xref: /openbmc/linux/drivers/net/veth.c (revision 6a143a7c)
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 		if (rcv_xdp)
306 			skb_record_rx_queue(skb, rxq);
307 	}
308 
309 	skb_tx_timestamp(skb);
310 	if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
311 		if (!rcv_xdp)
312 			dev_lstats_add(dev, length);
313 	} else {
314 drop:
315 		atomic64_inc(&priv->dropped);
316 	}
317 
318 	if (rcv_xdp)
319 		__veth_xdp_flush(rq);
320 
321 	rcu_read_unlock();
322 
323 	return NETDEV_TX_OK;
324 }
325 
326 static u64 veth_stats_tx(struct net_device *dev, u64 *packets, u64 *bytes)
327 {
328 	struct veth_priv *priv = netdev_priv(dev);
329 
330 	dev_lstats_read(dev, packets, bytes);
331 	return atomic64_read(&priv->dropped);
332 }
333 
334 static void veth_stats_rx(struct veth_stats *result, struct net_device *dev)
335 {
336 	struct veth_priv *priv = netdev_priv(dev);
337 	int i;
338 
339 	result->peer_tq_xdp_xmit_err = 0;
340 	result->xdp_packets = 0;
341 	result->xdp_tx_err = 0;
342 	result->xdp_bytes = 0;
343 	result->rx_drops = 0;
344 	for (i = 0; i < dev->num_rx_queues; i++) {
345 		u64 packets, bytes, drops, xdp_tx_err, peer_tq_xdp_xmit_err;
346 		struct veth_rq_stats *stats = &priv->rq[i].stats;
347 		unsigned int start;
348 
349 		do {
350 			start = u64_stats_fetch_begin_irq(&stats->syncp);
351 			peer_tq_xdp_xmit_err = stats->vs.peer_tq_xdp_xmit_err;
352 			xdp_tx_err = stats->vs.xdp_tx_err;
353 			packets = stats->vs.xdp_packets;
354 			bytes = stats->vs.xdp_bytes;
355 			drops = stats->vs.rx_drops;
356 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
357 		result->peer_tq_xdp_xmit_err += peer_tq_xdp_xmit_err;
358 		result->xdp_tx_err += xdp_tx_err;
359 		result->xdp_packets += packets;
360 		result->xdp_bytes += bytes;
361 		result->rx_drops += drops;
362 	}
363 }
364 
365 static void veth_get_stats64(struct net_device *dev,
366 			     struct rtnl_link_stats64 *tot)
367 {
368 	struct veth_priv *priv = netdev_priv(dev);
369 	struct net_device *peer;
370 	struct veth_stats rx;
371 	u64 packets, bytes;
372 
373 	tot->tx_dropped = veth_stats_tx(dev, &packets, &bytes);
374 	tot->tx_bytes = bytes;
375 	tot->tx_packets = packets;
376 
377 	veth_stats_rx(&rx, dev);
378 	tot->tx_dropped += rx.xdp_tx_err;
379 	tot->rx_dropped = rx.rx_drops + rx.peer_tq_xdp_xmit_err;
380 	tot->rx_bytes = rx.xdp_bytes;
381 	tot->rx_packets = rx.xdp_packets;
382 
383 	rcu_read_lock();
384 	peer = rcu_dereference(priv->peer);
385 	if (peer) {
386 		veth_stats_tx(peer, &packets, &bytes);
387 		tot->rx_bytes += bytes;
388 		tot->rx_packets += packets;
389 
390 		veth_stats_rx(&rx, peer);
391 		tot->tx_dropped += rx.peer_tq_xdp_xmit_err;
392 		tot->rx_dropped += rx.xdp_tx_err;
393 		tot->tx_bytes += rx.xdp_bytes;
394 		tot->tx_packets += rx.xdp_packets;
395 	}
396 	rcu_read_unlock();
397 }
398 
399 /* fake multicast ability */
400 static void veth_set_multicast_list(struct net_device *dev)
401 {
402 }
403 
404 static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
405 				      int buflen)
406 {
407 	struct sk_buff *skb;
408 
409 	skb = build_skb(head, buflen);
410 	if (!skb)
411 		return NULL;
412 
413 	skb_reserve(skb, headroom);
414 	skb_put(skb, len);
415 
416 	return skb;
417 }
418 
419 static int veth_select_rxq(struct net_device *dev)
420 {
421 	return smp_processor_id() % dev->real_num_rx_queues;
422 }
423 
424 static struct net_device *veth_peer_dev(struct net_device *dev)
425 {
426 	struct veth_priv *priv = netdev_priv(dev);
427 
428 	/* Callers must be under RCU read side. */
429 	return rcu_dereference(priv->peer);
430 }
431 
432 static int veth_xdp_xmit(struct net_device *dev, int n,
433 			 struct xdp_frame **frames,
434 			 u32 flags, bool ndo_xmit)
435 {
436 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
437 	int i, ret = -ENXIO, drops = 0;
438 	struct net_device *rcv;
439 	unsigned int max_len;
440 	struct veth_rq *rq;
441 
442 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
443 		return -EINVAL;
444 
445 	rcu_read_lock();
446 	rcv = rcu_dereference(priv->peer);
447 	if (unlikely(!rcv))
448 		goto out;
449 
450 	rcv_priv = netdev_priv(rcv);
451 	rq = &rcv_priv->rq[veth_select_rxq(rcv)];
452 	/* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive
453 	 * side. This means an XDP program is loaded on the peer and the peer
454 	 * device is up.
