[2/2] batman-adv: Use own timer for multicast TT and TVLV updates

Message ID 1469137112-24094-2-git-send-email-linus.luessing@c0d3.blue (mailing list archive)
State Superseded, archived
Delegated to: Marek Lindner
Headers

Commit Message

Linus Lüssing July 21, 2016, 9:38 p.m. UTC
  Instead of latching onto the OGM period, this patch introduces a worker
dedicated to multicast TT and TVLV updates.

The reasoning is, that upon roaming especially the translation table
should be updated timely to minimize connectivity issues.

With BATMAN V, the idea is to greatly increase the OGM interval to
reduce overhead. Unfortunately, right now this could lead to
a bad user experience if multicast traffic is involved.

Therefore this patch introduces a fixed 500ms update interval for
multicast TT entries and the multicast TVLV.

Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
---

Changes in RFC -> non-RFC:

* Move INIT_DELAYED_WORK() out of batadv_mcast_start_timer()
* lockdep fixes (fixes a lockdep warning):
  * Remove spinlock for tt.commit_lock in batadv_mcast_free()
  * Replace lockdep with WARN_ON(delayed_work_pending(...))
  -> we do not rely on tt-commit lock anymore
  * Clarify mla_list protection in kernel docs
* Now runtime tested
* Checked bat_events workqueue CPU usage:
  -> acc. to top, @500 bridged in mcast listeners, kworker/u2:0
     increased from ~0.3% to ~1.0% in kvm instance
     (bat_events kept slumbering at 0.0%)

 net/batman-adv/main.h              |  1 +
 net/batman-adv/multicast.c         | 54 +++++++++++++++++++++++++++++++++-----
 net/batman-adv/multicast.h         |  6 -----
 net/batman-adv/translation-table.c |  4 ---
 net/batman-adv/types.h             |  4 ++-
 5 files changed, 52 insertions(+), 17 deletions(-)
  

Comments

Linus Lüssing July 21, 2016, 9:53 p.m. UTC | #1
On Thu, Jul 21, 2016 at 11:38:32PM +0200, Linus Lüssing wrote:
> Instead of latching onto the OGM period, this patch introduces a worker
> dedicated to multicast TT and TVLV updates.
> 
> The reasoning is, that upon roaming especially the translation table
> should be updated timely to minimize connectivity issues.
> 
> With BATMAN V, the idea is to greatly increase the OGM interval to
> reduce overhead. Unfortunately, right now this could lead to
> a bad user experience if multicast traffic is involved.
> 
> Therefore this patch introduces a fixed 500ms update interval for
> multicast TT entries and the multicast TVLV.

PS: And as a side-effect, removes the funny locking dependency
with the TT commit lock. Had still been nagging me, this tight
coupling between MC and TT.
  
Sven Eckelmann July 21, 2016, 9:59 p.m. UTC | #2
On Donnerstag, 21. Juli 2016 23:38:32 CEST Linus Lüssing wrote:
> -       struct hlist_head mla_list;
> +       struct hlist_head mla_list; /* protected by mcast.work */

Protected by a workqueue item? How?

Kind regards,
	Sven
  
Sven Eckelmann July 21, 2016, 10:19 p.m. UTC | #3
On Donnerstag, 21. Juli 2016 23:59:30 CEST Sven Eckelmann wrote:
> On Donnerstag, 21. Juli 2016 23:38:32 CEST Linus Lüssing wrote:
> > -       struct hlist_head mla_list;
> > +       struct hlist_head mla_list; /* protected by mcast.work */
> 
> Protected by a workqueue item? How?

Looks like you wanted to say here that it is only modified inside the worker 
queue (via the mcast.work delayed workqueue item) - unless it is not done this 
way like in batadv_mcast_free after cancel_delayed_work_sync.

