/drivers/net/veth.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  drivers/net/veth.c
   4 *
   5 *  Copyright (C) 2007 OpenVZ http://openvz.org, SWsoft Inc
   6 *
   7 * Author: Pavel Emelianov <xemul@openvz.org>
   8 * Ethtool interface from: Eric W. Biederman <ebiederm@xmission.com>
   9 *
  10 */
  11
  12#include <linux/netdevice.h>
  13#include <linux/slab.h>
  14#include <linux/ethtool.h>
  15#include <linux/etherdevice.h>
  16#include <linux/u64_stats_sync.h>
  17
  18#include <net/rtnetlink.h>
  19#include <net/dst.h>
  20#include <net/xfrm.h>
  21#include <net/xdp.h>
  22#include <linux/veth.h>
  23#include <linux/module.h>
  24#include <linux/bpf.h>
  25#include <linux/filter.h>
  26#include <linux/ptr_ring.h>
  27#include <linux/bpf_trace.h>
  28#include <linux/net_tstamp.h>
  29
  30#define DRV_NAME	"veth"
  31#define DRV_VERSION	"1.0"
  32
  33#define VETH_XDP_FLAG		BIT(0)
  34#define VETH_RING_SIZE		256
  35#define VETH_XDP_HEADROOM	(XDP_PACKET_HEADROOM + NET_IP_ALIGN)
  36
  37#define VETH_XDP_TX_BULK_SIZE	16
  38
  39struct veth_stats {
  40	u64	rx_drops;
  41	/* xdp */
  42	u64	xdp_packets;
  43	u64	xdp_bytes;
  44	u64	xdp_redirect;
  45	u64	xdp_drops;
  46	u64	xdp_tx;
  47	u64	xdp_tx_err;
  48	u64	peer_tq_xdp_xmit;
  49	u64	peer_tq_xdp_xmit_err;
  50};
  51
  52struct veth_rq_stats {
  53	struct veth_stats	vs;
  54	struct u64_stats_sync	syncp;
  55};
  56
  57struct veth_rq {
  58	struct napi_struct	xdp_napi;
  59	struct net_device	*dev;
  60	struct bpf_prog __rcu	*xdp_prog;
  61	struct xdp_mem_info	xdp_mem;
  62	struct veth_rq_stats	stats;
  63	bool			rx_notify_masked;
  64	struct ptr_ring		xdp_ring;
  65	struct xdp_rxq_info	xdp_rxq;
  66};
  67
  68struct veth_priv {
  69	struct net_device __rcu	*peer;
  70	atomic64_t		dropped;
  71	struct bpf_prog		*_xdp_prog;
  72	struct veth_rq		*rq;
  73	unsigned int		requested_headroom;
  74};
  75
  76struct veth_xdp_tx_bq {
  77	struct xdp_frame *q[VETH_XDP_TX_BULK_SIZE];
  78	unsigned int count;
  79};
  80
  81/*
  82 * ethtool interface
  83 */
  84
  85struct veth_q_stat_desc {
  86	char	desc[ETH_GSTRING_LEN];
  87	size_t	offset;
  88};
  89
  90#define VETH_RQ_STAT(m)	offsetof(struct veth_stats, m)
  91
  92static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
  93	{ "xdp_packets",	VETH_RQ_STAT(xdp_packets) },
  94	{ "xdp_bytes",		VETH_RQ_STAT(xdp_bytes) },
  95	{ "drops",		VETH_RQ_STAT(rx_drops) },
  96	{ "xdp_redirect",	VETH_RQ_STAT(xdp_redirect) },
  97	{ "xdp_drops",		VETH_RQ_STAT(xdp_drops) },
  98	{ "xdp_tx",		VETH_RQ_STAT(xdp_tx) },
  99	{ "xdp_tx_errors",	VETH_RQ_STAT(xdp_tx_err) },
 100};
 101
 102#define VETH_RQ_STATS_LEN	ARRAY_SIZE(veth_rq_stats_desc)
 103
 104static const struct veth_q_stat_desc veth_tq_stats_desc[] = {
 105	{ "xdp_xmit",		VETH_RQ_STAT(peer_tq_xdp_xmit) },
 106	{ "xdp_xmit_errors",	VETH_RQ_STAT(peer_tq_xdp_xmit_err) },
 107};
 108
 109#define VETH_TQ_STATS_LEN	ARRAY_SIZE(veth_tq_stats_desc)
 110
 111static struct {
 112	const char string[ETH_GSTRING_LEN];
 113} ethtool_stats_keys[] = {
 114	{ "peer_ifindex" },
 115};
 116
 117static int veth_get_link_ksettings(struct net_device *dev,
 118				   struct ethtool_link_ksettings *cmd)
 119{
 120	cmd->base.speed		= SPEED_10000;
 121	cmd->base.duplex	= DUPLEX_FULL;
 122	cmd->base.port		= PORT_TP;
 123	cmd->base.autoneg	= AUTONEG_DISABLE;
 124	return 0;
 125}
 126
 127static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
 128{
 129	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
 130	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
 131}
 132
 133static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
 134{
 135	char *p = (char *)buf;
 136	int i, j;
 137
 138	switch(stringset) {
 139	case ETH_SS_STATS:
 140		memcpy(p, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
 141		p += sizeof(ethtool_stats_keys);
 142		for (i = 0; i < dev->real_num_rx_queues; i++) {
 143			for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
 144				snprintf(p, ETH_GSTRING_LEN,
 145					 "rx_queue_%u_%.18s",
 146					 i, veth_rq_stats_desc[j].desc);
 147				p += ETH_GSTRING_LEN;
 148			}
 149		}
 150		for (i = 0; i < dev->real_num_tx_queues; i++) {
 151			for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
 152				snprintf(p, ETH_GSTRING_LEN,
 153					 "tx_queue_%u_%.18s",
 154					 i, veth_tq_stats_desc[j].desc);
 155				p += ETH_GSTRING_LEN;
 156			}
 157		}
 158		break;
 159	}
 160}
 161
 162static int veth_get_sset_count(struct net_device *dev, int sset)
 163{
 164	switch (sset) {
 165	case ETH_SS_STATS:
 166		return ARRAY_SIZE(ethtool_stats_keys) +
 167		       VETH_RQ_STATS_LEN * dev->real_num_rx_queues +
 168		       VETH_TQ_STATS_LEN * dev->real_num_tx_queues;
 169	default:
 170		return -EOPNOTSUPP;
 171	}
 172}
 173
 174static void veth_get_ethtool_stats(struct net_device *dev,