455 	 */
456 	if (!rcu_access_pointer(rq->xdp_prog))
457 		goto out;
458 
459 	max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
460 
461 	spin_lock(&rq->xdp_ring.producer_lock);
462 	for (i = 0; i < n; i++) {
463 		struct xdp_frame *frame = frames[i];
464 		void *ptr = veth_xdp_to_ptr(frame);
465 
466 		if (unlikely(frame->len > max_len ||
467 			     __ptr_ring_produce(&rq->xdp_ring, ptr))) {
468 			xdp_return_frame_rx_napi(frame);
469 			drops++;
470 		}
471 	}
472 	spin_unlock(&rq->xdp_ring.producer_lock);
473 
474 	if (flags & XDP_XMIT_FLUSH)
475 		__veth_xdp_flush(rq);
476 
477 	ret = n - drops;
478 	if (ndo_xmit) {
479 		u64_stats_update_begin(&rq->stats.syncp);
480 		rq->stats.vs.peer_tq_xdp_xmit += n - drops;
481 		rq->stats.vs.peer_tq_xdp_xmit_err += drops;
482 		u64_stats_update_end(&rq->stats.syncp);
483 	}
484 
485 out:
486 	rcu_read_unlock();
487 
488 	return ret;
489 }
490 
491 static int veth_ndo_xdp_xmit(struct net_device *dev, int n,
492 			     struct xdp_frame **frames, u32 flags)
493 {
494 	int err;
495 
496 	err = veth_xdp_xmit(dev, n, frames, flags, true);
497 	if (err < 0) {
498 		struct veth_priv *priv = netdev_priv(dev);
499 
500 		atomic64_add(n, &priv->dropped);
501 	}
502 
503 	return err;
504 }
505 
506 static void veth_xdp_flush_bq(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
507 {
508 	int sent, i, err = 0;
509 
510 	sent = veth_xdp_xmit(rq->dev, bq->count, bq->q, 0, false);
511 	if (sent < 0) {
512 		err = sent;
513 		sent = 0;
514 		for (i = 0; i < bq->count; i++)
515 			xdp_return_frame(bq->q[i]);
516 	}
517 	trace_xdp_bulk_tx(rq->dev, sent, bq->count - sent, err);
518 
519 	u64_stats_update_begin(&rq->stats.syncp);
520 	rq->stats.vs.xdp_tx += sent;
521 	rq->stats.vs.xdp_tx_err += bq->count - sent;
522 	u64_stats_update_end(&rq->stats.syncp);
523 
524 	bq->count = 0;
525 }
526 
527 static void veth_xdp_flush(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
528 {
529 	struct veth_priv *rcv_priv, *priv = netdev_priv(rq->dev);
530 	struct net_device *rcv;
531 	struct veth_rq *rcv_rq;
532 
533 	rcu_read_lock();
534 	veth_xdp_flush_bq(rq, bq);
535 	rcv = rcu_dereference(priv->peer);
536 	if (unlikely(!rcv))
537 		goto out;
538 
539 	rcv_priv = netdev_priv(rcv);
540 	rcv_rq = &rcv_priv->rq[veth_select_rxq(rcv)];
541 	/* xdp_ring is initialized on receive side? */
542 	if (unlikely(!rcu_access_pointer(rcv_rq->xdp_prog)))
543 		goto out;
544 
545 	__veth_xdp_flush(rcv_rq);
546 out:
547 	rcu_read_unlock();
548 }
549 
550 static int veth_xdp_tx(struct veth_rq *rq, struct xdp_buff *xdp,
551 		       struct veth_xdp_tx_bq *bq)
552 {
553 	struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
554 
555 	if (unlikely(!frame))
556 		return -EOVERFLOW;
557 
558 	if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
559 		veth_xdp_flush_bq(rq, bq);
560 
561 	bq->q[bq->count++] = frame;
562 
563 	return 0;
564 }
565 
566 static struct xdp_frame *veth_xdp_rcv_one(struct veth_rq *rq,
567 					  struct xdp_frame *frame,
568 					  struct veth_xdp_tx_bq *bq,
569 					  struct veth_stats *stats)
570 {
571 	struct xdp_frame orig_frame;
572 	struct bpf_prog *xdp_prog;
573 
574 	rcu_read_lock();
575 	xdp_prog = rcu_dereference(rq->xdp_prog);
576 	if (likely(xdp_prog)) {
577 		struct xdp_buff xdp;
578 		u32 act;
579 
580 		xdp_convert_frame_to_buff(frame, &xdp);
581 		xdp.rxq = &rq->xdp_rxq;
582 
583 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
584 
585 		switch (act) {
586 		case XDP_PASS:
587 			if (xdp_update_frame_from_buff(&xdp, frame))
588 				goto err_xdp;
589 			break;
590 		case XDP_TX:
591 			orig_frame = *frame;
592 			xdp.rxq->mem = frame->mem;
593 			if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
594 				trace_xdp_exception(rq->dev, xdp_prog, act);
595 				frame = &orig_frame;
596 				stats->rx_drops++;
597 				goto err_xdp;
598 			}
599 			stats->xdp_tx++;
600 			rcu_read_unlock();
601 			goto xdp_xmit;
602 		case XDP_REDIRECT:
603 			orig_frame = *frame;
604 			xdp.rxq->mem = frame->mem;
605 			if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
