xref: /openbmc/linux/drivers/net/veth.c (revision 8b0adbe3e38dbe5aae9edf6f5159ffdca7cfbdf1)
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, nxmit = 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 			break;
468 		nxmit++;
469 	}
470 	spin_unlock(&rq->xdp_ring.producer_lock);
471 
472 	if (flags & XDP_XMIT_FLUSH)
473 		__veth_xdp_flush(rq);
474 
475 	ret = nxmit;
476 	if (ndo_xmit) {
477 		u64_stats_update_begin(&rq->stats.syncp);
478 		rq->stats.vs.peer_tq_xdp_xmit += nxmit;
479 		rq->stats.vs.peer_tq_xdp_xmit_err += n - nxmit;
480 		u64_stats_update_end(&rq->stats.syncp);
481 	}
482 
483 out:
484 	rcu_read_unlock();
485 
486 	return ret;
487 }
488 
489 static int veth_ndo_xdp_xmit(struct net_device *dev, int n,
490 			     struct xdp_frame **frames, u32 flags)
491 {
492 	int err;
493 
494 	err = veth_xdp_xmit(dev, n, frames, flags, true);
495 	if (err < 0) {
496 		struct veth_priv *priv = netdev_priv(dev);
497 
498 		atomic64_add(n, &priv->dropped);
499 	}
500 
501 	return err;
502 }
503 
504 static void veth_xdp_flush_bq(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
505 {
506 	int sent, i, err = 0, drops;
507 
508 	sent = veth_xdp_xmit(rq->dev, bq->count, bq->q, 0, false);
509 	if (sent < 0) {
510 		err = sent;
511 		sent = 0;
512 	}
513 
514 	for (i = sent; unlikely(i < bq->count); i++)
515 		xdp_return_frame(bq->q[i]);
516 
517 	drops = bq->count - sent;
518 	trace_xdp_bulk_tx(rq->dev, sent, drops, err);
519 
520 	u64_stats_update_begin(&rq->stats.syncp);
521 	rq->stats.vs.xdp_tx += sent;
522 	rq->stats.vs.xdp_tx_err += drops;
523 	u64_stats_update_end(&rq->stats.syncp);
524 
525 	bq->count = 0;
526 }
527 
528 static void veth_xdp_flush(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
529 {
530 	struct veth_priv *rcv_priv, *priv = netdev_priv(rq->dev);
531 	struct net_device *rcv;
532 	struct veth_rq *rcv_rq;
533 
534 	rcu_read_lock();
535 	veth_xdp_flush_bq(rq, bq);
536 	rcv = rcu_dereference(priv->peer);
537 	if (unlikely(!rcv))
538 		goto out;
539 
540 	rcv_priv = netdev_priv(rcv);
541 	rcv_rq = &rcv_priv->rq[veth_select_rxq(rcv)];
542 	/* xdp_ring is initialized on receive side? */
543 	if (unlikely(!rcu_access_pointer(rcv_rq->xdp_prog)))
544 		goto out;
545 
546 	__veth_xdp_flush(rcv_rq);
547 out:
548 	rcu_read_unlock();
549 }
550 
551 static int veth_xdp_tx(struct veth_rq *rq, struct xdp_buff *xdp,
552 		       struct veth_xdp_tx_bq *bq)
553 {
554 	struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
555 
556 	if (unlikely(!frame))
557 		return -EOVERFLOW;
558 
559 	if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
560 		veth_xdp_flush_bq(rq, bq);
561 
562 	bq->q[bq->count++] = frame;
563 
564 	return 0;
565 }
566 
567 static struct xdp_frame *veth_xdp_rcv_one(struct veth_rq *rq,
568 					  struct xdp_frame *frame,
569 					  struct veth_xdp_tx_bq *bq,
570 					  struct veth_stats *stats)
571 {
572 	struct xdp_frame orig_frame;
573 	struct bpf_prog *xdp_prog;
574 
575 	rcu_read_lock();
576 	xdp_prog = rcu_dereference(rq->xdp_prog);
577 	if (likely(xdp_prog)) {
578 		struct xdp_buff xdp;
579 		u32 act;
580 
581 		xdp_convert_frame_to_buff(frame, &xdp);
582 		xdp.rxq = &rq->xdp_rxq;
583 
584 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
585 
586 		switch (act) {
587 		case XDP_PASS:
588 			if (xdp_update_frame_from_buff(&xdp, frame))
589 				goto err_xdp;
590 			break;
591 		case XDP_TX:
