xref: /openbmc/linux/drivers/net/veth.c (revision 6cd70754)
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 #define VETH_XDP_BATCH		16
39 
40 struct veth_stats {
41 	u64	rx_drops;
42 	/* xdp */
43 	u64	xdp_packets;
44 	u64	xdp_bytes;
45 	u64	xdp_redirect;
46 	u64	xdp_drops;
47 	u64	xdp_tx;
48 	u64	xdp_tx_err;
49 	u64	peer_tq_xdp_xmit;
50 	u64	peer_tq_xdp_xmit_err;
51 };
52 
53 struct veth_rq_stats {
54 	struct veth_stats	vs;
55 	struct u64_stats_sync	syncp;
56 };
57 
58 struct veth_rq {
59 	struct napi_struct	xdp_napi;
60 	struct net_device	*dev;
61 	struct bpf_prog __rcu	*xdp_prog;
62 	struct xdp_mem_info	xdp_mem;
63 	struct veth_rq_stats	stats;
64 	bool			rx_notify_masked;
65 	struct ptr_ring		xdp_ring;
66 	struct xdp_rxq_info	xdp_rxq;
67 };
68 
69 struct veth_priv {
70 	struct net_device __rcu	*peer;
71 	atomic64_t		dropped;
72 	struct bpf_prog		*_xdp_prog;
73 	struct veth_rq		*rq;
74 	unsigned int		requested_headroom;
75 };
76 
77 struct veth_xdp_tx_bq {
78 	struct xdp_frame *q[VETH_XDP_TX_BULK_SIZE];
79 	unsigned int count;
80 };
81 
82 /*
83  * ethtool interface
84  */
85 
86 struct veth_q_stat_desc {
87 	char	desc[ETH_GSTRING_LEN];
88 	size_t	offset;
89 };
90 
91 #define VETH_RQ_STAT(m)	offsetof(struct veth_stats, m)
92 
93 static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
94 	{ "xdp_packets",	VETH_RQ_STAT(xdp_packets) },
95 	{ "xdp_bytes",		VETH_RQ_STAT(xdp_bytes) },
96 	{ "drops",		VETH_RQ_STAT(rx_drops) },
97 	{ "xdp_redirect",	VETH_RQ_STAT(xdp_redirect) },
98 	{ "xdp_drops",		VETH_RQ_STAT(xdp_drops) },
99 	{ "xdp_tx",		VETH_RQ_STAT(xdp_tx) },
100 	{ "xdp_tx_errors",	VETH_RQ_STAT(xdp_tx_err) },
101 };
102 
103 #define VETH_RQ_STATS_LEN	ARRAY_SIZE(veth_rq_stats_desc)
104 
105 static const struct veth_q_stat_desc veth_tq_stats_desc[] = {
106 	{ "xdp_xmit",		VETH_RQ_STAT(peer_tq_xdp_xmit) },
107 	{ "xdp_xmit_errors",	VETH_RQ_STAT(peer_tq_xdp_xmit_err) },
108 };
109 
110 #define VETH_TQ_STATS_LEN	ARRAY_SIZE(veth_tq_stats_desc)
111 
112 static struct {
113 	const char string[ETH_GSTRING_LEN];
114 } ethtool_stats_keys[] = {
115 	{ "peer_ifindex" },
116 };
117 
118 static int veth_get_link_ksettings(struct net_device *dev,
119 				   struct ethtool_link_ksettings *cmd)
120 {
121 	cmd->base.speed		= SPEED_10000;
122 	cmd->base.duplex	= DUPLEX_FULL;
123 	cmd->base.port		= PORT_TP;
124 	cmd->base.autoneg	= AUTONEG_DISABLE;
125 	return 0;
126 }
127 
128 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
129 {
130 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
131 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
132 }
133 
134 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
135 {
136 	char *p = (char *)buf;
137 	int i, j;
138 
139 	switch(stringset) {
140 	case ETH_SS_STATS:
141 		memcpy(p, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
142 		p += sizeof(ethtool_stats_keys);
143 		for (i = 0; i < dev->real_num_rx_queues; i++) {
144 			for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
145 				snprintf(p, ETH_GSTRING_LEN,
146 					 "rx_queue_%u_%.18s",
147 					 i, veth_rq_stats_desc[j].desc);
148 				p += ETH_GSTRING_LEN;
149 			}
150 		}
151 		for (i = 0; i < dev->real_num_tx_queues; i++) {
152 			for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
153 				snprintf(p, ETH_GSTRING_LEN,
154 					 "tx_queue_%u_%.18s",
155 					 i, veth_tq_stats_desc[j].desc);
156 				p += ETH_GSTRING_LEN;
157 			}
158 		}
159 		break;
160 	}
161 }
162 
163 static int veth_get_sset_count(struct net_device *dev, int sset)
