1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* drivers/net/ifb.c: 3 4 The purpose of this driver is to provide a device that allows 5 for sharing of resources: 6 7 1) qdiscs/policies that are per device as opposed to system wide. 8 ifb allows for a device which can be redirected to thus providing 9 an impression of sharing. 10 11 2) Allows for queueing incoming traffic for shaping instead of 12 dropping. 13 14 The original concept is based on what is known as the IMQ 15 driver initially written by Martin Devera, later rewritten 16 by Patrick McHardy and then maintained by Andre Correa. 17 18 You need the tc action mirror or redirect to feed this device 19 packets. 20 21 22 Authors: Jamal Hadi Salim (2005) 23 24 */ 25 26 27 #include <linux/module.h> 28 #include <linux/kernel.h> 29 #include <linux/netdevice.h> 30 #include <linux/etherdevice.h> 31 #include <linux/init.h> 32 #include <linux/interrupt.h> 33 #include <linux/moduleparam.h> 34 #include <linux/netfilter_netdev.h> 35 #include <net/pkt_sched.h> 36 #include <net/net_namespace.h> 37 38 #define TX_Q_LIMIT 32 39 struct ifb_q_private { 40 struct net_device *dev; 41 struct tasklet_struct ifb_tasklet; 42 int tasklet_pending; 43 int txqnum; 44 struct sk_buff_head rq; 45 u64 rx_packets; 46 u64 rx_bytes; 47 struct u64_stats_sync rsync; 48 49 struct u64_stats_sync tsync; 50 u64 tx_packets; 51 u64 tx_bytes; 52 struct sk_buff_head tq; 53 } ____cacheline_aligned_in_smp; 54 55 struct ifb_dev_private { 56 struct ifb_q_private *tx_private; 57 }; 58 59 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev); 60 static int ifb_open(struct net_device *dev); 61 static int ifb_close(struct net_device *dev); 62 63 static void ifb_ri_tasklet(struct tasklet_struct *t) 64 { 65 struct ifb_q_private *txp = from_tasklet(txp, t, ifb_tasklet); 66 struct netdev_queue *txq; 67 struct sk_buff *skb; 68 69 txq = netdev_get_tx_queue(txp->dev, txp->txqnum); 70 skb = skb_peek(&txp->tq); 71 if (!skb) { 72 if (!__netif_tx_trylock(txq)) 73 goto resched; 74 skb_queue_splice_tail_init(&txp->rq, &txp->tq); 75 __netif_tx_unlock(txq); 76 } 77 78 while ((skb = __skb_dequeue(&txp->tq)) != NULL) { 79 /* Skip tc and netfilter to prevent redirection loop. */ 80 skb->redirected = 0; 81 skb->tc_skip_classify = 1; 82 nf_skip_egress(skb, true); 83 84 u64_stats_update_begin(&txp->tsync); 85 txp->tx_packets++; 86 txp->tx_bytes += skb->len; 87 u64_stats_update_end(&txp->tsync); 88 89 rcu_read_lock(); 90 skb->dev = dev_get_by_index_rcu(dev_net(txp->dev), skb->skb_iif); 91 if (!skb->dev) { 92 rcu_read_unlock(); 93 dev_kfree_skb(skb); 94 txp->dev->stats.tx_dropped++; 95 if (skb_queue_len(&txp->tq) != 0) 96 goto resched; 97 break; 98 } 99 rcu_read_unlock(); 100 skb->skb_iif = txp->dev->ifindex; 101 102 if (!skb->from_ingress) { 103 dev_queue_xmit(skb); 104 } else { 105 skb_pull_rcsum(skb, skb->mac_len); 106 netif_receive_skb(skb); 107 } 108 } 109 110 if (__netif_tx_trylock(txq)) { 111 skb = skb_peek(&txp->rq); 112 if (!skb) { 113 txp->tasklet_pending = 0; 114 if (netif_tx_queue_stopped(txq)) 115 netif_tx_wake_queue(txq); 116 } else { 117 __netif_tx_unlock(txq); 118 goto resched; 119 } 120 __netif_tx_unlock(txq); 121 } else { 