592 			orig_frame = *frame;
593 			xdp.rxq->mem = frame->mem;
594 			if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
595 				trace_xdp_exception(rq->dev, xdp_prog, act);
596 				frame = &orig_frame;
597 				stats->rx_drops++;
598 				goto err_xdp;
599 			}
600 			stats->xdp_tx++;
601 			rcu_read_unlock();
602 			goto xdp_xmit;
603 		case XDP_REDIRECT:
604 			orig_frame = *frame;
605 			xdp.rxq->mem = frame->mem;
606 			if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
607 				frame = &orig_frame;
608 				stats->rx_drops++;
609 				goto err_xdp;
610 			}
611 			stats->xdp_redirect++;
612 			rcu_read_unlock();
613 			goto xdp_xmit;
614 		default:
615 			bpf_warn_invalid_xdp_action(act);
616 			fallthrough;
617 		case XDP_ABORTED:
618 			trace_xdp_exception(rq->dev, xdp_prog, act);
619 			fallthrough;
620 		case XDP_DROP:
621 			stats->xdp_drops++;
622 			goto err_xdp;
623 		}
624 	}
625 	rcu_read_unlock();
626 
627 	return frame;
628 err_xdp:
629 	rcu_read_unlock();
630 	xdp_return_frame(frame);
631 xdp_xmit:
632 	return NULL;
633 }
634 
635 /* frames array contains VETH_XDP_BATCH at most */
636 static void veth_xdp_rcv_bulk_skb(struct veth_rq *rq, void **frames,
637 				  int n_xdpf, struct veth_xdp_tx_bq *bq,
638 				  struct veth_stats *stats)
639 {
640 	void *skbs[VETH_XDP_BATCH];
641 	int i;
642 
643 	if (xdp_alloc_skb_bulk(skbs, n_xdpf,
644 			       GFP_ATOMIC | __GFP_ZERO) < 0) {
645 		for (i = 0; i < n_xdpf; i++)
646 			xdp_return_frame(frames[i]);
647 		stats->rx_drops += n_xdpf;
648 
649 		return;
650 	}
651 
652 	for (i = 0; i < n_xdpf; i++) {
653 		struct sk_buff *skb = skbs[i];
654 
655 		skb = __xdp_build_skb_from_frame(frames[i], skb,
656 						 rq->dev);
657 		if (!skb) {
658 			xdp_return_frame(frames[i]);
659 			stats->rx_drops++;
660 			continue;
661 		}
662 		napi_gro_receive(&rq->xdp_napi, skb);
663 	}
664 }
665 
666 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
667 					struct sk_buff *skb,
668 					struct veth_xdp_tx_bq *bq,
669 					struct veth_stats *stats)
670 {
671 	u32 pktlen, headroom, act, metalen, frame_sz;
672 	void *orig_data, *orig_data_end;
673 	struct bpf_prog *xdp_prog;
674 	int mac_len, delta, off;
675 	struct xdp_buff xdp;
676 
677 	skb_orphan(skb);
678 
679 	rcu_read_lock();
680 	xdp_prog = rcu_dereference(rq->xdp_prog);
681 	if (unlikely(!xdp_prog)) {
682 		rcu_read_unlock();
683 		goto out;
684 	}
685 
686 	mac_len = skb->data - skb_mac_header(skb);
687 	pktlen = skb->len + mac_len;
688 	headroom = skb_headroom(skb) - mac_len;
689 
690 	if (skb_shared(skb) || skb_head_is_locked(skb) ||
691 	    skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
692 		struct sk_buff *nskb;
693 		int size, head_off;
694 		void *head, *start;
695 		struct page *page;
696 
697 		size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
698 		       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
699 		if (size > PAGE_SIZE)
700 			goto drop;
701 
702 		page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
703 		if (!page)
704 			goto drop;
705 
706 		head = page_address(page);
707 		start = head + VETH_XDP_HEADROOM;
708 		if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
709 			page_frag_free(head);
710 			goto drop;
711 		}
712 
713 		nskb = veth_build_skb(head, VETH_XDP_HEADROOM + mac_len,
714 				      skb->len, PAGE_SIZE);
715 		if (!nskb) {
716 			page_frag_free(head);
