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