linux-2.6-525-sknid-elevator.patch

   1 diff -Nurb linux-2.6.22-510/net/core/dev.c linux-2.6.22-520/net/core/dev.c
   2 --- linux-2.6.22-510/net/core/dev.c     2008-06-06 17:07:48.000000000 -0400
   3 +++ linux-2.6.22-520/net/core/dev.c     2008-06-06 17:07:56.000000000 -0400
   4 @@ -1803,6 +1803,7 @@
   5   * the ingress scheduler, you just cant add policies on ingress.
   6   *
   7   */
   8 +
   9  static int ing_filter(struct sk_buff *skb)
  10  {
  11         struct Qdisc *q;
  12 @@ -1832,13 +1833,20 @@
  13  }
  14  #endif
  15
  16 +/* The code already makes the assumption that packet handlers run
  17 + * sequentially on the same CPU. -Sapan */
  18 +DEFINE_PER_CPU(int, sknid_elevator);
  19 +
  20  int netif_receive_skb(struct sk_buff *skb)
  21  {
  22         struct packet_type *ptype, *pt_prev;
  23         struct net_device *orig_dev;
  24         int ret = NET_RX_DROP;
  25 +       int *cur_elevator=&__get_cpu_var(sknid_elevator);
  26         __be16 type;
  27
  28 +       *cur_elevator = 0;
  29 +
  30         /* if we've gotten here through NAPI, check netpoll */
  31         if (skb->dev->poll && netpoll_rx(skb))
  32                 return NET_RX_DROP;
  33 @@ -1873,8 +1881,9 @@
  34
  35         list_for_each_entry_rcu(ptype, &ptype_all, list) {
  36                 if (!ptype->dev || ptype->dev == skb->dev) {
  37 -                       if (pt_prev)
  38 +                       if (pt_prev) {
  39                                 ret = deliver_skb(skb, pt_prev, orig_dev);
  40 +                       }
  41                         pt_prev = ptype;
  42                 }
  43         }
  44 @@ -1912,8 +1921,22 @@
  45                 }
  46         }
  47
  48 +       /* We don't want the packet handlers to throw the packet away
  49 +        * if we want the taps to treat it again - Sapan */
  50 +       if (*cur_elevator) {
  51 +               atomic_inc(&skb->users);
  52 +       }
  53 +
  54         if (pt_prev) {
  55                 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
  56 +                       if (*cur_elevator > 0) {
  57 +                               skb->skb_tag = *cur_elevator;
  58 +                               list_for_each_entry_rcu(ptype, &ptype_all, list) {
  59 +                                       if (!ptype->dev || ptype->dev == skb->dev) {
  60 +                                                       ret = deliver_skb(skb, ptype, orig_dev);
  61 +                                       }
  62 +                               }
  63 +                       }
  64         } else {
  65                 kfree_skb(skb);
  66                 /* Jamal, now you will not able to escape explaining
  67 @@ -1922,6 +1945,13 @@
  68                 ret = NET_RX_DROP;
  69         }
  70
  71 +       if (*cur_elevator) {
  72 +               /* We have a packet */
  73 +               kfree_skb(skb);
  74 +       }
  75 +
  76 +       *cur_elevator=0;
  77 +
  78  out:
  79         rcu_read_unlock();
  80         return ret;
  81 @@ -3780,6 +3810,7 @@
  82  EXPORT_SYMBOL(net_enable_timestamp);
  83  EXPORT_SYMBOL(net_disable_timestamp);
  84  EXPORT_SYMBOL(dev_get_flags);
  85 +EXPORT_PER_CPU_SYMBOL(sknid_elevator);
  86
  87  #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
  88  EXPORT_SYMBOL(br_handle_frame_hook);
  89 diff -Nurb linux-2.6.22-510/net/packet/af_packet.c linux-2.6.22-520/net/packet/af_packet.c
  90 --- linux-2.6.22-510/net/packet/af_packet.c     2007-07-08 19:32:17.000000000 -0400
  91 +++ linux-2.6.22-520/net/packet/af_packet.c     2008-06-07 18:30:41.000000000 -0400
  92 @@ -78,6 +78,7 @@
  93  #include <linux/poll.h>
  94  #include <linux/module.h>
  95  #include <linux/init.h>
  96 +#include <linux/vs_network.h>
  97
  98  #ifdef CONFIG_INET
  99  #include <net/inet_common.h>
 100 @@ -246,10 +247,13 @@
 101
 102  static const struct proto_ops packet_ops_spkt;
 103
 104 +extern DEFINE_PER_CPU(int, sknid_elevator);
 105  static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,  struct packet_type *pt, struct net_device *orig_dev)
 106  {
 107         struct sock *sk;
 108         struct sockaddr_pkt *spkt;
 109 +       int tag = skb->skb_tag;
 110 +       int *elevator=&__get_cpu_var(sknid_elevator);
 111
 112         /*
 113          *      When we registered the protocol we saved the socket in the data
 114 @@ -269,6 +273,22 @@
 115          *      so that this procedure is noop.
 116          */
 117
 118 +       /*
 119 +        * (18:05:41) daniel_hozac: where?
 120 +        * (18:05:58) daniel_hozac: we already have filters on PF_PACKET, don't we?
 121 +        * (18:05:58) er: in packet_rcv_skpt
 122 +        * (18:07:33) daniel_hozac: oh, that's evil.
 123 +        */
 124 +
 125 +       if (sk->sk_nx_info && !(tag == 1 || sk->sk_nid == tag)) {
 126 +               *elevator=-2;
 127 +               goto out;
 128 +       }
 129 +       else if (!sk->sk_nx_info && *elevator) {
 130 +               /* Root has already seen this packet */
 131 +               goto out;
 132 +       }
 133 +
 134         if (skb->pkt_type == PACKET_LOOPBACK)
 135                 goto out;
 136
 137 @@ -324,6 +344,9 @@
 138         __be16 proto=0;
 139         int err;
 140
 141 +       if (!nx_capable(CAP_NET_RAW, NXC_RAW_SEND))
 142 +               return -EPERM;
 143 +
 144         /*
 145          *      Get and verify the address.
 146          */
 147 @@ -420,6 +443,17 @@
 148                                       unsigned int res)
 149  {
 150         struct sk_filter *filter;
 151 +       int tag = skb->skb_tag;
 152 +       int *elevator=&__get_cpu_var(sknid_elevator);
 153 +
 154 +       if (sk->sk_nx_info && !(tag == 1 || sk->sk_nid == tag)) {
 155 +               *elevator=-2;
 156 +               return 0;
 157 +       }
 158 +       else if (!sk->sk_nx_info && *elevator) {
 159 +               /* Root has already seen this packet */
 160 +               return 0;
 161 +       }
 162
 163         rcu_read_lock_bh();
 164         filter = rcu_dereference(sk->sk_filter);
 165 @@ -711,6 +745,9 @@
 166         unsigned char *addr;
 167         int ifindex, err, reserve = 0;
 168
 169 +       if (!nx_capable(CAP_NET_RAW, NXC_RAW_SEND))
 170 +               return -EPERM;
 171 +
 172         /*
 173          *      Get and verify the address.
 174          */
 175 @@ -984,8 +1021,9 @@
 176         __be16 proto = (__force __be16)protocol; /* weird, but documented */
 177         int err;
 178
 179 -       if (!capable(CAP_NET_RAW))
 180 +       if (!nx_capable(CAP_NET_RAW, NXC_RAW_SOCKET))
 181                 return -EPERM;
 182 +
 183         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
 184             sock->type != SOCK_PACKET)
 185                 return -ESOCKTNOSUPPORT;