xref: /linux/drivers/net/veth.c (revision 9f6c532f59b20580acf8ede9409c9b8dce6e74e1)
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 #include <net/page_pool/helpers.h>
30 
31 #define DRV_NAME	"veth"
32 #define DRV_VERSION	"1.0"
33 
34 #define VETH_XDP_FLAG		BIT(0)
35 #define VETH_RING_SIZE		256
36 #define VETH_XDP_HEADROOM	(XDP_PACKET_HEADROOM + NET_IP_ALIGN)
37 
38 #define VETH_XDP_TX_BULK_SIZE	16
39 #define VETH_XDP_BATCH		16
40 
41 struct veth_stats {
42 	u64	rx_drops;
43 	/* xdp */
44 	u64	xdp_packets;
45 	u64	xdp_bytes;
46 	u64	xdp_redirect;
47 	u64	xdp_drops;
48 	u64	xdp_tx;
49 	u64	xdp_tx_err;
50 	u64	peer_tq_xdp_xmit;
51 	u64	peer_tq_xdp_xmit_err;
52 };
53 
54 struct veth_rq_stats {
55 	struct veth_stats	vs;
56 	struct u64_stats_sync	syncp;
57 };
58 
59 struct veth_rq {
60 	struct napi_struct	xdp_napi;
61 	struct napi_struct __rcu *napi; /* points to xdp_napi when the latter is initialized */
62 	struct net_device	*dev;
63 	struct bpf_prog __rcu	*xdp_prog;
64 	struct xdp_mem_info	xdp_mem;
65 	struct veth_rq_stats	stats;
66 	bool			rx_notify_masked;
67 	struct ptr_ring		xdp_ring;
68 	struct xdp_rxq_info	xdp_rxq;
69 	struct page_pool	*page_pool;
70 };
71 
72 struct veth_priv {
73 	struct net_device __rcu	*peer;
74 	atomic64_t		dropped;
75 	struct bpf_prog		*_xdp_prog;
76 	struct veth_rq		*rq;
77 	unsigned int		requested_headroom;
78 };
79 
80 struct veth_xdp_tx_bq {
81 	struct xdp_frame *q[VETH_XDP_TX_BULK_SIZE];
82 	unsigned int count;
83 };
84 
85 /*
86  * ethtool interface
87  */
88 
89 struct veth_q_stat_desc {
90 	char	desc[ETH_GSTRING_LEN];
91 	size_t	offset;
92 };
93 
94 #define VETH_RQ_STAT(m)	offsetof(struct veth_stats, m)
95 
96 static const struct veth_q_stat_desc veth_rq_stats_desc[] = {
97 	{ "xdp_packets",	VETH_RQ_STAT(xdp_packets) },
98 	{ "xdp_bytes",		VETH_RQ_STAT(xdp_bytes) },
99 	{ "drops",		VETH_RQ_STAT(rx_drops) },
100 	{ "xdp_redirect",	VETH_RQ_STAT(xdp_redirect) },
101 	{ "xdp_drops",		VETH_RQ_STAT(xdp_drops) },
102 	{ "xdp_tx",		VETH_RQ_STAT(xdp_tx) },
103 	{ "xdp_tx_errors",	VETH_RQ_STAT(xdp_tx_err) },
104 };
105 
106 #define VETH_RQ_STATS_LEN	ARRAY_SIZE(veth_rq_stats_desc)
107 
108 static const struct veth_q_stat_desc veth_tq_stats_desc[] = {
109 	{ "xdp_xmit",		VETH_RQ_STAT(peer_tq_xdp_xmit) },
110 	{ "xdp_xmit_errors",	VETH_RQ_STAT(peer_tq_xdp_xmit_err) },
111 };
112 
113 #define VETH_TQ_STATS_LEN	ARRAY_SIZE(veth_tq_stats_desc)
114 
115 static struct {
116 	const char string[ETH_GSTRING_LEN];
117 } ethtool_stats_keys[] = {
118 	{ "peer_ifindex" },
119 };
120 
121 struct veth_xdp_buff {
122 	struct xdp_buff xdp;
123 	struct sk_buff *skb;
124 };
125 
126 static int veth_get_link_ksettings(struct net_device *dev,
127 				   struct ethtool_link_ksettings *cmd)
128 {
129 	cmd->base.speed		= SPEED_10000;
130 	cmd->base.duplex	= DUPLEX_FULL;
131 	cmd->base.port		= PORT_TP;
132 	cmd->base.autoneg	= AUTONEG_DISABLE;
133 	return 0;
134 }
135 
136 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
137 {
138 	strscpy(info->driver, DRV_NAME, sizeof(info->driver));
139 	strscpy(info->version, DRV_VERSION, sizeof(info->version));
140 }
141 
142 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
143 {
144 	u8 *p = buf;
145 	int i, j;
146 
147 	switch(stringset) {
148 	case ETH_SS_STATS:
149 		memcpy(p, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
150 		p += sizeof(ethtool_stats_keys);
151 		for (i = 0; i < dev->real_num_rx_queues; i++)
152 			for (j = 0; j < VETH_RQ_STATS_LEN; j++)
153 				ethtool_sprintf(&p, "rx_queue_%u_%.18s",
154 						i, veth_rq_stats_desc[j].desc);
155 
156 		for (i = 0; i < dev->real_num_tx_queues; i++)
157 			for (j = 0; j < VETH_TQ_STATS_LEN; j++)
158 				ethtool_sprintf(&p, "tx_queue_%u_%.18s",
159 						i, veth_tq_stats_desc[j].desc);
160 
161 		page_pool_ethtool_stats_get_strings(p);
162 		break;
163 	}
164 }
165 
166 static int veth_get_sset_count(struct net_device *dev, int sset)
167 {
168 	switch (sset) {
169 	case ETH_SS_STATS:
170 		return ARRAY_SIZE(ethtool_stats_keys) +
171 		       VETH_RQ_STATS_LEN * dev->real_num_rx_queues +
172 		       VETH_TQ_STATS_LEN * dev->real_num_tx_queues +
173 		       page_pool_ethtool_stats_get_count();
174 	default:
175 		return -EOPNOTSUPP;
176 	}
177 }
178 
179 static void veth_get_page_pool_stats(struct net_device *dev, u64 *data)
180 {
181 #ifdef CONFIG_PAGE_POOL_STATS
182 	struct veth_priv *priv = netdev_priv(dev);
183 	struct page_pool_stats pp_stats = {};
184 	int i;
185 
186 	for (i = 0; i < dev->real_num_rx_queues; i++) {
187 		if (!priv->rq[i].page_pool)
188 			continue;
189 		page_pool_get_stats(priv->rq[i].page_pool, &pp_stats);
190 	}
191 	page_pool_ethtool_stats_get(data, &pp_stats);
192 #endif /* CONFIG_PAGE_POOL_STATS */
193 }
194 
195 static void veth_get_ethtool_stats(struct net_device *dev,
196 		struct ethtool_stats *stats, u64 *data)
197 {
198 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
199 	struct net_device *peer = rtnl_dereference(priv->peer);
200 	int i, j, idx, pp_idx;
201 
202 	data[0] = peer ? peer->ifindex : 0;
203 	idx = 1;
204 	for (i = 0; i < dev->real_num_rx_queues; i++) {
205 		const struct veth_rq_stats *rq_stats = &priv->rq[i].stats;
206 		const void *stats_base = (void *)&rq_stats->vs;
207 		unsigned int start;
208 		size_t offset;
209 
210 		do {
211 			start = u64_stats_fetch_begin(&rq_stats->syncp);
212 			for (j = 0; j < VETH_RQ_STATS_LEN; j++) {
213 				offset = veth_rq_stats_desc[j].offset;
214 				data[idx + j] = *(u64 *)(stats_base + offset);
215 			}
216 		} while (u64_stats_fetch_retry(&rq_stats->syncp, start));
217 		idx += VETH_RQ_STATS_LEN;
218 	}
219 	pp_idx = idx;
220 
221 	if (!peer)
222 		goto page_pool_stats;
223 
224 	rcv_priv = netdev_priv(peer);
225 	for (i = 0; i < peer->real_num_rx_queues; i++) {
226 		const struct veth_rq_stats *rq_stats = &rcv_priv->rq[i].stats;
227 		const void *base = (void *)&rq_stats->vs;
228 		unsigned int start, tx_idx = idx;
229 		size_t offset;
230 
231 		tx_idx += (i % dev->real_num_tx_queues) * VETH_TQ_STATS_LEN;
232 		do {
233 			start = u64_stats_fetch_begin(&rq_stats->syncp);
234 			for (j = 0; j < VETH_TQ_STATS_LEN; j++) {
235 				offset = veth_tq_stats_desc[j].offset;
236 				data[tx_idx + j] += *(u64 *)(base + offset);
237 			}
238 		} while (u64_stats_fetch_retry(&rq_stats->syncp, start));
239 		pp_idx = tx_idx + VETH_TQ_STATS_LEN;
240 	}
241 
242 page_pool_stats:
243 	veth_get_page_pool_stats(dev, &data[pp_idx]);
244 }
245 
246 static void veth_get_channels(struct net_device *dev,
247 			      struct ethtool_channels *channels)
248 {
249 	channels->tx_count = dev->real_num_tx_queues;
250 	channels->rx_count = dev->real_num_rx_queues;
251 	channels->max_tx = dev->num_tx_queues;
252 	channels->max_rx = dev->num_rx_queues;
253 }
254 
255 static int veth_set_channels(struct net_device *dev,
256 			     struct ethtool_channels *ch);
257 
258 static const struct ethtool_ops veth_ethtool_ops = {
259 	.get_drvinfo		= veth_get_drvinfo,
260 	.get_link		= ethtool_op_get_link,
261 	.get_strings		= veth_get_strings,
262 	.get_sset_count		= veth_get_sset_count,
263 	.get_ethtool_stats	= veth_get_ethtool_stats,
264 	.get_link_ksettings	= veth_get_link_ksettings,
265 	.get_ts_info		= ethtool_op_get_ts_info,
266 	.get_channels		= veth_get_channels,
267 	.set_channels		= veth_set_channels,
268 };
269 
270 /* general routines */
271 
272 static bool veth_is_xdp_frame(void *ptr)
273 {
274 	return (unsigned long)ptr & VETH_XDP_FLAG;
275 }
276 
277 static struct xdp_frame *veth_ptr_to_xdp(void *ptr)
278 {
279 	return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG);
280 }
281 
282 static void *veth_xdp_to_ptr(struct xdp_frame *xdp)
283 {
284 	return (void *)((unsigned long)xdp | VETH_XDP_FLAG);
285 }
286 
287 static void veth_ptr_free(void *ptr)
288 {
289 	if (veth_is_xdp_frame(ptr))
290 		xdp_return_frame(veth_ptr_to_xdp(ptr));
291 	else
292 		kfree_skb(ptr);
293 }
294 
295 static void __veth_xdp_flush(struct veth_rq *rq)
296 {
297 	/* Write ptr_ring before reading rx_notify_masked */
298 	smp_mb();
299 	if (!READ_ONCE(rq->rx_notify_masked) &&
300 	    napi_schedule_prep(&rq->xdp_napi)) {
301 		WRITE_ONCE(rq->rx_notify_masked, true);
302 		__napi_schedule(&rq->xdp_napi);
303 	}
304 }
305 
306 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
307 {
308 	if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
309 		dev_kfree_skb_any(skb);
310 		return NET_RX_DROP;
311 	}
312 
313 	return NET_RX_SUCCESS;
314 }
315 
316 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
317 			    struct veth_rq *rq, bool xdp)
318 {
319 	return __dev_forward_skb(dev, skb) ?: xdp ?