606 				frame = &orig_frame;
607 				stats->rx_drops++;
608 				goto err_xdp;
609 			}
610 			stats->xdp_redirect++;
611 			rcu_read_unlock();
612 			goto xdp_xmit;
613 		default:
614 			bpf_warn_invalid_xdp_action(act);
615 			fallthrough;
616 		case XDP_ABORTED:
617 			trace_xdp_exception(rq->dev, xdp_prog, act);
618 			fallthrough;
619 		case XDP_DROP:
620 			stats->xdp_drops++;
621 			goto err_xdp;
622 		}
623 	}
624 	rcu_read_unlock();
625 
626 	return frame;
627 err_xdp:
628 	rcu_read_unlock();
629 	xdp_return_frame(frame);
630 xdp_xmit:
631 	return NULL;
632 }
633 
634 /* frames array contains VETH_XDP_BATCH at most */
635 static void veth_xdp_rcv_bulk_skb(struct veth_rq *rq, void **frames,
636 				  int n_xdpf, struct veth_xdp_tx_bq *bq,
637 				  struct veth_stats *stats)
638 {
639 	void *skbs[VETH_XDP_BATCH];
640 	int i;
641 
642 	if (xdp_alloc_skb_bulk(skbs, n_xdpf,
643 			       GFP_ATOMIC | __GFP_ZERO) < 0) {
644 		for (i = 0; i < n_xdpf; i++)
645 			xdp_return_frame(frames[i]);
646 		stats->rx_drops += n_xdpf;
647 
648 		return;
649 	}
650 
651 	for (i = 0; i < n_xdpf; i++) {
652 		struct sk_buff *skb = skbs[i];
653 
654 		skb = __xdp_build_skb_from_frame(frames[i], skb,
655 						 rq->dev);
656 		if (!skb) {
657 			xdp_return_frame(frames[i]);
658 			stats->rx_drops++;
659 			continue;
660 		}
661 		napi_gro_receive(&rq->xdp_napi, skb);
662 	}
663 }
664 
665 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
666 					struct sk_buff *skb,
667 					struct veth_xdp_tx_bq *bq,
668 					struct veth_stats *stats)
669 {
670 	u32 pktlen, headroom, act, metalen, frame_sz;
671 	void *orig_data, *orig_data_end;
672 	struct bpf_prog *xdp_prog;
673 	int mac_len, delta, off;
674 	struct xdp_buff xdp;
675 
676 	skb_orphan(skb);
677 
678 	rcu_read_lock();
679 	xdp_prog = rcu_dereference(rq->xdp_prog);
680 	if (unlikely(!xdp_prog)) {
681 		rcu_read_unlock();
682 		goto out;
683 	}
684 
685 	mac_len = skb->data - skb_mac_header(skb);
686 	pktlen = skb->len + mac_len;
687 	headroom = skb_headroom(skb) - mac_len;
688 
689 	if (skb_shared(skb) || skb_head_is_locked(skb) ||
690 	    skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
691 		struct sk_buff *nskb;
692 		int size, head_off;
693 		void *head, *start;
694 		struct page *page;
695 
696 		size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
697 		       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
698 		if (size > PAGE_SIZE)
699 			goto drop;
700 
701 		page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
702 		if (!page)
703 			goto drop;
704 
705 		head = page_address(page);
706 		start = head + VETH_XDP_HEADROOM;
707 		if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
708 			page_frag_free(head);
709 			goto drop;
710 		}
711 
712 		nskb = veth_build_skb(head, VETH_XDP_HEADROOM + mac_len,
713 				      skb->len, PAGE_SIZE);
714 		if (!nskb) {
715 			page_frag_free(head);
716 			goto drop;
717 		}
718 
719 		skb_copy_header(nskb, skb);
720 		head_off = skb_headroom(nskb) - skb_headroom(skb);
721 		skb_headers_offset_update(nskb, head_off);
722 		consume_skb(skb);
723 		skb = nskb;
724 	}
725 
726 	/* SKB "head" area always have tailroom for skb_shared_info */
727 	frame_sz = skb_end_pointer(skb) - skb->head;
728 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
729 	xdp_init_buff(&xdp, frame_sz, &rq->xdp_rxq);
730 	xdp_prepare_buff(&xdp, skb->head, skb->mac_header, pktlen, true);
731 
732 	orig_data = xdp.data;
733 	orig_data_end = xdp.data_end;
734 
735 	act = bpf_prog_run_xdp(xdp_prog, &xdp);
736 
737 	switch (act) {
738 	case XDP_PASS:
739 		break;
740 	case XDP_TX:
741 		get_page(virt_to_page(xdp.data));
742 		consume_skb(skb);
743 		xdp.rxq->mem = rq->xdp_mem;
744 		if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
745 			trace_xdp_exception(rq->dev, xdp_prog, act);
746 			stats->rx_drops++;
747 			goto err_xdp;
748 		}
749 		stats->xdp_tx++;
750 		rcu_read_unlock();
751 		goto xdp_xmit;
752 	case XDP_REDIRECT:
753 		get_page(virt_to_page(xdp.data));
754 		consume_skb(skb);
755 		xdp.rxq->mem = rq->xdp_mem;
