linux-2.6-525-sknid-elevator.patch

   1 diff -Nurb linux-2.6.22-524/net/core/dev.c linux-2.6.22-525/net/core/dev.c
   2 --- linux-2.6.22-524/net/core/dev.c     2008-07-15 11:39:32.000000000 -0400
   3 +++ linux-2.6.22-525/net/core/dev.c     2008-07-21 16:22:33.000000000 -0400
   4 @@ -1131,7 +1131,7 @@
   5                 if ((ptype->dev == dev || !ptype->dev) &&
   6                     (ptype->af_packet_priv == NULL ||
   7                      (struct sock *)ptype->af_packet_priv != skb->sk)) {
   8 -                       struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
   9 +                       struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
  10                         if (!skb2)
  11                                 break;
  12
  13 @@ -1803,6 +1803,7 @@
  14   * the ingress scheduler, you just cant add policies on ingress.
  15   *
  16   */
  17 +
  18  static int ing_filter(struct sk_buff *skb)
  19  {
  20         struct Qdisc *q;
  21 @@ -1832,13 +1833,21 @@
  22  }
  23  #endif
  24
  25 +/* The code already makes the assumption that packet handlers run
  26 + * sequentially on the same CPU. -Sapan */
  27 +DEFINE_PER_CPU(int, sknid_elevator);
  28 +
  29  int netif_receive_skb(struct sk_buff *skb)
  30  {
  31         struct packet_type *ptype, *pt_prev;
  32         struct net_device *orig_dev;
  33         int ret = NET_RX_DROP;
  34 +       int *cur_elevator=&__get_cpu_var(sknid_elevator);
  35 +       struct sk_buff *skb2;
  36         __be16 type;
  37
  38 +       *cur_elevator = 0;
  39 +
  40         /* if we've gotten here through NAPI, check netpoll */
  41         if (skb->dev->poll && netpoll_rx(skb))
  42                 return NET_RX_DROP;
  43 @@ -1873,8 +1882,9 @@
  44
  45         list_for_each_entry_rcu(ptype, &ptype_all, list) {
  46                 if (!ptype->dev || ptype->dev == skb->dev) {
  47 -                       if (pt_prev)
  48 +                       if (pt_prev) {
  49                                 ret = deliver_skb(skb, pt_prev, orig_dev);
  50 +                       }
  51                         pt_prev = ptype;
  52                 }
  53         }
  54 @@ -1902,6 +1912,14 @@
  55         if (!skb)
  56                 goto out;
  57
  58 +       /* We don't want the packet handlers to throw the packet away
  59 +        * if we want the taps to treat it again - Sapan */
  60 +       if (*cur_elevator) {
  61 +               skb2 = skb_copy(skb,GFP_ATOMIC);
  62 +               if (!skb2) *cur_elevator=0;
  63 +       }
  64 +
  65 +
  66         type = skb->protocol;
  67         list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
  68                 if (ptype->type == type &&
  69 @@ -1914,6 +1932,7 @@
  70
  71         if (pt_prev) {
  72                 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
  73 +
  74         } else {
  75                 kfree_skb(skb);
  76                 /* Jamal, now you will not able to escape explaining
  77 @@ -1922,6 +1941,21 @@
  78                 ret = NET_RX_DROP;
  79         }
  80
  81 +       if (*cur_elevator>0) {
  82 +               skb2->skb_tag = *cur_elevator;
  83 +               list_for_each_entry_rcu(ptype, &ptype_all, list) {
  84 +                       if (!ptype->dev || ptype->dev == skb2->dev) {
  85 +                               ret = deliver_skb(skb2, ptype, orig_dev);
  86 +                       }
  87 +               }
  88 +       }
  89 +       if (*cur_elevator) {
  90 +               /* We have a packet */
  91 +               kfree_skb(skb2);
  92 +       }
  93 +
  94 +       *cur_elevator=0;
  95 +
  96  out:
  97         rcu_read_unlock();
  98         return ret;
  99 @@ -3780,6 +3814,7 @@
