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 #define VETH_XDP_TX_BULK_SIZE   16
  38 #define VETH_XDP_BATCH          16
  39
  40 struct veth_stats {
  41         u64     rx_drops;
  42         /* xdp */
  43         u64     xdp_packets;
  44         u64     xdp_bytes;
  45         u64     xdp_redirect;
  46         u64     xdp_drops;
  47         u64     xdp_tx;
  48         u64     xdp_tx_err;
  49         u64     peer_tq_xdp_xmit;
  50         u64     peer_tq_xdp_xmit_err;
  51 };
  52
  53 struct veth_rq_stats {
  54         struct veth_stats       vs;
  55         struct u64_stats_sync   syncp;
  56 };
  57
  58 struct veth_rq {
  59         struct napi_struct      xdp_napi;
  60         struct napi_struct __rcu *napi; /* points to xdp_napi when the latter is initialized */
  61         struct net_device       *dev;
  62         struct bpf_prog __rcu   *xdp_prog;
  63         struct xdp_mem_info     xdp_mem;
  64         struct veth_rq_stats    stats;
  65         bool                    rx_notify_masked;
  66         struct ptr_ring         xdp_ring;
  67         struct xdp_rxq_info     xdp_rxq;
  68 };
  69
  70 struct veth_priv {
  71         struct net_device __rcu *peer;
  72         atomic64_t              dropped;
  73         struct bpf_prog         *_xdp_prog;
  74         struct veth_rq          *rq;
  75         unsigned int            requested_headroom;
  76 };
  77
  78 struct veth_xdp_tx_bq {
  79         struct xdp_frame *q[VETH_XDP_TX_BULK_SIZE];
  80         unsigned int count;
  81 };
  82
  83 /*
  84  * ethtool interface
  85  */
  86
  87 struct veth_q_stat_desc {
  88         char    desc[ETH_GSTRING_LEN];
  89         size_t  offset;
  90 };
  91
  92 #define VETH_RQ_STAT(m) offsetof(struct veth_stats, m)
  93
  94 static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
  95         { "xdp_packets",        VETH_RQ_STAT(xdp_packets) },
  96         { "xdp_bytes",          VETH_RQ_STAT(xdp_bytes) },
  97         { "drops",              VETH_RQ_STAT(rx_drops) },
  98         { "xdp_redirect",       VETH_RQ_STAT(xdp_redirect) },
  99         { "xdp_drops",          VETH_RQ_STAT(xdp_drops) },
 100         { "xdp_tx",             VETH_RQ_STAT(xdp_tx) },
 101         { "xdp_tx_errors",      VETH_RQ_STAT(xdp_tx_err) },
 102 };
 103
 104 #define VETH_RQ_STATS_LEN       ARRAY_SIZE(veth_rq_stats_desc)
 105
 106 static const struct veth_q_stat_desc veth_tq_stats_desc[] = {
 107         { "xdp_xmit",           VETH_RQ_STAT(peer_tq_xdp_xmit) },
 108         { "xdp_xmit_errors",    VETH_RQ_STAT(peer_tq_xdp_xmit_err) },
 109 };
 110
 111 #define VETH_TQ_STATS_LEN       ARRAY_SIZE(veth_tq_stats_desc)
 112
 113 static struct {
 114         const char string[ETH_GSTRING_LEN];
 115 } ethtool_stats_keys[] = {
 116         { "peer_ifindex" },
 117 };
 118
 119 struct veth_xdp_buff {
 120         struct xdp_buff xdp;
 121         struct sk_buff *skb;
 122 };
 123
 124 static int veth_get_link_ksettings(struct net_device *dev,
 125                                    struct ethtool_link_ksettings *cmd)
 126 {
 127         cmd->base.speed         = SPEED_10000;
 128         cmd->base.duplex        = DUPLEX_FULL;
 129         cmd->base.port          = PORT_TP;
 130         cmd->base.autoneg       = AUTONEG_DISABLE;
 131         return 0;
 132 }
 133
 134 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
 135 {
 136         strscpy(info->driver, DRV_NAME, sizeof(info->driver));
 137         strscpy(info->version, DRV_VERSION, sizeof(info->version));
 138 }
 139
 140 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
 141 {
 142         u8 *p = buf;
 143         int i, j;
 144
 145         switch(stringset) {
 146         case ETH_SS_STATS:
 147                 memcpy(p, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
 148                 p += sizeof(ethtool_stats_keys);
 149                 for (i = 0; i < dev->real_num_rx_queues; i++)
 150                         for (j = 0; j < VETH_RQ_STATS_LEN; j++)
 151                                 ethtool_sprintf(&p, "rx_queue_%u_%.18s",
 152                                                 i, veth_rq_stats_desc[j].desc);
 153
 154                 for (i = 0; i < dev->real_num_tx_queues; i++)
 155                         for (j = 0; j < VETH_TQ_STATS_LEN; j++)
 156                                 ethtool_sprintf(&p, "tx_queue_%u_%.18s",
 157                                                 i, veth_tq_stats_desc[j].desc);
 158                 break;
 159         }
 160 }
 161
 162 static int veth_get_sset_count(struct net_device *dev, int sset)
 163 {
 164         switch (sset) {
 165         case ETH_SS_STATS:
 166                 return ARRAY_SIZE(ethtool_stats_keys) +
 167                        VETH_RQ_STATS_LEN * dev->real_num_rx_queues +
 168                        VETH_TQ_STATS_LEN * dev->real_num_tx_queues;
 169         default:
 170                 return -EOPNOTSUPP;
 171         }
 172 }
 173
 174 static void veth_get_ethtool_stats(struct net_device *dev,
 175                 struct ethtool_stats *stats, u64 *data)
 176 {
 177         struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 178         struct net_device *peer = rtnl_dereference(priv->peer);
 179         int i, j, idx;
 180
 181         data[0] = peer ? peer->ifindex : 0;
 182         idx = 1;
 183         for (i = 0; i < dev->real_num_rx_queues; i++) {
 184                 const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
 185                 const void *stats_base = (void *)&rq_stats->vs;
 186                 unsigned int start;
 187                 size_t offset;
 188
 189                 do {
 190                         start = u64_stats_fetch_begin(&rq_stats->syncp);
 191                         for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
 192                                 offset = veth_rq_stats_desc[j].offset;
 193                                 data[idx + j] = *(u64 *)(stats_base + offset);
 194                         }
 195                 } while (u64_stats_fetch_retry(&rq_stats->syncp, start));
 196                 idx += VETH_RQ_STATS_LEN;
 197         }
 198
 199         if (!peer)
 200                 return;
 201
 202         rcv_priv = netdev_priv(peer);
 203         for (i = 0; i < peer->real_num_rx_queues; i++) {
 204                 const struct veth_rq_stats *rq_stats = &rcv_priv->rq[i].stats;
 205                 const void *base = (void *)&rq_stats->vs;
 206                 unsigned int start, tx_idx = idx;
 207                 size_t offset;
 208
 209                 tx_idx += (i % dev->real_num_tx_queues) * VETH_TQ_STATS_LEN;
 210                 do {
 211                         start = u64_stats_fetch_begin(&rq_stats->syncp);
 212                         for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
 213                                 offset = veth_tq_stats_desc[j].offset;
 214                                 data[tx_idx + j] += *(u64 *)(base + offset);
 215                         }
 216                 } while (u64_stats_fetch_retry(&rq_stats->syncp, start));
 217         }
 218 }
 219
 220 static void veth_get_channels(struct net_device *dev,
 221                               struct ethtool_channels *channels)
 222 {
 223         channels->tx_count = dev->real_num_tx_queues;
 224         channels->rx_count = dev->real_num_rx_queues;
 225         channels->max_tx = dev->num_tx_queues;
 226         channels->max_rx = dev->num_rx_queues;
 227 }
 228
 229 static int veth_set_channels(struct net_device *dev,
 230                              struct ethtool_channels *ch);
 231
 232 static const struct ethtool_ops veth_ethtool_ops = {
 233         .get_drvinfo            = veth_get_drvinfo,
 234         .get_link               = ethtool_op_get_link,
 235         .get_strings            = veth_get_strings,
 236         .get_sset_count         = veth_get_sset_count,
 237         .get_ethtool_stats      = veth_get_ethtool_stats,
 238         .get_link_ksettings     = veth_get_link_ksettings,
 239         .get_ts_info            = ethtool_op_get_ts_info,
 240         .get_channels           = veth_get_channels,
 241         .set_channels           = veth_set_channels,
 242 };
 243
 244 /* general routines */
 245
 246 static bool veth_is_xdp_frame(void *ptr)
 247 {
 248         return (unsigned long)ptr & VETH_XDP_FLAG;
 249 }
 250
 251 static struct xdp_frame *veth_ptr_to_xdp(void *ptr)
 252 {
 253         return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
 254 }
 255
 256 static void *veth_xdp_to_ptr(struct xdp_frame *xdp)
 257 {
 258         return (void *)((unsigned long)xdp | VETH_XDP_FLAG);
 259 }
 260
 261 static void veth_ptr_free(void *ptr)
 262 {
 263         if (veth_is_xdp_frame(ptr))
 264                 xdp_return_frame(veth_ptr_to_xdp(ptr));
 265         else
 266                 kfree_skb(ptr);
 267 }
 268
 269 static void __veth_xdp_flush(struct veth_rq *rq)
 270 {
 271         /* Write ptr_ring before reading rx_notify_masked */
 272         smp_mb();
 273         if (!READ_ONCE(rq->rx_notify_masked) &&
 274             napi_schedule_prep(&rq->xdp_napi)) {
 275                 WRITE_ONCE(rq->rx_notify_masked, true);
 276                 __napi_schedule(&rq->xdp_napi);
 277         }
 278 }
 279
 280 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
 281 {
 282         if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
 283                 dev_kfree_skb_any(skb);
 284                 return NET_RX_DROP;
 285         }
 286
 287         return NET_RX_SUCCESS;
 288 }
 289
 290 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
 291                             struct veth_rq *rq, bool xdp)
 292 {
 293         return __dev_forward_skb(dev, skb) ?: xdp ?
