src/patches/suse-2.6.27.39/patches.fixes/fix-balance-alb-bonding-receive

   1 From: Jiri Pirko <jpirko@redhat.com>
   2 Date: Thu, 28 May 2009 01:05:00 +0000 (+0000)
   3 Subject: bonding: allow bond in mode balance-alb to work properly in bridge -try4.3
   4 Patch-mainline: v2.6.31-rc1
   5 Git-commit: 5d4e039b2cb1ca4de9774344ea7b61ad7fa1b0a1
   6 References: bnc#532443
   7
   8 bonding: allow bond in mode balance-alb to work properly in bridge -try4.3
   9
  10 [PATCH net-next] bonding: allow bond in mode balance-alb to work properly in bridge -try4.3
  11
  12 (updated)
  13 changes v4.2 -> v4.3
  14 - memcpy the address always, not just in case it differs from master->dev_addr
  15 - compare_ether_addr_64bits() is not used so there is no direct need to make new
  16   header file (I think it would be good to have bond stuff in separate file
  17   anyway).
  18
  19 changes v4.1 -> v4.2
  20 - use skb->pkt_type == PACKET_HOST compare rather then comparing skb dest addr
  21   against skb->dev->dev_addr
  22
  23 The problem is described in following bugzilla:
  24 https://bugzilla.redhat.com/show_bug.cgi?id=487763
  25
  26 Basically here's what's going on. In every mode, bonding interface uses the same
  27 mac address for all enslaved devices (except fail_over_mac). Only balance-alb
  28 will simultaneously use multiple MAC addresses across different slaves. When you
  29 put this kind of bond device into a bridge it will only add one of mac adresses
  30 into a hash list of mac addresses, say X. This mac address is marked as local.
  31 But this bonding interface also has mac address Y. Now then packet arrives with
  32 destination address Y, this address is not marked as local and the packed looks
  33 like it needs to be forwarded. This packet is then lost which is wrong.
  34
  35 Notice that interfaces can be added and removed from bond while it is in bridge.
  36
  37
  38 When the multiple addresses for bridge port approach failed to solve this issue
  39 due to STP I started to think other way to solve this. I returned to previous
  40 solution but tweaked one.
  41
  42 This patch solves the situation in the bonding without touching bridge code.
  43 For every incoming frame to bonding the destination address is compared to
  44 current address of the slave device from which tha packet came. If these two
  45 match destination address is replaced by mac address of the master. This address
  46 is known by bridge so it is delivered properly. Note that the comparsion is not
  47 made directly, it's used skb->pkt_type == PACKET_HOST instead. This is "set"
  48 previously in eth_type_trans().
  49
  50 I experimentally tried that this works as good as searching through the slave
  51 list (v4 of this patch).
  52
  53 Jirka
  54
  55 Signed-off-by: Jiri Pirko <jpirko@redhat.com>
  56 Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
  57 Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
  58 Signed-off-by: David S. Miller <davem@davemloft.net>
  59 Acked-by: Jeff Mahoney <jeffm@suse.com>
  60 ---
  61  include/linux/netdevice.h |   18 ++++++++++++++++++
  62  1 file changed, 18 insertions(+)
  63
  64 --- a/include/linux/netdevice.h
  65 +++ b/include/linux/netdevice.h
  66 @@ -1705,6 +1705,16 @@ static inline void netif_set_gso_max_siz
  67         dev->gso_max_size = size;
  68  }
  69
  70 +static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
  71 +                                             struct net_device *master)
  72 +{
  73 +       if (skb->pkt_type == PACKET_HOST) {
  74 +               u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
  75 +
  76 +               memcpy(dest, master->dev_addr, ETH_ALEN);
  77 +       }
  78 +}
  79 +
  80  /* On bonding slaves other than the currently active slave, suppress
  81   * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
  82   * ARP on active-backup slaves with arp_validate enabled.
  83 @@ -1718,6 +1728,14 @@ static inline int skb_bond_should_drop(s
  84                 if (master->priv_flags & IFF_MASTER_ARPMON)
  85                         dev->last_rx = jiffies;
  86
  87 +               if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
  88 +                       /* Do address unmangle. The local destination address
  89 +                        * will be always the one master has. Provides the right
  90 +                        * functionality in a bridge.
  91 +                        */
  92 +                       skb_bond_set_mac_by_master(skb, master);
  93 +               }
  94 +
  95                 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
  96                         if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
  97                             skb->protocol == __constant_htons(ETH_P_ARP))