drivers/net/veth.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  drivers/net/veth.c
   4  *
   5  *  Copyright (C) 2007 OpenVZ http://openvz.org, SWsoft Inc
   6  *
   7  * Author: Pavel Emelianov <xemul@openvz.org>
   8  * Ethtool interface from: Eric W. Biederman <ebiederm@xmission.com>
   9  *
  10  */
  11
  12 #include <linux/netdevice.h>
  13 #include <linux/slab.h>
  14 #include <linux/ethtool.h>
  15 #include <linux/etherdevice.h>
  16 #include <linux/u64_stats_sync.h>
  17
  18 #include <net/rtnetlink.h>
  19 #include <net/dst.h>
  20 #include <net/xfrm.h>
  21 #include <net/xdp.h>
  22 #include <linux/veth.h>
  23 #include <linux/module.h>
  24 #include <linux/bpf.h>
  25 #include <linux/filter.h>
  26 #include <linux/ptr_ring.h>
  27 #include <linux/bpf_trace.h>
  28 #include <linux/net_tstamp.h>
  29
  30 #define DRV_NAME        "veth"
  31 #define DRV_VERSION     "1.0"
  32
  33 #define VETH_XDP_FLAG           BIT(0)
  34 #define VETH_RING_SIZE          256
  35 #define VETH_XDP_HEADROOM       (XDP_PACKET_HEADROOM + NET_IP_ALIGN)
  36
  37 /* Separating two types of XDP xmit */
  38 #define VETH_XDP_TX             BIT(0)
  39 #define VETH_XDP_REDIR          BIT(1)
  40
  41 #define VETH_XDP_TX_BULK_SIZE   16
  42
  43 struct veth_rq_stats {
  44         u64                     xdp_packets;
  45         u64                     xdp_bytes;
  46         u64                     xdp_drops;
  47         struct u64_stats_sync   syncp;
  48 };
  49
  50 struct veth_rq {
  51         struct napi_struct      xdp_napi;
  52         struct net_device       *dev;
  53         struct bpf_prog __rcu   *xdp_prog;
  54         struct xdp_mem_info     xdp_mem;
  55         struct veth_rq_stats    stats;
  56         bool                    rx_notify_masked;
  57         struct ptr_ring         xdp_ring;
  58         struct xdp_rxq_info     xdp_rxq;
  59 };
  60
  61 struct veth_priv {
  62         struct net_device __rcu *peer;
  63         atomic64_t              dropped;
  64         struct bpf_prog         *_xdp_prog;
  65         struct veth_rq          *rq;
  66         unsigned int            requested_headroom;
  67 };
  68
  69 struct veth_xdp_tx_bq {
  70         struct xdp_frame *q[VETH_XDP_TX_BULK_SIZE];
  71         unsigned int count;
  72 };
  73
  74 /*
  75  * ethtool interface
  76  */
  77
  78 struct veth_q_stat_desc {
  79         char    desc[ETH_GSTRING_LEN];
  80         size_t  offset;
  81 };
  82
  83 #define VETH_RQ_STAT(m) offsetof(struct veth_rq_stats, m)
  84
  85 static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
  86         { "xdp_packets",        VETH_RQ_STAT(xdp_packets) },
  87         { "xdp_bytes",          VETH_RQ_STAT(xdp_bytes) },
  88         { "xdp_drops",          VETH_RQ_STAT(xdp_drops) },
  89 };
  90
  91 #define VETH_RQ_STATS_LEN       ARRAY_SIZE(veth_rq_stats_desc)
  92
  93 static struct {
  94         const char string[ETH_GSTRING_LEN];
  95 } ethtool_stats_keys[] = {
  96         { "peer_ifindex" },
  97 };
  98
  99 static int veth_get_link_ksettings(struct net_device *dev,
 100                                    struct ethtool_link_ksettings *cmd)
 101 {
 102         cmd->base.speed         = SPEED_10000;
 103         cmd->base.duplex        = DUPLEX_FULL;
 104         cmd->base.port          = PORT_TP;
 105         cmd->base.autoneg       = AUTONEG_DISABLE;
 106         return 0;
 107 }
 108
 109 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
 110 {
 111         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
 112         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
 113 }
 114
 115 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
 116 {
 117         char *p = (char *)buf;
 118         int i, j;
 119
 120         switch(stringset) {
 121         case ETH_SS_STATS:
 122                 memcpy(p, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
 123                 p += sizeof(ethtool_stats_keys);
 124                 for (i = 0; i < dev->real_num_rx_queues; i++) {
 125                         for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
 126                                 snprintf(p, ETH_GSTRING_LEN,
 127                                          "rx_queue_%u_%.11s",
 128                                          i, veth_rq_stats_desc[j].desc);
 129                                 p += ETH_GSTRING_LEN;
 130                         }
 131                 }
 132                 break;
 133         }
 134 }
 135
 136 static int veth_get_sset_count(struct net_device *dev, int sset)
 137 {
 138         switch (sset) {
 139         case ETH_SS_STATS:
 140                 return ARRAY_SIZE(ethtool_stats_keys) +
 141                        VETH_RQ_STATS_LEN * dev->real_num_rx_queues;
 142         default:
 143                 return -EOPNOTSUPP;
 144         }
 145 }
 146
 147 static void veth_get_ethtool_stats(struct net_device *dev,
 148                 struct ethtool_stats *stats, u64 *data)
 149 {
 150         struct veth_priv *priv = netdev_priv(dev);
 151         struct net_device *peer = rtnl_dereference(priv->peer);
 152         int i, j, idx;
 153
 154         data[0] = peer ? peer->ifindex : 0;
 155         idx = 1;
 156         for (i = 0; i < dev->real_num_rx_queues; i++) {
 157                 const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
 158                 const void *stats_base = (void *)rq_stats;
 159                 unsigned int start;
 160                 size_t offset;
 161
 162                 do {
 163                         start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
 164                         for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
 165                                 offset = veth_rq_stats_desc[j].offset;
 166                                 data[idx + j] = *(u64 *)(stats_base + offset);
 167                         }
