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 net_device_stats stats;
  44         struct tasklet_struct   ifb_tasklet;
  45         int     tasklet_pending;
  46         /* mostly debug stats leave in for now */
  47         unsigned long   st_task_enter; /* tasklet entered */
  48         unsigned long   st_txq_refl_try; /* transmit queue refill attempt */
  49         unsigned long   st_rxq_enter; /* receive queue entered */
  50         unsigned long   st_rx2tx_tran; /* receive to trasmit transfers */
  51         unsigned long   st_rxq_notenter; /*receiveQ not entered, resched */
  52         unsigned long   st_rx_frm_egr; /* received from egress path */
  53         unsigned long   st_rx_frm_ing; /* received from ingress path */
  54         unsigned long   st_rxq_check;
  55         unsigned long   st_rxq_rsch;
  56         struct sk_buff_head     rq;
  57         struct sk_buff_head     tq;
  58 };
  59
  60 static int numifbs = 2;
  61
  62 static void ri_tasklet(unsigned long dev);
  63 static int ifb_xmit(struct sk_buff *skb, struct net_device *dev);
  64 static struct net_device_stats *ifb_get_stats(struct net_device *dev);
  65 static int ifb_open(struct net_device *dev);
  66 static int ifb_close(struct net_device *dev);
  67
  68 static void ri_tasklet(unsigned long dev)
  69 {
  70
  71         struct net_device *_dev = (struct net_device *)dev;
  72         struct ifb_private *dp = netdev_priv(_dev);
  73         struct net_device_stats *stats = &dp->stats;
  74         struct sk_buff *skb;
  75
  76         dp->st_task_enter++;
  77         if ((skb = skb_peek(&dp->tq)) == NULL) {
  78                 dp->st_txq_refl_try++;
  79                 if (netif_tx_trylock(_dev)) {
  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(_dev);
  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(_dev)) {
 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(_dev);
 129                         goto resched;
 130                 }
 131                 netif_tx_unlock(_dev);
 132         } else {
 133 resched:
 134                 dp->tasklet_pending = 1;
 135                 tasklet_schedule(&dp->ifb_tasklet);
 136         }
 137
 138 }
 139
 140 static void ifb_setup(struct net_device *dev)
 141 {
 142         /* Initialize the device structure. */
 143         dev->get_stats = ifb_get_stats;
 144         dev->hard_start_xmit = ifb_xmit;
 145         dev->open = &ifb_open;
 146         dev->stop = &ifb_close;
 147         dev->destructor = free_netdev;
 148
 149         /* Fill in device structure with ethernet-generic values. */
 150         ether_setup(dev);
 151         dev->tx_queue_len = TX_Q_LIMIT;
 152         dev->change_mtu = NULL;
 153         dev->flags |= IFF_NOARP;
 154         dev->flags &= ~IFF_MULTICAST;
 155         SET_MODULE_OWNER(dev);
 156         random_ether_addr(dev->dev_addr);
 157 }
 158
 159 static int ifb_xmit(struct sk_buff *skb, struct net_device *dev)
 160 {
 161         struct ifb_private *dp = netdev_priv(dev);
 162         struct net_device_stats *stats = &dp->stats;
 163         int ret = 0;
 164         u32 from = G_TC_FROM(skb->tc_verd);
 165
 166         stats->rx_packets++;
 167         stats->rx_bytes+=skb->len;
 168
 169         if (!(from & (AT_INGRESS|AT_EGRESS)) || !skb->iif) {
 170                 dev_kfree_skb(skb);
 171                 stats->rx_dropped++;
 172                 return ret;
 173         }
 174
 175         if (skb_queue_len(&dp->rq) >= dev->tx_queue_len) {
 176                 netif_stop_queue(dev);
 177         }
 178
 179         dev->trans_start = jiffies;
 180         skb_queue_tail(&dp->rq, skb);