 100  EXPORT_SYMBOL(net_enable_timestamp);
 101  EXPORT_SYMBOL(net_disable_timestamp);
 102  EXPORT_SYMBOL(dev_get_flags);
 103 +EXPORT_PER_CPU_SYMBOL(sknid_elevator);
 104
 105  #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
 106  EXPORT_SYMBOL(br_handle_frame_hook);
 107 diff -Nurb linux-2.6.22-524/net/packet/af_packet.c linux-2.6.22-525/net/packet/af_packet.c
 108 --- linux-2.6.22-524/net/packet/af_packet.c     2007-07-08 19:32:17.000000000 -0400
 109 +++ linux-2.6.22-525/net/packet/af_packet.c     2008-07-15 11:40:11.000000000 -0400
 110 @@ -78,6 +78,7 @@
 111  #include <linux/poll.h>
 112  #include <linux/module.h>
 113  #include <linux/init.h>
 114 +#include <linux/vs_network.h>
 115
 116  #ifdef CONFIG_INET
 117  #include <net/inet_common.h>
 118 @@ -246,10 +247,13 @@
 119
 120  static const struct proto_ops packet_ops_spkt;
 121
 122 +extern DEFINE_PER_CPU(int, sknid_elevator);
 123  static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,  struct packet_type *pt, struct net_device *orig_dev)
 124  {
 125         struct sock *sk;
 126         struct sockaddr_pkt *spkt;
 127 +       int tag = skb->skb_tag;
 128 +       int *elevator=&__get_cpu_var(sknid_elevator);
 129
 130         /*
 131          *      When we registered the protocol we saved the socket in the data
 132 @@ -269,6 +273,22 @@
 133          *      so that this procedure is noop.
 134          */
 135
 136 +       /*
 137 +        * (18:05:41) daniel_hozac: where?
 138 +        * (18:05:58) daniel_hozac: we already have filters on PF_PACKET, don't we?
 139 +        * (18:05:58) er: in packet_rcv_skpt
 140 +        * (18:07:33) daniel_hozac: oh, that's evil.
 141 +        */
 142 +
 143 +       if (sk->sk_nx_info && !(tag == 1 || sk->sk_nid == tag)) {
 144 +               *elevator=-2;
 145 +               goto out;
 146 +       }
 147 +       else if (!sk->sk_nx_info && *elevator) {
 148 +               /* Root has already seen this packet */
 149 +               goto out;
 150 +       }
 151 +
 152         if (skb->pkt_type == PACKET_LOOPBACK)
 153                 goto out;
 154
 155 @@ -324,6 +344,9 @@
 156         __be16 proto=0;
 157         int err;
 158
 159 +       if (!nx_capable(CAP_NET_RAW, NXC_RAW_SEND))
 160 +               return -EPERM;
 161 +
 162         /*
 163          *      Get and verify the address.
 164          */
 165 @@ -420,6 +443,17 @@
 166                                       unsigned int res)
 167  {
 168         struct sk_filter *filter;
 169 +       int tag = skb->skb_tag;
 170 +       int *elevator=&__get_cpu_var(sknid_elevator);
 171 +
 172 +       if (sk->sk_nx_info && !(tag == 1 || sk->sk_nid == tag)) {
 173 +               *elevator=-2;
 174 +               return 0;
 175 +       }
 176 +       else if (!sk->sk_nx_info && *elevator) {
 177 +               /* Root has already seen this packet */
 178 +               return 0;
 179 +       }
 180
 181         rcu_read_lock_bh();
 182         filter = rcu_dereference(sk->sk_filter);
 183 @@ -711,6 +745,9 @@
 184         unsigned char *addr;
 185         int ifindex, err, reserve = 0;
 186
 187 +       if (!nx_capable(CAP_NET_RAW, NXC_RAW_SEND))
 188 +               return -EPERM;
 189 +
 190         /*
 191          *      Get and verify the address.
 192          */
 193 @@ -984,8 +1021,9 @@
 194         __be16 proto = (__force __be16)protocol; /* weird, but documented */
 195         int err;
 196
 197 -       if (!capable(CAP_NET_RAW))
 198 +       if (!nx_capable(CAP_NET_RAW, NXC_RAW_SOCKET))
 199                 return -EPERM;
 200 +
 201         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
 202             sock->type != SOCK_PACKET)
 203                 return -ESOCKTNOSUPPORT;