 294                 veth_xdp_rx(rq, skb) :
 295                 __netif_rx(skb);
 296 }
 297
 298 /* return true if the specified skb has chances of GRO aggregation
 299  * Don't strive for accuracy, but try to avoid GRO overhead in the most
 300  * common scenarios.
 301  * When XDP is enabled, all traffic is considered eligible, as the xmit
 302  * device has TSO off.
 303  * When TSO is enabled on the xmit device, we are likely interested only
 304  * in UDP aggregation, explicitly check for that if the skb is suspected
 305  * - the sock_wfree destructor is used by UDP, ICMP and XDP sockets -
 306  * to belong to locally generated UDP traffic.
 307  */
 308 static bool veth_skb_is_eligible_for_gro(const struct net_device *dev,
 309                                          const struct net_device *rcv,
 310                                          const struct sk_buff *skb)
 311 {
 312         return !(dev->features & NETIF_F_ALL_TSO) ||
 313                 (skb->destructor == sock_wfree &&
 314                  rcv->features & (NETIF_F_GRO_FRAGLIST | NETIF_F_GRO_UDP_FWD));
 315 }
 316
 317 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 318 {
 319         struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 320         struct veth_rq *rq = NULL;
 321         struct net_device *rcv;
 322         int length = skb->len;
 323         bool use_napi = false;
 324         int rxq;
 325
 326         rcu_read_lock();
 327         rcv = rcu_dereference(priv->peer);
 328         if (unlikely(!rcv) || !pskb_may_pull(skb, ETH_HLEN)) {
 329                 kfree_skb(skb);
 330                 goto drop;
 331         }
 332
 333         rcv_priv = netdev_priv(rcv);
 334         rxq = skb_get_queue_mapping(skb);
 335         if (rxq < rcv->real_num_rx_queues) {
 336                 rq = &rcv_priv->rq[rxq];
 337
 338                 /* The napi pointer is available when an XDP program is
 339                  * attached or when GRO is enabled
 340                  * Don't bother with napi/GRO if the skb can't be aggregated
 341                  */
 342                 use_napi = rcu_access_pointer(rq->napi) &&
 343                            veth_skb_is_eligible_for_gro(dev, rcv, skb);
 344         }
 345
 346         skb_tx_timestamp(skb);
 347         if (likely(veth_forward_skb(rcv, skb, rq, use_napi) == NET_RX_SUCCESS)) {
 348                 if (!use_napi)
 349                         dev_lstats_add(dev, length);
 350         } else {
 351 drop:
 352                 atomic64_inc(&priv->dropped);
 353         }
 354
 355         if (use_napi)
 356                 __veth_xdp_flush(rq);
 357
 358         rcu_read_unlock();
 359
 360         return NETDEV_TX_OK;
 361 }
 362
 363 static u64 veth_stats_tx(struct net_device *dev, u64 *packets, u64 *bytes)
 364 {
 365         struct veth_priv *priv = netdev_priv(dev);
 366
 367         dev_lstats_read(dev, packets, bytes);
 368         return atomic64_read(&priv->dropped);
 369 }
 370
 371 static void veth_stats_rx(struct veth_stats *result, struct net_device *dev)
 372 {
 373         struct veth_priv *priv = netdev_priv(dev);
 374         int i;
 375
 376         result->peer_tq_xdp_xmit_err = 0;
 377         result->xdp_packets = 0;
 378         result->xdp_tx_err = 0;
 379         result->xdp_bytes = 0;
 380         result->rx_drops = 0;
 381         for (i = 0; i < dev->num_rx_queues; i++) {
 382                 u64 packets, bytes, drops, xdp_tx_err, peer_tq_xdp_xmit_err;
 383                 struct veth_rq_stats *stats = &priv->rq[i].stats;
 384                 unsigned int start;
 385
 386                 do {
 387                         start = u64_stats_fetch_begin(&stats->syncp);
 388                         peer_tq_xdp_xmit_err = stats->vs.peer_tq_xdp_xmit_err;
 389                         xdp_tx_err = stats->vs.xdp_tx_err;
 390                         packets = stats->vs.xdp_packets;
 391                         bytes = stats->vs.xdp_bytes;
 392                         drops = stats->vs.rx_drops;
 393                 } while (u64_stats_fetch_retry(&stats->syncp, start));
 394                 result->peer_tq_xdp_xmit_err += peer_tq_xdp_xmit_err;
 395                 result->xdp_tx_err += xdp_tx_err;
 396                 result->xdp_packets += packets;
 397                 result->xdp_bytes += bytes;
 398                 result->rx_drops += drops;
 399         }
 400 }
 401
 402 static void veth_get_stats64(struct net_device *dev,
 403                              struct rtnl_link_stats64 *tot)
 404 {
 405         struct veth_priv *priv = netdev_priv(dev);
 406         struct net_device *peer;
 407         struct veth_stats rx;
 408         u64 packets, bytes;
 409
 410         tot->tx_dropped = veth_stats_tx(dev, &packets, &bytes);
 411         tot->tx_bytes = bytes;
 412         tot->tx_packets = packets;
 413
 414         veth_stats_rx(&rx, dev);
 415         tot->tx_dropped += rx.xdp_tx_err;
 416         tot->rx_dropped = rx.rx_drops + rx.peer_tq_xdp_xmit_err;
 417         tot->rx_bytes = rx.xdp_bytes;
 418         tot->rx_packets = rx.xdp_packets;
 419
 420         rcu_read_lock();
 421         peer = rcu_dereference(priv->peer);
 422         if (peer) {
 423                 veth_stats_tx(peer, &packets, &bytes);
 424                 tot->rx_bytes += bytes;
 425                 tot->rx_packets += packets;
 426
 427                 veth_stats_rx(&rx, peer);
 428                 tot->tx_dropped += rx.peer_tq_xdp_xmit_err;
 429                 tot->rx_dropped += rx.xdp_tx_err;
 430                 tot->tx_bytes += rx.xdp_bytes;
 431                 tot->tx_packets += rx.xdp_packets;
 432         }
 433         rcu_read_unlock();
 434 }
 435
 436 /* fake multicast ability */
 437 static void veth_set_multicast_list(struct net_device *dev)
 438 {
 439 }
 440
 441 static int veth_select_rxq(struct net_device *dev)
 442 {
 443         return smp_processor_id() % dev->real_num_rx_queues;
 444 }
 445
 446 static struct net_device *veth_peer_dev(struct net_device *dev)
 447 {
 448         struct veth_priv *priv = netdev_priv(dev);
 449
 450         /* Callers must be under RCU read side. */
 451         return rcu_dereference(priv->peer);
 452 }
 453
 454 static int veth_xdp_xmit(struct net_device *dev, int n,
 455                          struct xdp_frame **frames,
 456                          u32 flags, bool ndo_xmit)
 457 {
 458         struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 459         int i, ret = -ENXIO, nxmit = 0;
 460         struct net_device *rcv;
 461         unsigned int max_len;
 462         struct veth_rq *rq;
 463
 464         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 465                 return -EINVAL;
 466
 467         rcu_read_lock();
 468         rcv = rcu_dereference(priv->peer);
 469         if (unlikely(!rcv))
 470                 goto out;
 471
 472         rcv_priv = netdev_priv(rcv);
 473         rq = &rcv_priv->rq[veth_select_rxq(rcv)];
 474         /* The napi pointer is set if NAPI is enabled, which ensures that
 475          * xdp_ring is initialized on receive side and the peer device is up.
