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 38 #define TX_TIMEOUT (2*HZ) 39 40 #define TX_Q_LIMIT 32 41 struct ifb_private { 42 struct net_device_stats stats; 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 struct net_device_stats *ifb_get_stats(struct net_device *dev); 64 static int ifb_open(struct net_device *dev); 65 static int ifb_close(struct net_device *dev); 66 67 static void ri_tasklet(unsigned long dev) 68 { 69 70 struct net_device *_dev = (struct net_device *)dev; 71 struct ifb_private *dp = netdev_priv(_dev); 72 struct net_device_stats *stats = &dp->stats; 73 struct sk_buff *skb; 74 75 dp->st_task_enter++; 76 if ((skb = skb_peek(&dp->tq)) == NULL) { 77 dp->st_txq_refl_try++; 78 if (netif_tx_trylock(_dev)) { 79 dp->st_rxq_enter++; 80 while ((skb = skb_dequeue(&dp->rq)) != NULL) { 81 skb_queue_tail(&dp->tq, skb); 82 dp->st_rx2tx_tran++; 83 } 84 netif_tx_unlock(_dev); 85 } else { 86 /* reschedule */ 87 dp->st_rxq_notenter++; 88 goto resched; 89 } 90 } 91 92 while ((skb = skb_dequeue(&dp->tq)) != NULL) { 93 u32 from = G_TC_FROM(skb->tc_verd); 94 95 skb->tc_verd = 0; 96 skb->tc_verd = SET_TC_NCLS(skb->tc_verd); 97 stats->tx_packets++; 98 stats->tx_bytes +=skb->len; 99 100 skb->dev = __dev_get_by_index(skb->iif); 101 if (!skb->dev) { 102 dev_kfree_skb(skb); 103 stats->tx_dropped++; 104 break; 105 } 106 skb->iif = _dev->ifindex; 107 108 if (from & AT_EGRESS) { 109 dp->st_rx_frm_egr++; 110 dev_queue_xmit(skb); 111 } else if (from & AT_INGRESS) { 112 dp->st_rx_frm_ing++; 113 skb_pull(skb, skb->dev->hard_header_len); 114 netif_rx(skb); 115 } else 116 BUG(); 117 } 118 119 if (netif_tx_trylock(_dev)) { 120 dp->st_rxq_check++; 121 if ((skb = skb_peek(&dp->rq)) == NULL) { 122 dp->tasklet_pending = 0; 123 if (netif_queue_stopped(_dev)) 124 netif_wake_queue(_dev); 125 } else { 126 dp->st_rxq_rsch++; 127 netif_tx_unlock(_dev); 128 goto resched; 129 } 130 netif_tx_unlock(_dev); 131 } else { 132 resched: 133 dp->tasklet_pending = 1; 134 tasklet_schedule(&dp->ifb_tasklet); 135 } 136 137 } 138 139 static void ifb_setup(struct net_device *dev) 140 { 141 /* Initialize the device structure. */ 142 dev->get_stats = ifb_get_stats; 143 dev->hard_start_xmit = ifb_xmit; 144 dev->open = &ifb_open; 145 dev->stop = &ifb_close; 146 dev->destructor = free_netdev; 147 148 /* Fill in device structure with ethernet-generic values. */ 149 ether_setup(dev); 150 dev->tx_queue_len = TX_Q_LIMIT; 151 dev->change_mtu = NULL; 152 dev->flags |= IFF_NOARP; 153 dev->flags &= ~IFF_MULTICAST; 154 SET_MODULE_OWNER(dev); 155 random_ether_addr(dev->dev_addr); 156 } 157 158 static int ifb_xmit(struct sk_buff *skb, struct net_device *dev) 159 { 160 struct ifb_private *dp = netdev_priv(dev); 161 struct net_device_stats *stats = &dp->stats; 162 int ret = 0; 163 u32 from = G_TC_FROM(skb->tc_verd); 164 165 stats->rx_packets++; 166 stats->rx_bytes+=skb->len; 167 168 if (!