320 		veth_xdp_rx(rq, skb) :
321 		__netif_rx(skb);
322 }
323 
324 /* return true if the specified skb has chances of GRO aggregation
325  * Don't strive for accuracy, but try to avoid GRO overhead in the most
326  * common scenarios.
327  * When XDP is enabled, all traffic is considered eligible, as the xmit
328  * device has TSO off.
329  * When TSO is enabled on the xmit device, we are likely interested only
330  * in UDP aggregation, explicitly check for that if the skb is suspected
331  * - the sock_wfree destructor is used by UDP, ICMP and XDP sockets -
332  * to belong to locally generated UDP traffic.
333  */
334 static bool veth_skb_is_eligible_for_gro(const struct net_device *dev,
335 					 const struct net_device *rcv,
336 					 const struct sk_buff *skb)
337 {
338 	return !(dev->features & NETIF_F_ALL_TSO) ||
339 		(skb->destructor == sock_wfree &&
340 		 rcv->features & (NETIF_F_GRO_FRAGLIST | NETIF_F_GRO_UDP_FWD));
341 }
342 
343 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
344 {
345 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
346 	struct veth_rq *rq = NULL;
347 	int ret = NETDEV_TX_OK;
348 	struct net_device *rcv;
349 	int length = skb->len;
350 	bool use_napi = false;
351 	int rxq;
352 
353 	rcu_read_lock();
354 	rcv = rcu_dereference(priv->peer);
355 	if (unlikely(!rcv) || !pskb_may_pull(skb, ETH_HLEN)) {
356 		kfree_skb(skb);
357 		goto drop;
358 	}
359 
360 	rcv_priv = netdev_priv(rcv);
361 	rxq = skb_get_queue_mapping(skb);
362 	if (rxq < rcv->real_num_rx_queues) {
363 		rq = &rcv_priv->rq[rxq];
364 
365 		/* The napi pointer is available when an XDP program is
366 		 * attached or when GRO is enabled
367 		 * Don't bother with napi/GRO if the skb can't be aggregated
368 		 */
369 		use_napi = rcu_access_pointer(rq->napi) &&
370 			   veth_skb_is_eligible_for_gro(dev, rcv, skb);
371 	}
372 
373 	skb_tx_timestamp(skb);
374 	if (likely(veth_forward_skb(rcv, skb, rq, use_napi) == NET_RX_SUCCESS)) {
375 		if (!use_napi)
376 			dev_lstats_add(dev, length);
377 		else
378 			__veth_xdp_flush(rq);
379 	} else {
380 drop:
381 		atomic64_inc(&priv->dropped);
382 		ret = NET_XMIT_DROP;
383 	}
384 
385 	rcu_read_unlock();
386 
387 	return ret;
388 }
389 
390 static u64 veth_stats_tx(struct net_device *dev, u64 *packets, u64 *bytes)
391 {
392 	struct veth_priv *priv = netdev_priv(dev);
393 
394 	dev_lstats_read(dev, packets, bytes);
395 	return atomic64_read(&priv->dropped);
396 }
397 
398 static void veth_stats_rx(struct veth_stats *result, struct net_device *dev)
399 {
400 	struct veth_priv *priv = netdev_priv(dev);
401 	int i;
402 
403 	result->peer_tq_xdp_xmit_err = 0;
404 	result->xdp_packets = 0;
405 	result->xdp_tx_err = 0;
406 	result->xdp_bytes = 0;
407 	result->rx_drops = 0;
408 	for (i = 0; i < dev->num_rx_queues; i++) {
409 		u64 packets, bytes, drops, xdp_tx_err, peer_tq_xdp_xmit_err;
410 		struct veth_rq_stats *stats = &priv->rq[i].stats;
411 		unsigned int start;
412 
413 		do {
414 			start = u64_stats_fetch_begin(&stats->syncp);
415 			peer_tq_xdp_xmit_err = stats->vs.peer_tq_xdp_xmit_err;
416 			xdp_tx_err = stats->vs.xdp_tx_err;
417 			packets = stats->vs.xdp_packets;
418 			bytes = stats->vs.xdp_bytes;
419 			drops = stats->vs.rx_drops;
420 		} while (u64_stats_fetch_retry(&stats->syncp, start));
421 		result->peer_tq_xdp_xmit_err += peer_tq_xdp_xmit_err;
422 		result->xdp_tx_err += xdp_tx_err;
423 		result->xdp_packets += packets;
424 		result->xdp_bytes += bytes;
425 		result->rx_drops += drops;
426 	}
427 }
428 
429 static void veth_get_stats64(struct net_device *dev,
430 			     struct rtnl_link_stats64 *tot)
431 {
432 	struct veth_priv *priv = netdev_priv(dev);
433 	struct net_device *peer;
434 	struct veth_stats rx;
435 	u64 packets, bytes;
436 
437 	tot->tx_dropped = veth_stats_tx(dev, &packets, &bytes);
438 	tot->tx_bytes = bytes;
439 	tot->tx_packets = packets;
440 
441 	veth_stats_rx(&rx, dev);
442 	tot->tx_dropped += rx.xdp_tx_err;
443 	tot->rx_dropped = rx.rx_drops + rx.peer_tq_xdp_xmit_err;
444 	tot->rx_bytes = rx.xdp_bytes;
445 	tot->rx_packets = rx.xdp_packets;
446 
447 	rcu_read_lock();
448 	peer = rcu_dereference(priv->peer);
449 	if (peer) {
450 		veth_stats_tx(peer, &packets, &bytes);
451 		tot->rx_bytes += bytes;
452 		tot->rx_packets += packets;
453 
454 		veth_stats_rx(&rx, peer);
455 		tot->tx_dropped += rx.peer_tq_xdp_xmit_err;
456 		tot->rx_dropped += rx.xdp_tx_err;
457 		tot->tx_bytes += rx.xdp_bytes;
458 		tot->tx_packets += rx.xdp_packets;
459 	}
460 	rcu_read_unlock();
461 }
462 
463 /* fake multicast ability */
464 static void veth_set_multicast_list(struct net_device *dev)
465 {
466 }
467 
468 static int veth_select_rxq(struct net_device *dev)
469 {
470 	return smp_processor_id() % dev->real_num_rx_queues;
471 }
472 
473 static struct net_device *veth_peer_dev(struct net_device *dev)
474 {
475 	struct veth_priv *priv = netdev_priv(dev);
476 
477 	/* Callers must be under RCU read side. */
478 	return rcu_dereference(priv->peer);
479 }
480 
481 static int veth_xdp_xmit(struct net_device *dev, int n,
482 			 struct xdp_frame **frames,
483 			 u32 flags, bool ndo_xmit)
484 {
485 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
486 	int i, ret = -ENXIO, nxmit = 0;
487 	struct net_device *rcv;
488 	unsigned int max_len;
489 	struct veth_rq *rq;
490 
491 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
492 		return -EINVAL;
493 
494 	rcu_read_lock();
495 	rcv = rcu_dereference(priv->peer);
496 	if (unlikely(!rcv))
497 		goto out;
498 
499 	rcv_priv = netdev_priv(rcv);
500 	rq = &rcv_priv->rq[veth_select_rxq(rcv)];
501 	/* The napi pointer is set if NAPI is enabled, which ensures that
502 	 * xdp_ring is initialized on receive side and the peer device is up.
