xref: /linux/drivers/net/veth.c (revision 05ee19c18c2bb3dea69e29219017367c4a77e65a)
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 void *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(void *ptr)
243 {
244 	return (void *)((unsigned long)ptr | 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 int veth_xdp_xmit(struct net_device *dev, int n,
424 			 struct xdp_frame **frames,
425 			 u32 flags, bool ndo_xmit)
426 {
427 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
428 	int i, ret = -ENXIO, drops = 0;
429 	struct net_device *rcv;
430 	unsigned int max_len;
431 	struct veth_rq *rq;
432 
433 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
434 		return -EINVAL;
435 
436 	rcu_read_lock();
437 	rcv = rcu_dereference(priv->peer);
438 	if (unlikely(!rcv))
439 		goto out;
440 
441 	rcv_priv = netdev_priv(rcv);
442 	rq = &rcv_priv->rq[veth_select_rxq(rcv)];
443 	/* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
444 	 * side. This means an XDP program is loaded on the peer and the peer
445 	 * device is up.
446 	 */
447 	if (!rcu_access_pointer(rq->xdp_prog))
448 		goto out;
449 
450 	max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
451 
452 	spin_lock(&rq->xdp_ring.producer_lock);
453 	for (i = 0; i < n; i++) {
454 		struct xdp_frame *frame = frames[i];
455 		void *ptr = veth_xdp_to_ptr(frame);
456 
457 		if (unlikely(frame->len > max_len ||
458 			     __ptr_ring_produce(&rq->xdp_ring, ptr))) {
459 			xdp_return_frame_rx_napi(frame);
460 			drops++;
461 		}
462 	}
463 	spin_unlock(&rq->xdp_ring.producer_lock);
464 
465 	if (flags & XDP_XMIT_FLUSH)
466 		__veth_xdp_flush(rq);
467 
468 	ret = n - drops;
469 	if (ndo_xmit) {
470 		u64_stats_update_begin(&rq->stats.syncp);
471 		rq->stats.vs.peer_tq_xdp_xmit += n - drops;
472 		rq->stats.vs.peer_tq_xdp_xmit_err += drops;
473 		u64_stats_update_end(&rq->stats.syncp);
474 	}
475 
476 out:
477 	rcu_read_unlock();
478 
479 	return ret;
480 }
481 
482 static int veth_ndo_xdp_xmit(struct net_device *dev, int n,
483 			     struct xdp_frame **frames, u32 flags)
484 {
485 	int err;
486 
487 	err = veth_xdp_xmit(dev, n, frames, flags, true);
488 	if (err < 0) {
489 		struct veth_priv *priv = netdev_priv(dev);
490 
491 		atomic64_add(n, &priv->dropped);
492 	}
493 
494 	return err;
495 }
496 
497 static void veth_xdp_flush_bq(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
498 {
499 	int sent, i, err = 0;
500 
501 	sent = veth_xdp_xmit(rq->dev, bq->count, bq->q, 0, false);
502 	if (sent < 0) {
503 		err = sent;
504 		sent = 0;
505 		for (i = 0; i < bq->count; i++)
506 			xdp_return_frame(bq->q[i]);
507 	}
508 	trace_xdp_bulk_tx(rq->dev, sent, bq->count - sent, err);
509 
510 	u64_stats_update_begin(&rq->stats.syncp);
511 	rq->stats.vs.xdp_tx += sent;
512 	rq->stats.vs.xdp_tx_err += bq->count - sent;
513 	u64_stats_update_end(&rq->stats.syncp);
514 
515 	bq->count = 0;
516 }
517 
518 static void veth_xdp_flush(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
519 {
520 	struct veth_priv *rcv_priv, *priv = netdev_priv(rq->dev);
521 	struct net_device *rcv;
522 	struct veth_rq *rcv_rq;
523 
524 	rcu_read_lock();
525 	veth_xdp_flush_bq(rq, bq);
526 	rcv = rcu_dereference(priv->peer);
527 	if (unlikely(!rcv))
528 		goto out;
529 
530 	rcv_priv = netdev_priv(rcv);
531 	rcv_rq = &rcv_priv->rq[veth_select_rxq(rcv)];
532 	/* xdp_ring is initialized on receive side? */
533 	if (unlikely(!rcu_access_pointer(rcv_rq->xdp_prog)))
534 		goto out;
535 
536 	__veth_xdp_flush(rcv_rq);
537 out:
538 	rcu_read_unlock();
539 }