(from & (AT_INGRESS|AT_EGRESS)) || !skb->iif) { 169 dev_kfree_skb(skb); 170 stats->rx_dropped++; 171 return ret; 172 } 173 174 if (skb_queue_len(&dp->rq) >= dev->tx_queue_len) { 175 netif_stop_queue(dev); 176 } 177 178 dev->trans_start = jiffies; 179 skb_queue_tail(&dp->rq, skb); 180 if (!dp->tasklet_pending) { 181 dp->tasklet_pending = 1; 182 tasklet_schedule(&dp->ifb_tasklet); 183 } 184 185 return ret; 186 } 187 188 static struct net_device_stats *ifb_get_stats(struct net_device *dev) 189 { 190 struct ifb_private *dp = netdev_priv(dev); 191 struct net_device_stats *stats = &dp->stats; 192 193 pr_debug("tasklets stats %ld:%ld:%ld:%ld:%ld:%ld:%ld:%ld:%ld \n", 194 dp->st_task_enter, dp->st_txq_refl_try, dp->st_rxq_enter, 195 dp->st_rx2tx_tran, dp->st_rxq_notenter, dp->st_rx_frm_egr, 196 dp->st_rx_frm_ing, dp->st_rxq_check, dp->st_rxq_rsch); 197 198 return stats; 199 } 200 201 static int ifb_close(struct net_device *dev) 202 { 203 struct ifb_private *dp = netdev_priv(dev); 204 205 tasklet_kill(&dp->ifb_tasklet); 206 netif_stop_queue(dev); 207 skb_queue_purge(&dp->rq); 208 skb_queue_purge(&dp->tq); 209 return 0; 210 } 211 212 static int ifb_open(struct net_device *dev) 213 { 214 struct ifb_private *dp = netdev_priv(dev); 215 216 tasklet_init(&dp->ifb_tasklet, ri_tasklet, (unsigned long)dev); 217 skb_queue_head_init(&dp->rq); 218 skb_queue_head_init(&dp->tq); 219 netif_start_queue(dev); 220 221 return 0; 222 } 223 224 static int ifb_validate(struct nlattr *tb[], struct nlattr *data[]) 225 { 226 if (tb[IFLA_ADDRESS]) { 227 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 228 return -EINVAL; 229 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 230 return -EADDRNOTAVAIL; 231 } 232 return 0; 233 } 234 235 static struct rtnl_link_ops ifb_link_ops __read_mostly = { 236 .kind = "ifb", 237 .priv_size = sizeof(struct ifb_private), 238 .setup = ifb_setup, 239 .validate = ifb_validate, 240 }; 241 242 /* Number of ifb devices to be set up by this module. */ 243 module_param(numifbs, int, 0); 244 MODULE_PARM_DESC(numifbs, "Number of ifb devices"); 245 246 static int __init ifb_init_one(int index) 247 { 248 struct net_device *dev_ifb; 249 int err; 250 251 dev_ifb = alloc_netdev(sizeof(struct ifb_private), 252 "ifb%d", ifb_setup); 253 254 if (!dev_ifb) 255 return -ENOMEM; 256 257 err = dev_alloc_name(dev_ifb, dev_ifb->name); 258 if (err < 0) 259 goto err; 260 261 dev_ifb->rtnl_link_ops = &ifb_link_ops; 262 err = register_netdevice(dev_ifb); 263 if (err < 0) 264 goto err; 265 return 0; 266 267 err: 268 free_netdev(dev_ifb); 269 return err; 270 } 271 272 static int __init ifb_init_module(void) 273 { 274 int i, err; 275 276 rtnl_lock(); 277 err = __rtnl_link_register(&ifb_link_ops); 278 279 for (i = 0; i < numifbs && !err; i++) 280 err = ifb_init_one(i); 281 if (err) 282 __rtnl_link_unregister(&ifb_link_ops); 283 rtnl_unlock(); 284 285 return err; 286 } 287 288 static void __exit ifb_cleanup_module(void) 289 { 290 rtnl_link_unregister(&ifb_link_ops); 291 } 292 293 module_init(ifb_init_module); 294 module_exit(ifb_cleanup_module); 295 MODULE_LICENSE("GPL"); 296 MODULE_AUTHOR("Jamal Hadi Salim"); 297 MODULE_ALIAS_RTNL_LINK("ifb"); 298