 168                 } while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
 169                 idx += VETH_RQ_STATS_LEN;
 170         }
 171 }
 172
 173 static const struct ethtool_ops veth_ethtool_ops = {
 174         .get_drvinfo            = veth_get_drvinfo,
 175         .get_link               = ethtool_op_get_link,
 176         .get_strings            = veth_get_strings,
 177         .get_sset_count         = veth_get_sset_count,
 178         .get_ethtool_stats      = veth_get_ethtool_stats,
 179         .get_link_ksettings     = veth_get_link_ksettings,
 180         .get_ts_info            = ethtool_op_get_ts_info,
 181 };
 182
 183 /* general routines */
 184
 185 static bool veth_is_xdp_frame(void *ptr)
 186 {
 187         return (unsigned long)ptr & VETH_XDP_FLAG;
 188 }
 189
 190 static void *veth_ptr_to_xdp(void *ptr)
 191 {
 192         return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
 193 }
 194
 195 static void *veth_xdp_to_ptr(void *ptr)
 196 {
 197         return (void *)((unsigned long)ptr | VETH_XDP_FLAG);
 198 }
 199
 200 static void veth_ptr_free(void *ptr)
 201 {
 202         if (veth_is_xdp_frame(ptr))
 203                 xdp_return_frame(veth_ptr_to_xdp(ptr));
 204         else
 205                 kfree_skb(ptr);
 206 }
 207
 208 static void __veth_xdp_flush(struct veth_rq *rq)
 209 {
 210         /* Write ptr_ring before reading rx_notify_masked */
 211         smp_mb();
 212         if (!rq->rx_notify_masked) {
 213                 rq->rx_notify_masked = true;
 214                 napi_schedule(&rq->xdp_napi);
 215         }
 216 }
 217
 218 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
 219 {
 220         if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
 221                 dev_kfree_skb_any(skb);
 222                 return NET_RX_DROP;
 223         }
 224
 225         return NET_RX_SUCCESS;
 226 }
 227
 228 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
 229                             struct veth_rq *rq, bool xdp)
 230 {
 231         return __dev_forward_skb(dev, skb) ?: xdp ?
 232                 veth_xdp_rx(rq, skb) :
 233                 netif_rx(skb);
 234 }
 235
 236 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 237 {
 238         struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 239         struct veth_rq *rq = NULL;
 240         struct net_device *rcv;
 241         int length = skb->len;
 242         bool rcv_xdp = false;
 243         int rxq;
 244
 245         rcu_read_lock();
 246         rcv = rcu_dereference(priv->peer);
 247         if (unlikely(!rcv)) {
 248                 kfree_skb(skb);
 249                 goto drop;
 250         }
 251
 252         rcv_priv = netdev_priv(rcv);
 253         rxq = skb_get_queue_mapping(skb);
 254         if (rxq < rcv->real_num_rx_queues) {
 255                 rq = &rcv_priv->rq[rxq];
 256                 rcv_xdp = rcu_access_pointer(rq->xdp_prog);
 257                 if (rcv_xdp)
 258                         skb_record_rx_queue(skb, rxq);
 259         }
 260
 261         skb_tx_timestamp(skb);
 262         if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
 263                 if (!rcv_xdp) {
 264                         struct pcpu_lstats *stats = this_cpu_ptr(dev->lstats);
 265
 266                         u64_stats_update_begin(&stats->syncp);
 267                         stats->bytes += length;
 268                         stats->packets++;
 269                         u64_stats_update_end(&stats->syncp);
 270                 }
 271         } else {
 272 drop:
 273                 atomic64_inc(&priv->dropped);
 274         }
 275
 276         if (rcv_xdp)
 277                 __veth_xdp_flush(rq);
 278
 279         rcu_read_unlock();
 280
 281         return NETDEV_TX_OK;
 282 }
 283
 284 static u64 veth_stats_tx(struct pcpu_lstats *result, struct net_device *dev)
 285 {
 286         struct veth_priv *priv = netdev_priv(dev);
 287         int cpu;
 288
 289         result->packets = 0;
 290         result->bytes = 0;
 291         for_each_possible_cpu(cpu) {
 292                 struct pcpu_lstats *stats = per_cpu_ptr(dev->lstats, cpu);
 293                 u64 packets, bytes;
 294                 unsigned int start;
 295
 296                 do {
 297                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 298                         packets = stats->packets;
 299                         bytes = stats->bytes;
 300                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 301                 result->packets += packets;
 302                 result->bytes += bytes;
 303         }
 304         return atomic64_read(&priv->dropped);
 305 }
 306
 307 static void veth_stats_rx(struct veth_rq_stats *result, struct net_device *dev)
 308 {
 309         struct veth_priv *priv = netdev_priv(dev);
 310         int i;
 311
 312         result->xdp_packets = 0;
 313         result->xdp_bytes = 0;
 314         result->xdp_drops = 0;
 315         for (i = 0; i < dev->num_rx_queues; i++) {
 316                 struct veth_rq_stats *stats = &priv->rq[i].stats;
 317                 u64 packets, bytes, drops;
 318                 unsigned int start;
 319
 320                 do {
 321                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 322                         packets = stats->xdp_packets;
 323                         bytes = stats->xdp_bytes;
 324                         drops = stats->xdp_drops;
 325                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 326                 result->xdp_packets += packets;
 327                 result->xdp_bytes += bytes;
 328                 result->xdp_drops += drops;
 329         }
 330 }
 331
 332 static void veth_get_stats64(struct net_device *dev,
 333                              struct rtnl_link_stats64 *tot)
 334 {
 335         struct veth_priv *priv = netdev_priv(dev);
