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_TIMEOUT  (2*HZ)
  40
  41 #define TX_Q_LIMIT    32
  42 struct ifb_private {
  43         struct tasklet_struct   ifb_tasklet;
  44         int     tasklet_pending;
  45         /* mostly debug stats leave in for now */
  46         unsigned long   st_task_enter; /* tasklet entered */
  47         unsigned long   st_txq_refl_try; /* transmit queue refill attempt */
  48         unsigned long   st_rxq_enter; /* receive queue entered */
  49         unsigned long   st_rx2tx_tran; /* receive to trasmit transfers */
  50         unsigned long   st_rxq_notenter; /*receiveQ not entered, resched */
  51         unsigned long   st_rx_frm_egr; /* received from egress path */
  52         unsigned long   st_rx_frm_ing; /* received from ingress path */
  53         unsigned long   st_rxq_check;
  54         unsigned long   st_rxq_rsch;
  55         struct sk_buff_head     rq;
  56         struct sk_buff_head     tq;
  57 };
  58
  59 static int numifbs = 2;
  60
  61 static void ri_tasklet(unsigned long dev);
  62 static int ifb_xmit(struct sk_buff *skb, struct net_device *dev);
  63 static int ifb_open(struct net_device *dev);
  64 static int ifb_close(struct net_device *dev);
  65
  66 static void ri_tasklet(unsigned long dev)
  67 {
  68
  69         struct net_device *_dev = (struct net_device *)dev;
  70         struct ifb_private *dp = netdev_priv(_dev);
  71         struct net_device_stats *stats = &_dev->stats;
  72         struct netdev_queue *txq;
  73         struct sk_buff *skb;
  74
  75         txq = netdev_get_tx_queue(_dev, 0);
  76         dp->st_task_enter++;
  77         if ((skb = skb_peek(&dp->tq)) == NULL) {
  78                 dp->st_txq_refl_try++;
  79                 if (__netif_tx_trylock(txq)) {
  80                         dp->st_rxq_enter++;
  81                         while ((skb = skb_dequeue(&dp->rq)) != NULL) {
  82                                 skb_queue_tail(&dp->tq, skb);
  83                                 dp->st_rx2tx_tran++;
  84                         }
  85                         __netif_tx_unlock(txq);
  86                 } else {
  87                         /* reschedule */
  88                         dp->st_rxq_notenter++;
  89                         goto resched;
  90                 }
  91         }
  92
  93         while ((skb = skb_dequeue(&dp->tq)) != NULL) {
  94                 u32 from = G_TC_FROM(skb->tc_verd);
  95
  96                 skb->tc_verd = 0;
  97                 skb->tc_verd = SET_TC_NCLS(skb->tc_verd);
  98                 stats->tx_packets++;
  99                 stats->tx_bytes +=skb->len;
 100
 101                 skb->dev = __dev_get_by_index(&init_net, skb->iif);
 102                 if (!skb->dev) {
 103                         dev_kfree_skb(skb);
 104                         stats->tx_dropped++;
 105                         break;
 106                 }
 107                 skb->iif = _dev->ifindex;
 108
 109                 if (from & AT_EGRESS) {
 110                         dp->st_rx_frm_egr++;
 111                         dev_queue_xmit(skb);
 112                 } else if (from & AT_INGRESS) {
 113                         dp->st_rx_frm_ing++;
 114                         skb_pull(skb, skb->dev->hard_header_len);
 115                         netif_rx(skb);
 116                 } else
 117                         BUG();
 118         }
 119
 120         if (__netif_tx_trylock(txq)) {
 121                 dp->st_rxq_check++;
 122                 if ((skb = skb_peek(&dp->rq)) == NULL) {
 123                         dp->tasklet_pending = 0;
 124                         if (netif_queue_stopped(_dev))
 125                                 netif_wake_queue(_dev);
 126                 } else {
 127                         dp->st_rxq_rsch++;
 128                         __netif_tx_unlock(txq);
 129                         goto resched;
 130                 }
 131                 __netif_tx_unlock(txq);
 132         } else {
 133 resched:
 134                 dp->tasklet_pending = 1;
 135                 tasklet_schedule(&dp->ifb_tasklet);
 136         }
 137
 138 }
 139
 140 static const struct net_device_ops ifb_netdev_ops = {
 141         .ndo_open       = ifb_open,
 142         .ndo_stop       = ifb_close,
 143         .ndo_start_xmit = ifb_xmit,
 144         .ndo_validate_addr = eth_validate_addr,
 145 };
 146