Kind regards,
	Sven
  
Linus Lüssing Aug. 1, 2016, 8:57 p.m. UTC | #4
On Fri, Jul 22, 2016 at 12:19:54AM +0200, Sven Eckelmann wrote:
> On Donnerstag, 21. Juli 2016 23:59:30 CEST Sven Eckelmann wrote:
> > On Donnerstag, 21. Juli 2016 23:38:32 CEST Linus Lüssing wrote:
> > > -       struct hlist_head mla_list;
> > > +       struct hlist_head mla_list; /* protected by mcast.work */
> > 
> > Protected by a workqueue item? How?
> 
> Looks like you wanted to say here that it is only modified inside the worker 
> queue (via the mcast.work delayed workqueue item) - unless it is not done this 
> way like in batadv_mcast_free after cancel_delayed_work_sync.

Exactly :). I tried to "summarize" that a little to make it fit as an
inline comment ;). (although there are some further kerneldoc remarks on
the functions using it)

Is it ok now or do you think I should add something like your
explanation somewhere else?
  
Sven Eckelmann Aug. 2, 2016, 7:44 a.m. UTC | #5
On Montag, 1. August 2016 22:57:16 CEST Linus Lüssing wrote:
> On Fri, Jul 22, 2016 at 12:19:54AM +0200, Sven Eckelmann wrote:
> > On Donnerstag, 21. Juli 2016 23:59:30 CEST Sven Eckelmann wrote:
> > > On Donnerstag, 21. Juli 2016 23:38:32 CEST Linus Lüssing wrote:
> > > > -       struct hlist_head mla_list;
> > > > +       struct hlist_head mla_list; /* protected by mcast.work */
> > > 
> > > Protected by a workqueue item? How?
> > 
> > Looks like you wanted to say here that it is only modified inside the worker 
> > queue (via the mcast.work delayed workqueue item) - unless it is not done this 
> > way like in batadv_mcast_free after cancel_delayed_work_sync.
> 
> Exactly :). I tried to "summarize" that a little to make it fit as an
> inline comment ;). (although there are some further kerneldoc remarks on
> the functions using it)
> 
> Is it ok now or do you think I should add something like your
> explanation somewhere else?

I personally don't like the statement "protected by mcast.work". Maybe you can
add something like "see ...." to reference the functions/kernel-doc describing
why it "protects" it.

Kind regards,
	Sven
  

Patch

diff --git a/net/batman-adv/main.h b/net/batman-adv/main.h
index 06a8608..edb8427 100644
--- a/net/batman-adv/main.h
+++ b/net/batman-adv/main.h
@@ -48,6 +48,7 @@ 
 #define BATADV_TT_CLIENT_TEMP_TIMEOUT 600000 /* in milliseconds */
 #define BATADV_TT_WORK_PERIOD 5000 /* 5 seconds */
 #define BATADV_ORIG_WORK_PERIOD 1000 /* 1 second */
+#define BATADV_MCAST_WORK_PERIOD 500 /* 0.5 seconds */
 #define BATADV_DAT_ENTRY_TIMEOUT (5 * 60000) /* 5 mins in milliseconds */
 /* sliding packet range of received originator messages in sequence numbers
  * (should be a multiple of our word size)
diff --git a/net/batman-adv/multicast.c b/net/batman-adv/multicast.c
index 9c3dd3e..af037f6 100644
--- a/net/batman-adv/multicast.c
+++ b/net/batman-adv/multicast.c
@@ -60,6 +60,18 @@ 
 #include "translation-table.h"
 #include "tvlv.h"
 
+static void batadv_mcast_mla_update(struct work_struct *work);
+
+/**
+ * batadv_mcast_start_timer - schedule the multicast periodic worker
+ * @bat_priv: the bat priv with all the soft interface information
+ */
+static void batadv_mcast_start_timer(struct batadv_priv *bat_priv)
+{
+	queue_delayed_work(batadv_event_workqueue, &bat_priv->mcast.work,
+			   msecs_to_jiffies(BATADV_MCAST_WORK_PERIOD));
+}
+
 /**
  * batadv_mcast_get_bridge - get the bridge on top of the softif if it exists
  * @soft_iface: netdev struct of the mesh interface
@@ -255,6 +267,8 @@  static void batadv_mcast_mla_list_free(struct hlist_head *mcast_list)
  * translation table except the ones listed in the given mcast_list.
  *
  * If mcast_list is NULL then all are retracted.
+ *
+ * Do not call outside of the mcast worker! (or cancel mcast worker first)
  */
 static void batadv_mcast_mla_tt_retract(struct batadv_priv *bat_priv,
 					struct hlist_head *mcast_list)
@@ -262,7 +276,7 @@  static void batadv_mcast_mla_tt_retract(struct batadv_priv *bat_priv,
 	struct batadv_hw_addr *mcast_entry;
 	struct hlist_node *tmp;
 