 336         struct net_device *peer;
 337         struct veth_rq_stats rx;
 338         struct pcpu_lstats tx;
 339
 340         tot->tx_dropped = veth_stats_tx(&tx, dev);
 341         tot->tx_bytes = tx.bytes;
 342         tot->tx_packets = tx.packets;
 343
 344         veth_stats_rx(&rx, dev);
 345         tot->rx_dropped = rx.xdp_drops;
 346         tot->rx_bytes = rx.xdp_bytes;
 347         tot->rx_packets = rx.xdp_packets;
 348
 349         rcu_read_lock();
 350         peer = rcu_dereference(priv->peer);
 351         if (peer) {
 352                 tot->rx_dropped += veth_stats_tx(&tx, peer);
 353                 tot->rx_bytes += tx.bytes;
 354                 tot->rx_packets += tx.packets;
 355
 356                 veth_stats_rx(&rx, peer);
 357                 tot->tx_bytes += rx.xdp_bytes;
 358                 tot->tx_packets += rx.xdp_packets;
 359         }
 360         rcu_read_unlock();
 361 }
 362
 363 /* fake multicast ability */
 364 static void veth_set_multicast_list(struct net_device *dev)
 365 {
 366 }
 367
 368 static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
 369                                       int buflen)
 370 {
 371         struct sk_buff *skb;
 372
 373         if (!buflen) {
 374                 buflen = SKB_DATA_ALIGN(headroom + len) +
 375                          SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 376         }
 377         skb = build_skb(head, buflen);
 378         if (!skb)
 379                 return NULL;
 380
 381         skb_reserve(skb, headroom);
 382         skb_put(skb, len);
 383
 384         return skb;
 385 }
 386
 387 static int veth_select_rxq(struct net_device *dev)
 388 {
 389         return smp_processor_id() % dev->real_num_rx_queues;
 390 }
 391
 392 static int veth_xdp_xmit(struct net_device *dev, int n,
 393                          struct xdp_frame **frames, u32 flags)
 394 {
 395         struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 396         struct net_device *rcv;
 397         int i, ret, drops = n;
 398         unsigned int max_len;
 399         struct veth_rq *rq;
 400
 401         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
 402                 ret = -EINVAL;
 403                 goto drop;
 404         }
 405
 406         rcv = rcu_dereference(priv->peer);
 407         if (unlikely(!rcv)) {
 408                 ret = -ENXIO;
 409                 goto drop;
 410         }
 411
 412         rcv_priv = netdev_priv(rcv);
 413         rq = &rcv_priv->rq[veth_select_rxq(rcv)];
 414         /* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
 415          * side. This means an XDP program is loaded on the peer and the peer
 416          * device is up.
 417          */
 418         if (!rcu_access_pointer(rq->xdp_prog)) {
 419                 ret = -ENXIO;
 420                 goto drop;
 421         }
 422
 423         drops = 0;
 424         max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
 425
 426         spin_lock(&rq->xdp_ring.producer_lock);
 427         for (i = 0; i < n; i++) {
 428                 struct xdp_frame *frame = frames[i];
 429                 void *ptr = veth_xdp_to_ptr(frame);
 430
 431                 if (unlikely(frame->len > max_len ||
 432                              __ptr_ring_produce(&rq->xdp_ring, ptr))) {
 433                         xdp_return_frame_rx_napi(frame);
 434                         drops++;
 435                 }
 436         }
 437         spin_unlock(&rq->xdp_ring.producer_lock);
 438
 439         if (flags & XDP_XMIT_FLUSH)
 440                 __veth_xdp_flush(rq);
 441
 442         if (likely(!drops))
 443                 return n;
 444
 445         ret = n - drops;
 446 drop:
 447         atomic64_add(drops, &priv->dropped);
 448
 449         return ret;
 450 }
 451
 452 static void veth_xdp_flush_bq(struct net_device *dev, struct veth_xdp_tx_bq *bq)
 453 {
 454         int sent, i, err = 0;
 455
 456         sent = veth_xdp_xmit(dev, bq->count, bq->q, 0);
 457         if (sent < 0) {
 458                 err = sent;
 459                 sent = 0;
 460                 for (i = 0; i < bq->count; i++)
 461                         xdp_return_frame(bq->q[i]);
 462         }
 463         trace_xdp_bulk_tx(dev, sent, bq->count - sent, err);
 464
 465         bq->count = 0;
 466 }
 467
 468 static void veth_xdp_flush(struct net_device *dev, struct veth_xdp_tx_bq *bq)
 469 {
 470         struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 471         struct net_device *rcv;
 472         struct veth_rq *rq;
 473
 474         rcu_read_lock();
 475         veth_xdp_flush_bq(dev, bq);
 476         rcv = rcu_dereference(priv->peer);
 477         if (unlikely(!rcv))
 478                 goto out;
 479
 480         rcv_priv = netdev_priv(rcv);
 481         rq = &rcv_priv->rq[veth_select_rxq(rcv)];
 482         /* xdp_ring is initialized on receive side? */
 483         if (unlikely(!rcu_access_pointer(rq->xdp_prog)))
 484                 goto out;
 485
 486         __veth_xdp_flush(rq);
 487 out:
 488         rcu_read_unlock();
 489 }
 490
 491 static int veth_xdp_tx(struct net_device *dev, struct xdp_buff *xdp,
 492                        struct veth_xdp_tx_bq *bq)
 493 {
 494         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
 495
 496         if (unlikely(!frame))
 497                 return -EOVERFLOW;
 498
 499         if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
 500                 veth_xdp_flush_bq(dev, bq);
 501
 502         bq->q[bq->count++] = frame;
 503
 504         return 0;
 505 }
 506
 507 static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
 508                                         struct xdp_frame *frame,
 509                                         unsigned int *xdp_xmit,
 510                                         struct veth_xdp_tx_bq *bq)
 511 {
 512         void *hard_start = frame->data - frame->headroom;
