Linux Audio

Check our new training course

Loading...
v6.8
  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
  3 */
  4
 
 
 
  5#include "ipvlan.h"
  6
  7static u32 ipvlan_jhash_secret __read_mostly;
  8
  9void ipvlan_init_secret(void)
 10{
 11	net_get_random_once(&ipvlan_jhash_secret, sizeof(ipvlan_jhash_secret));
 12}
 13
 14void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
 15			    unsigned int len, bool success, bool mcast)
 16{
 17	if (likely(success)) {
 18		struct ipvl_pcpu_stats *pcptr;
 19
 20		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 21		u64_stats_update_begin(&pcptr->syncp);
 22		u64_stats_inc(&pcptr->rx_pkts);
 23		u64_stats_add(&pcptr->rx_bytes, len);
 24		if (mcast)
 25			u64_stats_inc(&pcptr->rx_mcast);
 26		u64_stats_update_end(&pcptr->syncp);
 27	} else {
 28		this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
 29	}
 30}
 31EXPORT_SYMBOL_GPL(ipvlan_count_rx);
 32
 33#if IS_ENABLED(CONFIG_IPV6)
 34static u8 ipvlan_get_v6_hash(const void *iaddr)
 35{
 36	const struct in6_addr *ip6_addr = iaddr;
 37
 38	return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
 39	       IPVLAN_HASH_MASK;
 40}
 41#else
 42static u8 ipvlan_get_v6_hash(const void *iaddr)
 43{
 44	return 0;
 45}
 46#endif
 47
 48static u8 ipvlan_get_v4_hash(const void *iaddr)
 49{
 50	const struct in_addr *ip4_addr = iaddr;
 51
 52	return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
 53	       IPVLAN_HASH_MASK;
 54}
 55
 56static bool addr_equal(bool is_v6, struct ipvl_addr *addr, const void *iaddr)
 57{
 58	if (!is_v6 && addr->atype == IPVL_IPV4) {
 59		struct in_addr *i4addr = (struct in_addr *)iaddr;
 60
 61		return addr->ip4addr.s_addr == i4addr->s_addr;
 62#if IS_ENABLED(CONFIG_IPV6)
 63	} else if (is_v6 && addr->atype == IPVL_IPV6) {
 64		struct in6_addr *i6addr = (struct in6_addr *)iaddr;
 65
 66		return ipv6_addr_equal(&addr->ip6addr, i6addr);
 67#endif
 68	}
 69
 70	return false;
 71}
 72
 73static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
 74					       const void *iaddr, bool is_v6)
 75{
 76	struct ipvl_addr *addr;
 77	u8 hash;
 78
 79	hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
 80	       ipvlan_get_v4_hash(iaddr);
 81	hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode)
 82		if (addr_equal(is_v6, addr, iaddr))
 83			return addr;
 84	return NULL;
 85}
 86
 87void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
 88{
 89	struct ipvl_port *port = ipvlan->port;
 90	u8 hash;
 91
 92	hash = (addr->atype == IPVL_IPV6) ?
 93	       ipvlan_get_v6_hash(&addr->ip6addr) :
 94	       ipvlan_get_v4_hash(&addr->ip4addr);
 95	if (hlist_unhashed(&addr->hlnode))
 96		hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
 97}
 98
 99void ipvlan_ht_addr_del(struct ipvl_addr *addr)
100{
101	hlist_del_init_rcu(&addr->hlnode);
102}
103
104struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
105				   const void *iaddr, bool is_v6)
106{
107	struct ipvl_addr *addr, *ret = NULL;
108
109	rcu_read_lock();
110	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode) {
111		if (addr_equal(is_v6, addr, iaddr)) {
112			ret = addr;
113			break;
114		}
115	}
116	rcu_read_unlock();
117	return ret;
118}
119
120bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
121{
122	struct ipvl_dev *ipvlan;
123	bool ret = false;
124
125	rcu_read_lock();
126	list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
127		if (ipvlan_find_addr(ipvlan, iaddr, is_v6)) {
128			ret = true;
129			break;
130		}
131	}
132	rcu_read_unlock();
133	return ret;
134}
135
136void *ipvlan_get_L3_hdr(struct ipvl_port *port, struct sk_buff *skb, int *type)
137{
138	void *lyr3h = NULL;
139
140	switch (skb->protocol) {
141	case htons(ETH_P_ARP): {
142		struct arphdr *arph;
143
144		if (unlikely(!pskb_may_pull(skb, arp_hdr_len(port->dev))))
145			return NULL;
146
147		arph = arp_hdr(skb);
148		*type = IPVL_ARP;
149		lyr3h = arph;
150		break;
151	}
152	case htons(ETH_P_IP): {
153		u32 pktlen;
154		struct iphdr *ip4h;
155
156		if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
157			return NULL;
158
159		ip4h = ip_hdr(skb);
160		pktlen = skb_ip_totlen(skb);
161		if (ip4h->ihl < 5 || ip4h->version != 4)
162			return NULL;
163		if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
164			return NULL;
165
166		*type = IPVL_IPV4;
167		lyr3h = ip4h;
168		break;
169	}
170#if IS_ENABLED(CONFIG_IPV6)
171	case htons(ETH_P_IPV6): {
172		struct ipv6hdr *ip6h;
173
174		if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
175			return NULL;
176
177		ip6h = ipv6_hdr(skb);
178		if (ip6h->version != 6)
179			return NULL;
180
181		*type = IPVL_IPV6;
182		lyr3h = ip6h;
183		/* Only Neighbour Solicitation pkts need different treatment */
184		if (ipv6_addr_any(&ip6h->saddr) &&
185		    ip6h->nexthdr == NEXTHDR_ICMP) {
186			struct icmp6hdr	*icmph;
187
188			if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph))))
189				return NULL;
190
191			ip6h = ipv6_hdr(skb);
192			icmph = (struct icmp6hdr *)(ip6h + 1);
193
194			if (icmph->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
195				/* Need to access the ipv6 address in body */
196				if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph)
197						+ sizeof(struct in6_addr))))
198					return NULL;
199
200				ip6h = ipv6_hdr(skb);
201				icmph = (struct icmp6hdr *)(ip6h + 1);
202			}
203
204			*type = IPVL_ICMPV6;
205			lyr3h = icmph;