-	lockdep_assert_held(&bat_priv->tt.commit_lock);
+	WARN_ON(delayed_work_pending(&bat_priv->mcast.work));
 
 	hlist_for_each_entry_safe(mcast_entry, tmp, &bat_priv->mcast.mla_list,
 				  list) {
@@ -287,6 +301,8 @@  static void batadv_mcast_mla_tt_retract(struct batadv_priv *bat_priv,
  *
  * Adds multicast listener announcements from the given mcast_list to the
  * translation table if they have not been added yet.
+ *
+ * Do not call outside of the mcast worker! (or cancel mcast worker first)
  */
 static void batadv_mcast_mla_tt_add(struct batadv_priv *bat_priv,
 				    struct hlist_head *mcast_list)
@@ -294,7 +310,7 @@  static void batadv_mcast_mla_tt_add(struct batadv_priv *bat_priv,
 	struct batadv_hw_addr *mcast_entry;
 	struct hlist_node *tmp;
 
-	lockdep_assert_held(&bat_priv->tt.commit_lock);
+	WARN_ON(delayed_work_pending(&bat_priv->mcast.work));
 
 	if (!mcast_list)
 		return;
@@ -528,13 +544,13 @@  update:
 }
 
 /**
- * batadv_mcast_mla_update - update the own MLAs
+ * __batadv_mcast_mla_update - update the own MLAs
  * @bat_priv: the bat priv with all the soft interface information
  *
  * Updates the own multicast listener announcements in the translation
  * table as well as the own, announced multicast tvlv container.
  */
-void batadv_mcast_mla_update(struct batadv_priv *bat_priv)
+static void __batadv_mcast_mla_update(struct batadv_priv *bat_priv)
 {
 	struct net_device *soft_iface = bat_priv->soft_iface;
 	struct hlist_head mcast_list = HLIST_HEAD_INIT;
@@ -560,6 +576,29 @@  out:
 }
 
 /**
+ * batadv_mcast_mla_update - update the own MLAs
+ * @bat_priv: the bat priv with all the soft interface information
+ *
+ * Updates the own multicast listener announcements in the translation
+ * table as well as the own, announced multicast tvlv container.
+ *
+ * In the end, reschedules the work timer.
+ */
+static void batadv_mcast_mla_update(struct work_struct *work)
+{
+	struct delayed_work *delayed_work;
+	struct batadv_priv_mcast *priv_mcast;
+	struct batadv_priv *bat_priv;
+
+	delayed_work = to_delayed_work(work);
+	priv_mcast = container_of(delayed_work, struct batadv_priv_mcast, work);
+	bat_priv = container_of(priv_mcast, struct batadv_priv, mcast);
+
+	__batadv_mcast_mla_update(bat_priv);
+	batadv_mcast_start_timer(bat_priv);
+}
+
+/**
  * batadv_mcast_is_report_ipv4 - check for IGMP reports
  * @skb: the ethernet frame destined for the mesh
  *
@@ -1128,6 +1167,9 @@  void batadv_mcast_init(struct batadv_priv *bat_priv)
 	batadv_tvlv_handler_register(bat_priv, batadv_mcast_tvlv_ogm_handler,
 				     NULL, BATADV_TVLV_MCAST, 2,
 				     BATADV_TVLV_HANDLER_OGM_CIFNOTFND);
+
+	INIT_DELAYED_WORK(&bat_priv->mcast.work, batadv_mcast_mla_update);
+	batadv_mcast_start_timer(bat_priv);
 }
 