 513         void *head = hard_start - sizeof(struct xdp_frame);
 514         int len = frame->len, delta = 0;
 515         struct xdp_frame orig_frame;
 516         struct bpf_prog *xdp_prog;
 517         unsigned int headroom;
 518         struct sk_buff *skb;
 519
 520         rcu_read_lock();
 521         xdp_prog = rcu_dereference(rq->xdp_prog);
 522         if (likely(xdp_prog)) {
 523                 struct xdp_buff xdp;
 524                 u32 act;
 525
 526                 xdp.data_hard_start = hard_start;
 527                 xdp.data = frame->data;
 528                 xdp.data_end = frame->data + frame->len;
 529                 xdp.data_meta = frame->data - frame->metasize;
 530                 xdp.rxq = &rq->xdp_rxq;
 531
 532                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
 533
 534                 switch (act) {
 535                 case XDP_PASS:
 536                         delta = frame->data - xdp.data;
 537                         len = xdp.data_end - xdp.data;
 538                         break;
 539                 case XDP_TX:
 540                         orig_frame = *frame;
 541                         xdp.data_hard_start = head;
 542                         xdp.rxq->mem = frame->mem;
 543                         if (unlikely(veth_xdp_tx(rq->dev, &xdp, bq) < 0)) {
 544                                 trace_xdp_exception(rq->dev, xdp_prog, act);
 545                                 frame = &orig_frame;
 546                                 goto err_xdp;
 547                         }
 548                         *xdp_xmit |= VETH_XDP_TX;
 549                         rcu_read_unlock();
 550                         goto xdp_xmit;
 551                 case XDP_REDIRECT:
 552                         orig_frame = *frame;
 553                         xdp.data_hard_start = head;
 554                         xdp.rxq->mem = frame->mem;
 555                         if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
 556                                 frame = &orig_frame;
 557                                 goto err_xdp;
 558                         }
 559                         *xdp_xmit |= VETH_XDP_REDIR;
 560                         rcu_read_unlock();
 561                         goto xdp_xmit;
 562                 default:
 563                         bpf_warn_invalid_xdp_action(act);
 564                         /* fall through */
 565                 case XDP_ABORTED:
 566                         trace_xdp_exception(rq->dev, xdp_prog, act);
 567                         /* fall through */
 568                 case XDP_DROP:
 569                         goto err_xdp;
 570                 }
 571         }
 572         rcu_read_unlock();
 573
 574         headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
 575         skb = veth_build_skb(head, headroom, len, 0);
 576         if (!skb) {
 577                 xdp_return_frame(frame);
 578                 goto err;
 579         }
 580
 581         xdp_release_frame(frame);
 582         xdp_scrub_frame(frame);
 583         skb->protocol = eth_type_trans(skb, rq->dev);
 584 err:
 585         return skb;
 586 err_xdp:
 587         rcu_read_unlock();
 588         xdp_return_frame(frame);
 589 xdp_xmit:
 590         return NULL;
 591 }
 592
 593 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq, struct sk_buff *skb,
 594                                         unsigned int *xdp_xmit,
 595                                         struct veth_xdp_tx_bq *bq)
 596 {
 597         u32 pktlen, headroom, act, metalen;
 598         void *orig_data, *orig_data_end;
 599         struct bpf_prog *xdp_prog;
 600         int mac_len, delta, off;
 601         struct xdp_buff xdp;
 602
 603         skb_orphan(skb);
 604
 605         rcu_read_lock();
 606         xdp_prog = rcu_dereference(rq->xdp_prog);
 607         if (unlikely(!xdp_prog)) {
 608                 rcu_read_unlock();
 609                 goto out;
 610         }
 611
 612         mac_len = skb->data - skb_mac_header(skb);
 613         pktlen = skb->len + mac_len;
 614         headroom = skb_headroom(skb) - mac_len;
 615
 616         if (skb_shared(skb) || skb_head_is_locked(skb) ||
 617             skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
 618                 struct sk_buff *nskb;
 619                 int size, head_off;
 620                 void *head, *start;
 621                 struct page *page;
 622
 623                 size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
 624                        SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 625                 if (size > PAGE_SIZE)
 626                         goto drop;
 627
 628                 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
 629                 if (!page)
 630                         goto drop;
 631
 632                 head = page_address(page);
 633                 start = head + VETH_XDP_HEADROOM;
 634                 if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
 635                         page_frag_free(head);
 636                         goto drop;
 637                 }
 638
 639                 nskb = veth_build_skb(head,
 640                                       VETH_XDP_HEADROOM + mac_len, skb->len,
 641                                       PAGE_SIZE);
 642                 if (!nskb) {
 643                         page_frag_free(head);
 644                         goto drop;
 645                 }
 646
 647                 skb_copy_header(nskb, skb);
 648                 head_off = skb_headroom(nskb) - skb_headroom(skb);
 649                 skb_headers_offset_update(nskb, head_off);
 650                 consume_skb(skb);
 651                 skb = nskb;
 652         }
 653
 654         xdp.data_hard_start = skb->head;
 655         xdp.data = skb_mac_header(skb);
