xref: /linux/drivers/net/veth.c (revision 8137a49e1567726eb10fcf55ad141ac19804ca6b)
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 napi_struct __rcu *napi; /* points to xdp_napi when the latter is initialized */
61 	struct net_device	*dev;
62 	struct bpf_prog __rcu	*xdp_prog;
63 	struct xdp_mem_info	xdp_mem;
64 	struct veth_rq_stats	stats;
65 	bool			rx_notify_masked;
66 	struct ptr_ring		xdp_ring;
67 	struct xdp_rxq_info	xdp_rxq;
68 };
69 
70 struct veth_priv {
71 	struct net_device __rcu	*peer;
72 	atomic64_t		dropped;
73 	struct bpf_prog		*_xdp_prog;
74 	struct veth_rq		*rq;
75 	unsigned int		requested_headroom;
76 };
77 
78 struct veth_xdp_tx_bq {
79 	struct xdp_frame *q[VETH_XDP_TX_BULK_SIZE];
80 	unsigned int count;
81 };
82 
83 /*
84  * ethtool interface
85  */
86 
87 struct veth_q_stat_desc {
88 	char	desc[ETH_GSTRING_LEN];
89 	size_t	offset;
90 };
91 
92 #define VETH_RQ_STAT(m)	offsetof(struct veth_stats, m)
93 
94 static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
95 	{ "xdp_packets",	VETH_RQ_STAT(xdp_packets) },
96 	{ "xdp_bytes",		VETH_RQ_STAT(xdp_bytes) },
97 	{ "drops",		VETH_RQ_STAT(rx_drops) },
98 	{ "xdp_redirect",	VETH_RQ_STAT(xdp_redirect) },
99 	{ "xdp_drops",		VETH_RQ_STAT(xdp_drops) },
100 	{ "xdp_tx",		VETH_RQ_STAT(xdp_tx) },
101 	{ "xdp_tx_errors",	VETH_RQ_STAT(xdp_tx_err) },
102 };
103 
104 #define VETH_RQ_STATS_LEN	ARRAY_SIZE(veth_rq_stats_desc)
105 
106 static const struct veth_q_stat_desc veth_tq_stats_desc[] = {
107 	{ "xdp_xmit",		VETH_RQ_STAT(peer_tq_xdp_xmit) },
108 	{ "xdp_xmit_errors",	VETH_RQ_STAT(peer_tq_xdp_xmit_err) },
109 };
110 
111 #define VETH_TQ_STATS_LEN	ARRAY_SIZE(veth_tq_stats_desc)
112 
113 static struct {
114 	const char string[ETH_GSTRING_LEN];
115 } ethtool_stats_keys[] = {
116 	{ "peer_ifindex" },
117 };
118 
119 static int veth_get_link_ksettings(struct net_device *dev,
120 				   struct ethtool_link_ksettings *cmd)
121 {
122 	cmd->base.speed		= SPEED_10000;
123 	cmd->base.duplex	= DUPLEX_FULL;
124 	cmd->base.port		= PORT_TP;
125 	cmd->base.autoneg	= AUTONEG_DISABLE;
126 	return 0;
127 }
128 
129 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
130 {
131 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
132 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
133 }
134 
135 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
136 {
137 	char *p = (char *)buf;
138 	int i, j;
139 
140 	switch(stringset) {
141 	case ETH_SS_STATS:
142 		memcpy(p, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
143 		p += sizeof(ethtool_stats_keys);
144 		for (i = 0; i < dev->real_num_rx_queues; i++) {
145 			for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
146 				snprintf(p, ETH_GSTRING_LEN,
147 					 "rx_queue_%u_%.18s",
148 					 i, veth_rq_stats_desc[j].desc);
149 				p += ETH_GSTRING_LEN;
150 			}
151 		}
152 		for (i = 0; i < dev->real_num_tx_queues; i++) {
153 			for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
154 				snprintf(p, ETH_GSTRING_LEN,
155 					 "tx_queue_%u_%.18s",
156 					 i, veth_tq_stats_desc[j].desc);
157 				p += ETH_GSTRING_LEN;
158 			}
159 		}
160 		break;
161 	}
162 }
163 
164 static int veth_get_sset_count(struct net_device *dev, int sset)
165 {
166 	switch (sset) {
167 	case ETH_SS_STATS:
168 		return ARRAY_SIZE(ethtool_stats_keys) +
169 		       VETH_RQ_STATS_LEN * dev->real_num_rx_queues +
170 		       VETH_TQ_STATS_LEN * dev->real_num_tx_queues;
171 	default:
172 		return -EOPNOTSUPP;
173 	}
174 }
175 
176 static void veth_get_ethtool_stats(struct net_device *dev,
177 		struct ethtool_stats *stats, u64 *data)
178 {
179 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
180 	struct net_device *peer = rtnl_dereference(priv->peer);
181 	int i, j, idx;
182 
183 	data[0] = peer ? peer->ifindex : 0;
184 	idx = 1;
185 	for (i = 0; i < dev->real_num_rx_queues; i++) {
186 		const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
187 		const void *stats_base = (void *)&rq_stats->vs;
188 		unsigned int start;
189 		size_t offset;
190 
191 		do {
192 			start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
193 			for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
194 				offset = veth_rq_stats_desc[j].offset;
195 				data[idx + j] = *(u64 *)(stats_base + offset);
196 			}
197 		} while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
198 		idx += VETH_RQ_STATS_LEN;
199 	}
200 
201 	if (!peer)
202 		return;
203 
204 	rcv_priv = netdev_priv(peer);
205 	for (i = 0; i < peer->real_num_rx_queues; i++) {
206 		const struct veth_rq_stats *rq_stats = &rcv_priv->rq[i].stats;
207 		const void *base = (void *)&rq_stats->vs;
208 		unsigned int start, tx_idx = idx;
209 		size_t offset;
210 
211 		tx_idx += (i % dev->real_num_tx_queues) * VETH_TQ_STATS_LEN;
212 		do {
213 			start = u64_stats_fetch_begin_irq(&rq_stats->syncp);
214 			for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
215 				offset = veth_tq_stats_desc[j].offset;
216 				data[tx_idx + j] += *(u64 *)(base + offset);
217 			}
218 		} while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start));
219 	}
220 }
221 
222 static void veth_get_channels(struct net_device *dev,
223 			      struct ethtool_channels *channels)
224 {
225 	channels->tx_count = dev->real_num_tx_queues;
226 	channels->rx_count = dev->real_num_rx_queues;
227 	channels->max_tx = dev->real_num_tx_queues;
228 	channels->max_rx = dev->real_num_rx_queues;
229 	channels->combined_count = min(dev->real_num_rx_queues, dev->real_num_tx_queues);
230 	channels->max_combined = min(dev->real_num_rx_queues, dev->real_num_tx_queues);
231 }
232 
233 static const struct ethtool_ops veth_ethtool_ops = {
234 	.get_drvinfo		= veth_get_drvinfo,
235 	.get_link		= ethtool_op_get_link,
236 	.get_strings		= veth_get_strings,
237 	.get_sset_count		= veth_get_sset_count,
238 	.get_ethtool_stats	= veth_get_ethtool_stats,
239 	.get_link_ksettings	= veth_get_link_ksettings,
240 	.get_ts_info		= ethtool_op_get_ts_info,
241 	.get_channels		= veth_get_channels,
242 };
243 
244 /* general routines */
245 
246 static bool veth_is_xdp_frame(void *ptr)
247 {
248 	return (unsigned long)ptr & VETH_XDP_FLAG;
249 }
250 
251 static struct xdp_frame *veth_ptr_to_xdp(void *ptr)
252 {
253 	return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
254 }
255 
256 static void *veth_xdp_to_ptr(struct xdp_frame *xdp)
257 {
258 	return (void *)((unsigned long)xdp | VETH_XDP_FLAG);
259 }
260 
261 static void veth_ptr_free(void *ptr)
262 {
263 	if (veth_is_xdp_frame(ptr))
264 		xdp_return_frame(veth_ptr_to_xdp(ptr));
265 	else
266 		kfree_skb(ptr);
267 }
268 
269 static void __veth_xdp_flush(struct veth_rq *rq)
270 {
271 	/* Write ptr_ring before reading rx_notify_masked */
272 	smp_mb();
273 	if (!rq->rx_notify_masked) {
274 		rq->rx_notify_masked = true;
275 		napi_schedule(&rq->xdp_napi);
276 	}
277 }
278 
279 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
280 {
281 	if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
282 		dev_kfree_skb_any(skb);
283 		return NET_RX_DROP;
284 	}
285 
286 	return NET_RX_SUCCESS;
287 }
288 
289 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
290 			    struct veth_rq *rq, bool xdp)
291 {
292 	return __dev_forward_skb(dev, skb) ?: xdp ?