 147 static void ifb_setup(struct net_device *dev)
 148 {
 149         /* Initialize the device structure. */
 150         dev->destructor = free_netdev;
 151         dev->netdev_ops = &ifb_netdev_ops;
 152
 153         /* Fill in device structure with ethernet-generic values. */
 154         ether_setup(dev);
 155         dev->tx_queue_len = TX_Q_LIMIT;
 156
 157         dev->flags |= IFF_NOARP;
 158         dev->flags &= ~IFF_MULTICAST;
 159         dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
 160         random_ether_addr(dev->dev_addr);
 161 }
 162
 163 static int ifb_xmit(struct sk_buff *skb, struct net_device *dev)
 164 {
 165         struct ifb_private *dp = netdev_priv(dev);
 166         struct net_device_stats *stats = &dev->stats;
 167         int ret = 0;
 168         u32 from = G_TC_FROM(skb->tc_verd);
 169
 170         stats->rx_packets++;
 171         stats->rx_bytes+=skb->len;
 172
 173         if (!(from & (AT_INGRESS|AT_EGRESS)) || !skb->iif) {
 174                 dev_kfree_skb(skb);
 175                 stats->rx_dropped++;
 176                 return ret;
 177         }
 178
 179         if (skb_queue_len(&dp->rq) >= dev->tx_queue_len) {
 180                 netif_stop_queue(dev);
 181         }
 182
 183         dev->trans_start = jiffies;
 184         skb_queue_tail(&dp->rq, skb);
 185         if (!dp->tasklet_pending) {
 186                 dp->tasklet_pending = 1;
 187                 tasklet_schedule(&dp->ifb_tasklet);
 188         }
 189
 190         return ret;
 191 }
 192
 193 static int ifb_close(struct net_device *dev)
 194 {
 195         struct ifb_private *dp = netdev_priv(dev);
 196
 197         tasklet_kill(&dp->ifb_tasklet);
 198         netif_stop_queue(dev);
 199         skb_queue_purge(&dp->rq);
 200         skb_queue_purge(&dp->tq);
 201         return 0;
 202 }
 203
 204 static int ifb_open(struct net_device *dev)
 205 {
 206         struct ifb_private *dp = netdev_priv(dev);
 207
 208         tasklet_init(&dp->ifb_tasklet, ri_tasklet, (unsigned long)dev);
 209         skb_queue_head_init(&dp->rq);
 210         skb_queue_head_init(&dp->tq);
 211         netif_start_queue(dev);
 212
 213         return 0;
 214 }
 215
 216 static int ifb_validate(struct nlattr *tb[], struct nlattr *data[])
 217 {
 218         if (tb[IFLA_ADDRESS]) {
 219                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 220                         return -EINVAL;
 221                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 222                         return -EADDRNOTAVAIL;
 223         }
 224         return 0;
 225 }
 226
 227 static struct rtnl_link_ops ifb_link_ops __read_mostly = {
 228         .kind           = "ifb",
 229         .priv_size      = sizeof(struct ifb_private),
 230         .setup          = ifb_setup,
 231         .validate       = ifb_validate,
 232 };
 233
 234 /* Number of ifb devices to be set up by this module. */
 235 module_param(numifbs, int, 0);
 236 MODULE_PARM_DESC(numifbs, "Number of ifb devices");
 237
 238 static int __init ifb_init_one(int index)
 239 {
 240         struct net_device *dev_ifb;
 241         int err;
 242
 243         dev_ifb = alloc_netdev(sizeof(struct ifb_private),
 244                                  "ifb%d", ifb_setup);
 245
 246         if (!dev_ifb)
 247                 return -ENOMEM;
 248
 249         err = dev_alloc_name(dev_ifb, dev_ifb->name);
 250         if (err < 0)
 251                 goto err;
 252
 253         dev_ifb->rtnl_link_ops = &ifb_link_ops;
 254         err = register_netdevice(dev_ifb);
 255         if (err < 0)
 256                 goto err;
 257
 258         return 0;
 259
 260 err:
 261         free_netdev(dev_ifb);
 262         return err;
 263 }
 264
 265 static int __init ifb_init_module(void)
 266 {
 267         int i, err;
 268
 269         rtnl_lock();
 270         err = __rtnl_link_register(&ifb_link_ops);
 271
 272         for (i = 0; i < numifbs && !err; i++)
 273                 err = ifb_init_one(i);
 274         if (err)
 275                 __rtnl_link_unregister(&ifb_link_ops);
 276         rtnl_unlock();
 277
 278         return err;
 279 }
 280
 281 static void __exit ifb_cleanup_module(void)
 282 {
 283         rtnl_link_unregister(&ifb_link_ops);
 284 }
 285
 286 module_init(ifb_init_module);
 287 module_exit(ifb_cleanup_module);
 288 MODULE_LICENSE("GPL");
 289 MODULE_AUTHOR("Jamal Hadi Salim");
 290 MODULE_ALIAS_RTNL_LINK("ifb");