 656         xdp.data_end = xdp.data + pktlen;
 657         xdp.data_meta = xdp.data;
 658         xdp.rxq = &rq->xdp_rxq;
 659         orig_data = xdp.data;
 660         orig_data_end = xdp.data_end;
 661
 662         act = bpf_prog_run_xdp(xdp_prog, &xdp);
 663
 664         switch (act) {
 665         case XDP_PASS:
 666                 break;
 667         case XDP_TX:
 668                 get_page(virt_to_page(xdp.data));
 669                 consume_skb(skb);
 670                 xdp.rxq->mem = rq->xdp_mem;
 671                 if (unlikely(veth_xdp_tx(rq->dev, &xdp, bq) < 0)) {
 672                         trace_xdp_exception(rq->dev, xdp_prog, act);
 673                         goto err_xdp;
 674                 }
 675                 *xdp_xmit |= VETH_XDP_TX;
 676                 rcu_read_unlock();
 677                 goto xdp_xmit;
 678         case XDP_REDIRECT:
 679                 get_page(virt_to_page(xdp.data));
 680                 consume_skb(skb);
 681                 xdp.rxq->mem = rq->xdp_mem;
 682                 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog))
 683                         goto err_xdp;
 684                 *xdp_xmit |= VETH_XDP_REDIR;
 685                 rcu_read_unlock();
 686                 goto xdp_xmit;
 687         default:
 688                 bpf_warn_invalid_xdp_action(act);
 689                 /* fall through */
 690         case XDP_ABORTED:
 691                 trace_xdp_exception(rq->dev, xdp_prog, act);
 692                 /* fall through */
 693         case XDP_DROP:
 694                 goto drop;
 695         }
 696         rcu_read_unlock();
 697
 698         delta = orig_data - xdp.data;
 699         off = mac_len + delta;
 700         if (off > 0)
 701                 __skb_push(skb, off);
 702         else if (off < 0)
 703                 __skb_pull(skb, -off);
 704         skb->mac_header -= delta;
 705         off = xdp.data_end - orig_data_end;
 706         if (off != 0)
 707                 __skb_put(skb, off);
 708         skb->protocol = eth_type_trans(skb, rq->dev);
 709
 710         metalen = xdp.data - xdp.data_meta;
 711         if (metalen)
 712                 skb_metadata_set(skb, metalen);
 713 out:
 714         return skb;
 715 drop:
 716         rcu_read_unlock();
 717         kfree_skb(skb);
 718         return NULL;
 719 err_xdp:
 720         rcu_read_unlock();
 721         page_frag_free(xdp.data);
 722 xdp_xmit:
 723         return NULL;
 724 }
 725
 726 static int veth_xdp_rcv(struct veth_rq *rq, int budget, unsigned int *xdp_xmit,
 727                         struct veth_xdp_tx_bq *bq)
 728 {
 729         int i, done = 0, drops = 0, bytes = 0;
 730
 731         for (i = 0; i < budget; i++) {
 732                 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
 733                 unsigned int xdp_xmit_one = 0;
 734                 struct sk_buff *skb;
 735
 736                 if (!ptr)
 737                         break;
 738
 739                 if (veth_is_xdp_frame(ptr)) {
 740                         struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
 741
 742                         bytes += frame->len;
 743                         skb = veth_xdp_rcv_one(rq, frame, &xdp_xmit_one, bq);
 744                 } else {
 745                         skb = ptr;
 746                         bytes += skb->len;
 747                         skb = veth_xdp_rcv_skb(rq, skb, &xdp_xmit_one, bq);
 748                 }
 749                 *xdp_xmit |= xdp_xmit_one;
 750
 751                 if (skb)
 752                         napi_gro_receive(&rq->xdp_napi, skb);
 753                 else if (!xdp_xmit_one)
 754                         drops++;
 755
 756                 done++;
 757         }
 758
 759         u64_stats_update_begin(&rq->stats.syncp);
 760         rq->stats.xdp_packets += done;
 761         rq->stats.xdp_bytes += bytes;
 762         rq->stats.xdp_drops += drops;
 763         u64_stats_update_end(&rq->stats.syncp);
 764
 765         return done;
 766 }
 767
 768 static int veth_poll(struct napi_struct *napi, int budget)
 769 {
 770         struct veth_rq *rq =
 771                 container_of(napi, struct veth_rq, xdp_napi);
 772         unsigned int xdp_xmit = 0;
 773         struct veth_xdp_tx_bq bq;
 774         int done;
 775
 776         bq.count = 0;
 777
 778         xdp_set_return_frame_no_direct();
 779         done = veth_xdp_rcv(rq, budget, &xdp_xmit, &bq);
 780
 781         if (done < budget && napi_complete_done(napi, done)) {
 782                 /* Write rx_notify_masked before reading ptr_ring */
 783                 smp_store_mb(rq->rx_notify_masked, false);
 784                 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
 785                         rq->rx_notify_masked = true;
 786                         napi_schedule(&rq->xdp_napi);
 787                 }
 788         }
 789
 790         if (xdp_xmit & VETH_XDP_TX)
 791                 veth_xdp_flush(rq->dev, &bq);
 792         if (xdp_xmit & VETH_XDP_REDIR)
 793                 xdp_do_flush_map();
 794         xdp_clear_return_frame_no_direct();
 795
 796         return done;
 797 }
 798
 799 static int veth_napi_add(struct net_device *dev)
 800 {
 801         struct veth_priv *priv = netdev_priv(dev);
 802         int err, i;
 803
 804         for (i = 0; i < dev->real_num_rx_queues; i++) {
 805                 struct veth_rq *rq = &priv->rq[i];
 806
 807                 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
 808                 if (err)
 809                         goto err_xdp_ring;
 810         }
 811
 812         for (i = 0; i < dev->real_num_rx_queues; i++) {
 813                 struct veth_rq *rq = &priv->rq[i];
 814
 815                 netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
 816                 napi_enable(&rq->xdp_napi);
 817         }
 818
 819         return 0;