293 		veth_xdp_rx(rq, skb) :
294 		netif_rx(skb);
295 }
296 
297 /* return true if the specified skb has chances of GRO aggregation
298  * Don't strive for accuracy, but try to avoid GRO overhead in the most
299  * common scenarios.
300  * When XDP is enabled, all traffic is considered eligible, as the xmit
301  * device has TSO off.
302  * When TSO is enabled on the xmit device, we are likely interested only
303  * in UDP aggregation, explicitly check for that if the skb is suspected
304  * - the sock_wfree destructor is used by UDP, ICMP and XDP sockets -
305  * to belong to locally generated UDP traffic.
306  */
307 static bool veth_skb_is_eligible_for_gro(const struct net_device *dev,
308 					 const struct net_device *rcv,
309 					 const struct sk_buff *skb)
310 {
311 	return !(dev->features & NETIF_F_ALL_TSO) ||
312 		(skb->destructor == sock_wfree &&
313 		 rcv->features & (NETIF_F_GRO_FRAGLIST | NETIF_F_GRO_UDP_FWD));
314 }
315 
316 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
317 {
318 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
319 	struct veth_rq *rq = NULL;
320 	struct net_device *rcv;
321 	int length = skb->len;
322 	bool use_napi = false;
323 	int rxq;
324 
325 	rcu_read_lock();
326 	rcv = rcu_dereference(priv->peer);
327 	if (unlikely(!rcv)) {
328 		kfree_skb(skb);
329 		goto drop;
330 	}
331 
332 	rcv_priv = netdev_priv(rcv);
333 	rxq = skb_get_queue_mapping(skb);
334 	if (rxq < rcv->real_num_rx_queues) {
335 		rq = &rcv_priv->rq[rxq];
336 
337 		/* The napi pointer is available when an XDP program is
338 		 * attached or when GRO is enabled
339 		 * Don't bother with napi/GRO if the skb can't be aggregated
340 		 */
341 		use_napi = rcu_access_pointer(rq->napi) &&
342 			   veth_skb_is_eligible_for_gro(dev, rcv, skb);
343 		skb_record_rx_queue(skb, rxq);
344 	}
345 
346 	skb_tx_timestamp(skb);
347 	if (likely(veth_forward_skb(rcv, skb, rq, use_napi) == NET_RX_SUCCESS)) {
348 		if (!use_napi)
349 			dev_lstats_add(dev, length);
350 	} else {
351 drop:
352 		atomic64_inc(&priv->dropped);
353 	}
354 
355 	if (use_napi)
356 		__veth_xdp_flush(rq);
357 
358 	rcu_read_unlock();
359 
360 	return NETDEV_TX_OK;
361 }
362 
363 static u64 veth_stats_tx(struct net_device *dev, u64 *packets, u64 *bytes)
364 {
365 	struct veth_priv *priv = netdev_priv(dev);
366 
367 	dev_lstats_read(dev, packets, bytes);
368 	return atomic64_read(&priv->dropped);
369 }
370 
371 static void veth_stats_rx(struct veth_stats *result, struct net_device *dev)
372 {
373 	struct veth_priv *priv = netdev_priv(dev);
374 	int i;
375 
376 	result->peer_tq_xdp_xmit_err = 0;
377 	result->xdp_packets = 0;
378 	result->xdp_tx_err = 0;
379 	result->xdp_bytes = 0;
380 	result->rx_drops = 0;
381 	for (i = 0; i < dev->num_rx_queues; i++) {
382 		u64 packets, bytes, drops, xdp_tx_err, peer_tq_xdp_xmit_err;
383 		struct veth_rq_stats *stats = &priv->rq[i].stats;
384 		unsigned int start;
385 
386 		do {
387 			start = u64_stats_fetch_begin_irq(&stats->syncp);
388 			peer_tq_xdp_xmit_err = stats->vs.peer_tq_xdp_xmit_err;
389 			xdp_tx_err = stats->vs.xdp_tx_err;
390 			packets = stats->vs.xdp_packets;
391 			bytes = stats->vs.xdp_bytes;
392 			drops = stats->vs.rx_drops;
393 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
394 		result->peer_tq_xdp_xmit_err += peer_tq_xdp_xmit_err;
395 		result->xdp_tx_err += xdp_tx_err;
396 		result->xdp_packets += packets;
397 		result->xdp_bytes += bytes;
398 		result->rx_drops += drops;
399 	}
400 }
401 
402 static void veth_get_stats64(struct net_device *dev,
403 			     struct rtnl_link_stats64 *tot)
404 {
405 	struct veth_priv *priv = netdev_priv(dev);
406 	struct net_device *peer;
407 	struct veth_stats rx;
408 	u64 packets, bytes;
409 
410 	tot->tx_dropped = veth_stats_tx(dev, &packets, &bytes);
411 	tot->tx_bytes = bytes;
412 	tot->tx_packets = packets;
413 
414 	veth_stats_rx(&rx, dev);
415 	tot->tx_dropped += rx.xdp_tx_err;
416 	tot->rx_dropped = rx.rx_drops + rx.peer_tq_xdp_xmit_err;
417 	tot->rx_bytes = rx.xdp_bytes;
418 	tot->rx_packets = rx.xdp_packets;
419 
420 	rcu_read_lock();
421 	peer = rcu_dereference(priv->peer);
422 	if (peer) {
423 		veth_stats_tx(peer, &packets, &bytes);
424 		tot->rx_bytes += bytes;
425 		tot->rx_packets += packets;
426 
427 		veth_stats_rx(&rx, peer);
428 		tot->tx_dropped += rx.peer_tq_xdp_xmit_err;
429 		tot->rx_dropped += rx.xdp_tx_err;
430 		tot->tx_bytes += rx.xdp_bytes;
431 		tot->tx_packets += rx.xdp_packets;
432 	}
433 	rcu_read_unlock();
434 }
435 
436 /* fake multicast ability */
437 static void veth_set_multicast_list(struct net_device *dev)
438 {
439 }
440 
441 static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
442 				      int buflen)
443 {
444 	struct sk_buff *skb;
445 
446 	skb = build_skb(head, buflen);
447 	if (!skb)
448 		return NULL;
449 
450 	skb_reserve(skb, headroom);
451 	skb_put(skb, len);
452 
453 	return skb;
454 }
455 
456 static int veth_select_rxq(struct net_device *dev)
457 {
458 	return smp_processor_id() % dev->real_num_rx_queues;
