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             skb_headroom(skb) < XDP_PACKET_HEADROOM) {
 713                 u32 size, len, max_head_size, off;
 714                 struct sk_buff *nskb;
 715                 struct page *page;
 716                 int i, head_off;
 717
 718                 /* We need a private copy of the skb and data buffers since
 719                  * the ebpf program can modify it. We segment the original skb
 720                  * into order-0 pages without linearize it.
 721                  *
 722                  * Make sure we have enough space for linear and paged area
 723                  */
 724                 max_head_size = SKB_WITH_OVERHEAD(PAGE_SIZE -
 725                                                   VETH_XDP_HEADROOM);
 726                 if (skb->len > PAGE_SIZE * MAX_SKB_FRAGS + max_head_size)
 727                         goto drop;
 728
 729                 /* Allocate skb head */
 730                 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
 731                 if (!page)
 732                         goto drop;
 733
 734                 nskb = build_skb(page_address(page), PAGE_SIZE);
 735                 if (!nskb) {
 736                         put_page(page);
 737                         goto drop;
 738                 }
 739
 740                 skb_reserve(nskb, VETH_XDP_HEADROOM);
 741                 size = min_t(u32, skb->len, max_head_size);
 742                 if (skb_copy_bits(skb, 0, nskb->data, size)) {
 743                         consume_skb(nskb);
 744                         goto drop;
 745                 }
 746                 skb_put(nskb, size);
 747
 748                 skb_copy_header(nskb, skb);
 749                 head_off = skb_headroom(nskb) - skb_headroom(skb);
 750                 skb_headers_offset_update(nskb, head_off);
 751
 752                 /* Allocate paged area of new skb */
 753                 off = size;
 754                 len = skb->len - off;
 755
 756                 for (i = 0; i < MAX_SKB_FRAGS && off < skb->len; i++) {
 757                         page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
 758                         if (!page) {
 759                                 consume_skb(nskb);
 760                                 goto drop;
 761                         }
 762
 763                         size = min_t(u32, len, PAGE_SIZE);
 764                         skb_add_rx_frag(nskb, i, page, 0, size, PAGE_SIZE);
 765                         if (skb_copy_bits(skb, off, page_address(page),
 766                                           size)) {
 767                                 consume_skb(nskb);
 768                                 goto drop;
 769                         }
 770
 771                         len -= size;
 772                         off += size;
 773                 }
 774
 775                 consume_skb(skb);
 776                 skb = nskb;
 777         }
 778
 779         /* SKB "head" area always have tailroom for skb_shared_info */
 780         frame_sz = skb_end_pointer(skb) - skb->head;
 781         frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 782         xdp_init_buff(xdp, frame_sz, &rq->xdp_rxq);
 783         xdp_prepare_buff(xdp, skb->head, skb_headroom(skb),
 784                          skb_headlen(skb), true);
 785
 786         if (skb_is_nonlinear(skb)) {
 787                 skb_shinfo(skb)->xdp_frags_size = skb->data_len;
 788                 xdp_buff_set_frags_flag(xdp);
 789         } else {
 790                 xdp_buff_clear_frags_flag(xdp);
 791         }
 792         *pskb = skb;
 793
 794         return 0;
 795 drop:
 796         consume_skb(skb);
 797         *pskb = NULL;
 798
 799         return -ENOMEM;
 800 }
 801
 802 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
 803                                         struct sk_buff *skb,
 804                                         struct veth_xdp_tx_bq *bq,
 805                                         struct veth_stats *stats)
 806 {
 807         void *orig_data, *orig_data_end;
 808         struct bpf_prog *xdp_prog;
 809         struct veth_xdp_buff vxbuf;
 810         struct xdp_buff *xdp = &vxbuf.xdp;
 811         u32 act, metalen;
 812         int off;
 813
 814         skb_prepare_for_gro(skb);
 815
 816         rcu_read_lock();
 817         xdp_prog = rcu_dereference(rq->xdp_prog);
 818         if (unlikely(!xdp_prog)) {
 819                 rcu_read_unlock();
 820                 goto out;
 821         }
 822
 823         __skb_push(skb, skb->data - skb_mac_header(skb));
 824         if (veth_convert_skb_to_xdp_buff(rq, xdp, &skb))
 825                 goto drop;
 826         vxbuf.skb = skb;
 827
 828         orig_data = xdp->data;
 829         orig_data_end = xdp->data_end;
 830
 831         act = bpf_prog_run_xdp(xdp_prog, xdp);
 832
 833         switch (act) {
 834         case XDP_PASS:
 835                 break;
 836         case XDP_TX:
 837                 veth_xdp_get(xdp);
 838                 consume_skb(skb);
 839                 xdp->rxq->mem = rq->xdp_mem;
 840                 if (unlikely(veth_xdp_tx(rq, xdp, bq) < 0)) {
 841                         trace_xdp_exception(rq->dev, xdp_prog, act);
 842                         stats->rx_drops++;
 843                         goto err_xdp;
 844                 }
 845                 stats->xdp_tx++;
 846                 rcu_read_unlock();
 847                 goto xdp_xmit;
 848         case XDP_REDIRECT:
 849                 veth_xdp_get(xdp);