717 			goto drop;
718 		}
719 
720 		skb_copy_header(nskb, skb);
721 		head_off = skb_headroom(nskb) - skb_headroom(skb);
722 		skb_headers_offset_update(nskb, head_off);
723 		consume_skb(skb);
724 		skb = nskb;
725 	}
726 
727 	/* SKB "head" area always have tailroom for skb_shared_info */
728 	frame_sz = skb_end_pointer(skb) - skb->head;
729 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
730 	xdp_init_buff(&xdp, frame_sz, &rq->xdp_rxq);
731 	xdp_prepare_buff(&xdp, skb->head, skb->mac_header, pktlen, true);
732 
733 	orig_data = xdp.data;
734 	orig_data_end = xdp.data_end;
735 
736 	act = bpf_prog_run_xdp(xdp_prog, &xdp);
737 
738 	switch (act) {
739 	case XDP_PASS:
740 		break;
741 	case XDP_TX:
742 		get_page(virt_to_page(xdp.data));
743 		consume_skb(skb);
744 		xdp.rxq->mem = rq->xdp_mem;
745 		if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
746 			trace_xdp_exception(rq->dev, xdp_prog, act);
747 			stats->rx_drops++;
748 			goto err_xdp;
749 		}
750 		stats->xdp_tx++;
751 		rcu_read_unlock();
752 		goto xdp_xmit;
753 	case XDP_REDIRECT:
754 		get_page(virt_to_page(xdp.data));
755 		consume_skb(skb);
756 		xdp.rxq->mem = rq->xdp_mem;
757 		if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
758 			stats->rx_drops++;
759 			goto err_xdp;
760 		}
761 		stats->xdp_redirect++;
762 		rcu_read_unlock();
763 		goto xdp_xmit;
764 	default:
765 		bpf_warn_invalid_xdp_action(act);
766 		fallthrough;
767 	case XDP_ABORTED:
768 		trace_xdp_exception(rq->dev, xdp_prog, act);
769 		fallthrough;
770 	case XDP_DROP:
771 		stats->xdp_drops++;
772 		goto xdp_drop;
773 	}
774 	rcu_read_unlock();
775 
776 	/* check if bpf_xdp_adjust_head was used */
777 	delta = orig_data - xdp.data;
778 	off = mac_len + delta;
779 	if (off > 0)
780 		__skb_push(skb, off);
781 	else if (off < 0)
782 		__skb_pull(skb, -off);
783 	skb->mac_header -= delta;
784 
785 	/* check if bpf_xdp_adjust_tail was used */
786 	off = xdp.data_end - orig_data_end;
787 	if (off != 0)
788 		__skb_put(skb, off); /* positive on grow, negative on shrink */
789 	skb->protocol = eth_type_trans(skb, rq->dev);
790 
791 	metalen = xdp.data - xdp.data_meta;
792 	if (metalen)
793 		skb_metadata_set(skb, metalen);
794 out:
795 	return skb;
796 drop:
797 	stats->rx_drops++;
798 xdp_drop:
799 	rcu_read_unlock();
800 	kfree_skb(skb);
801 	return NULL;
802 err_xdp:
803 	rcu_read_unlock();
804 	page_frag_free(xdp.data);
805 xdp_xmit:
806 	return NULL;
807 }
808 
809 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
810 			struct veth_xdp_tx_bq *bq,
811 			struct veth_stats *stats)
812 {
813 	int i, done = 0, n_xdpf = 0;
814 	void *xdpf[VETH_XDP_BATCH];
815 
816 	for (i = 0; i < budget; i++) {
817 		void *ptr = __ptr_ring_consume(&rq->xdp_ring);
818 
819 		if (!ptr)
820 			break;
821 
822 		if (veth_is_xdp_frame(ptr)) {
823 			/* ndo_xdp_xmit */
824 			struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
825 
826 			stats->xdp_bytes += frame->len;
827 			frame = veth_xdp_rcv_one(rq, frame, bq, stats);
828 			if (frame) {
829 				/* XDP_PASS */
830 				xdpf[n_xdpf++] = frame;
831 				if (n_xdpf == VETH_XDP_BATCH) {
832 					veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf,
833 							      bq, stats);
834 					n_xdpf = 0;
835 				}
836 			}
837 		} else {
838 			/* ndo_start_xmit */
839 			struct sk_buff *skb = ptr;
840 
841 			stats->xdp_bytes += skb->len;