459 }
460 
461 static struct net_device *veth_peer_dev(struct net_device *dev)
462 {
463 	struct veth_priv *priv = netdev_priv(dev);
464 
465 	/* Callers must be under RCU read side. */
466 	return rcu_dereference(priv->peer);
467 }
468 
469 static int veth_xdp_xmit(struct net_device *dev, int n,
470 			 struct xdp_frame **frames,
471 			 u32 flags, bool ndo_xmit)
472 {
473 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
474 	int i, ret = -ENXIO, nxmit = 0;
475 	struct net_device *rcv;
476 	unsigned int max_len;
477 	struct veth_rq *rq;
478 
479 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
480 		return -EINVAL;
481 
482 	rcu_read_lock();
483 	rcv = rcu_dereference(priv->peer);
484 	if (unlikely(!rcv))
485 		goto out;
486 
487 	rcv_priv = netdev_priv(rcv);
488 	rq = &rcv_priv->rq[veth_select_rxq(rcv)];
489 	/* The napi pointer is set if NAPI is enabled, which ensures that
490 	 * xdp_ring is initialized on receive side and the peer device is up.
491 	 */
492 	if (!rcu_access_pointer(rq->napi))
493 		goto out;
494 
495 	max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
496 
497 	spin_lock(&rq->xdp_ring.producer_lock);
498 	for (i = 0; i < n; i++) {
499 		struct xdp_frame *frame = frames[i];
500 		void *ptr = veth_xdp_to_ptr(frame);
501 
502 		if (unlikely(frame->len > max_len ||
503 			     __ptr_ring_produce(&rq->xdp_ring, ptr)))
504 			break;
505 		nxmit++;
506 	}
507 	spin_unlock(&rq->xdp_ring.producer_lock);
508 
509 	if (flags & XDP_XMIT_FLUSH)
510 		__veth_xdp_flush(rq);
511 
512 	ret = nxmit;
513 	if (ndo_xmit) {
514 		u64_stats_update_begin(&rq->stats.syncp);
515 		rq->stats.vs.peer_tq_xdp_xmit += nxmit;
516 		rq->stats.vs.peer_tq_xdp_xmit_err += n - nxmit;
517 		u64_stats_update_end(&rq->stats.syncp);
518 	}
519 
520 out:
521 	rcu_read_unlock();
522 
523 	return ret;
524 }
525 
526 static int veth_ndo_xdp_xmit(struct net_device *dev, int n,
527 			     struct xdp_frame **frames, u32 flags)
528 {
529 	int err;
530 
531 	err = veth_xdp_xmit(dev, n, frames, flags, true);
532 	if (err < 0) {
533 		struct veth_priv *priv = netdev_priv(dev);
534 
535 		atomic64_add(n, &priv->dropped);
536 	}
537 
538 	return err;
539 }
540 
541 static void veth_xdp_flush_bq(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
542 {
543 	int sent, i, err = 0, drops;
544 
545 	sent = veth_xdp_xmit(rq->dev, bq->count, bq->q, 0, false);
546 	if (sent < 0) {
547 		err = sent;
548 		sent = 0;
549 	}
550 
551 	for (i = sent; unlikely(i < bq->count); i++)
552 		xdp_return_frame(bq->q[i]);
553 
554 	drops = bq->count - sent;
555 	trace_xdp_bulk_tx(rq->dev, sent, drops, err);
556 
557 	u64_stats_update_begin(&rq->stats.syncp);
558 	rq->stats.vs.xdp_tx += sent;
559 	rq->stats.vs.xdp_tx_err += drops;
560 	u64_stats_update_end(&rq->stats.syncp);
561 
562 	bq->count = 0;
563 }
564 
565 static void veth_xdp_flush(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
566 {
567 	struct veth_priv *rcv_priv, *priv = netdev_priv(rq->dev);
568 	struct net_device *rcv;
569 	struct veth_rq *rcv_rq;
570 
571 	rcu_read_lock();
572 	veth_xdp_flush_bq(rq, bq);
573 	rcv = rcu_dereference(priv->peer);
574 	if (unlikely(!rcv))
575 		goto out;
576 
577 	rcv_priv = netdev_priv(rcv);
578 	rcv_rq = &rcv_priv->rq[veth_select_rxq(rcv)];
579 	/* xdp_ring is initialized on receive side? */
580 	if (unlikely(!rcu_access_pointer(rcv_rq->xdp_prog)))
581 		goto out;
582 
583 	__veth_xdp_flush(rcv_rq);
584 out:
585 	rcu_read_unlock();
586 }
587 
588 static int veth_xdp_tx(struct veth_rq *rq, struct xdp_buff *xdp,
589 		       struct veth_xdp_tx_bq *bq)
590 {
591 	struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
592 
593 	if (unlikely(!frame))
594 		return -EOVERFLOW;
595 
596 	if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
597 		veth_xdp_flush_bq(rq, bq);
598 
599 	bq->q[bq->count++] = frame;
600 
601 	return 0;
602 }
603 
604 static struct xdp_frame *veth_xdp_rcv_one(struct veth_rq *rq,
605 					  struct xdp_frame *frame,
606 					  struct veth_xdp_tx_bq *bq,
607 					  struct veth_stats *stats)
608 {
609 	struct xdp_frame orig_frame;
610 	struct bpf_prog *xdp_prog;
611 
612 	rcu_read_lock();
613 	xdp_prog = rcu_dereference(rq->xdp_prog);
614 	if (likely(xdp_prog)) {
615 		struct xdp_buff xdp;
616 		u32 act;
617 
618 		xdp_convert_frame_to_buff(frame, &xdp);
619 		xdp.rxq = &rq->xdp_rxq;
620 
621 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
622 
623 		switch (act) {
624 		case XDP_PASS:
625 			if (xdp_update_frame_from_buff(&xdp, frame))
626 				goto err_xdp;
627 			break;
628 		case XDP_TX:
629 			orig_frame = *frame;
630 			xdp.rxq->mem = frame->mem;
631 			if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
632 				trace_xdp_exception(rq->dev, xdp_prog, act);
633 				frame = &orig_frame;
634 				stats->rx_drops++;
635 				goto err_xdp;
636 			}
637 			stats->xdp_tx++;
638 			rcu_read_unlock();
639 			goto xdp_xmit;
640 		case XDP_REDIRECT:
641 			orig_frame = *frame;
642 			xdp.rxq->mem = frame->mem;