540 
541 static int veth_xdp_tx(struct veth_rq *rq, struct xdp_buff *xdp,
542 		       struct veth_xdp_tx_bq *bq)
543 {
544 	struct xdp_frame *frame = convert_to_xdp_frame(xdp);
545 
546 	if (unlikely(!frame))
547 		return -EOVERFLOW;
548 
549 	if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
550 		veth_xdp_flush_bq(rq, bq);
551 
552 	bq->q[bq->count++] = frame;
553 
554 	return 0;
555 }
556 
557 static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
558 					struct xdp_frame *frame,
559 					struct veth_xdp_tx_bq *bq,
560 					struct veth_stats *stats)
561 {
562 	void *hard_start = frame->data - frame->headroom;
563 	int len = frame->len, delta = 0;
564 	struct xdp_frame orig_frame;
565 	struct bpf_prog *xdp_prog;
566 	unsigned int headroom;
567 	struct sk_buff *skb;
568 
569 	/* bpf_xdp_adjust_head() assures BPF cannot access xdp_frame area */
570 	hard_start -= sizeof(struct xdp_frame);
571 
572 	rcu_read_lock();
573 	xdp_prog = rcu_dereference(rq->xdp_prog);
574 	if (likely(xdp_prog)) {
575 		struct xdp_buff xdp;
576 		u32 act;
577 
578 		xdp.data_hard_start = hard_start;
579 		xdp.data = frame->data;
580 		xdp.data_end = frame->data + frame->len;
581 		xdp.data_meta = frame->data - frame->metasize;
582 		xdp.frame_sz = frame->frame_sz;
583 		xdp.rxq = &rq->xdp_rxq;
584 
585 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
586 
587 		switch (act) {
588 		case XDP_PASS:
589 			delta = frame->data - xdp.data;
590 			len = xdp.data_end - xdp.data;
591 			break;
592 		case XDP_TX:
593 			orig_frame = *frame;
594 			xdp.rxq->mem = frame->mem;
595 			if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
596 				trace_xdp_exception(rq->dev, xdp_prog, act);
597 				frame = &orig_frame;
598 				stats->rx_drops++;
599 				goto err_xdp;
600 			}
601 			stats->xdp_tx++;
602 			rcu_read_unlock();
603 			goto xdp_xmit;
604 		case XDP_REDIRECT:
605 			orig_frame = *frame;
606 			xdp.rxq->mem = frame->mem;
607 			if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
608 				frame = &orig_frame;
609 				stats->rx_drops++;
610 				goto err_xdp;
611 			}
612 			stats->xdp_redirect++;
613 			rcu_read_unlock();
614 			goto xdp_xmit;
615 		default:
616 			bpf_warn_invalid_xdp_action(act);
617 			/* fall through */
618 		case XDP_ABORTED:
619 			trace_xdp_exception(rq->dev, xdp_prog, act);
620 			/* fall through */
621 		case XDP_DROP:
622 			stats->xdp_drops++;
623 			goto err_xdp;
624 		}
625 	}
626 	rcu_read_unlock();
627 
628 	headroom = sizeof(struct xdp_frame) + frame->headroom - delta;
629 	skb = veth_build_skb(hard_start, headroom, len, frame->frame_sz);
630 	if (!skb) {
631 		xdp_return_frame(frame);
632 		stats->rx_drops++;
633 		goto err;
634 	}
635 
636 	xdp_release_frame(frame);
637 	xdp_scrub_frame(frame);
638 	skb->protocol = eth_type_trans(skb, rq->dev);
639 err:
640 	return skb;
641 err_xdp:
642 	rcu_read_unlock();
643 	xdp_return_frame(frame);
644 xdp_xmit:
645 	return NULL;
646 }
647 
648 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
649 					struct sk_buff *skb,
650 					struct veth_xdp_tx_bq *bq,
651 					struct veth_stats *stats)
652 {
653 	u32 pktlen, headroom, act, metalen;
654 	void *orig_data, *orig_data_end;
655 	struct bpf_prog *xdp_prog;
656 	int mac_len, delta, off;
657 	struct xdp_buff xdp;
658 
659 	skb_orphan(skb);
660 
661 	rcu_read_lock();
662 	xdp_prog = rcu_dereference(rq->xdp_prog);
663 	if (unlikely(!xdp_prog)) {
664 		rcu_read_unlock();
665 		goto out;
666 	}
667 
668 	mac_len = skb->data - skb_mac_header(skb);
669 	pktlen = skb->len + mac_len;