 175		struct ethtool_stats *stats, u64 *data)
 176{
 177	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 178	struct net_device *peer = rtnl_dereference(priv->peer);
 179	int i, j, idx;
 180
 181	data[0] = peer ? peer->ifindex : 0;
 182	idx = 1;
 183	for (i = 0; i < dev->real_num_rx_queues; i++) {
 184		const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
 185		const void *stats_base = (void *)&rq_stats->vs;
 186		unsigned int start;
 187		size_t offset;
 188
 189		do {
 190			start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
 191			for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
 192				offset = veth_rq_stats_desc[j].offset;
 193				data[idx + j] = *(u64 *)(stats_base + offset);
 194			}
 195		} while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
 196		idx += VETH_RQ_STATS_LEN;
 197	}
 198
 199	if (!peer)
 200		return;
 201
 202	rcv_priv = netdev_priv(peer);
 203	for (i = 0; i < peer->real_num_rx_queues; i++) {
 204		const struct veth_rq_stats *rq_stats = &rcv_priv->rq[i].stats;
 205		const void *base = (void *)&rq_stats->vs;
 206		unsigned int start, tx_idx = idx;
 207		size_t offset;
 208
 209		tx_idx += (i % dev->real_num_tx_queues) * VETH_TQ_STATS_LEN;
 210		do {
 211			start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
 212			for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
 213				offset = veth_tq_stats_desc[j].offset;
 214				data[tx_idx + j] += *(u64 *)(base + offset);
 215			}
 216		} while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
 217	}
 218}
 219
 220static const struct ethtool_ops veth_ethtool_ops = {
 221	.get_drvinfo		= veth_get_drvinfo,
 222	.get_link		= ethtool_op_get_link,
 223	.get_strings		= veth_get_strings,
 224	.get_sset_count		= veth_get_sset_count,
 225	.get_ethtool_stats	= veth_get_ethtool_stats,
 226	.get_link_ksettings	= veth_get_link_ksettings,
 227	.get_ts_info		= ethtool_op_get_ts_info,
 228};
 229
 230/* general routines */
 231
 232static bool veth_is_xdp_frame(void *ptr)
 233{
 234	return (unsigned long)ptr & VETH_XDP_FLAG;
 235}
 236
 237static void *veth_ptr_to_xdp(void *ptr)
 238{
 239	return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
 240}
 241
 242static void *veth_xdp_to_ptr(void *ptr)
 243{
 244	return (void *)((unsigned long)ptr | VETH_XDP_FLAG);
 245}
 246
 247static void veth_ptr_free(void *ptr)
 248{
 249	if (veth_is_xdp_frame(ptr))
 250		xdp_return_frame(veth_ptr_to_xdp(ptr));
 251	else
 252		kfree_skb(ptr);
 253}
 254
 255static void __veth_xdp_flush(struct veth_rq *rq)
 256{
 257	/* Write ptr_ring before reading rx_notify_masked */
 258	smp_mb();
 259	if (!rq->rx_notify_masked) {
 260		rq->rx_notify_masked = true;
 261		napi_schedule(&rq->xdp_napi);
 262	}
 263}
 264
 265static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
 266{
 267	if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
 268		dev_kfree_skb_any(skb);
 269		return NET_RX_DROP;
 270	}
 271
 272	return NET_RX_SUCCESS;
 273}
 274
 275static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
 276			    struct veth_rq *rq, bool xdp)
 277{
 278	return __dev_forward_skb(dev, skb) ?: xdp ?
 279		veth_xdp_rx(rq, skb) :
 280		netif_rx(skb);
 281}
 282
 283static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 284{
 285	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 286	struct veth_rq *rq = NULL;
 287	struct net_device *rcv;
 288	int length = skb->len;
 289	bool rcv_xdp = false;
 290	int rxq;
 291
 292	rcu_read_lock();
 293	rcv = rcu_dereference(priv->peer);
 294	if (unlikely(!rcv)) {
 295		kfree_skb(skb);
 296		goto drop;
 297	}
 298
 299	rcv_priv = netdev_priv(rcv);
 300	rxq = skb_get_queue_mapping(skb);
 301	if (rxq < rcv->real_num_rx_queues) {
 302		rq = &rcv_priv->rq[rxq];
 303		rcv_xdp = rcu_access_pointer(rq->xdp_prog);
 304		if (rcv_xdp)
 305			skb_record_rx_queue(skb, rxq);
 306	}
 307
 308	skb_tx_timestamp(skb);
 309	if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
 310		if (!rcv_xdp)
 311			dev_lstats_add(dev, length);
 312	} else {
 313drop:
 314		atomic64_inc(&priv->dropped);
 315	}
 316
 317	if (rcv_xdp)
 318		__veth_xdp_flush(rq);
 319
 320	rcu_read_unlock();
 321
 322	return NETDEV_TX_OK;
 323}
 324
 325static u64 veth_stats_tx(struct net_device *dev, u64 *packets, u64 *bytes)
 326{
 327	struct veth_priv *priv = netdev_priv(dev);
 328
 329	dev_lstats_read(dev, packets, bytes);
 330	return atomic64_read(&priv->dropped);
 331}
 332
 333static void veth_stats_rx(struct veth_stats *result, struct net_device *dev)
 334{
 335	struct veth_priv *priv = netdev_priv(dev);
 336	int i;
 337
 338	result->peer_tq_xdp_xmit_err = 0;
 339	result->xdp_packets = 0;
 340	result->xdp_tx_err = 0;
 341	result->xdp_bytes = 0;
 342	result->rx_drops = 0;
 343	for (i = 0; i < dev->num_rx_queues; i++) {
 344		u64 packets, bytes, drops, xdp_tx_err, peer_tq_xdp_xmit_err;