 850                 consume_skb(skb);
 851                 xdp->rxq->mem = rq->xdp_mem;
 852                 if (xdp_do_redirect(rq->dev, xdp, xdp_prog)) {
 853                         stats->rx_drops++;
 854                         goto err_xdp;
 855                 }
 856                 stats->xdp_redirect++;
 857                 rcu_read_unlock();
 858                 goto xdp_xmit;
 859         default:
 860                 bpf_warn_invalid_xdp_action(rq->dev, xdp_prog, act);
 861                 fallthrough;
 862         case XDP_ABORTED:
 863                 trace_xdp_exception(rq->dev, xdp_prog, act);
 864                 fallthrough;
 865         case XDP_DROP:
 866                 stats->xdp_drops++;
 867                 goto xdp_drop;
 868         }
 869         rcu_read_unlock();
 870
 871         /* check if bpf_xdp_adjust_head was used */
 872         off = orig_data - xdp->data;
 873         if (off > 0)
 874                 __skb_push(skb, off);
 875         else if (off < 0)
 876                 __skb_pull(skb, -off);
 877
 878         skb_reset_mac_header(skb);
 879
 880         /* check if bpf_xdp_adjust_tail was used */
 881         off = xdp->data_end - orig_data_end;
 882         if (off != 0)
 883                 __skb_put(skb, off); /* positive on grow, negative on shrink */
 884
 885         /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
 886          * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
 887          */
 888         if (xdp_buff_has_frags(xdp))
 889                 skb->data_len = skb_shinfo(skb)->xdp_frags_size;
 890         else
 891                 skb->data_len = 0;
 892
 893         skb->protocol = eth_type_trans(skb, rq->dev);
 894
 895         metalen = xdp->data - xdp->data_meta;
 896         if (metalen)
 897                 skb_metadata_set(skb, metalen);
 898 out:
 899         return skb;
 900 drop:
 901         stats->rx_drops++;
 902 xdp_drop:
 903         rcu_read_unlock();
 904         kfree_skb(skb);
 905         return NULL;
 906 err_xdp:
 907         rcu_read_unlock();
 908         xdp_return_buff(xdp);
 909 xdp_xmit:
 910         return NULL;
 911 }
 912
 913 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
 914                         struct veth_xdp_tx_bq *bq,
 915                         struct veth_stats *stats)
 916 {
 917         int i, done = 0, n_xdpf = 0;
 918         void *xdpf[VETH_XDP_BATCH];
 919
 920         for (i = 0; i < budget; i++) {
 921                 void *ptr = __ptr_ring_consume(&rq->xdp_ring);
 922
 923                 if (!ptr)
 924                         break;
 925
 926                 if (veth_is_xdp_frame(ptr)) {
 927                         /* ndo_xdp_xmit */
 928                         struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
 929
 930                         stats->xdp_bytes += xdp_get_frame_len(frame);
 931                         frame = veth_xdp_rcv_one(rq, frame, bq, stats);
 932                         if (frame) {
 933                                 /* XDP_PASS */
 934                                 xdpf[n_xdpf++] = frame;
 935                                 if (n_xdpf == VETH_XDP_BATCH) {
 936                                         veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf,
 937                                                               bq, stats);
 938                                         n_xdpf = 0;
 939                                 }
 940                         }
 941                 } else {
 942                         /* ndo_start_xmit */
 943                         struct sk_buff *skb = ptr;
 944
 945                         stats->xdp_bytes += skb->len;
 946                         skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
 947                         if (skb) {
 948                                 if (skb_shared(skb) || skb_unclone(skb, GFP_ATOMIC))
 949                                         netif_receive_skb(skb);
 950                                 else
 951                                         napi_gro_receive(&rq->xdp_napi, skb);
 952                         }
 953                 }
 954                 done++;
 955         }
 956
 957         if (n_xdpf)
 958                 veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf, bq, stats);
 959
 960         u64_stats_update_begin(&rq->stats.syncp);
 961         rq->stats.vs.xdp_redirect += stats->xdp_redirect;
 962         rq->stats.vs.xdp_bytes += stats->xdp_bytes;
 963         rq->stats.vs.xdp_drops += stats->xdp_drops;
 964         rq->stats.vs.rx_drops += stats->rx_drops;
 965         rq->stats.vs.xdp_packets += done;
 966         u64_stats_update_end(&rq->stats.syncp);
 967
 968         return done;
 969 }
 970
 971 static int veth_poll(struct napi_struct *napi, int budget)
 972 {
 973         struct veth_rq *rq =
 974                 container_of(napi, struct veth_rq, xdp_napi);
 975         struct veth_stats stats = {};
 976         struct veth_xdp_tx_bq bq;
 977         int done;
 978
 979         bq.count = 0;
 980
 981         xdp_set_return_frame_no_direct();
 982         done = veth_xdp_rcv(rq, budget, &bq, &stats);
 983
 984         if (stats.xdp_redirect > 0)
 985                 xdp_do_flush();
 986
 987         if (done < budget && napi_complete_done(napi, done)) {
 988                 /* Write rx_notify_masked before reading ptr_ring */
 989                 smp_store_mb(rq->rx_notify_masked, false);
 990                 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