164 {
165 	switch (sset) {
166 	case ETH_SS_STATS:
167 		return ARRAY_SIZE(ethtool_stats_keys) +
168 		       VETH_RQ_STATS_LEN * dev->real_num_rx_queues +
169 		       VETH_TQ_STATS_LEN * dev->real_num_tx_queues;
170 	default:
171 		return -EOPNOTSUPP;
172 	}
173 }
174 
175 static void veth_get_ethtool_stats(struct net_device *dev,
176 		struct ethtool_stats *stats, u64 *data)
177 {
178 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
179 	struct net_device *peer = rtnl_dereference(priv->peer);
180 	int i, j, idx;
181 
182 	data[0] = peer ? peer->ifindex : 0;
183 	idx = 1;
184 	for (i = 0; i < dev->real_num_rx_queues; i++) {
185 		const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
186 		const void *stats_base = (void *)&rq_stats->vs;
187 		unsigned int start;
188 		size_t offset;
189 
190 		do {
191 			start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
192 			for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
193 				offset = veth_rq_stats_desc[j].offset;
194 				data[idx + j] = *(u64 *)(stats_base + offset);
195 			}
196 		} while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
197 		idx += VETH_RQ_STATS_LEN;
198 	}
199 
200 	if (!peer)
201 		return;
202 
203 	rcv_priv = netdev_priv(peer);
204 	for (i = 0; i < peer->real_num_rx_queues; i++) {
205 		const struct veth_rq_stats *rq_stats = &rcv_priv->rq[i].stats;
206 		const void *base = (void *)&rq_stats->vs;
207 		unsigned int start, tx_idx = idx;
208 		size_t offset;
209 
210 		tx_idx += (i % dev->real_num_tx_queues) * VETH_TQ_STATS_LEN;
211 		do {
212 			start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
213 			for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
214 				offset = veth_tq_stats_desc[j].offset;
215 				data[tx_idx + j] += *(u64 *)(base + offset);
216 			}
217 		} while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
218 	}
219 }
220 
221 static const struct ethtool_ops veth_ethtool_ops = {
222 	.get_drvinfo		= veth_get_drvinfo,
223 	.get_link		= ethtool_op_get_link,
224 	.get_strings		= veth_get_strings,
225 	.get_sset_count		= veth_get_sset_count,
226 	.get_ethtool_stats	= veth_get_ethtool_stats,
227 	.get_link_ksettings	= veth_get_link_ksettings,
228 	.get_ts_info		= ethtool_op_get_ts_info,
229 };
230 
231 /* general routines */
232 
233 static bool veth_is_xdp_frame(void *ptr)
234 {
235 	return (unsigned long)ptr & VETH_XDP_FLAG;
236 }
237 
238 static struct xdp_frame *veth_ptr_to_xdp(void *ptr)
239 {
240 	return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
241 }
242 
243 static void *veth_xdp_to_ptr(struct xdp_frame *xdp)
244 {
245 	return (void *)((unsigned long)xdp | VETH_XDP_FLAG);
246 }
247 
248 static void veth_ptr_free(void *ptr)
249 {
250 	if (veth_is_xdp_frame(ptr))
251 		xdp_return_frame(veth_ptr_to_xdp(ptr));
252 	else
253 		kfree_skb(ptr);
254 }
255 
256 static void __veth_xdp_flush(struct veth_rq *rq)
257 {
258 	/* Write ptr_ring before reading rx_notify_masked */
259 	smp_mb();
260 	if (!rq->rx_notify_masked) {
261 		rq->rx_notify_masked = true;
262 		napi_schedule(&rq->xdp_napi);
263 	}
264 }
265 
266 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
267 {
268 	if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
269 		dev_kfree_skb_any(skb);
270 		return NET_RX_DROP;
271 	}
272 
273 	return NET_RX_SUCCESS;
274 }
275 
276 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
277 			    struct veth_rq *rq, bool xdp)
278 {
279 	return __dev_forward_skb(dev, skb) ?: xdp ?