 /**
@@ -1237,12 +1279,12 @@  int batadv_mcast_flags_seq_print_text(struct seq_file *seq, void *offset)
  */
 void batadv_mcast_free(struct batadv_priv *bat_priv)
 {
+	cancel_delayed_work_sync(&bat_priv->mcast.work);
+
 	batadv_tvlv_container_unregister(bat_priv, BATADV_TVLV_MCAST, 2);
 	batadv_tvlv_handler_unregister(bat_priv, BATADV_TVLV_MCAST, 2);
 
-	spin_lock_bh(&bat_priv->tt.commit_lock);
 	batadv_mcast_mla_tt_retract(bat_priv, NULL);
-	spin_unlock_bh(&bat_priv->tt.commit_lock);
 }
 
 /**
diff --git a/net/batman-adv/multicast.h b/net/batman-adv/multicast.h
index 1fb00ba..2cddaf5 100644
--- a/net/batman-adv/multicast.h
+++ b/net/batman-adv/multicast.h
@@ -39,8 +39,6 @@  enum batadv_forw_mode {
 
 #ifdef CONFIG_BATMAN_ADV_MCAST
 
-void batadv_mcast_mla_update(struct batadv_priv *bat_priv);
-
 enum batadv_forw_mode
 batadv_mcast_forw_mode(struct batadv_priv *bat_priv, struct sk_buff *skb,
 		       struct batadv_orig_node **mcast_single_orig);
@@ -55,10 +53,6 @@  void batadv_mcast_purge_orig(struct batadv_orig_node *orig_node);
 
 #else
 
-static inline void batadv_mcast_mla_update(struct batadv_priv *bat_priv)
-{
-}
-
 static inline enum batadv_forw_mode
 batadv_mcast_forw_mode(struct batadv_priv *bat_priv, struct sk_buff *skb,
 		       struct batadv_orig_node **mcast_single_orig)
diff --git a/net/batman-adv/translation-table.c b/net/batman-adv/translation-table.c
index d94e298..d9ec32d 100644
--- a/net/batman-adv/translation-table.c
+++ b/net/batman-adv/translation-table.c
@@ -56,7 +56,6 @@ 
 #include "hard-interface.h"
 #include "hash.h"
 #include "log.h"
-#include "multicast.h"
 #include "netlink.h"
 #include "originator.h"
 #include "packet.h"
@@ -3791,9 +3790,6 @@  static void batadv_tt_local_commit_changes_nolock(struct batadv_priv *bat_priv)
 {
 	lockdep_assert_held(&bat_priv->tt.commit_lock);
 
-	/* Update multicast addresses in local translation table */
-	batadv_mcast_mla_update(bat_priv);
-
 	if (atomic_read(&bat_priv->tt.local_changes) < 1) {
 		if (!batadv_atomic_dec_not_zero(&bat_priv->tt.ogm_append_cnt))
 			batadv_tt_tvlv_container_update(bat_priv);
diff --git a/net/batman-adv/types.h b/net/batman-adv/types.h
index b5f01a3..90e4006 100644
--- a/net/batman-adv/types.h
+++ b/net/batman-adv/types.h
@@ -785,9 +785,10 @@  struct batadv_mcast_querier_state {
  * @num_want_all_ipv6: counter for items in want_all_ipv6_list
  * @want_lists_lock: lock for protecting modifications to mcast want lists
  *  (traversals are rcu-locked)
+ * @work: work queue callback item for multicast TT and TVLV updates
  */
 struct batadv_priv_mcast {
-	struct hlist_head mla_list;
+	struct hlist_head mla_list; /* protected by mcast.work */
 	struct hlist_head want_all_unsnoopables_list;
 	struct hlist_head want_all_ipv4_list;
 	struct hlist_head want_all_ipv6_list;
@@ -802,6 +803,7 @@  struct batadv_priv_mcast {
 	atomic_t num_want_all_ipv6;
 	/* protects want_all_{unsnoopables,ipv4,ipv6}_list */
 	spinlock_t want_lists_lock;
+	struct delayed_work work;
 };
 #endif