 991                         if (napi_schedule_prep(&rq->xdp_napi)) {
 992                                 WRITE_ONCE(rq->rx_notify_masked, true);
 993                                 __napi_schedule(&rq->xdp_napi);
 994                         }
 995                 }
 996         }
 997
 998         if (stats.xdp_tx > 0)
 999                 veth_xdp_flush(rq, &bq);
1000         xdp_clear_return_frame_no_direct();
1001
1002         return done;
1003 }
1004
1005 static int __veth_napi_enable_range(struct net_device *dev, int start, int end)
1006 {
1007         struct veth_priv *priv = netdev_priv(dev);
1008         int err, i;
1009
1010         for (i = start; i < end; i++) {
1011                 struct veth_rq *rq = &priv->rq[i];
1012
1013                 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
1014                 if (err)
1015                         goto err_xdp_ring;
1016         }
1017
1018         for (i = start; i < end; i++) {
1019                 struct veth_rq *rq = &priv->rq[i];
1020
1021                 napi_enable(&rq->xdp_napi);
1022                 rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1023         }
1024
1025         return 0;
1026
1027 err_xdp_ring:
1028         for (i--; i >= start; i--)
1029                 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
1030
1031         return err;
1032 }
1033
1034 static int __veth_napi_enable(struct net_device *dev)
1035 {
1036         return __veth_napi_enable_range(dev, 0, dev->real_num_rx_queues);
1037 }
1038
1039 static void veth_napi_del_range(struct net_device *dev, int start, int end)
1040 {
1041         struct veth_priv *priv = netdev_priv(dev);
1042         int i;
1043
1044         for (i = start; i < end; i++) {
1045                 struct veth_rq *rq = &priv->rq[i];
1046
1047                 rcu_assign_pointer(priv->rq[i].napi, NULL);
1048                 napi_disable(&rq->xdp_napi);
1049                 __netif_napi_del(&rq->xdp_napi);
1050         }
1051         synchronize_net();
1052
1053         for (i = start; i < end; i++) {
1054                 struct veth_rq *rq = &priv->rq[i];
1055
1056                 rq->rx_notify_masked = false;
1057                 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
1058         }
1059 }
1060
1061 static void veth_napi_del(struct net_device *dev)
1062 {
1063         veth_napi_del_range(dev, 0, dev->real_num_rx_queues);
1064 }
1065
1066 static bool veth_gro_requested(const struct net_device *dev)
1067 {
1068         return !!(dev->wanted_features & NETIF_F_GRO);
1069 }
1070
1071 static int veth_enable_xdp_range(struct net_device *dev, int start, int end,
1072                                  bool napi_already_on)
1073 {
1074         struct veth_priv *priv = netdev_priv(dev);
1075         int err, i;
1076
1077         for (i = start; i < end; i++) {
1078                 struct veth_rq *rq = &priv->rq[i];
1079
1080                 if (!napi_already_on)
1081                         netif_napi_add(dev, &rq->xdp_napi, veth_poll);
1082                 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i, rq->xdp_napi.napi_id);
1083                 if (err < 0)
1084                         goto err_rxq_reg;
1085
1086                 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
1087                                                  MEM_TYPE_PAGE_SHARED,
1088                                                  NULL);
1089                 if (err < 0)
1090                         goto err_reg_mem;
1091
1092                 /* Save original mem info as it can be overwritten */
1093                 rq->xdp_mem = rq->xdp_rxq.mem;
1094         }
1095         return 0;
1096
1097 err_reg_mem:
1098         xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
1099 err_rxq_reg:
1100         for (i--; i >= start; i--) {
1101                 struct veth_rq *rq = &priv->rq[i];
1102
1103                 xdp_rxq_info_unreg(&rq->xdp_rxq);
1104                 if (!napi_already_on)
1105                         netif_napi_del(&rq->xdp_napi);
1106         }
1107
1108         return err;
1109 }
1110
1111 static void veth_disable_xdp_range(struct net_device *dev, int start, int end,
1112                                    bool delete_napi)
1113 {
1114         struct veth_priv *priv = netdev_priv(dev);
1115         int i;
1116
1117         for (i = start; i < end; i++) {
1118                 struct veth_rq *rq = &priv->rq[i];
1119
1120                 rq->xdp_rxq.mem = rq->xdp_mem;
1121                 xdp_rxq_info_unreg(&rq->xdp_rxq);
1122
1123                 if (delete_napi)
1124                         netif_napi_del(&rq->xdp_napi);
1125         }
1126 }
1127
1128 static int veth_enable_xdp(struct net_device *dev)
1129 {
1130         bool napi_already_on = veth_gro_requested(dev) && (dev->flags & IFF_UP);
1131         struct veth_priv *priv = netdev_priv(dev);
1132         int err, i;
1133
1134         if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
1135                 err = veth_enable_xdp_range(dev, 0, dev->real_num_rx_queues, napi_already_on);
1136                 if (err)
1137                         return err;
1138
1139                 if (!napi_already_on) {
1140                         err = __veth_napi_enable(dev);
1141                         if (err) {
1142                                 veth_disable_xdp_range(dev, 0, dev->real_num_rx_queues, true);
1143                                 return err;
1144                         }
1145
1146                         if (!veth_gro_requested(dev)) {