643 			if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
644 				frame = &orig_frame;
645 				stats->rx_drops++;
646 				goto err_xdp;
647 			}
648 			stats->xdp_redirect++;
649 			rcu_read_unlock();
650 			goto xdp_xmit;
651 		default:
652 			bpf_warn_invalid_xdp_action(act);
653 			fallthrough;
654 		case XDP_ABORTED:
655 			trace_xdp_exception(rq->dev, xdp_prog, act);
656 			fallthrough;
657 		case XDP_DROP:
658 			stats->xdp_drops++;
659 			goto err_xdp;
660 		}
661 	}
662 	rcu_read_unlock();
663 
664 	return frame;
665 err_xdp:
666 	rcu_read_unlock();
667 	xdp_return_frame(frame);
668 xdp_xmit:
669 	return NULL;
670 }
671 
672 /* frames array contains VETH_XDP_BATCH at most */
673 static void veth_xdp_rcv_bulk_skb(struct veth_rq *rq, void **frames,
674 				  int n_xdpf, struct veth_xdp_tx_bq *bq,
675 				  struct veth_stats *stats)
676 {
677 	void *skbs[VETH_XDP_BATCH];
678 	int i;
679 
680 	if (xdp_alloc_skb_bulk(skbs, n_xdpf,
681 			       GFP_ATOMIC | __GFP_ZERO) < 0) {
682 		for (i = 0; i < n_xdpf; i++)
683 			xdp_return_frame(frames[i]);
684 		stats->rx_drops += n_xdpf;
685 
686 		return;
687 	}
688 
689 	for (i = 0; i < n_xdpf; i++) {
690 		struct sk_buff *skb = skbs[i];
691 
692 		skb = __xdp_build_skb_from_frame(frames[i], skb,
693 						 rq->dev);
694 		if (!skb) {
695 			xdp_return_frame(frames[i]);
696 			stats->rx_drops++;
697 			continue;
698 		}
699 		napi_gro_receive(&rq->xdp_napi, skb);
700 	}
701 }
702 
703 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
704 					struct sk_buff *skb,
705 					struct veth_xdp_tx_bq *bq,
706 					struct veth_stats *stats)
707 {
708 	u32 pktlen, headroom, act, metalen, frame_sz;
709 	void *orig_data, *orig_data_end;
710 	struct bpf_prog *xdp_prog;
711 	int mac_len, delta, off;
712 	struct xdp_buff xdp;
713 
714 	skb_orphan_partial(skb);
715 
716 	rcu_read_lock();
717 	xdp_prog = rcu_dereference(rq->xdp_prog);
718 	if (unlikely(!xdp_prog)) {
719 		rcu_read_unlock();
720 		goto out;
721 	}
722 
723 	mac_len = skb->data - skb_mac_header(skb);
724 	pktlen = skb->len + mac_len;
725 	headroom = skb_headroom(skb) - mac_len;
726 
727 	if (skb_shared(skb) || skb_head_is_locked(skb) ||
728 	    skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
729 		struct sk_buff *nskb;
730 		int size, head_off;
731 		void *head, *start;
732 		struct page *page;
733 
734 		size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
735 		       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
736 		if (size > PAGE_SIZE)
737 			goto drop;
738 
739 		page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
740 		if (!page)
741 			goto drop;
742 
743 		head = page_address(page);
744 		start = head + VETH_XDP_HEADROOM;
745 		if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
746 			page_frag_free(head);
747 			goto drop;
748 		}
749 
750 		nskb = veth_build_skb(head, VETH_XDP_HEADROOM + mac_len,
751 				      skb->len, PAGE_SIZE);
752 		if (!nskb) {
753 			page_frag_free(head);
754 			goto drop;
755 		}
756 
757 		skb_copy_header(nskb, skb);
758 		head_off = skb_headroom(nskb) - skb_headroom(skb);
759 		skb_headers_offset_update(nskb, head_off);
760 		consume_skb(skb);
761 		skb = nskb;
762 	}
763 
764 	/* SKB "head" area always have tailroom for skb_shared_info */
765 	frame_sz = skb_end_pointer(skb) - skb->head;
766 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
767 	xdp_init_buff(&xdp, frame_sz, &rq->xdp_rxq);
768 	xdp_prepare_buff(&xdp, skb->head, skb->mac_header, pktlen, true);
769 
770 	orig_data = xdp.data;
771 	orig_data_end = xdp.data_end;
772 
773 	act = bpf_prog_run_xdp(xdp_prog, &xdp);
774 
775 	switch (act) {
776 	case XDP_PASS:
777 		break;
778 	case XDP_TX:
779 		get_page(virt_to_page(xdp.data));
780 		consume_skb(skb);
781 		xdp.rxq->mem = rq->xdp_mem;
782 		if (unlikely(veth_xdp_tx(rq, &xdp, bq) < 0)) {
783 			trace_xdp_exception(rq->dev, xdp_prog, act);
784 			stats->rx_drops++;
785 			goto err_xdp;
786 		}
787 		stats->xdp_tx++;
788 		rcu_read_unlock();
789 		goto xdp_xmit;
790 	case XDP_REDIRECT:
791 		get_page(virt_to_page(xdp.data));
792 		consume_skb(skb);
793 		xdp.rxq->mem = rq->xdp_mem;
794 		if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
795 			stats->rx_drops++;
796 			goto err_xdp;
797 		}
798 		stats->xdp_redirect++;
799 		rcu_read_unlock();
800 		goto xdp_xmit;
801 	default:
802 		bpf_warn_invalid_xdp_action(act);
803 		fallthrough;
804 	case XDP_ABORTED:
805 		trace_xdp_exception(rq->dev, xdp_prog, act);
806 		fallthrough;
807 	case XDP_DROP:
808 		stats->xdp_drops++;
809 		goto xdp_drop;
810 	}
811 	rcu_read_unlock();
812 
813 	/* check if bpf_xdp_adjust_head was used */
814 	delta = orig_data - xdp.data;
815 	off = mac_len + delta;
816 	if (off > 0)
817 		__skb_push(skb, off);
818 	else if (off < 0)
819 		__skb_pull(skb, -off);
820 	skb->mac_header -= delta;
821 
822 	/* check if bpf_xdp_adjust_tail was used */
823 	off = xdp.data_end - orig_data_end;
824 	if (off != 0)
825 		__skb_put(skb, off); /* positive on grow, negative on shrink */