842 			skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
843 			if (skb)
844 				napi_gro_receive(&rq->xdp_napi, skb);
845 		}
846 		done++;
847 	}
848 
849 	if (n_xdpf)
850 		veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf, bq, stats);
851 
852 	u64_stats_update_begin(&rq->stats.syncp);
853 	rq->stats.vs.xdp_redirect += stats->xdp_redirect;
854 	rq->stats.vs.xdp_bytes += stats->xdp_bytes;
855 	rq->stats.vs.xdp_drops += stats->xdp_drops;
856 	rq->stats.vs.rx_drops += stats->rx_drops;
857 	rq->stats.vs.xdp_packets += done;
858 	u64_stats_update_end(&rq->stats.syncp);
859 
860 	return done;
861 }
862 
863 static int veth_poll(struct napi_struct *napi, int budget)
864 {
865 	struct veth_rq *rq =
866 		container_of(napi, struct veth_rq, xdp_napi);
867 	struct veth_stats stats = {};
868 	struct veth_xdp_tx_bq bq;
869 	int done;
870 
871 	bq.count = 0;
872 
873 	xdp_set_return_frame_no_direct();
874 	done = veth_xdp_rcv(rq, budget, &bq, &stats);
875 
876 	if (done < budget && napi_complete_done(napi, done)) {
877 		/* Write rx_notify_masked before reading ptr_ring */
878 		smp_store_mb(rq->rx_notify_masked, false);
879 		if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
880 			rq->rx_notify_masked = true;
881 			napi_schedule(&rq->xdp_napi);
882 		}
883 	}
884 
885 	if (stats.xdp_tx > 0)
886 		veth_xdp_flush(rq, &bq);
887 	if (stats.xdp_redirect > 0)
888 		xdp_do_flush();
889 	xdp_clear_return_frame_no_direct();
890 
891 	return done;
892 }
893 
894 static int veth_napi_add(struct net_device *dev)
895 {
896 	struct veth_priv *priv = netdev_priv(dev);
897 	int err, i;
898 
899 	for (i = 0; i < dev->real_num_rx_queues; i++) {
900 		struct veth_rq *rq = &priv->rq[i];
901 
902 		err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
903 		if (err)
904 			goto err_xdp_ring;
905 	}
906 
907 	for (i = 0; i < dev->real_num_rx_queues; i++) {
908 		struct veth_rq *rq = &priv->rq[i];
909 
910 		napi_enable(&rq->xdp_napi);
911 	}
912 
913 	return 0;
914 err_xdp_ring:
915 	for (i--; i >= 0; i--)
916 		ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
917 
918 	return err;
919 }
920 
921 static void veth_napi_del(struct net_device *dev)
922 {
923 	struct veth_priv *priv = netdev_priv(dev);
924 	int i;
925 
926 	for (i = 0; i < dev->real_num_rx_queues; i++) {
927 		struct veth_rq *rq = &priv->rq[i];
928 
929 		napi_disable(&rq->xdp_napi);
930 		__netif_napi_del(&rq->xdp_napi);
931 	}
932 	synchronize_net();
933 
934 	for (i = 0; i < dev->real_num_rx_queues; i++) {
935 		struct veth_rq *rq = &priv->rq[i];
936 
937 		rq->rx_notify_masked = false;
938 		ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
939 	}
940 }
941 
942 static int veth_enable_xdp(struct net_device *dev)
943 {
944 	struct veth_priv *priv = netdev_priv(dev);
945 	int err, i;
946 
947 	if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
948 		for (i = 0; i < dev->real_num_rx_queues; i++) {
949 			struct veth_rq *rq = &priv->rq[i];
950 
951 			netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
952 			err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i, rq->xdp_napi.napi_id);
953 			if (err < 0)
954 				goto err_rxq_reg;
955 
956 			err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
957 							 MEM_TYPE_PAGE_SHARED,
958 							 NULL);
959 			if (err < 0)
960 				goto err_reg_mem;
961 
962 			/* Save original mem info as it can be overwritten */