756 		if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
757 			stats->rx_drops++;
758 			goto err_xdp;
759 		}
760 		stats->xdp_redirect++;
761 		rcu_read_unlock();
762 		goto xdp_xmit;
763 	default:
764 		bpf_warn_invalid_xdp_action(act);
765 		fallthrough;
766 	case XDP_ABORTED:
767 		trace_xdp_exception(rq->dev, xdp_prog, act);
768 		fallthrough;
769 	case XDP_DROP:
770 		stats->xdp_drops++;
771 		goto xdp_drop;
772 	}
773 	rcu_read_unlock();
774 
775 	/* check if bpf_xdp_adjust_head was used */
776 	delta = orig_data - xdp.data;
777 	off = mac_len + delta;
778 	if (off > 0)
779 		__skb_push(skb, off);
780 	else if (off < 0)
781 		__skb_pull(skb, -off);
782 	skb->mac_header -= delta;
783 
784 	/* check if bpf_xdp_adjust_tail was used */
785 	off = xdp.data_end - orig_data_end;
786 	if (off != 0)
787 		__skb_put(skb, off); /* positive on grow, negative on shrink */
788 	skb->protocol = eth_type_trans(skb, rq->dev);
789 
790 	metalen = xdp.data - xdp.data_meta;
791 	if (metalen)
792 		skb_metadata_set(skb, metalen);
793 out:
794 	return skb;
795 drop:
796 	stats->rx_drops++;
797 xdp_drop:
798 	rcu_read_unlock();
799 	kfree_skb(skb);
800 	return NULL;
801 err_xdp:
802 	rcu_read_unlock();
803 	page_frag_free(xdp.data);
804 xdp_xmit:
805 	return NULL;
806 }
807 
808 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
809 			struct veth_xdp_tx_bq *bq,
810 			struct veth_stats *stats)
811 {
812 	int i, done = 0, n_xdpf = 0;
813 	void *xdpf[VETH_XDP_BATCH];
814 
815 	for (i = 0; i < budget; i++) {
816 		void *ptr = __ptr_ring_consume(&rq->xdp_ring);
817 
818 		if (!ptr)
819 			break;
820 
821 		if (veth_is_xdp_frame(ptr)) {
822 			/* ndo_xdp_xmit */
823 			struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
824 
825 			stats->xdp_bytes += frame->len;
826 			frame = veth_xdp_rcv_one(rq, frame, bq, stats);
827 			if (frame) {
828 				/* XDP_PASS */
829 				xdpf[n_xdpf++] = frame;
830 				if (n_xdpf == VETH_XDP_BATCH) {
831 					veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf,
832 							      bq, stats);
833 					n_xdpf = 0;
834 				}
835 			}
836 		} else {
837 			/* ndo_start_xmit */
838 			struct sk_buff *skb = ptr;
839 
840 			stats->xdp_bytes += skb->len;
841 			skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
842 			if (skb)
843 				napi_gro_receive(&rq->xdp_napi, skb);
844 		}
845 		done++;
846 	}
847 
848 	if (n_xdpf)
849 		veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf, bq, stats);
850 
851 	u64_stats_update_begin(&rq->stats.syncp);
852 	rq->stats.vs.xdp_redirect += stats->xdp_redirect;
853 	rq->stats.vs.xdp_bytes += stats->xdp_bytes;
854 	rq->stats.vs.xdp_drops += stats->xdp_drops;
855 	rq->stats.vs.rx_drops += stats->rx_drops;
856 	rq->stats.vs.xdp_packets += done;
857 	u64_stats_update_end(&rq->stats.syncp);
858 
859 	return done;
860 }
861 
862 static int veth_poll(struct napi_struct *napi, int budget)
863 {
864 	struct veth_rq *rq =
865 		container_of(napi, struct veth_rq, xdp_napi);
866 	struct veth_stats stats = {};
867 	struct veth_xdp_tx_bq bq;
868 	int done;
869 
870 	bq.count = 0;
871 
872 	xdp_set_return_frame_no_direct();
873 	done = veth_xdp_rcv(rq, budget, &bq, &stats);
874 
875 	if (done < budget && napi_complete_done(napi, done)) {
876 		/* Write rx_notify_masked before reading ptr_ring */
877 		smp_store_mb(rq->rx_notify_masked, false);
878 		if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
879 			rq->rx_notify_masked = true;
880 			napi_schedule(&rq->xdp_napi);
881 		}
882 	}
883 
884 	if (stats.xdp_tx > 0)
885 		veth_xdp_flush(rq, &bq);
886 	if (stats.xdp_redirect > 0)
887 		xdp_do_flush();
888 	xdp_clear_return_frame_no_direct();
889 
890 	return done;
891 }
892 
893 static int veth_napi_add(struct net_device *dev)
894 {
895 	struct veth_priv *priv = netdev_priv(dev);
896 	int err, i;
897 
898 	for (i = 0; i < dev->real_num_rx_queues; i++) {
899 		struct veth_rq *rq = &priv->rq[i];
900 
901 		err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
902 		if (err)
903 			goto err_xdp_ring;
904 	}
905 
906 	for (i = 0; i < dev->real_num_rx_queues; i++) {