 476          */
 477         if (!rcu_access_pointer(rq->napi))
 478                 goto out;
 479
 480         max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
 481
 482         spin_lock(&rq->xdp_ring.producer_lock);
 483         for (i = 0; i < n; i++) {
 484                 struct xdp_frame *frame = frames[i];
 485                 void *ptr = veth_xdp_to_ptr(frame);
 486
 487                 if (unlikely(xdp_get_frame_len(frame) > max_len ||
 488                              __ptr_ring_produce(&rq->xdp_ring, ptr)))
 489                         break;
 490                 nxmit++;
 491         }
 492         spin_unlock(&rq->xdp_ring.producer_lock);
 493
 494         if (flags & XDP_XMIT_FLUSH)
 495                 __veth_xdp_flush(rq);
 496
 497         ret = nxmit;
 498         if (ndo_xmit) {
 499                 u64_stats_update_begin(&rq->stats.syncp);
 500                 rq->stats.vs.peer_tq_xdp_xmit += nxmit;
 501                 rq->stats.vs.peer_tq_xdp_xmit_err += n - nxmit;
 502                 u64_stats_update_end(&rq->stats.syncp);
 503         }
 504
 505 out:
 506         rcu_read_unlock();
 507
 508         return ret;
 509 }
 510
 511 static int veth_ndo_xdp_xmit(struct net_device *dev, int n,
 512                              struct xdp_frame **frames, u32 flags)
 513 {
 514         int err;
 515
 516         err = veth_xdp_xmit(dev, n, frames, flags, true);
 517         if (err < 0) {
 518                 struct veth_priv *priv = netdev_priv(dev);
 519
 520                 atomic64_add(n, &priv->dropped);
 521         }
 522
 523         return err;
 524 }
 525
 526 static void veth_xdp_flush_bq(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
 527 {
 528         int sent, i, err = 0, drops;
 529
 530         sent = veth_xdp_xmit(rq->dev, bq->count, bq->q, 0, false);
 531         if (sent < 0) {
 532                 err = sent;
 533                 sent = 0;
 534         }
 535
 536         for (i = sent; unlikely(i < bq->count); i++)
 537                 xdp_return_frame(bq->q[i]);
 538
 539         drops = bq->count - sent;
 540         trace_xdp_bulk_tx(rq->dev, sent, drops, err);
 541
 542         u64_stats_update_begin(&rq->stats.syncp);
 543         rq->stats.vs.xdp_tx += sent;
 544         rq->stats.vs.xdp_tx_err += drops;
 545         u64_stats_update_end(&rq->stats.syncp);
 546
 547         bq->count = 0;
 548 }
 549
 550 static void veth_xdp_flush(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
 551 {
 552         struct veth_priv *rcv_priv, *priv = netdev_priv(rq->dev);
 553         struct net_device *rcv;
 554         struct veth_rq *rcv_rq;
 555
 556         rcu_read_lock();
 557         veth_xdp_flush_bq(rq, bq);
 558         rcv = rcu_dereference(priv->peer);
 559         if (unlikely(!rcv))
 560                 goto out;
 561
 562         rcv_priv = netdev_priv(rcv);
 563         rcv_rq = &rcv_priv->rq[veth_select_rxq(rcv)];
 564         /* xdp_ring is initialized on receive side? */
 565         if (unlikely(!rcu_access_pointer(rcv_rq->xdp_prog)))
 566                 goto out;
 567
 568         __veth_xdp_flush(rcv_rq);
 569 out:
 570         rcu_read_unlock();
 571 }
 572
 573 static int veth_xdp_tx(struct veth_rq *rq, struct xdp_buff *xdp,
 574                        struct veth_xdp_tx_bq *bq)
 575 {
 576         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
 577
 578         if (unlikely(!frame))
 579                 return -EOVERFLOW;
 580
 581         if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
 582                 veth_xdp_flush_bq(rq, bq);
 583
 584         bq->q[bq->count++] = frame;
 585
 586         return 0;
 587 }
 588
 589 static struct xdp_frame *veth_xdp_rcv_one(struct veth_rq *rq,
 590                                           struct xdp_frame *frame,
 591                                           struct veth_xdp_tx_bq *bq,
 592                                           struct veth_stats *stats)
 593 {
 594         struct xdp_frame orig_frame;
 595         struct bpf_prog *xdp_prog;
 596
 597         rcu_read_lock();
 598         xdp_prog = rcu_dereference(rq->xdp_prog);
 599         if (likely(xdp_prog)) {
 600                 struct veth_xdp_buff vxbuf;
 601                 struct xdp_buff *xdp = &vxbuf.xdp;
 602                 u32 act;
 603
 604                 xdp_convert_frame_to_buff(frame, xdp);
 605                 xdp->rxq = &rq->xdp_rxq;
 606                 vxbuf.skb = NULL;
 607
 608                 act = bpf_prog_run_xdp(xdp_prog, xdp);
 609
 610                 switch (act) {
 611                 case XDP_PASS:
 612                         if (xdp_update_frame_from_buff(xdp, frame))
 613                                 goto err_xdp;
 614                         break;
 615                 case XDP_TX:
 616                         orig_frame = *frame;
 617                         xdp->rxq->mem = frame->mem;
 618                         if (unlikely(veth_xdp_tx(rq, xdp, bq) < 0)) {
 619                                 trace_xdp_exception(rq->dev, xdp_prog, act);
 620                                 frame = &orig_frame;
 621                                 stats->rx_drops++;
 622                                 goto err_xdp;
 623                         }
 624                         stats->xdp_tx++;
 625                         rcu_read_unlock();
 626                         goto xdp_xmit;
 627                 case XDP_REDIRECT:
 628                         orig_frame = *frame;
 629                         xdp->rxq->mem = frame->mem;
 630                         if (xdp_do_redirect(rq->dev, xdp, xdp_prog)) {
 631                                 frame = &orig_frame;
 632                                 stats->rx_drops++;
 633                                 goto err_xdp;
 634                         }
 635                         stats->xdp_redirect++;
 636                         rcu_read_unlock();
 637                         goto xdp_xmit;
 638                 default:
 639                         bpf_warn_invalid_xdp_action(rq->dev, xdp_prog, act);
 640                         fallthrough;
 641                 case XDP_ABORTED:
 642                         trace_xdp_exception(rq->dev, xdp_prog, act);
 643                         fallthrough;
 644                 case XDP_DROP:
 645                         stats->xdp_drops++;
 646                         goto err_xdp;
 647                 }
 648         }
 649         rcu_read_unlock();
 650
 651         return frame;
 652 err_xdp:
 653         rcu_read_unlock();
 654         xdp_return_frame(frame);
 655 xdp_xmit:
 656         return NULL;
 657 }
 658
 659 /* frames array contains VETH_XDP_BATCH at most */
 660 static void veth_xdp_rcv_bulk_skb(struct veth_rq *rq, void **frames,
 661                                   int n_xdpf, struct veth_xdp_tx_bq *bq,
 662                                   struct veth_stats *stats)
 663 {
 664         void *skbs[VETH_XDP_BATCH];
 665         int i;
 666
 667         if (xdp_alloc_skb_bulk(skbs, n_xdpf,
 668                                GFP_ATOMIC | __GFP_ZERO) < 0) {
 669                 for (i = 0; i < n_xdpf; i++)
 670                         xdp_return_frame(frames[i]);
 671                 stats->rx_drops += n_xdpf;
 672
 673                 return;
 674         }
 675
 676         for (i = 0; i < n_xdpf; i++) {
 677                 struct sk_buff *skb = skbs[i];
 678
 679                 skb = __xdp_build_skb_from_frame(frames[i], skb,
 680                                                  rq->dev);
 681                 if (!skb) {
 682                         xdp_return_frame(frames[i]);
 683                         stats->rx_drops++;
 684                         continue;
 685                 }
 686                 napi_gro_receive(&rq->xdp_napi, skb);
 687         }
 688 }
 689
 690 static void veth_xdp_get(struct xdp_buff *xdp)
 691 {
 692         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
 693         int i;
 694
 695         get_page(virt_to_page(xdp->data));
 696         if (likely(!xdp_buff_has_frags(xdp)))
 697                 return;
 698
 699         for (i = 0; i < sinfo->nr_frags; i++)
 700                 __skb_frag_ref(&sinfo->frags[i]);
 701 }
 702
 703 static int veth_convert_skb_to_xdp_buff(struct veth_rq *rq,
 704                                         struct xdp_buff *xdp,
 705                                         struct sk_buff **pskb)
 706 {
 707         struct sk_buff *skb = *pskb;
 708         u32 frame_sz;
 709
 710         if (skb_shared(skb) || skb_head_is_locked(skb) ||
 711             skb_shinfo(skb)->nr_frags) {
 712                 u32 size, len, max_head_size, off;
 713                 struct sk_buff *nskb;
 714                 struct page *page;
 715                 int i, head_off;
 716
 717                 /* We need a private copy of the skb and data buffers since
 718                  * the ebpf program can modify it. We segment the original skb
 719                  * into order-0 pages without linearize it.