206		}
207		break;
208	}
209#endif
210	default:
211		return NULL;
212	}
213
214	return lyr3h;
215}
216
217unsigned int ipvlan_mac_hash(const unsigned char *addr)
218{
219	u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
220			       ipvlan_jhash_secret);
221
222	return hash & IPVLAN_MAC_FILTER_MASK;
223}
224
225void ipvlan_process_multicast(struct work_struct *work)
226{
227	struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
228	struct ethhdr *ethh;
229	struct ipvl_dev *ipvlan;
230	struct sk_buff *skb, *nskb;
231	struct sk_buff_head list;
232	unsigned int len;
233	unsigned int mac_hash;
234	int ret;
235	u8 pkt_type;
236	bool tx_pkt;
237
238	__skb_queue_head_init(&list);
239
240	spin_lock_bh(&port->backlog.lock);
241	skb_queue_splice_tail_init(&port->backlog, &list);
242	spin_unlock_bh(&port->backlog.lock);
243
244	while ((skb = __skb_dequeue(&list)) != NULL) {
245		struct net_device *dev = skb->dev;
246		bool consumed = false;
247
248		ethh = eth_hdr(skb);
249		tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
250		mac_hash = ipvlan_mac_hash(ethh->h_dest);
251
252		if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
253			pkt_type = PACKET_BROADCAST;
254		else
255			pkt_type = PACKET_MULTICAST;
256
257		rcu_read_lock();
258		list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
259			if (tx_pkt && (ipvlan->dev == skb->dev))
260				continue;
261			if (!test_bit(mac_hash, ipvlan->mac_filters))
262				continue;
263			if (!(ipvlan->dev->flags & IFF_UP))
264				continue;
265			ret = NET_RX_DROP;
266			len = skb->len + ETH_HLEN;
267			nskb = skb_clone(skb, GFP_ATOMIC);
268			local_bh_disable();
269			if (nskb) {
270				consumed = true;
271				nskb->pkt_type = pkt_type;
272				nskb->dev = ipvlan->dev;
273				if (tx_pkt)
274					ret = dev_forward_skb(ipvlan->dev, nskb);
275				else
276					ret = netif_rx(nskb);
277			}
278			ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
279			local_bh_enable();
280		}
281		rcu_read_unlock();
282
283		if (tx_pkt) {
284			/* If the packet originated here, send it out. */
285			skb->dev = port->dev;
286			skb->pkt_type = pkt_type;
287			dev_queue_xmit(skb);
288		} else {
289			if (consumed)
290				consume_skb(skb);
291			else
292				kfree_skb(skb);
293		}
294		dev_put(dev);
295		cond_resched();
296	}
297}
298
299static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
300{
301	bool xnet = true;
302
303	if (dev)
304		xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
305
306	skb_scrub_packet(skb, xnet);
307	if (dev)
308		skb->dev = dev;
309}
310
311static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
312			    bool local)
313{
314	struct ipvl_dev *ipvlan = addr->master;
315	struct net_device *dev = ipvlan->dev;
316	unsigned int len;
317	rx_handler_result_t ret = RX_HANDLER_CONSUMED;
318	bool success = false;
319	struct sk_buff *skb = *pskb;
320
321	len = skb->len + ETH_HLEN;
322	/* Only packets exchanged between two local slaves need to have
323	 * device-up check as well as skb-share check.
324	 */
325	if (local) {
326		if (unlikely(!(dev->flags & IFF_UP))) {
327			kfree_skb(skb);
328			goto out;
329		}
330
331		skb = skb_share_check(skb, GFP_ATOMIC);
332		if (!skb)
333			goto out;
334
335		*pskb = skb;
336	}
337
338	if (local) {
339		skb->pkt_type = PACKET_HOST;
340		if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
341			success = true;
342	} else {
343		skb->dev = dev;
344		ret = RX_HANDLER_ANOTHER;
345		success = true;
346	}
347
348out:
349	ipvlan_count_rx(ipvlan, len, success, false);
350	return ret;
351}
352
353struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port, void *lyr3h,
354				     int addr_type, bool use_dest)
355{
356	struct ipvl_addr *addr = NULL;
357
358	switch (addr_type) {
359#if IS_ENABLED(CONFIG_IPV6)
360	case IPVL_IPV6: {
361		struct ipv6hdr *ip6h;
362		struct in6_addr *i6addr;
363
364		ip6h = (struct ipv6hdr *)lyr3h;
365		i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
366		addr = ipvlan_ht_addr_lookup(port, i6addr, true);
367		break;
368	}
369	case IPVL_ICMPV6: {
370		struct nd_msg *ndmh;
371		struct in6_addr *i6addr;
372
373		/* Make sure that the NeighborSolicitation ICMPv6 packets
374		 * are handled to avoid DAD issue.
375		 */
376		ndmh = (struct nd_msg *)lyr3h;
377		if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
378			i6addr = &ndmh->target;
379			addr = ipvlan_ht_addr_lookup(port, i6addr, true);
380		}
381		break;
382	}
383#endif
384	case IPVL_IPV4: {
385		struct iphdr *ip4h;
386		__be32 *i4addr;
387
388		ip4h = (struct iphdr *)lyr3h;
389		i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
390		addr = ipvlan_ht_addr_lookup(port, i4addr, false);
391		break;
392	}
393	case IPVL_ARP: {
394		struct arphdr *arph;
395		unsigned char *arp_ptr;
396		__be32 dip;
397
398		arph = (struct arphdr *)lyr3h;
399		arp_ptr = (unsigned char *)(arph + 1);
400		if (use_dest)
401			arp_ptr += (2 * port->dev->addr_len) + 4;
402		else
403			arp_ptr += port->dev->addr_len;
404
405		memcpy(&dip, arp_ptr, 4);
406		addr = ipvlan_ht_addr_lookup(port, &dip, false);
407		break;
408	}
409	}
410
411	return addr;
412}
413
414static noinline_for_stack int ipvlan_process_v4_outbound(struct sk_buff *skb)
415{
416	const struct iphdr *ip4h = ip_hdr(skb);
417	struct net_device *dev = skb->dev;
418	struct net *net = dev_net(dev);
419	struct rtable *rt;
420	int err, ret = NET_XMIT_DROP;
 