1147                                 /* user-space did not require GRO, but adding XDP
1148                                  * is supposed to get GRO working
1149                                  */
1150                                 dev->features |= NETIF_F_GRO;
1151                                 netdev_features_change(dev);
1152                         }
1153                 }
1154         }
1155
1156         for (i = 0; i < dev->real_num_rx_queues; i++) {
1157                 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
1158                 rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1159         }
1160
1161         return 0;
1162 }
1163
1164 static void veth_disable_xdp(struct net_device *dev)
1165 {
1166         struct veth_priv *priv = netdev_priv(dev);
1167         int i;
1168
1169         for (i = 0; i < dev->real_num_rx_queues; i++)
1170                 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
1171
1172         if (!netif_running(dev) || !veth_gro_requested(dev)) {
1173                 veth_napi_del(dev);
1174
1175                 /* if user-space did not require GRO, since adding XDP
1176                  * enabled it, clear it now
1177                  */
1178                 if (!veth_gro_requested(dev) && netif_running(dev)) {
1179                         dev->features &= ~NETIF_F_GRO;
1180                         netdev_features_change(dev);
1181                 }
1182         }
1183
1184         veth_disable_xdp_range(dev, 0, dev->real_num_rx_queues, false);
1185 }
1186
1187 static int veth_napi_enable_range(struct net_device *dev, int start, int end)
1188 {
1189         struct veth_priv *priv = netdev_priv(dev);
1190         int err, i;
1191
1192         for (i = start; i < end; i++) {
1193                 struct veth_rq *rq = &priv->rq[i];
1194
1195                 netif_napi_add(dev, &rq->xdp_napi, veth_poll);
1196         }
1197
1198         err = __veth_napi_enable_range(dev, start, end);
1199         if (err) {
1200                 for (i = start; i < end; i++) {
1201                         struct veth_rq *rq = &priv->rq[i];
1202
1203                         netif_napi_del(&rq->xdp_napi);
1204                 }
1205                 return err;
1206         }
1207         return err;
1208 }
1209
1210 static int veth_napi_enable(struct net_device *dev)
1211 {
1212         return veth_napi_enable_range(dev, 0, dev->real_num_rx_queues);
1213 }
1214
1215 static void veth_disable_range_safe(struct net_device *dev, int start, int end)
1216 {
1217         struct veth_priv *priv = netdev_priv(dev);
1218
1219         if (start >= end)
1220                 return;
1221
1222         if (priv->_xdp_prog) {
1223                 veth_napi_del_range(dev, start, end);
1224                 veth_disable_xdp_range(dev, start, end, false);
1225         } else if (veth_gro_requested(dev)) {
1226                 veth_napi_del_range(dev, start, end);
1227         }
1228 }
1229
1230 static int veth_enable_range_safe(struct net_device *dev, int start, int end)
1231 {
1232         struct veth_priv *priv = netdev_priv(dev);
1233         int err;
1234
1235         if (start >= end)
1236                 return 0;
1237
1238         if (priv->_xdp_prog) {
1239                 /* these channels are freshly initialized, napi is not on there even
1240                  * when GRO is requeste
1241                  */
1242                 err = veth_enable_xdp_range(dev, start, end, false);
1243                 if (err)
1244                         return err;
1245
1246                 err = __veth_napi_enable_range(dev, start, end);
1247                 if (err) {
1248                         /* on error always delete the newly added napis */
1249                         veth_disable_xdp_range(dev, start, end, true);
1250                         return err;
1251                 }
1252         } else if (veth_gro_requested(dev)) {
1253                 return veth_napi_enable_range(dev, start, end);
1254         }
1255         return 0;
1256 }
1257
1258 static void veth_set_xdp_features(struct net_device *dev)
1259 {
1260         struct veth_priv *priv = netdev_priv(dev);
1261         struct net_device *peer;
1262
1263         peer = rtnl_dereference(priv->peer);
1264         if (peer && peer->real_num_tx_queues <= dev->real_num_rx_queues) {
1265                 xdp_features_t val = NETDEV_XDP_ACT_BASIC |
1266                                      NETDEV_XDP_ACT_REDIRECT |
1267                                      NETDEV_XDP_ACT_RX_SG;
1268
1269                 if (priv->_xdp_prog || veth_gro_requested(dev))
1270                         val |= NETDEV_XDP_ACT_NDO_XMIT |
1271                                NETDEV_XDP_ACT_NDO_XMIT_SG;
1272                 xdp_set_features_flag(dev, val);
1273         } else {
1274                 xdp_clear_features_flag(dev);
1275         }
1276 }
1277
1278 static int veth_set_channels(struct net_device *dev,
1279                              struct ethtool_channels *ch)
1280 {
1281         struct veth_priv *priv = netdev_priv(dev);
1282         unsigned int old_rx_count, new_rx_count;
1283         struct veth_priv *peer_priv;
1284         struct net_device *peer;
1285         int err;
1286
1287         /* sanity check. Upper bounds are already enforced by the caller */
1288         if (!ch->rx_count || !ch->tx_count)
1289                 return -EINVAL;
1290
1291         /* avoid braking XDP, if that is enabled */
1292         peer = rtnl_dereference(priv->peer);
1293         peer_priv = peer ? netdev_priv(peer) : NULL;
