drivers/net/ifb.c

   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/interrupt.h>
  36 #include <linux/moduleparam.h>
  37 #include <net/pkt_sched.h>
  38 #include <net/net_namespace.h>
  39
  40 #define TX_Q_LIMIT    32
  41 struct ifb_q_private {
  42         struct net_device       *dev;
  43         struct tasklet_struct   ifb_tasklet;
  44         int                     tasklet_pending;
  45         int                     txqnum;
  46         struct sk_buff_head     rq;
  47         u64                     rx_packets;
  48         u64                     rx_bytes;
  49         struct u64_stats_sync   rsync;
  50
  51         struct u64_stats_sync   tsync;
  52         u64                     tx_packets;
  53         u64                     tx_bytes;
  54         struct sk_buff_head     tq;
  55 } ____cacheline_aligned_in_smp;
  56
  57 struct ifb_dev_private {
  58         struct ifb_q_private *tx_private;
  59 };
  60
  61 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev);
  62 static int ifb_open(struct net_device *dev);
  63 static int ifb_close(struct net_device *dev);
  64
  65 static void ifb_ri_tasklet(unsigned long _txp)
  66 {
  67         struct ifb_q_private *txp = (struct ifb_q_private *)_txp;
  68         struct netdev_queue *txq;
  69         struct sk_buff *skb;
  70
  71         txq = netdev_get_tx_queue(txp->dev, txp->txqnum);
  72         skb = skb_peek(&txp->tq);
  73         if (!skb) {
  74                 if (!__netif_tx_trylock(txq))
  75                         goto resched;
  76                 skb_queue_splice_tail_init(&txp->rq, &txp->tq);
  77                 __netif_tx_unlock(txq);
  78         }
  79
  80         while ((skb = __skb_dequeue(&txp->tq)) != NULL) {
  81                 skb->tc_redirected = 0;
  82                 skb->tc_skip_classify = 1;
  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->tc_from_ingress) {
 103                         dev_queue_xmit(skb);
 104                 } else {
 105                         skb_pull(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_init(&txp->ifb_tasklet, ifb_ri_tasklet,
 177                              (unsigned long)txp);
 178                 netif_tx_start_queue(netdev_get_tx_queue(dev, i));
 179         }
 180         return 0;
 181 }
 182
 183 static const struct net_device_ops ifb_netdev_ops = {
 184         .ndo_open       = ifb_open,
 185         .ndo_stop       = ifb_close,
 186         .ndo_get_stats64 = ifb_stats64,
 187         .ndo_start_xmit = ifb_xmit,
 188         .ndo_validate_addr = eth_validate_addr,
 189         .ndo_init       = ifb_dev_init,
 190 };
 191
 192 #define IFB_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG  | NETIF_F_FRAGLIST  | \
 193                       NETIF_F_TSO_ECN | NETIF_F_TSO | NETIF_F_TSO6      | \
 194                       NETIF_F_GSO_ENCAP_ALL                             | \
 195                       NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX         | \
 196                       NETIF_F_HW_VLAN_STAG_TX)
 197
 198 static void ifb_dev_free(struct net_device *dev)
 199 {
 200         struct ifb_dev_private *dp = netdev_priv(dev);
 201         struct ifb_q_private *txp = dp->tx_private;
 202         int i;
 203
 204         for (i = 0; i < dev->num_tx_queues; i++,txp++) {
 205                 tasklet_kill(&txp->ifb_tasklet);
 206                 __skb_queue_purge(&txp->rq);
 207                 __skb_queue_purge(&txp->tq);
 208         }
 209         kfree(dp->tx_private);
 210 }
 211
 212 static void ifb_setup(struct net_device *dev)
 213 {
 214         /* Initialize the device structure. */
 215         dev->netdev_ops = &ifb_netdev_ops;
 216
 217         /* Fill in device structure with ethernet-generic values. */
 218         ether_setup(dev);
 219         dev->tx_queue_len = TX_Q_LIMIT;
 220
 221         dev->features |= IFB_FEATURES;
 222         dev->hw_features |= dev->features;
 223         dev->hw_enc_features |= dev->features;