122 resched: 123 txp->tasklet_pending = 1; 124 tasklet_schedule(&txp->ifb_tasklet); 125 } 126 127 } 128 129 static void ifb_stats64(struct net_device *dev, 130 struct rtnl_link_stats64 *stats) 131 { 132 struct ifb_dev_private *dp = netdev_priv(dev); 133 struct ifb_q_private *txp = dp->tx_private; 134 unsigned int start; 135 u64 packets, bytes; 136 int i; 137 138 for (i = 0; i < dev->num_tx_queues; i++,txp++) { 139 do { 140 start = u64_stats_fetch_begin_irq(&txp->rsync); 141 packets = txp->rx_packets; 142 bytes = txp->rx_bytes; 143 } while (u64_stats_fetch_retry_irq(&txp->rsync, start)); 144 stats->rx_packets += packets; 145 stats->rx_bytes += bytes; 146 147 do { 148 start = u64_stats_fetch_begin_irq(&txp->tsync); 149 packets = txp->tx_packets; 150 bytes = txp->tx_bytes; 151 } while (u64_stats_fetch_retry_irq(&txp->tsync, start)); 152 stats->tx_packets += packets; 153 stats->tx_bytes += bytes; 154 } 155 stats->rx_dropped = dev->stats.rx_dropped; 156 stats->tx_dropped = dev->stats.tx_dropped; 157 } 158 159 static int ifb_dev_init(struct net_device *dev) 160 { 161 struct ifb_dev_private *dp = netdev_priv(dev); 162 struct ifb_q_private *txp; 163 int i; 164 165 txp = kcalloc(dev->num_tx_queues, sizeof(*txp), GFP_KERNEL); 166 if (!txp) 167 return -ENOMEM; 168 dp->tx_private = txp; 169 for (i = 0; i < dev->num_tx_queues; i++,txp++) { 170 txp->txqnum = i; 171 txp->dev = dev; 172 __skb_queue_head_init(&txp->rq); 173 __skb_queue_head_init(&txp->tq); 174 u64_stats_init(&txp->rsync); 175 u64_stats_init(&txp->tsync); 176 tasklet_setup(&txp->ifb_tasklet, ifb_ri_tasklet); 177 netif_tx_start_queue(netdev_get_tx_queue(dev, i)); 178 } 179 return 0; 180 } 181 182 static const struct net_device_ops ifb_netdev_ops = { 183 .ndo_open = ifb_open, 184 .ndo_stop = ifb_close, 185 .ndo_get_stats64 = ifb_stats64, 186 .ndo_start_xmit = ifb_xmit, 187 .ndo_validate_addr = eth_validate_addr, 188 .ndo_init = ifb_dev_init, 189 }; 190 191 #define IFB_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | NETIF_F_FRAGLIST | \ 192 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \ 193 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX | \ 194 NETIF_F_HW_VLAN_STAG_TX) 195 196 static void ifb_dev_free(struct net_device *dev) 197 { 198 struct ifb_dev_private *dp = netdev_priv(dev); 199 struct ifb_q_private *txp = dp->tx_private; 200 int i; 201 202 for (i = 0; i < dev->num_tx_queues; i++,txp++) { 203 tasklet_kill(&txp->ifb_tasklet); 204 __skb_queue_purge(&txp->rq); 205 __skb_queue_purge(&txp->tq); 206 } 207 kfree(dp->tx_private); 208 } 209 210 static void ifb_setup(struct net_device *dev) 211 { 212 /* Initialize the device structure. */ 213 dev->netdev_ops = &ifb_netdev_ops; 214 215 /* Fill in device structure with ethernet-generic values. */ 216 ether_setup(dev); 217 dev->tx_queue_len = TX_Q_LIMIT; 218 219 dev->features |= IFB_FEATURES; 220 dev->hw_features |= dev->features; 221 dev->hw_enc_features |= dev->features; 222 dev->vlan_features |= IFB_FEATURES & ~(NETIF_F_HW_VLAN_CTAG_TX | 223 NETIF_F_HW_VLAN_STAG_TX); 224 225 dev->flags |= IFF_NOARP; 226 dev->flags &= ~IFF_MULTICAST; 227 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 228 netif_keep_dst(dev); 