 1/2 batman-adv: Fix broadcast queue limit on a removed interface
When removing a single interface while a broadcast or ogm packet is still pending then we will free the forward packet without releasing the queue slots again.
This patch is supposed to fix this issue.
--- send.c | 6 ++++++ 1 file changed, 6 insertions(+)
send.c +6 -0
@@ -356,6 +356,9 @@ batadv_purge_outstanding_packets(struct batadv_priv *bat_priv,if (pending) {hlist_del (&forw_packet->list); +if (!forw_packet->own) +atomic_inc (&bat_priv->bcast_queue_left); +batadv_forw_packet_free (forw_packet); } }
@@ -382,6 +385,9 @@ batadv_purge_outstanding_packets(struct batadv_priv *bat_priv,if (pending) {hlist_del (&forw_packet->list); +if (!forw_packet->own) +atomic_inc (&bat_priv->batman_queue_left); +batadv_forw_packet_free (forw_packet); } }
Signature · 1 line
1.7.10.4
PATCH 2/2 batman-adv: Refactor forward packet creation/destruction
This patch abstracts the forward packet creation and destruction in the functions batadv_forw_packet_alloc() and batadv_forw_packet_free().
That way there is less complexity to wrap the head around when freeing a forward packet.
For instance broadcast/ogm queue left counting will not need to be done with "manual" atomic_inc/dec calls anymore, which should greatly reduce the risk of having or reintroducing another queue-left-miscounting again.
--- bat_iv_ogm.c | 32 ++++-------------- send.c | 103 +++++++++++++++++++++++++++++++++++++++++----------------- send.h | 5 +++ types.h | 1 + 4 files changed, 86 insertions(+), 55 deletions(-)
bat_iv_ogm.c +6 -26
@@ -427,26 +427,13 @@ static void batadv_iv_ogm_aggregate_new(const unsigned char *packet_buff,struct batadv_forw_packet *forw_packet_aggr;unsigned char *skb_buff;unsigned int skb_size; +atomic_t *queue_left = own_packet ?NULL : &bat_priv->batman_queue_left; -if (!atomic_inc_not_zero (&if_incoming->refcount)) + forw_packet_aggr =batadv_forw_packet_alloc (if_incoming, queue_left, + bat_priv); +if (!forw_packet_aggr)return ; -/* own packet should always be scheduled */ -if (!own_packet) { -if (!batadv_atomic_dec_not_zero (&bat_priv->batman_queue_left)) { -batadv_dbg (BATADV_DBG_BATMAN , bat_priv, -"batman packet queue full\n" ); - goto out; - } - } - - forw_packet_aggr =kmalloc (sizeof (*forw_packet_aggr),GFP_ATOMIC ); -if (!forw_packet_aggr) { -if (!own_packet) -atomic_inc (&bat_priv->batman_queue_left); - goto out; - } -if ((atomic_read (&bat_priv->aggregated_ogms)) && (packet_len <BATADV_MAX_AGGREGATION_BYTES )) skb_size =BATADV_MAX_AGGREGATION_BYTES ;
@@ -457,10 +444,8 @@ static void batadv_iv_ogm_aggregate_new(const unsigned char *packet_buff, forw_packet_aggr->skb =dev_alloc_skb (skb_size);if (!forw_packet_aggr->skb) { -if (!own_packet) -atomic_inc (&bat_priv->batman_queue_left); -kfree (forw_packet_aggr); - goto out; +batadv_forw_packet_free (forw_packet_aggr); +return ; }skb_reserve (forw_packet_aggr->skb,ETH_HLEN +NET_IP_ALIGN );
@@ -471,7 +456,6 @@ static void batadv_iv_ogm_aggregate_new(const unsigned char *packet_buff,memcpy (skb_buff, packet_buff, packet_len); forw_packet_aggr->own = own_packet; - forw_packet_aggr->if_incoming = if_incoming; forw_packet_aggr->num_packets =0 ; forw_packet_aggr->direct_link_flags =BATADV_NO_FLAGS ; forw_packet_aggr->send_time = send_time;
@@ -491,10 +475,6 @@ static void batadv_iv_ogm_aggregate_new(const unsigned char *packet_buff,queue_delayed_work (batadv_event_workqueue, &forw_packet_aggr->delayed_work, send_time - jiffies); - -return ; -out: -batadv_hardif_free_ref (if_incoming); }/* aggregate a new packet into the existing ogm packet */
send.c +74 -29
165 lines not on this page · raw message
send.h +5 -0