 345		struct veth_rq_stats *stats = &priv->rq[i].stats;
 346		unsigned int start;
 347
 348		do {
 349			start = u64_stats_fetch_begin_irq(&stats->syncp);
 350			peer_tq_xdp_xmit_err = stats->vs.peer_tq_xdp_xmit_err;
 351			xdp_tx_err = stats->vs.xdp_tx_err;
 352			packets = stats->vs.xdp_packets;
 353			bytes = stats->vs.xdp_bytes;
 354			drops = stats->vs.rx_drops;
 355		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 356		result->peer_tq_xdp_xmit_err += peer_tq_xdp_xmit_err;
 357		result->xdp_tx_err += xdp_tx_err;
 358		result->xdp_packets += packets;
 359		result->xdp_bytes += bytes;
 360		result->rx_drops += drops;
 361	}
 362}
 363
 364static void veth_get_stats64(struct net_device *dev,
 365			     struct rtnl_link_stats64 *tot)
 366{
 367	struct veth_priv *priv = netdev_priv(dev);
 368	struct net_device *peer;
 369	struct veth_stats rx;
 370	u64 packets, bytes;
 371
 372	tot->tx_dropped = veth_stats_tx(dev, &packets, &bytes);
 373	tot->tx_bytes = bytes;
 374	tot->tx_packets = packets;
 375
 376	veth_stats_rx(&rx, dev);
 377	tot->tx_dropped += rx.xdp_tx_err;
 378	tot->rx_dropped = rx.rx_drops + rx.peer_tq_xdp_xmit_err;
 379	tot->rx_bytes = rx.xdp_bytes;
 380	tot->rx_packets = rx.xdp_packets;
 381
 382	rcu_read_lock();
 383	peer = rcu_dereference(priv->peer);
 384	if (peer) {
 385		veth_stats_tx(peer, &packets, &bytes);
 386		tot->rx_bytes += bytes;
 387		tot->rx_packets += packets;
 388
 389		veth_stats_rx(&rx, peer);
 390		tot->tx_dropped += rx.peer_tq_xdp_xmit_err;
 391		tot->rx_dropped += rx.xdp_tx_err;
 392		tot->tx_bytes += rx.xdp_bytes;
 393		tot->tx_packets += rx.xdp_packets;
 394	}
 395	rcu_read_unlock();
 396}
 397
 398/* fake multicast ability */
 399static void veth_set_multicast_list(struct net_device *dev)
 400{
 401}
 402
 403static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
 404				      int buflen)
 405{
 406	struct sk_buff *skb;
 407
 408	if (!buflen) {
 409		buflen = SKB_DATA_ALIGN(headroom + len) +
 410			 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 411	}
 412	skb = build_skb(head, buflen);
 413	if (!skb)
 414		return NULL;
 415
 416	skb_reserve(skb, headroom);
 417	skb_put(skb, len);
 418
 419	return skb;
 420}
 421
 422static int veth_select_rxq(struct net_device *dev)
 423{
 424	return smp_processor_id() % dev->real_num_rx_queues;
 425}
 426
 427static int veth_xdp_xmit(struct net_device *dev, int n,
 428			 struct xdp_frame **frames,
 429			 u32 flags, bool ndo_xmit)
 430{
 431	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 432	int i, ret = -ENXIO, drops = 0;
 433	struct net_device *rcv;
 434	unsigned int max_len;
 435	struct veth_rq *rq;
 436
 437	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 438		return -EINVAL;
 439
 440	rcu_read_lock();
 441	rcv = rcu_dereference(priv->peer);
 442	if (unlikely(!rcv))
 443		goto out;
 444
 445	rcv_priv = netdev_priv(rcv);
 446	rq = &rcv_priv->rq[veth_select_rxq(rcv)];
 447	/* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
 448	 * side. This means an XDP program is loaded on the peer and the peer
 449	 * device is up.
 450	 */
 451	if (!rcu_access_pointer(rq->xdp_prog))
 452		goto out;
 453
 454	max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
 455
 456	spin_lock(&rq->xdp_ring.producer_lock);
 457	for (i = 0; i < n; i++) {
 458		struct xdp_frame *frame = frames[i];
 459		void *ptr = veth_xdp_to_ptr(frame);
 460
 461		if (unlikely(frame->len > max_len ||
 462			     __ptr_ring_produce(&rq->xdp_ring, ptr))) {
 463			xdp_return_frame_rx_napi(frame);
 464			drops++;
 465		}
 466	}
 467	spin_unlock(&rq->xdp_ring.producer_lock);
 468
 469	if (flags & XDP_XMIT_FLUSH)
 470		__veth_xdp_flush(rq);
 471
 472	ret = n - drops;
 473	if (ndo_xmit) {
 474		u64_stats_update_begin(&rq->stats.syncp);
 475		rq->stats.vs.peer_tq_xdp_xmit += n - drops;
 476		rq->stats.vs.peer_tq_xdp_xmit_err += drops;
 477		u64_stats_update_end(&rq->stats.syncp);
 478	}
 479
 480out:
 481	rcu_read_unlock();
 482
 483	return ret;
 484}
 485
 486static int veth_ndo_xdp_xmit(struct net_device *dev, int n,
 487			     struct xdp_frame **frames, u32 flags)
 488{
 489	int err;
 490
 491	err = veth_xdp_xmit(dev, n, frames, flags, true);
 492	if (err < 0) {
 493		struct veth_priv *priv = netdev_priv(dev);
 494
 495		atomic64_add(n, &priv->dropped);
 496	}
 497
 498	return err;
 499}
 500
 501static void veth_xdp_flush_bq(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
 502{
 503	int sent, i, err = 0;
 504
 505	sent = veth_xdp_xmit(rq->dev, bq->count, bq->q, 0, false);
 506	if (sent < 0) {
 507		err = sent;
 508		sent = 0;
 509		for (i = 0; i < bq->count; i++)
 510			xdp_return_frame(bq->q[i]);
 511	}
 512	trace_xdp_bulk_tx(rq->dev, sent, bq->count - sent, err);
 513
 514	u64_stats_update_begin(&rq->stats.syncp);
 515	rq->stats.vs.xdp_tx += sent;