907 		struct veth_rq *rq = &priv->rq[i];
908 
909 		napi_enable(&rq->xdp_napi);
910 	}
911 
912 	return 0;
913 err_xdp_ring:
914 	for (i--; i >= 0; i--)
915 		ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
916 
917 	return err;
918 }
919 
920 static void veth_napi_del(struct net_device *dev)
921 {
922 	struct veth_priv *priv = netdev_priv(dev);
923 	int i;
924 
925 	for (i = 0; i < dev->real_num_rx_queues; i++) {
926 		struct veth_rq *rq = &priv->rq[i];
927 
928 		napi_disable(&rq->xdp_napi);
929 		__netif_napi_del(&rq->xdp_napi);
930 	}
931 	synchronize_net();
932 
933 	for (i = 0; i < dev->real_num_rx_queues; i++) {
934 		struct veth_rq *rq = &priv->rq[i];
935 
936 		rq->rx_notify_masked = false;
937 		ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
938 	}
939 }
940 
941 static int veth_enable_xdp(struct net_device *dev)
942 {
943 	struct veth_priv *priv = netdev_priv(dev);
944 	int err, i;
945 
946 	if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
947 		for (i = 0; i < dev->real_num_rx_queues; i++) {
948 			struct veth_rq *rq = &priv->rq[i];
949 
950 			netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
951 			err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i, rq->xdp_napi.napi_id);
952 			if (err < 0)
953 				goto err_rxq_reg;
954 
955 			err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
956 							 MEM_TYPE_PAGE_SHARED,
957 							 NULL);
958 			if (err < 0)
959 				goto err_reg_mem;
960 
961 			/* Save original mem info as it can be overwritten */
962 			rq->xdp_mem = rq->xdp_rxq.mem;
963 		}
964 
965 		err = veth_napi_add(dev);
966 		if (err)
967 			goto err_rxq_reg;
968 	}
969 
970 	for (i = 0; i < dev->real_num_rx_queues; i++)
971 		rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
972 
973 	return 0;
974 err_reg_mem:
975 	xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
976 err_rxq_reg:
977 	for (i--; i >= 0; i--) {
978 		struct veth_rq *rq = &priv->rq[i];
979 
980 		xdp_rxq_info_unreg(&rq->xdp_rxq);
981 		netif_napi_del(&rq->xdp_napi);
982 	}
983 
984 	return err;
985 }
986 
987 static void veth_disable_xdp(struct net_device *dev)
988 {
989 	struct veth_priv *priv = netdev_priv(dev);
990 	int i;
991 
992 	for (i = 0; i < dev->real_num_rx_queues; i++)
993 		rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
994 	veth_napi_del(dev);
995 	for (i = 0; i < dev->real_num_rx_queues; i++) {
996 		struct veth_rq *rq = &priv->rq[i];
997 
998 		rq->xdp_rxq.mem = rq->xdp_mem;
999 		xdp_rxq_info_unreg(&rq->xdp_rxq);
1000 	}
1001 }
1002 
1003 static int veth_open(struct net_device *dev)
1004 {
1005 	struct veth_priv *priv = netdev_priv(dev);
1006 	struct net_device *peer = rtnl_dereference(priv->peer);
1007 	int err;
1008 
1009 	if (!peer)
1010 		return -ENOTCONN;
1011 
1012 	if (priv->_xdp_prog) {
1013 		err = veth_enable_xdp(dev);
1014 		if (err)
1015 			return err;
1016 	}
1017 
1018 	if (peer->flags & IFF_UP) {
1019 		netif_carrier_on(dev);
1020 		netif_carrier_on(peer);
1021 	}
1022 
1023 	return 0;
1024 }
1025 
1026 static int veth_close(struct net_device *dev)
1027 {
1028 	struct veth_priv *priv = netdev_priv(dev);
1029 	struct net_device *peer = rtnl_dereference(priv->peer);
1030 
1031 	netif_carrier_off(dev);
1032 	if (peer)
1033 		netif_carrier_off(peer);
1034 
1035 	if (priv->_xdp_prog)
1036 		veth_disable_xdp(dev);
1037 
1038 	return 0;
1039 }
1040 
1041 static int is_valid_veth_mtu(int mtu)
1042 {
1043 	return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1044 }
1045 
1046 static int veth_alloc_queues(struct net_device *dev)
1047 {
1048 	struct veth_priv *priv = netdev_priv(dev);
1049 	int i;
1050 
1051 	priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
1052 	if (!priv->rq)
1053 		return -ENOMEM;
1054 
1055 	for (i = 0; i < dev->num_rx_queues; i++) {
1056 		priv->rq[i].dev = dev;
1057 		u64_stats_init(&priv->rq[i].stats.syncp);
1058 	}
1059 
1060 	return 0;
1061 }
1062 
1063 static void veth_free_queues(struct net_device *dev)
1064 {
1065 	struct veth_priv *priv = netdev_priv(dev);
1066 
1067 	kfree(priv->rq);
1068 }
