Staging: batman-adv: receive packets directly using skbs

This patch removes the (ugly and racy) packet receiving thread and the
kernel socket usage. Instead, packets are received directly by registering
the ethernet type and handling skbs instead of self-allocated buffers.

Some consequences and comments:

 * we don't copy the payload data when forwarding/sending/receiving data
   anymore. This should boost performance.
 * packets from/to different interfaces can be (theoretically) processed
   simultaneously. Only the big originator hash lock might be in the way.
 * no more polling or sleeping/wakeup/scheduling issues when receiving
   packets
 * this might introduce new race conditions.
 * aggregation and vis code still use packet buffers and are not (yet)
   converted.
 * all spinlocks were converted to irqsave/restore versions to solve
   some lifelock issues when preempted. This might be overkill, some
   of these locks might be reverted later.
 * skb copies are only done if neccesary to avoid overhead

performance differences:

 * we made some "benchmarks" with intel laptops.
 * bandwidth on Gigabit Ethernet increased from ~500 MBit/s to ~920 MBit/s
 * ping latency decresed from ~2ms to ~0.2 ms

I did some tests on my 9 node qemu environment and could confirm that
usual sending/receiving, forwarding, vis, batctl ping etc works.

Signed-off-by: Simon Wunderlich <siwu@hrz.tu-chemnitz.de>
Acked-by: Sven Eckelmann <sven.eckelmann@gmx.de>
Acked-by: Marek Lindner <lindner_marek@yahoo.de>
Acked-by: Linus Lüssing <linus.luessing@web.de>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

1 files changed, 38 insertions, 38 deletions

diff --git a/drivers/staging/batman-adv/soft-interface.c b/drivers/staging/batman-adv/soft-interface.c
index 168a4e195a1..8ae3483a625 100644
--- a/drivers/staging/batman-adv/soft-interface.c
+++ b/drivers/staging/batman-adv/soft-interface.c
@@ -34,7 +34,6 @@ static uint16_t bcast_seqno = 1; /* give own bcast messages seq numbers to avoid
                                   * broadcast storms */
 static int32_t skb_packets;
 static int32_t skb_bad_packets;
-static int32_t lock_dropped;
 
 unsigned char mainIfAddr[ETH_ALEN];
 static unsigned char mainIfAddr_default[ETH_ALEN];
@@ -67,12 +66,12 @@ int main_if_was_up(void)
         return (memcmp(mainIfAddr, mainIfAddr_default, ETH_ALEN) != 0 ? 1 : 0);
 }
 
-static int my_skb_push(struct sk_buff *skb, unsigned int len)
+int my_skb_push(struct sk_buff *skb, unsigned int len)
 {
         int result = 0;
 
         skb_packets++;
-        if (skb->data - len < skb->head) {
+        if (skb_headroom(skb) < len) {
                 skb_bad_packets++;
                 result = pskb_expand_head(skb, len, 0, GFP_ATOMIC);
 
@@ -169,7 +168,10 @@ int interface_tx(struct sk_buff *skb, struct net_device *dev)
         struct orig_node *orig_node;
         struct ethhdr *ethhdr = (struct ethhdr *)skb->data;
         struct bat_priv *priv = netdev_priv(dev);
+        struct batman_if *batman_if;
+        uint8_t dstaddr[6];
         int data_len = skb->len;
+        unsigned long flags;
 
         if (atomic_read(&module_state) != MODULE_ACTIVE)
                 goto dropped;
@@ -185,7 +187,6 @@ int interface_tx(struct sk_buff *skb, struct net_device *dev)
                         goto dropped;
 
                 bcast_packet = (struct bcast_packet *)skb->data;
-
                 bcast_packet->version = COMPAT_VERSION;
 
                 /* batman packet type: broadcast */
@@ -194,27 +195,21 @@ int interface_tx(struct sk_buff *skb, struct net_device *dev)
                 /* hw address of first interface is the orig mac because only
                  * this mac is known throughout the mesh */
                 memcpy(bcast_packet->orig, mainIfAddr, ETH_ALEN);
+
                 /* set broadcast sequence number */
                 bcast_packet->seqno = htons(bcast_seqno);
 
                 bcast_seqno++;
 