 516	rq->stats.vs.xdp_tx_err += bq->count - sent;
 517	u64_stats_update_end(&rq->stats.syncp);
 518
 519	bq->count = 0;
 520}
 521
 522static void veth_xdp_flush(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
 523{
 524	struct veth_priv *rcv_priv, *priv = netdev_priv(rq->dev);
 525	struct net_device *rcv;
 526	struct veth_rq *rcv_rq;
 527
 528	rcu_read_lock();
 529	veth_xdp_flush_bq(rq, bq);
 530	rcv = rcu_dereference(priv->peer);
 531	if (unlikely(!rcv))
 532		goto out;
 533
 534	rcv_priv = netdev_priv(rcv);
 535	rcv_rq = &rcv_priv->rq[veth_select_rxq(rcv)];
 536	/* xdp_ring is initialized on receive side? */
 537	if (unlikely(!rcu_access_pointer(rcv_rq->xdp_prog)))
 538		goto out;
 539
 540	__veth_xdp_flush(rcv_rq);
 541out:
 542	rcu_read_unlock();
 543}
 544
 545static int veth_xdp_tx(struct veth_rq *rq, struct xdp_buff *xdp,
 546		       struct veth_xdp_tx_bq *bq)
 547{
 548	struct xdp_frame *frame = convert_to_xdp_frame(xdp);
 549
 550	if (unlikely(!frame))
 551		return -EOVERFLOW;
 552
 553	if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
 554		veth_xdp_flush_bq(rq, bq);
 555
 556	bq->q[bq->count++] = frame;
 557
 558	return 0;
 559}
 560
 561static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
 562					struct xdp_frame *frame,
 563					struct veth_xdp_tx_bq *bq,
 564					struct veth_stats *stats)
 565{
 566	void *hard_start = frame->data - frame->headroom;
 567	void *head = hard_start - sizeof(struct xdp_frame);
 568	int len = frame->len, delta = 0;
 569	struct xdp_frame orig_frame;
 570	struct bpf_prog *xdp_prog;
 571	unsigned int headroom;
 572	struct sk_buff *skb;
 573
 574	rcu_read_lock();
 575	xdp_prog = rcu_dereference(rq->xdp_prog);
 576	if (likely(xdp_prog)) {
 577		struct xdp_buff xdp;
 578		u32 act;
 579
 580		xdp.data_hard_start = hard_start;
 581		xdp.data = frame->data;
 582		xdp.data_end = frame->data + frame->len;
 583		xdp.data_meta = frame->data - frame->metasize;
 584		xdp.rxq = &rq->xdp_rxq;
 585
 586		act = bpf_prog_run_xdp(xdp_prog, &xdp);
 587
 588		switch (act) {
 589		case XDP_PASS:
 590			delta = frame->data - xdp.data;
 591			len = xdp.data_end - xdp.data;
 592			break;
 593		case XDP_TX:
 594			orig_frame = *frame;
 595			xdp.data_hard_start = head;
 596			xdp.rxq->mem = frame->mem;
 597			if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
 598				trace_xdp_exception(rq->dev, xdp_prog, act);
 599				frame = &orig_frame;
 600				stats->rx_drops++;
 601				goto err_xdp;
 602			}
 603			stats->xdp_tx++;
 604			rcu_read_unlock();
 605			goto xdp_xmit;
 606		case XDP_REDIRECT:
 607			orig_frame = *frame;
 608			xdp.data_hard_start = head;
 609			xdp.rxq->mem = frame->mem;
 610			if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
 611				frame = &orig_frame;
 612				stats->rx_drops++;
 613				goto err_xdp;
 614			}
 615			stats->xdp_redirect++;
 616			rcu_read_unlock();
 617			goto xdp_xmit;
 618		default:
 619			bpf_warn_invalid_xdp_action(act);
 620			/* fall through */
 621		case XDP_ABORTED:
 622			trace_xdp_exception(rq->dev, xdp_prog, act);
 623			/* fall through */
 624		case XDP_DROP:
 625			stats->xdp_drops++;
 626			goto err_xdp;
 627		}
 628	}
 629	rcu_read_unlock();
 630
 631	headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
 632	skb = veth_build_skb(head, headroom, len, 0);
 633	if (!skb) {
 634		xdp_return_frame(frame);
 635		stats->rx_drops++;
 636		goto err;
 637	}
 638
 639	xdp_release_frame(frame);
 640	xdp_scrub_frame(frame);
 641	skb->protocol = eth_type_trans(skb, rq->dev);
 642err:
 643	return skb;
 644err_xdp:
 645	rcu_read_unlock();
 646	xdp_return_frame(frame);
 647xdp_xmit:
 648	return NULL;
 649}
 650
 651static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
 652					struct sk_buff *skb,
 653					struct veth_xdp_tx_bq *bq,
 654					struct veth_stats *stats)
 655{
 656	u32 pktlen, headroom, act, metalen;
 657	void *orig_data, *orig_data_end;
 658	struct bpf_prog *xdp_prog;
 659	int mac_len, delta, off;
 660	struct xdp_buff xdp;
 661
 662	skb_orphan(skb);
 663
 664	rcu_read_lock();
 665	xdp_prog = rcu_dereference(rq->xdp_prog);
 666	if (unlikely(!xdp_prog)) {
 667		rcu_read_unlock();
 668		goto out;
 669	}
 670
 671	mac_len = skb->data - skb_mac_header(skb);
 672	pktlen = skb->len + mac_len;
 673	headroom = skb_headroom(skb) - mac_len;
 674
 675	if (skb_shared(skb) || skb_head_is_locked(skb) ||
 676	    skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
 677		struct sk_buff *nskb;
 678		int size, head_off;
 679		void *head, *start;
 680		struct page *page;
 681
 682		size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
 683		       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 684		if (size > PAGE_SIZE)
 685			goto drop;
 686
 687		page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
 688		if (!page)
 689			goto drop;