826 	skb->protocol = eth_type_trans(skb, rq->dev);
827 
828 	metalen = xdp.data - xdp.data_meta;
829 	if (metalen)
830 		skb_metadata_set(skb, metalen);
831 out:
832 	return skb;
833 drop:
834 	stats->rx_drops++;
835 xdp_drop:
836 	rcu_read_unlock();
837 	kfree_skb(skb);
838 	return NULL;
839 err_xdp:
840 	rcu_read_unlock();
841 	page_frag_free(xdp.data);
842 xdp_xmit:
843 	return NULL;
844 }
845 
846 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
847 			struct veth_xdp_tx_bq *bq,
848 			struct veth_stats *stats)
849 {
850 	int i, done = 0, n_xdpf = 0;
851 	void *xdpf[VETH_XDP_BATCH];
852 
853 	for (i = 0; i < budget; i++) {
854 		void *ptr = __ptr_ring_consume(&rq->xdp_ring);
855 
856 		if (!ptr)
857 			break;
858 
859 		if (veth_is_xdp_frame(ptr)) {
860 			/* ndo_xdp_xmit */
861 			struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
862 
863 			stats->xdp_bytes += frame->len;
864 			frame = veth_xdp_rcv_one(rq, frame, bq, stats);
865 			if (frame) {
866 				/* XDP_PASS */
867 				xdpf[n_xdpf++] = frame;
868 				if (n_xdpf == VETH_XDP_BATCH) {
869 					veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf,
870 							      bq, stats);
871 					n_xdpf = 0;
872 				}
873 			}
874 		} else {
875 			/* ndo_start_xmit */
876 			struct sk_buff *skb = ptr;
877 
878 			stats->xdp_bytes += skb->len;
879 			skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
880 			if (skb)
881 				napi_gro_receive(&rq->xdp_napi, skb);
882 		}
883 		done++;
884 	}
885 
886 	if (n_xdpf)
887 		veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf, bq, stats);
888 
889 	u64_stats_update_begin(&rq->stats.syncp);
890 	rq->stats.vs.xdp_redirect += stats->xdp_redirect;
891 	rq->stats.vs.xdp_bytes += stats->xdp_bytes;
892 	rq->stats.vs.xdp_drops += stats->xdp_drops;
893 	rq->stats.vs.rx_drops += stats->rx_drops;
894 	rq->stats.vs.xdp_packets += done;
895 	u64_stats_update_end(&rq->stats.syncp);
896 
897 	return done;
898 }
899 
900 static int veth_poll(struct napi_struct *napi, int budget)
901 {
902 	struct veth_rq *rq =
903 		container_of(napi, struct veth_rq, xdp_napi);
904 	struct veth_stats stats = {};
905 	struct veth_xdp_tx_bq bq;
906 	int done;
907 
908 	bq.count = 0;
909 
910 	xdp_set_return_frame_no_direct();
911 	done = veth_xdp_rcv(rq, budget, &bq, &stats);
912 
913 	if (done < budget && napi_complete_done(napi, done)) {
914 		/* Write rx_notify_masked before reading ptr_ring */
915 		smp_store_mb(rq->rx_notify_masked, false);
916 		if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
917 			rq->rx_notify_masked = true;
918 			napi_schedule(&rq->xdp_napi);
919 		}
920 	}
921 
922 	if (stats.xdp_tx > 0)
923 		veth_xdp_flush(rq, &bq);
924 	if (stats.xdp_redirect > 0)
925 		xdp_do_flush();
926 	xdp_clear_return_frame_no_direct();
927 
928 	return done;
929 }
930 
931 static int __veth_napi_enable(struct net_device *dev)
932 {
933 	struct veth_priv *priv = netdev_priv(dev);
934 	int err, i;
935 
936 	for (i = 0; i < dev->real_num_rx_queues; i++) {
937 		struct veth_rq *rq = &priv->rq[i];
938 
939 		err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
940 		if (err)
941 			goto err_xdp_ring;
942 	}
943 
944 	for (i = 0; i < dev->real_num_rx_queues; i++) {
945 		struct veth_rq *rq = &priv->rq[i];
946 
947 		napi_enable(&rq->xdp_napi);
948 		rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
949 	}
950 
951 	return 0;
952 err_xdp_ring:
953 	for (i--; i >= 0; i--)
954 		ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
955 
956 	return err;
957 }
958 
959 static void veth_napi_del(struct net_device *dev)
960 {
961 	struct veth_priv *priv = netdev_priv(dev);
962 	int i;
963 
964 	for (i = 0; i < dev->real_num_rx_queues; i++) {
965 		struct veth_rq *rq = &priv->rq[i];
966 
967 		rcu_assign_pointer(priv->rq[i].napi, NULL);
968 		napi_disable(&rq->xdp_napi);
969 		__netif_napi_del(&rq->xdp_napi);
970 	}
971 	synchronize_net();
972 
973 	for (i = 0; i < dev->real_num_rx_queues; i++) {
974 		struct veth_rq *rq = &priv->rq[i];
975 
976 		rq->rx_notify_masked = false;
977 		ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
978 	}
979 }
980 
981 static bool veth_gro_requested(const struct net_device *dev)
982 {
983 	return !!(dev->wanted_features & NETIF_F_GRO);
984 }
985 
986 static int veth_enable_xdp(struct net_device *dev)
987 {
988 	bool napi_already_on = veth_gro_requested(dev) && (dev->flags & IFF_UP);
989 	struct veth_priv *priv = netdev_priv(dev);
990 	int err, i;
991 
992 	if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
993 		for (i = 0; i < dev->real_num_rx_queues; i++) {
994 			struct veth_rq *rq = &priv->rq[i];
995 
996 			if (!napi_already_on)
997 				netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
998 			err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i, rq->xdp_napi.napi_id);
999 			if (err < 0)
1000 				goto err_rxq_reg;
1001 
1002 			err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
1003 							 MEM_TYPE_PAGE_SHARED,
1004 							 NULL);
1005 			if (err < 0)