1294         if (priv->_xdp_prog && peer && ch->rx_count < peer->real_num_tx_queues)
1295                 return -EINVAL;
1296
1297         if (peer && peer_priv && peer_priv->_xdp_prog && ch->tx_count > peer->real_num_rx_queues)
1298                 return -EINVAL;
1299
1300         old_rx_count = dev->real_num_rx_queues;
1301         new_rx_count = ch->rx_count;
1302         if (netif_running(dev)) {
1303                 /* turn device off */
1304                 netif_carrier_off(dev);
1305                 if (peer)
1306                         netif_carrier_off(peer);
1307
1308                 /* try to allocate new resurces, as needed*/
1309                 err = veth_enable_range_safe(dev, old_rx_count, new_rx_count);
1310                 if (err)
1311                         goto out;
1312         }
1313
1314         err = netif_set_real_num_rx_queues(dev, ch->rx_count);
1315         if (err)
1316                 goto revert;
1317
1318         err = netif_set_real_num_tx_queues(dev, ch->tx_count);
1319         if (err) {
1320                 int err2 = netif_set_real_num_rx_queues(dev, old_rx_count);
1321
1322                 /* this error condition could happen only if rx and tx change
1323                  * in opposite directions (e.g. tx nr raises, rx nr decreases)
1324                  * and we can't do anything to fully restore the original
1325                  * status
1326                  */
1327                 if (err2)
1328                         pr_warn("Can't restore rx queues config %d -> %d %d",
1329                                 new_rx_count, old_rx_count, err2);
1330                 else
1331                         goto revert;
1332         }
1333
1334 out:
1335         if (netif_running(dev)) {
1336                 /* note that we need to swap the arguments WRT the enable part
1337                  * to identify the range we have to disable
1338                  */
1339                 veth_disable_range_safe(dev, new_rx_count, old_rx_count);
1340                 netif_carrier_on(dev);
1341                 if (peer)
1342                         netif_carrier_on(peer);
1343         }
1344
1345         /* update XDP supported features */
1346         veth_set_xdp_features(dev);
1347         if (peer)
1348                 veth_set_xdp_features(peer);
1349
1350         return err;
1351
1352 revert:
1353         new_rx_count = old_rx_count;
1354         old_rx_count = ch->rx_count;
1355         goto out;
1356 }
1357
1358 static int veth_open(struct net_device *dev)
1359 {
1360         struct veth_priv *priv = netdev_priv(dev);
1361         struct net_device *peer = rtnl_dereference(priv->peer);
1362         int err;
1363
1364         if (!peer)
1365                 return -ENOTCONN;
1366
1367         if (priv->_xdp_prog) {
1368                 err = veth_enable_xdp(dev);
1369                 if (err)
1370                         return err;
1371         } else if (veth_gro_requested(dev)) {
1372                 err = veth_napi_enable(dev);
1373                 if (err)
1374                         return err;
1375         }
1376
1377         if (peer->flags & IFF_UP) {
1378                 netif_carrier_on(dev);
1379                 netif_carrier_on(peer);
1380         }
1381
1382         return 0;
1383 }
1384
1385 static int veth_close(struct net_device *dev)
1386 {
1387         struct veth_priv *priv = netdev_priv(dev);
1388         struct net_device *peer = rtnl_dereference(priv->peer);
1389
1390         netif_carrier_off(dev);
1391         if (peer)
1392                 netif_carrier_off(peer);
1393
1394         if (priv->_xdp_prog)
1395                 veth_disable_xdp(dev);
1396         else if (veth_gro_requested(dev))
1397                 veth_napi_del(dev);
1398
1399         return 0;
1400 }
1401
1402 static int is_valid_veth_mtu(int mtu)
1403 {
1404         return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1405 }
1406
1407 static int veth_alloc_queues(struct net_device *dev)
1408 {
1409         struct veth_priv *priv = netdev_priv(dev);
1410         int i;
1411
1412         priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL_ACCOUNT);
1413         if (!priv->rq)
1414                 return -ENOMEM;
1415
1416         for (i = 0; i < dev->num_rx_queues; i++) {
1417                 priv->rq[i].dev = dev;
1418                 u64_stats_init(&priv->rq[i].stats.syncp);
1419         }
1420
1421         return 0;
1422 }
1423
1424 static void veth_free_queues(struct net_device *dev)
1425 {
1426         struct veth_priv *priv = netdev_priv(dev);
1427
1428         kfree(priv->rq);
1429 }
1430
1431 static int veth_dev_init(struct net_device *dev)
1432 {
1433         int err;
1434
1435         dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1436         if (!dev->lstats)
1437                 return -ENOMEM;
1438
1439         err = veth_alloc_queues(dev);
1440         if (err) {
1441                 free_percpu(dev->lstats);
1442                 return err;
1443         }
1444
1445         return 0;
1446 }
1447
1448 static void veth_dev_free(struct net_device *dev)
1449 {
1450         veth_free_queues(dev);
1451         free_percpu(dev->lstats);
1452 }
1453
1454 #ifdef CONFIG_NET_POLL_CONTROLLER
1455 static void veth_poll_controller(struct net_device *dev)
1456 {
1457         /* veth only receives frames when its peer sends one
1458          * Since it has nothing to do with disabling irqs, we are guaranteed
1459          * never to have pending data when we poll for it so
1460          * there is nothing to do here.