503 	 */
504 	if (!rcu_access_pointer(rq->napi))
505 		goto out;
506 
507 	max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN;
508 
509 	spin_lock(&rq->xdp_ring.producer_lock);
510 	for (i = 0; i < n; i++) {
511 		struct xdp_frame *frame = frames[i];
512 		void *ptr = veth_xdp_to_ptr(frame);
513 
514 		if (unlikely(xdp_get_frame_len(frame) > max_len ||
515 			     __ptr_ring_produce(&rq->xdp_ring, ptr)))
516 			break;
517 		nxmit++;
518 	}
519 	spin_unlock(&rq->xdp_ring.producer_lock);
520 
521 	if (flags & XDP_XMIT_FLUSH)
522 		__veth_xdp_flush(rq);
523 
524 	ret = nxmit;
525 	if (ndo_xmit) {
526 		u64_stats_update_begin(&rq->stats.syncp);
527 		rq->stats.vs.peer_tq_xdp_xmit += nxmit;
528 		rq->stats.vs.peer_tq_xdp_xmit_err += n - nxmit;
529 		u64_stats_update_end(&rq->stats.syncp);
530 	}
531 
532 out:
533 	rcu_read_unlock();
534 
535 	return ret;
536 }
537 
538 static int veth_ndo_xdp_xmit(struct net_device *dev, int n,
539 			     struct xdp_frame **frames, u32 flags)
540 {
541 	int err;
542 
543 	err = veth_xdp_xmit(dev, n, frames, flags, true);
544 	if (err < 0) {
545 		struct veth_priv *priv = netdev_priv(dev);
546 
547 		atomic64_add(n, &priv->dropped);
548 	}
549 
550 	return err;
551 }
552 
553 static void veth_xdp_flush_bq(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
554 {
555 	int sent, i, err = 0, drops;
556 
557 	sent = veth_xdp_xmit(rq->dev, bq->count, bq->q, 0, false);
558 	if (sent < 0) {
559 		err = sent;
560 		sent = 0;
561 	}
562 
563 	for (i = sent; unlikely(i < bq->count); i++)
564 		xdp_return_frame(bq->q[i]);
565 
566 	drops = bq->count - sent;
567 	trace_xdp_bulk_tx(rq->dev, sent, drops, err);
568 
569 	u64_stats_update_begin(&rq->stats.syncp);
570 	rq->stats.vs.xdp_tx += sent;
571 	rq->stats.vs.xdp_tx_err += drops;
572 	u64_stats_update_end(&rq->stats.syncp);
573 
574 	bq->count = 0;
575 }
576 
577 static void veth_xdp_flush(struct veth_rq *rq, struct veth_xdp_tx_bq *bq)
578 {
579 	struct veth_priv *rcv_priv, *priv = netdev_priv(rq->dev);
580 	struct net_device *rcv;
581 	struct veth_rq *rcv_rq;
582 
583 	rcu_read_lock();
584 	veth_xdp_flush_bq(rq, bq);
585 	rcv = rcu_dereference(priv->peer);
586 	if (unlikely(!rcv))
587 		goto out;
588 
589 	rcv_priv = netdev_priv(rcv);
590 	rcv_rq = &rcv_priv->rq[veth_select_rxq(rcv)];
591 	/* xdp_ring is initialized on receive side? */
592 	if (unlikely(!rcu_access_pointer(rcv_rq->xdp_prog)))
593 		goto out;
594 
595 	__veth_xdp_flush(rcv_rq);
596 out:
597 	rcu_read_unlock();
598 }
599 
600 static int veth_xdp_tx(struct veth_rq *rq, struct xdp_buff *xdp,
601 		       struct veth_xdp_tx_bq *bq)
602 {
603 	struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
604 
605 	if (unlikely(!frame))
606 		return -EOVERFLOW;
607 
608 	if (unlikely(bq->count == VETH_XDP_TX_BULK_SIZE))
609 		veth_xdp_flush_bq(rq, bq);
610 
611 	bq->q[bq->count++] = frame;
612 
613 	return 0;
614 }
615 
616 static struct xdp_frame *veth_xdp_rcv_one(struct veth_rq *rq,
617 					  struct xdp_frame *frame,
618 					  struct veth_xdp_tx_bq *bq,
619 					  struct veth_stats *stats)
620 {
621 	struct xdp_frame orig_frame;
622 	struct bpf_prog *xdp_prog;
623 
624 	rcu_read_lock();
625 	xdp_prog = rcu_dereference(rq->xdp_prog);
626 	if (likely(xdp_prog)) {
627 		struct veth_xdp_buff vxbuf;
628 		struct xdp_buff *xdp = &vxbuf.xdp;
629 		u32 act;
630 
631 		xdp_convert_frame_to_buff(frame, xdp);
632 		xdp->rxq = &rq->xdp_rxq;
633 		vxbuf.skb = NULL;
634 
635 		act = bpf_prog_run_xdp(xdp_prog, xdp);
636 
637 		switch (act) {
638 		case XDP_PASS:
639 			if (xdp_update_frame_from_buff(xdp, frame))
640 				goto err_xdp;
641 			break;
642 		case XDP_TX:
643 			orig_frame = *frame;
644 			xdp->rxq->mem = frame->mem;
645 			if (unlikely(veth_xdp_tx(rq, xdp, bq) < 0)) {
646 				trace_xdp_exception(rq->dev, xdp_prog, act);
647 				frame = &orig_frame;
648 				stats->rx_drops++;
649 				goto err_xdp;
650 			}
651 			stats->xdp_tx++;
652 			rcu_read_unlock();
653 			goto xdp_xmit;
654 		case XDP_REDIRECT:
655 			orig_frame = *frame;
656 			xdp->rxq->mem = frame->mem;
657 			if (xdp_do_redirect(rq->dev, xdp, xdp_prog)) {
658 				frame = &orig_frame;
659 				stats->rx_drops++;
660 				goto err_xdp;
661 			}
662 			stats->xdp_redirect++;
663 			rcu_read_unlock();
664 			goto xdp_xmit;
665 		default:
666 			bpf_warn_invalid_xdp_action(rq->dev, xdp_prog, act);
667 			fallthrough;
668 		case XDP_ABORTED:
669 			trace_xdp_exception(rq->dev, xdp_prog, act);
670 			fallthrough;
671 		case XDP_DROP:
672 			stats->xdp_drops++;
673 			goto err_xdp;
674 		}
675 	}
676 	rcu_read_unlock();
677 
678 	return frame;
679 err_xdp:
680 	rcu_read_unlock();
681 	xdp_return_frame(frame);
682 xdp_xmit:
683 	return NULL;
684 }
685 
686 /* frames array contains VETH_XDP_BATCH at most */
687 static void veth_xdp_rcv_bulk_skb(struct veth_rq *rq, void **frames,
688 				  int n_xdpf, struct veth_xdp_tx_bq *bq,
689 				  struct veth_stats *stats)
690 {
691 	void *skbs[VETH_XDP_BATCH];
692 	int i;
693 
694 	if (xdp_alloc_skb_bulk(skbs, n_xdpf,
695 			       GFP_ATOMIC | __GFP_ZERO) < 0) {
696 		for (i = 0; i < n_xdpf; i++)
697 			xdp_return_frame(frames[i]);
698 		stats->rx_drops += n_xdpf;
699 
700 		return;
701 	}
702 
703 	for (i = 0; i < n_xdpf; i++) {
704 		struct sk_buff *skb = skbs[i];
705 
706 		skb = __xdp_build_skb_from_frame(frames[i], skb,
707 						 rq->dev);
708 		if (!skb) {
709 			xdp_return_frame(frames[i]);
710 			stats->rx_drops++;
711 			continue;
712 		}
713 		napi_gro_receive(&rq->xdp_napi, skb);
714 	}
715 }
716 
717 static void veth_xdp_get(struct xdp_buff *xdp)
718 {
719 	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
720 	int i;
721 
722 	get_page(virt_to_page(xdp->data));
723 	if (likely(!xdp_buff_has_frags(xdp)))
724 		return;
725 
726 	for (i = 0; i < sinfo->nr_frags; i++)
727 		__skb_frag_ref(&sinfo->frags[i]);
728 }
729 
730 static int veth_convert_skb_to_xdp_buff(struct veth_rq *rq,
731 					struct xdp_buff *xdp,
732 					struct sk_buff **pskb)
733 {
734 	struct sk_buff *skb = *pskb;
735 	u32 frame_sz;
736 
737 	if (skb_shared(skb) || skb_head_is_locked(skb) ||
738 	    skb_shinfo(skb)->nr_frags ||
739 	    skb_headroom(skb) < XDP_PACKET_HEADROOM) {
740 		u32 size, len, max_head_size, off;
741 		struct sk_buff *nskb;
742 		struct page *page;
743 		int i, head_off;
744 
745 		/* We need a private copy of the skb and data buffers since
746 		 * the ebpf program can modify it. We segment the original skb
747 		 * into order-0 pages without linearize it.