 
421	struct flowi4 fl4 = {
422		.flowi4_oif = dev->ifindex,
423		.flowi4_tos = RT_TOS(ip4h->tos),
424		.flowi4_flags = FLOWI_FLAG_ANYSRC,
425		.flowi4_mark = skb->mark,
426		.daddr = ip4h->daddr,
427		.saddr = ip4h->saddr,
428	};
429
 
 
 
 
 
 
 
 
430	rt = ip_route_output_flow(net, &fl4, NULL);
431	if (IS_ERR(rt))
432		goto err;
433
434	if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
435		ip_rt_put(rt);
436		goto err;
437	}
438	skb_dst_set(skb, &rt->dst);
439
440	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
441
442	err = ip_local_out(net, skb->sk, skb);
443	if (unlikely(net_xmit_eval(err)))
444		DEV_STATS_INC(dev, tx_errors);
445	else
446		ret = NET_XMIT_SUCCESS;
447	goto out;
448err:
449	DEV_STATS_INC(dev, tx_errors);
450	kfree_skb(skb);
451out:
452	return ret;
453}
454
455#if IS_ENABLED(CONFIG_IPV6)
456
457static noinline_for_stack int
458ipvlan_route_v6_outbound(struct net_device *dev, struct sk_buff *skb)
459{
460	const struct ipv6hdr *ip6h = ipv6_hdr(skb);
461	struct flowi6 fl6 = {
462		.flowi6_oif = dev->ifindex,
463		.daddr = ip6h->daddr,
464		.saddr = ip6h->saddr,
465		.flowi6_flags = FLOWI_FLAG_ANYSRC,
466		.flowlabel = ip6_flowinfo(ip6h),
467		.flowi6_mark = skb->mark,
468		.flowi6_proto = ip6h->nexthdr,
469	};
470	struct dst_entry *dst;
471	int err;
472
473	dst = ip6_route_output(dev_net(dev), NULL, &fl6);
474	err = dst->error;
475	if (err) {
476		dst_release(dst);
477		return err;
478	}
479	skb_dst_set(skb, dst);
480	return 0;
481}
482
483static int ipvlan_process_v6_outbound(struct sk_buff *skb)
484{
485	struct net_device *dev = skb->dev;
486	int err, ret = NET_XMIT_DROP;
487
 
 
 
 
 
