xref: /openbmc/linux/drivers/net/veth.c (revision 6c9111bc)
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 
39 struct 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 
52 struct veth_rq_stats {
53 	struct veth_stats	vs;
54 	struct u64_stats_sync	syncp;
55 };
56 
57 struct 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 
68 struct 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 
76 struct 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 
85 struct 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 
92 static 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 
104 static 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 
111 static struct {
112 	const char string[ETH_GSTRING_LEN];
113 } ethtool_stats_keys[] = {
114 	{ "peer_ifindex" },
115 };
116 
117 static 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 
127 static 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 
133 static 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 
162 static int veth_get_sset_count(struct net_device *dev, int sset)
163 {
164 	switch (sset) {
165 	case ETH_SS_STATS:
166 		return ARRAY_SIZE(ethtool_stats_keys) +
167 		       VETH_RQ_STATS_LEN * dev->real_num_rx_queues +
168 		       VETH_TQ_STATS_LEN * dev->real_num_tx_queues;
169 	default:
170 		return -EOPNOTSUPP;
171 	}
172 }
173 
174 static void veth_get_ethtool_stats(struct net_device *dev,
175 		struct ethtool_stats *stats, u64 *data)
176 {
177 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
178 	struct net_device *peer = rtnl_dereference(priv->peer);
179 	int i, j, idx;
180 
181 	data[0] = peer ? peer->ifindex : 0;
182 	idx = 1;
183 	for (i = 0; i < dev->real_num_rx_queues; i++) {
184 		const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
185 		const void *stats_base = (void *)&rq_stats->vs;
186 		unsigned int start;
187 		size_t offset;
188 
189 		do {
190 			start = u64_stats_fetch_begin_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 
220 static 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 
232 static bool veth_is_xdp_frame(void *ptr)
233 {
234 	return (unsigned long)ptr & VETH_XDP_FLAG;
235 }
236 
237 static struct xdp_frame *veth_ptr_to_xdp(void *ptr)
238 {
239 	return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
240 }
241 
242 static void *veth_xdp_to_ptr(struct xdp_frame *xdp)
243 {
244 	return (void *)((unsigned long)xdp | VETH_XDP_FLAG);
245 }
246 
247 static 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 
255 static 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 
265 static 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 
275 static 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 
283 static 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 {
313 drop:
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 
325 static 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 
333 static 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 
364 static 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 */
399 static void veth_set_multicast_list(struct net_device *dev)
400 {
401 }
402 
403 static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
404 				      int buflen)
405 {
406 	struct sk_buff *skb;
407 
408 	skb = build_skb(head, buflen);
409 	if (!skb)
410 		return NULL;
411 
412 	skb_reserve(skb, headroom);
413 	skb_put(skb, len);
414 
415 	return skb;
416 }
417 
418 static int veth_select_rxq(struct net_device *dev)
419 {
420 	return smp_processor_id() % dev->real_num_rx_queues;
421 }
422 
423 static struct net_device *veth_peer_dev(struct net_device *dev)
424 {
425 	struct veth_priv *priv = netdev_priv(dev);
426 
427 	/* Callers must be under RCU read side. */
428 	return rcu_dereference(priv->peer);
429 }
430 
431 static int veth_xdp_xmit(struct net_device *dev, int n,
432 			 struct xdp_frame **frames,
433 			 u32 flags, bool ndo_xmit)
434 {
435 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
436 	int i, ret = -ENXIO, drops = 0;
437 	struct net_device *rcv;
438 	unsigned int max_len;
439 	struct veth_rq *rq;
440 
441 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
442 		return -EINVAL;
443 
444 	rcu_read_lock();
445 	rcv = rcu_dereference(priv->peer);
446 	if (unlikely(!rcv))
447 		goto out;
448 
449 	rcv_priv = netdev_priv(rcv);
450 	rq = &rcv_priv->rq[veth_select_rxq(rcv)];
451 	/* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
452 	 * side. This means an XDP program is loaded on the peer and the peer
453 	 * device is up.