748 		 *
749 		 * Make sure we have enough space for linear and paged area
750 		 */
751 		max_head_size = SKB_WITH_OVERHEAD(PAGE_SIZE -
752 						  VETH_XDP_HEADROOM);
753 		if (skb->len > PAGE_SIZE * MAX_SKB_FRAGS + max_head_size)
754 			goto drop;
755 
756 		/* Allocate skb head */
757 		page = page_pool_dev_alloc_pages(rq->page_pool);
758 		if (!page)
759 			goto drop;
760 
761 		nskb = napi_build_skb(page_address(page), PAGE_SIZE);
762 		if (!nskb) {
763 			page_pool_put_full_page(rq->page_pool, page, true);
764 			goto drop;
765 		}
766 
767 		skb_reserve(nskb, VETH_XDP_HEADROOM);
768 		skb_copy_header(nskb, skb);
769 		skb_mark_for_recycle(nskb);
770 
771 		size = min_t(u32, skb->len, max_head_size);
772 		if (skb_copy_bits(skb, 0, nskb->data, size)) {
773 			consume_skb(nskb);
774 			goto drop;
775 		}
776 		skb_put(nskb, size);
777 
778 		head_off = skb_headroom(nskb) - skb_headroom(skb);
779 		skb_headers_offset_update(nskb, head_off);
780 
781 		/* Allocate paged area of new skb */
782 		off = size;
783 		len = skb->len - off;
784 
785 		for (i = 0; i < MAX_SKB_FRAGS && off < skb->len; i++) {
786 			page = page_pool_dev_alloc_pages(rq->page_pool);
787 			if (!page) {
788 				consume_skb(nskb);
789 				goto drop;
790 			}
791 
792 			size = min_t(u32, len, PAGE_SIZE);
793 			skb_add_rx_frag(nskb, i, page, 0, size, PAGE_SIZE);
794 			if (skb_copy_bits(skb, off, page_address(page),
795 					  size)) {
796 				consume_skb(nskb);
797 				goto drop;
798 			}
799 
800 			len -= size;
801 			off += size;
802 		}
803 
804 		consume_skb(skb);
805 		skb = nskb;
806 	}
807 
808 	/* SKB "head" area always have tailroom for skb_shared_info */
809 	frame_sz = skb_end_pointer(skb) - skb->head;
810 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
811 	xdp_init_buff(xdp, frame_sz, &rq->xdp_rxq);
812 	xdp_prepare_buff(xdp, skb->head, skb_headroom(skb),
813 			 skb_headlen(skb), true);
814 
815 	if (skb_is_nonlinear(skb)) {
816 		skb_shinfo(skb)->xdp_frags_size = skb->data_len;
817 		xdp_buff_set_frags_flag(xdp);
818 	} else {
819 		xdp_buff_clear_frags_flag(xdp);
820 	}
821 	*pskb = skb;
822 
823 	return 0;
824 drop:
825 	consume_skb(skb);
826 	*pskb = NULL;
827 
828 	return -ENOMEM;
829 }
830 
831 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
832 					struct sk_buff *skb,
833 					struct veth_xdp_tx_bq *bq,
834 					struct veth_stats *stats)
835 {
836 	void *orig_data, *orig_data_end;
837 	struct bpf_prog *xdp_prog;
838 	struct veth_xdp_buff vxbuf;
839 	struct xdp_buff *xdp = &vxbuf.xdp;
840 	u32 act, metalen;
841 	int off;
842 
843 	skb_prepare_for_gro(skb);
844 
845 	rcu_read_lock();
846 	xdp_prog = rcu_dereference(rq->xdp_prog);
847 	if (unlikely(!xdp_prog)) {
848 		rcu_read_unlock();
849 		goto out;
850 	}
851 
852 	__skb_push(skb, skb->data - skb_mac_header(skb));
853 	if (veth_convert_skb_to_xdp_buff(rq, xdp, &skb))
854 		goto drop;
855 	vxbuf.skb = skb;
856 
857 	orig_data = xdp->data;
858 	orig_data_end = xdp->data_end;
859 
860 	act = bpf_prog_run_xdp(xdp_prog, xdp);
861 
862 	switch (act) {
863 	case XDP_PASS:
864 		break;
865 	case XDP_TX:
866 		veth_xdp_get(xdp);
867 		consume_skb(skb);
868 		xdp->rxq->mem = rq->xdp_mem;
869 		if (unlikely(veth_xdp_tx(rq, xdp, bq) < 0)) {
870 			trace_xdp_exception(rq->dev, xdp_prog, act);
871 			stats->rx_drops++;
872 			goto err_xdp;
873 		}
874 		stats->xdp_tx++;
875 		rcu_read_unlock();
876 		goto xdp_xmit;
877 	case XDP_REDIRECT:
878 		veth_xdp_get(xdp);
879 		consume_skb(skb);
880 		xdp->rxq->mem = rq->xdp_mem;
881 		if (xdp_do_redirect(rq->dev, xdp, xdp_prog)) {
882 			stats->rx_drops++;
883 			goto err_xdp;
884 		}
885 		stats->xdp_redirect++;
886 		rcu_read_unlock();
887 		goto xdp_xmit;
888 	default:
889 		bpf_warn_invalid_xdp_action(rq->dev, xdp_prog, act);
890 		fallthrough;
891 	case XDP_ABORTED:
892 		trace_xdp_exception(rq->dev, xdp_prog, act);
893 		fallthrough;
894 	case XDP_DROP:
895 		stats->xdp_drops++;
896 		goto xdp_drop;
897 	}
898 	rcu_read_unlock();
899 
900 	/* check if bpf_xdp_adjust_head was used */
901 	off = orig_data - xdp->data;
902 	if (off > 0)
903 		__skb_push(skb, off);
904 	else if (off < 0)
905 		__skb_pull(skb, -off);
906 
907 	skb_reset_mac_header(skb);
908 
909 	/* check if bpf_xdp_adjust_tail was used */
910 	off = xdp->data_end - orig_data_end;
911 	if (off != 0)
912 		__skb_put(skb, off); /* positive on grow, negative on shrink */
913 
914 	/* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
915 	 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
916 	 */
917 	if (xdp_buff_has_frags(xdp))
918 		skb->data_len = skb_shinfo(skb)->xdp_frags_size;
919 	else
920 		skb->data_len = 0;
921 
922 	skb->protocol = eth_type_trans(skb, rq->dev);
923 
924 	metalen = xdp->data - xdp->data_meta;
925 	if (metalen)
926 		skb_metadata_set(skb, metalen);
927 out:
928 	return skb;
929 drop:
930 	stats->rx_drops++;
931 xdp_drop:
932 	rcu_read_unlock();
933 	kfree_skb(skb);
934 	return NULL;
935 err_xdp:
936 	rcu_read_unlock();
937 	xdp_return_buff(xdp);
938 xdp_xmit:
939 	return NULL;
940 }
941 
942 static int veth_xdp_rcv(struct veth_rq *rq, int budget,
943 			struct veth_xdp_tx_bq *bq,
944 			struct veth_stats *stats)
945 {
946 	int i, done = 0, n_xdpf = 0;
947 	void *xdpf[VETH_XDP_BATCH];
948 
949 	for (i = 0; i < budget; i++) {
950 		void *ptr = __ptr_ring_consume(&rq->xdp_ring);
951 
952 		if (!ptr)
953 			break;
954 
955 		if (veth_is_xdp_frame(ptr)) {
956 			/* ndo_xdp_xmit */
957 			struct xdp_frame *frame = veth_ptr_to_xdp(ptr);
958 
959 			stats->xdp_bytes += xdp_get_frame_len(frame);
960 			frame = veth_xdp_rcv_one(rq, frame, bq, stats);
961 			if (frame) {
962 				/* XDP_PASS */
963 				xdpf[n_xdpf++] = frame;
964 				if (n_xdpf == VETH_XDP_BATCH) {
965 					veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf,
966 							      bq, stats);
967 					n_xdpf = 0;
968 				}
969 			}
970 		} else {
971 			/* ndo_start_xmit */
972 			struct sk_buff *skb = ptr;
973 
974 			stats->xdp_bytes += skb->len;
975 			skb = veth_xdp_rcv_skb(rq, skb, bq, stats);
976 			if (skb) {
977 				if (skb_shared(skb) || skb_unclone(skb, GFP_ATOMIC))
978 					netif_receive_skb(skb);
979 				else
980 					napi_gro_receive(&rq->xdp_napi, skb);
981 			}
982 		}
983 		done++;
984 	}
985 
986 	if (n_xdpf)
987 		veth_xdp_rcv_bulk_skb(rq, xdpf, n_xdpf, bq, stats);
988 
989 	u64_stats_update_begin(&rq->stats.syncp);
990 	rq->stats.vs.xdp_redirect += stats->xdp_redirect;
991 	rq->stats.vs.xdp_bytes += stats->xdp_bytes;
992 	rq->stats.vs.xdp_drops += stats->xdp_drops;
993 	rq->stats.vs.rx_drops += stats->rx_drops;
994 	rq->stats.vs.xdp_packets += done;
995 	u64_stats_update_end(&rq->stats.syncp);
996 
997 	return done;
998 }
999 
1000 static int veth_poll(struct napi_struct *napi, int budget)
1001 {
1002 	struct veth_rq *rq =
1003 		container_of(napi, struct veth_rq, xdp_napi);
1004 	struct veth_stats stats = {};
1005 	struct veth_xdp_tx_bq bq;
1006 	int done;
1007 
1008 	bq.count = 0;
1009 
1010 	xdp_set_return_frame_no_direct();
1011 	done = veth_xdp_rcv(rq, budget, &bq, &stats);
1012 
1013 	if (stats.xdp_redirect > 0)
1014 		xdp_do_flush();
1015 