 
488	err = ipvlan_route_v6_outbound(dev, skb);
489	if (unlikely(err)) {
490		DEV_STATS_INC(dev, tx_errors);
491		kfree_skb(skb);
492		return err;
493	}
494
495	memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
496
497	err = ip6_local_out(dev_net(dev), skb->sk, skb);
498	if (unlikely(net_xmit_eval(err)))
499		DEV_STATS_INC(dev, tx_errors);
500	else
501		ret = NET_XMIT_SUCCESS;
502	return ret;
503}
504#else
505static int ipvlan_process_v6_outbound(struct sk_buff *skb)
506{
507	return NET_XMIT_DROP;
508}
509#endif
510
511static int ipvlan_process_outbound(struct sk_buff *skb)
512{
513	int ret = NET_XMIT_DROP;
514
515	/* The ipvlan is a pseudo-L2 device, so the packets that we receive
516	 * will have L2; which need to discarded and processed further
517	 * in the net-ns of the main-device.
518	 */
519	if (skb_mac_header_was_set(skb)) {
520		/* In this mode we dont care about
521		 * multicast and broadcast traffic */
522		struct ethhdr *ethh = eth_hdr(skb);
523
524		if (is_multicast_ether_addr(ethh->h_dest)) {
525			pr_debug_ratelimited(
526				"Dropped {multi|broad}cast of type=[%x]\n",
527				ntohs(skb->protocol));
528			kfree_skb(skb);
529			goto out;
530		}
531
532		skb_pull(skb, sizeof(*ethh));
533		skb->mac_header = (typeof(skb->mac_header))~0U;
534		skb_reset_network_header(skb);
535	}
536
537	if (skb->protocol == htons(ETH_P_IPV6))
538		ret = ipvlan_process_v6_outbound(skb);
539	else if (skb->protocol == htons(ETH_P_IP))
540		ret = ipvlan_process_v4_outbound(skb);
541	else {
542		pr_warn_ratelimited("Dropped outbound packet type=%x\n",
543				    ntohs(skb->protocol));
544		kfree_skb(skb);
545	}
546out:
547	return ret;
548}
549
550static void ipvlan_multicast_enqueue(struct ipvl_port *port,
551				     struct sk_buff *skb, bool tx_pkt)
552{
553	if (skb->protocol == htons(ETH_P_PAUSE)) {
554		kfree_skb(skb);
555		return;
556	}
557
558	/* Record that the deferred packet is from TX or RX path. By
559	 * looking at mac-addresses on packet will lead to erronus decisions.
560	 * (This would be true for a loopback-mode on master device or a
561	 * hair-pin mode of the switch.)
562	 */
563	IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
564
565	spin_lock(&port->backlog.lock);
566	if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
567		dev_hold(skb->dev);
568		__skb_queue_tail(&port->backlog, skb);
569		spin_unlock(&port->backlog.lock);
570		schedule_work(&port->wq);
571	} else {
572		spin_unlock(&port->backlog.lock);
573		dev_core_stats_rx_dropped_inc(skb->dev);
574		kfree_skb(skb);
575	}
576}
577
578static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
579{
580	const struct ipvl_dev *ipvlan = netdev_priv(dev);
581	void *lyr3h;
582	struct ipvl_addr *addr;
583	int addr_type;
584
585	lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
586	if (!lyr3h)
587		goto out;
588
589	if (!ipvlan_is_vepa(ipvlan->port)) {
590		addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
591		if (addr) {
592			if (ipvlan_is_private(ipvlan->port)) {
593				consume_skb(skb);
594				return NET_XMIT_DROP;
595			}
596			ipvlan_rcv_frame(addr, &skb, true);
597			return NET_XMIT_SUCCESS;
598		}
599	}
600out:
601	ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
602	return ipvlan_process_outbound(skb);
603}
604
605static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
606{
607	const struct ipvl_dev *ipvlan = netdev_priv(dev);
608	struct ethhdr *eth = skb_eth_hdr(skb);
609	struct ipvl_addr *addr;
610	void *lyr3h;
611	int addr_type;
612
613	if (!ipvlan_is_vepa(ipvlan->port) &&
614	    ether_addr_equal(eth->h_dest, eth->h_source)) {
615		lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
616		if (lyr3h) {
617			addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
618			if (addr) {
619				if (ipvlan_is_private(ipvlan->port)) {
620					consume_skb(skb);
621					return NET_XMIT_DROP;
622				}
623				ipvlan_rcv_frame(addr, &skb, true);
624				return NET_XMIT_SUCCESS;
625			}
626		}
627		skb = skb_share_check(skb, GFP_ATOMIC);
628		if (!skb)
629			return NET_XMIT_DROP;
630
631		/* Packet definitely does not belong to any of the
632		 * virtual devices, but the dest is local. So forward
633		 * the skb for the main-dev. At the RX side we just return
634		 * RX_PASS for it to be processed further on the stack.
635		 */
636		dev_forward_skb(ipvlan->phy_dev, skb);
637		return NET_XMIT_SUCCESS;
638
639	} else if (is_multicast_ether_addr(eth->h_dest)) {
640		skb_reset_mac_header(skb);
641		ipvlan_skb_crossing_ns(skb, NULL);
642		ipvlan_multicast_enqueue(ipvlan->port, skb, true);
643		return NET_XMIT_SUCCESS;
644	}
645
646	skb->dev = ipvlan->phy_dev;
647	return dev_queue_xmit(skb);
648}
649
650int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
651{
652	struct ipvl_dev *ipvlan = netdev_priv(dev);
653	struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
654
655	if (!port)
656		goto out;
657
658	if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
659		goto out;
660
661	switch(port->mode) {
662	case IPVLAN_MODE_L2:
663		return ipvlan_xmit_mode_l2(skb, dev);
664	case IPVLAN_MODE_L3:
665#ifdef CONFIG_IPVLAN_L3S
666	case IPVLAN_MODE_L3S:
667#endif
668		return ipvlan_xmit_mode_l3(skb, dev);
669	}
670
671	/* Should not reach here */
672	WARN_ONCE(true, "%s called for mode = [%x]\n", __func__, port->mode);
673out:
674	kfree_skb(skb);
675	return NET_XMIT_DROP;
676}
677
678static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
679{
680	struct ethhdr *eth = eth_hdr(skb);
681	struct ipvl_addr *addr;
682	void *lyr3h;
683	int addr_type;
684
685	if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
686		lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
687		if (!lyr3h)
688			return true;
689
690		addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
691		if (addr)
692			return false;
693	}
694
695	return true;
696}
697
698static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
699						 struct ipvl_port *port)
700{
701	void *lyr3h;
702	int addr_type;
703	struct ipvl_addr *addr;
704	struct sk_buff *skb = *pskb;
705	rx_handler_result_t ret = RX_HANDLER_PASS;
706
707	lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
708	if (!lyr3h)
709		goto out;
