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 sk_buff *skb;
  73
  74         dp->st_task_enter++;
  75         if ((skb = skb_peek(&dp->tq)) == NULL) {
  76                 dp->st_txq_refl_try++;
  77                 if (netif_tx_trylock(_dev)) {
  78                         dp->st_rxq_enter++;
  79                         while ((skb = skb_dequeue(&dp->rq)) != NULL) {
  80                                 skb_queue_tail(&dp->tq, skb);
  81                                 dp->st_rx2tx_tran++;
  82                         }
  83                         netif_tx_unlock(_dev);
  84                 } else {
  85                         /* reschedule */
  86                         dp->st_rxq_notenter++;
  87                         goto resched;
  88                 }
  89         }
  90
  91         while ((skb = skb_dequeue(&dp->tq)) != NULL) {
  92                 u32 from = G_TC_FROM(skb->tc_verd);
  93
  94                 skb->tc_verd = 0;
  95                 skb->tc_verd = SET_TC_NCLS(skb->tc_verd);
  96                 stats->tx_packets++;
  97                 stats->tx_bytes +=skb->len;
  98
  99                 skb->dev = __dev_get_by_index(&init_net, skb->iif);
 100                 if (!skb->dev) {
 101                         dev_kfree_skb(skb);
 102                         stats->tx_dropped++;
 103                         break;
 104                 }
 105                 skb->iif = _dev->ifindex;
 106
 107                 if (from & AT_EGRESS) {
 108                         dp->st_rx_frm_egr++;
 109                         dev_queue_xmit(skb);
 110                 } else if (from & AT_INGRESS) {
 111                         dp->st_rx_frm_ing++;
 112                         skb_pull(skb, skb->dev->hard_header_len);
 113                         netif_rx(skb);
 114                 } else
 115                         BUG();
 116         }
 117
 118         if (netif_tx_trylock(_dev)) {
 119                 dp->st_rxq_check++;
 120                 if ((skb = skb_peek(&dp->rq)) == NULL) {
 121                         dp->tasklet_pending = 0;
 122                         if (netif_queue_stopped(_dev))
 123                                 netif_wake_queue(_dev);
 124                 } else {
 125                         dp->st_rxq_rsch++;
 126                         netif_tx_unlock(_dev);
 127                         goto resched;
 128                 }
 129                 netif_tx_unlock(_dev);
 130         } else {
 131 resched:
 132                 dp->tasklet_pending = 1;
 133                 tasklet_schedule(&dp->ifb_tasklet);
 134         }
 135
 136 }
 137
 138 static void ifb_setup(struct net_device *dev)
 139 {
 140         /* Initialize the device structure. */
 141         dev->hard_start_xmit = ifb_xmit;
 142         dev->open = &ifb_open;
 143         dev->stop = &ifb_close;
 144         dev->destructor = free_netdev;
 145
 146         /* Fill in device structure with ethernet-generic values. */
 147         ether_setup(dev);
 148         dev->tx_queue_len = TX_Q_LIMIT;
 149         dev->change_mtu = NULL;
 150         dev->flags |= IFF_NOARP;
 151         dev->flags &= ~IFF_MULTICAST;
 152         random_ether_addr(dev->dev_addr);
 153 }
 154
 155 static int ifb_xmit(struct sk_buff *skb, struct net_device *dev)
 156 {
 157         struct ifb_private *dp = netdev_priv(dev);
 158         struct net_device_stats *stats = &dev->stats;
 159         int ret = 0;
 160         u32 from = G_TC_FROM(skb->tc_verd);
 161
 162         stats->rx_packets++;
 163         stats->rx_bytes+=skb->len;
 164
 165         if (!(from & (AT_INGRESS|AT_EGRESS)) || !skb->iif) {
 166                 dev_kfree_skb(skb);
 167                 stats->rx_dropped++;
 168                 return ret;
 169         }
 170
 171         if (skb_queue_len(&dp->rq) >= dev->tx_queue_len) {
 172                 netif_stop_queue(dev);
 173         }
 174
 175         dev->trans_start = jiffies;
 176         skb_queue_tail(&dp->rq, skb);
 177         if (!dp->tasklet_pending) {
 178                 dp->tasklet_pending = 1;
 179                 tasklet_schedule(&dp->ifb_tasklet);
 180         }
 181
 182         return ret;
 183 }
 184
 185 static int ifb_close(struct net_device *dev)
 186 {
 187         struct ifb_private *dp = netdev_priv(dev);
 188
 189         tasklet_kill(&dp->ifb_tasklet);
 190         netif_stop_queue(dev);
 191         skb_queue_purge(&dp->rq);
 192         skb_queue_purge(&dp->tq);
 193         return 0;
 194 }
 195
 196 static int ifb_open(struct net_device *dev)
 197 {
 198         struct ifb_private *dp = netdev_priv(dev);
 199
 200         tasklet_init(&dp->ifb_tasklet, ri_tasklet, (unsigned long)dev);
 201         skb_queue_head_init(&dp->rq);
 202         skb_queue_head_init(&dp->tq);
 203         netif_start_queue(dev);
 204
 205         return 0;
 206 }
 207
 208 static int ifb_validate(struct nlattr *tb[], struct nlattr *data[])
 209 {
 210         if (tb[IFLA_ADDRESS]) {
 211                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 212                         return -EINVAL;
 213                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 214                         return -EADDRNOTAVAIL;
 215         }
 216         return 0;
 217 }
 218
 219 static struct rtnl_link_ops ifb_link_ops __read_mostly = {
 220         .kind           = "ifb",
 221         .priv_size      = sizeof(struct ifb_private),
 222         .setup          = ifb_setup,
 223         .validate       = ifb_validate,
 224 };
 225
 226 /* Number of ifb devices to be set up by this module. */
 227 module_param(numifbs, int, 0);
 228 MODULE_PARM_DESC(numifbs, "Number of ifb devices");
 229
 230 static int __init ifb_init_one(int index)
 231 {
 232         struct net_device *dev_ifb;
 233         int err;
 234
 235         dev_ifb = alloc_netdev(sizeof(struct ifb_private),
 236                                  "ifb%d", ifb_setup);
 237
 238         if (!dev_ifb)
 239                 return -ENOMEM;
 240
 241         err = dev_alloc_name(dev_ifb, dev_ifb->name);
 242         if (err < 0)
 243                 goto err;
 244
 245         dev_ifb->rtnl_link_ops = &ifb_link_ops;
 246         err = register_netdevice(dev_ifb);
 247         if (err < 0)
 248                 goto err;
 249         return 0;
 250
 251 err:
 252         free_netdev(dev_ifb);
 253         return err;
 254 }
 255
 256 static int __init ifb_init_module(void)
 257 {
 258         int i, err;
 259
 260         rtnl_lock();
 261         err = __rtnl_link_register(&ifb_link_ops);
 262
 263         for (i = 0; i < numifbs && !err; i++)
 264                 err = ifb_init_one(i);
 265         if (err)
 266                 __rtnl_link_unregister(&ifb_link_ops);
 267         rtnl_unlock();
 268
 269         return err;
 270 }
 271
 272 static void __exit ifb_cleanup_module(void)
 273 {
 274         rtnl_link_unregister(&ifb_link_ops);
 275 }
 276
 277 module_init(ifb_init_module);
 278 module_exit(ifb_cleanup_module);
 279 MODULE_LICENSE("GPL");
 280 MODULE_AUTHOR("Jamal Hadi Salim");
 281 MODULE_ALIAS_RTNL_LINK("ifb");