 690
 691		head = page_address(page);
 692		start = head + VETH_XDP_HEADROOM;
 693		if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
 694			page_frag_free(head);
 695			goto drop;
 696		}
 697
 698		nskb = veth_build_skb(head,
 699				      VETH_XDP_HEADROOM + mac_len, skb->len,
 700				      PAGE_SIZE);
 701		if (!nskb) {
 702			page_frag_free(head);
 703			goto drop;
 704		}
 705
 706		skb_copy_header(nskb, skb);
 707		head_off = skb_headroom(nskb) - skb_headroom(skb);
 708		skb_headers_offset_update(nskb, head_off);
 709		consume_skb(skb);
 710		skb = nskb;
 711	}
 712
 713	xdp.data_hard_start = skb->head;
 714	xdp.data = skb_mac_header(skb);
 715	xdp.data_end = xdp.data + pktlen;
 716	xdp.data_meta = xdp.data;
 717	xdp.rxq = &rq->xdp_rxq;
 718	orig_data = xdp.data;
 719	orig_data_end = xdp.data_end;
 720
 721	act = bpf_prog_run_xdp(xdp_prog, &xdp);
 722
 723	switch (act) {
 724	case XDP_PASS:
 725		break;
 726	case XDP_TX:
 727		get_page(virt_to_page(xdp.data));
 728		consume_skb(skb);
 729		xdp.rxq->mem = rq->xdp_mem;
 730		if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
 731			trace_xdp_exception(rq->dev, xdp_prog, act);
 732			stats->rx_drops++;
 733			goto err_xdp;
 734		}
 735		stats->xdp_tx++;
 736		rcu_read_unlock();
 737		goto xdp_xmit;
 738	case XDP_REDIRECT:
 739		get_page(virt_to_page(xdp.data));
 740		consume_skb(skb);
 741		xdp.rxq->mem = rq->xdp_mem;
 742		if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
 743			stats->rx_drops++;
 744			goto err_xdp;
 745		}
 746		stats->xdp_redirect++;
 747		rcu_read_unlock();
 748		goto xdp_xmit;
 749	default:
 750		bpf_warn_invalid_xdp_action(act);
 751		/* fall through */
 752	case XDP_ABORTED:
 753		trace_xdp_exception(rq->dev, xdp_prog, act);
 754		/* fall through */
 755	case XDP_DROP:
 756		stats->xdp_drops++;
 757		goto xdp_drop;
 758	}
 759	rcu_read_unlock();
 760
 761	delta = orig_data - xdp.data;
 762	off = mac_len + delta;
 763	if (off > 0)
 764		__skb_push(skb, off);
 765	else if (off < 0)
 766		__skb_pull(skb, -off);
 767	skb->mac_header -= delta;
 768	off = xdp.data_end - orig_data_end;
 769	if (off != 0)
 770		__skb_put(skb, off);
 771	skb->protocol = eth_type_trans(skb, rq->dev);
 772
 773	metalen = xdp.data - xdp.data_meta;
 774	if (metalen)
 775		skb_metadata_set(skb, metalen);
 776out:
 777	return skb;
 778drop:
 779	stats->rx_drops++;
 780xdp_drop:
 781	rcu_read_unlock();
 782	kfree_skb(skb);
 783	return NULL;
 784err_xdp:
 785	rcu_read_unlock();
 786	page_frag_free(xdp.data);
 787xdp_xmit:
 788	return NULL;
 789}
 790
 791static int veth_xdp_rcv(struct veth_rq *rq, int budget,
 792			struct veth_xdp_tx_bq *bq,
 793			struct veth_stats *stats)
 794{
 795	int i, done = 0;
 796
 797	for (i = 0; i < budget; i++) {
 798		void *ptr = __ptr_ring_consume(&rq->xdp_ring);
 799		struct sk_buff *skb;
 800
 801		if (!ptr)
 802			break;
 803
 804		if (veth_is_xdp_frame(ptr)) {
 805			struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
 806
 807			stats->xdp_bytes += frame->len;
 808			skb = veth_xdp_rcv_one(rq, frame, bq, stats);
 809		} else {
 810			skb = ptr;
 811			stats->xdp_bytes += skb->len;
 812			skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
 813		}
 814
 815		if (skb)
 816			napi_gro_receive(&rq->xdp_napi, skb);
 817
 818		done++;
 819	}
 820
 821	u64_stats_update_begin(&rq->stats.syncp);
 822	rq->stats.vs.xdp_redirect += stats->xdp_redirect;
 823	rq->stats.vs.xdp_bytes += stats->xdp_bytes;
 824	rq->stats.vs.xdp_drops += stats->xdp_drops;
 825	rq->stats.vs.rx_drops += stats->rx_drops;
 826	rq->stats.vs.xdp_packets += done;
 827	u64_stats_update_end(&rq->stats.syncp);
 828
 829	return done;
 830}
 831
 832static int veth_poll(struct napi_struct *napi, int budget)
 833{
 834	struct veth_rq *rq =
 835		container_of(napi, struct veth_rq, xdp_napi);
 836	struct veth_stats stats = {};
 837	struct veth_xdp_tx_bq bq;
 838	int done;
 839
 840	bq.count = 0;
 841
 842	xdp_set_return_frame_no_direct();
 843	done = veth_xdp_rcv(rq, budget, &bq, &stats);
 844
 845	if (done < budget && napi_complete_done(napi, done)) {
 846		/* Write rx_notify_masked before reading ptr_ring */
 847		smp_store_mb(rq->rx_notify_masked, false);
 848		if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
 849			rq->rx_notify_masked = true;
 850			napi_schedule(&rq->xdp_napi);
 851		}
 852	}
 853
 854	if (stats.xdp_tx > 0)
 855		veth_xdp_flush(rq, &bq);
 856	if (stats.xdp_redirect > 0)
 857		xdp_do_flush();
 858	xdp_clear_return_frame_no_direct();
 859
 860	return done;
 861}
 862
 863static int veth_napi_add(struct net_device *dev)
 864{
 865	struct veth_priv *priv = netdev_priv(dev);
 866	int err, i;
 867
 868	for (i = 0; i < dev->real_num_rx_queues; i++) {
 869		struct veth_rq *rq = &priv->rq[i];
 870
 871		err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