1006 				goto err_reg_mem;
1007 
1008 			/* Save original mem info as it can be overwritten */
1009 			rq->xdp_mem = rq->xdp_rxq.mem;
1010 		}
1011 
1012 		if (!napi_already_on) {
1013 			err = __veth_napi_enable(dev);
1014 			if (err)
1015 				goto err_rxq_reg;
1016 
1017 			if (!veth_gro_requested(dev)) {
1018 				/* user-space did not require GRO, but adding XDP
1019 				 * is supposed to get GRO working
1020 				 */
1021 				dev->features |= NETIF_F_GRO;
1022 				netdev_features_change(dev);
1023 			}
1024 		}
1025 	}
1026 
1027 	for (i = 0; i < dev->real_num_rx_queues; i++) {
1028 		rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
1029 		rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1030 	}
1031 
1032 	return 0;
1033 err_reg_mem:
1034 	xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
1035 err_rxq_reg:
1036 	for (i--; i >= 0; i--) {
1037 		struct veth_rq *rq = &priv->rq[i];
1038 
1039 		xdp_rxq_info_unreg(&rq->xdp_rxq);
1040 		if (!napi_already_on)
1041 			netif_napi_del(&rq->xdp_napi);
1042 	}
1043 
1044 	return err;
1045 }
1046 
1047 static void veth_disable_xdp(struct net_device *dev)
1048 {
1049 	struct veth_priv *priv = netdev_priv(dev);
1050 	int i;
1051 
1052 	for (i = 0; i < dev->real_num_rx_queues; i++)
1053 		rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
1054 
1055 	if (!netif_running(dev) || !veth_gro_requested(dev)) {
1056 		veth_napi_del(dev);
1057 
1058 		/* if user-space did not require GRO, since adding XDP
1059 		 * enabled it, clear it now
1060 		 */
1061 		if (!veth_gro_requested(dev) && netif_running(dev)) {
1062 			dev->features &= ~NETIF_F_GRO;
1063 			netdev_features_change(dev);
1064 		}
1065 	}
1066 
1067 	for (i = 0; i < dev->real_num_rx_queues; i++) {
1068 		struct veth_rq *rq = &priv->rq[i];
1069 
1070 		rq->xdp_rxq.mem = rq->xdp_mem;
1071 		xdp_rxq_info_unreg(&rq->xdp_rxq);
1072 	}
1073 }
1074 
1075 static int veth_napi_enable(struct net_device *dev)
1076 {
1077 	struct veth_priv *priv = netdev_priv(dev);
1078 	int err, i;
1079 
1080 	for (i = 0; i < dev->real_num_rx_queues; i++) {
1081 		struct veth_rq *rq = &priv->rq[i];
1082 
1083 		netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
1084 	}
1085 
1086 	err = __veth_napi_enable(dev);
1087 	if (err) {
1088 		for (i = 0; i < dev->real_num_rx_queues; i++) {
1089 			struct veth_rq *rq = &priv->rq[i];
1090 
1091 			netif_napi_del(&rq->xdp_napi);
1092 		}
1093 		return err;
1094 	}
1095 	return err;
1096 }
1097 
1098 static int veth_open(struct net_device *dev)
1099 {
1100 	struct veth_priv *priv = netdev_priv(dev);
1101 	struct net_device *peer = rtnl_dereference(priv->peer);
1102 	int err;
1103 
1104 	if (!peer)
1105 		return -ENOTCONN;
1106 
1107 	if (priv->_xdp_prog) {
1108 		err = veth_enable_xdp(dev);
1109 		if (err)
1110 			return err;
1111 	} else if (veth_gro_requested(dev)) {
1112 		err = veth_napi_enable(dev);
1113 		if (err)
1114 			return err;
1115 	}
1116 
1117 	if (peer->flags & IFF_UP) {
1118 		netif_carrier_on(dev);
1119 		netif_carrier_on(peer);
1120 	}
1121 
1122 	return 0;
1123 }
1124 
1125 static int veth_close(struct net_device *dev)
1126 {
1127 	struct veth_priv *priv = netdev_priv(dev);
1128 	struct net_device *peer = rtnl_dereference(priv->peer);
1129 
1130 	netif_carrier_off(dev);
1131 	if (peer)
1132 		netif_carrier_off(peer);
1133 
1134 	if (priv->_xdp_prog)
1135 		veth_disable_xdp(dev);
1136 	else if (veth_gro_requested(dev))
1137 		veth_napi_del(dev);
1138 
1139 	return 0;
1140 }
1141 
1142 static int is_valid_veth_mtu(int mtu)
1143 {
1144 	return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1145 }
1146 
1147 static int veth_alloc_queues(struct net_device *dev)
1148 {
1149 	struct veth_priv *priv = netdev_priv(dev);
1150 	int i;
1151 
1152 	priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
1153 	if (!priv->rq)
1154 		return -ENOMEM;
1155 
1156 	for (i = 0; i < dev->num_rx_queues; i++) {
1157 		priv->rq[i].dev = dev;
1158 		u64_stats_init(&priv->rq[i].stats.syncp);
1159 	}
1160 
1161 	return 0;
1162 }
1163 
1164 static void veth_free_queues(struct net_device *dev)
1165 {
1166 	struct veth_priv *priv = netdev_priv(dev);
1167 
1168 	kfree(priv->rq);
1169 }
1170 
1171 static int veth_dev_init(struct net_device *dev)
1172 {
1173 	int err;
1174 
1175 	dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1176 	if (!dev->lstats)
1177 		return -ENOMEM;
1178 
1179 	err = veth_alloc_queues(dev);
1180 	if (err) {
1181 		free_percpu(dev->lstats);
1182 		return err;
1183 	}
1184 
1185 	return 0;
1186 }
1187 
1188 static void veth_dev_free(struct net_device *dev)
1189 {
1190 	veth_free_queues(dev);
1191 	free_percpu(dev->lstats);
1192 }
1193 
1194 #ifdef CONFIG_NET_POLL_CONTROLLER
1195 static void veth_poll_controller(struct net_device *dev)
1196 {
1197 	/* veth only receives frames when its peer sends one
1198 	 * Since it has nothing to do with disabling irqs, we are guaranteed
1199 	 * never to have pending data when we poll for it so
1200 	 * there is nothing to do here.