454 	 */
455 	if (!rcu_access_pointer(rq->xdp_prog))
456 		goto out;
457 
458 	max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
459 
460 	spin_lock(&rq->xdp_ring.producer_lock);
461 	for (i = 0; i < n; i++) {
462 		struct xdp_frame *frame = frames[i];
463 		void *ptr = veth_xdp_to_ptr(frame);
464 
465 		if (unlikely(frame->len > max_len ||
466 			     __ptr_ring_produce(&rq->xdp_ring, ptr))) {
467 			xdp_return_frame_rx_napi(frame);
468 			drops++;
469 		}
470 	}
471 	spin_unlock(&rq->xdp_ring.producer_lock);
472 
473 	if (flags & XDP_XMIT_FLUSH)
474 		__veth_xdp_flush(rq);
475 
476 	ret = n - drops;
477 	if (ndo_xmit) {
478 		u64_stats_update_begin(&rq->stats.syncp);
479 		rq->stats.vs.peer_tq_xdp_xmit += n - drops;
480 		rq->stats.vs.peer_tq_xdp_xmit_err += drops;
481 		u64_stats_update_end(&rq->stats.syncp);
482 	}
483 
484 out:
485 	rcu_read_unlock();
486 
487 	return ret;
488 }
489 
490 static int veth_ndo_xdp_xmit(struct net_device *dev, int n,
491 			     struct xdp_frame **frames, u32 flags)
492 {
493 	int err;
494 
495 	err = veth_xdp_xmit(dev, n, frames, flags, true);
496 	if (err < 0) {
497 		struct veth_priv *priv = netdev_priv(dev);
498 
499 		atomic64_add(n, &priv->dropped);
500 	}
501 
502 	return err;
503 }
504 
505 static void veth_xdp_flush_bq(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
506 {
507 	int sent, i, err = 0;
508 
509 	sent = veth_xdp_xmit(rq->dev, bq->count, bq->q, 0, false);
510 	if (sent < 0) {
511 		err = sent;
512 		sent = 0;
513 		for (i = 0; i < bq->count; i++)
514 			xdp_return_frame(bq->q[i]);
515 	}
516 	trace_xdp_bulk_tx(rq->dev, sent, bq->count - sent, err);
517 
518 	u64_stats_update_begin(&rq->stats.syncp);
519 	rq->stats.vs.xdp_tx += sent;
520 	rq->stats.vs.xdp_tx_err += bq->count - sent;
521 	u64_stats_update_end(&rq->stats.syncp);
522 
523 	bq->count = 0;
524 }
525 
526 static void veth_xdp_flush(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
527 {
528 	struct veth_priv *rcv_priv, *priv = netdev_priv(rq->dev);
529 	struct net_device *rcv;
530 	struct veth_rq *rcv_rq;
531 
532 	rcu_read_lock();
533 	veth_xdp_flush_bq(rq, bq);
534 	rcv = rcu_dereference(priv->peer);
535 	if (unlikely(!rcv))
536 		goto out;
537 
538 	rcv_priv = netdev_priv(rcv);
539 	rcv_rq = &rcv_priv->rq[veth_select_rxq(rcv)];
540 	/* xdp_ring is initialized on receive side? */
541 	if (unlikely(!rcu_access_pointer(rcv_rq->xdp_prog)))
542 		goto out;
543 
544 	__veth_xdp_flush(rcv_rq);
545 out:
546 	rcu_read_unlock();
547 }
548 
549 static int veth_xdp_tx(struct veth_rq *rq, struct xdp_buff *xdp,
550 		       struct veth_xdp_tx_bq *bq)
551 {
552 	struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
553 
554 	if (unlikely(!frame))
555 		return -EOVERFLOW;
556 
557 	if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
558 		veth_xdp_flush_bq(rq, bq);
559 
560 	bq->q[bq->count++] = frame;
561 
562 	return 0;
563 }
564 
565 static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
566 					struct xdp_frame *frame,
567 					struct veth_xdp_tx_bq *bq,
568 					struct veth_stats *stats)
569 {
570 	void *hard_start = frame->data - frame->headroom;
571 	int len = frame->len, delta = 0;
572 	struct xdp_frame orig_frame;
573 	struct bpf_prog *xdp_prog;
574 	unsigned int headroom;
575 	struct sk_buff *skb;
576 
577 	/* bpf_xdp_adjust_head() assures BPF cannot access xdp_frame area */
578 	hard_start -= sizeof(struct xdp_frame);
579 
580 	rcu_read_lock();
581 	xdp_prog = rcu_dereference(rq->xdp_prog);
582 	if (likely(xdp_prog)) {
583 		struct xdp_buff xdp;
584 		u32 act;
585 
586 		xdp_convert_frame_to_buff(frame, &xdp);
587 		xdp.rxq = &rq->xdp_rxq;
588 
589 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
590 
591 		switch (act) {
592 		case XDP_PASS:
593 			delta = frame->data - xdp.data;
594 			len = xdp.data_end - xdp.data;
595 			break;
596 		case XDP_TX:
597 			orig_frame = *frame;
598 			xdp.rxq->mem = frame->mem;
599 			if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