229 eth_hw_addr_random(dev); 230 dev->needs_free_netdev = true; 231 dev->priv_destructor = ifb_dev_free; 232 233 dev->min_mtu = 0; 234 dev->max_mtu = 0; 235 } 236 237 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev) 238 { 239 struct ifb_dev_private *dp = netdev_priv(dev); 240 struct ifb_q_private *txp = dp->tx_private + skb_get_queue_mapping(skb); 241 242 u64_stats_update_begin(&txp->rsync); 243 txp->rx_packets++; 244 txp->rx_bytes += skb->len; 245 u64_stats_update_end(&txp->rsync); 246 247 if (!skb->redirected || !skb->skb_iif) { 248 dev_kfree_skb(skb); 249 dev->stats.rx_dropped++; 250 return NETDEV_TX_OK; 251 } 252 253 if (skb_queue_len(&txp->rq) >= dev->tx_queue_len) 254 netif_tx_stop_queue(netdev_get_tx_queue(dev, txp->txqnum)); 255 256 __skb_queue_tail(&txp->rq, skb); 257 if (!txp->tasklet_pending) { 258 txp->tasklet_pending = 1; 259 tasklet_schedule(&txp->ifb_tasklet); 260 } 261 262 return NETDEV_TX_OK; 263 } 264 265 static int ifb_close(struct net_device *dev) 266 { 267 netif_tx_stop_all_queues(dev); 268 return 0; 269 } 270 271 static int ifb_open(struct net_device *dev) 272 { 273 netif_tx_start_all_queues(dev); 274 return 0; 275 } 276 277 static int ifb_validate(struct nlattr *tb[], struct nlattr *data[], 278 struct netlink_ext_ack *extack) 279 { 280 if (tb[IFLA_ADDRESS]) { 281 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 282 return -EINVAL; 283 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 284 return -EADDRNOTAVAIL; 285 } 286 return 0; 287 } 288 289 static struct rtnl_link_ops ifb_link_ops __read_mostly = { 290 .kind = "ifb", 291 .priv_size = sizeof(struct ifb_dev_private), 292 .setup = ifb_setup, 293 .validate = ifb_validate, 294 }; 295 296 /* Number of ifb devices to be set up by this module. 297 * Note that these legacy devices have one queue. 298 * Prefer something like : ip link add ifb10 numtxqueues 8 type ifb 299 */ 300 static int numifbs = 2; 301 module_param(numifbs, int, 0); 302 MODULE_PARM_DESC(numifbs, "Number of ifb devices"); 303 304 static int __init ifb_init_one(int index) 305 { 306 struct net_device *dev_ifb; 307 int err; 308 309 dev_ifb = alloc_netdev(sizeof(struct ifb_dev_private), "ifb%d", 310 NET_NAME_UNKNOWN, ifb_setup); 311 312 if (!dev_ifb) 313 return -ENOMEM; 314 315 dev_ifb->rtnl_link_ops = &ifb_link_ops; 316 err = register_netdevice(dev_ifb); 317 if (err < 0) 318 goto err; 319 320 return 0; 321 322 err: 323 free_netdev(dev_ifb); 324 return err; 325 } 326 327 static int __init ifb_init_module(void) 328 { 329 int i, err; 330 331 down_write(&pernet_ops_rwsem); 332 rtnl_lock(); 333 err = __rtnl_link_register(&ifb_link_ops); 334 if (err < 0) 335 goto out; 336 337 for (i = 0; i < numifbs && !err; i++) { 338 err = ifb_init_one(i); 339 cond_resched(); 340 } 341 if (err) 342 __rtnl_link_unregister(&ifb_link_ops); 343 344 out: 345 rtnl_unlock(); 346 up_write(&pernet_ops_rwsem); 347 348 return err; 349 } 350 351 static void __exit ifb_cleanup_module(void) 352 { 353 rtnl_link_unregister(&ifb_link_ops); 354 } 355 356 module_init(ifb_init_module); 357 module_exit(ifb_cleanup_module); 358 MODULE_LICENSE("GPL"); 359 MODULE_AUTHOR("Jamal Hadi Salim"); 360 MODULE_ALIAS_RTNL_LINK("ifb"); 361