 720                  *
 721                  * Make sure we have enough space for linear and paged area
 722                  */
 723                 max_head_size = SKB_WITH_OVERHEAD(PAGE_SIZE -
 724                                                   VETH_XDP_HEADROOM);
 725                 if (skb->len > PAGE_SIZE * MAX_SKB_FRAGS + max_head_size)
 726                         goto drop;
 727
 728                 /* Allocate skb head */
 729                 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
 730                 if (!page)
 731                         goto drop;
 732
 733                 nskb = build_skb(page_address(page), PAGE_SIZE);
 734                 if (!nskb) {
 735                         put_page(page);
 736                         goto drop;
 737                 }
 738
 739                 skb_reserve(nskb, VETH_XDP_HEADROOM);
 740                 size = min_t(u32, skb->len, max_head_size);
 741                 if (skb_copy_bits(skb, 0, nskb->data, size)) {
 742                         consume_skb(nskb);
 743                         goto drop;
 744                 }
 745                 skb_put(nskb, size);
 746
 747                 skb_copy_header(nskb, skb);
 748                 head_off = skb_headroom(nskb) - skb_headroom(skb);
 749                 skb_headers_offset_update(nskb, head_off);
 750
 751                 /* Allocate paged area of new skb */
 752                 off = size;
 753                 len = skb->len - off;
 754
 755                 for (i = 0; i < MAX_SKB_FRAGS && off < skb->len; i++) {
 756                         page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
 757                         if (!page) {
 758                                 consume_skb(nskb);
 759                                 goto drop;
 760                         }
 761
 762                         size = min_t(u32, len, PAGE_SIZE);
 763                         skb_add_rx_frag(nskb, i, page, 0, size, PAGE_SIZE);
 764                         if (skb_copy_bits(skb, off, page_address(page),
 765                                           size)) {
 766                                 consume_skb(nskb);
 767                                 goto drop;
 768                         }
 769
 770                         len -= size;
 771                         off += size;
 772                 }
 773
 774                 consume_skb(skb);
 775                 skb = nskb;
 776         } else if (skb_headroom(skb) < XDP_PACKET_HEADROOM &&
 777                    pskb_expand_head(skb, VETH_XDP_HEADROOM, 0, GFP_ATOMIC)) {
 778                 goto drop;
 779         }
 780
 781         /* SKB "head" area always have tailroom for skb_shared_info */
 782         frame_sz = skb_end_pointer(skb) - skb->head;
 783         frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 784         xdp_init_buff(xdp, frame_sz, &rq->xdp_rxq);
 785         xdp_prepare_buff(xdp, skb->head, skb_headroom(skb),
 786                          skb_headlen(skb), true);
 787
 788         if (skb_is_nonlinear(skb)) {
 789                 skb_shinfo(skb)->xdp_frags_size = skb->data_len;
 790                 xdp_buff_set_frags_flag(xdp);
 791         } else {
 792                 xdp_buff_clear_frags_flag(xdp);
 793         }
 794         *pskb = skb;
 795
 796         return 0;
 797 drop:
 798         consume_skb(skb);
 799         *pskb = NULL;
 800
 801         return -ENOMEM;
 802 }
 803
 804 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
 805                                         struct sk_buff *skb,
 806                                         struct veth_xdp_tx_bq *bq,
 807                                         struct veth_stats *stats)
 808 {
 809         void *orig_data, *orig_data_end;
 810         struct bpf_prog *xdp_prog;
 811         struct veth_xdp_buff vxbuf;
 812         struct xdp_buff *xdp = &vxbuf.xdp;
 813         u32 act, metalen;
 814         int off;
 815
 816         skb_prepare_for_gro(skb);
 817
 818         rcu_read_lock();
 819         xdp_prog = rcu_dereference(rq->xdp_prog);
 820         if (unlikely(!xdp_prog)) {
 821                 rcu_read_unlock();
 822                 goto out;
 823         }
 824
 825         __skb_push(skb, skb->data - skb_mac_header(skb));
 826         if (veth_convert_skb_to_xdp_buff(rq, xdp, &skb))
 827                 goto drop;
 828         vxbuf.skb = skb;
 829
 830         orig_data = xdp->data;
 831         orig_data_end = xdp->data_end;
 832
 833         act = bpf_prog_run_xdp(xdp_prog, xdp);
 834
 835         switch (act) {
 836         case XDP_PASS:
 837                 break;
 838         case XDP_TX:
 839                 veth_xdp_get(xdp);
 840                 consume_skb(skb);
 841                 xdp->rxq->mem = rq->xdp_mem;
 842                 if (unlikely(veth_xdp_tx(rq, xdp, bq) < 0)) {
 843                         trace_xdp_exception(rq->dev, xdp_prog, act);
 844                         stats->rx_drops++;
 845                         goto err_xdp;
 846                 }
 847                 stats->xdp_tx++;
 848                 rcu_read_unlock();
 849                 goto xdp_xmit;
 850         case XDP_REDIRECT:
 851                 veth_xdp_get(xdp);
 852                 consume_skb(skb);
 853                 xdp->rxq->mem = rq->xdp_mem;
 854                 if (xdp_do_redirect(rq->dev, xdp, xdp_prog)) {
 855                         stats->rx_drops++;
 856                         goto err_xdp;
 857                 }
 858                 stats->xdp_redirect++;
 859                 rcu_read_unlock();
 860                 goto xdp_xmit;
 861         default:
 862                 bpf_warn_invalid_xdp_action(rq->dev, xdp_prog, act);
 863                 fallthrough;
 864         case XDP_ABORTED:
 865                 trace_xdp_exception(rq->dev, xdp_prog, act);
 866                 fallthrough;
 867         case XDP_DROP:
 868                 stats->xdp_drops++;
 869                 goto xdp_drop;
 870         }
 871         rcu_read_unlock();
 872
 873         /* check if bpf_xdp_adjust_head was used */
 874         off = orig_data - xdp->data;
 875         if (off > 0)
 876                 __skb_push(skb, off);
 877         else if (off < 0)
 878                 __skb_pull(skb, -off);
 879
 880         skb_reset_mac_header(skb);
 881
 882         /* check if bpf_xdp_adjust_tail was used */
 883         off = xdp->data_end - orig_data_end;
 884         if (off != 0)
 885                 __skb_put(skb, off); /* positive on grow, negative on shrink */
 886
 887         /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
 888          * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
 889          */
 890         if (xdp_buff_has_frags(xdp))
 891                 skb->data_len = skb_shinfo(skb)->xdp_frags_size;
 892         else
 893                 skb->data_len = 0;
 894
 895         skb->protocol = eth_type_trans(skb, rq->dev);
 896
 897         metalen = xdp->data - xdp->data_meta;
 898         if (metalen)
 899                 skb_metadata_set(skb, metalen);
 900 out:
 901         return skb;
 902 drop:
 903         stats->rx_drops++;
 904 xdp_drop:
 905         rcu_read_unlock();
 906         kfree_skb(skb);
 907         return NULL;
 908 err_xdp:
 909         rcu_read_unlock();
 910         xdp_return_buff(xdp);
 911 xdp_xmit:
 912         return NULL;
 913 }
 914
 915 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
 916                         struct veth_xdp_tx_bq *bq,
 917                         struct veth_stats *stats)
 918 {
 919         int i, done = 0, n_xdpf = 0;
 920         void *xdpf[VETH_XDP_BATCH];
 921
 922         for (i = 0; i < budget; i++) {
 923                 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
 924
 925                 if (!ptr)
 926                         break;
 927
 928                 if (veth_is_xdp_frame(ptr)) {
 929                         /* ndo_xdp_xmit */
 930                         struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
 931
 932                         stats->xdp_bytes += xdp_get_frame_len(frame);
 933                         frame = veth_xdp_rcv_one(rq, frame, bq, stats);
 934                         if (frame) {
 935                                 /* XDP_PASS */
 936                                 xdpf[n_xdpf++] = frame;
 937                                 if (n_xdpf == VETH_XDP_BATCH) {
 938                                         veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf,
 939                                                               bq, stats);
 940                                         n_xdpf = 0;
 941                                 }
 942                         }
 943                 } else {
 944                         /* ndo_start_xmit */