 872		if (err)
 873			goto err_xdp_ring;
 874	}
 875
 876	for (i = 0; i < dev->real_num_rx_queues; i++) {
 877		struct veth_rq *rq = &priv->rq[i];
 878
 879		netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
 880		napi_enable(&rq->xdp_napi);
 881	}
 882
 883	return 0;
 884err_xdp_ring:
 885	for (i--; i >= 0; i--)
 886		ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
 887
 888	return err;
 889}
 890
 891static void veth_napi_del(struct net_device *dev)
 892{
 893	struct veth_priv *priv = netdev_priv(dev);
 894	int i;
 895
 896	for (i = 0; i < dev->real_num_rx_queues; i++) {
 897		struct veth_rq *rq = &priv->rq[i];
 898
 899		napi_disable(&rq->xdp_napi);
 900		napi_hash_del(&rq->xdp_napi);
 901	}
 902	synchronize_net();
 903
 904	for (i = 0; i < dev->real_num_rx_queues; i++) {
 905		struct veth_rq *rq = &priv->rq[i];
 906
 907		netif_napi_del(&rq->xdp_napi);
 908		rq->rx_notify_masked = false;
 909		ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
 910	}
 911}
 912
 913static int veth_enable_xdp(struct net_device *dev)
 914{
 915	struct veth_priv *priv = netdev_priv(dev);
 916	int err, i;
 917
 918	if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
 919		for (i = 0; i < dev->real_num_rx_queues; i++) {
 920			struct veth_rq *rq = &priv->rq[i];
 921
 922			err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
 923			if (err < 0)
 924				goto err_rxq_reg;
 925
 926			err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
 927							 MEM_TYPE_PAGE_SHARED,
 928							 NULL);
 929			if (err < 0)
 930				goto err_reg_mem;
 931
 932			/* Save original mem info as it can be overwritten */
 933			rq->xdp_mem = rq->xdp_rxq.mem;
 934		}
 935
 936		err = veth_napi_add(dev);
 937		if (err)
 938			goto err_rxq_reg;
 939	}
 940
 941	for (i = 0; i < dev->real_num_rx_queues; i++)
 942		rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
 943
 944	return 0;
 945err_reg_mem:
 946	xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
 947err_rxq_reg:
 948	for (i--; i >= 0; i--)
 949		xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
 950
 951	return err;
 952}
 953
 954static void veth_disable_xdp(struct net_device *dev)
 955{
 956	struct veth_priv *priv = netdev_priv(dev);
 957	int i;
 958
 959	for (i = 0; i < dev->real_num_rx_queues; i++)
 960		rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
 961	veth_napi_del(dev);
 962	for (i = 0; i < dev->real_num_rx_queues; i++) {
 963		struct veth_rq *rq = &priv->rq[i];
 964
 965		rq->xdp_rxq.mem = rq->xdp_mem;
 966		xdp_rxq_info_unreg(&rq->xdp_rxq);
 967	}
 968}
 969
 970static int veth_open(struct net_device *dev)
 971{
 972	struct veth_priv *priv = netdev_priv(dev);
 973	struct net_device *peer = rtnl_dereference(priv->peer);
 974	int err;
 975
 976	if (!peer)
 977		return -ENOTCONN;
 978
 979	if (priv->_xdp_prog) {
 980		err = veth_enable_xdp(dev);
 981		if (err)
 982			return err;
 983	}
 984
 985	if (peer->flags & IFF_UP) {
 986		netif_carrier_on(dev);
 987		netif_carrier_on(peer);
 988	}
 989
 990	return 0;
 991}
 992
 993static int veth_close(struct net_device *dev)
 994{
 995	struct veth_priv *priv = netdev_priv(dev);
 996	struct net_device *peer = rtnl_dereference(priv->peer);
 997
 998	netif_carrier_off(dev);
 999	if (peer)
1000		netif_carrier_off(peer);
1001
1002	if (priv->_xdp_prog)
1003		veth_disable_xdp(dev);
1004
1005	return 0;
1006}
1007
1008static int is_valid_veth_mtu(int mtu)
1009{
1010	return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1011}
1012
1013static int veth_alloc_queues(struct net_device *dev)
1014{
1015	struct veth_priv *priv = netdev_priv(dev);
1016	int i;
1017
1018	priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
1019	if (!priv->rq)
1020		return -ENOMEM;
1021
1022	for (i = 0; i < dev->num_rx_queues; i++) {
1023		priv->rq[i].dev = dev;
1024		u64_stats_init(&priv->rq[i].stats.syncp);
1025	}
1026
1027	return 0;
1028}
1029
1030static void veth_free_queues(struct net_device *dev)
1031{
1032	struct veth_priv *priv = netdev_priv(dev);
1033
1034	kfree(priv->rq);
1035}
1036
1037static int veth_dev_init(struct net_device *dev)
1038{
1039	int err;
1040
1041	dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1042	if (!dev->lstats)
1043		return -ENOMEM;
1044
1045	err = veth_alloc_queues(dev);
1046	if (err) {
1047		free_percpu(dev->lstats);
1048		return err;
1049	}
1050
1051	return 0;
1052}
1053
1054static void veth_dev_free(struct net_device *dev)
1055{
1056	veth_free_queues(dev);
1057	free_percpu(dev->lstats);
1058}
1059
1060#ifdef CONFIG_NET_POLL_CONTROLLER
1061static void veth_poll_controller(struct net_device *dev)
1062{
1063	/* veth only receives frames when its peer sends one
1064	 * Since it has nothing to do with disabling irqs, we are guaranteed
1065	 * never to have pending data when we poll for it so
1066	 * there is nothing to do here.