670 	headroom = skb_headroom(skb) - mac_len;
671 
672 	if (skb_shared(skb) || skb_head_is_locked(skb) ||
673 	    skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
674 		struct sk_buff *nskb;
675 		int size, head_off;
676 		void *head, *start;
677 		struct page *page;
678 
679 		size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
680 		       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
681 		if (size > PAGE_SIZE)
682 			goto drop;
683 
684 		page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
685 		if (!page)
686 			goto drop;
687 
688 		head = page_address(page);
689 		start = head + VETH_XDP_HEADROOM;
690 		if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
691 			page_frag_free(head);
692 			goto drop;
693 		}
694 
695 		nskb = veth_build_skb(head, VETH_XDP_HEADROOM + mac_len,
696 				      skb->len, PAGE_SIZE);
697 		if (!nskb) {
698 			page_frag_free(head);
699 			goto drop;
700 		}
701 
702 		skb_copy_header(nskb, skb);
703 		head_off = skb_headroom(nskb) - skb_headroom(skb);
704 		skb_headers_offset_update(nskb, head_off);
705 		consume_skb(skb);
706 		skb = nskb;
707 	}
708 
709 	xdp.data_hard_start = skb->head;
710 	xdp.data = skb_mac_header(skb);
711 	xdp.data_end = xdp.data + pktlen;
712 	xdp.data_meta = xdp.data;
713 	xdp.rxq = &rq->xdp_rxq;
714 
715 	/* SKB "head" area always have tailroom for skb_shared_info */
716 	xdp.frame_sz = (void *)skb_end_pointer(skb) - xdp.data_hard_start;
717 	xdp.frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
718 
719 	orig_data = xdp.data;
720 	orig_data_end = xdp.data_end;
721 
722 	act = bpf_prog_run_xdp(xdp_prog, &xdp);
723 
724 	switch (act) {
725 	case XDP_PASS:
726 		break;
727 	case XDP_TX:
728 		get_page(virt_to_page(xdp.data));
729 		consume_skb(skb);
730 		xdp.rxq->mem = rq->xdp_mem;
731 		if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
732 			trace_xdp_exception(rq->dev, xdp_prog, act);
733 			stats->rx_drops++;
734 			goto err_xdp;
735 		}
736 		stats->xdp_tx++;
737 		rcu_read_unlock();
738 		goto xdp_xmit;
739 	case XDP_REDIRECT:
740 		get_page(virt_to_page(xdp.data));
741 		consume_skb(skb);
742 		xdp.rxq->mem = rq->xdp_mem;
743 		if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
744 			stats->rx_drops++;
745 			goto err_xdp;
746 		}
747 		stats->xdp_redirect++;
748 		rcu_read_unlock();
749 		goto xdp_xmit;
750 	default:
751 		bpf_warn_invalid_xdp_action(act);
752 		/* fall through */
753 	case XDP_ABORTED:
754 		trace_xdp_exception(rq->dev, xdp_prog, act);
755 		/* fall through */
756 	case XDP_DROP:
757 		stats->xdp_drops++;
758 		goto xdp_drop;
759 	}
760 	rcu_read_unlock();
761 
762 	/* check if bpf_xdp_adjust_head was used */
763 	delta = orig_data - xdp.data;
764 	off = mac_len + delta;
765 	if (off > 0)
766 		__skb_push(skb, off);
767 	else if (off < 0)
768 		__skb_pull(skb, -off);
769 	skb->mac_header -= delta;
770 
771 	/* check if bpf_xdp_adjust_tail was used */
772 	off = xdp.data_end - orig_data_end;
773 	if (off != 0)
774 		__skb_put(skb, off); /* positive on grow, negative on shrink */
775 	skb->protocol = eth_type_trans(skb, rq->dev);
776 
777 	metalen = xdp.data - xdp.data_meta;
778 	if (metalen)
779 		skb_metadata_set(skb, metalen);
780 out:
781 	return skb;
782 drop:
783 	stats->rx_drops++;
784 xdp_drop:
785 	rcu_read_unlock();
786 	kfree_skb(skb);
787 	return NULL;
788 err_xdp:
789 	rcu_read_unlock();
790 	page_frag_free(xdp.data);
791 xdp_xmit:
792 	return NULL;
793 }
794 
795 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