                 /* broadcast packet */
-                add_bcast_packet_to_list(skb->data, skb->len);
+                add_bcast_packet_to_list(skb);
+                /* a copy is stored in the bcast list, therefore removing
+                 * the original skb. */
+                kfree_skb(skb);
 
         /* unicast packet */
         } else {
-
-                /* simply spin_lock()ing can deadlock when the lock is already
-                 * hold. */
-                /* TODO: defer the work in a working queue instead of
-                 * dropping */
-                if (!spin_trylock(&orig_hash_lock)) {
-                        lock_dropped++;
-                        printk(KERN_WARNING "batman-adv:%d packets dropped because lock was hold\n", lock_dropped);
-                        goto dropped;
-                }
-
+                spin_lock_irqsave(&orig_hash_lock, flags);
                 /* get routing information */
                 orig_node = ((struct orig_node *)hash_find(orig_hash,
                                                            ethhdr->h_dest));
@@ -243,14 +238,17 @@ int interface_tx(struct sk_buff *skb, struct net_device *dev)
                         if (orig_node->batman_if->if_active != IF_ACTIVE)
                                 goto unlock;
 
-                        send_raw_packet(skb->data, skb->len,
-                                        orig_node->batman_if,
-                                        orig_node->router->addr);
+                        /* don't lock while sending the packets ... we therefore
+                         * copy the required data before sending */
+
+                        batman_if = orig_node->batman_if;
+                        memcpy(dstaddr, orig_node->router->addr, ETH_ALEN);
+                        spin_unlock_irqrestore(&orig_hash_lock, flags);
+
+                        send_skb_packet(skb, batman_if, dstaddr);
                 } else {
                         goto unlock;
                 }
-
-                spin_unlock(&orig_hash_lock);
         }
 
         priv->stats.tx_packets++;
@@ -258,42 +256,44 @@ int interface_tx(struct sk_buff *skb, struct net_device *dev)
         goto end;
 
 unlock:
-        spin_unlock(&orig_hash_lock);
+        spin_unlock_irqrestore(&orig_hash_lock, flags);
 dropped:
         priv->stats.tx_dropped++;
 end:
-        kfree_skb(skb);
         return 0;
 }
 
-void interface_rx(struct net_device *dev, void *packet, int packet_len)
+void interface_rx(struct sk_buff *skb, int hdr_size)
 {
-        struct sk_buff *skb;
+        struct net_device *dev = soft_device;
         struct bat_priv *priv = netdev_priv(dev);
 
-        skb = dev_alloc_skb(packet_len);
-
-        if (!skb) {
-                priv->stats.rx_dropped++;
-                goto out;
+        /* check if enough space is available for pulling, and pull */
+        if (!pskb_may_pull(skb, hdr_size)) {
+                kfree_skb(skb);
+                return;
         }
+        skb_pull_rcsum(skb, hdr_size);
+/*      skb_set_mac_header(skb, -sizeof(struct ethhdr));*/
 
-        memcpy(skb_put(skb, packet_len), packet, packet_len);
-
-        /* Write metadata, and then pass to the receive level */
         skb->dev = dev;
         skb->protocol = eth_type_trans(skb, dev);
-        skb->ip_summed = CHECKSUM_UNNECESSARY;
+
+        /* should not be neccesary anymore as we use skb_pull_rcsum()
+         * TODO: please verify this and remove this TODO
+         * -- Dec 21st 2009, Simon Wunderlich */
+
+/*      skb->ip_summed = CHECKSUM_UNNECESSARY;*/
+
+        /* TODO: set skb->pkt_type to PACKET_BROADCAST, PACKET_MULTICAST,
+         * PACKET_OTHERHOST or PACKET_HOST */
 
         priv->stats.rx_packets++;
-        priv->stats.rx_bytes += packet_len;
+        priv->stats.rx_bytes += skb->len;
 
         dev->last_rx = jiffies;
 
         netif_rx(skb);
-
-out:
-        return;
 }
 
 /* ethtool */