drivers/net/ifb.c

   1 /* drivers/net/ifb.c:
   2
   3         The purpose of this driver is to provide a device that allows
   4         for sharing of resources:
   5
   6         1) qdiscs/policies that are per device as opposed to system wide.
   7         ifb allows for a device which can be redirected to thus providing
   8         an impression of sharing.
   9
  10         2) Allows for queueing incoming traffic for shaping instead of
  11         dropping.
  12
  13         The original concept is based on what is known as the IMQ
  14         driver initially written by Martin Devera, later rewritten
  15         by Patrick McHardy and then maintained by Andre Correa.
  16
  17         You need the tc action  mirror or redirect to feed this device
  18         packets.
  19
  20         This program is free software; you can redistribute it and/or
  21         modify it under the terms of the GNU General Public License
  22         as published by the Free Software Foundation; either version
  23         2 of the License, or (at your option) any later version.
  24
  25         Authors:        Jamal Hadi Salim (2005)
  26
  27 */
  28
  29
  30 #include <linux/module.h>
  31 #include <linux/kernel.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/etherdevice.h>
  34 #include <linux/init.h>
  35 #include <linux/moduleparam.h>
  36 #include <net/pkt_sched.h>
  37 #include <net/net_namespace.h>
  38
  39 #define TX_Q_LIMIT    32
  40 struct ifb_private {
  41         struct tasklet_struct   ifb_tasklet;
  42         int     tasklet_pending;
  43         struct sk_buff_head     rq;
  44         struct sk_buff_head     tq;
  45 };
  46
  47 static int numifbs = 2;
  48
  49 static void ri_tasklet(unsigned long dev);
  50 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev);
  51 static int ifb_open(struct net_device *dev);
  52 static int ifb_close(struct net_device *dev);
  53
  54 static void ri_tasklet(unsigned long dev)
  55 {
  56
  57         struct net_device *_dev = (struct net_device *)dev;
  58         struct ifb_private *dp = netdev_priv(_dev);
  59         struct net_device_stats *stats = &_dev->stats;
  60         struct netdev_queue *txq;
  61         struct sk_buff *skb;
  62
  63         txq = netdev_get_tx_queue(_dev, 0);
  64         if ((skb = skb_peek(&dp->tq)) == NULL) {
  65                 if (__netif_tx_trylock(txq)) {
  66                         skb_queue_splice_tail_init(&dp->rq, &dp->tq);
  67                         __netif_tx_unlock(txq);
  68                 } else {
  69                         /* reschedule */
  70                         goto resched;
  71                 }
  72         }
  73
  74         while ((skb = __skb_dequeue(&dp->tq)) != NULL) {
  75                 u32 from = G_TC_FROM(skb->tc_verd);
  76
  77                 skb->tc_verd = 0;
  78                 skb->tc_verd = SET_TC_NCLS(skb->tc_verd);
  79                 stats->tx_packets++;
  80                 stats->tx_bytes +=skb->len;
  81
  82                 rcu_read_lock();
  83                 skb->dev = dev_get_by_index_rcu(&init_net, skb->skb_iif);
  84                 if (!skb->dev) {
  85                         rcu_read_unlock();
  86                         dev_kfree_skb(skb);
  87                         stats->tx_dropped++;
  88                         if (skb_queue_len(&dp->tq) != 0)
  89                                 goto resched;
  90                         break;
  91                 }
  92                 rcu_read_unlock();
  93                 skb->skb_iif = _dev->ifindex;
  94
  95                 if (from & AT_EGRESS) {
  96                         dev_queue_xmit(skb);
  97                 } else if (from & AT_INGRESS) {
  98                         skb_pull(skb, skb->dev->hard_header_len);
  99                         netif_receive_skb(skb);
 100                 } else
 101                         BUG();
 102         }
 103
 104         if (__netif_tx_trylock(txq)) {
 105                 if ((skb = skb_peek(&dp->rq)) == NULL) {
 106                         dp->tasklet_pending = 0;
 107                         if (netif_queue_stopped(_dev))
 108                                 netif_wake_queue(_dev);
 109                 } else {
 110                         __netif_tx_unlock(txq);
 111                         goto resched;
 112                 }
 113                 __netif_tx_unlock(txq);
 114         } else {
 115 resched:
 116                 dp->tasklet_pending = 1;
 117                 tasklet_schedule(&dp->ifb_tasklet);
 118         }
 119
 120 }
 121
 122 static const struct net_device_ops ifb_netdev_ops = {