@@ -27,6 +27,11 @@ bool batadv_send_skb_to_orig(struct sk_buff *skb,struct batadv_orig_node *orig_node, struct batadv_hard_iface *recv_if);void batadv_schedule_bat_ogm (struct batadv_hard_iface *hard_iface); +struct batadv_forw_packet *batadv_forw_packet_alloc ( +struct batadv_hard_iface *if_incoming, +atomic_t *queue_left, +struct batadv_priv *bat_priv); +void batadv_forw_packet_free (struct batadv_forw_packet *forw_packet);int batadv_add_bcast_packet_to_list (struct batadv_priv *bat_priv,const struct sk_buff *skb,unsigned long delay);
types.h +1 -0
@@ -863,6 +863,7 @@ struct batadv_forw_packet {uint8_t num_packets;struct delayed_work delayed_work;struct batadv_hard_iface *if_incoming; +atomic_t *queue_left; };/**
Signature · 1 line
1.7.10.4
Re: [B.A.T.M.A.N.] [PATCH 2/2] batman-adv: Refactor forward packet creation/destruction
Quoted 6 lines · On Wednesday 17 April 2013 23:54:33 Linus Lüssing wrote:
This patch abstracts the forward packet creation and destruction in the functions
batadv_forw_packet_alloc()andbatadv_forw_packet_free().That way there is less complexity to wrap the head around when freeing a forward packet.
Quoted 7 lines
For instance broadcast/ogm queue left counting will not need to be done with "manual" atomic_inc/dec calls anymore, which should greatly reduce the risk of having or reintroducing another queue-left-miscounting again.
Signed-off-by Linus Lüssing <[email protected]>---
It looks like this patch doesn't apply anymore. Can you please resent it or mark it correctly in patchwork [1].
Thanks,
Sven
Re: [B.A.T.M.A.N.] [PATCH 2/2] batman-adv: Refactor forward packet creation/destruction
On Thursday 10 March 2016 18:08:27 Sven Eckelmann wrote:
[...]
Quoted 7 lines
It looks like this patch doesn't apply anymore. Can you please resent it or mark it correctly in patchwork [1].
Thanks,
Sven
What is the state here?
Kind regards,
Sven
PATCH v2 1/2 batman-adv: Fix broadcast/ogm queue limit on a removed interface
In reply to Linus LüssingWhen removing a single interface while a broadcast or ogm packet is still pending then we will free the forward packet without releasing the queue slots again.
This patch is supposed to fix this issue.
--- send.c | 6 ++++++ 1 file changed, 6 insertions(+)
* v2: Fix summary line of commit message: This issue can happen for both OGMs and broadcast packets.
send.c +6 -0
@@ -356,6 +356,9 @@ batadv_purge_outstanding_packets(struct batadv_priv *bat_priv,if (pending) {hlist_del (&forw_packet->list); +if (!forw_packet->own) +atomic_inc (&bat_priv->bcast_queue_left); +batadv_forw_packet_free (forw_packet); } }
@@ -382,6 +385,9 @@ batadv_purge_outstanding_packets(struct batadv_priv *bat_priv,if (pending) {hlist_del (&forw_packet->list); +if (!forw_packet->own) +atomic_inc (&bat_priv->batman_queue_left); +batadv_forw_packet_free (forw_packet); } }
Signature · 1 line
1.7.10.4
Re: [B.A.T.M.A.N.] [PATCHv2 1/2] batman-adv: Fix broadcast/ogm queue limit on a removed interface
Quoted 8 lines · On Thursday 18 April 2013 00:08:36 Linus Lüssing wrote:
When removing a single interface while a broadcast or ogm packet is still pending then we will free the forward packet without releasing the queue slots again.
This patch is supposed to fix this issue.
Signed-off-by Linus Lüssing <[email protected]>---
You are talking a batman-adv interface when it contains multiple slave interfaces, right? So batadv_purge_outstanding_packets would be called in batadv_hardif_disable_interface and not in batadv_mesh_free (which is only done when the batX interface will be removed).
This at least sounds legit and I cannot find where else this imbalance would be fixed.
Kind regards,
Sven
PS: This patch only requires the path change from / to /net/batman-adv/ to apply via git-am