600 				trace_xdp_exception(rq->dev, xdp_prog, act);
601 				frame = &orig_frame;
602 				stats->rx_drops++;
603 				goto err_xdp;
604 			}
605 			stats->xdp_tx++;
606 			rcu_read_unlock();
607 			goto xdp_xmit;
608 		case XDP_REDIRECT:
609 			orig_frame = *frame;
610 			xdp.rxq->mem = frame->mem;
611 			if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
612 				frame = &orig_frame;
613 				stats->rx_drops++;
614 				goto err_xdp;
615 			}
616 			stats->xdp_redirect++;
617 			rcu_read_unlock();
618 			goto xdp_xmit;
619 		default:
620 			bpf_warn_invalid_xdp_action(act);
621 			fallthrough;
622 		case XDP_ABORTED:
623 			trace_xdp_exception(rq->dev, xdp_prog, act);
624 			fallthrough;
625 		case XDP_DROP:
626 			stats->xdp_drops++;
627 			goto err_xdp;
628 		}
629 	}
630 	rcu_read_unlock();
631 
632 	headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
633 	skb = veth_build_skb(hard_start, headroom, len, frame->frame_sz);
634 	if (!skb) {
635 		xdp_return_frame(frame);
636 		stats->rx_drops++;
637 		goto err;
638 	}
639 
640 	xdp_release_frame(frame);
641 	xdp_scrub_frame(frame);
642 	skb->protocol = eth_type_trans(skb, rq->dev);
643 err:
644 	return skb;
645 err_xdp:
646 	rcu_read_unlock();
647 	xdp_return_frame(frame);
648 xdp_xmit:
649 	return NULL;
650 }
651 
652 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
653 					struct sk_buff *skb,
654 					struct veth_xdp_tx_bq *bq,
655 					struct veth_stats *stats)
656 {
657 	u32 pktlen, headroom, act, metalen;
658 	void *orig_data, *orig_data_end;
659 	struct bpf_prog *xdp_prog;
660 	int mac_len, delta, off;
661 	struct xdp_buff xdp;
662 
663 	skb_orphan(skb);
664 
665 	rcu_read_lock();
666 	xdp_prog = rcu_dereference(rq->xdp_prog);
667 	if (unlikely(!xdp_prog)) {
668 		rcu_read_unlock();
669 		goto out;
670 	}
671 
672 	mac_len = skb->data - skb_mac_header(skb);
673 	pktlen = skb->len + mac_len;
674 	headroom = skb_headroom(skb) - mac_len;
675 
676 	if (skb_shared(skb) || skb_head_is_locked(skb) ||
677 	    skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
678 		struct sk_buff *nskb;
679 		int size, head_off;
680 		void *head, *start;
681 		struct page *page;
682 
683 		size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
684 		       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
685 		if (size > PAGE_SIZE)
686 			goto drop;
687 
688 		page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
689 		if (!page)
690 			goto drop;
691 
692 		head = page_address(page);
693 		start = head + VETH_XDP_HEADROOM;
694 		if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
695 			page_frag_free(head);
696 			goto drop;
697 		}
698 
699 		nskb = veth_build_skb(head, VETH_XDP_HEADROOM + mac_len,
700 				      skb->len, 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 
719 	/* SKB "head" area always have tailroom for skb_shared_info */
720 	xdp.frame_sz = (void *)skb_end_pointer(skb) - xdp.data_hard_start;
721 	xdp.frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
722 
723 	orig_data = xdp.data;
724 	orig_data_end = xdp.data_end;
725 
726 	act = bpf_prog_run_xdp(xdp_prog, &xdp);
727 
728 	switch (act) {
729 	case XDP_PASS:
730 		break;
731 	case XDP_TX:
732 		get_page(virt_to_page(xdp.data));
733 		consume_skb(skb);
734 		xdp.rxq->mem = rq->xdp_mem;
735 		if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
736 			trace_xdp_exception(rq->dev, xdp_prog, act);
737 			stats->rx_drops++;
738 			goto err_xdp;
739 		}
740 		stats->xdp_tx++;
741 		rcu_read_unlock();
742 		goto xdp_xmit;
743 	case XDP_REDIRECT:
744 		get_page(virt_to_page(xdp.data));
745 		consume_skb(skb);
746 		xdp.rxq->mem = rq->xdp_mem;
747 		if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
748 			stats->rx_drops++;
749 			goto err_xdp;
750 		}
751 		stats->xdp_redirect++;
752 		rcu_read_unlock();
753 		goto xdp_xmit;
754 	default:
755 		bpf_warn_invalid_xdp_action(act);
756 		fallthrough;
757 	case XDP_ABORTED:
758 		trace_xdp_exception(rq->dev, xdp_prog, act);
759 		fallthrough;
760 	case XDP_DROP:
761 		stats->xdp_drops++;