1016 	if (done < budget && napi_complete_done(napi, done)) {
1017 		/* Write rx_notify_masked before reading ptr_ring */
1018 		smp_store_mb(rq->rx_notify_masked, false);
1019 		if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) {
1020 			if (napi_schedule_prep(&rq->xdp_napi)) {
1021 				WRITE_ONCE(rq->rx_notify_masked, true);
1022 				__napi_schedule(&rq->xdp_napi);
1023 			}
1024 		}
1025 	}
1026 
1027 	if (stats.xdp_tx > 0)
1028 		veth_xdp_flush(rq, &bq);
1029 	xdp_clear_return_frame_no_direct();
1030 
1031 	return done;
1032 }
1033 
1034 static int veth_create_page_pool(struct veth_rq *rq)
1035 {
1036 	struct page_pool_params pp_params = {
1037 		.order = 0,
1038 		.pool_size = VETH_RING_SIZE,
1039 		.nid = NUMA_NO_NODE,
1040 		.dev = &rq->dev->dev,
1041 	};
1042 
1043 	rq->page_pool = page_pool_create(&pp_params);
1044 	if (IS_ERR(rq->page_pool)) {
1045 		int err = PTR_ERR(rq->page_pool);
1046 
1047 		rq->page_pool = NULL;
1048 		return err;
1049 	}
1050 
1051 	return 0;
1052 }
1053 
1054 static int __veth_napi_enable_range(struct net_device *dev, int start, int end)
1055 {
1056 	struct veth_priv *priv = netdev_priv(dev);
1057 	int err, i;
1058 
1059 	for (i = start; i < end; i++) {
1060 		err = veth_create_page_pool(&priv->rq[i]);
1061 		if (err)
1062 			goto err_page_pool;
1063 	}
1064 
1065 	for (i = start; i < end; i++) {
1066 		struct veth_rq *rq = &priv->rq[i];
1067 
1068 		err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
1069 		if (err)
1070 			goto err_xdp_ring;
1071 	}
1072 
1073 	for (i = start; i < end; i++) {
1074 		struct veth_rq *rq = &priv->rq[i];
1075 
1076 		napi_enable(&rq->xdp_napi);
1077 		rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1078 	}
1079 
1080 	return 0;
1081 
1082 err_xdp_ring:
1083 	for (i--; i >= start; i--)
1084 		ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free);
1085 	i = end;
1086 err_page_pool:
1087 	for (i--; i >= start; i--) {
1088 		page_pool_destroy(priv->rq[i].page_pool);
1089 		priv->rq[i].page_pool = NULL;
1090 	}
1091 
1092 	return err;
1093 }
1094 
1095 static int __veth_napi_enable(struct net_device *dev)
1096 {
1097 	return __veth_napi_enable_range(dev, 0, dev->real_num_rx_queues);
1098 }
1099 
1100 static void veth_napi_del_range(struct net_device *dev, int start, int end)
1101 {
1102 	struct veth_priv *priv = netdev_priv(dev);
1103 	int i;
1104 
1105 	for (i = start; i < end; i++) {
1106 		struct veth_rq *rq = &priv->rq[i];
1107 
1108 		rcu_assign_pointer(priv->rq[i].napi, NULL);
1109 		napi_disable(&rq->xdp_napi);
1110 		__netif_napi_del(&rq->xdp_napi);
1111 	}
1112 	synchronize_net();
1113 
1114 	for (i = start; i < end; i++) {
1115 		struct veth_rq *rq = &priv->rq[i];
1116 
1117 		rq->rx_notify_masked = false;
1118 		ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free);
1119 	}
1120 
1121 	for (i = start; i < end; i++) {
1122 		page_pool_destroy(priv->rq[i].page_pool);
1123 		priv->rq[i].page_pool = NULL;
1124 	}
1125 }
1126 
1127 static void veth_napi_del(struct net_device *dev)
1128 {
1129 	veth_napi_del_range(dev, 0, dev->real_num_rx_queues);
1130 }
1131 
1132 static bool veth_gro_requested(const struct net_device *dev)
1133 {
1134 	return !!(dev->wanted_features & NETIF_F_GRO);
1135 }
1136 
1137 static int veth_enable_xdp_range(struct net_device *dev, int start, int end,
1138 				 bool napi_already_on)
1139 {
1140 	struct veth_priv *priv = netdev_priv(dev);
1141 	int err, i;
1142 
1143 	for (i = start; i < end; i++) {
1144 		struct veth_rq *rq = &priv->rq[i];
1145 
1146 		if (!napi_already_on)
1147 			netif_napi_add(dev, &rq->xdp_napi, veth_poll);
1148 		err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i, rq->xdp_napi.napi_id);
1149 		if (err < 0)
1150 			goto err_rxq_reg;
1151 
1152 		err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
1153 						 MEM_TYPE_PAGE_SHARED,
1154 						 NULL);
1155 		if (err < 0)
1156 			goto err_reg_mem;
1157 
1158 		/* Save original mem info as it can be overwritten */
1159 		rq->xdp_mem = rq->xdp_rxq.mem;
1160 	}
1161 	return 0;
1162 
1163 err_reg_mem:
1164 	xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
1165 err_rxq_reg:
1166 	for (i--; i >= start; i--) {
1167 		struct veth_rq *rq = &priv->rq[i];
1168 
1169 		xdp_rxq_info_unreg(&rq->xdp_rxq);
1170 		if (!napi_already_on)
1171 			netif_napi_del(&rq->xdp_napi);
1172 	}
1173 
1174 	return err;
1175 }
1176 
1177 static void veth_disable_xdp_range(struct net_device *dev, int start, int end,
1178 				   bool delete_napi)
1179 {
1180 	struct veth_priv *priv = netdev_priv(dev);
1181 	int i;
1182 
1183 	for (i = start; i < end; i++) {
1184 		struct veth_rq *rq = &priv->rq[i];
1185 
1186 		rq->xdp_rxq.mem = rq->xdp_mem;
1187 		xdp_rxq_info_unreg(&rq->xdp_rxq);
1188 
1189 		if (delete_napi)
1190 			netif_napi_del(&rq->xdp_napi);
1191 	}
1192 }
1193 
1194 static int veth_enable_xdp(struct net_device *dev)
1195 {
1196 	bool napi_already_on = veth_gro_requested(dev) && (dev->flags & IFF_UP);
1197 	struct veth_priv *priv = netdev_priv(dev);
1198 	int err, i;
1199 
1200 	if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) {
1201 		err = veth_enable_xdp_range(dev, 0, dev->real_num_rx_queues, napi_already_on);
1202 		if (err)
1203 			return err;
1204 
1205 		if (!napi_already_on) {
1206 			err = __veth_napi_enable(dev);
1207 			if (err) {
1208 				veth_disable_xdp_range(dev, 0, dev->real_num_rx_queues, true);
1209 				return err;
1210 			}
1211 
1212 			if (!veth_gro_requested(dev)) {
1213 				/* user-space did not require GRO, but adding XDP
1214 				 * is supposed to get GRO working
1215 				 */
1216 				dev->features |= NETIF_F_GRO;
1217 				netdev_features_change(dev);
1218 			}
1219 		}
1220 	}
1221 
1222 	for (i = 0; i < dev->real_num_rx_queues; i++) {
1223 		rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog);
1224 		rcu_assign_pointer(priv->rq[i].napi, &priv->rq[i].xdp_napi);
1225 	}
1226 
1227 	return 0;
1228 }
1229 
1230 static void veth_disable_xdp(struct net_device *dev)
1231 {
1232 	struct veth_priv *priv = netdev_priv(dev);
1233 	int i;
1234 
1235 	for (i = 0; i < dev->real_num_rx_queues; i++)
1236 		rcu_assign_pointer(priv->rq[i].xdp_prog, NULL);
1237 
1238 	if (!netif_running(dev) || !veth_gro_requested(dev)) {
1239 		veth_napi_del(dev);
1240 
1241 		/* if user-space did not require GRO, since adding XDP
1242 		 * enabled it, clear it now
1243 		 */
1244 		if (!veth_gro_requested(dev) && netif_running(dev)) {
1245 			dev->features &= ~NETIF_F_GRO;
1246 			netdev_features_change(dev);
1247 		}
1248 	}
1249 
1250 	veth_disable_xdp_range(dev, 0, dev->real_num_rx_queues, false);
1251 }
1252 
1253 static int veth_napi_enable_range(struct net_device *dev, int start, int end)
1254 {
1255 	struct veth_priv *priv = netdev_priv(dev);
1256 	int err, i;
1257 
1258 	for (i = start; i < end; i++) {
1259 		struct veth_rq *rq = &priv->rq[i];
1260 
1261 		netif_napi_add(dev, &rq->xdp_napi, veth_poll);
1262 	}
1263 
1264 	err = __veth_napi_enable_range(dev, start, end);
1265 	if (err) {
1266 		for (i = start; i < end; i++) {
1267 			struct veth_rq *rq = &priv->rq[i];
1268 
1269 			netif_napi_del(&rq->xdp_napi);
1270 		}
1271 		return err;
1272 	}
1273 	return err;
1274 }