280 		veth_xdp_rx(rq, skb) :
281 		netif_rx(skb);
282 }
283 
284 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
285 {
286 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
287 	struct veth_rq *rq = NULL;
288 	struct net_device *rcv;
289 	int length = skb->len;
290 	bool rcv_xdp = false;
291 	int rxq;
292 
293 	rcu_read_lock();
294 	rcv = rcu_dereference(priv->peer);
295 	if (unlikely(!rcv)) {
296 		kfree_skb(skb);
297 		goto drop;
298 	}
299 
300 	rcv_priv = netdev_priv(rcv);
301 	rxq = skb_get_queue_mapping(skb);
302 	if (rxq < rcv->real_num_rx_queues) {
303 		rq = &rcv_priv->rq[rxq];
304 		rcv_xdp = rcu_access_pointer(rq->xdp_prog);
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 xdp_frame *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 	struct xdp_frame orig_frame;
571 	struct bpf_prog *xdp_prog;
572 
573 	rcu_read_lock();
574 	xdp_prog = rcu_dereference(rq->xdp_prog);
575 	if (likely(xdp_prog)) {
576 		struct xdp_buff xdp;
577 		u32 act;
578 
579 		xdp_convert_frame_to_buff(frame, &xdp);
580 		xdp.rxq = &rq->xdp_rxq;
581 
582 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
583 
584 		switch (act) {
585 		case XDP_PASS:
586 			if (xdp_update_frame_from_buff(&xdp, frame))
587 				goto err_xdp;
588 			break;
589 		case XDP_TX:
590 			orig_frame = *frame;
591 			xdp.rxq->mem = frame->mem;
592 			if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
593 				trace_xdp_exception(rq->dev, xdp_prog, act);
594 				frame = &orig_frame;
595 				stats->rx_drops++;
596 				goto err_xdp;
597 			}
598 			stats->xdp_tx++;
599 			rcu_read_unlock();
600 			goto xdp_xmit;
601 		case XDP_REDIRECT:
602 			orig_frame = *frame;
603 			xdp.rxq->mem = frame->mem;
604 			if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
605 				frame = &orig_frame;
606 				stats->rx_drops++;
607 				goto err_xdp;
608 			}
609 			stats->xdp_redirect++;
610 			rcu_read_unlock();
611 			goto xdp_xmit;
612 		default:
613 			bpf_warn_invalid_xdp_action(act);
614 			fallthrough;
615 		case XDP_ABORTED:
616 			trace_xdp_exception(rq->dev, xdp_prog, act);
617 			fallthrough;
618 		case XDP_DROP:
619 			stats->xdp_drops++;
620 			goto err_xdp;
621 		}
622 	}
623 	rcu_read_unlock();
624 
625 	return frame;
626 err_xdp:
627 	rcu_read_unlock();
628 	xdp_return_frame(frame);
629 xdp_xmit:
630 	return NULL;
631 }
632 
633 /* frames array contains VETH_XDP_BATCH at most */
634 static void veth_xdp_rcv_bulk_skb(struct veth_rq *rq, void **frames,
635 				  int n_xdpf, struct veth_xdp_tx_bq *bq,
636 				  struct veth_stats *stats)
637 {
638 	void *skbs[VETH_XDP_BATCH];
639 	int i;
640 
641 	if (xdp_alloc_skb_bulk(skbs, n_xdpf,
642 			       GFP_ATOMIC | __GFP_ZERO) < 0) {
643 		for (i = 0; i < n_xdpf; i++)
644 			xdp_return_frame(frames[i]);
645 		stats->rx_drops += n_xdpf;
646 
647 		return;
648 	}
649 
650 	for (i = 0; i < n_xdpf; i++) {
651 		struct sk_buff *skb = skbs[i];
652 
653 		skb = __xdp_build_skb_from_frame(frames[i], skb,
654 						 rq->dev);
655 		if (!skb) {
656 			xdp_return_frame(frames[i]);
657 			stats->rx_drops++;
658 			continue;
659 		}
660 		napi_gro_receive(&rq->xdp_napi, skb);
661 	}
662 }
663 
664 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
665 					struct sk_buff *skb,
666 					struct veth_xdp_tx_bq *bq,
667 					struct veth_stats *stats)
668 {
669 	u32 pktlen, headroom, act, metalen, frame_sz;
670 	void *orig_data, *orig_data_end;
671 	struct bpf_prog *xdp_prog;
672 	int mac_len, delta, off;
673 	struct xdp_buff xdp;
674 
675 	skb_orphan(skb);
676 
677 	rcu_read_lock();
678 	xdp_prog = rcu_dereference(rq->xdp_prog);
679 	if (unlikely(!xdp_prog)) {
680 		rcu_read_unlock();
681 		goto out;
682 	}
683 
684 	mac_len = skb->data - skb_mac_header(skb);
685 	pktlen = skb->len + mac_len;
686 	headroom = skb_headroom(skb) - mac_len;
687 
688 	if (skb_shared(skb) || skb_head_is_locked(skb) ||
689 	    skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
690 		struct sk_buff *nskb;
691 		int size, head_off;
692 		void *head, *start;
693 		struct page *page;
694 
695 		size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
696 		       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
697 		if (size > PAGE_SIZE)
698 			goto drop;
699 
700 		page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
701 		if (!page)
702 			goto drop;
703 
704 		head = page_address(page);