762 		goto xdp_drop;
763 	}
764 	rcu_read_unlock();
765 
766 	/* check if bpf_xdp_adjust_head was used */
767 	delta = orig_data - xdp.data;
768 	off = mac_len + delta;
769 	if (off > 0)
770 		__skb_push(skb, off);
771 	else if (off < 0)
772 		__skb_pull(skb, -off);
773 	skb->mac_header -= delta;
774 
775 	/* check if bpf_xdp_adjust_tail was used */
776 	off = xdp.data_end - orig_data_end;
777 	if (off != 0)
778 		__skb_put(skb, off); /* positive on grow, negative on shrink */
779 	skb->protocol = eth_type_trans(skb, rq->dev);
780 
781 	metalen = xdp.data - xdp.data_meta;
782 	if (metalen)
783 		skb_metadata_set(skb, metalen);
784 out:
785 	return skb;
786 drop:
787 	stats->rx_drops++;
788 xdp_drop:
789 	rcu_read_unlock();
790 	kfree_skb(skb);
791 	return NULL;
792 err_xdp:
793 	rcu_read_unlock();
794 	page_frag_free(xdp.data);
795 xdp_xmit:
796 	return NULL;
797 }
798 
799 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
800 			struct veth_xdp_tx_bq *bq,
801 			struct veth_stats *stats)
802 {
803 	int i, done = 0;
804 
805 	for (i = 0; i < budget; i++) {
806 		void *ptr = __ptr_ring_consume(&rq->xdp_ring);
807 		struct sk_buff *skb;
808 
809 		if (!ptr)
810 			break;
811 
812 		if (veth_is_xdp_frame(ptr)) {
813 			struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
814 
815 			stats->xdp_bytes += frame->len;
816 			skb = veth_xdp_rcv_one(rq, frame, bq, stats);
817 		} else {
818 			skb = ptr;
819 			stats->xdp_bytes += skb->len;
820 			skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
821 		}
822 
823 		if (skb)
824 			napi_gro_receive(&rq->xdp_napi, skb);
825 
826 		done++;
827 	}
828 
829 	u64_stats_update_begin(&rq->stats.syncp);
830 	rq->stats.vs.xdp_redirect += stats->xdp_redirect;
831 	rq->stats.vs.xdp_bytes += stats->xdp_bytes;
832 	rq->stats.vs.xdp_drops += stats->xdp_drops;
833 	rq->stats.vs.rx_drops += stats->rx_drops;
834 	rq->stats.vs.xdp_packets += done;
835 	u64_stats_update_end(&rq->stats.syncp);
836 
837 	return done;
838 }
839 
840 static int veth_poll(struct napi_struct *napi, int budget)
841 {
842 	struct veth_rq *rq =
843 		container_of(napi, struct veth_rq, xdp_napi);
844 	struct veth_stats stats = {};
845 	struct veth_xdp_tx_bq bq;
846 	int done;
847 
848 	bq.count = 0;
849 
850 	xdp_set_return_frame_no_direct();
851 	done = veth_xdp_rcv(rq, budget, &bq, &stats);
852 
853 	if (done < budget && napi_complete_done(napi, done)) {
854 		/* Write rx_notify_masked before reading ptr_ring */
855 		smp_store_mb(rq->rx_notify_masked, false);
856 		if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
857 			rq->rx_notify_masked = true;
858 			napi_schedule(&rq->xdp_napi);
859 		}
860 	}
861 
862 	if (stats.xdp_tx > 0)
863 		veth_xdp_flush(rq, &bq);
864 	if (stats.xdp_redirect > 0)
865 		xdp_do_flush();
866 	xdp_clear_return_frame_no_direct();
867 
868 	return done;
869 }
870 
871 static int veth_napi_add(struct net_device *dev)
872 {
873 	struct veth_priv *priv = netdev_priv(dev);
874 	int err, i;
875 
876 	for (i = 0; i < dev->real_num_rx_queues; i++) {
877 		struct veth_rq *rq = &priv->rq[i];
878 
879 		err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
880 		if (err)
881 			goto err_xdp_ring;
882 	}
883 
884 	for (i = 0; i < dev->real_num_rx_queues; i++) {
885 		struct veth_rq *rq = &priv->rq[i];
886 
887 		netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
888 		napi_enable(&rq->xdp_napi);
889 	}
890 
891 	return 0;
892 err_xdp_ring:
893 	for (i--; i >= 0; i--)
894 		ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
895 
896 	return err;
897 }
898 
899 static void veth_napi_del(struct net_device *dev)
900 {
901 	struct veth_priv *priv = netdev_priv(dev);
902 	int i;
903 
904 	for (i = 0; i < dev->real_num_rx_queues; i++) {
905 		struct veth_rq *rq = &priv->rq[i];
906 
907 		napi_disable(&rq->xdp_napi);
908 		__netif_napi_del(&rq->xdp_napi);
909 	}
910 	synchronize_net();
911 
912 	for (i = 0; i < dev->real_num_rx_queues; i++) {
913 		struct veth_rq *rq = &priv->rq[i];
914 