1201 	 *
1202 	 * We need this though so netpoll recognizes us as an interface that
1203 	 * supports polling, which enables bridge devices in virt setups to
1204 	 * still use netconsole
1205 	 */
1206 }
1207 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1208 
1209 static int veth_get_iflink(const struct net_device *dev)
1210 {
1211 	struct veth_priv *priv = netdev_priv(dev);
1212 	struct net_device *peer;
1213 	int iflink;
1214 
1215 	rcu_read_lock();
1216 	peer = rcu_dereference(priv->peer);
1217 	iflink = peer ? peer->ifindex : 0;
1218 	rcu_read_unlock();
1219 
1220 	return iflink;
1221 }
1222 
1223 static netdev_features_t veth_fix_features(struct net_device *dev,
1224 					   netdev_features_t features)
1225 {
1226 	struct veth_priv *priv = netdev_priv(dev);
1227 	struct net_device *peer;
1228 
1229 	peer = rtnl_dereference(priv->peer);
1230 	if (peer) {
1231 		struct veth_priv *peer_priv = netdev_priv(peer);
1232 
1233 		if (peer_priv->_xdp_prog)
1234 			features &= ~NETIF_F_GSO_SOFTWARE;
1235 	}
1236 	if (priv->_xdp_prog)
1237 		features |= NETIF_F_GRO;
1238 
1239 	return features;
1240 }
1241 
1242 static int veth_set_features(struct net_device *dev,
1243 			     netdev_features_t features)
1244 {
1245 	netdev_features_t changed = features ^ dev->features;
1246 	struct veth_priv *priv = netdev_priv(dev);
1247 	int err;
1248 
1249 	if (!(changed & NETIF_F_GRO) || !(dev->flags & IFF_UP) || priv->_xdp_prog)
1250 		return 0;
1251 
1252 	if (features & NETIF_F_GRO) {
1253 		err = veth_napi_enable(dev);
1254 		if (err)
1255 			return err;
1256 	} else {
1257 		veth_napi_del(dev);
1258 	}
1259 	return 0;
1260 }
1261 
1262 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1263 {
1264 	struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1265 	struct net_device *peer;
1266 
1267 	if (new_hr < 0)
1268 		new_hr = 0;
1269 
1270 	rcu_read_lock();
1271 	peer = rcu_dereference(priv->peer);
1272 	if (unlikely(!peer))
1273 		goto out;
1274 
1275 	peer_priv = netdev_priv(peer);
1276 	priv->requested_headroom = new_hr;
1277 	new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1278 	dev->needed_headroom = new_hr;
1279 	peer->needed_headroom = new_hr;
1280 
1281 out:
1282 	rcu_read_unlock();
1283 }
1284 
1285 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1286 			struct netlink_ext_ack *extack)
1287 {
1288 	struct veth_priv *priv = netdev_priv(dev);
1289 	struct bpf_prog *old_prog;
1290 	struct net_device *peer;
1291 	unsigned int max_mtu;
1292 	int err;
1293 
1294 	old_prog = priv->_xdp_prog;
1295 	priv->_xdp_prog = prog;
1296 	peer = rtnl_dereference(priv->peer);
1297 
1298 	if (prog) {
1299 		if (!peer) {
1300 			NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1301 			err = -ENOTCONN;
1302 			goto err;
1303 		}
1304 
1305 		max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
1306 			  peer->hard_header_len -
1307 			  SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1308 		if (peer->mtu > max_mtu) {
1309 			NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1310 			err = -ERANGE;
1311 			goto err;
1312 		}
1313 
1314 		if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1315 			NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1316 			err = -ENOSPC;
1317 			goto err;
1318 		}
1319 
1320 		if (dev->flags & IFF_UP) {
1321 			err = veth_enable_xdp(dev);
1322 			if (err) {
1323 				NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1324 				goto err;
1325 			}
1326 		}
1327 
1328 		if (!old_prog) {
1329 			peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1330 			peer->max_mtu = max_mtu;
1331 		}
1332 	}
1333 
1334 	if (old_prog) {
1335 		if (!prog) {
1336 			if (dev->flags & IFF_UP)
1337 				veth_disable_xdp(dev);
1338 
1339 			if (peer) {
1340 				peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1341 				peer->max_mtu = ETH_MAX_MTU;
1342 			}
1343 		}
1344 		bpf_prog_put(old_prog);
1345 	}
1346 
1347 	if ((!!old_prog ^ !!prog) && peer)
1348 		netdev_update_features(peer);
1349 
1350 	return 0;
1351 err:
1352 	priv->_xdp_prog = old_prog;
1353 
1354 	return err;
1355 }
1356 
1357 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1358 {
1359 	switch (xdp->command) {
1360 	case XDP_SETUP_PROG:
1361 		return veth_xdp_set(dev, xdp->prog, xdp->extack);
1362 	default:
1363 		return -EINVAL;
1364 	}
1365 }
1366 
1367 static const struct net_device_ops veth_netdev_ops = {
1368 	.ndo_init            = veth_dev_init,
1369 	.ndo_open            = veth_open,
1370 	.ndo_stop            = veth_close,
1371 	.ndo_start_xmit      = veth_xmit,
1372 	.ndo_get_stats64     = veth_get_stats64,
1373 	.ndo_set_rx_mode     = veth_set_multicast_list,
1374 	.ndo_set_mac_address = eth_mac_addr,
1375 #ifdef CONFIG_NET_POLL_CONTROLLER
1376 	.ndo_poll_controller	= veth_poll_controller,
1377 #endif
1378 	.ndo_get_iflink		= veth_get_iflink,
1379 	.ndo_fix_features	= veth_fix_features,
1380 	.ndo_set_features	= veth_set_features,
1381 	.ndo_features_check	= passthru_features_check,
1382 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
1383 	.ndo_bpf		= veth_xdp,
1384 	.ndo_xdp_xmit		= veth_ndo_xdp_xmit,
1385 	.ndo_get_peer_dev	= veth_peer_dev,
1386 };
1387 
1388 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1389 		       NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1390 		       NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1391 		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1392 		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1393 
1394 static void veth_setup(struct net_device *dev)
1395 {
1396 	ether_setup(dev);
1397 
1398 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1399 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1400 	dev->priv_flags |= IFF_NO_QUEUE;
1401 	dev->priv_flags |= IFF_PHONY_HEADROOM;
1402 
1403 	dev->netdev_ops = &veth_netdev_ops;
1404 	dev->ethtool_ops = &veth_ethtool_ops;
1405 	dev->features |= NETIF_F_LLTX;
1406 	dev->features |= VETH_FEATURES;
1407 	dev->vlan_features = dev->features &
1408 			     ~(NETIF_F_HW_VLAN_CTAG_TX |
1409 			       NETIF_F_HW_VLAN_STAG_TX |
1410 			       NETIF_F_HW_VLAN_CTAG_RX |
1411 			       NETIF_F_HW_VLAN_STAG_RX);
1412 	dev->needs_free_netdev = true;