 820 err_xdp_ring:
 821         for (i--; i >= 0; i--)
 822                 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
 823
 824         return err;
 825 }
 826
 827 static void veth_napi_del(struct net_device *dev)
 828 {
 829         struct veth_priv *priv = netdev_priv(dev);
 830         int i;
 831
 832         for (i = 0; i < dev->real_num_rx_queues; i++) {
 833                 struct veth_rq *rq = &priv->rq[i];
 834
 835                 napi_disable(&rq->xdp_napi);
 836                 napi_hash_del(&rq->xdp_napi);
 837         }
 838         synchronize_net();
 839
 840         for (i = 0; i < dev->real_num_rx_queues; i++) {
 841                 struct veth_rq *rq = &priv->rq[i];
 842
 843                 netif_napi_del(&rq->xdp_napi);
 844                 rq->rx_notify_masked = false;
 845                 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
 846         }
 847 }
 848
 849 static int veth_enable_xdp(struct net_device *dev)
 850 {
 851         struct veth_priv *priv = netdev_priv(dev);
 852         int err, i;
 853
 854         if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
 855                 for (i = 0; i < dev->real_num_rx_queues; i++) {
 856                         struct veth_rq *rq = &priv->rq[i];
 857
 858                         err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
 859                         if (err < 0)
 860                                 goto err_rxq_reg;
 861
 862                         err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
 863                                                          MEM_TYPE_PAGE_SHARED,
 864                                                          NULL);
 865                         if (err < 0)
 866                                 goto err_reg_mem;
 867
 868                         /* Save original mem info as it can be overwritten */
 869                         rq->xdp_mem = rq->xdp_rxq.mem;
 870                 }
 871
 872                 err = veth_napi_add(dev);
 873                 if (err)
 874                         goto err_rxq_reg;
 875         }
 876
 877         for (i = 0; i < dev->real_num_rx_queues; i++)
 878                 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
 879
 880         return 0;
 881 err_reg_mem:
 882         xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
 883 err_rxq_reg:
 884         for (i--; i >= 0; i--)
 885                 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
 886
 887         return err;
 888 }
 889
 890 static void veth_disable_xdp(struct net_device *dev)
 891 {
 892         struct veth_priv *priv = netdev_priv(dev);
 893         int i;
 894
 895         for (i = 0; i < dev->real_num_rx_queues; i++)
 896                 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
 897         veth_napi_del(dev);
 898         for (i = 0; i < dev->real_num_rx_queues; i++) {
 899                 struct veth_rq *rq = &priv->rq[i];
 900
 901                 rq->xdp_rxq.mem = rq->xdp_mem;
 902                 xdp_rxq_info_unreg(&rq->xdp_rxq);
 903         }
 904 }
 905
 906 static int veth_open(struct net_device *dev)
 907 {
 908         struct veth_priv *priv = netdev_priv(dev);
 909         struct net_device *peer = rtnl_dereference(priv->peer);
 910         int err;
 911
 912         if (!peer)
 913                 return -ENOTCONN;
 914
 915         if (priv->_xdp_prog) {
 916                 err = veth_enable_xdp(dev);
 917                 if (err)
 918                         return err;
 919         }
 920
 921         if (peer->flags & IFF_UP) {
 922                 netif_carrier_on(dev);
 923                 netif_carrier_on(peer);
 924         }
 925
 926         return 0;
 927 }
 928
 929 static int veth_close(struct net_device *dev)
 930 {
 931         struct veth_priv *priv = netdev_priv(dev);
 932         struct net_device *peer = rtnl_dereference(priv->peer);
 933
 934         netif_carrier_off(dev);
 935         if (peer)
 936                 netif_carrier_off(peer);
 937
 938         if (priv->_xdp_prog)
 939                 veth_disable_xdp(dev);
 940
 941         return 0;
 942 }
 943
 944 static int is_valid_veth_mtu(int mtu)
 945 {
 946         return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
 947 }
 948
 949 static int veth_alloc_queues(struct net_device *dev)
 950 {
 951         struct veth_priv *priv = netdev_priv(dev);
 952         int i;
 953
 954         priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
 955         if (!priv->rq)
 956                 return -ENOMEM;
 957
 958         for (i = 0; i < dev->num_rx_queues; i++) {
 959                 priv->rq[i].dev = dev;
 960                 u64_stats_init(&priv->rq[i].stats.syncp);
 961         }
 962
 963         return 0;
 964 }
 965
 966 static void veth_free_queues(struct net_device *dev)
 967 {
 968         struct veth_priv *priv = netdev_priv(dev);
 969
 970         kfree(priv->rq);
 971 }
 972
 973 static int veth_dev_init(struct net_device *dev)
 974 {
 975         int err;
 976
 977         dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
 978         if (!dev->lstats)
 979                 return -ENOMEM;
 980
 981         err = veth_alloc_queues(dev);
 982         if (err) {
 983                 free_percpu(dev->lstats);
 984                 return err;
 985         }
 986
 987         return 0;
 988 }
 989
 990 static void veth_dev_free(struct net_device *dev)
 991 {
 992         veth_free_queues(dev);
 993         free_percpu(dev->lstats);
 994 }
 995
 996 #ifdef CONFIG_NET_POLL_CONTROLLER
 997 static void veth_poll_controller(struct net_device *dev)
 998 {
 999         /* veth only receives frames when its peer sends one
1000          * Since it has nothing to do with disabling irqs, we are guaranteed
1001          * never to have pending data when we poll for it so
1002          * there is nothing to do here.