915 		rq->rx_notify_masked = false;
916 		ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
917 	}
918 }
919 
920 static int veth_enable_xdp(struct net_device *dev)
921 {
922 	struct veth_priv *priv = netdev_priv(dev);
923 	int err, i;
924 
925 	if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
926 		for (i = 0; i < dev->real_num_rx_queues; i++) {
927 			struct veth_rq *rq = &priv->rq[i];
928 
929 			err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
930 			if (err < 0)
931 				goto err_rxq_reg;
932 
933 			err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
934 							 MEM_TYPE_PAGE_SHARED,
935 							 NULL);
936 			if (err < 0)
937 				goto err_reg_mem;
938 
939 			/* Save original mem info as it can be overwritten */
940 			rq->xdp_mem = rq->xdp_rxq.mem;
941 		}
942 
943 		err = veth_napi_add(dev);
944 		if (err)
945 			goto err_rxq_reg;
946 	}
947 
948 	for (i = 0; i < dev->real_num_rx_queues; i++)
949 		rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
950 
951 	return 0;
952 err_reg_mem:
953 	xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
954 err_rxq_reg:
955 	for (i--; i >= 0; i--)
956 		xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
957 
958 	return err;
959 }
960 
961 static void veth_disable_xdp(struct net_device *dev)
962 {
963 	struct veth_priv *priv = netdev_priv(dev);
964 	int i;
965 
966 	for (i = 0; i < dev->real_num_rx_queues; i++)
967 		rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
968 	veth_napi_del(dev);
969 	for (i = 0; i < dev->real_num_rx_queues; i++) {
970 		struct veth_rq *rq = &priv->rq[i];
971 
972 		rq->xdp_rxq.mem = rq->xdp_mem;
973 		xdp_rxq_info_unreg(&rq->xdp_rxq);
974 	}
975 }
976 
977 static int veth_open(struct net_device *dev)
978 {
979 	struct veth_priv *priv = netdev_priv(dev);
980 	struct net_device *peer = rtnl_dereference(priv->peer);
981 	int err;
982 
983 	if (!peer)
984 		return -ENOTCONN;
985 
986 	if (priv->_xdp_prog) {
987 		err = veth_enable_xdp(dev);
988 		if (err)
989 			return err;
990 	}
991 
992 	if (peer->flags & IFF_UP) {
993 		netif_carrier_on(dev);
994 		netif_carrier_on(peer);
995 	}
996 
997 	return 0;
998 }
999 
1000 static int veth_close(struct net_device *dev)
1001 {
1002 	struct veth_priv *priv = netdev_priv(dev);
1003 	struct net_device *peer = rtnl_dereference(priv->peer);
1004 
1005 	netif_carrier_off(dev);
1006 	if (peer)
1007 		netif_carrier_off(peer);
1008 
1009 	if (priv->_xdp_prog)
1010 		veth_disable_xdp(dev);
1011 
1012 	return 0;
1013 }
1014 
1015 static int is_valid_veth_mtu(int mtu)
1016 {
1017 	return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1018 }
1019 
1020 static int veth_alloc_queues(struct net_device *dev)
1021 {
1022 	struct veth_priv *priv = netdev_priv(dev);
1023 	int i;
1024 
1025 	priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
1026 	if (!priv->rq)
1027 		return -ENOMEM;
1028 
1029 	for (i = 0; i < dev->num_rx_queues; i++) {
1030 		priv->rq[i].dev = dev;
1031 		u64_stats_init(&priv->rq[i].stats.syncp);
1032 	}
1033 
1034 	return 0;
1035 }
1036 
1037 static void veth_free_queues(struct net_device *dev)
1038 {
1039 	struct veth_priv *priv = netdev_priv(dev);
1040 
1041 	kfree(priv->rq);
1042 }
1043 
1044 static int veth_dev_init(struct net_device *dev)
1045 {
1046 	int err;
1047 
1048 	dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1049 	if (!dev->lstats)
1050 		return -ENOMEM;
1051 
1052 	err = veth_alloc_queues(dev);
1053 	if (err) {
1054 		free_percpu(dev->lstats);
1055 		return err;
1056 	}
1057 
1058 	return 0;
1059 }
1060 
1061 static void veth_dev_free(struct net_device *dev)
1062 {
1063 	veth_free_queues(dev);
1064 	free_percpu(dev->lstats);
1065 }
1066 
1067 #ifdef CONFIG_NET_POLL_CONTROLLER
1068 static void veth_poll_controller(struct net_device *dev)
1069 {
1070 	/* veth only receives frames when its peer sends one
1071 	 * Since it has nothing to do with disabling irqs, we are guaranteed
1072 	 * never to have pending data when we poll for it so
1073 	 * there is nothing to do here.