963 			rq->xdp_mem = rq->xdp_rxq.mem;
964 		}
965 
966 		err = veth_napi_add(dev);
967 		if (err)
968 			goto err_rxq_reg;
969 	}
970 
971 	for (i = 0; i < dev->real_num_rx_queues; i++)
972 		rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
973 
974 	return 0;
975 err_reg_mem:
976 	xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
977 err_rxq_reg:
978 	for (i--; i >= 0; i--) {
979 		struct veth_rq *rq = &priv->rq[i];
980 
981 		xdp_rxq_info_unreg(&rq->xdp_rxq);
982 		netif_napi_del(&rq->xdp_napi);
983 	}
984 
985 	return err;
986 }
987 
988 static void veth_disable_xdp(struct net_device *dev)
989 {
990 	struct veth_priv *priv = netdev_priv(dev);
991 	int i;
992 
993 	for (i = 0; i < dev->real_num_rx_queues; i++)
994 		rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
995 	veth_napi_del(dev);
996 	for (i = 0; i < dev->real_num_rx_queues; i++) {
997 		struct veth_rq *rq = &priv->rq[i];
998 
999 		rq->xdp_rxq.mem = rq->xdp_mem;
1000 		xdp_rxq_info_unreg(&rq->xdp_rxq);
1001 	}
1002 }
1003 
1004 static int veth_open(struct net_device *dev)
1005 {
1006 	struct veth_priv *priv = netdev_priv(dev);
1007 	struct net_device *peer = rtnl_dereference(priv->peer);
1008 	int err;
1009 
1010 	if (!peer)
1011 		return -ENOTCONN;
1012 
1013 	if (priv->_xdp_prog) {
1014 		err = veth_enable_xdp(dev);
1015 		if (err)
1016 			return err;
1017 	}
1018 
1019 	if (peer->flags & IFF_UP) {
1020 		netif_carrier_on(dev);
1021 		netif_carrier_on(peer);
1022 	}
1023 
1024 	return 0;
1025 }
1026 
1027 static int veth_close(struct net_device *dev)
1028 {
1029 	struct veth_priv *priv = netdev_priv(dev);
1030 	struct net_device *peer = rtnl_dereference(priv->peer);
1031 
1032 	netif_carrier_off(dev);
1033 	if (peer)
1034 		netif_carrier_off(peer);
1035 
1036 	if (priv->_xdp_prog)
1037 		veth_disable_xdp(dev);
1038 
1039 	return 0;
1040 }
1041 
1042 static int is_valid_veth_mtu(int mtu)
1043 {
1044 	return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1045 }
1046 
1047 static int veth_alloc_queues(struct net_device *dev)
1048 {
1049 	struct veth_priv *priv = netdev_priv(dev);
1050 	int i;
1051 
1052 	priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
1053 	if (!priv->rq)
1054 		return -ENOMEM;
1055 
1056 	for (i = 0; i < dev->num_rx_queues; i++) {
1057 		priv->rq[i].dev = dev;
1058 		u64_stats_init(&priv->rq[i].stats.syncp);
1059 	}
1060 
1061 	return 0;
1062 }
1063 
1064 static void veth_free_queues(struct net_device *dev)
1065 {
1066 	struct veth_priv *priv = netdev_priv(dev);
1067 
1068 	kfree(priv->rq);
1069 }
1070 
1071 static int veth_dev_init(struct net_device *dev)
1072 {
1073 	int err;
1074 
1075 	dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1076 	if (!dev->lstats)
1077 		return -ENOMEM;
1078 
1079 	err = veth_alloc_queues(dev);
1080 	if (err) {
1081 		free_percpu(dev->lstats);
1082 		return err;
1083 	}
1084 
1085 	return 0;
1086 }
1087 
1088 static void veth_dev_free(struct net_device *dev)
1089 {
1090 	veth_free_queues(dev);
1091 	free_percpu(dev->lstats);
1092 }
1093 
1094 #ifdef CONFIG_NET_POLL_CONTROLLER
1095 static void veth_poll_controller(struct net_device *dev)
1096 {
1097 	/* veth only receives frames when its peer sends one
1098 	 * Since it has nothing to do with disabling irqs, we are guaranteed
1099 	 * never to have pending data when we poll for it so
1100 	 * there is nothing to do here.