796 			struct veth_xdp_tx_bq *bq,
797 			struct veth_stats *stats)
798 {
799 	int i, done = 0;
800 
801 	for (i = 0; i < budget; i++) {
802 		void *ptr = __ptr_ring_consume(&rq->xdp_ring);
803 		struct sk_buff *skb;
804 
805 		if (!ptr)
806 			break;
807 
808 		if (veth_is_xdp_frame(ptr)) {
809 			struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
810 
811 			stats->xdp_bytes += frame->len;
812 			skb = veth_xdp_rcv_one(rq, frame, bq, stats);
813 		} else {
814 			skb = ptr;
815 			stats->xdp_bytes += skb->len;
816 			skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
817 		}
818 
819 		if (skb)
820 			napi_gro_receive(&rq->xdp_napi, skb);
821 
822 		done++;
823 	}
824 
825 	u64_stats_update_begin(&rq->stats.syncp);
826 	rq->stats.vs.xdp_redirect += stats->xdp_redirect;
827 	rq->stats.vs.xdp_bytes += stats->xdp_bytes;
828 	rq->stats.vs.xdp_drops += stats->xdp_drops;
829 	rq->stats.vs.rx_drops += stats->rx_drops;
830 	rq->stats.vs.xdp_packets += done;
831 	u64_stats_update_end(&rq->stats.syncp);
832 
833 	return done;
834 }
835 
836 static int veth_poll(struct napi_struct *napi, int budget)
837 {
838 	struct veth_rq *rq =
839 		container_of(napi, struct veth_rq, xdp_napi);
840 	struct veth_stats stats = {};
841 	struct veth_xdp_tx_bq bq;
842 	int done;
843 
844 	bq.count = 0;
845 
846 	xdp_set_return_frame_no_direct();
847 	done = veth_xdp_rcv(rq, budget, &bq, &stats);
848 
849 	if (done < budget && napi_complete_done(napi, done)) {
850 		/* Write rx_notify_masked before reading ptr_ring */
851 		smp_store_mb(rq->rx_notify_masked, false);
852 		if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
853 			rq->rx_notify_masked = true;
854 			napi_schedule(&rq->xdp_napi);
855 		}
856 	}
857 
858 	if (stats.xdp_tx > 0)
859 		veth_xdp_flush(rq, &bq);
860 	if (stats.xdp_redirect > 0)
861 		xdp_do_flush();
862 	xdp_clear_return_frame_no_direct();
863 
864 	return done;
865 }
866 
867 static int veth_napi_add(struct net_device *dev)
868 {
869 	struct veth_priv *priv = netdev_priv(dev);
870 	int err, i;
871 
872 	for (i = 0; i < dev->real_num_rx_queues; i++) {
873 		struct veth_rq *rq = &priv->rq[i];
874 
875 		err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
876 		if (err)
877 			goto err_xdp_ring;
878 	}
879 
880 	for (i = 0; i < dev->real_num_rx_queues; i++) {
881 		struct veth_rq *rq = &priv->rq[i];
882 
883 		netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
884 		napi_enable(&rq->xdp_napi);
885 	}
886 
887 	return 0;
888 err_xdp_ring:
889 	for (i--; i >= 0; i--)
890 		ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
891 
892 	return err;
893 }
894 
895 static void veth_napi_del(struct net_device *dev)
896 {
897 	struct veth_priv *priv = netdev_priv(dev);
898 	int i;
899 
900 	for (i = 0; i < dev->real_num_rx_queues; i++) {
901 		struct veth_rq *rq = &priv->rq[i];
902 
903 		napi_disable(&rq->xdp_napi);
904 		napi_hash_del(&rq->xdp_napi);
905 	}
906 	synchronize_net();
907 
908 	for (i = 0; i < dev->real_num_rx_queues; i++) {
909 		struct veth_rq *rq = &priv->rq[i];
910 
911 		netif_napi_del(&rq->xdp_napi);
912 		rq->rx_notify_masked = false;
913 		ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
914 	}
915 }
916 
917 static int veth_enable_xdp(struct net_device *dev)
918 {
919 	struct veth_priv *priv = netdev_priv(dev);
920 	int err, i;
921 
922 	if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
923 		for (i = 0; i < dev->real_num_rx_queues; i++) {
924 			struct veth_rq *rq = &priv->rq[i];
925 
926 			err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
