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 42struct 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 59static int numifbs = 2; 60 61static void ri_tasklet(unsigned long dev); 62static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev); 63static int ifb_open(struct net_device *dev); 64static int ifb_close(struct net_device *dev); 65 66static 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 if (skb_queue_len(&dp->tq) != 0) 108 goto resched; 109 break; 110 } 111 rcu_read_unlock(); 112 skb->skb_iif = _dev->ifindex; 113 114 if (from & AT_EGRESS) { 115 dp->st_rx_frm_egr++; 116 dev_queue_xmit(skb); 117 } else if (from & AT_INGRESS) { 118 dp->st_rx_frm_ing++; 119 skb_pull(skb, skb->dev->hard_header_len); 120 netif_rx(skb); 121 } else 122 BUG(); 123 } 124 125 if (__netif_tx_trylock(txq)) { 126 dp->st_rxq_check++; 127 if ((skb = skb_peek(&dp->rq)) == NULL) { 128 dp->tasklet_pending = 0; 129 if (netif_queue_stopped(_dev)) 130 netif_wake_queue(_dev); 131 } else { 132 dp->st_rxq_rsch++; 133 __netif_tx_unlock(txq); 134 goto resched; 135 } 136 __netif_tx_unlock(txq); 137 } else { 138resched: 139 dp->tasklet_pending = 1; 140 tasklet_schedule(&dp->ifb_tasklet); 141 } 142 143} 144 145static const struct net_device_ops ifb_netdev_ops = { 146 .ndo_open = ifb_open, 147 .ndo_stop = ifb_close, 148 .ndo_start_xmit = ifb_xmit, 149 .ndo_validate_addr = eth_validate_addr, 150}; 151 152static void ifb_setup(struct net_device *dev) 153{ 154 /* Initialize the device structure. */ 155 dev->destructor = free_netdev; 156 dev->netdev_ops = &ifb_netdev_ops; 157 158 /* Fill in device structure with ethernet-generic values. */ 159 ether_setup(dev); 160 dev->tx_queue_len = TX_Q_LIMIT; 161 162 dev->flags |= IFF_NOARP; 163 dev->flags &= ~IFF_MULTICAST; 164 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 165 random_ether_addr(dev->dev_addr); 166} 167 168static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev) 169{ 170 struct ifb_private *dp = netdev_priv(dev); 171 struct net_device_stats *stats = &dev->stats; 172 u32 from = G_TC_FROM(skb->tc_verd); 173 174 stats->rx_packets++; 175 stats->rx_bytes+=skb->len; 176 177 if (!(from & (AT_INGRESS|AT_EGRESS)) || !skb->skb_iif) { 178 dev_kfree_skb(skb); 179 stats->rx_dropped++; 180 return NETDEV_TX_OK; 181 } 182 183 if (skb_queue_len(&dp->rq) >= dev->tx_queue_len) { 184 netif_stop_queue(dev); 185 } 186 187 skb_queue_tail(&dp->rq, skb); 188 if (!dp->tasklet_pending) { 189 dp->tasklet_pending = 1; 190 tasklet_schedule(&dp->ifb_tasklet); 191 } 192 193 return NETDEV_TX_OK; 194} 195 196static int ifb_close(struct net_device *dev) 197{ 198 struct ifb_private *dp = netdev_priv(dev); 199 200 tasklet_kill(&dp->ifb_tasklet); 201 netif_stop_queue(dev); 202 skb_queue_purge(&dp->rq); 203 skb_queue_purge(&dp->tq); 204 return 0; 205} 206 207static int ifb_open(struct net_device *dev) 208{ 209 struct ifb_private *dp = netdev_priv(dev); 210 211 tasklet_init(&dp->ifb_tasklet, ri_tasklet, (unsigned long)dev); 212 skb_queue_head_init(&dp->rq); 213 skb_queue_head_init(&dp->tq); 214 netif_start_queue(dev); 215 216 return 0; 217} 218 219static int ifb_validate(struct nlattr *tb[], struct nlattr *data[]) 220{ 221 if (tb[IFLA_ADDRESS]) { 222 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 223 return -EINVAL; 224 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 225 return -EADDRNOTAVAIL; 226 } 227 return 0; 228} 229 230static struct rtnl_link_ops ifb_link_ops __read_mostly = { 231 .kind = "ifb", 232 .priv_size = sizeof(struct ifb_private), 233 .setup = ifb_setup, 234 .validate = ifb_validate, 235}; 236 237/* Number of ifb devices to be set up by this module. */ 238module_param(numifbs, int, 0); 239MODULE_PARM_DESC(numifbs, "Number of ifb devices"); 240 241static int __init ifb_init_one(int index) 242{ 243 struct net_device *dev_ifb; 244 int err; 245 246 dev_ifb = alloc_netdev(sizeof(struct ifb_private), 247 "ifb%d", ifb_setup); 248 249 if (!dev_ifb) 250 return -ENOMEM; 251 252 err = dev_alloc_name(dev_ifb, dev_ifb->name); 253 if (err < 0) 254 goto err; 255 256 dev_ifb->rtnl_link_ops = &ifb_link_ops; 257 err = register_netdevice(dev_ifb); 258 if (err < 0) 259 goto err; 260 261 return 0; 262 263err: 264 free_netdev(dev_ifb); 265 return err; 266} 267 268static int __init ifb_init_module(void) 269{ 270 int i, err; 271 272 rtnl_lock(); 273 err = __rtnl_link_register(&ifb_link_ops); 274 275 for (i = 0; i < numifbs && !err; i++) 276 err = ifb_init_one(i); 277 if (err) 278 __rtnl_link_unregister(&ifb_link_ops); 279 rtnl_unlock(); 280 281 return err; 282} 283 284static void __exit ifb_cleanup_module(void) 285{ 286 rtnl_link_unregister(&ifb_link_ops); 287} 288 289module_init(ifb_init_module); 290module_exit(ifb_cleanup_module); 291MODULE_LICENSE("GPL"); 292MODULE_AUTHOR("Jamal Hadi Salim"); 293MODULE_ALIAS_RTNL_LINK("ifb"); 294