 945                         struct sk_buff *skb = ptr;
 946
 947                         stats->xdp_bytes += skb->len;
 948                         skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
 949                         if (skb) {
 950                                 if (skb_shared(skb) || skb_unclone(skb, GFP_ATOMIC))
 951                                         netif_receive_skb(skb);
 952                                 else
 953                                         napi_gro_receive(&rq->xdp_napi, skb);
 954                         }
 955                 }
 956                 done++;
 957         }
 958
 959         if (n_xdpf)
 960                 veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf, bq, stats);
 961
 962         u64_stats_update_begin(&rq->stats.syncp);
 963         rq->stats.vs.xdp_redirect += stats->xdp_redirect;
 964         rq->stats.vs.xdp_bytes += stats->xdp_bytes;
 965         rq->stats.vs.xdp_drops += stats->xdp_drops;
 966         rq->stats.vs.rx_drops += stats->rx_drops;
 967         rq->stats.vs.xdp_packets += done;
 968         u64_stats_update_end(&rq->stats.syncp);
 969
 970         return done;
 971 }
 972
 973 static int veth_poll(struct napi_struct *napi, int budget)
 974 {
 975         struct veth_rq *rq =
 976                 container_of(napi, struct veth_rq, xdp_napi);
 977         struct veth_stats stats = {};
 978         struct veth_xdp_tx_bq bq;
 979         int done;
 980
 981         bq.count = 0;
 982
 983         xdp_set_return_frame_no_direct();
 984         done = veth_xdp_rcv(rq, budget, &bq, &stats);
 985
 986         if (stats.xdp_redirect > 0)
 987                 xdp_do_flush();
 988
 989         if (done < budget && napi_complete_done(napi, done)) {
 990                 /* Write rx_notify_masked before reading ptr_ring */
 991                 smp_store_mb(rq->rx_notify_masked, false);
 992                 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
 993                         if (napi_schedule_prep(&rq->xdp_napi)) {
 994                                 WRITE_ONCE(rq->rx_notify_masked, true);
 995                                 __napi_schedule(&rq->xdp_napi);
 996                         }
 997                 }
 998         }
 999
1000         if (stats.xdp_tx > 0)
1001                 veth_xdp_flush(rq, &bq);
1002         xdp_clear_return_frame_no_direct();
1003
1004         return done;
1005 }
1006
1007 static int __veth_napi_enable_range(struct net_device *dev, int start, int end)
1008 {
1009         struct veth_priv *priv = netdev_priv(dev);
1010         int err, i;
1011
1012         for (i = start; i < end; i++) {
1013                 struct veth_rq *rq = &priv->rq[i];
1014
1015                 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
1016                 if (err)
1017                         goto err_xdp_ring;
1018         }
1019
1020         for (i = start; i < end; i++) {
1021                 struct veth_rq *rq = &priv->rq[i];
1022
1023                 napi_enable(&rq->xdp_napi);
1024                 rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1025         }
1026
1027         return 0;
1028
1029 err_xdp_ring:
1030         for (i--; i >= start; i--)
1031                 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
1032
1033         return err;
1034 }
1035
1036 static int __veth_napi_enable(struct net_device *dev)
1037 {
1038         return __veth_napi_enable_range(dev, 0, dev->real_num_rx_queues);
1039 }
1040
1041 static void veth_napi_del_range(struct net_device *dev, int start, int end)
1042 {
1043         struct veth_priv *priv = netdev_priv(dev);
1044         int i;
1045
1046         for (i = start; i < end; i++) {
1047                 struct veth_rq *rq = &priv->rq[i];
1048
1049                 rcu_assign_pointer(priv->rq[i].napi, NULL);
1050                 napi_disable(&rq->xdp_napi);
1051                 __netif_napi_del(&rq->xdp_napi);
1052         }
1053         synchronize_net();
1054
1055         for (i = start; i < end; i++) {
1056                 struct veth_rq *rq = &priv->rq[i];
1057
1058                 rq->rx_notify_masked = false;
1059                 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
1060         }
1061 }
1062
1063 static void veth_napi_del(struct net_device *dev)
1064 {
1065         veth_napi_del_range(dev, 0, dev->real_num_rx_queues);
1066 }
1067
1068 static bool veth_gro_requested(const struct net_device *dev)
1069 {
1070         return !!(dev->wanted_features & NETIF_F_GRO);
1071 }
1072
1073 static int veth_enable_xdp_range(struct net_device *dev, int start, int end,
1074                                  bool napi_already_on)
1075 {
1076         struct veth_priv *priv = netdev_priv(dev);
1077         int err, i;
1078
1079         for (i = start; i < end; i++) {
1080                 struct veth_rq *rq = &priv->rq[i];
1081
1082                 if (!napi_already_on)
1083                         netif_napi_add(dev, &rq->xdp_napi, veth_poll);
1084                 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i, rq->xdp_napi.napi_id);
1085                 if (err < 0)
1086                         goto err_rxq_reg;
1087
1088                 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
1089                                                  MEM_TYPE_PAGE_SHARED,
1090                                                  NULL);
1091                 if (err < 0)
1092                         goto err_reg_mem;
1093
1094                 /* Save original mem info as it can be overwritten */
1095                 rq->xdp_mem = rq->xdp_rxq.mem;
1096         }
1097         return 0;
1098
1099 err_reg_mem:
1100         xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
1101 err_rxq_reg:
1102         for (i--; i >= start; i--) {
1103                 struct veth_rq *rq = &priv->rq[i];
1104
1105                 xdp_rxq_info_unreg(&rq->xdp_rxq);
1106                 if (!napi_already_on)
1107                         netif_napi_del(&rq->xdp_napi);
1108         }
1109
1110         return err;
1111 }
1112
1113 static void veth_disable_xdp_range(struct net_device *dev, int start, int end,
1114                                    bool delete_napi)
1115 {
1116         struct veth_priv *priv = netdev_priv(dev);
1117         int i;
1118
1119         for (i = start; i < end; i++) {
1120                 struct veth_rq *rq = &priv->rq[i];
1121
1122                 rq->xdp_rxq.mem = rq->xdp_mem;
1123                 xdp_rxq_info_unreg(&rq->xdp_rxq);
1124
1125                 if (delete_napi)
1126                         netif_napi_del(&rq->xdp_napi);
1127         }
1128 }
1129
1130 static int veth_enable_xdp(struct net_device *dev)
1131 {
1132         bool napi_already_on = veth_gro_requested(dev) && (dev->flags & IFF_UP);
1133         struct veth_priv *priv = netdev_priv(dev);
1134         int err, i;
1135
1136         if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
1137                 err = veth_enable_xdp_range(dev, 0, dev->real_num_rx_queues, napi_already_on);
1138                 if (err)
1139                         return err;
1140
1141                 if (!napi_already_on) {
1142                         err = __veth_napi_enable(dev);
1143                         if (err) {
1144                                 veth_disable_xdp_range(dev, 0, dev->real_num_rx_queues, true);
1145                                 return err;
1146                         }
1147
1148                         if (!veth_gro_requested(dev)) {
1149                                 /* user-space did not require GRO, but adding XDP
1150                                  * is supposed to get GRO working
1151                                  */
1152                                 dev->features |= NETIF_F_GRO;
1153                                 netdev_features_change(dev);
1154                         }
1155                 }
1156         }
1157
1158         for (i = 0; i < dev->real_num_rx_queues; i++) {
1159                 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
1160                 rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1161         }
1162
1163         return 0;
1164 }
1165
1166 static void veth_disable_xdp(struct net_device *dev)
1167 {
1168         struct veth_priv *priv = netdev_priv(dev);
1169         int i;
1170
1171         for (i = 0; i < dev->real_num_rx_queues; i++)
1172                 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
1173
1174         if (!netif_running(dev) || !veth_gro_requested(dev)) {
1175                 veth_napi_del(dev);
1176
1177                 /* if user-space did not require GRO, since adding XDP
1178                  * enabled it, clear it now
1179                  */
1180                 if (!veth_gro_requested(dev) && netif_running(dev)) {