1003          *
1004          * We need this though so netpoll recognizes us as an interface that
1005          * supports polling, which enables bridge devices in virt setups to
1006          * still use netconsole
1007          */
1008 }
1009 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1010
1011 static int veth_get_iflink(const struct net_device *dev)
1012 {
1013         struct veth_priv *priv = netdev_priv(dev);
1014         struct net_device *peer;
1015         int iflink;
1016
1017         rcu_read_lock();
1018         peer = rcu_dereference(priv->peer);
1019         iflink = peer ? peer->ifindex : 0;
1020         rcu_read_unlock();
1021
1022         return iflink;
1023 }
1024
1025 static netdev_features_t veth_fix_features(struct net_device *dev,
1026                                            netdev_features_t features)
1027 {
1028         struct veth_priv *priv = netdev_priv(dev);
1029         struct net_device *peer;
1030
1031         peer = rtnl_dereference(priv->peer);
1032         if (peer) {
1033                 struct veth_priv *peer_priv = netdev_priv(peer);
1034
1035                 if (peer_priv->_xdp_prog)
1036                         features &= ~NETIF_F_GSO_SOFTWARE;
1037         }
1038
1039         return features;
1040 }
1041
1042 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1043 {
1044         struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1045         struct net_device *peer;
1046
1047         if (new_hr < 0)
1048                 new_hr = 0;
1049
1050         rcu_read_lock();
1051         peer = rcu_dereference(priv->peer);
1052         if (unlikely(!peer))
1053                 goto out;
1054
1055         peer_priv = netdev_priv(peer);
1056         priv->requested_headroom = new_hr;
1057         new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1058         dev->needed_headroom = new_hr;
1059         peer->needed_headroom = new_hr;
1060
1061 out:
1062         rcu_read_unlock();
1063 }
1064
1065 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1066                         struct netlink_ext_ack *extack)
1067 {
1068         struct veth_priv *priv = netdev_priv(dev);
1069         struct bpf_prog *old_prog;
1070         struct net_device *peer;
1071         unsigned int max_mtu;
1072         int err;
1073
1074         old_prog = priv->_xdp_prog;
1075         priv->_xdp_prog = prog;
1076         peer = rtnl_dereference(priv->peer);
1077
1078         if (prog) {
1079                 if (!peer) {
1080                         NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1081                         err = -ENOTCONN;
1082                         goto err;
1083                 }
1084
1085                 max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1086                           peer->hard_header_len -
1087                           SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1088                 if (peer->mtu > max_mtu) {
1089                         NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1090                         err = -ERANGE;
1091                         goto err;
1092                 }
1093
1094                 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1095                         NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1096                         err = -ENOSPC;
1097                         goto err;
1098                 }
1099
1100                 if (dev->flags & IFF_UP) {
1101                         err = veth_enable_xdp(dev);
1102                         if (err) {
1103                                 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1104                                 goto err;
1105                         }
1106                 }
1107
1108                 if (!old_prog) {
1109                         peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1110                         peer->max_mtu = max_mtu;
1111                 }
1112         }
1113
1114         if (old_prog) {
1115                 if (!prog) {
1116                         if (dev->flags & IFF_UP)
1117                                 veth_disable_xdp(dev);
1118
1119                         if (peer) {
1120                                 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1121                                 peer->max_mtu = ETH_MAX_MTU;
1122                         }
1123                 }
1124                 bpf_prog_put(old_prog);
1125         }
1126
1127         if ((!!old_prog ^ !!prog) && peer)
1128                 netdev_update_features(peer);
1129
1130         return 0;
1131 err:
1132         priv->_xdp_prog = old_prog;
1133
1134         return err;
1135 }
1136
1137 static u32 veth_xdp_query(struct net_device *dev)
1138 {
1139         struct veth_priv *priv = netdev_priv(dev);
1140         const struct bpf_prog *xdp_prog;
1141
1142         xdp_prog = priv->_xdp_prog;
1143         if (xdp_prog)
1144                 return xdp_prog->aux->id;
1145
1146         return 0;
1147 }
1148
1149 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1150 {
1151         switch (xdp->command) {
1152         case XDP_SETUP_PROG:
1153                 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1154         case XDP_QUERY_PROG:
1155                 xdp->prog_id = veth_xdp_query(dev);
1156                 return 0;
1157         default:
1158                 return -EINVAL;
1159         }
1160 }
1161
1162 static const struct net_device_ops veth_netdev_ops = {
1163         .ndo_init            = veth_dev_init,
1164         .ndo_open            = veth_open,
1165         .ndo_stop            = veth_close,
1166         .ndo_start_xmit      = veth_xmit,
1167         .ndo_get_stats64     = veth_get_stats64,
1168         .ndo_set_rx_mode     = veth_set_multicast_list,
1169         .ndo_set_mac_address = eth_mac_addr,
1170 #ifdef CONFIG_NET_POLL_CONTROLLER
1171         .ndo_poll_controller    = veth_poll_controller,
1172 #endif
1173         .ndo_get_iflink         = veth_get_iflink,
1174         .ndo_fix_features       = veth_fix_features,
1175         .ndo_features_check     = passthru_features_check,
1176         .ndo_set_rx_headroom    = veth_set_rx_headroom,
1177         .ndo_bpf                = veth_xdp,
1178         .ndo_xdp_xmit           = veth_xdp_xmit,
1179 };
1180
1181 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1182                        NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1183                        NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1184                        NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1185                        NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1186
1187 static void veth_setup(struct net_device *dev)
1188 {
1189         ether_setup(dev);
1190
1191         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1192         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1193         dev->priv_flags |= IFF_NO_QUEUE;
1194         dev->priv_flags |= IFF_PHONY_HEADROOM;
1195
1196         dev->netdev_ops = &veth_netdev_ops;
1197         dev->ethtool_ops = &veth_ethtool_ops;
1198         dev->features |= NETIF_F_LLTX;
1199         dev->features |= VETH_FEATURES;
1200         dev->vlan_features = dev->features &
1201                              ~(NETIF_F_HW_VLAN_CTAG_TX |
1202                                NETIF_F_HW_VLAN_STAG_TX |