1275 
1276 static int veth_napi_enable(struct net_device *dev)
1277 {
1278 	return veth_napi_enable_range(dev, 0, dev->real_num_rx_queues);
1279 }
1280 
1281 static void veth_disable_range_safe(struct net_device *dev, int start, int end)
1282 {
1283 	struct veth_priv *priv = netdev_priv(dev);
1284 
1285 	if (start >= end)
1286 		return;
1287 
1288 	if (priv->_xdp_prog) {
1289 		veth_napi_del_range(dev, start, end);
1290 		veth_disable_xdp_range(dev, start, end, false);
1291 	} else if (veth_gro_requested(dev)) {
1292 		veth_napi_del_range(dev, start, end);
1293 	}
1294 }
1295 
1296 static int veth_enable_range_safe(struct net_device *dev, int start, int end)
1297 {
1298 	struct veth_priv *priv = netdev_priv(dev);
1299 	int err;
1300 
1301 	if (start >= end)
1302 		return 0;
1303 
1304 	if (priv->_xdp_prog) {
1305 		/* these channels are freshly initialized, napi is not on there even
1306 		 * when GRO is requeste
1307 		 */
1308 		err = veth_enable_xdp_range(dev, start, end, false);
1309 		if (err)
1310 			return err;
1311 
1312 		err = __veth_napi_enable_range(dev, start, end);
1313 		if (err) {
1314 			/* on error always delete the newly added napis */
1315 			veth_disable_xdp_range(dev, start, end, true);
1316 			return err;
1317 		}
1318 	} else if (veth_gro_requested(dev)) {
1319 		return veth_napi_enable_range(dev, start, end);
1320 	}
1321 	return 0;
1322 }
1323 
1324 static void veth_set_xdp_features(struct net_device *dev)
1325 {
1326 	struct veth_priv *priv = netdev_priv(dev);
1327 	struct net_device *peer;
1328 
1329 	peer = rtnl_dereference(priv->peer);
1330 	if (peer && peer->real_num_tx_queues <= dev->real_num_rx_queues) {
1331 		struct veth_priv *priv_peer = netdev_priv(peer);
1332 		xdp_features_t val = NETDEV_XDP_ACT_BASIC |
1333 				     NETDEV_XDP_ACT_REDIRECT |
1334 				     NETDEV_XDP_ACT_RX_SG;
1335 
1336 		if (priv_peer->_xdp_prog || veth_gro_requested(peer))
1337 			val |= NETDEV_XDP_ACT_NDO_XMIT |
1338 			       NETDEV_XDP_ACT_NDO_XMIT_SG;
1339 		xdp_set_features_flag(dev, val);
1340 	} else {
1341 		xdp_clear_features_flag(dev);
1342 	}
1343 }
1344 
1345 static int veth_set_channels(struct net_device *dev,
1346 			     struct ethtool_channels *ch)
1347 {
1348 	struct veth_priv *priv = netdev_priv(dev);
1349 	unsigned int old_rx_count, new_rx_count;
1350 	struct veth_priv *peer_priv;
1351 	struct net_device *peer;
1352 	int err;
1353 
1354 	/* sanity check. Upper bounds are already enforced by the caller */
1355 	if (!ch->rx_count || !ch->tx_count)
1356 		return -EINVAL;
1357 
1358 	/* avoid braking XDP, if that is enabled */
1359 	peer = rtnl_dereference(priv->peer);
1360 	peer_priv = peer ? netdev_priv(peer) : NULL;
1361 	if (priv->_xdp_prog && peer && ch->rx_count < peer->real_num_tx_queues)
1362 		return -EINVAL;
1363 
1364 	if (peer && peer_priv && peer_priv->_xdp_prog && ch->tx_count > peer->real_num_rx_queues)
1365 		return -EINVAL;
1366 
1367 	old_rx_count = dev->real_num_rx_queues;
1368 	new_rx_count = ch->rx_count;
1369 	if (netif_running(dev)) {
1370 		/* turn device off */
1371 		netif_carrier_off(dev);
1372 		if (peer)
1373 			netif_carrier_off(peer);
1374 
1375 		/* try to allocate new resurces, as needed*/
1376 		err = veth_enable_range_safe(dev, old_rx_count, new_rx_count);
1377 		if (err)
1378 			goto out;
1379 	}
1380 
1381 	err = netif_set_real_num_rx_queues(dev, ch->rx_count);
1382 	if (err)
1383 		goto revert;
1384 
1385 	err = netif_set_real_num_tx_queues(dev, ch->tx_count);
1386 	if (err) {
1387 		int err2 = netif_set_real_num_rx_queues(dev, old_rx_count);
1388 
1389 		/* this error condition could happen only if rx and tx change
1390 		 * in opposite directions (e.g. tx nr raises, rx nr decreases)
1391 		 * and we can't do anything to fully restore the original
1392 		 * status
1393 		 */
1394 		if (err2)
1395 			pr_warn("Can't restore rx queues config %d -> %d %d",
1396 				new_rx_count, old_rx_count, err2);
1397 		else
1398 			goto revert;
1399 	}
1400 
1401 out:
1402 	if (netif_running(dev)) {
1403 		/* note that we need to swap the arguments WRT the enable part
1404 		 * to identify the range we have to disable
1405 		 */
1406 		veth_disable_range_safe(dev, new_rx_count, old_rx_count);
1407 		netif_carrier_on(dev);
1408 		if (peer)
1409 			netif_carrier_on(peer);
1410 	}
1411 
1412 	/* update XDP supported features */
1413 	veth_set_xdp_features(dev);
1414 	if (peer)
1415 		veth_set_xdp_features(peer);
1416 
1417 	return err;
1418 
1419 revert:
1420 	new_rx_count = old_rx_count;
1421 	old_rx_count = ch->rx_count;
1422 	goto out;
1423 }
1424 
1425 static int veth_open(struct net_device *dev)
1426 {
1427 	struct veth_priv *priv = netdev_priv(dev);
1428 	struct net_device *peer = rtnl_dereference(priv->peer);
1429 	int err;
1430 
1431 	if (!peer)
1432 		return -ENOTCONN;
1433 
1434 	if (priv->_xdp_prog) {
1435 		err = veth_enable_xdp(dev);
1436 		if (err)
1437 			return err;
1438 	} else if (veth_gro_requested(dev)) {
1439 		err = veth_napi_enable(dev);
1440 		if (err)
1441 			return err;
1442 	}
1443 
1444 	if (peer->flags & IFF_UP) {
1445 		netif_carrier_on(dev);
1446 		netif_carrier_on(peer);
1447 	}
1448 
1449 	veth_set_xdp_features(dev);
1450 
1451 	return 0;
1452 }
1453 
1454 static int veth_close(struct net_device *dev)
1455 {
1456 	struct veth_priv *priv = netdev_priv(dev);
1457 	struct net_device *peer = rtnl_dereference(priv->peer);
1458 
1459 	netif_carrier_off(dev);
1460 	if (peer)
1461 		netif_carrier_off(peer);
1462 
1463 	if (priv->_xdp_prog)
1464 		veth_disable_xdp(dev);
1465 	else if (veth_gro_requested(dev))
1466 		veth_napi_del(dev);
1467 
1468 	return 0;
1469 }
1470 
1471 static int is_valid_veth_mtu(int mtu)
1472 {
1473 	return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
1474 }
1475 
1476 static int veth_alloc_queues(struct net_device *dev)
1477 {
1478 	struct veth_priv *priv = netdev_priv(dev);
1479 	int i;
1480 
1481 	priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL_ACCOUNT);
1482 	if (!priv->rq)
1483 		return -ENOMEM;
1484 
1485 	for (i = 0; i < dev->num_rx_queues; i++) {
1486 		priv->rq[i].dev = dev;
1487 		u64_stats_init(&priv->rq[i].stats.syncp);
1488 	}
1489 
1490 	return 0;
1491 }
1492 
1493 static void veth_free_queues(struct net_device *dev)
1494 {
1495 	struct veth_priv *priv = netdev_priv(dev);
1496 
1497 	kfree(priv->rq);
1498 }
1499 
1500 static int veth_dev_init(struct net_device *dev)
1501 {
1502 	int err;
1503 
1504 	dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
1505 	if (!dev->lstats)
1506 		return -ENOMEM;
1507 
1508 	err = veth_alloc_queues(dev);
1509 	if (err) {
1510 		free_percpu(dev->lstats);
1511 		return err;
1512 	}
1513 
1514 	return 0;
1515 }
1516 
1517 static void veth_dev_free(struct net_device *dev)
1518 {
1519 	veth_free_queues(dev);
1520 	free_percpu(dev->lstats);
1521 }
1522 
1523 #ifdef CONFIG_NET_POLL_CONTROLLER
1524 static void veth_poll_controller(struct net_device *dev)
1525 {
1526 	/* veth only receives frames when its peer sends one
1527 	 * Since it has nothing to do with disabling irqs, we are guaranteed
1528 	 * never to have pending data when we poll for it so
1529 	 * there is nothing to do here.