1181                         dev->features &= ~NETIF_F_GRO;
1182                         netdev_features_change(dev);
1183                 }
1184         }
1185
1186         veth_disable_xdp_range(dev, 0, dev->real_num_rx_queues, false);
1187 }
1188
1189 static int veth_napi_enable_range(struct net_device *dev, int start, int end)
1190 {
1191         struct veth_priv *priv = netdev_priv(dev);
1192         int err, i;
1193
1194         for (i = start; i < end; i++) {
1195                 struct veth_rq *rq = &priv->rq[i];
1196
1197                 netif_napi_add(dev, &rq->xdp_napi, veth_poll);
1198         }
1199
1200         err = __veth_napi_enable_range(dev, start, end);
1201         if (err) {
1202                 for (i = start; i < end; i++) {
1203                         struct veth_rq *rq = &priv->rq[i];
1204
1205                         netif_napi_del(&rq->xdp_napi);
1206                 }
1207                 return err;
1208         }
1209         return err;
1210 }
1211
1212 static int veth_napi_enable(struct net_device *dev)
1213 {
1214         return veth_napi_enable_range(dev, 0, dev->real_num_rx_queues);
1215 }
1216
1217 static void veth_disable_range_safe(struct net_device *dev, int start, int end)
1218 {
1219         struct veth_priv *priv = netdev_priv(dev);
1220
1221         if (start >= end)
1222                 return;
1223
1224         if (priv->_xdp_prog) {
1225                 veth_napi_del_range(dev, start, end);
1226                 veth_disable_xdp_range(dev, start, end, false);
1227         } else if (veth_gro_requested(dev)) {
1228                 veth_napi_del_range(dev, start, end);
1229         }
1230 }
1231
1232 static int veth_enable_range_safe(struct net_device *dev, int start, int end)
1233 {
1234         struct veth_priv *priv = netdev_priv(dev);
1235         int err;
1236
1237         if (start >= end)
1238                 return 0;
1239
1240         if (priv->_xdp_prog) {
1241                 /* these channels are freshly initialized, napi is not on there even
1242                  * when GRO is requeste
1243                  */
1244                 err = veth_enable_xdp_range(dev, start, end, false);
1245                 if (err)
1246                         return err;
1247
1248                 err = __veth_napi_enable_range(dev, start, end);
1249                 if (err) {
1250                         /* on error always delete the newly added napis */
1251                         veth_disable_xdp_range(dev, start, end, true);
1252                         return err;
1253                 }
1254         } else if (veth_gro_requested(dev)) {
1255                 return veth_napi_enable_range(dev, start, end);
1256         }
1257         return 0;
1258 }
1259
1260 static int veth_set_channels(struct net_device *dev,
1261                              struct ethtool_channels *ch)
1262 {
1263         struct veth_priv *priv = netdev_priv(dev);
1264         unsigned int old_rx_count, new_rx_count;
1265         struct veth_priv *peer_priv;
1266         struct net_device *peer;
1267         int err;
1268
1269         /* sanity check. Upper bounds are already enforced by the caller */
1270         if (!ch->rx_count || !ch->tx_count)
1271                 return -EINVAL;
1272
1273         /* avoid braking XDP, if that is enabled */
1274         peer = rtnl_dereference(priv->peer);
1275         peer_priv = peer ? netdev_priv(peer) : NULL;
1276         if (priv->_xdp_prog && peer && ch->rx_count < peer->real_num_tx_queues)
1277                 return -EINVAL;
1278
1279         if (peer && peer_priv && peer_priv->_xdp_prog && ch->tx_count > peer->real_num_rx_queues)
1280                 return -EINVAL;
1281
1282         old_rx_count = dev->real_num_rx_queues;
1283         new_rx_count = ch->rx_count;
1284         if (netif_running(dev)) {
1285                 /* turn device off */
1286                 netif_carrier_off(dev);
1287                 if (peer)
1288                         netif_carrier_off(peer);
1289
1290                 /* try to allocate new resurces, as needed*/
1291                 err = veth_enable_range_safe(dev, old_rx_count, new_rx_count);
1292                 if (err)
1293                         goto out;
1294         }
1295
1296         err = netif_set_real_num_rx_queues(dev, ch->rx_count);
1297         if (err)
1298                 goto revert;
1299
1300         err = netif_set_real_num_tx_queues(dev, ch->tx_count);
1301         if (err) {
1302                 int err2 = netif_set_real_num_rx_queues(dev, old_rx_count);
1303
1304                 /* this error condition could happen only if rx and tx change
1305                  * in opposite directions (e.g. tx nr raises, rx nr decreases)
1306                  * and we can't do anything to fully restore the original
1307                  * status
1308                  */
1309                 if (err2)
1310                         pr_warn("Can't restore rx queues config %d -> %d %d",
1311                                 new_rx_count, old_rx_count, err2);
1312                 else
1313                         goto revert;
1314         }
1315
1316 out:
1317         if (netif_running(dev)) {
1318                 /* note that we need to swap the arguments WRT the enable part
1319                  * to identify the range we have to disable
1320                  */
1321                 veth_disable_range_safe(dev, new_rx_count, old_rx_count);
1322                 netif_carrier_on(dev);
1323                 if (peer)
1324                         netif_carrier_on(peer);
1325         }
1326         return err;
1327
1328 revert:
1329         new_rx_count = old_rx_count;
1330         old_rx_count = ch->rx_count;
1331         goto out;
1332 }
1333
1334 static int veth_open(struct net_device *dev)
1335 {
1336         struct veth_priv *priv = netdev_priv(dev);
1337         struct net_device *peer = rtnl_dereference(priv->peer);
1338         int err;
1339
1340         if (!peer)
1341                 return -ENOTCONN;
1342
1343         if (priv->_xdp_prog) {
1344                 err = veth_enable_xdp(dev);
1345                 if (err)
1346                         return err;
1347         } else if (veth_gro_requested(dev)) {
1348                 err = veth_napi_enable(dev);
1349                 if (err)
1350                         return err;
1351         }
1352
1353         if (peer->flags & IFF_UP) {
1354                 netif_carrier_on(dev);
1355                 netif_carrier_on(peer);
1356         }
1357
1358         return 0;
1359 }
1360
1361 static int veth_close(struct net_device *dev)
1362 {
1363         struct veth_priv *priv = netdev_priv(dev);
1364         struct net_device *peer = rtnl_dereference(priv->peer);
1365
1366         netif_carrier_off(dev);
1367         if (peer)
1368                 netif_carrier_off(peer);
1369
1370         if (priv->_xdp_prog)
1371                 veth_disable_xdp(dev);
1372         else if (veth_gro_requested(dev))
1373                 veth_napi_del(dev);
1374
1375         return 0;
1376 }
1377
1378 static int is_valid_veth_mtu(int mtu)
1379 {
1380         return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1381 }
1382
1383 static int veth_alloc_queues(struct net_device *dev)
1384 {
1385         struct veth_priv *priv = netdev_priv(dev);
1386         int i;
1387
1388         priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL_ACCOUNT);
1389         if (!priv->rq)
1390                 return -ENOMEM;
1391
1392         for (i = 0; i < dev->num_rx_queues; i++) {
1393                 priv->rq[i].dev = dev;
1394                 u64_stats_init(&priv->rq[i].stats.syncp);
1395         }
1396
1397         return 0;
1398 }
1399
1400 static void veth_free_queues(struct net_device *dev)
1401 {
1402         struct veth_priv *priv = netdev_priv(dev);
1403
1404         kfree(priv->rq);
1405 }
1406
1407 static int veth_dev_init(struct net_device *dev)
1408 {
1409         int err;
1410
1411         dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1412         if (!dev->lstats)
1413                 return -ENOMEM;
1414
1415         err = veth_alloc_queues(dev);
1416         if (err) {
1417                 free_percpu(dev->lstats);
1418                 return err;
1419         }
1420
1421         return 0;
1422 }
1423
1424 static void veth_dev_free(struct net_device *dev)
1425 {
1426         veth_free_queues(dev);
1427         free_percpu(dev->lstats);
1428 }
1429
1430 #ifdef CONFIG_NET_POLL_CONTROLLER
1431 static void veth_poll_controller(struct net_device *dev)
1432 {
1433         /* veth only receives frames when its peer sends one
1434          * Since it has nothing to do with disabling irqs, we are guaranteed
1435          * never to have pending data when we poll for it so
1436          * there is nothing to do here.