927 			if (err < 0)
928 				goto err_rxq_reg;
929 
930 			err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
931 							 MEM_TYPE_PAGE_SHARED,
932 							 NULL);
933 			if (err < 0)
934 				goto err_reg_mem;
935 
936 			/* Save original mem info as it can be overwritten */
937 			rq->xdp_mem = rq->xdp_rxq.mem;
938 		}
939 
940 		err = veth_napi_add(dev);
941 		if (err)
942 			goto err_rxq_reg;
943 	}
944 
945 	for (i = 0; i < dev->real_num_rx_queues; i++)
946 		rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
947 
948 	return 0;
949 err_reg_mem:
950 	xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
951 err_rxq_reg:
952 	for (i--; i >= 0; i--)
953 		xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
954 
955 	return err;
956 }
957 
958 static void veth_disable_xdp(struct net_device *dev)
959 {
960 	struct veth_priv *priv = netdev_priv(dev);
961 	int i;
962 
963 	for (i = 0; i < dev->real_num_rx_queues; i++)
964 		rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
965 	veth_napi_del(dev);
966 	for (i = 0; i < dev->real_num_rx_queues; i++) {
967 		struct veth_rq *rq = &priv->rq[i];
968 
969 		rq->xdp_rxq.mem = rq->xdp_mem;
970 		xdp_rxq_info_unreg(&rq->xdp_rxq);
971 	}
972 }
973 
974 static int veth_open(struct net_device *dev)
975 {
976 	struct veth_priv *priv = netdev_priv(dev);
977 	struct net_device *peer = rtnl_dereference(priv->peer);
978 	int err;
979 
980 	if (!peer)
981 		return -ENOTCONN;
982 
983 	if (priv->_xdp_prog) {
984 		err = veth_enable_xdp(dev);
985 		if (err)
986 			return err;
987 	}
988 
989 	if (peer->flags & IFF_UP) {
990 		netif_carrier_on(dev);
991 		netif_carrier_on(peer);
992 	}
993 
994 	return 0;
995 }
996 
997 static int veth_close(struct net_device *dev)
998 {
999 	struct veth_priv *priv = netdev_priv(dev);
1000 	struct net_device *peer = rtnl_dereference(priv->peer);
1001 
1002 	netif_carrier_off(dev);
1003 	if (peer)
1004 		netif_carrier_off(peer);
1005 
1006 	if (priv->_xdp_prog)
1007 		veth_disable_xdp(dev);
1008 
1009 	return 0;
1010 }
1011 
1012 static int is_valid_veth_mtu(int mtu)
1013 {
1014 	return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1015 }
1016 
1017 static int veth_alloc_queues(struct net_device *dev)
1018 {
1019 	struct veth_priv *priv = netdev_priv(dev);
1020 	int i;
1021 
1022 	priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
1023 	if (!priv->rq)
1024 		return -ENOMEM;
1025 
1026 	for (i = 0; i < dev->num_rx_queues; i++) {
1027 		priv->rq[i].dev = dev;
1028 		u64_stats_init(&priv->rq[i].stats.syncp);
1029 	}
1030 
1031 	return 0;
1032 }
1033 
1034 static void veth_free_queues(struct net_device *dev)
1035 {
1036 	struct veth_priv *priv = netdev_priv(dev);
1037 
1038 	kfree(priv->rq);
1039 }
1040 
1041 static int veth_dev_init(struct net_device *dev)
1042 {
1043 	int err;
1044 
1045 	dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1046 	if (!dev->lstats)
1047 		return -ENOMEM;
1048 
1049 	err = veth_alloc_queues(dev);
1050 	if (err) {
1051 		free_percpu(dev->lstats);
1052 		return err;
1053 	}
1054 
1055 	return 0;
1056 }
1057 
1058 static void veth_dev_free(struct net_device *dev)
1059 {
1060 	veth_free_queues(dev);
1061 	free_percpu(dev->lstats);
1062 }
1063 
1064 #ifdef CONFIG_NET_POLL_CONTROLLER
1065 static void veth_poll_controller(struct net_device *dev)
1066 {
1067 	/* veth only receives frames when its peer sends one
1068 	 * Since it has nothing to do with disabling irqs, we are guaranteed
1069 	 * never to have pending data when we poll for it so
1070 	 * there is nothing to do here.