1101 	 *
1102 	 * We need this though so netpoll recognizes us as an interface that
1103 	 * supports polling, which enables bridge devices in virt setups to
1104 	 * still use netconsole
1105 	 */
1106 }
1107 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1108 
1109 static int veth_get_iflink(const struct net_device *dev)
1110 {
1111 	struct veth_priv *priv = netdev_priv(dev);
1112 	struct net_device *peer;
1113 	int iflink;
1114 
1115 	rcu_read_lock();
1116 	peer = rcu_dereference(priv->peer);
1117 	iflink = peer ? peer->ifindex : 0;
1118 	rcu_read_unlock();
1119 
1120 	return iflink;
1121 }
1122 
1123 static netdev_features_t veth_fix_features(struct net_device *dev,
1124 					   netdev_features_t features)
1125 {
1126 	struct veth_priv *priv = netdev_priv(dev);
1127 	struct net_device *peer;
1128 
1129 	peer = rtnl_dereference(priv->peer);
1130 	if (peer) {
1131 		struct veth_priv *peer_priv = netdev_priv(peer);
1132 
1133 		if (peer_priv->_xdp_prog)
1134 			features &= ~NETIF_F_GSO_SOFTWARE;
1135 	}
1136 
1137 	return features;
1138 }
1139 
1140 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1141 {
1142 	struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1143 	struct net_device *peer;
1144 
1145 	if (new_hr < 0)
1146 		new_hr = 0;
1147 
1148 	rcu_read_lock();
1149 	peer = rcu_dereference(priv->peer);
1150 	if (unlikely(!peer))
1151 		goto out;
1152 
1153 	peer_priv = netdev_priv(peer);
1154 	priv->requested_headroom = new_hr;
1155 	new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1156 	dev->needed_headroom = new_hr;
1157 	peer->needed_headroom = new_hr;
1158 
1159 out:
1160 	rcu_read_unlock();
1161 }
1162 
1163 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1164 			struct netlink_ext_ack *extack)
1165 {
1166 	struct veth_priv *priv = netdev_priv(dev);
1167 	struct bpf_prog *old_prog;
1168 	struct net_device *peer;
1169 	unsigned int max_mtu;
1170 	int err;
1171 
1172 	old_prog = priv->_xdp_prog;
1173 	priv->_xdp_prog = prog;
1174 	peer = rtnl_dereference(priv->peer);
1175 
1176 	if (prog) {
1177 		if (!peer) {
1178 			NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1179 			err = -ENOTCONN;
1180 			goto err;
1181 		}
1182 
1183 		max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1184 			  peer->hard_header_len -
1185 			  SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1186 		if (peer->mtu > max_mtu) {
1187 			NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1188 			err = -ERANGE;
1189 			goto err;
1190 		}
1191 
1192 		if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1193 			NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1194 			err = -ENOSPC;
1195 			goto err;
1196 		}
1197 
1198 		if (dev->flags & IFF_UP) {
1199 			err = veth_enable_xdp(dev);
1200 			if (err) {
1201 				NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1202 				goto err;
1203 			}
1204 		}
1205 
1206 		if (!old_prog) {
1207 			peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1208 			peer->max_mtu = max_mtu;
1209 		}
1210 	}
1211 
1212 	if (old_prog) {
1213 		if (!prog) {
1214 			if (dev->flags & IFF_UP)
1215 				veth_disable_xdp(dev);
1216 
1217 			if (peer) {
1218 				peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1219 				peer->max_mtu = ETH_MAX_MTU;
1220 			}
1221 		}
1222 		bpf_prog_put(old_prog);
1223 	}
1224 
1225 	if ((!!old_prog ^ !!prog) && peer)
1226 		netdev_update_features(peer);
1227 
1228 	return 0;
1229 err:
1230 	priv->_xdp_prog = old_prog;
1231 
1232 	return err;
1233 }
1234 
1235 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1236 {
1237 	switch (xdp->command) {
1238 	case XDP_SETUP_PROG:
1239 		return veth_xdp_set(dev, xdp->prog, xdp->extack);
1240 	default:
1241 		return -EINVAL;
1242 	}
1243 }
1244 
1245 static const struct net_device_ops veth_netdev_ops = {