1203                                NETIF_F_HW_VLAN_CTAG_RX |
1204                                NETIF_F_HW_VLAN_STAG_RX);
1205         dev->needs_free_netdev = true;
1206         dev->priv_destructor = veth_dev_free;
1207         dev->max_mtu = ETH_MAX_MTU;
1208
1209         dev->hw_features = VETH_FEATURES;
1210         dev->hw_enc_features = VETH_FEATURES;
1211         dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1212 }
1213
1214 /*
1215  * netlink interface
1216  */
1217
1218 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1219                          struct netlink_ext_ack *extack)
1220 {
1221         if (tb[IFLA_ADDRESS]) {
1222                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1223                         return -EINVAL;
1224                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1225                         return -EADDRNOTAVAIL;
1226         }
1227         if (tb[IFLA_MTU]) {
1228                 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1229                         return -EINVAL;
1230         }
1231         return 0;
1232 }
1233
1234 static struct rtnl_link_ops veth_link_ops;
1235
1236 static int veth_newlink(struct net *src_net, struct net_device *dev,
1237                         struct nlattr *tb[], struct nlattr *data[],
1238                         struct netlink_ext_ack *extack)
1239 {
1240         int err;
1241         struct net_device *peer;
1242         struct veth_priv *priv;
1243         char ifname[IFNAMSIZ];
1244         struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1245         unsigned char name_assign_type;
1246         struct ifinfomsg *ifmp;
1247         struct net *net;
1248
1249         /*
1250          * create and register peer first
1251          */
1252         if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1253                 struct nlattr *nla_peer;
1254
1255                 nla_peer = data[VETH_INFO_PEER];
1256                 ifmp = nla_data(nla_peer);
1257                 err = rtnl_nla_parse_ifla(peer_tb,
1258                                           nla_data(nla_peer) + sizeof(struct ifinfomsg),
1259                                           nla_len(nla_peer) - sizeof(struct ifinfomsg),
1260                                           NULL);
1261                 if (err < 0)
1262                         return err;
1263
1264                 err = veth_validate(peer_tb, NULL, extack);
1265                 if (err < 0)
1266                         return err;
1267
1268                 tbp = peer_tb;
1269         } else {
1270                 ifmp = NULL;
1271                 tbp = tb;
1272         }
1273
1274         if (ifmp && tbp[IFLA_IFNAME]) {
1275                 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1276                 name_assign_type = NET_NAME_USER;
1277         } else {
1278                 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1279                 name_assign_type = NET_NAME_ENUM;
1280         }
1281
1282         net = rtnl_link_get_net(src_net, tbp);
1283         if (IS_ERR(net))
1284                 return PTR_ERR(net);
1285
1286         peer = rtnl_create_link(net, ifname, name_assign_type,
1287                                 &veth_link_ops, tbp, extack);
1288         if (IS_ERR(peer)) {
1289                 put_net(net);
1290                 return PTR_ERR(peer);
1291         }
1292
1293         if (!ifmp || !tbp[IFLA_ADDRESS])
1294                 eth_hw_addr_random(peer);
1295
1296         if (ifmp && (dev->ifindex != 0))
1297                 peer->ifindex = ifmp->ifi_index;
1298
1299         peer->gso_max_size = dev->gso_max_size;
1300         peer->gso_max_segs = dev->gso_max_segs;
1301
1302         err = register_netdevice(peer);
1303         put_net(net);
1304         net = NULL;
1305         if (err < 0)
1306                 goto err_register_peer;
1307
1308         netif_carrier_off(peer);
1309
1310         err = rtnl_configure_link(peer, ifmp);
1311         if (err < 0)
1312                 goto err_configure_peer;
1313
1314         /*
1315          * register dev last
1316          *
1317          * note, that since we've registered new device the dev's name
1318          * should be re-allocated
1319          */
1320
1321         if (tb[IFLA_ADDRESS] == NULL)
1322                 eth_hw_addr_random(dev);
1323
1324         if (tb[IFLA_IFNAME])
1325                 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1326         else
1327                 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1328
1329         err = register_netdevice(dev);
1330         if (err < 0)
1331                 goto err_register_dev;
1332
1333         netif_carrier_off(dev);
1334
1335         /*
1336          * tie the deviced together
1337          */
1338
1339         priv = netdev_priv(dev);
1340         rcu_assign_pointer(priv->peer, peer);
1341
1342         priv = netdev_priv(peer);
1343         rcu_assign_pointer(priv->peer, dev);
1344
1345         return 0;
1346
1347 err_register_dev:
1348         /* nothing to do */
1349 err_configure_peer:
1350         unregister_netdevice(peer);
1351         return err;
1352
1353 err_register_peer:
1354         free_netdev(peer);
1355         return err;
1356 }
1357
1358 static void veth_dellink(struct net_device *dev, struct list_head *head)
1359 {
1360         struct veth_priv *priv;
1361         struct net_device *peer;
1362
1363         priv = netdev_priv(dev);
1364         peer = rtnl_dereference(priv->peer);
1365
1366         /* Note : dellink() is called from default_device_exit_batch(),
1367          * before a rcu_synchronize() point. The devices are guaranteed
1368          * not being freed before one RCU grace period.
1369          */
1370         RCU_INIT_POINTER(priv->peer, NULL);
1371         unregister_netdevice_queue(dev, head);
1372
1373         if (peer) {
1374                 priv = netdev_priv(peer);
1375                 RCU_INIT_POINTER(priv->peer, NULL);
1376                 unregister_netdevice_queue(peer, head);
1377         }
1378 }
1379
1380 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1381         [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
1382 };
1383
1384 static struct net *veth_get_link_net(const struct net_device *dev)
1385 {
1386         struct veth_priv *priv = netdev_priv(dev);
1387         struct net_device *peer = rtnl_dereference(priv->peer);
1388
1389         return peer ? dev_net(peer) : dev_net(dev);
1390 }
1391
1392 static struct rtnl_link_ops veth_link_ops = {
1393         .kind           = DRV_NAME,
1394         .priv_size      = sizeof(struct veth_priv),
1395         .setup          = veth_setup,
1396         .validate       = veth_validate,
1397         .newlink        = veth_newlink,
1398         .dellink        = veth_dellink,
1399         .policy         = veth_policy,
1400         .maxtype        = VETH_INFO_MAX,
1401         .get_link_net   = veth_get_link_net,
1402 };
1403
1404 /*
1405  * init/fini
1406  */
1407
1408 static __init int veth_init(void)
1409 {
1410         return rtnl_link_register(&veth_link_ops);
1411 }
1412
1413 static __exit void veth_exit(void)
1414 {
1415         rtnl_link_unregister(&veth_link_ops);
1416 }
1417
1418 module_init(veth_init);
1419 module_exit(veth_exit);
1420
1421 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1422 MODULE_LICENSE("GPL v2");
1423 MODULE_ALIAS_RTNL_LINK(DRV_NAME);