1074 	 *
1075 	 * We need this though so netpoll recognizes us as an interface that
1076 	 * supports polling, which enables bridge devices in virt setups to
1077 	 * still use netconsole
1078 	 */
1079 }
1080 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1081 
1082 static int veth_get_iflink(const struct net_device *dev)
1083 {
1084 	struct veth_priv *priv = netdev_priv(dev);
1085 	struct net_device *peer;
1086 	int iflink;
1087 
1088 	rcu_read_lock();
1089 	peer = rcu_dereference(priv->peer);
1090 	iflink = peer ? peer->ifindex : 0;
1091 	rcu_read_unlock();
1092 
1093 	return iflink;
1094 }
1095 
1096 static netdev_features_t veth_fix_features(struct net_device *dev,
1097 					   netdev_features_t features)
1098 {
1099 	struct veth_priv *priv = netdev_priv(dev);
1100 	struct net_device *peer;
1101 
1102 	peer = rtnl_dereference(priv->peer);
1103 	if (peer) {
1104 		struct veth_priv *peer_priv = netdev_priv(peer);
1105 
1106 		if (peer_priv->_xdp_prog)
1107 			features &= ~NETIF_F_GSO_SOFTWARE;
1108 	}
1109 
1110 	return features;
1111 }
1112 
1113 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1114 {
1115 	struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1116 	struct net_device *peer;
1117 
1118 	if (new_hr < 0)
1119 		new_hr = 0;
1120 
1121 	rcu_read_lock();
1122 	peer = rcu_dereference(priv->peer);
1123 	if (unlikely(!peer))
1124 		goto out;
1125 
1126 	peer_priv = netdev_priv(peer);
1127 	priv->requested_headroom = new_hr;
1128 	new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1129 	dev->needed_headroom = new_hr;
1130 	peer->needed_headroom = new_hr;
1131 
1132 out:
1133 	rcu_read_unlock();
1134 }
1135 
1136 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1137 			struct netlink_ext_ack *extack)
1138 {
1139 	struct veth_priv *priv = netdev_priv(dev);
1140 	struct bpf_prog *old_prog;
1141 	struct net_device *peer;
1142 	unsigned int max_mtu;
1143 	int err;
1144 
1145 	old_prog = priv->_xdp_prog;
1146 	priv->_xdp_prog = prog;
1147 	peer = rtnl_dereference(priv->peer);
1148 
1149 	if (prog) {
1150 		if (!peer) {
1151 			NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1152 			err = -ENOTCONN;
1153 			goto err;
1154 		}
1155 
1156 		max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1157 			  peer->hard_header_len -
1158 			  SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1159 		if (peer->mtu > max_mtu) {
1160 			NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1161 			err = -ERANGE;
1162 			goto err;
1163 		}
1164 
1165 		if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1166 			NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1167 			err = -ENOSPC;
1168 			goto err;
1169 		}
1170 
1171 		if (dev->flags & IFF_UP) {
1172 			err = veth_enable_xdp(dev);
1173 			if (err) {
1174 				NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1175 				goto err;
1176 			}
1177 		}
1178 
1179 		if (!old_prog) {
1180 			peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1181 			peer->max_mtu = max_mtu;
1182 		}
1183 	}
1184 
1185 	if (old_prog) {
1186 		if (!prog) {
1187 			if (dev->flags & IFF_UP)
1188 				veth_disable_xdp(dev);
1189 
1190 			if (peer) {
1191 				peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1192 				peer->max_mtu = ETH_MAX_MTU;
1193 			}
1194 		}
1195 		bpf_prog_put(old_prog);
1196 	}
1197 
1198 	if ((!!old_prog ^ !!prog) && peer)
1199 		netdev_update_features(peer);
1200 
1201 	return 0;
1202 err:
1203 	priv->_xdp_prog = old_prog;
1204 
1205 	return err;
1206 }
1207 
1208 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1209 {
1210 	switch (xdp->command) {
1211 	case XDP_SETUP_PROG:
1212 		return veth_xdp_set(dev, xdp->prog, xdp->extack);
1213 	default:
1214 		return -EINVAL;
1215 	}
1216 }
1217 
1218 static const struct net_device_ops veth_netdev_ops = {