710
711	addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
712	if (addr)
713		ret = ipvlan_rcv_frame(addr, pskb, false);
714
715out:
716	return ret;
717}
718
719static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
720						 struct ipvl_port *port)
721{
722	struct sk_buff *skb = *pskb;
723	struct ethhdr *eth = eth_hdr(skb);
724	rx_handler_result_t ret = RX_HANDLER_PASS;
725
726	if (is_multicast_ether_addr(eth->h_dest)) {
727		if (ipvlan_external_frame(skb, port)) {
728			struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
729
730			/* External frames are queued for device local
731			 * distribution, but a copy is given to master
732			 * straight away to avoid sending duplicates later
733			 * when work-queue processes this frame. This is
734			 * achieved by returning RX_HANDLER_PASS.
735			 */
736			if (nskb) {
737				ipvlan_skb_crossing_ns(nskb, NULL);
738				ipvlan_multicast_enqueue(port, nskb, false);
739			}
740		}
741	} else {
742		/* Perform like l3 mode for non-multicast packet */
743		ret = ipvlan_handle_mode_l3(pskb, port);
744	}
745
746	return ret;
747}
748
749rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
750{
751	struct sk_buff *skb = *pskb;
752	struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
753
754	if (!port)
755		return RX_HANDLER_PASS;
756
757	switch (port->mode) {
758	case IPVLAN_MODE_L2:
759		return ipvlan_handle_mode_l2(pskb, port);
760	case IPVLAN_MODE_L3:
761		return ipvlan_handle_mode_l3(pskb, port);
762#ifdef CONFIG_IPVLAN_L3S
763	case IPVLAN_MODE_L3S:
764		return RX_HANDLER_PASS;
765#endif
766	}
767
768	/* Should not reach here */
769	WARN_ONCE(true, "%s called for mode = [%x]\n", __func__, port->mode);
770	kfree_skb(skb);
771	return RX_HANDLER_CONSUMED;
772}
v6.13.7
  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
  3 */
  4
  5#include <net/inet_dscp.h>
  6#include <net/ip.h>
  7
  8#include "ipvlan.h"
  9
 10static u32 ipvlan_jhash_secret __read_mostly;
 11
 12void ipvlan_init_secret(void)
 13{
 14	net_get_random_once(&ipvlan_jhash_secret, sizeof(ipvlan_jhash_secret));
 15}
 16
 17void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
 18			    unsigned int len, bool success, bool mcast)
 19{
 20	if (likely(success)) {
 21		struct ipvl_pcpu_stats *pcptr;
 22
 23		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 24		u64_stats_update_begin(&pcptr->syncp);
 25		u64_stats_inc(&pcptr->rx_pkts);
 26		u64_stats_add(&pcptr->rx_bytes, len);
 27		if (mcast)
 28			u64_stats_inc(&pcptr->rx_mcast);
 29		u64_stats_update_end(&pcptr->syncp);
 30	} else {
 31		this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
 32	}
 33}
 34EXPORT_SYMBOL_GPL(ipvlan_count_rx);
 35
 36#if IS_ENABLED(CONFIG_IPV6)
 37static u8 ipvlan_get_v6_hash(const void *iaddr)
 38{
 39	const struct in6_addr *ip6_addr = iaddr;
 40
 41	return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
 42	       IPVLAN_HASH_MASK;
 43}
 44#else
 45static u8 ipvlan_get_v6_hash(const void *iaddr)
 46{
 47	return 0;
 48}
 49#endif
 50
 51static u8 ipvlan_get_v4_hash(const void *iaddr)
 52{
 53	const struct in_addr *ip4_addr = iaddr;
 54
 55	return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
 56	       IPVLAN_HASH_MASK;
 57}
 58
 59static bool addr_equal(bool is_v6, struct ipvl_addr *addr, const void *iaddr)
 60{
 61	if (!is_v6 && addr->atype == IPVL_IPV4) {
 62		struct in_addr *i4addr = (struct in_addr *)iaddr;
 63
 64		return addr->ip4addr.s_addr == i4addr->s_addr;
 65#if IS_ENABLED(CONFIG_IPV6)
 66	} else if (is_v6 && addr->atype == IPVL_IPV6) {
 67		struct in6_addr *i6addr = (struct in6_addr *)iaddr;
 68
 69		return ipv6_addr_equal(&addr->ip6addr, i6addr);
 70#endif
 71	}
 72
 73	return false;
 74}
 75
 76static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
 77					       const void *iaddr, bool is_v6)
 78{
 79	struct ipvl_addr *addr;
 80	u8 hash;
 81
 82	hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
 83	       ipvlan_get_v4_hash(iaddr);
 84	hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode)
 85		if (addr_equal(is_v6, addr, iaddr))
 86			return addr;
 87	return NULL;
 88}
 89
 90void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
 91{
 92	struct ipvl_port *port = ipvlan->port;
 93	u8 hash;
 94
 95	hash = (addr->atype == IPVL_IPV6) ?
 96	       ipvlan_get_v6_hash(&addr->ip6addr) :
 97	       ipvlan_get_v4_hash(&addr->ip4addr);
 98	if (hlist_unhashed(&addr->hlnode))
 99		hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
100}
101
102void ipvlan_ht_addr_del(struct ipvl_addr *addr)
103{
104	hlist_del_init_rcu(&addr->hlnode);
105}
106
107struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
108				   const void *iaddr, bool is_v6)
109{
110	struct ipvl_addr *addr, *ret = NULL;
111
112	rcu_read_lock();
113	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode) {
114		if (addr_equal(is_v6, addr, iaddr)) {
115			ret = addr;
116			break;
117		}
118	}
119	rcu_read_unlock();
120	return ret;
121}
122
123bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
124{
125	struct ipvl_dev *ipvlan;
126	bool ret = false;
127
128	rcu_read_lock();
129	list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
130		if (ipvlan_find_addr(ipvlan, iaddr, is_v6)) {
131			ret = true;
132			break;
133		}
134	}
135	rcu_read_unlock();
136	return ret;
137}
138
139void *ipvlan_get_L3_hdr(struct ipvl_port *port, struct sk_buff *skb, int *type)
140{
141	void *lyr3h = NULL;
142
143	switch (skb->protocol) {
144	case htons(ETH_P_ARP): {
145		struct arphdr *arph;
146
147		if (unlikely(!pskb_may_pull(skb, arp_hdr_len(port->dev))))
148			return NULL;
149
150		arph = arp_hdr(skb);
151		*type = IPVL_ARP;
152		lyr3h = arph;
153		break;
154	}
155	case htons(ETH_P_IP): {
156		u32 pktlen;
157		struct iphdr *ip4h;
158
159		if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
160			return NULL;
161
162		ip4h = ip_hdr(skb);
163		pktlen = skb_ip_totlen(skb);
164		if (ip4h->ihl < 5 || ip4h->version != 4)
165			return NULL;
166		if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
167			return NULL;
168
169		*type = IPVL_IPV4;
170		lyr3h = ip4h;
171		break;
172	}
173#if IS_ENABLED(CONFIG_IPV6)
174	case htons(ETH_P_IPV6): {