1067	 *
1068	 * We need this though so netpoll recognizes us as an interface that
1069	 * supports polling, which enables bridge devices in virt setups to
1070	 * still use netconsole
1071	 */
1072}
1073#endif	/* CONFIG_NET_POLL_CONTROLLER */
1074
1075static int veth_get_iflink(const struct net_device *dev)
1076{
1077	struct veth_priv *priv = netdev_priv(dev);
1078	struct net_device *peer;
1079	int iflink;
1080
1081	rcu_read_lock();
1082	peer = rcu_dereference(priv->peer);
1083	iflink = peer ? peer->ifindex : 0;
1084	rcu_read_unlock();
1085
1086	return iflink;
1087}
1088
1089static netdev_features_t veth_fix_features(struct net_device *dev,
1090					   netdev_features_t features)
1091{
1092	struct veth_priv *priv = netdev_priv(dev);
1093	struct net_device *peer;
1094
1095	peer = rtnl_dereference(priv->peer);
1096	if (peer) {
1097		struct veth_priv *peer_priv = netdev_priv(peer);
1098
1099		if (peer_priv->_xdp_prog)
1100			features &= ~NETIF_F_GSO_SOFTWARE;
1101	}
1102
1103	return features;
1104}
1105
1106static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1107{
1108	struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1109	struct net_device *peer;
1110
1111	if (new_hr < 0)
1112		new_hr = 0;
1113
1114	rcu_read_lock();
1115	peer = rcu_dereference(priv->peer);
1116	if (unlikely(!peer))
1117		goto out;
1118
1119	peer_priv = netdev_priv(peer);
1120	priv->requested_headroom = new_hr;
1121	new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1122	dev->needed_headroom = new_hr;
1123	peer->needed_headroom = new_hr;
1124
1125out:
1126	rcu_read_unlock();
1127}
1128
1129static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1130			struct netlink_ext_ack *extack)
1131{
1132	struct veth_priv *priv = netdev_priv(dev);
1133	struct bpf_prog *old_prog;
1134	struct net_device *peer;
1135	unsigned int max_mtu;
1136	int err;
1137
1138	old_prog = priv->_xdp_prog;
1139	priv->_xdp_prog = prog;
1140	peer = rtnl_dereference(priv->peer);
1141
1142	if (prog) {
1143		if (!peer) {
1144			NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1145			err = -ENOTCONN;
1146			goto err;
1147		}
1148
1149		max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1150			  peer->hard_header_len -
1151			  SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1152		if (peer->mtu > max_mtu) {
1153			NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1154			err = -ERANGE;
1155			goto err;
1156		}
1157
1158		if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1159			NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1160			err = -ENOSPC;
1161			goto err;
1162		}
1163
1164		if (dev->flags & IFF_UP) {
1165			err = veth_enable_xdp(dev);
1166			if (err) {
1167				NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1168				goto err;
1169			}
1170		}
1171
1172		if (!old_prog) {
1173			peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1174			peer->max_mtu = max_mtu;
1175		}
1176	}
1177
1178	if (old_prog) {
1179		if (!prog) {
1180			if (dev->flags & IFF_UP)
1181				veth_disable_xdp(dev);
1182
1183			if (peer) {
1184				peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1185				peer->max_mtu = ETH_MAX_MTU;
1186			}
1187		}
1188		bpf_prog_put(old_prog);
1189	}
1190
1191	if ((!!old_prog ^ !!prog) && peer)
1192		netdev_update_features(peer);
1193
1194	return 0;
1195err:
1196	priv->_xdp_prog = old_prog;
1197
1198	return err;
1199}
1200
1201static u32 veth_xdp_query(struct net_device *dev)
1202{
1203	struct veth_priv *priv = netdev_priv(dev);
1204	const struct bpf_prog *xdp_prog;
1205
1206	xdp_prog = priv->_xdp_prog;
1207	if (xdp_prog)
1208		return xdp_prog->aux->id;
1209
1210	return 0;
1211}
1212
1213static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1214{
1215	switch (xdp->command) {
1216	case XDP_SETUP_PROG:
1217		return veth_xdp_set(dev, xdp->prog, xdp->extack);
1218	case XDP_QUERY_PROG:
1219		xdp->prog_id = veth_xdp_query(dev);
1220		return 0;
1221	default:
1222		return -EINVAL;
1223	}
1224}
1225
1226static const struct net_device_ops veth_netdev_ops = {
1227	.ndo_init            = veth_dev_init,
1228	.ndo_open            = veth_open,
1229	.ndo_stop            = veth_close,
1230	.ndo_start_xmit      = veth_xmit,
1231	.ndo_get_stats64     = veth_get_stats64,
1232	.ndo_set_rx_mode     = veth_set_multicast_list,
1233	.ndo_set_mac_address = eth_mac_addr,
1234#ifdef CONFIG_NET_POLL_CONTROLLER
1235	.ndo_poll_controller	= veth_poll_controller,
1236#endif
1237	.ndo_get_iflink		= veth_get_iflink,
1238	.ndo_fix_features	= veth_fix_features,
1239	.ndo_features_check	= passthru_features_check,
1240	.ndo_set_rx_headroom	= veth_set_rx_headroom,
1241	.ndo_bpf		= veth_xdp,
1242	.ndo_xdp_xmit		= veth_ndo_xdp_xmit,
1243};
1244
1245#define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1246		       NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1247		       NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1248		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1249		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1250
1251static void veth_setup(struct net_device *dev)
1252{
1253	ether_setup(dev);
1254
1255	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1256	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1257	dev->priv_flags |= IFF_NO_QUEUE;
1258	dev->priv_flags |= IFF_PHONY_HEADROOM;
1259
1260	dev->netdev_ops = &veth_netdev_ops;
1261	dev->ethtool_ops = &veth_ethtool_ops;
1262	dev->features |= NETIF_F_LLTX;
1263	dev->features |= VETH_FEATURES;
1264	dev->vlan_features = dev->features &
1265			     ~(NETIF_F_HW_VLAN_CTAG_TX |
1266			       NETIF_F_HW_VLAN_STAG_TX |
1267			       NETIF_F_HW_VLAN_CTAG_RX |
1268			       NETIF_F_HW_VLAN_STAG_RX);