1069 
1070 static int veth_dev_init(struct net_device *dev)
1071 {
1072 	int err;
1073 
1074 	dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1075 	if (!dev->lstats)
1076 		return -ENOMEM;
1077 
1078 	err = veth_alloc_queues(dev);
1079 	if (err) {
1080 		free_percpu(dev->lstats);
1081 		return err;
1082 	}
1083 
1084 	return 0;
1085 }
1086 
1087 static void veth_dev_free(struct net_device *dev)
1088 {
1089 	veth_free_queues(dev);
1090 	free_percpu(dev->lstats);
1091 }
1092 
1093 #ifdef CONFIG_NET_POLL_CONTROLLER
1094 static void veth_poll_controller(struct net_device *dev)
1095 {
1096 	/* veth only receives frames when its peer sends one
1097 	 * Since it has nothing to do with disabling irqs, we are guaranteed
1098 	 * never to have pending data when we poll for it so
1099 	 * there is nothing to do here.
1100 	 *
1101 	 * We need this though so netpoll recognizes us as an interface that
1102 	 * supports polling, which enables bridge devices in virt setups to
1103 	 * still use netconsole
1104 	 */
1105 }
1106 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1107 
1108 static int veth_get_iflink(const struct net_device *dev)
1109 {
1110 	struct veth_priv *priv = netdev_priv(dev);
1111 	struct net_device *peer;
1112 	int iflink;
1113 
1114 	rcu_read_lock();
1115 	peer = rcu_dereference(priv->peer);
1116 	iflink = peer ? peer->ifindex : 0;
1117 	rcu_read_unlock();
1118 
1119 	return iflink;
1120 }
1121 
1122 static netdev_features_t veth_fix_features(struct net_device *dev,
1123 					   netdev_features_t features)
1124 {
1125 	struct veth_priv *priv = netdev_priv(dev);
1126 	struct net_device *peer;
1127 
1128 	peer = rtnl_dereference(priv->peer);
1129 	if (peer) {
1130 		struct veth_priv *peer_priv = netdev_priv(peer);
1131 
1132 		if (peer_priv->_xdp_prog)
1133 			features &= ~NETIF_F_GSO_SOFTWARE;
1134 	}
1135 
1136 	return features;
1137 }
1138 
1139 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1140 {
1141 	struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1142 	struct net_device *peer;
1143 
1144 	if (new_hr < 0)
1145 		new_hr = 0;
1146 
1147 	rcu_read_lock();
1148 	peer = rcu_dereference(priv->peer);
1149 	if (unlikely(!peer))
1150 		goto out;
1151 
1152 	peer_priv = netdev_priv(peer);
1153 	priv->requested_headroom = new_hr;
1154 	new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1155 	dev->needed_headroom = new_hr;
1156 	peer->needed_headroom = new_hr;
1157 
1158 out:
1159 	rcu_read_unlock();
1160 }
1161 
1162 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1163 			struct netlink_ext_ack *extack)
1164 {
1165 	struct veth_priv *priv = netdev_priv(dev);
1166 	struct bpf_prog *old_prog;
1167 	struct net_device *peer;
1168 	unsigned int max_mtu;
1169 	int err;
1170 
1171 	old_prog = priv->_xdp_prog;
1172 	priv->_xdp_prog = prog;
1173 	peer = rtnl_dereference(priv->peer);
1174 
1175 	if (prog) {
1176 		if (!peer) {
1177 			NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1178 			err = -ENOTCONN;
1179 			goto err;
1180 		}
1181 
1182 		max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1183 			  peer->hard_header_len -
1184 			  SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1185 		if (peer->mtu > max_mtu) {
1186 			NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1187 			err = -ERANGE;
1188 			goto err;
1189 		}
1190 
1191 		if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1192 			NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1193 			err = -ENOSPC;
1194 			goto err;
1195 		}
1196 
1197 		if (dev->flags & IFF_UP) {
1198 			err = veth_enable_xdp(dev);
1199 			if (err) {
1200 				NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1201 				goto err;
1202 			}
1203 		}
1204 
1205 		if (!old_prog) {
1206 			peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1207 			peer->max_mtu = max_mtu;
1208 		}
1209 	}
1210 
1211 	if (old_prog) {
1212 		if (!prog) {
1213 			if (dev->flags & IFF_UP)
1214 				veth_disable_xdp(dev);
1215 
1216 			if (peer) {
1217 				peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1218 				peer->max_mtu = ETH_MAX_MTU;