705 		start = head + VETH_XDP_HEADROOM;
706 		if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
707 			page_frag_free(head);
708 			goto drop;
709 		}
710 
711 		nskb = veth_build_skb(head, VETH_XDP_HEADROOM + mac_len,
712 				      skb->len, PAGE_SIZE);
713 		if (!nskb) {
714 			page_frag_free(head);
715 			goto drop;
716 		}
717 
718 		skb_copy_header(nskb, skb);
719 		head_off = skb_headroom(nskb) - skb_headroom(skb);
720 		skb_headers_offset_update(nskb, head_off);
721 		consume_skb(skb);
722 		skb = nskb;
723 	}
724 
725 	/* SKB "head" area always have tailroom for skb_shared_info */
726 	frame_sz = skb_end_pointer(skb) - skb->head;
727 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
728 	xdp_init_buff(&xdp, frame_sz, &rq->xdp_rxq);
729 	xdp_prepare_buff(&xdp, skb->head, skb->mac_header, pktlen, true);
730 
731 	orig_data = xdp.data;
732 	orig_data_end = xdp.data_end;
733 
734 	act = bpf_prog_run_xdp(xdp_prog, &xdp);
735 
736 	switch (act) {
737 	case XDP_PASS:
738 		break;
739 	case XDP_TX:
740 		get_page(virt_to_page(xdp.data));
741 		consume_skb(skb);
742 		xdp.rxq->mem = rq->xdp_mem;
743 		if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
744 			trace_xdp_exception(rq->dev, xdp_prog, act);
745 			stats->rx_drops++;
746 			goto err_xdp;
747 		}
748 		stats->xdp_tx++;
749 		rcu_read_unlock();
750 		goto xdp_xmit;
751 	case XDP_REDIRECT:
752 		get_page(virt_to_page(xdp.data));
753 		consume_skb(skb);
754 		xdp.rxq->mem = rq->xdp_mem;
755 		if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
756 			stats->rx_drops++;
757 			goto err_xdp;
758 		}
759 		stats->xdp_redirect++;
760 		rcu_read_unlock();
761 		goto xdp_xmit;
762 	default:
763 		bpf_warn_invalid_xdp_action(act);
764 		fallthrough;
765 	case XDP_ABORTED:
766 		trace_xdp_exception(rq->dev, xdp_prog, act);
767 		fallthrough;
768 	case XDP_DROP:
769 		stats->xdp_drops++;
770 		goto xdp_drop;
771 	}
772 	rcu_read_unlock();
773 
774 	/* check if bpf_xdp_adjust_head was used */
775 	delta = orig_data - xdp.data;
776 	off = mac_len + delta;
777 	if (off > 0)
778 		__skb_push(skb, off);
779 	else if (off < 0)
780 		__skb_pull(skb, -off);
781 	skb->mac_header -= delta;
782 
783 	/* check if bpf_xdp_adjust_tail was used */
784 	off = xdp.data_end - orig_data_end;
785 	if (off != 0)
786 		__skb_put(skb, off); /* positive on grow, negative on shrink */
787 	skb->protocol = eth_type_trans(skb, rq->dev);
788 
789 	metalen = xdp.data - xdp.data_meta;
790 	if (metalen)
791 		skb_metadata_set(skb, metalen);
792 out:
793 	return skb;
794 drop:
795 	stats->rx_drops++;
796 xdp_drop:
797 	rcu_read_unlock();
798 	kfree_skb(skb);
799 	return NULL;
800 err_xdp:
801 	rcu_read_unlock();
802 	page_frag_free(xdp.data);
803 xdp_xmit:
804 	return NULL;
805 }
806 
807 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
808 			struct veth_xdp_tx_bq *bq,
809 			struct veth_stats *stats)
810 {
811 	int i, done = 0, n_xdpf = 0;
812 	void *xdpf[VETH_XDP_BATCH];
813 
814 	for (i = 0; i < budget; i++) {
815 		void *ptr = __ptr_ring_consume(&rq->xdp_ring);
816 
817 		if (!ptr)
818 			break;
819 
820 		if (veth_is_xdp_frame(ptr)) {
821 			/* ndo_xdp_xmit */
822 			struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
823 
824 			stats->xdp_bytes += frame->len;
825 			frame = veth_xdp_rcv_one(rq, frame, bq, stats);
826 			if (frame) {
827 				/* XDP_PASS */
828 				xdpf[n_xdpf++] = frame;
829 				if (n_xdpf == VETH_XDP_BATCH) {
830 					veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf,
831 							      bq, stats);
832 					n_xdpf = 0;
833 				}
834 			}
835 		} else {
836 			/* ndo_start_xmit */
837 			struct sk_buff *skb = ptr;
838 
839 			stats->xdp_bytes += skb->len;
840 			skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
841 			if (skb)
842 				napi_gro_receive(&rq->xdp_napi, skb);
843 		}
844 		done++;
845 	}
846 
847 	if (n_xdpf)
848 		veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf, bq, stats);
849 
850 	u64_stats_update_begin(&rq->stats.syncp);
851 	rq->stats.vs.xdp_redirect += stats->xdp_redirect;
852 	rq->stats.vs.xdp_bytes += stats->xdp_bytes;
853 	rq->stats.vs.xdp_drops += stats->xdp_drops;
854 	rq->stats.vs.rx_drops += stats->rx_drops;
855 	rq->stats.vs.xdp_packets += done;
856 	u64_stats_update_end(&rq->stats.syncp);
857 
858 	return done;
859 }
860 
861 static int veth_poll(struct napi_struct *napi, int budget)
862 {