175		struct ipv6hdr *ip6h;
176
177		if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
178			return NULL;
179
180		ip6h = ipv6_hdr(skb);
181		if (ip6h->version != 6)
182			return NULL;
183
184		*type = IPVL_IPV6;
185		lyr3h = ip6h;
186		/* Only Neighbour Solicitation pkts need different treatment */
187		if (ipv6_addr_any(&ip6h->saddr) &&
188		    ip6h->nexthdr == NEXTHDR_ICMP) {
189			struct icmp6hdr	*icmph;
190
191			if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph))))
192				return NULL;
193
194			ip6h = ipv6_hdr(skb);
195			icmph = (struct icmp6hdr *)(ip6h + 1);
196
197			if (icmph->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
198				/* Need to access the ipv6 address in body */
199				if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph)
200						+ sizeof(struct in6_addr))))
201					return NULL;
202
203				ip6h = ipv6_hdr(skb);
204				icmph = (struct icmp6hdr *)(ip6h + 1);
205			}
206
207			*type = IPVL_ICMPV6;
208			lyr3h = icmph;
209		}
210		break;
211	}
212#endif
213	default:
214		return NULL;
215	}
216
217	return lyr3h;
218}
219
220unsigned int ipvlan_mac_hash(const unsigned char *addr)
221{
222	u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
223			       ipvlan_jhash_secret);
224
225	return hash & IPVLAN_MAC_FILTER_MASK;
226}
227
228void ipvlan_process_multicast(struct work_struct *work)
229{
230	struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
231	struct ethhdr *ethh;
232	struct ipvl_dev *ipvlan;
233	struct sk_buff *skb, *nskb;
234	struct sk_buff_head list;
235	unsigned int len;
236	unsigned int mac_hash;
237	int ret;
238	u8 pkt_type;
239	bool tx_pkt;
240
241	__skb_queue_head_init(&list);
242
243	spin_lock_bh(&port->backlog.lock);
244	skb_queue_splice_tail_init(&port->backlog, &list);
245	spin_unlock_bh(&port->backlog.lock);
246
247	while ((skb = __skb_dequeue(&list)) != NULL) {
248		struct net_device *dev = skb->dev;
249		bool consumed = false;
250
251		ethh = eth_hdr(skb);
252		tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
253		mac_hash = ipvlan_mac_hash(ethh->h_dest);
254
255		if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
256			pkt_type = PACKET_BROADCAST;
257		else
258			pkt_type = PACKET_MULTICAST;
259
260		rcu_read_lock();
261		list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
262			if (tx_pkt && (ipvlan->dev == skb->dev))
263				continue;
264			if (!test_bit(mac_hash, ipvlan->mac_filters))
265				continue;
266			if (!(ipvlan->dev->flags & IFF_UP))
267				continue;
268			ret = NET_RX_DROP;
269			len = skb->len + ETH_HLEN;
270			nskb = skb_clone(skb, GFP_ATOMIC);
271			local_bh_disable();
272			if (nskb) {
273				consumed = true;
274				nskb->pkt_type = pkt_type;
275				nskb->dev = ipvlan->dev;
276				if (tx_pkt)
277					ret = dev_forward_skb(ipvlan->dev, nskb);
278				else
279					ret = netif_rx(nskb);
280			}
281			ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
282			local_bh_enable();
283		}
284		rcu_read_unlock();
285
286		if (tx_pkt) {
287			/* If the packet originated here, send it out. */
288			skb->dev = port->dev;
289			skb->pkt_type = pkt_type;
290			dev_queue_xmit(skb);
291		} else {
292			if (consumed)
293				consume_skb(skb);
294			else
295				kfree_skb(skb);
296		}
297		dev_put(dev);
298		cond_resched();
299	}
300}
301
302static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
303{
304	bool xnet = true;
305
306	if (dev)
307		xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
308
309	skb_scrub_packet(skb, xnet);
310	if (dev)
311		skb->dev = dev;
312}
313
314static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
315			    bool local)
316{
317	struct ipvl_dev *ipvlan = addr->master;
318	struct net_device *dev = ipvlan->dev;
319	unsigned int len;
320	rx_handler_result_t ret = RX_HANDLER_CONSUMED;
321	bool success = false;
322	struct sk_buff *skb = *pskb;
323
324	len = skb->len + ETH_HLEN;
325	/* Only packets exchanged between two local slaves need to have
326	 * device-up check as well as skb-share check.
327	 */
328	if (local) {
329		if (unlikely(!(dev->flags & IFF_UP))) {
330			kfree_skb(skb);
331			goto out;
332		}
333
334		skb = skb_share_check(skb, GFP_ATOMIC);
335		if (!skb)
336			goto out;
337
338		*pskb = skb;
339	}
340
341	if (local) {
342		skb->pkt_type = PACKET_HOST;
343		if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
344			success = true;
345	} else {
346		skb->dev = dev;
347		ret = RX_HANDLER_ANOTHER;
348		success = true;
349	}
350
351out:
352	ipvlan_count_rx(ipvlan, len, success, false);
353	return ret;
354}
355
356struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port, void *lyr3h,
357				     int addr_type, bool use_dest)
358{
359	struct ipvl_addr *addr = NULL;
360
361	switch (addr_type) {
362#if IS_ENABLED(CONFIG_IPV6)
363	case IPVL_IPV6: {
364		struct ipv6hdr *ip6h;
365		struct in6_addr *i6addr;
366
367		ip6h = (struct ipv6hdr *)lyr3h;
368		i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
369		addr = ipvlan_ht_addr_lookup(port, i6addr, true);
370		break;
371	}
372	case IPVL_ICMPV6: {
373		struct nd_msg *ndmh;
374		struct in6_addr *i6addr;
375
376		/* Make sure that the NeighborSolicitation ICMPv6 packets
377		 * are handled to avoid DAD issue.
378		 */
379		ndmh = (struct nd_msg *)lyr3h;
380		if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
381			i6addr = &ndmh->target;
382			addr = ipvlan_ht_addr_lookup(port, i6addr, true);
383		}
384		break;
385	}
386#endif
387	case IPVL_IPV4: {
388		struct iphdr *ip4h;
389		__be32 *i4addr;
390
391		ip4h = (struct iphdr *)lyr3h;
392		i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
393		addr = ipvlan_ht_addr_lookup(port, i4addr, false);
394		break;
395	}
396	case IPVL_ARP: {
397		struct arphdr *arph;
398		unsigned char *arp_ptr;
399		__be32 dip;
400
401		arph = (struct arphdr *)lyr3h;
402		arp_ptr = (unsigned char *)(arph + 1);
403		if (use_dest)
404			arp_ptr += (2 * port->dev->addr_len) + 4;
405		else
406			arp_ptr += port->dev->addr_len;
407
408		memcpy(&dip, arp_ptr, 4);
409		addr = ipvlan_ht_addr_lookup(port, &dip, false);
410		break;
411	}
412	}
413
414	return addr;
415}
416
417static noinline_for_stack int ipvlan_process_v4_outbound(struct sk_buff *skb)
418{
 