1219 	.ndo_init            = veth_dev_init,
1220 	.ndo_open            = veth_open,
1221 	.ndo_stop            = veth_close,
1222 	.ndo_start_xmit      = veth_xmit,
1223 	.ndo_get_stats64     = veth_get_stats64,
1224 	.ndo_set_rx_mode     = veth_set_multicast_list,
1225 	.ndo_set_mac_address = eth_mac_addr,
1226 #ifdef CONFIG_NET_POLL_CONTROLLER
1227 	.ndo_poll_controller	= veth_poll_controller,
1228 #endif
1229 	.ndo_get_iflink		= veth_get_iflink,
1230 	.ndo_fix_features	= veth_fix_features,
1231 	.ndo_features_check	= passthru_features_check,
1232 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
1233 	.ndo_bpf		= veth_xdp,
1234 	.ndo_xdp_xmit		= veth_ndo_xdp_xmit,
1235 	.ndo_get_peer_dev	= veth_peer_dev,
1236 };
1237 
1238 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1239 		       NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1240 		       NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1241 		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1242 		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1243 
1244 static void veth_setup(struct net_device *dev)
1245 {
1246 	ether_setup(dev);
1247 
1248 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1249 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1250 	dev->priv_flags |= IFF_NO_QUEUE;
1251 	dev->priv_flags |= IFF_PHONY_HEADROOM;
1252 
1253 	dev->netdev_ops = &veth_netdev_ops;
1254 	dev->ethtool_ops = &veth_ethtool_ops;
1255 	dev->features |= NETIF_F_LLTX;
1256 	dev->features |= VETH_FEATURES;
1257 	dev->vlan_features = dev->features &
1258 			     ~(NETIF_F_HW_VLAN_CTAG_TX |
1259 			       NETIF_F_HW_VLAN_STAG_TX |
1260 			       NETIF_F_HW_VLAN_CTAG_RX |
1261 			       NETIF_F_HW_VLAN_STAG_RX);
1262 	dev->needs_free_netdev = true;
1263 	dev->priv_destructor = veth_dev_free;
1264 	dev->max_mtu = ETH_MAX_MTU;
1265 
1266 	dev->hw_features = VETH_FEATURES;
1267 	dev->hw_enc_features = VETH_FEATURES;
1268 	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1269 }
1270 
1271 /*
1272  * netlink interface
1273  */
1274 
1275 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1276 			 struct netlink_ext_ack *extack)
1277 {
1278 	if (tb[IFLA_ADDRESS]) {
1279 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1280 			return -EINVAL;
1281 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1282 			return -EADDRNOTAVAIL;
1283 	}
1284 	if (tb[IFLA_MTU]) {
1285 		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1286 			return -EINVAL;
1287 	}
1288 	return 0;
1289 }
1290 
1291 static struct rtnl_link_ops veth_link_ops;
1292 
1293 static int veth_newlink(struct net *src_net, struct net_device *dev,
1294 			struct nlattr *tb[], struct nlattr *data[],
1295 			struct netlink_ext_ack *extack)
1296 {
1297 	int err;
1298 	struct net_device *peer;
1299 	struct veth_priv *priv;
1300 	char ifname[IFNAMSIZ];
1301 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1302 	unsigned char name_assign_type;
1303 	struct ifinfomsg *ifmp;
1304 	struct net *net;
1305 
1306 	/*
1307 	 * create and register peer first
1308 	 */
1309 	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1310 		struct nlattr *nla_peer;
1311 
1312 		nla_peer = data[VETH_INFO_PEER];
1313 		ifmp = nla_data(nla_peer);
1314 		err = rtnl_nla_parse_ifla(peer_tb,
1315 					  nla_data(nla_peer) + sizeof(struct ifinfomsg),
1316 					  nla_len(nla_peer) - sizeof(struct ifinfomsg),
1317 					  NULL);
1318 		if (err < 0)
1319 			return err;
1320 
1321 		err = veth_validate(peer_tb, NULL, extack);
1322 		if (err < 0)
1323 			return err;
1324 
1325 		tbp = peer_tb;
1326 	} else {
1327 		ifmp = NULL;
1328 		tbp = tb;
1329 	}
1330 
1331 	if (ifmp && tbp[IFLA_IFNAME]) {
1332 		nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1333 		name_assign_type = NET_NAME_USER;
1334 	} else {