1269	dev->needs_free_netdev = true;
1270	dev->priv_destructor = veth_dev_free;
1271	dev->max_mtu = ETH_MAX_MTU;
1272
1273	dev->hw_features = VETH_FEATURES;
1274	dev->hw_enc_features = VETH_FEATURES;
1275	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1276}
1277
1278/*
1279 * netlink interface
1280 */
1281
1282static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1283			 struct netlink_ext_ack *extack)
1284{
1285	if (tb[IFLA_ADDRESS]) {
1286		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1287			return -EINVAL;
1288		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1289			return -EADDRNOTAVAIL;
1290	}
1291	if (tb[IFLA_MTU]) {
1292		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1293			return -EINVAL;
1294	}
1295	return 0;
1296}
1297
1298static struct rtnl_link_ops veth_link_ops;
1299
1300static int veth_newlink(struct net *src_net, struct net_device *dev,
1301			struct nlattr *tb[], struct nlattr *data[],
1302			struct netlink_ext_ack *extack)
1303{
1304	int err;
1305	struct net_device *peer;
1306	struct veth_priv *priv;
1307	char ifname[IFNAMSIZ];
1308	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1309	unsigned char name_assign_type;
1310	struct ifinfomsg *ifmp;
1311	struct net *net;
1312
1313	/*
1314	 * create and register peer first
1315	 */
1316	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1317		struct nlattr *nla_peer;
1318
1319		nla_peer = data[VETH_INFO_PEER];
1320		ifmp = nla_data(nla_peer);
1321		err = rtnl_nla_parse_ifla(peer_tb,
1322					  nla_data(nla_peer) + sizeof(struct ifinfomsg),
1323					  nla_len(nla_peer) - sizeof(struct ifinfomsg),
1324					  NULL);
1325		if (err < 0)
1326			return err;
1327
1328		err = veth_validate(peer_tb, NULL, extack);
1329		if (err < 0)
1330			return err;
1331
1332		tbp = peer_tb;
1333	} else {
1334		ifmp = NULL;
1335		tbp = tb;
1336	}
1337
1338	if (ifmp && tbp[IFLA_IFNAME]) {
1339		nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1340		name_assign_type = NET_NAME_USER;
1341	} else {
1342		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1343		name_assign_type = NET_NAME_ENUM;
1344	}
1345
1346	net = rtnl_link_get_net(src_net, tbp);
1347	if (IS_ERR(net))
1348		return PTR_ERR(net);
1349
1350	peer = rtnl_create_link(net, ifname, name_assign_type,
1351				&veth_link_ops, tbp, extack);
1352	if (IS_ERR(peer)) {
1353		put_net(net);
1354		return PTR_ERR(peer);
1355	}
1356
1357	if (!ifmp || !tbp[IFLA_ADDRESS])
1358		eth_hw_addr_random(peer);
1359
1360	if (ifmp && (dev->ifindex != 0))
1361		peer->ifindex = ifmp->ifi_index;
1362
1363	peer->gso_max_size = dev->gso_max_size;
1364	peer->gso_max_segs = dev->gso_max_segs;
1365
1366	err = register_netdevice(peer);
1367	put_net(net);
1368	net = NULL;
1369	if (err < 0)
1370		goto err_register_peer;
1371
1372	netif_carrier_off(peer);
1373
1374	err = rtnl_configure_link(peer, ifmp);
1375	if (err < 0)
1376		goto err_configure_peer;
1377
1378	/*
1379	 * register dev last
1380	 *
1381	 * note, that since we've registered new device the dev's name
1382	 * should be re-allocated
1383	 */
1384
1385	if (tb[IFLA_ADDRESS] == NULL)
1386		eth_hw_addr_random(dev);
1387
1388	if (tb[IFLA_IFNAME])
1389		nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1390	else
1391		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1392
1393	err = register_netdevice(dev);
1394	if (err < 0)
1395		goto err_register_dev;
1396
1397	netif_carrier_off(dev);
1398
1399	/*
1400	 * tie the deviced together
1401	 */
1402
1403	priv = netdev_priv(dev);
1404	rcu_assign_pointer(priv->peer, peer);
1405
1406	priv = netdev_priv(peer);
1407	rcu_assign_pointer(priv->peer, dev);
1408
1409	return 0;
1410
1411err_register_dev:
1412	/* nothing to do */
1413err_configure_peer:
1414	unregister_netdevice(peer);
1415	return err;
1416
1417err_register_peer:
1418	free_netdev(peer);
1419	return err;
1420}
1421
1422static void veth_dellink(struct net_device *dev, struct list_head *head)
1423{
1424	struct veth_priv *priv;
1425	struct net_device *peer;
1426
1427	priv = netdev_priv(dev);
1428	peer = rtnl_dereference(priv->peer);
1429
1430	/* Note : dellink() is called from default_device_exit_batch(),
1431	 * before a rcu_synchronize() point. The devices are guaranteed
1432	 * not being freed before one RCU grace period.
1433	 */
1434	RCU_INIT_POINTER(priv->peer, NULL);
1435	unregister_netdevice_queue(dev, head);
1436
1437	if (peer) {
1438		priv = netdev_priv(peer);
1439		RCU_INIT_POINTER(priv->peer, NULL);
1440		unregister_netdevice_queue(peer, head);
1441	}
1442}
1443
1444static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1445	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
1446};
1447
1448static struct net *veth_get_link_net(const struct net_device *dev)
1449{
1450	struct veth_priv *priv = netdev_priv(dev);
1451	struct net_device *peer = rtnl_dereference(priv->peer);
1452
1453	return peer ? dev_net(peer) : dev_net(dev);
1454}
1455
1456static struct rtnl_link_ops veth_link_ops = {
1457	.kind		= DRV_NAME,
1458	.priv_size	= sizeof(struct veth_priv),
1459	.setup		= veth_setup,
1460	.validate	= veth_validate,
1461	.newlink	= veth_newlink,
1462	.dellink	= veth_dellink,
1463	.policy		= veth_policy,
1464	.maxtype	= VETH_INFO_MAX,
1465	.get_link_net	= veth_get_link_net,
1466};
1467
1468/*
1469 * init/fini
1470 */
1471
1472static __init int veth_init(void)
1473{
1474	return rtnl_link_register(&veth_link_ops);
1475}
1476
1477static __exit void veth_exit(void)
1478{
1479	rtnl_link_unregister(&veth_link_ops);
1480}
1481
1482module_init(veth_init);
1483module_exit(veth_exit);
1484
1485MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1486MODULE_LICENSE("GPL v2");
1487MODULE_ALIAS_RTNL_LINK(DRV_NAME);