1437          *
1438          * We need this though so netpoll recognizes us as an interface that
1439          * supports polling, which enables bridge devices in virt setups to
1440          * still use netconsole
1441          */
1442 }
1443 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1444
1445 static int veth_get_iflink(const struct net_device *dev)
1446 {
1447         struct veth_priv *priv = netdev_priv(dev);
1448         struct net_device *peer;
1449         int iflink;
1450
1451         rcu_read_lock();
1452         peer = rcu_dereference(priv->peer);
1453         iflink = peer ? peer->ifindex : 0;
1454         rcu_read_unlock();
1455
1456         return iflink;
1457 }
1458
1459 static netdev_features_t veth_fix_features(struct net_device *dev,
1460                                            netdev_features_t features)
1461 {
1462         struct veth_priv *priv = netdev_priv(dev);
1463         struct net_device *peer;
1464
1465         peer = rtnl_dereference(priv->peer);
1466         if (peer) {
1467                 struct veth_priv *peer_priv = netdev_priv(peer);
1468
1469                 if (peer_priv->_xdp_prog)
1470                         features &= ~NETIF_F_GSO_SOFTWARE;
1471         }
1472         if (priv->_xdp_prog)
1473                 features |= NETIF_F_GRO;
1474
1475         return features;
1476 }
1477
1478 static int veth_set_features(struct net_device *dev,
1479                              netdev_features_t features)
1480 {
1481         netdev_features_t changed = features ^ dev->features;
1482         struct veth_priv *priv = netdev_priv(dev);
1483         int err;
1484
1485         if (!(changed & NETIF_F_GRO) || !(dev->flags & IFF_UP) || priv->_xdp_prog)
1486                 return 0;
1487
1488         if (features & NETIF_F_GRO) {
1489                 err = veth_napi_enable(dev);
1490                 if (err)
1491                         return err;
1492         } else {
1493                 veth_napi_del(dev);
1494         }
1495         return 0;
1496 }
1497
1498 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1499 {
1500         struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1501         struct net_device *peer;
1502
1503         if (new_hr < 0)
1504                 new_hr = 0;
1505
1506         rcu_read_lock();
1507         peer = rcu_dereference(priv->peer);
1508         if (unlikely(!peer))
1509                 goto out;
1510
1511         peer_priv = netdev_priv(peer);
1512         priv->requested_headroom = new_hr;
1513         new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1514         dev->needed_headroom = new_hr;
1515         peer->needed_headroom = new_hr;
1516
1517 out:
1518         rcu_read_unlock();
1519 }
1520
1521 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1522                         struct netlink_ext_ack *extack)
1523 {
1524         struct veth_priv *priv = netdev_priv(dev);
1525         struct bpf_prog *old_prog;
1526         struct net_device *peer;
1527         unsigned int max_mtu;
1528         int err;
1529
1530         old_prog = priv->_xdp_prog;
1531         priv->_xdp_prog = prog;
1532         peer = rtnl_dereference(priv->peer);
1533
1534         if (prog) {
1535                 if (!peer) {
1536                         NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1537                         err = -ENOTCONN;
1538                         goto err;
1539                 }
1540
1541                 max_mtu = SKB_WITH_OVERHEAD(PAGE_SIZE - VETH_XDP_HEADROOM) -
1542                           peer->hard_header_len;
1543                 /* Allow increasing the max_mtu if the program supports
1544                  * XDP fragments.
1545                  */
1546                 if (prog->aux->xdp_has_frags)
1547                         max_mtu += PAGE_SIZE * MAX_SKB_FRAGS;
1548
1549                 if (peer->mtu > max_mtu) {
1550                         NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1551                         err = -ERANGE;
1552                         goto err;
1553                 }
1554
1555                 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1556                         NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1557                         err = -ENOSPC;
1558                         goto err;
1559                 }
1560
1561                 if (dev->flags & IFF_UP) {
1562                         err = veth_enable_xdp(dev);
1563                         if (err) {
1564                                 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1565                                 goto err;
1566                         }
1567                 }
1568
1569                 if (!old_prog) {
1570                         peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1571                         peer->max_mtu = max_mtu;
1572                 }
1573         }
1574
1575         if (old_prog) {
1576                 if (!prog) {
1577                         if (dev->flags & IFF_UP)
1578                                 veth_disable_xdp(dev);
1579
1580                         if (peer) {
1581                                 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1582                                 peer->max_mtu = ETH_MAX_MTU;
1583                         }
1584                 }
1585                 bpf_prog_put(old_prog);
1586         }
1587
1588         if ((!!old_prog ^ !!prog) && peer)
1589                 netdev_update_features(peer);
1590
1591         return 0;
1592 err:
1593         priv->_xdp_prog = old_prog;
1594
1595         return err;
1596 }
1597
1598 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1599 {
1600         switch (xdp->command) {
1601         case XDP_SETUP_PROG:
1602                 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1603         default:
1604                 return -EINVAL;
1605         }
1606 }
1607
1608 static int veth_xdp_rx_timestamp(const struct xdp_md *ctx, u64 *timestamp)
1609 {
1610         struct veth_xdp_buff *_ctx = (void *)ctx;
1611
1612         if (!_ctx->skb)
1613                 return -EOPNOTSUPP;
1614
1615         *timestamp = skb_hwtstamps(_ctx->skb)->hwtstamp;
1616         return 0;
1617 }
1618
1619 static int veth_xdp_rx_hash(const struct xdp_md *ctx, u32 *hash)
1620 {
1621         struct veth_xdp_buff *_ctx = (void *)ctx;
1622
1623         if (!_ctx->skb)
1624                 return -EOPNOTSUPP;
1625
1626         *hash = skb_get_hash(_ctx->skb);
1627         return 0;
1628 }
1629
1630 static const struct net_device_ops veth_netdev_ops = {
1631         .ndo_init            = veth_dev_init,
1632         .ndo_open            = veth_open,
1633         .ndo_stop            = veth_close,
1634         .ndo_start_xmit      = veth_xmit,
1635         .ndo_get_stats64     = veth_get_stats64,
1636         .ndo_set_rx_mode     = veth_set_multicast_list,
1637         .ndo_set_mac_address = eth_mac_addr,
1638 #ifdef CONFIG_NET_POLL_CONTROLLER
1639         .ndo_poll_controller    = veth_poll_controller,
1640 #endif
1641         .ndo_get_iflink         = veth_get_iflink,
1642         .ndo_fix_features       = veth_fix_features,
1643         .ndo_set_features       = veth_set_features,
1644         .ndo_features_check     = passthru_features_check,
1645         .ndo_set_rx_headroom    = veth_set_rx_headroom,
1646         .ndo_bpf                = veth_xdp,
1647         .ndo_xdp_xmit           = veth_ndo_xdp_xmit,
1648         .ndo_get_peer_dev       = veth_peer_dev,
1649 };
1650
1651 static const struct xdp_metadata_ops veth_xdp_metadata_ops = {
1652         .xmo_rx_timestamp               = veth_xdp_rx_timestamp,
1653         .xmo_rx_hash                    = veth_xdp_rx_hash,
1654 };
1655
1656 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1657                        NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1658                        NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1659                        NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1660                        NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1661
1662 static void veth_setup(struct net_device *dev)
1663 {
1664         ether_setup(dev);
1665
1666         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1667         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1668         dev->priv_flags |= IFF_NO_QUEUE;
1669         dev->priv_flags |= IFF_PHONY_HEADROOM;
1670
1671         dev->netdev_ops = &veth_netdev_ops;
1672         dev->xdp_metadata_ops = &veth_xdp_metadata_ops;
1673         dev->ethtool_ops = &veth_ethtool_ops;
1674         dev->features |= NETIF_F_LLTX;
1675         dev->features |= VETH_FEATURES;
1676         dev->vlan_features = dev->features &
1677                              ~(NETIF_F_HW_VLAN_CTAG_TX |
1678                                NETIF_F_HW_VLAN_STAG_TX |
1679                                NETIF_F_HW_VLAN_CTAG_RX |
1680                                NETIF_F_HW_VLAN_STAG_RX);
1681         dev->needs_free_netdev = true;
1682         dev->priv_destructor = veth_dev_free;