1461          *
1462          * We need this though so netpoll recognizes us as an interface that
1463          * supports polling, which enables bridge devices in virt setups to
1464          * still use netconsole
1465          */
1466 }
1467 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1468
1469 static int veth_get_iflink(const struct net_device *dev)
1470 {
1471         struct veth_priv *priv = netdev_priv(dev);
1472         struct net_device *peer;
1473         int iflink;
1474
1475         rcu_read_lock();
1476         peer = rcu_dereference(priv->peer);
1477         iflink = peer ? peer->ifindex : 0;
1478         rcu_read_unlock();
1479
1480         return iflink;
1481 }
1482
1483 static netdev_features_t veth_fix_features(struct net_device *dev,
1484                                            netdev_features_t features)
1485 {
1486         struct veth_priv *priv = netdev_priv(dev);
1487         struct net_device *peer;
1488
1489         peer = rtnl_dereference(priv->peer);
1490         if (peer) {
1491                 struct veth_priv *peer_priv = netdev_priv(peer);
1492
1493                 if (peer_priv->_xdp_prog)
1494                         features &= ~NETIF_F_GSO_SOFTWARE;
1495         }
1496         if (priv->_xdp_prog)
1497                 features |= NETIF_F_GRO;
1498
1499         return features;
1500 }
1501
1502 static int veth_set_features(struct net_device *dev,
1503                              netdev_features_t features)
1504 {
1505         netdev_features_t changed = features ^ dev->features;
1506         struct veth_priv *priv = netdev_priv(dev);
1507         int err;
1508
1509         if (!(changed & NETIF_F_GRO) || !(dev->flags & IFF_UP) || priv->_xdp_prog)
1510                 return 0;
1511
1512         if (features & NETIF_F_GRO) {
1513                 err = veth_napi_enable(dev);
1514                 if (err)
1515                         return err;
1516
1517                 xdp_features_set_redirect_target(dev, true);
1518         } else {
1519                 xdp_features_clear_redirect_target(dev);
1520                 veth_napi_del(dev);
1521         }
1522         return 0;
1523 }
1524
1525 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1526 {
1527         struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1528         struct net_device *peer;
1529
1530         if (new_hr < 0)
1531                 new_hr = 0;
1532
1533         rcu_read_lock();
1534         peer = rcu_dereference(priv->peer);
1535         if (unlikely(!peer))
1536                 goto out;
1537
1538         peer_priv = netdev_priv(peer);
1539         priv->requested_headroom = new_hr;
1540         new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1541         dev->needed_headroom = new_hr;
1542         peer->needed_headroom = new_hr;
1543
1544 out:
1545         rcu_read_unlock();
1546 }
1547
1548 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1549                         struct netlink_ext_ack *extack)
1550 {
1551         struct veth_priv *priv = netdev_priv(dev);
1552         struct bpf_prog *old_prog;
1553         struct net_device *peer;
1554         unsigned int max_mtu;
1555         int err;
1556
1557         old_prog = priv->_xdp_prog;
1558         priv->_xdp_prog = prog;
1559         peer = rtnl_dereference(priv->peer);
1560
1561         if (prog) {
1562                 if (!peer) {
1563                         NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1564                         err = -ENOTCONN;
1565                         goto err;
1566                 }
1567
1568                 max_mtu = SKB_WITH_OVERHEAD(PAGE_SIZE - VETH_XDP_HEADROOM) -
1569                           peer->hard_header_len;
1570                 /* Allow increasing the max_mtu if the program supports
1571                  * XDP fragments.
1572                  */
1573                 if (prog->aux->xdp_has_frags)
1574                         max_mtu += PAGE_SIZE * MAX_SKB_FRAGS;
1575
1576                 if (peer->mtu > max_mtu) {
1577                         NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1578                         err = -ERANGE;
1579                         goto err;
1580                 }
1581
1582                 if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1583                         NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1584                         err = -ENOSPC;
1585                         goto err;
1586                 }
1587
1588                 if (dev->flags & IFF_UP) {
1589                         err = veth_enable_xdp(dev);
1590                         if (err) {
1591                                 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1592                                 goto err;
1593                         }
1594                 }
1595
1596                 if (!old_prog) {
1597                         peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1598                         peer->max_mtu = max_mtu;
1599                 }
1600
1601                 xdp_features_set_redirect_target(dev, true);
1602         }
1603
1604         if (old_prog) {
1605                 if (!prog) {
1606                         if (!veth_gro_requested(dev))
1607                                 xdp_features_clear_redirect_target(dev);
1608
1609                         if (dev->flags & IFF_UP)
1610                                 veth_disable_xdp(dev);
1611
1612                         if (peer) {
1613                                 peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1614                                 peer->max_mtu = ETH_MAX_MTU;
1615                         }
1616                 }
1617                 bpf_prog_put(old_prog);
1618         }
1619
1620         if ((!!old_prog ^ !!prog) && peer)
1621                 netdev_update_features(peer);
1622
1623         return 0;
1624 err:
1625         priv->_xdp_prog = old_prog;
1626
1627         return err;
1628 }
1629
1630 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1631 {
1632         switch (xdp->command) {
1633         case XDP_SETUP_PROG:
1634                 return veth_xdp_set(dev, xdp->prog, xdp->extack);