1071 	 *
1072 	 * We need this though so netpoll recognizes us as an interface that
1073 	 * supports polling, which enables bridge devices in virt setups to
1074 	 * still use netconsole
1075 	 */
1076 }
1077 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1078 
1079 static int veth_get_iflink(const struct net_device *dev)
1080 {
1081 	struct veth_priv *priv = netdev_priv(dev);
1082 	struct net_device *peer;
1083 	int iflink;
1084 
1085 	rcu_read_lock();
1086 	peer = rcu_dereference(priv->peer);
1087 	iflink = peer ? peer->ifindex : 0;
1088 	rcu_read_unlock();
1089 
1090 	return iflink;
1091 }
1092 
1093 static netdev_features_t veth_fix_features(struct net_device *dev,
1094 					   netdev_features_t features)
1095 {
1096 	struct veth_priv *priv = netdev_priv(dev);
1097 	struct net_device *peer;
1098 
1099 	peer = rtnl_dereference(priv->peer);
1100 	if (peer) {
1101 		struct veth_priv *peer_priv = netdev_priv(peer);
1102 
1103 		if (peer_priv->_xdp_prog)
1104 			features &= ~NETIF_F_GSO_SOFTWARE;
1105 	}
1106 
1107 	return features;
1108 }
1109 
1110 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1111 {
1112 	struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1113 	struct net_device *peer;
1114 
1115 	if (new_hr < 0)
1116 		new_hr = 0;
1117 
1118 	rcu_read_lock();
1119 	peer = rcu_dereference(priv->peer);
1120 	if (unlikely(!peer))
1121 		goto out;
1122 
1123 	peer_priv = netdev_priv(peer);
1124 	priv->requested_headroom = new_hr;
1125 	new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1126 	dev->needed_headroom = new_hr;
1127 	peer->needed_headroom = new_hr;
1128 
1129 out:
1130 	rcu_read_unlock();
1131 }
1132 
1133 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1134 			struct netlink_ext_ack *extack)
1135 {
1136 	struct veth_priv *priv = netdev_priv(dev);
1137 	struct bpf_prog *old_prog;
1138 	struct net_device *peer;
1139 	unsigned int max_mtu;
1140 	int err;
1141 
1142 	old_prog = priv->_xdp_prog;
1143 	priv->_xdp_prog = prog;
1144 	peer = rtnl_dereference(priv->peer);
1145 
1146 	if (prog) {
1147 		if (!peer) {
1148 			NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1149 			err = -ENOTCONN;
1150 			goto err;
1151 		}
1152 
1153 		max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1154 			  peer->hard_header_len -
1155 			  SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1156 		if (peer->mtu > max_mtu) {
1157 			NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1158 			err = -ERANGE;
1159 			goto err;
1160 		}
1161 
1162 		if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1163 			NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1164 			err = -ENOSPC;
1165 			goto err;
1166 		}
1167 
1168 		if (dev->flags & IFF_UP) {
1169 			err = veth_enable_xdp(dev);
1170 			if (err) {
1171 				NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1172 				goto err;
1173 			}
1174 		}
1175 
1176 		if (!old_prog) {
1177 			peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1178 			peer->max_mtu = max_mtu;
1179 		}
1180 	}
1181 
1182 	if (old_prog) {
1183 		if (!prog) {
1184 			if (dev->flags & IFF_UP)
1185 				veth_disable_xdp(dev);
1186 
1187 			if (peer) {
1188 				peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1189 				peer->max_mtu = ETH_MAX_MTU;
1190 			}
1191 		}
1192 		bpf_prog_put(old_prog);
1193 	}
1194 
1195 	if ((!!old_prog ^ !!prog) && peer)
1196 		netdev_update_features(peer);
1197 
1198 	return 0;
1199 err:
1200 	priv->_xdp_prog = old_prog;
1201 
1202 	return err;
1203 }
1204 
1205 static u32 veth_xdp_query(struct net_device *dev)
1206 {
1207 	struct veth_priv *priv = netdev_priv(dev);
1208 	const struct bpf_prog *xdp_prog;
1209 
1210 	xdp_prog = priv->_xdp_prog;
1211 	if (xdp_prog)
1212 		return xdp_prog->aux->id;
1213 
1214 	return 0;
1215 }
1216 
1217 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1218 {
1219 	switch (xdp->command) {
1220 	case XDP_SETUP_PROG:
