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