1530 	 *
1531 	 * We need this though so netpoll recognizes us as an interface that
1532 	 * supports polling, which enables bridge devices in virt setups to
1533 	 * still use netconsole
1534 	 */
1535 }
1536 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1537 
1538 static int veth_get_iflink(const struct net_device *dev)
1539 {
1540 	struct veth_priv *priv = netdev_priv(dev);
1541 	struct net_device *peer;
1542 	int iflink;
1543 
1544 	rcu_read_lock();
1545 	peer = rcu_dereference(priv->peer);
1546 	iflink = peer ? peer->ifindex : 0;
1547 	rcu_read_unlock();
1548 
1549 	return iflink;
1550 }
1551 
1552 static netdev_features_t veth_fix_features(struct net_device *dev,
1553 					   netdev_features_t features)
1554 {
1555 	struct veth_priv *priv = netdev_priv(dev);
1556 	struct net_device *peer;
1557 
1558 	peer = rtnl_dereference(priv->peer);
1559 	if (peer) {
1560 		struct veth_priv *peer_priv = netdev_priv(peer);
1561 
1562 		if (peer_priv->_xdp_prog)
1563 			features &= ~NETIF_F_GSO_SOFTWARE;
1564 	}
1565 	if (priv->_xdp_prog)
1566 		features |= NETIF_F_GRO;
1567 
1568 	return features;
1569 }
1570 
1571 static int veth_set_features(struct net_device *dev,
1572 			     netdev_features_t features)
1573 {
1574 	netdev_features_t changed = features ^ dev->features;
1575 	struct veth_priv *priv = netdev_priv(dev);
1576 	struct net_device *peer;
1577 	int err;
1578 
1579 	if (!(changed & NETIF_F_GRO) || !(dev->flags & IFF_UP) || priv->_xdp_prog)
1580 		return 0;
1581 
1582 	peer = rtnl_dereference(priv->peer);
1583 	if (features & NETIF_F_GRO) {
1584 		err = veth_napi_enable(dev);
1585 		if (err)
1586 			return err;
1587 
1588 		if (peer)
1589 			xdp_features_set_redirect_target(peer, true);
1590 	} else {
1591 		if (peer)
1592 			xdp_features_clear_redirect_target(peer);
1593 		veth_napi_del(dev);
1594 	}
1595 	return 0;
1596 }
1597 
1598 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
1599 {
1600 	struct veth_priv *peer_priv, *priv = netdev_priv(dev);
1601 	struct net_device *peer;
1602 
1603 	if (new_hr < 0)
1604 		new_hr = 0;
1605 
1606 	rcu_read_lock();
1607 	peer = rcu_dereference(priv->peer);
1608 	if (unlikely(!peer))
1609 		goto out;
1610 
1611 	peer_priv = netdev_priv(peer);
1612 	priv->requested_headroom = new_hr;
1613 	new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
1614 	dev->needed_headroom = new_hr;
1615 	peer->needed_headroom = new_hr;
1616 
1617 out:
1618 	rcu_read_unlock();
1619 }
1620 
1621 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1622 			struct netlink_ext_ack *extack)
1623 {
1624 	struct veth_priv *priv = netdev_priv(dev);
1625 	struct bpf_prog *old_prog;
1626 	struct net_device *peer;
1627 	unsigned int max_mtu;
1628 	int err;
1629 
1630 	old_prog = priv->_xdp_prog;
1631 	priv->_xdp_prog = prog;
1632 	peer = rtnl_dereference(priv->peer);
1633 
1634 	if (prog) {
1635 		if (!peer) {
1636 			NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached");
1637 			err = -ENOTCONN;
1638 			goto err;
1639 		}
1640 
1641 		max_mtu = SKB_WITH_OVERHEAD(PAGE_SIZE - VETH_XDP_HEADROOM) -
1642 			  peer->hard_header_len;
1643 		/* Allow increasing the max_mtu if the program supports
1644 		 * XDP fragments.
1645 		 */
1646 		if (prog->aux->xdp_has_frags)
1647 			max_mtu += PAGE_SIZE * MAX_SKB_FRAGS;
1648 
1649 		if (peer->mtu > max_mtu) {
1650 			NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
1651 			err = -ERANGE;
1652 			goto err;
1653 		}
1654 
1655 		if (dev->real_num_rx_queues < peer->real_num_tx_queues) {
1656 			NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues");
1657 			err = -ENOSPC;
1658 			goto err;
1659 		}
1660 
1661 		if (dev->flags & IFF_UP) {
1662 			err = veth_enable_xdp(dev);
1663 			if (err) {
1664 				NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed");
1665 				goto err;
1666 			}
1667 		}
1668 
1669 		if (!old_prog) {
1670 			peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
1671 			peer->max_mtu = max_mtu;
1672 		}
1673 
1674 		xdp_features_set_redirect_target(peer, true);
1675 	}
1676 
1677 	if (old_prog) {
1678 		if (!prog) {
1679 			if (peer && !veth_gro_requested(dev))
1680 				xdp_features_clear_redirect_target(peer);
1681 
1682 			if (dev->flags & IFF_UP)
1683 				veth_disable_xdp(dev);
1684 
1685 			if (peer) {
1686 				peer->hw_features |= NETIF_F_GSO_SOFTWARE;
1687 				peer->max_mtu = ETH_MAX_MTU;
1688 			}
1689 		}
1690 		bpf_prog_put(old_prog);
1691 	}
1692 
1693 	if ((!!old_prog ^ !!prog) && peer)
1694 		netdev_update_features(peer);
1695 
1696 	return 0;
1697 err:
1698 	priv->_xdp_prog = old_prog;
1699 
1700 	return err;
1701 }
1702 
1703 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1704 {
1705 	switch (xdp->command) {
1706 	case XDP_SETUP_PROG:
1707 		return veth_xdp_set(dev, xdp->prog, xdp->extack);
1708 	default:
1709 		return -EINVAL;
1710 	}
1711 }
1712 
1713 static int veth_xdp_rx_timestamp(const struct xdp_md *ctx, u64 *timestamp)
1714 {
1715 	struct veth_xdp_buff *_ctx = (void *)ctx;
1716 
1717 	if (!_ctx->skb)
1718 		return -ENODATA;
1719 
1720 	*timestamp = skb_hwtstamps(_ctx->skb)->hwtstamp;
1721 	return 0;
1722 }
1723 
1724 static int veth_xdp_rx_hash(const struct xdp_md *ctx, u32 *hash,
1725 			    enum xdp_rss_hash_type *rss_type)
1726 {
1727 	struct veth_xdp_buff *_ctx = (void *)ctx;
1728 	struct sk_buff *skb = _ctx->skb;
1729 
1730 	if (!skb)
1731 		return -ENODATA;
1732 
1733 	*hash = skb_get_hash(skb);
1734 	*rss_type = skb->l4_hash ? XDP_RSS_TYPE_L4_ANY : XDP_RSS_TYPE_NONE;
1735 
1736 	return 0;
1737 }
1738 
1739 static const struct net_device_ops veth_netdev_ops = {
1740 	.ndo_init            = veth_dev_init,
1741 	.ndo_open            = veth_open,
1742 	.ndo_stop            = veth_close,
1743 	.ndo_start_xmit      = veth_xmit,
1744 	.ndo_get_stats64     = veth_get_stats64,
1745 	.ndo_set_rx_mode     = veth_set_multicast_list,
1746 	.ndo_set_mac_address = eth_mac_addr,
1747 #ifdef CONFIG_NET_POLL_CONTROLLER
1748 	.ndo_poll_controller	= veth_poll_controller,
1749 #endif
1750 	.ndo_get_iflink		= veth_get_iflink,
1751 	.ndo_fix_features	= veth_fix_features,
1752 	.ndo_set_features	= veth_set_features,
1753 	.ndo_features_check	= passthru_features_check,
1754 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
1755 	.ndo_bpf		= veth_xdp,
1756 	.ndo_xdp_xmit		= veth_ndo_xdp_xmit,
1757 	.ndo_get_peer_dev	= veth_peer_dev,
1758 };
1759 
1760 static const struct xdp_metadata_ops veth_xdp_metadata_ops = {
1761 	.xmo_rx_timestamp		= veth_xdp_rx_timestamp,
1762 	.xmo_rx_hash			= veth_xdp_rx_hash,
1763 };
1764 
1765 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
1766 		       NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
1767 		       NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
1768 		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
1769 		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
1770 
1771 static void veth_setup(struct net_device *dev)
1772 {
1773 	ether_setup(dev);
1774 
1775 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1776 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1777 	dev->priv_flags |= IFF_NO_QUEUE;
1778 	dev->priv_flags |= IFF_PHONY_HEADROOM;
1779 
1780 	dev->netdev_ops = &veth_netdev_ops;
1781 	dev->xdp_metadata_ops = &veth_xdp_metadata_ops;
1782 	dev->ethtool_ops = &veth_ethtool_ops;
1783 	dev->features |= NETIF_F_LLTX;
1784 	dev->features |= VETH_FEATURES;
1785 	dev->vlan_features = dev->features &
