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 netdev_tx_t 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 rcu_read_lock(); 102 skb->dev = dev_get_by_index_rcu(&init_net, skb->skb_iif); 103 if (!skb->dev) { 104 rcu_read_unlock(); 105 dev_kfree_skb(skb); 106 stats->tx_dropped++; 107 break; 108 } 109 rcu_read_unlock(); 110 skb->skb_iif = _dev->ifindex; 111 112 if (from & AT_EGRESS) { 113 dp->st_rx_frm_egr++; 114 dev_queue_xmit(skb); 115 } else if (from & AT_INGRESS) { 116 dp->st_rx_frm_ing++; 117 skb_pull(skb, skb->dev->hard_header_len); 118 netif_rx(skb); 119 } else 120 BUG(); 121 } 122 123 if (__netif_tx_trylock(txq)) { 124 dp->st_rxq_check++; 125 if ((skb = skb_peek(&dp->rq)) == NULL) { 126 dp->tasklet_pending = 0; 127 if (netif_queue_stopped(_dev)) 128 netif_wake_queue(_dev); 129 } else { 130 dp->st_rxq_rsch++; 131 __netif_tx_unlock(txq); 132 goto resched; 133 } 134 __netif_tx_unlock(txq); 135 } else { 136 resched: 137 dp->tasklet_pending = 1; 138 tasklet_schedule(&dp->ifb_tasklet); 139 } 140 141 } 142 143 static const struct net_device_ops ifb_netdev_ops = { 144 .ndo_open = ifb_open, 145 .ndo_stop = ifb_close, 146 .ndo_start_xmit = ifb_xmit, 147 .ndo_validate_addr = eth_validate_addr, 148 }; 149 150 static void ifb_setup(struct net_device *dev) 151 { 152 /* Initialize the device structure. */ 153 dev->destructor = free_netdev; 154 dev->netdev_ops = &ifb_netdev_ops; 155 156 /* Fill in device structure with ethernet-generic values. */ 157 ether_setup(dev); 158 dev->tx_queue_len = TX_Q_LIMIT; 159 160 dev->flags |= IFF_NOARP; 161 dev->flags &= ~IFF_MULTICAST; 162 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 163 random_ether_addr(dev->dev_addr); 164 } 165 166 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev) 167 { 168 struct ifb_private *dp = netdev_priv(dev); 169 struct net_device_stats *stats = &dev->stats; 170 u32 from = G_TC_FROM(skb->tc_verd); 171 172 stats->rx_packets++; 173 stats->rx_bytes+=skb->len; 174 175 if (!(from & (AT_INGRESS|AT_EGRESS)) || !skb->skb_iif) { 176 dev_kfree_skb(skb); 177 stats->rx_dropped++; 178 return NETDEV_TX_OK; 179 } 180 181 if (skb_queue_len(&dp->rq) >= dev->tx_queue_len) { 182 netif_stop_queue(dev); 183 } 184 185 dev->trans_start = jiffies; 186 skb_queue_tail(&dp->rq, skb); 187 if (!dp->tasklet_pending) { 188 dp->tasklet_pending = 1; 189 tasklet_schedule(&dp->ifb_tasklet); 190 } 191 192 return NETDEV_TX_OK; 193 } 194 195 static int ifb_close(struct net_device *dev) 196 { 197 struct ifb_private *dp = netdev_priv(dev); 198 199 tasklet_kill(&dp->ifb_tasklet); 200 netif_stop_queue(dev); 201 skb_queue_purge(&dp->rq); 202 skb_queue_purge(&dp->tq); 203 return 0; 204 } 205 206 static int ifb_open(struct net_device *dev) 207 { 208 struct ifb_private *dp = netdev_priv(dev); 209 210 tasklet_init(&dp->ifb_tasklet, ri_tasklet, (unsigned long)dev); 211 skb_queue_head_init(&dp->rq); 212 skb_queue_head_init(&dp->tq); 213 netif_start_queue(dev); 214 215 return 0; 216 } 217 218 static int ifb_validate(struct nlattr *tb[], struct nlattr *data[]) 219 { 220 if (tb[IFLA_ADDRESS]) { 221 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 222 return -EINVAL; 223 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 224 return -EADDRNOTAVAIL; 225 } 226 return 0; 227 } 228 229 static struct rtnl_link_ops ifb_link_ops __read_mostly = { 230 .kind = "ifb", 231 .priv_size = sizeof(struct ifb_private), 232 .setup = ifb_setup, 233 .validate = ifb_validate, 234 }; 235 236 /* Number of ifb devices to be set up by this module. */ 237 module_param(numifbs, int, 0); 238 MODULE_PARM_DESC(numifbs, "Number of ifb devices"); 239 240 static int __init ifb_init_one(int index) 241 { 242 struct net_device *dev_ifb; 243 int err; 244 245 dev_ifb = alloc_netdev(sizeof(struct ifb_private), 246 "ifb%d", ifb_setup); 247 248 if (!dev_ifb) 249 return -ENOMEM; 250 251 err = dev_alloc_name(dev_ifb, dev_ifb->name); 252 if (err < 0) 253 goto err; 254 255 dev_ifb->rtnl_link_ops = &ifb_link_ops; 256 err = register_netdevice(dev_ifb); 257 if (err < 0) 258 goto err; 259 260 return 0; 261 262 err: 263 free_netdev(dev_ifb); 264 return err; 265 } 266 267 static int __init ifb_init_module(void) 268 { 269 int i, err; 270 271 rtnl_lock(); 272 err = __rtnl_link_register(&ifb_link_ops); 273 274 for (i = 0; i < numifbs && !err; i++) 275 err = ifb_init_one(i); 276 if (err) 277 __rtnl_link_unregister(&ifb_link_ops); 278 rtnl_unlock(); 279 280 return err; 281 } 282 283 static void __exit ifb_cleanup_module(void) 284 { 285 rtnl_link_unregister(&ifb_link_ops); 286 } 287 288 module_init(ifb_init_module); 289 module_exit(ifb_cleanup_module); 290 MODULE_LICENSE("GPL"); 291 MODULE_AUTHOR("Jamal Hadi Salim"); 292 MODULE_ALIAS_RTNL_LINK("ifb"); 293