863 	struct veth_rq *rq =
864 		container_of(napi, struct veth_rq, xdp_napi);
865 	struct veth_stats stats = {};
866 	struct veth_xdp_tx_bq bq;
867 	int done;
868 
869 	bq.count = 0;
870 
871 	xdp_set_return_frame_no_direct();
872 	done = veth_xdp_rcv(rq, budget, &bq, &stats);
873 
874 	if (done < budget && napi_complete_done(napi, done)) {
875 		/* Write rx_notify_masked before reading ptr_ring */
876 		smp_store_mb(rq->rx_notify_masked, false);
877 		if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
878 			rq->rx_notify_masked = true;
879 			napi_schedule(&rq->xdp_napi);
880 		}
881 	}
882 
883 	if (stats.xdp_tx > 0)
884 		veth_xdp_flush(rq, &bq);
885 	if (stats.xdp_redirect > 0)
886 		xdp_do_flush();
887 	xdp_clear_return_frame_no_direct();
888 
889 	return done;
890 }
891 
892 static int veth_napi_add(struct net_device *dev)
893 {
894 	struct veth_priv *priv = netdev_priv(dev);
895 	int err, i;
896 
897 	for (i = 0; i < dev->real_num_rx_queues; i++) {
898 		struct veth_rq *rq = &priv->rq[i];
899 
900 		err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
901 		if (err)
902 			goto err_xdp_ring;
903 	}
904 
905 	for (i = 0; i < dev->real_num_rx_queues; i++) {
906 		struct veth_rq *rq = &priv->rq[i];
907 
908 		napi_enable(&rq->xdp_napi);
909 	}
910 
911 	return 0;
912 err_xdp_ring:
913 	for (i--; i >= 0; i--)
914 		ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
915 
916 	return err;
917 }
918 
919 static void veth_napi_del(struct net_device *dev)
920 {
921 	struct veth_priv *priv = netdev_priv(dev);
922 	int i;
923 
924 	for (i = 0; i < dev->real_num_rx_queues; i++) {
925 		struct veth_rq *rq = &priv->rq[i];
926 
927 		napi_disable(&rq->xdp_napi);
928 		__netif_napi_del(&rq->xdp_napi);
929 	}
930 	synchronize_net();
931 
932 	for (i = 0; i < dev->real_num_rx_queues; i++) {
933 		struct veth_rq *rq = &priv->rq[i];
934 
935 		rq->rx_notify_masked = false;
936 		ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
937 	}
938 }
939 
940 static int veth_enable_xdp(struct net_device *dev)
941 {
942 	struct veth_priv *priv = netdev_priv(dev);
943 	int err, i;
944 
945 	if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
946 		for (i = 0; i < dev->real_num_rx_queues; i++) {
947 			struct veth_rq *rq = &priv->rq[i];
948 
949 			netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
950 			err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i, rq->xdp_napi.napi_id);
951 			if (err < 0)
952 				goto err_rxq_reg;
953 
954 			err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
955 							 MEM_TYPE_PAGE_SHARED,
956 							 NULL);
957 			if (err < 0)
958 				goto err_reg_mem;
959 
960 			/* Save original mem info as it can be overwritten */
961 			rq->xdp_mem = rq->xdp_rxq.mem;
962 		}
963 
964 		err = veth_napi_add(dev);
965 		if (err)
966 			goto err_rxq_reg;
967 	}
968 
969 	for (i = 0; i < dev->real_num_rx_queues; i++)
970 		rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
971 
972 	return 0;
973 err_reg_mem:
974 	xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
975 err_rxq_reg:
976 	for (i--; i >= 0; i--) {
977 		struct veth_rq *rq = &priv->rq[i];
978 
979 		xdp_rxq_info_unreg(&rq->xdp_rxq);
980 		netif_napi_del(&rq->xdp_napi);
981 	}
982 
983 	return err;
984 }
985 
986 static void veth_disable_xdp(struct net_device *dev)
987 {
988 	struct veth_priv *priv = netdev_priv(dev);
989 	int i;
990 
991 	for (i = 0; i < dev->real_num_rx_queues; i++)
992 		rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
993 	veth_napi_del(dev);
994 	for (i = 0; i < dev->real_num_rx_queues; i++) {
995 		struct veth_rq *rq = &priv->rq[i];
996 
997 		rq->xdp_rxq.mem = rq->xdp_mem;
998 		xdp_rxq_info_unreg(&rq->xdp_rxq);
999 	}
1000 }
1001 
1002 static int veth_open(struct net_device *dev)
1003 {
1004 	struct veth_priv *priv = netdev_priv(dev);
1005 	struct net_device *peer = rtnl_dereference(priv->peer);
1006 	int err;
1007 
1008 	if (!peer)
1009 		return -ENOTCONN;
1010 
1011 	if (priv->_xdp_prog) {
1012 		err = veth_enable_xdp(dev);
1013 		if (err)
1014 			return err;
1015 	}
1016 
1017 	if (peer->flags & IFF_UP) {
1018 		netif_carrier_on(dev);
1019 		netif_carrier_on(peer);
1020 	}
1021 
1022 	return 0;
1023 }
1024 
1025 static int veth_close(struct net_device *dev)
1026 {
1027 	struct veth_priv *priv = netdev_priv(dev);
1028 	struct net_device *peer = rtnl_dereference(priv->peer);