1635         default:
1636                 return -EINVAL;
1637         }
1638 }
1639
1640 static int veth_xdp_rx_timestamp(const struct xdp_md *ctx, u64 *timestamp)
1641 {
1642         struct veth_xdp_buff *_ctx = (void *)ctx;
1643
1644         if (!_ctx->skb)
1645                 return -EOPNOTSUPP;
1646
1647         *timestamp = skb_hwtstamps(_ctx->skb)->hwtstamp;
1648         return 0;
1649 }
1650
1651 static int veth_xdp_rx_hash(const struct xdp_md *ctx, u32 *hash)
1652 {
1653         struct veth_xdp_buff *_ctx = (void *)ctx;
1654
1655         if (!_ctx->skb)
1656                 return -EOPNOTSUPP;
1657
1658         *hash = skb_get_hash(_ctx->skb);
1659         return 0;
1660 }
1661
1662 static const struct net_device_ops veth_netdev_ops = {
1663         .ndo_init            = veth_dev_init,
1664         .ndo_open            = veth_open,
1665         .ndo_stop            = veth_close,
1666         .ndo_start_xmit      = veth_xmit,
1667         .ndo_get_stats64     = veth_get_stats64,
1668         .ndo_set_rx_mode     = veth_set_multicast_list,
1669         .ndo_set_mac_address = eth_mac_addr,
1670 #ifdef CONFIG_NET_POLL_CONTROLLER
1671         .ndo_poll_controller    = veth_poll_controller,
1672 #endif
1673         .ndo_get_iflink         = veth_get_iflink,
1674         .ndo_fix_features       = veth_fix_features,
1675         .ndo_set_features       = veth_set_features,
1676         .ndo_features_check     = passthru_features_check,
1677         .ndo_set_rx_headroom    = veth_set_rx_headroom,
1678         .ndo_bpf                = veth_xdp,
1679         .ndo_xdp_xmit           = veth_ndo_xdp_xmit,
1680         .ndo_get_peer_dev       = veth_peer_dev,
1681 };
1682
1683 static const struct xdp_metadata_ops veth_xdp_metadata_ops = {
1684         .xmo_rx_timestamp               = veth_xdp_rx_timestamp,
1685         .xmo_rx_hash                    = veth_xdp_rx_hash,
1686 };
1687
1688 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1689                        NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1690                        NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1691                        NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1692                        NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1693
1694 static void veth_setup(struct net_device *dev)
1695 {
1696         ether_setup(dev);
1697
1698         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1699         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1700         dev->priv_flags |= IFF_NO_QUEUE;
1701         dev->priv_flags |= IFF_PHONY_HEADROOM;
1702
1703         dev->netdev_ops = &veth_netdev_ops;
1704         dev->xdp_metadata_ops = &veth_xdp_metadata_ops;
1705         dev->ethtool_ops = &veth_ethtool_ops;
1706         dev->features |= NETIF_F_LLTX;
1707         dev->features |= VETH_FEATURES;
1708         dev->vlan_features = dev->features &
1709                              ~(NETIF_F_HW_VLAN_CTAG_TX |
1710                                NETIF_F_HW_VLAN_STAG_TX |
1711                                NETIF_F_HW_VLAN_CTAG_RX |
1712                                NETIF_F_HW_VLAN_STAG_RX);
1713         dev->needs_free_netdev = true;
1714         dev->priv_destructor = veth_dev_free;
1715         dev->max_mtu = ETH_MAX_MTU;
1716
1717         dev->hw_features = VETH_FEATURES;
1718         dev->hw_enc_features = VETH_FEATURES;
1719         dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1720         netif_set_tso_max_size(dev, GSO_MAX_SIZE);
1721 }
1722
1723 /*
1724  * netlink interface
1725  */
1726
1727 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1728                          struct netlink_ext_ack *extack)
1729 {
1730         if (tb[IFLA_ADDRESS]) {
1731                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1732                         return -EINVAL;
1733                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1734                         return -EADDRNOTAVAIL;
1735         }
1736         if (tb[IFLA_MTU]) {
1737                 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1738                         return -EINVAL;
1739         }
1740         return 0;
1741 }
1742
1743 static struct rtnl_link_ops veth_link_ops;
1744
1745 static void veth_disable_gro(struct net_device *dev)
1746 {
1747         dev->features &= ~NETIF_F_GRO;
1748         dev->wanted_features &= ~NETIF_F_GRO;
1749         netdev_update_features(dev);
1750 }
1751
1752 static int veth_init_queues(struct net_device *dev, struct nlattr *tb[])
1753 {
1754         int err;
1755
1756         if (!tb[IFLA_NUM_TX_QUEUES] && dev->num_tx_queues > 1) {
1757                 err = netif_set_real_num_tx_queues(dev, 1);
1758                 if (err)
1759                         return err;
1760         }
1761         if (!tb[IFLA_NUM_RX_QUEUES] && dev->num_rx_queues > 1) {
1762                 err = netif_set_real_num_rx_queues(dev, 1);
1763                 if (err)
1764                         return err;
1765         }
1766         return 0;
1767 }
1768
1769 static int veth_newlink(struct net *src_net, struct net_device *dev,
1770                         struct nlattr *tb[], struct nlattr *data[],
1771                         struct netlink_ext_ack *extack)
1772 {
1773         int err;
1774         struct net_device *peer;
1775         struct veth_priv *priv;
1776         char ifname[IFNAMSIZ];
1777         struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1778         unsigned char name_assign_type;
1779         struct ifinfomsg *ifmp;
1780         struct net *net;
1781
1782         /*
1783          * create and register peer first
1784          */
1785         if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1786                 struct nlattr *nla_peer;
1787
1788                 nla_peer = data[VETH_INFO_PEER];
1789                 ifmp = nla_data(nla_peer);
1790                 err = rtnl_nla_parse_ifla(peer_tb,
1791                                           nla_data(nla_peer) + sizeof(struct ifinfomsg),
1792                                           nla_len(nla_peer) - sizeof(struct ifinfomsg),