 224         dev->vlan_features |= IFB_FEATURES & ~(NETIF_F_HW_VLAN_CTAG_TX |
 225                                                NETIF_F_HW_VLAN_STAG_TX);
 226
 227         dev->flags |= IFF_NOARP;
 228         dev->flags &= ~IFF_MULTICAST;
 229         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 230         netif_keep_dst(dev);
 231         eth_hw_addr_random(dev);
 232         dev->needs_free_netdev = true;
 233         dev->priv_destructor = ifb_dev_free;
 234 }
 235
 236 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev)
 237 {
 238         struct ifb_dev_private *dp = netdev_priv(dev);
 239         struct ifb_q_private *txp = dp->tx_private + skb_get_queue_mapping(skb);
 240
 241         u64_stats_update_begin(&txp->rsync);
 242         txp->rx_packets++;
 243         txp->rx_bytes += skb->len;
 244         u64_stats_update_end(&txp->rsync);
 245
 246         if (!skb->tc_redirected || !skb->skb_iif) {
 247                 dev_kfree_skb(skb);
 248                 dev->stats.rx_dropped++;
 249                 return NETDEV_TX_OK;
 250         }
 251
 252         if (skb_queue_len(&txp->rq) >= dev->tx_queue_len)
 253                 netif_tx_stop_queue(netdev_get_tx_queue(dev, txp->txqnum));
 254
 255         __skb_queue_tail(&txp->rq, skb);
 256         if (!txp->tasklet_pending) {
 257                 txp->tasklet_pending = 1;
 258                 tasklet_schedule(&txp->ifb_tasklet);
 259         }
 260
 261         return NETDEV_TX_OK;
 262 }
 263
 264 static int ifb_close(struct net_device *dev)
 265 {
 266         netif_tx_stop_all_queues(dev);
 267         return 0;
 268 }
 269
 270 static int ifb_open(struct net_device *dev)
 271 {
 272         netif_tx_start_all_queues(dev);
 273         return 0;
 274 }
 275
 276 static int ifb_validate(struct nlattr *tb[], struct nlattr *data[],
 277                         struct netlink_ext_ack *extack)
 278 {
 279         if (tb[IFLA_ADDRESS]) {
 280                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 281                         return -EINVAL;
 282                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 283                         return -EADDRNOTAVAIL;
 284         }
 285         return 0;
 286 }
 287
 288 static struct rtnl_link_ops ifb_link_ops __read_mostly = {
 289         .kind           = "ifb",
 290         .priv_size      = sizeof(struct ifb_dev_private),
 291         .setup          = ifb_setup,
 292         .validate       = ifb_validate,
 293 };
 294
 295 /* Number of ifb devices to be set up by this module.
 296  * Note that these legacy devices have one queue.
 297  * Prefer something like : ip link add ifb10 numtxqueues 8 type ifb
 298  */
 299 static int numifbs = 2;
 300 module_param(numifbs, int, 0);
 301 MODULE_PARM_DESC(numifbs, "Number of ifb devices");
 302
 303 static int __init ifb_init_one(int index)
 304 {
 305         struct net_device *dev_ifb;
 306         int err;
 307
 308         dev_ifb = alloc_netdev(sizeof(struct ifb_dev_private), "ifb%d",
 309                                NET_NAME_UNKNOWN, ifb_setup);
 310
 311         if (!dev_ifb)
 312                 return -ENOMEM;
 313
 314         dev_ifb->rtnl_link_ops = &ifb_link_ops;
 315         err = register_netdevice(dev_ifb);
 316         if (err < 0)
 317                 goto err;
 318
 319         return 0;
 320
 321 err:
 322         free_netdev(dev_ifb);
 323         return err;
 324 }
 325
 326 static int __init ifb_init_module(void)
 327 {
 328         int i, err;
 329
 330         rtnl_lock();
 331         err = __rtnl_link_register(&ifb_link_ops);
 332         if (err < 0)
 333                 goto out;
 334
 335         for (i = 0; i < numifbs && !err; i++) {
 336                 err = ifb_init_one(i);
 337                 cond_resched();
 338         }
 339         if (err)
 340                 __rtnl_link_unregister(&ifb_link_ops);
 341
 342 out:
 343         rtnl_unlock();
 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");