1221 		return veth_xdp_set(dev, xdp->prog, xdp->extack);
1222 	case XDP_QUERY_PROG:
1223 		xdp->prog_id = veth_xdp_query(dev);
1224 		return 0;
1225 	default:
1226 		return -EINVAL;
1227 	}
1228 }
1229 
1230 static const struct net_device_ops veth_netdev_ops = {
1231 	.ndo_init            = veth_dev_init,
1232 	.ndo_open            = veth_open,
1233 	.ndo_stop            = veth_close,
1234 	.ndo_start_xmit      = veth_xmit,
1235 	.ndo_get_stats64     = veth_get_stats64,
1236 	.ndo_set_rx_mode     = veth_set_multicast_list,
1237 	.ndo_set_mac_address = eth_mac_addr,
1238 #ifdef CONFIG_NET_POLL_CONTROLLER
1239 	.ndo_poll_controller	= veth_poll_controller,
1240 #endif
1241 	.ndo_get_iflink		= veth_get_iflink,
1242 	.ndo_fix_features	= veth_fix_features,
1243 	.ndo_features_check	= passthru_features_check,
1244 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
1245 	.ndo_bpf		= veth_xdp,
1246 	.ndo_xdp_xmit		= veth_ndo_xdp_xmit,
1247 };
1248 
1249 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1250 		       NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1251 		       NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1252 		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1253 		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1254 
1255 static void veth_setup(struct net_device *dev)
1256 {
1257 	ether_setup(dev);
1258 
1259 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1260 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1261 	dev->priv_flags |= IFF_NO_QUEUE;
1262 	dev->priv_flags |= IFF_PHONY_HEADROOM;
1263 
1264 	dev->netdev_ops = &veth_netdev_ops;
1265 	dev->ethtool_ops = &veth_ethtool_ops;
1266 	dev->features |= NETIF_F_LLTX;
1267 	dev->features |= VETH_FEATURES;
1268 	dev->vlan_features = dev->features &
1269 			     ~(NETIF_F_HW_VLAN_CTAG_TX |
1270 			       NETIF_F_HW_VLAN_STAG_TX |
1271 			       NETIF_F_HW_VLAN_CTAG_RX |
1272 			       NETIF_F_HW_VLAN_STAG_RX);
1273 	dev->needs_free_netdev = true;
1274 	dev->priv_destructor = veth_dev_free;
1275 	dev->max_mtu = ETH_MAX_MTU;
1276 
1277 	dev->hw_features = VETH_FEATURES;
1278 	dev->hw_enc_features = VETH_FEATURES;
1279 	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1280 }
1281 
1282 /*
1283  * netlink interface
1284  */
1285 
1286 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1287 			 struct netlink_ext_ack *extack)
1288 {
1289 	if (tb[IFLA_ADDRESS]) {
1290 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1291 			return -EINVAL;
1292 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1293 			return -EADDRNOTAVAIL;
1294 	}
1295 	if (tb[IFLA_MTU]) {
1296 		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1297 			return -EINVAL;
1298 	}
1299 	return 0;
1300 }
1301 
1302 static struct rtnl_link_ops veth_link_ops;
1303 
1304 static int veth_newlink(struct net *src_net, struct net_device *dev,
1305 			struct nlattr *tb[], struct nlattr *data[],
1306 			struct netlink_ext_ack *extack)
1307 {
1308 	int err;
1309 	struct net_device *peer;
1310 	struct veth_priv *priv;
1311 	char ifname[IFNAMSIZ];
1312 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1313 	unsigned char name_assign_type;
1314 	struct ifinfomsg *ifmp;
1315 	struct net *net;
1316 
1317 	/*
1318 	 * create and register peer first
1319 	 */
1320 	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1321 		struct nlattr *nla_peer;
1322 
1323 		nla_peer = data[VETH_INFO_PEER];
1324 		ifmp = nla_data(nla_peer);
1325 		err = rtnl_nla_parse_ifla(peer_tb,
1326 					  nla_data(nla_peer) + sizeof(struct ifinfomsg),
1327 					  nla_len(nla_peer) - sizeof(struct ifinfomsg),
1328 					  NULL);
1329 		if (err < 0)
1330 			return err;
1331 
1332 		err = veth_validate(peer_tb, NULL, extack);
1333 		if (err < 0)
1334 			return err;
1335 
1336 		tbp = peer_tb;
1337 	} else {
1338 		ifmp = NULL;
1339 		tbp = tb;
1340 	}
1341 
1342 	if (ifmp && tbp[IFLA_IFNAME]) {
