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