1683         dev->max_mtu = ETH_MAX_MTU;
1684
1685         dev->hw_features = VETH_FEATURES;
1686         dev->hw_enc_features = VETH_FEATURES;
1687         dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1688         netif_set_tso_max_size(dev, GSO_MAX_SIZE);
1689
1690         dev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
1691                             NETDEV_XDP_ACT_NDO_XMIT | NETDEV_XDP_ACT_RX_SG |
1692                             NETDEV_XDP_ACT_NDO_XMIT_SG;
1693 }
1694
1695 /*
1696  * netlink interface
1697  */
1698
1699 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1700                          struct netlink_ext_ack *extack)
1701 {
1702         if (tb[IFLA_ADDRESS]) {
1703                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1704                         return -EINVAL;
1705                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1706                         return -EADDRNOTAVAIL;
1707         }
1708         if (tb[IFLA_MTU]) {
1709                 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1710                         return -EINVAL;
1711         }
1712         return 0;
1713 }
1714
1715 static struct rtnl_link_ops veth_link_ops;
1716
1717 static void veth_disable_gro(struct net_device *dev)
1718 {
1719         dev->features &= ~NETIF_F_GRO;
1720         dev->wanted_features &= ~NETIF_F_GRO;
1721         netdev_update_features(dev);
1722 }
1723
1724 static int veth_init_queues(struct net_device *dev, struct nlattr *tb[])
1725 {
1726         int err;
1727
1728         if (!tb[IFLA_NUM_TX_QUEUES] && dev->num_tx_queues > 1) {
1729                 err = netif_set_real_num_tx_queues(dev, 1);
1730                 if (err)
1731                         return err;
1732         }
1733         if (!tb[IFLA_NUM_RX_QUEUES] && dev->num_rx_queues > 1) {
1734                 err = netif_set_real_num_rx_queues(dev, 1);
1735                 if (err)
1736                         return err;
1737         }
1738         return 0;
1739 }
1740
1741 static int veth_newlink(struct net *src_net, struct net_device *dev,
1742                         struct nlattr *tb[], struct nlattr *data[],
1743                         struct netlink_ext_ack *extack)
1744 {
1745         int err;
1746         struct net_device *peer;
1747         struct veth_priv *priv;
1748         char ifname[IFNAMSIZ];
1749         struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1750         unsigned char name_assign_type;
1751         struct ifinfomsg *ifmp;
1752         struct net *net;
1753
1754         /*
1755          * create and register peer first
1756          */
1757         if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1758                 struct nlattr *nla_peer;
1759
1760                 nla_peer = data[VETH_INFO_PEER];
1761                 ifmp = nla_data(nla_peer);
1762                 err = rtnl_nla_parse_ifla(peer_tb,
1763                                           nla_data(nla_peer) + sizeof(struct ifinfomsg),
1764                                           nla_len(nla_peer) - sizeof(struct ifinfomsg),
1765                                           NULL);
1766                 if (err < 0)
1767                         return err;
1768
1769                 err = veth_validate(peer_tb, NULL, extack);
1770                 if (err < 0)
1771                         return err;
1772
1773                 tbp = peer_tb;
1774         } else {
1775                 ifmp = NULL;
1776                 tbp = tb;
1777         }
1778
1779         if (ifmp && tbp[IFLA_IFNAME]) {
1780                 nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1781                 name_assign_type = NET_NAME_USER;
1782         } else {
1783                 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1784                 name_assign_type = NET_NAME_ENUM;
1785         }
1786
1787         net = rtnl_link_get_net(src_net, tbp);
1788         if (IS_ERR(net))
1789                 return PTR_ERR(net);
1790
1791         peer = rtnl_create_link(net, ifname, name_assign_type,
1792                                 &veth_link_ops, tbp, extack);
1793         if (IS_ERR(peer)) {
1794                 put_net(net);
1795                 return PTR_ERR(peer);
1796         }
1797
1798         if (!ifmp || !tbp[IFLA_ADDRESS])
1799                 eth_hw_addr_random(peer);
1800
1801         if (ifmp && (dev->ifindex != 0))
1802                 peer->ifindex = ifmp->ifi_index;
1803
1804         netif_inherit_tso_max(peer, dev);
1805
1806         err = register_netdevice(peer);
1807         put_net(net);
1808         net = NULL;
1809         if (err < 0)
1810                 goto err_register_peer;
1811
1812         /* keep GRO disabled by default to be consistent with the established
1813          * veth behavior
1814          */
1815         veth_disable_gro(peer);
1816         netif_carrier_off(peer);
1817
1818         err = rtnl_configure_link(peer, ifmp, 0, NULL);
1819         if (err < 0)
1820                 goto err_configure_peer;
1821
1822         /*
1823          * register dev last
1824          *
1825          * note, that since we've registered new device the dev's name
1826          * should be re-allocated
1827          */
1828
1829         if (tb[IFLA_ADDRESS] == NULL)
1830                 eth_hw_addr_random(dev);
1831
1832         if (tb[IFLA_IFNAME])
1833                 nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1834         else
1835                 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1836
1837         err = register_netdevice(dev);
1838         if (err < 0)
1839                 goto err_register_dev;
1840
1841         netif_carrier_off(dev);
1842
1843         /*
1844          * tie the deviced together
1845          */
1846
1847         priv = netdev_priv(dev);
1848         rcu_assign_pointer(priv->peer, peer);
1849         err = veth_init_queues(dev, tb);
1850         if (err)
1851                 goto err_queues;
1852
1853         priv = netdev_priv(peer);
1854         rcu_assign_pointer(priv->peer, dev);
1855         err = veth_init_queues(peer, tb);
1856         if (err)
1857                 goto err_queues;
1858
1859         veth_disable_gro(dev);
1860         return 0;
1861
1862 err_queues:
1863         unregister_netdevice(dev);
1864 err_register_dev:
1865         /* nothing to do */
1866 err_configure_peer:
1867         unregister_netdevice(peer);
1868         return err;
1869
1870 err_register_peer:
1871         free_netdev(peer);
1872         return err;
1873 }
1874
1875 static void veth_dellink(struct net_device *dev, struct list_head *head)
1876 {
1877         struct veth_priv *priv;
1878         struct net_device *peer;
1879
1880         priv = netdev_priv(dev);
1881         peer = rtnl_dereference(priv->peer);
1882
1883         /* Note : dellink() is called from default_device_exit_batch(),
1884          * before a rcu_synchronize() point. The devices are guaranteed
1885          * not being freed before one RCU grace period.
1886          */
1887         RCU_INIT_POINTER(priv->peer, NULL);
1888         unregister_netdevice_queue(dev, head);
1889
1890         if (peer) {
1891                 priv = netdev_priv(peer);
1892                 RCU_INIT_POINTER(priv->peer, NULL);
1893                 unregister_netdevice_queue(peer, head);
1894         }
1895 }
1896
1897 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1898         [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
1899 };
1900
1901 static struct net *veth_get_link_net(const struct net_device *dev)
1902 {
1903         struct veth_priv *priv = netdev_priv(dev);
1904         struct net_device *peer = rtnl_dereference(priv->peer);
1905
1906         return peer ? dev_net(peer) : dev_net(dev);
1907 }
1908
1909 static unsigned int veth_get_num_queues(void)
1910 {
1911         /* enforce the same queue limit as rtnl_create_link */
1912         int queues = num_possible_cpus();
1913
1914         if (queues > 4096)
1915                 queues = 4096;
1916         return queues;
1917 }
1918
1919 static struct rtnl_link_ops veth_link_ops = {
1920         .kind           = DRV_NAME,
1921         .priv_size      = sizeof(struct veth_priv),
1922         .setup          = veth_setup,
1923         .validate       = veth_validate,
1924         .newlink        = veth_newlink,
1925         .dellink        = veth_dellink,
1926         .policy         = veth_policy,
1927         .maxtype        = VETH_INFO_MAX,
1928         .get_link_net   = veth_get_link_net,
1929         .get_num_tx_queues      = veth_get_num_queues,
1930         .get_num_rx_queues      = veth_get_num_queues,
1931 };
1932
1933 /*
1934  * init/fini
1935  */
1936
1937 static __init int veth_init(void)
1938 {
1939         return rtnl_link_register(&veth_link_ops);
1940 }
1941
1942 static __exit void veth_exit(void)
1943 {
1944         rtnl_link_unregister(&veth_link_ops);
1945 }
1946
1947 module_init(veth_init);
1948 module_exit(veth_exit);
1949
1950 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1951 MODULE_LICENSE("GPL v2");
1952 MODULE_ALIAS_RTNL_LINK(DRV_NAME);