1029 
1030 	netif_carrier_off(dev);
1031 	if (peer)
1032 		netif_carrier_off(peer);
1033 
1034 	if (priv->_xdp_prog)
1035 		veth_disable_xdp(dev);
1036 
1037 	return 0;
1038 }
1039 
1040 static int is_valid_veth_mtu(int mtu)
1041 {
1042 	return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1043 }
1044 
1045 static int veth_alloc_queues(struct net_device *dev)
1046 {
1047 	struct veth_priv *priv = netdev_priv(dev);
1048 	int i;
1049 
1050 	priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
1051 	if (!priv->rq)
1052 		return -ENOMEM;
1053 
1054 	for (i = 0; i < dev->num_rx_queues; i++) {
1055 		priv->rq[i].dev = dev;
1056 		u64_stats_init(&priv->rq[i].stats.syncp);
1057 	}
1058 
1059 	return 0;
1060 }
1061 
1062 static void veth_free_queues(struct net_device *dev)
1063 {
1064 	struct veth_priv *priv = netdev_priv(dev);
1065 
1066 	kfree(priv->rq);
1067 }
1068 
1069 static int veth_dev_init(struct net_device *dev)
1070 {
1071 	int err;
1072 
1073 	dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1074 	if (!dev->lstats)
1075 		return -ENOMEM;
1076 
1077 	err = veth_alloc_queues(dev);
1078 	if (err) {
1079 		free_percpu(dev->lstats);
1080 		return err;
1081 	}
1082 
1083 	return 0;
1084 }
1085 
1086 static void veth_dev_free(struct net_device *dev)
1087 {
1088 	veth_free_queues(dev);
1089 	free_percpu(dev->lstats);
1090 }
1091 
1092 #ifdef CONFIG_NET_POLL_CONTROLLER
1093 static void veth_poll_controller(struct net_device *dev)
1094 {
1095 	/* veth only receives frames when its peer sends one
1096 	 * Since it has nothing to do with disabling irqs, we are guaranteed
1097 	 * never to have pending data when we poll for it so
1098 	 * there is nothing to do here.
1099 	 *
1100 	 * We need this though so netpoll recognizes us as an interface that
1101 	 * supports polling, which enables bridge devices in virt setups to
1102 	 * still use netconsole
1103 	 */
1104 }
1105 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1106 
1107 static int veth_get_iflink(const struct net_device *dev)
1108 {
1109 	struct veth_priv *priv = netdev_priv(dev);
1110 	struct net_device *peer;
1111 	int iflink;
1112 
1113 	rcu_read_lock();
1114 	peer = rcu_dereference(priv->peer);
1115 	iflink = peer ? peer->ifindex : 0;
1116 	rcu_read_unlock();
1117 
1118 	return iflink;
1119 }
1120 
1121 static netdev_features_t veth_fix_features(struct net_device *dev,
1122 					   netdev_features_t features)
1123 {
1124 	struct veth_priv *priv = netdev_priv(dev);
1125 	struct net_device *peer;
1126 
1127 	peer = rtnl_dereference(priv->peer);
1128 	if (peer) {
1129 		struct veth_priv *peer_priv = netdev_priv(peer);
1130 
1131 		if (peer_priv->_xdp_prog)
1132 			features &= ~NETIF_F_GSO_SOFTWARE;
1133 	}
1134 
1135 	return features;
1136 }
1137 
1138 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1139 {
1140 	struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1141 	struct net_device *peer;
1142 
1143 	if (new_hr < 0)
1144 		new_hr = 0;
1145 
1146 	rcu_read_lock();
1147 	peer = rcu_dereference(priv->peer);
1148 	if (unlikely(!peer))
1149 		goto out;
1150 
1151 	peer_priv = netdev_priv(peer);
1152 	priv->requested_headroom = new_hr;
1153 	new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1154 	dev->needed_headroom = new_hr;
1155 	peer->needed_headroom = new_hr;
1156 
1157 out:
1158 	rcu_read_unlock();
1159 }
1160 
1161 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1162 			struct netlink_ext_ack *extack)
1163 {
1164 	struct veth_priv *priv = netdev_priv(dev);
1165 	struct bpf_prog *old_prog;
1166 	struct net_device *peer;
1167 	unsigned int max_mtu;
1168 	int err;
1169 
1170 	old_prog = priv->_xdp_prog;
1171 	priv->_xdp_prog = prog;
1172 	peer = rtnl_dereference(priv->peer);
1173 
1174 	if (prog) {
1175 		if (!peer) {
1176 			NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1177 			err = -ENOTCONN;
1178 			goto err;
1179 		}
1180 
1181 		max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1182 			  peer->hard_header_len -
1183 			  SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1184 		if (peer->mtu > max_mtu) {
1185 			NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1186 			err = -ERANGE;
1187 			goto err;
1188 		}
1189 
1190 		if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1191 			NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1192 			err = -ENOSPC;
1193 			goto err;