1786 			     ~(NETIF_F_HW_VLAN_CTAG_TX |
1787 			       NETIF_F_HW_VLAN_STAG_TX |
1788 			       NETIF_F_HW_VLAN_CTAG_RX |
1789 			       NETIF_F_HW_VLAN_STAG_RX);
1790 	dev->needs_free_netdev = true;
1791 	dev->priv_destructor = veth_dev_free;
1792 	dev->max_mtu = ETH_MAX_MTU;
1793 
1794 	dev->hw_features = VETH_FEATURES;
1795 	dev->hw_enc_features = VETH_FEATURES;
1796 	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
1797 	netif_set_tso_max_size(dev, GSO_MAX_SIZE);
1798 }
1799 
1800 /*
1801  * netlink interface
1802  */
1803 
1804 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
1805 			 struct netlink_ext_ack *extack)
1806 {
1807 	if (tb[IFLA_ADDRESS]) {
1808 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1809 			return -EINVAL;
1810 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1811 			return -EADDRNOTAVAIL;
1812 	}
1813 	if (tb[IFLA_MTU]) {
1814 		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
1815 			return -EINVAL;
1816 	}
1817 	return 0;
1818 }
1819 
1820 static struct rtnl_link_ops veth_link_ops;
1821 
1822 static void veth_disable_gro(struct net_device *dev)
1823 {
1824 	dev->features &= ~NETIF_F_GRO;
1825 	dev->wanted_features &= ~NETIF_F_GRO;
1826 	netdev_update_features(dev);
1827 }
1828 
1829 static int veth_init_queues(struct net_device *dev, struct nlattr *tb[])
1830 {
1831 	int err;
1832 
1833 	if (!tb[IFLA_NUM_TX_QUEUES] && dev->num_tx_queues > 1) {
1834 		err = netif_set_real_num_tx_queues(dev, 1);
1835 		if (err)
1836 			return err;
1837 	}
1838 	if (!tb[IFLA_NUM_RX_QUEUES] && dev->num_rx_queues > 1) {
1839 		err = netif_set_real_num_rx_queues(dev, 1);
1840 		if (err)
1841 			return err;
1842 	}
1843 	return 0;
1844 }
1845 
1846 static int veth_newlink(struct net *src_net, struct net_device *dev,
1847 			struct nlattr *tb[], struct nlattr *data[],
1848 			struct netlink_ext_ack *extack)
1849 {
1850 	int err;
1851 	struct net_device *peer;
1852 	struct veth_priv *priv;
1853 	char ifname[IFNAMSIZ];
1854 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
1855 	unsigned char name_assign_type;
1856 	struct ifinfomsg *ifmp;
1857 	struct net *net;
1858 
1859 	/*
1860 	 * create and register peer first
1861 	 */
1862 	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
1863 		struct nlattr *nla_peer;
1864 
1865 		nla_peer = data[VETH_INFO_PEER];
1866 		ifmp = nla_data(nla_peer);
1867 		err = rtnl_nla_parse_ifinfomsg(peer_tb, nla_peer, extack);
1868 		if (err < 0)
1869 			return err;
1870 
1871 		err = veth_validate(peer_tb, NULL, extack);
1872 		if (err < 0)
1873 			return err;
1874 
1875 		tbp = peer_tb;
1876 	} else {
1877 		ifmp = NULL;
1878 		tbp = tb;
1879 	}
1880 
1881 	if (ifmp && tbp[IFLA_IFNAME]) {
1882 		nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
1883 		name_assign_type = NET_NAME_USER;
1884 	} else {
1885 		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
1886 		name_assign_type = NET_NAME_ENUM;
1887 	}
1888 
1889 	net = rtnl_link_get_net(src_net, tbp);
1890 	if (IS_ERR(net))
1891 		return PTR_ERR(net);
1892 
1893 	peer = rtnl_create_link(net, ifname, name_assign_type,
1894 				&veth_link_ops, tbp, extack);
1895 	if (IS_ERR(peer)) {
1896 		put_net(net);
1897 		return PTR_ERR(peer);
1898 	}
1899 
1900 	if (!ifmp || !tbp[IFLA_ADDRESS])
1901 		eth_hw_addr_random(peer);
1902 
1903 	if (ifmp && (dev->ifindex != 0))
1904 		peer->ifindex = ifmp->ifi_index;
1905 
1906 	netif_inherit_tso_max(peer, dev);
1907 
1908 	err = register_netdevice(peer);
1909 	put_net(net);
1910 	net = NULL;
1911 	if (err < 0)
1912 		goto err_register_peer;
1913 
1914 	/* keep GRO disabled by default to be consistent with the established
1915 	 * veth behavior
1916 	 */
1917 	veth_disable_gro(peer);
1918 	netif_carrier_off(peer);
1919 
1920 	err = rtnl_configure_link(peer, ifmp, 0, NULL);
1921 	if (err < 0)
1922 		goto err_configure_peer;
1923 
1924 	/*
1925 	 * register dev last
1926 	 *
1927 	 * note, that since we've registered new device the dev's name
1928 	 * should be re-allocated
1929 	 */
1930 
1931 	if (tb[IFLA_ADDRESS] == NULL)
1932 		eth_hw_addr_random(dev);
1933 
1934 	if (tb[IFLA_IFNAME])
1935 		nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
1936 	else
1937 		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
1938 
1939 	err = register_netdevice(dev);
1940 	if (err < 0)
1941 		goto err_register_dev;
1942 
1943 	netif_carrier_off(dev);
1944 
1945 	/*
1946 	 * tie the deviced together
1947 	 */
1948 
1949 	priv = netdev_priv(dev);
1950 	rcu_assign_pointer(priv->peer, peer);
1951 	err = veth_init_queues(dev, tb);
1952 	if (err)
1953 		goto err_queues;
1954 
1955 	priv = netdev_priv(peer);
1956 	rcu_assign_pointer(priv->peer, dev);
1957 	err = veth_init_queues(peer, tb);
1958 	if (err)
1959 		goto err_queues;
1960 
1961 	veth_disable_gro(dev);
1962 	/* update XDP supported features */
1963 	veth_set_xdp_features(dev);
1964 	veth_set_xdp_features(peer);
1965 
1966 	return 0;
1967 
1968 err_queues:
1969 	unregister_netdevice(dev);
1970 err_register_dev:
1971 	/* nothing to do */
1972 err_configure_peer:
1973 	unregister_netdevice(peer);
1974 	return err;
1975 
1976 err_register_peer:
1977 	free_netdev(peer);
1978 	return err;
1979 }
1980 
1981 static void veth_dellink(struct net_device *dev, struct list_head *head)
1982 {
1983 	struct veth_priv *priv;
1984 	struct net_device *peer;
1985 
1986 	priv = netdev_priv(dev);
1987 	peer = rtnl_dereference(priv->peer);
1988 
1989 	/* Note : dellink() is called from default_device_exit_batch(),
1990 	 * before a rcu_synchronize() point. The devices are guaranteed
1991 	 * not being freed before one RCU grace period.
1992 	 */
1993 	RCU_INIT_POINTER(priv->peer, NULL);
1994 	unregister_netdevice_queue(dev, head);
1995 
1996 	if (peer) {
1997 		priv = netdev_priv(peer);
1998 		RCU_INIT_POINTER(priv->peer, NULL);
1999 		unregister_netdevice_queue(peer, head);
2000 	}
2001 }
2002 
2003 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
2004 	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
2005 };
2006 
2007 static struct net *veth_get_link_net(const struct net_device *dev)
2008 {
2009 	struct veth_priv *priv = netdev_priv(dev);
2010 	struct net_device *peer = rtnl_dereference(priv->peer);
2011 
2012 	return peer ? dev_net(peer) : dev_net(dev);
2013 }
2014 
2015 static unsigned int veth_get_num_queues(void)
2016 {
2017 	/* enforce the same queue limit as rtnl_create_link */
2018 	int queues = num_possible_cpus();
2019 
2020 	if (queues > 4096)
2021 		queues = 4096;
2022 	return queues;
2023 }
2024 
2025 static struct rtnl_link_ops veth_link_ops = {
2026 	.kind		= DRV_NAME,
2027 	.priv_size	= sizeof(struct veth_priv),
2028 	.setup		= veth_setup,
2029 	.validate	= veth_validate,
2030 	.newlink	= veth_newlink,
2031 	.dellink	= veth_dellink,
2032 	.policy		= veth_policy,
2033 	.maxtype	= VETH_INFO_MAX,
2034 	.get_link_net	= veth_get_link_net,
2035 	.get_num_tx_queues	= veth_get_num_queues,
2036 	.get_num_rx_queues	= veth_get_num_queues,
2037 };
2038 
2039 /*
2040  * init/fini
2041  */
2042 
2043 static __init int veth_init(void)
2044 {
2045 	return rtnl_link_register(&veth_link_ops);
2046 }
2047 
2048 static __exit void veth_exit(void)
2049 {
2050 	rtnl_link_unregister(&veth_link_ops);
2051 }
2052 
2053 module_init(veth_init);
2054 module_exit(veth_exit);
2055 
2056 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
2057 MODULE_LICENSE("GPL v2");
2058 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
2059