1793                                           NULL);
1794                 if (err < 0)
1795                         return err;
1796
1797                 err = veth_validate(peer_tb, NULL, extack);
1798                 if (err < 0)
1799                         return err;
1800
1801                 tbp = peer_tb;
1802         } else {
1803                 ifmp = NULL;
1804                 tbp = tb;
1805         }
1806
1807         if (ifmp && tbp[IFLA_IFNAME]) {
1808                 nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1809                 name_assign_type = NET_NAME_USER;
1810         } else {
1811                 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1812                 name_assign_type = NET_NAME_ENUM;
1813         }
1814
1815         net = rtnl_link_get_net(src_net, tbp);
1816         if (IS_ERR(net))
1817                 return PTR_ERR(net);
1818
1819         peer = rtnl_create_link(net, ifname, name_assign_type,
1820                                 &veth_link_ops, tbp, extack);
1821         if (IS_ERR(peer)) {
1822                 put_net(net);
1823                 return PTR_ERR(peer);
1824         }
1825
1826         if (!ifmp || !tbp[IFLA_ADDRESS])
1827                 eth_hw_addr_random(peer);
1828
1829         if (ifmp && (dev->ifindex != 0))
1830                 peer->ifindex = ifmp->ifi_index;
1831
1832         netif_inherit_tso_max(peer, dev);
1833
1834         err = register_netdevice(peer);
1835         put_net(net);
1836         net = NULL;
1837         if (err < 0)
1838                 goto err_register_peer;
1839
1840         /* keep GRO disabled by default to be consistent with the established
1841          * veth behavior
1842          */
1843         veth_disable_gro(peer);
1844         netif_carrier_off(peer);
1845
1846         err = rtnl_configure_link(peer, ifmp, 0, NULL);
1847         if (err < 0)
1848                 goto err_configure_peer;
1849
1850         /*
1851          * register dev last
1852          *
1853          * note, that since we've registered new device the dev's name
1854          * should be re-allocated
1855          */
1856
1857         if (tb[IFLA_ADDRESS] == NULL)
1858                 eth_hw_addr_random(dev);
1859
1860         if (tb[IFLA_IFNAME])
1861                 nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1862         else
1863                 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1864
1865         err = register_netdevice(dev);
1866         if (err < 0)
1867                 goto err_register_dev;
1868
1869         netif_carrier_off(dev);
1870
1871         /*
1872          * tie the deviced together
1873          */
1874
1875         priv = netdev_priv(dev);
1876         rcu_assign_pointer(priv->peer, peer);
1877         err = veth_init_queues(dev, tb);
1878         if (err)
1879                 goto err_queues;
1880
1881         priv = netdev_priv(peer);
1882         rcu_assign_pointer(priv->peer, dev);
1883         err = veth_init_queues(peer, tb);
1884         if (err)
1885                 goto err_queues;
1886
1887         veth_disable_gro(dev);
1888         /* update XDP supported features */
1889         veth_set_xdp_features(dev);
1890         veth_set_xdp_features(peer);
1891
1892         return 0;
1893
1894 err_queues:
1895         unregister_netdevice(dev);
1896 err_register_dev:
1897         /* nothing to do */
1898 err_configure_peer:
1899         unregister_netdevice(peer);
1900         return err;
1901
1902 err_register_peer:
1903         free_netdev(peer);
1904         return err;
1905 }
1906
1907 static void veth_dellink(struct net_device *dev, struct list_head *head)
1908 {
1909         struct veth_priv *priv;
1910         struct net_device *peer;
1911
1912         priv = netdev_priv(dev);
1913         peer = rtnl_dereference(priv->peer);
1914
1915         /* Note : dellink() is called from default_device_exit_batch(),
1916          * before a rcu_synchronize() point. The devices are guaranteed
1917          * not being freed before one RCU grace period.
1918          */
1919         RCU_INIT_POINTER(priv->peer, NULL);
1920         unregister_netdevice_queue(dev, head);
1921
1922         if (peer) {
1923                 priv = netdev_priv(peer);
1924                 RCU_INIT_POINTER(priv->peer, NULL);
1925                 unregister_netdevice_queue(peer, head);
1926         }
1927 }
1928
1929 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1930         [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
1931 };
1932
1933 static struct net *veth_get_link_net(const struct net_device *dev)
1934 {
1935         struct veth_priv *priv = netdev_priv(dev);
1936         struct net_device *peer = rtnl_dereference(priv->peer);
1937
1938         return peer ? dev_net(peer) : dev_net(dev);
1939 }
1940
1941 static unsigned int veth_get_num_queues(void)
1942 {
1943         /* enforce the same queue limit as rtnl_create_link */
1944         int queues = num_possible_cpus();
1945
1946         if (queues > 4096)
1947                 queues = 4096;
1948         return queues;
1949 }
1950
1951 static struct rtnl_link_ops veth_link_ops = {
1952         .kind           = DRV_NAME,
1953         .priv_size      = sizeof(struct veth_priv),
1954         .setup          = veth_setup,
1955         .validate       = veth_validate,
1956         .newlink        = veth_newlink,
1957         .dellink        = veth_dellink,
1958         .policy         = veth_policy,
1959         .maxtype        = VETH_INFO_MAX,
1960         .get_link_net   = veth_get_link_net,
1961         .get_num_tx_queues      = veth_get_num_queues,
1962         .get_num_rx_queues      = veth_get_num_queues,
1963 };
1964
1965 /*
1966  * init/fini
1967  */
1968
1969 static __init int veth_init(void)
1970 {
1971         return rtnl_link_register(&veth_link_ops);
1972 }
1973
1974 static __exit void veth_exit(void)
1975 {
1976         rtnl_link_unregister(&veth_link_ops);
1977 }
1978
1979 module_init(veth_init);
1980 module_exit(veth_exit);
1981
1982 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1983 MODULE_LICENSE("GPL v2");
1984 MODULE_ALIAS_RTNL_LINK(DRV_NAME);