1343 		nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1344 		name_assign_type = NET_NAME_USER;
1345 	} else {
1346 		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1347 		name_assign_type = NET_NAME_ENUM;
1348 	}
1349 
1350 	net = rtnl_link_get_net(src_net, tbp);
1351 	if (IS_ERR(net))
1352 		return PTR_ERR(net);
1353 
1354 	peer = rtnl_create_link(net, ifname, name_assign_type,
1355 				&veth_link_ops, tbp, extack);
1356 	if (IS_ERR(peer)) {
1357 		put_net(net);
1358 		return PTR_ERR(peer);
1359 	}
1360 
1361 	if (!ifmp || !tbp[IFLA_ADDRESS])
1362 		eth_hw_addr_random(peer);
1363 
1364 	if (ifmp && (dev->ifindex != 0))
1365 		peer->ifindex = ifmp->ifi_index;
1366 
1367 	peer->gso_max_size = dev->gso_max_size;
1368 	peer->gso_max_segs = dev->gso_max_segs;
1369 
1370 	err = register_netdevice(peer);
1371 	put_net(net);
1372 	net = NULL;
1373 	if (err < 0)
1374 		goto err_register_peer;
1375 
1376 	netif_carrier_off(peer);
1377 
1378 	err = rtnl_configure_link(peer, ifmp);
1379 	if (err < 0)
1380 		goto err_configure_peer;
1381 
1382 	/*
1383 	 * register dev last
1384 	 *
1385 	 * note, that since we've registered new device the dev's name
1386 	 * should be re-allocated
1387 	 */
1388 
1389 	if (tb[IFLA_ADDRESS] == NULL)
1390 		eth_hw_addr_random(dev);
1391 
1392 	if (tb[IFLA_IFNAME])
1393 		nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1394 	else
1395 		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1396 
1397 	err = register_netdevice(dev);
1398 	if (err < 0)
1399 		goto err_register_dev;
1400 
1401 	netif_carrier_off(dev);
1402 
1403 	/*
1404 	 * tie the deviced together
1405 	 */
1406 
1407 	priv = netdev_priv(dev);
1408 	rcu_assign_pointer(priv->peer, peer);
1409 
1410 	priv = netdev_priv(peer);
1411 	rcu_assign_pointer(priv->peer, dev);
1412 
1413 	return 0;
1414 
1415 err_register_dev:
1416 	/* nothing to do */
1417 err_configure_peer:
1418 	unregister_netdevice(peer);
1419 	return err;
1420 
1421 err_register_peer:
1422 	free_netdev(peer);
1423 	return err;
1424 }
1425 
1426 static void veth_dellink(struct net_device *dev, struct list_head *head)
1427 {
1428 	struct veth_priv *priv;
1429 	struct net_device *peer;
1430 
1431 	priv = netdev_priv(dev);
1432 	peer = rtnl_dereference(priv->peer);
1433 
1434 	/* Note : dellink() is called from default_device_exit_batch(),
1435 	 * before a rcu_synchronize() point. The devices are guaranteed
1436 	 * not being freed before one RCU grace period.
1437 	 */
1438 	RCU_INIT_POINTER(priv->peer, NULL);
1439 	unregister_netdevice_queue(dev, head);
1440 
1441 	if (peer) {
1442 		priv = netdev_priv(peer);
1443 		RCU_INIT_POINTER(priv->peer, NULL);
1444 		unregister_netdevice_queue(peer, head);
1445 	}
1446 }
1447 
1448 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1449 	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
1450 };
1451 
1452 static struct net *veth_get_link_net(const struct net_device *dev)
1453 {
1454 	struct veth_priv *priv = netdev_priv(dev);
1455 	struct net_device *peer = rtnl_dereference(priv->peer);
1456 
1457 	return peer ? dev_net(peer) : dev_net(dev);
1458 }
1459 
1460 static struct rtnl_link_ops veth_link_ops = {
1461 	.kind		= DRV_NAME,
1462 	.priv_size	= sizeof(struct veth_priv),
1463 	.setup		= veth_setup,
1464 	.validate	= veth_validate,
1465 	.newlink	= veth_newlink,
1466 	.dellink	= veth_dellink,
1467 	.policy		= veth_policy,
1468 	.maxtype	= VETH_INFO_MAX,
1469 	.get_link_net	= veth_get_link_net,
1470 };
1471 
1472 /*
1473  * init/fini
1474  */
1475 
1476 static __init int veth_init(void)
1477 {
1478 	return rtnl_link_register(&veth_link_ops);
1479 }
1480 
1481 static __exit void veth_exit(void)
1482 {
1483 	rtnl_link_unregister(&veth_link_ops);
1484 }
1485 
1486 module_init(veth_init);
1487 module_exit(veth_exit);
1488 
1489 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1490 MODULE_LICENSE("GPL v2");
1491 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1492