1246 	.ndo_init            = veth_dev_init,
1247 	.ndo_open            = veth_open,
1248 	.ndo_stop            = veth_close,
1249 	.ndo_start_xmit      = veth_xmit,
1250 	.ndo_get_stats64     = veth_get_stats64,
1251 	.ndo_set_rx_mode     = veth_set_multicast_list,
1252 	.ndo_set_mac_address = eth_mac_addr,
1253 #ifdef CONFIG_NET_POLL_CONTROLLER
1254 	.ndo_poll_controller	= veth_poll_controller,
1255 #endif
1256 	.ndo_get_iflink		= veth_get_iflink,
1257 	.ndo_fix_features	= veth_fix_features,
1258 	.ndo_features_check	= passthru_features_check,
1259 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
1260 	.ndo_bpf		= veth_xdp,
1261 	.ndo_xdp_xmit		= veth_ndo_xdp_xmit,
1262 	.ndo_get_peer_dev	= veth_peer_dev,
1263 };
1264 
1265 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1266 		       NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1267 		       NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1268 		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1269 		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1270 
1271 static void veth_setup(struct net_device *dev)
1272 {
1273 	ether_setup(dev);
1274 
1275 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1276 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1277 	dev->priv_flags |= IFF_NO_QUEUE;
1278 	dev->priv_flags |= IFF_PHONY_HEADROOM;
1279 
1280 	dev->netdev_ops = &veth_netdev_ops;
1281 	dev->ethtool_ops = &veth_ethtool_ops;
1282 	dev->features |= NETIF_F_LLTX;
1283 	dev->features |= VETH_FEATURES;
1284 	dev->vlan_features = dev->features &
1285 			     ~(NETIF_F_HW_VLAN_CTAG_TX |
1286 			       NETIF_F_HW_VLAN_STAG_TX |
1287 			       NETIF_F_HW_VLAN_CTAG_RX |
1288 			       NETIF_F_HW_VLAN_STAG_RX);
1289 	dev->needs_free_netdev = true;
1290 	dev->priv_destructor = veth_dev_free;
1291 	dev->max_mtu = ETH_MAX_MTU;
1292 
1293 	dev->hw_features = VETH_FEATURES;
1294 	dev->hw_enc_features = VETH_FEATURES;
1295 	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1296 }
1297 
1298 /*
1299  * netlink interface
1300  */
1301 
1302 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1303 			 struct netlink_ext_ack *extack)
1304 {
1305 	if (tb[IFLA_ADDRESS]) {
1306 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1307 			return -EINVAL;
1308 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1309 			return -EADDRNOTAVAIL;
1310 	}
1311 	if (tb[IFLA_MTU]) {
1312 		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1313 			return -EINVAL;
1314 	}
1315 	return 0;
1316 }
1317 
1318 static struct rtnl_link_ops veth_link_ops;
1319 
1320 static int veth_newlink(struct net *src_net, struct net_device *dev,
1321 			struct nlattr *tb[], struct nlattr *data[],
1322 			struct netlink_ext_ack *extack)
1323 {
1324 	int err;
1325 	struct net_device *peer;
1326 	struct veth_priv *priv;
1327 	char ifname[IFNAMSIZ];
1328 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1329 	unsigned char name_assign_type;
1330 	struct ifinfomsg *ifmp;
1331 	struct net *net;
1332 
1333 	/*
1334 	 * create and register peer first
1335 	 */
1336 	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1337 		struct nlattr *nla_peer;
1338 
1339 		nla_peer = data[VETH_INFO_PEER];
1340 		ifmp = nla_data(nla_peer);
1341 		err = rtnl_nla_parse_ifla(peer_tb,
1342 					  nla_data(nla_peer) + sizeof(struct ifinfomsg),
1343 					  nla_len(nla_peer) - sizeof(struct ifinfomsg),
1344 					  NULL);
1345 		if (err < 0)
1346 			return err;
1347 
1348 		err = veth_validate(peer_tb, NULL, extack);
1349 		if (err < 0)
1350 			return err;
1351 
1352 		tbp = peer_tb;
1353 	} else {
1354 		ifmp = NULL;
1355 		tbp = tb;
1356 	}
1357 
1358 	if (ifmp && tbp[IFLA_IFNAME]) {
1359 		nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1360 		name_assign_type = NET_NAME_USER;
1361 	} else {