419	struct net_device *dev = skb->dev;
420	struct net *net = dev_net(dev);
 
421	int err, ret = NET_XMIT_DROP;
422	const struct iphdr *ip4h;
423	struct rtable *rt;
424	struct flowi4 fl4 = {
425		.flowi4_oif = dev->ifindex,
 
426		.flowi4_flags = FLOWI_FLAG_ANYSRC,
427		.flowi4_mark = skb->mark,
 
 
428	};
429
430	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)))
431		goto err;
432
433	ip4h = ip_hdr(skb);
434	fl4.daddr = ip4h->daddr;
435	fl4.saddr = ip4h->saddr;
436	fl4.flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip4h));
437
438	rt = ip_route_output_flow(net, &fl4, NULL);
439	if (IS_ERR(rt))
440		goto err;
441
442	if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
443		ip_rt_put(rt);
444		goto err;
445	}
446	skb_dst_set(skb, &rt->dst);
447
448	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
449
450	err = ip_local_out(net, NULL, skb);
451	if (unlikely(net_xmit_eval(err)))
452		DEV_STATS_INC(dev, tx_errors);
453	else
454		ret = NET_XMIT_SUCCESS;
455	goto out;
456err:
457	DEV_STATS_INC(dev, tx_errors);
458	kfree_skb(skb);
459out:
460	return ret;
461}
462
463#if IS_ENABLED(CONFIG_IPV6)
464
465static noinline_for_stack int
466ipvlan_route_v6_outbound(struct net_device *dev, struct sk_buff *skb)
467{
468	const struct ipv6hdr *ip6h = ipv6_hdr(skb);
469	struct flowi6 fl6 = {
470		.flowi6_oif = dev->ifindex,
471		.daddr = ip6h->daddr,
472		.saddr = ip6h->saddr,
473		.flowi6_flags = FLOWI_FLAG_ANYSRC,
474		.flowlabel = ip6_flowinfo(ip6h),
475		.flowi6_mark = skb->mark,
476		.flowi6_proto = ip6h->nexthdr,
477	};
478	struct dst_entry *dst;
479	int err;
480
481	dst = ip6_route_output(dev_net(dev), NULL, &fl6);
482	err = dst->error;
483	if (err) {
484		dst_release(dst);
485		return err;
486	}
487	skb_dst_set(skb, dst);
488	return 0;
489}
490
491static int ipvlan_process_v6_outbound(struct sk_buff *skb)
492{
493	struct net_device *dev = skb->dev;
494	int err, ret = NET_XMIT_DROP;
495
496	if (!pskb_network_may_pull(skb, sizeof(struct ipv6hdr))) {
497		DEV_STATS_INC(dev, tx_errors);
498		kfree_skb(skb);
499		return ret;
500	}
501
502	err = ipvlan_route_v6_outbound(dev, skb);
503	if (unlikely(err)) {
504		DEV_STATS_INC(dev, tx_errors);
505		kfree_skb(skb);
506		return err;
507	}
508
509	memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
510
511	err = ip6_local_out(dev_net(dev), NULL, skb);
512	if (unlikely(net_xmit_eval(err)))
513		DEV_STATS_INC(dev, tx_errors);
514	else
515		ret = NET_XMIT_SUCCESS;
516	return ret;
517}
518#else
519static int ipvlan_process_v6_outbound(struct sk_buff *skb)
520{
521	return NET_XMIT_DROP;
522}
523#endif
524
525static int ipvlan_process_outbound(struct sk_buff *skb)
526{
527	int ret = NET_XMIT_DROP;
528
529	/* The ipvlan is a pseudo-L2 device, so the packets that we receive
530	 * will have L2; which need to discarded and processed further
531	 * in the net-ns of the main-device.
532	 */
533	if (skb_mac_header_was_set(skb)) {
534		/* In this mode we dont care about
535		 * multicast and broadcast traffic */
536		struct ethhdr *ethh = eth_hdr(skb);
537
538		if (is_multicast_ether_addr(ethh->h_dest)) {
539			pr_debug_ratelimited(
540				"Dropped {multi|broad}cast of type=[%x]\n",
541				ntohs(skb->protocol));
542			kfree_skb(skb);
543			goto out;
544		}
545
546		skb_pull(skb, sizeof(*ethh));
547		skb->mac_header = (typeof(skb->mac_header))~0U;
548		skb_reset_network_header(skb);
549	}
550
551	if (skb->protocol == htons(ETH_P_IPV6))
552		ret = ipvlan_process_v6_outbound(skb);
553	else if (skb->protocol == htons(ETH_P_IP))
554		ret = ipvlan_process_v4_outbound(skb);
555	else {
556		pr_warn_ratelimited("Dropped outbound packet type=%x\n",
557				    ntohs(skb->protocol));
558		kfree_skb(skb);
559	}
560out:
561	return ret;
562}
563
564static void ipvlan_multicast_enqueue(struct ipvl_port *port,
565				     struct sk_buff *skb, bool tx_pkt)
566{
567	if (skb->protocol == htons(ETH_P_PAUSE)) {
568		kfree_skb(skb);
569		return;
570	}
571
572	/* Record that the deferred packet is from TX or RX path. By
573	 * looking at mac-addresses on packet will lead to erronus decisions.
574	 * (This would be true for a loopback-mode on master device or a
575	 * hair-pin mode of the switch.)
576	 */
577	IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
578
579	spin_lock(&port->backlog.lock);
580	if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
581		dev_hold(skb->dev);
582		__skb_queue_tail(&port->backlog, skb);
583		spin_unlock(&port->backlog.lock);