1219 			}
1220 		}
1221 		bpf_prog_put(old_prog);
1222 	}
1223 
1224 	if ((!!old_prog ^ !!prog) && peer)
1225 		netdev_update_features(peer);
1226 
1227 	return 0;
1228 err:
1229 	priv->_xdp_prog = old_prog;
1230 
1231 	return err;
1232 }
1233 
1234 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1235 {
1236 	switch (xdp->command) {
1237 	case XDP_SETUP_PROG:
1238 		return veth_xdp_set(dev, xdp->prog, xdp->extack);
1239 	default:
1240 		return -EINVAL;
1241 	}
1242 }
1243 
1244 static const struct net_device_ops veth_netdev_ops = {
1245 	.ndo_init            = veth_dev_init,
1246 	.ndo_open            = veth_open,
1247 	.ndo_stop            = veth_close,
1248 	.ndo_start_xmit      = veth_xmit,
1249 	.ndo_get_stats64     = veth_get_stats64,
1250 	.ndo_set_rx_mode     = veth_set_multicast_list,
1251 	.ndo_set_mac_address = eth_mac_addr,
1252 #ifdef CONFIG_NET_POLL_CONTROLLER
1253 	.ndo_poll_controller	= veth_poll_controller,
1254 #endif
1255 	.ndo_get_iflink		= veth_get_iflink,
1256 	.ndo_fix_features	= veth_fix_features,
1257 	.ndo_features_check	= passthru_features_check,
1258 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
1259 	.ndo_bpf		= veth_xdp,
1260 	.ndo_xdp_xmit		= veth_ndo_xdp_xmit,
1261 	.ndo_get_peer_dev	= veth_peer_dev,
1262 };
1263 
1264 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1265 		       NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1266 		       NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1267 		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1268 		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1269 
1270 static void veth_setup(struct net_device *dev)
1271 {
1272 	ether_setup(dev);
1273 
1274 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1275 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1276 	dev->priv_flags |= IFF_NO_QUEUE;
1277 	dev->priv_flags |= IFF_PHONY_HEADROOM;
1278 
1279 	dev->netdev_ops = &veth_netdev_ops;
1280 	dev->ethtool_ops = &veth_ethtool_ops;
1281 	dev->features |= NETIF_F_LLTX;
1282 	dev->features |= VETH_FEATURES;
1283 	dev->vlan_features = dev->features &
1284 			     ~(NETIF_F_HW_VLAN_CTAG_TX |
1285 			       NETIF_F_HW_VLAN_STAG_TX |
1286 			       NETIF_F_HW_VLAN_CTAG_RX |
1287 			       NETIF_F_HW_VLAN_STAG_RX);
1288 	dev->needs_free_netdev = true;
1289 	dev->priv_destructor = veth_dev_free;
1290 	dev->max_mtu = ETH_MAX_MTU;
1291 
1292 	dev->hw_features = VETH_FEATURES;
1293 	dev->hw_enc_features = VETH_FEATURES;
1294 	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1295 }
1296 
1297 /*
1298  * netlink interface
1299  */
1300 
1301 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1302 			 struct netlink_ext_ack *extack)
1303 {
1304 	if (tb[IFLA_ADDRESS]) {
1305 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1306 			return -EINVAL;
1307 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1308 			return -EADDRNOTAVAIL;
1309 	}
1310 	if (tb[IFLA_MTU]) {
1311 		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1312 			return -EINVAL;
1313 	}
1314 	return 0;
1315 }
1316 
1317 static struct rtnl_link_ops veth_link_ops;
1318 
1319 static int veth_newlink(struct net *src_net, struct net_device *dev,
1320 			struct nlattr *tb[], struct nlattr *data[],
1321 			struct netlink_ext_ack *extack)
1322 {
1323 	int err;
1324 	struct net_device *peer;
1325 	struct veth_priv *priv;
1326 	char ifname[IFNAMSIZ];
1327 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1328 	unsigned char name_assign_type;
1329 	struct ifinfomsg *ifmp;
1330 	struct net *net;
1331 
1332 	/*
1333 	 * create and register peer first
1334 	 */
1335 	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1336 		struct nlattr *nla_peer;
1337 
1338 		nla_peer = data[VETH_INFO_PEER];
1339 		ifmp = nla_data(nla_peer);
1340 		err = rtnl_nla_parse_ifla(peer_tb,
1341 					  nla_data(nla_peer) + sizeof(struct ifinfomsg),
1342 					  nla_len(nla_peer) - sizeof(struct ifinfomsg),
1343 					  NULL);
1344 		if (err < 0)
1345 			return err;
1346 
1347 		err = veth_validate(peer_tb, NULL, extack);
1348 		if (err < 0)
1349 			return err;