1362 		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1363 		name_assign_type = NET_NAME_ENUM;
1364 	}
1365 
1366 	net = rtnl_link_get_net(src_net, tbp);
1367 	if (IS_ERR(net))
1368 		return PTR_ERR(net);
1369 
1370 	peer = rtnl_create_link(net, ifname, name_assign_type,
1371 				&veth_link_ops, tbp, extack);
1372 	if (IS_ERR(peer)) {
1373 		put_net(net);
1374 		return PTR_ERR(peer);
1375 	}
1376 
1377 	if (!ifmp || !tbp[IFLA_ADDRESS])
1378 		eth_hw_addr_random(peer);
1379 
1380 	if (ifmp && (dev->ifindex != 0))
1381 		peer->ifindex = ifmp->ifi_index;
1382 
1383 	peer->gso_max_size = dev->gso_max_size;
1384 	peer->gso_max_segs = dev->gso_max_segs;
1385 
1386 	err = register_netdevice(peer);
1387 	put_net(net);
1388 	net = NULL;
1389 	if (err < 0)
1390 		goto err_register_peer;
1391 
1392 	netif_carrier_off(peer);
1393 
1394 	err = rtnl_configure_link(peer, ifmp);
1395 	if (err < 0)
1396 		goto err_configure_peer;
1397 
1398 	/*
1399 	 * register dev last
1400 	 *
1401 	 * note, that since we've registered new device the dev's name
1402 	 * should be re-allocated
1403 	 */
1404 
1405 	if (tb[IFLA_ADDRESS] == NULL)
1406 		eth_hw_addr_random(dev);
1407 
1408 	if (tb[IFLA_IFNAME])
1409 		nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1410 	else
1411 		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1412 
1413 	err = register_netdevice(dev);
1414 	if (err < 0)
1415 		goto err_register_dev;
1416 
1417 	netif_carrier_off(dev);
1418 
1419 	/*
1420 	 * tie the deviced together
1421 	 */
1422 
1423 	priv = netdev_priv(dev);
1424 	rcu_assign_pointer(priv->peer, peer);
1425 
1426 	priv = netdev_priv(peer);
1427 	rcu_assign_pointer(priv->peer, dev);
1428 
1429 	return 0;
1430 
1431 err_register_dev:
1432 	/* nothing to do */
1433 err_configure_peer:
1434 	unregister_netdevice(peer);
1435 	return err;
1436 
1437 err_register_peer:
1438 	free_netdev(peer);
1439 	return err;
1440 }
1441 
1442 static void veth_dellink(struct net_device *dev, struct list_head *head)
1443 {
1444 	struct veth_priv *priv;
1445 	struct net_device *peer;
1446 
1447 	priv = netdev_priv(dev);
1448 	peer = rtnl_dereference(priv->peer);
1449 
1450 	/* Note : dellink() is called from default_device_exit_batch(),
1451 	 * before a rcu_synchronize() point. The devices are guaranteed
1452 	 * not being freed before one RCU grace period.
1453 	 */
1454 	RCU_INIT_POINTER(priv->peer, NULL);
1455 	unregister_netdevice_queue(dev, head);
1456 
1457 	if (peer) {
1458 		priv = netdev_priv(peer);
1459 		RCU_INIT_POINTER(priv->peer, NULL);
1460 		unregister_netdevice_queue(peer, head);
1461 	}
1462 }
1463 
1464 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1465 	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
1466 };
1467 
1468 static struct net *veth_get_link_net(const struct net_device *dev)
1469 {
1470 	struct veth_priv *priv = netdev_priv(dev);
1471 	struct net_device *peer = rtnl_dereference(priv->peer);
1472 
1473 	return peer ? dev_net(peer) : dev_net(dev);
1474 }
1475 
1476 static struct rtnl_link_ops veth_link_ops = {
1477 	.kind		= DRV_NAME,
1478 	.priv_size	= sizeof(struct veth_priv),
1479 	.setup		= veth_setup,
1480 	.validate	= veth_validate,
1481 	.newlink	= veth_newlink,
1482 	.dellink	= veth_dellink,
1483 	.policy		= veth_policy,
1484 	.maxtype	= VETH_INFO_MAX,
1485 	.get_link_net	= veth_get_link_net,
1486 };
1487 
1488 /*
1489  * init/fini
1490  */
1491 
1492 static __init int veth_init(void)
1493 {
1494 	return rtnl_link_register(&veth_link_ops);
1495 }
1496 
1497 static __exit void veth_exit(void)
1498 {
1499 	rtnl_link_unregister(&veth_link_ops);
1500 }
1501 
1502 module_init(veth_init);
1503 module_exit(veth_exit);
1504 
1505 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1506 MODULE_LICENSE("GPL v2");
1507 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1508