1335 		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1336 		name_assign_type = NET_NAME_ENUM;
1337 	}
1338 
1339 	net = rtnl_link_get_net(src_net, tbp);
1340 	if (IS_ERR(net))
1341 		return PTR_ERR(net);
1342 
1343 	peer = rtnl_create_link(net, ifname, name_assign_type,
1344 				&veth_link_ops, tbp, extack);
1345 	if (IS_ERR(peer)) {
1346 		put_net(net);
1347 		return PTR_ERR(peer);
1348 	}
1349 
1350 	if (!ifmp || !tbp[IFLA_ADDRESS])
1351 		eth_hw_addr_random(peer);
1352 
1353 	if (ifmp && (dev->ifindex != 0))
1354 		peer->ifindex = ifmp->ifi_index;
1355 
1356 	peer->gso_max_size = dev->gso_max_size;
1357 	peer->gso_max_segs = dev->gso_max_segs;
1358 
1359 	err = register_netdevice(peer);
1360 	put_net(net);
1361 	net = NULL;
1362 	if (err < 0)
1363 		goto err_register_peer;
1364 
1365 	netif_carrier_off(peer);
1366 
1367 	err = rtnl_configure_link(peer, ifmp);
1368 	if (err < 0)
1369 		goto err_configure_peer;
1370 
1371 	/*
1372 	 * register dev last
1373 	 *
1374 	 * note, that since we've registered new device the dev's name
1375 	 * should be re-allocated
1376 	 */
1377 
1378 	if (tb[IFLA_ADDRESS] == NULL)
1379 		eth_hw_addr_random(dev);
1380 
1381 	if (tb[IFLA_IFNAME])
1382 		nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1383 	else
1384 		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1385 
1386 	err = register_netdevice(dev);
1387 	if (err < 0)
1388 		goto err_register_dev;
1389 
1390 	netif_carrier_off(dev);
1391 
1392 	/*
1393 	 * tie the deviced together
1394 	 */
1395 
1396 	priv = netdev_priv(dev);
1397 	rcu_assign_pointer(priv->peer, peer);
1398 
1399 	priv = netdev_priv(peer);
1400 	rcu_assign_pointer(priv->peer, dev);
1401 
1402 	return 0;
1403 
1404 err_register_dev:
1405 	/* nothing to do */
1406 err_configure_peer:
1407 	unregister_netdevice(peer);
1408 	return err;
1409 
1410 err_register_peer:
1411 	free_netdev(peer);
1412 	return err;
1413 }
1414 
1415 static void veth_dellink(struct net_device *dev, struct list_head *head)
1416 {
1417 	struct veth_priv *priv;
1418 	struct net_device *peer;
1419 
1420 	priv = netdev_priv(dev);
1421 	peer = rtnl_dereference(priv->peer);
1422 
1423 	/* Note : dellink() is called from default_device_exit_batch(),
1424 	 * before a rcu_synchronize() point. The devices are guaranteed
1425 	 * not being freed before one RCU grace period.
1426 	 */
1427 	RCU_INIT_POINTER(priv->peer, NULL);
1428 	unregister_netdevice_queue(dev, head);
1429 
1430 	if (peer) {
1431 		priv = netdev_priv(peer);
1432 		RCU_INIT_POINTER(priv->peer, NULL);
1433 		unregister_netdevice_queue(peer, head);
1434 	}
1435 }
1436 
1437 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1438 	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
1439 };
1440 
1441 static struct net *veth_get_link_net(const struct net_device *dev)
1442 {
1443 	struct veth_priv *priv = netdev_priv(dev);
1444 	struct net_device *peer = rtnl_dereference(priv->peer);
1445 
1446 	return peer ? dev_net(peer) : dev_net(dev);
1447 }
1448 
1449 static struct rtnl_link_ops veth_link_ops = {
1450 	.kind		= DRV_NAME,
1451 	.priv_size	= sizeof(struct veth_priv),
1452 	.setup		= veth_setup,
1453 	.validate	= veth_validate,
1454 	.newlink	= veth_newlink,
1455 	.dellink	= veth_dellink,
1456 	.policy		= veth_policy,
1457 	.maxtype	= VETH_INFO_MAX,
1458 	.get_link_net	= veth_get_link_net,
1459 };
1460 
1461 /*
1462  * init/fini
1463  */
1464 
1465 static __init int veth_init(void)
1466 {
1467 	return rtnl_link_register(&veth_link_ops);
1468 }
1469 
1470 static __exit void veth_exit(void)
1471 {
1472 	rtnl_link_unregister(&veth_link_ops);
1473 }
1474 
1475 module_init(veth_init);
1476 module_exit(veth_exit);
1477 
1478 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1479 MODULE_LICENSE("GPL v2");
1480 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1481