 181         if (!dp->tasklet_pending) {
 182                 dp->tasklet_pending = 1;
 183                 tasklet_schedule(&dp->ifb_tasklet);
 184         }
 185
 186         return ret;
 187 }
 188
 189 static struct net_device_stats *ifb_get_stats(struct net_device *dev)
 190 {
 191         struct ifb_private *dp = netdev_priv(dev);
 192         struct net_device_stats *stats = &dp->stats;
 193
 194         pr_debug("tasklets stats %ld:%ld:%ld:%ld:%ld:%ld:%ld:%ld:%ld \n",
 195                 dp->st_task_enter, dp->st_txq_refl_try, dp->st_rxq_enter,
 196                 dp->st_rx2tx_tran, dp->st_rxq_notenter, dp->st_rx_frm_egr,
 197                 dp->st_rx_frm_ing, dp->st_rxq_check, dp->st_rxq_rsch);
 198
 199         return stats;
 200 }
 201
 202 static int ifb_close(struct net_device *dev)
 203 {
 204         struct ifb_private *dp = netdev_priv(dev);
 205
 206         tasklet_kill(&dp->ifb_tasklet);
 207         netif_stop_queue(dev);
 208         skb_queue_purge(&dp->rq);
 209         skb_queue_purge(&dp->tq);
 210         return 0;
 211 }
 212
 213 static int ifb_open(struct net_device *dev)
 214 {
 215         struct ifb_private *dp = netdev_priv(dev);
 216
 217         tasklet_init(&dp->ifb_tasklet, ri_tasklet, (unsigned long)dev);
 218         skb_queue_head_init(&dp->rq);
 219         skb_queue_head_init(&dp->tq);
 220         netif_start_queue(dev);
 221
 222         return 0;
 223 }
 224
 225 static int ifb_validate(struct nlattr *tb[], struct nlattr *data[])
 226 {
 227         if (tb[IFLA_ADDRESS]) {
 228                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 229                         return -EINVAL;
 230                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 231                         return -EADDRNOTAVAIL;
 232         }
 233         return 0;
 234 }
 235
 236 static struct rtnl_link_ops ifb_link_ops __read_mostly = {
 237         .kind           = "ifb",
 238         .priv_size      = sizeof(struct ifb_private),
 239         .setup          = ifb_setup,
 240         .validate       = ifb_validate,
 241 };
 242
 243 /* Number of ifb devices to be set up by this module. */
 244 module_param(numifbs, int, 0);
 245 MODULE_PARM_DESC(numifbs, "Number of ifb devices");
 246
 247 static int __init ifb_init_one(int index)
 248 {
 249         struct net_device *dev_ifb;
 250         int err;
 251
 252         dev_ifb = alloc_netdev(sizeof(struct ifb_private),
 253                                  "ifb%d", ifb_setup);
 254
 255         if (!dev_ifb)
 256                 return -ENOMEM;
 257
 258         err = dev_alloc_name(dev_ifb, dev_ifb->name);
 259         if (err < 0)
 260                 goto err;
 261
 262         dev_ifb->rtnl_link_ops = &ifb_link_ops;
 263         err = register_netdevice(dev_ifb);
 264         if (err < 0)
 265                 goto err;
 266         return 0;
 267
 268 err:
 269         free_netdev(dev_ifb);
 270         return err;
 271 }
 272
 273 static int __init ifb_init_module(void)
 274 {
 275         int i, err;
 276
 277         rtnl_lock();
 278         err = __rtnl_link_register(&ifb_link_ops);
 279
 280         for (i = 0; i < numifbs && !err; i++)
 281                 err = ifb_init_one(i);
 282         if (err)
 283                 __rtnl_link_unregister(&ifb_link_ops);
 284         rtnl_unlock();
 285
 286         return err;
 287 }
 288
 289 static void __exit ifb_cleanup_module(void)
 290 {
 291         rtnl_link_unregister(&ifb_link_ops);
 292 }
 293
 294 module_init(ifb_init_module);
 295 module_exit(ifb_cleanup_module);
 296 MODULE_LICENSE("GPL");
 297 MODULE_AUTHOR("Jamal Hadi Salim");
 298 MODULE_ALIAS_RTNL_LINK("ifb");