drivers/net/veth.c

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