 123         .ndo_open       = ifb_open,
 124         .ndo_stop       = ifb_close,
 125         .ndo_start_xmit = ifb_xmit,
 126         .ndo_validate_addr = eth_validate_addr,
 127 };
 128
 129 static void ifb_setup(struct net_device *dev)
 130 {
 131         /* Initialize the device structure. */
 132         dev->destructor = free_netdev;
 133         dev->netdev_ops = &ifb_netdev_ops;
 134
 135         /* Fill in device structure with ethernet-generic values. */
 136         ether_setup(dev);
 137         dev->tx_queue_len = TX_Q_LIMIT;
 138
 139         dev->flags |= IFF_NOARP;
 140         dev->flags &= ~IFF_MULTICAST;
 141         dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
 142         random_ether_addr(dev->dev_addr);
 143 }
 144
 145 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev)
 146 {
 147         struct ifb_private *dp = netdev_priv(dev);
 148         struct net_device_stats *stats = &dev->stats;
 149         u32 from = G_TC_FROM(skb->tc_verd);
 150
 151         stats->rx_packets++;
 152         stats->rx_bytes+=skb->len;
 153
 154         if (!(from & (AT_INGRESS|AT_EGRESS)) || !skb->skb_iif) {
 155                 dev_kfree_skb(skb);
 156                 stats->rx_dropped++;
 157                 return NETDEV_TX_OK;
 158         }
 159
 160         if (skb_queue_len(&dp->rq) >= dev->tx_queue_len) {
 161                 netif_stop_queue(dev);
 162         }
 163
 164         __skb_queue_tail(&dp->rq, skb);
 165         if (!dp->tasklet_pending) {
 166                 dp->tasklet_pending = 1;
 167                 tasklet_schedule(&dp->ifb_tasklet);
 168         }
 169
 170         return NETDEV_TX_OK;
 171 }
 172
 173 static int ifb_close(struct net_device *dev)
 174 {
 175         struct ifb_private *dp = netdev_priv(dev);
 176
 177         tasklet_kill(&dp->ifb_tasklet);
 178         netif_stop_queue(dev);
 179         __skb_queue_purge(&dp->rq);
 180         __skb_queue_purge(&dp->tq);
 181         return 0;
 182 }
 183
 184 static int ifb_open(struct net_device *dev)
 185 {
 186         struct ifb_private *dp = netdev_priv(dev);
 187
 188         tasklet_init(&dp->ifb_tasklet, ri_tasklet, (unsigned long)dev);
 189         __skb_queue_head_init(&dp->rq);
 190         __skb_queue_head_init(&dp->tq);
 191         netif_start_queue(dev);
 192
 193         return 0;
 194 }
 195
 196 static int ifb_validate(struct nlattr *tb[], struct nlattr *data[])
 197 {
 198         if (tb[IFLA_ADDRESS]) {
 199                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 200                         return -EINVAL;
 201                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 202                         return -EADDRNOTAVAIL;
 203         }
 204         return 0;
 205 }
 206
 207 static struct rtnl_link_ops ifb_link_ops __read_mostly = {
 208         .kind           = "ifb",
 209         .priv_size      = sizeof(struct ifb_private),
 210         .setup          = ifb_setup,
 211         .validate       = ifb_validate,
 212 };
 213
 214 /* Number of ifb devices to be set up by this module. */
 215 module_param(numifbs, int, 0);
 216 MODULE_PARM_DESC(numifbs, "Number of ifb devices");
 217
 218 static int __init ifb_init_one(int index)
 219 {
 220         struct net_device *dev_ifb;
 221         int err;
 222
 223         dev_ifb = alloc_netdev(sizeof(struct ifb_private),
 224                                  "ifb%d", ifb_setup);
 225
 226         if (!dev_ifb)
 227                 return -ENOMEM;
 228
 229         err = dev_alloc_name(dev_ifb, dev_ifb->name);
 230         if (err < 0)
 231                 goto err;
 232
 233         dev_ifb->rtnl_link_ops = &ifb_link_ops;
 234         err = register_netdevice(dev_ifb);
 235         if (err < 0)
 236                 goto err;
 237
 238         return 0;
 239
 240 err:
 241         free_netdev(dev_ifb);
 242         return err;
 243 }
 244
 245 static int __init ifb_init_module(void)
 246 {
 247         int i, err;
 248
 249         rtnl_lock();
 250         err = __rtnl_link_register(&ifb_link_ops);
 251
 252         for (i = 0; i < numifbs && !err; i++)
 253                 err = ifb_init_one(i);
 254         if (err)
 255                 __rtnl_link_unregister(&ifb_link_ops);
 256         rtnl_unlock();
 257
 258         return err;
 259 }
 260
 261 static void __exit ifb_cleanup_module(void)
 262 {
 263         rtnl_link_unregister(&ifb_link_ops);
 264 }
 265
 266 module_init(ifb_init_module);
 267 module_exit(ifb_cleanup_module);
 268 MODULE_LICENSE("GPL");
 269 MODULE_AUTHOR("Jamal Hadi Salim");
 270 MODULE_ALIAS_RTNL_LINK("ifb");