1413 	dev->priv_destructor = veth_dev_free;
1414 	dev->max_mtu = ETH_MAX_MTU;
1415 
1416 	dev->hw_features = VETH_FEATURES;
1417 	dev->hw_enc_features = VETH_FEATURES;
1418 	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1419 }
1420 
1421 /*
1422  * netlink interface
1423  */
1424 
1425 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1426 			 struct netlink_ext_ack *extack)
1427 {
1428 	if (tb[IFLA_ADDRESS]) {
1429 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1430 			return -EINVAL;
1431 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1432 			return -EADDRNOTAVAIL;
1433 	}
1434 	if (tb[IFLA_MTU]) {
1435 		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1436 			return -EINVAL;
1437 	}
1438 	return 0;
1439 }
1440 
1441 static struct rtnl_link_ops veth_link_ops;
1442 
1443 static void veth_disable_gro(struct net_device *dev)
1444 {
1445 	dev->features &= ~NETIF_F_GRO;
1446 	dev->wanted_features &= ~NETIF_F_GRO;
1447 	netdev_update_features(dev);
1448 }
1449 
1450 static int veth_newlink(struct net *src_net, struct net_device *dev,
1451 			struct nlattr *tb[], struct nlattr *data[],
1452 			struct netlink_ext_ack *extack)
1453 {
1454 	int err;
1455 	struct net_device *peer;
1456 	struct veth_priv *priv;
1457 	char ifname[IFNAMSIZ];
1458 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1459 	unsigned char name_assign_type;
1460 	struct ifinfomsg *ifmp;
1461 	struct net *net;
1462 
1463 	/*
1464 	 * create and register peer first
1465 	 */
1466 	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1467 		struct nlattr *nla_peer;
1468 
1469 		nla_peer = data[VETH_INFO_PEER];
1470 		ifmp = nla_data(nla_peer);
1471 		err = rtnl_nla_parse_ifla(peer_tb,
1472 					  nla_data(nla_peer) + sizeof(struct ifinfomsg),
1473 					  nla_len(nla_peer) - sizeof(struct ifinfomsg),
1474 					  NULL);
1475 		if (err < 0)
1476 			return err;
1477 
1478 		err = veth_validate(peer_tb, NULL, extack);
1479 		if (err < 0)
1480 			return err;
1481 
1482 		tbp = peer_tb;
1483 	} else {
1484 		ifmp = NULL;
1485 		tbp = tb;
1486 	}
1487 
1488 	if (ifmp && tbp[IFLA_IFNAME]) {
1489 		nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1490 		name_assign_type = NET_NAME_USER;
1491 	} else {
1492 		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1493 		name_assign_type = NET_NAME_ENUM;
1494 	}
1495 
1496 	net = rtnl_link_get_net(src_net, tbp);
1497 	if (IS_ERR(net))
1498 		return PTR_ERR(net);
1499 
1500 	peer = rtnl_create_link(net, ifname, name_assign_type,
1501 				&veth_link_ops, tbp, extack);
1502 	if (IS_ERR(peer)) {
1503 		put_net(net);
1504 		return PTR_ERR(peer);
1505 	}
1506 
1507 	if (!ifmp || !tbp[IFLA_ADDRESS])
1508 		eth_hw_addr_random(peer);
1509 
1510 	if (ifmp && (dev->ifindex != 0))
1511 		peer->ifindex = ifmp->ifi_index;
1512 
1513 	peer->gso_max_size = dev->gso_max_size;
1514 	peer->gso_max_segs = dev->gso_max_segs;
1515 
1516 	err = register_netdevice(peer);
1517 	put_net(net);
1518 	net = NULL;
1519 	if (err < 0)
1520 		goto err_register_peer;
1521 
1522 	/* keep GRO disabled by default to be consistent with the established
1523 	 * veth behavior
1524 	 */
1525 	veth_disable_gro(peer);
1526 	netif_carrier_off(peer);
1527 
1528 	err = rtnl_configure_link(peer, ifmp);
1529 	if (err < 0)
1530 		goto err_configure_peer;
1531 
1532 	/*
1533 	 * register dev last
1534 	 *
1535 	 * note, that since we've registered new device the dev's name
1536 	 * should be re-allocated
1537 	 */
1538 
1539 	if (tb[IFLA_ADDRESS] == NULL)
1540 		eth_hw_addr_random(dev);
1541 
1542 	if (tb[IFLA_IFNAME])
1543 		nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1544 	else
1545 		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1546 
1547 	err = register_netdevice(dev);
1548 	if (err < 0)
1549 		goto err_register_dev;
1550 
1551 	netif_carrier_off(dev);
1552 
1553 	/*
1554 	 * tie the deviced together
1555 	 */
1556 
1557 	priv = netdev_priv(dev);
1558 	rcu_assign_pointer(priv->peer, peer);
1559 
1560 	priv = netdev_priv(peer);
1561 	rcu_assign_pointer(priv->peer, dev);
1562 
1563 	veth_disable_gro(dev);
1564 	return 0;
1565 
1566 err_register_dev:
1567 	/* nothing to do */
1568 err_configure_peer:
1569 	unregister_netdevice(peer);
1570 	return err;
1571 
1572 err_register_peer:
1573 	free_netdev(peer);
1574 	return err;
1575 }
1576 
1577 static void veth_dellink(struct net_device *dev, struct list_head *head)
1578 {
1579 	struct veth_priv *priv;
1580 	struct net_device *peer;
1581 
1582 	priv = netdev_priv(dev);
1583 	peer = rtnl_dereference(priv->peer);
1584 
1585 	/* Note : dellink() is called from default_device_exit_batch(),
1586 	 * before a rcu_synchronize() point. The devices are guaranteed
1587 	 * not being freed before one RCU grace period.
1588 	 */
1589 	RCU_INIT_POINTER(priv->peer, NULL);
1590 	unregister_netdevice_queue(dev, head);
1591 
1592 	if (peer) {
1593 		priv = netdev_priv(peer);
1594 		RCU_INIT_POINTER(priv->peer, NULL);
1595 		unregister_netdevice_queue(peer, head);
1596 	}
1597 }
1598 
1599 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
1600 	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
1601 };
1602 
1603 static struct net *veth_get_link_net(const struct net_device *dev)
1604 {
1605 	struct veth_priv *priv = netdev_priv(dev);
1606 	struct net_device *peer = rtnl_dereference(priv->peer);
1607 
1608 	return peer ? dev_net(peer) : dev_net(dev);
1609 }
1610 
1611 static struct rtnl_link_ops veth_link_ops = {
1612 	.kind		= DRV_NAME,
1613 	.priv_size	= sizeof(struct veth_priv),
1614 	.setup		= veth_setup,
1615 	.validate	= veth_validate,
1616 	.newlink	= veth_newlink,
1617 	.dellink	= veth_dellink,
1618 	.policy		= veth_policy,
1619 	.maxtype	= VETH_INFO_MAX,
1620 	.get_link_net	= veth_get_link_net,
1621 };
1622 
1623 /*
1624  * init/fini
1625  */
1626 
1627 static __init int veth_init(void)
1628 {
1629 	return rtnl_link_register(&veth_link_ops);
1630 }
1631 
1632 static __exit void veth_exit(void)
1633 {
1634 	rtnl_link_unregister(&veth_link_ops);
1635 }
1636 
1637 module_init(veth_init);
1638 module_exit(veth_exit);
1639 
1640 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
1641 MODULE_LICENSE("GPL v2");
1642 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1643