1350 
1351 		tbp = peer_tb;
1352 	} else {
1353 		ifmp = NULL;
1354 		tbp = tb;
1355 	}
1356 
1357 	if (ifmp && tbp[IFLA_IFNAME]) {
1358 		nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1359 		name_assign_type = NET_NAME_USER;
1360 	} else {
1361 		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1362 		name_assign_type = NET_NAME_ENUM;
1363 	}
1364 
1365 	net = rtnl_link_get_net(src_net, tbp);
1366 	if (IS_ERR(net))
1367 		return PTR_ERR(net);
1368 
1369 	peer = rtnl_create_link(net, ifname, name_assign_type,
1370 				&veth_link_ops, tbp, extack);
1371 	if (IS_ERR(peer)) {
1372 		put_net(net);
1373 		return PTR_ERR(peer);
1374 	}
1375 
1376 	if (!ifmp || !tbp[IFLA_ADDRESS])
1377 		eth_hw_addr_random(peer);
1378 
1379 	if (ifmp && (dev->ifindex != 0))
1380 		peer->ifindex = ifmp->ifi_index;
1381 
1382 	peer->gso_max_size = dev->gso_max_size;
1383 	peer->gso_max_segs = dev->gso_max_segs;
1384 
1385 	err = register_netdevice(peer);
1386 	put_net(net);
1387 	net = NULL;
1388 	if (err < 0)
1389 		goto err_register_peer;
1390 
1391 	netif_carrier_off(peer);
1392 
1393 	err = rtnl_configure_link(peer, ifmp);
1394 	if (err < 0)
1395 		goto err_configure_peer;
1396 
1397 	/*
1398 	 * register dev last
1399 	 *
1400 	 * note, that since we've registered new device the dev's name
1401 	 * should be re-allocated
1402 	 */
1403 
1404 	if (tb[IFLA_ADDRESS] == NULL)
1405 		eth_hw_addr_random(dev);
1406 
1407 	if (tb[IFLA_IFNAME])
1408 		nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1409 	else
1410 		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1411 
1412 	err = register_netdevice(dev);
1413 	if (err < 0)
1414 		goto err_register_dev;
1415 
1416 	netif_carrier_off(dev);
1417 
1418 	/*
1419 	 * tie the deviced together
1420 	 */
1421 
1422 	priv = netdev_priv(dev);
1423 	rcu_assign_pointer(priv->peer, peer);
1424 
1425 	priv = netdev_priv(peer);
1426 	rcu_assign_pointer(priv->peer, dev);
1427 
1428 	return 0;
1429 
1430 err_register_dev:
1431 	/* nothing to do */
1432 err_configure_peer:
1433 	unregister_netdevice(peer);
1434 	return err;
1435 
1436 err_register_peer:
1437 	free_netdev(peer);
1438 	return err;
1439 }
1440 
1441 static void veth_dellink(struct net_device *dev, struct list_head *head)
1442 {
1443 	struct veth_priv *priv;
1444 	struct net_device *peer;
1445 
1446 	priv = netdev_priv(dev);
1447 	peer = rtnl_dereference(priv->peer);
1448 
1449 	/* Note : dellink() is called from default_device_exit_batch(),
1450 	 * before a rcu_synchronize() point. The devices are guaranteed
1451 	 * not being freed before one RCU grace period.
1452 	 */
1453 	RCU_INIT_POINTER(priv->peer, NULL);
1454 	unregister_netdevice_queue(dev, head);
1455 
1456 	if (peer) {
1457 		priv = netdev_priv(peer);
1458 		RCU_INIT_POINTER(priv->peer, NULL);
1459 		unregister_netdevice_queue(peer, head);
1460 	}
1461 }
1462 
1463 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1464 	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
1465 };
1466 
1467 static struct net *veth_get_link_net(const struct net_device *dev)
1468 {
1469 	struct veth_priv *priv = netdev_priv(dev);
1470 	struct net_device *peer = rtnl_dereference(priv->peer);
1471 
1472 	return peer ? dev_net(peer) : dev_net(dev);
1473 }
1474 
1475 static struct rtnl_link_ops veth_link_ops = {
1476 	.kind		= DRV_NAME,
1477 	.priv_size	= sizeof(struct veth_priv),
1478 	.setup		= veth_setup,
1479 	.validate	= veth_validate,
1480 	.newlink	= veth_newlink,
1481 	.dellink	= veth_dellink,
1482 	.policy		= veth_policy,
1483 	.maxtype	= VETH_INFO_MAX,
1484 	.get_link_net	= veth_get_link_net,
1485 };
1486 
1487 /*
1488  * init/fini
1489  */
1490 
1491 static __init int veth_init(void)
1492 {
1493 	return rtnl_link_register(&veth_link_ops);
1494 }
1495 
1496 static __exit void veth_exit(void)
1497 {
1498 	rtnl_link_unregister(&veth_link_ops);
1499 }
1500 
1501 module_init(veth_init);
1502 module_exit(veth_exit);
1503 
1504 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1505 MODULE_LICENSE("GPL v2");
1506 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1507