1194 		}
1195 
1196 		if (dev->flags & IFF_UP) {
1197 			err = veth_enable_xdp(dev);
1198 			if (err) {
1199 				NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1200 				goto err;
1201 			}
1202 		}
1203 
1204 		if (!old_prog) {
1205 			peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1206 			peer->max_mtu = max_mtu;
1207 		}
1208 	}
1209 
1210 	if (old_prog) {
1211 		if (!prog) {
1212 			if (dev->flags & IFF_UP)
1213 				veth_disable_xdp(dev);
1214 
1215 			if (peer) {
1216 				peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1217 				peer->max_mtu = ETH_MAX_MTU;
1218 			}
1219 		}
1220 		bpf_prog_put(old_prog);
1221 	}
1222 
1223 	if ((!!old_prog ^ !!prog) && peer)
1224 		netdev_update_features(peer);
1225 
1226 	return 0;
1227 err:
1228 	priv->_xdp_prog = old_prog;
1229 
1230 	return err;
1231 }
1232 
1233 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1234 {
1235 	switch (xdp->command) {
1236 	case XDP_SETUP_PROG:
1237 		return veth_xdp_set(dev, xdp->prog, xdp->extack);
1238 	default:
1239 		return -EINVAL;
1240 	}
1241 }
1242 
1243 static const struct net_device_ops veth_netdev_ops = {
1244 	.ndo_init            = veth_dev_init,
1245 	.ndo_open            = veth_open,
1246 	.ndo_stop            = veth_close,
1247 	.ndo_start_xmit      = veth_xmit,
1248 	.ndo_get_stats64     = veth_get_stats64,
1249 	.ndo_set_rx_mode     = veth_set_multicast_list,
1250 	.ndo_set_mac_address = eth_mac_addr,
1251 #ifdef CONFIG_NET_POLL_CONTROLLER
1252 	.ndo_poll_controller	= veth_poll_controller,
1253 #endif
1254 	.ndo_get_iflink		= veth_get_iflink,
1255 	.ndo_fix_features	= veth_fix_features,
1256 	.ndo_features_check	= passthru_features_check,
1257 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
1258 	.ndo_bpf		= veth_xdp,
1259 	.ndo_xdp_xmit		= veth_ndo_xdp_xmit,
1260 	.ndo_get_peer_dev	= veth_peer_dev,
1261 };
1262 
1263 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1264 		       NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1265 		       NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1266 		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1267 		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1268 
1269 static void veth_setup(struct net_device *dev)
1270 {
1271 	ether_setup(dev);
1272 
1273 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1274 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1275 	dev->priv_flags |= IFF_NO_QUEUE;
1276 	dev->priv_flags |= IFF_PHONY_HEADROOM;
1277 
1278 	dev->netdev_ops = &veth_netdev_ops;
1279 	dev->ethtool_ops = &veth_ethtool_ops;
1280 	dev->features |= NETIF_F_LLTX;
1281 	dev->features |= VETH_FEATURES;
1282 	dev->vlan_features = dev->features &
1283 			     ~(NETIF_F_HW_VLAN_CTAG_TX |
1284 			       NETIF_F_HW_VLAN_STAG_TX |
1285 			       NETIF_F_HW_VLAN_CTAG_RX |
1286 			       NETIF_F_HW_VLAN_STAG_RX);
1287 	dev->needs_free_netdev = true;
1288 	dev->priv_destructor = veth_dev_free;
1289 	dev->max_mtu = ETH_MAX_MTU;
1290 
1291 	dev->hw_features = VETH_FEATURES;
1292 	dev->hw_enc_features = VETH_FEATURES;
1293 	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1294 }
1295 
1296 /*
1297  * netlink interface
1298  */
1299 
1300 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1301 			 struct netlink_ext_ack *extack)
1302 {
1303 	if (tb[IFLA_ADDRESS]) {
1304 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1305 			return -EINVAL;
1306 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1307 			return -EADDRNOTAVAIL;
1308 	}
1309 	if (tb[IFLA_MTU]) {
1310 		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1311 			return -EINVAL;
1312 	}
1313 	return 0;
1314 }
1315 
1316 static struct rtnl_link_ops veth_link_ops;
1317 
1318 static int veth_newlink(struct net *src_net, struct net_device *dev,
1319 			struct nlattr *tb[], struct nlattr *data[],
1320 			struct netlink_ext_ack *extack)
1321 {
1322 	int err;
1323 	struct net_device *peer;
1324 	struct veth_priv *priv;
1325 	char ifname[IFNAMSIZ];
1326 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1327 	unsigned char name_assign_type;
1328 	struct ifinfomsg *ifmp;
1329 	struct net *net;
1330 
1331 	/*
1332 	 * create and register peer first
1333 	 */
1334 	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1335 		struct nlattr *nla_peer;
1336 
1337 		nla_peer = data[VETH_INFO_PEER];
1338 		ifmp = nla_data(nla_peer);
1339 		err = rtnl_nla_parse_ifla(peer_tb,
1340 					  nla_data(nla_peer) + sizeof(struct ifinfomsg),
1341 					  nla_len(nla_peer) - sizeof(struct ifinfomsg),
1342 					  NULL);
1343 		if (err < 0)
1344 			return err;
1345 
1346 		err = veth_validate(peer_tb, NULL, extack);
1347 		if (err < 0)
1348 			return err;
1349 
1350 		tbp = peer_tb;
1351 	} else {
1352 		ifmp = NULL;
1353 		tbp = tb;
1354 	}
1355 
1356 	if (ifmp && tbp[IFLA_IFNAME]) {
1357 		nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1358 		name_assign_type = NET_NAME_USER;
1359 	} else {
1360 		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1361 		name_assign_type = NET_NAME_ENUM;
1362 	}
1363 
1364 	net = rtnl_link_get_net(src_net, tbp);
1365 	if (IS_ERR(net))
1366 		return PTR_ERR(net);
1367 
1368 	peer = rtnl_create_link(net, ifname, name_assign_type,
1369 				&veth_link_ops, tbp, extack);
1370 	if (IS_ERR(peer)) {
1371 		put_net(net);
1372 		return PTR_ERR(peer);
1373 	}
1374 
1375 	if (!ifmp || !tbp[IFLA_ADDRESS])
1376 		eth_hw_addr_random(peer);
1377 
1378 	if (ifmp && (dev->ifindex != 0))
1379 		peer->ifindex = ifmp->ifi_index;
1380 
1381 	peer->gso_max_size = dev->gso_max_size;
1382 	peer->gso_max_segs = dev->gso_max_segs;
1383 
1384 	err = register_netdevice(peer);
1385 	put_net(net);
1386 	net = NULL;
1387 	if (err < 0)
1388 		goto err_register_peer;
1389 
1390 	netif_carrier_off(peer);
1391 
1392 	err = rtnl_configure_link(peer, ifmp);
1393 	if (err < 0)
1394 		goto err_configure_peer;
1395 
1396 	/*
1397 	 * register dev last
1398 	 *
1399 	 * note, that since we've registered new device the dev's name
1400 	 * should be re-allocated
1401 	 */
1402 
1403 	if (tb[IFLA_ADDRESS] == NULL)
1404 		eth_hw_addr_random(dev);
1405 
1406 	if (tb[IFLA_IFNAME])
1407 		nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1408 	else
1409 		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1410 
1411 	err = register_netdevice(dev);
1412 	if (err < 0)
1413 		goto err_register_dev;
1414 
1415 	netif_carrier_off(dev);
1416 
1417 	/*
1418 	 * tie the deviced together
1419 	 */
1420 
1421 	priv = netdev_priv(dev);
1422 	rcu_assign_pointer(priv->peer, peer);
1423 
1424 	priv = netdev_priv(peer);
1425 	rcu_assign_pointer(priv->peer, dev);
1426 
1427 	return 0;
1428 
1429 err_register_dev:
1430 	/* nothing to do */
1431 err_configure_peer:
1432 	unregister_netdevice(peer);
1433 	return err;
1434 
1435 err_register_peer:
1436 	free_netdev(peer);
1437 	return err;
1438 }
1439 
1440 static void veth_dellink(struct net_device *dev, struct list_head *head)
1441 {
1442 	struct veth_priv *priv;
1443 	struct net_device *peer;
1444 
1445 	priv = netdev_priv(dev);
1446 	peer = rtnl_dereference(priv->peer);
1447 
1448 	/* Note : dellink() is called from default_device_exit_batch(),
1449 	 * before a rcu_synchronize() point. The devices are guaranteed
1450 	 * not being freed before one RCU grace period.
1451 	 */
1452 	RCU_INIT_POINTER(priv->peer, NULL);
1453 	unregister_netdevice_queue(dev, head);
1454 
1455 	if (peer) {
1456 		priv = netdev_priv(peer);
1457 		RCU_INIT_POINTER(priv->peer, NULL);
1458 		unregister_netdevice_queue(peer, head);
1459 	}
1460 }
1461 
1462 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1463 	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
1464 };
1465 
1466 static struct net *veth_get_link_net(const struct net_device *dev)
1467 {
1468 	struct veth_priv *priv = netdev_priv(dev);
1469 	struct net_device *peer = rtnl_dereference(priv->peer);
1470 
1471 	return peer ? dev_net(peer) : dev_net(dev);
1472 }
1473 
1474 static struct rtnl_link_ops veth_link_ops = {
1475 	.kind		= DRV_NAME,
1476 	.priv_size	= sizeof(struct veth_priv),
1477 	.setup		= veth_setup,
1478 	.validate	= veth_validate,
1479 	.newlink	= veth_newlink,
1480 	.dellink	= veth_dellink,
1481 	.policy		= veth_policy,
1482 	.maxtype	= VETH_INFO_MAX,
1483 	.get_link_net	= veth_get_link_net,
1484 };
1485 
1486 /*
1487  * init/fini
1488  */
1489 
1490 static __init int veth_init(void)
1491 {
1492 	return rtnl_link_register(&veth_link_ops);
1493 }
1494 
1495 static __exit void veth_exit(void)
1496 {
1497 	rtnl_link_unregister(&veth_link_ops);
1498 }
1499 
1500 module_init(veth_init);
1501 module_exit(veth_exit);
1502 
1503 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1504 MODULE_LICENSE("GPL v2");
1505 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1506