584		schedule_work(&port->wq);
585	} else {
586		spin_unlock(&port->backlog.lock);
587		dev_core_stats_rx_dropped_inc(skb->dev);
588		kfree_skb(skb);
589	}
590}
591
592static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
593{
594	const struct ipvl_dev *ipvlan = netdev_priv(dev);
595	void *lyr3h;
596	struct ipvl_addr *addr;
597	int addr_type;
598
599	lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
600	if (!lyr3h)
601		goto out;
602
603	if (!ipvlan_is_vepa(ipvlan->port)) {
604		addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
605		if (addr) {
606			if (ipvlan_is_private(ipvlan->port)) {
607				consume_skb(skb);
608				return NET_XMIT_DROP;
609			}
610			ipvlan_rcv_frame(addr, &skb, true);
611			return NET_XMIT_SUCCESS;
612		}
613	}
614out:
615	ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
616	return ipvlan_process_outbound(skb);
617}
618
619static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
620{
621	const struct ipvl_dev *ipvlan = netdev_priv(dev);
622	struct ethhdr *eth = skb_eth_hdr(skb);
623	struct ipvl_addr *addr;
624	void *lyr3h;
625	int addr_type;
626
627	if (!ipvlan_is_vepa(ipvlan->port) &&
628	    ether_addr_equal(eth->h_dest, eth->h_source)) {
629		lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
630		if (lyr3h) {
631			addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
632			if (addr) {
633				if (ipvlan_is_private(ipvlan->port)) {
634					consume_skb(skb);
635					return NET_XMIT_DROP;
636				}
637				ipvlan_rcv_frame(addr, &skb, true);
638				return NET_XMIT_SUCCESS;
639			}
640		}
641		skb = skb_share_check(skb, GFP_ATOMIC);
642		if (!skb)
643			return NET_XMIT_DROP;
644
645		/* Packet definitely does not belong to any of the
646		 * virtual devices, but the dest is local. So forward
647		 * the skb for the main-dev. At the RX side we just return
648		 * RX_PASS for it to be processed further on the stack.
649		 */
650		dev_forward_skb(ipvlan->phy_dev, skb);
651		return NET_XMIT_SUCCESS;
652
653	} else if (is_multicast_ether_addr(eth->h_dest)) {
654		skb_reset_mac_header(skb);
655		ipvlan_skb_crossing_ns(skb, NULL);
656		ipvlan_multicast_enqueue(ipvlan->port, skb, true);
657		return NET_XMIT_SUCCESS;
658	}
659
660	skb->dev = ipvlan->phy_dev;
661	return dev_queue_xmit(skb);
662}
663
664int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
665{
666	struct ipvl_dev *ipvlan = netdev_priv(dev);
667	struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
668
669	if (!port)
670		goto out;
671
672	if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
673		goto out;
674
675	switch(port->mode) {
676	case IPVLAN_MODE_L2:
677		return ipvlan_xmit_mode_l2(skb, dev);
678	case IPVLAN_MODE_L3:
679#ifdef CONFIG_IPVLAN_L3S
680	case IPVLAN_MODE_L3S:
681#endif
682		return ipvlan_xmit_mode_l3(skb, dev);
683	}
684
685	/* Should not reach here */
686	WARN_ONCE(true, "%s called for mode = [%x]\n", __func__, port->mode);
687out:
688	kfree_skb(skb);
689	return NET_XMIT_DROP;
690}
691
692static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
693{
694	struct ethhdr *eth = eth_hdr(skb);
695	struct ipvl_addr *addr;
696	void *lyr3h;
697	int addr_type;
698
699	if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
700		lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
701		if (!lyr3h)
702			return true;
703
704		addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
705		if (addr)
706			return false;
707	}
708
709	return true;
710}
711
712static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
713						 struct ipvl_port *port)
714{
715	void *lyr3h;
716	int addr_type;
717	struct ipvl_addr *addr;
718	struct sk_buff *skb = *pskb;
719	rx_handler_result_t ret = RX_HANDLER_PASS;
720
721	lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
722	if (!lyr3h)
723		goto out;
724
725	addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
726	if (addr)
727		ret = ipvlan_rcv_frame(addr, pskb, false);
728
729out:
730	return ret;
731}
732
733static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
734						 struct ipvl_port *port)
735{
736	struct sk_buff *skb = *pskb;
737	struct ethhdr *eth = eth_hdr(skb);
738	rx_handler_result_t ret = RX_HANDLER_PASS;
739
740	if (is_multicast_ether_addr(eth->h_dest)) {
741		if (ipvlan_external_frame(skb, port)) {
742			struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
743
744			/* External frames are queued for device local
745			 * distribution, but a copy is given to master
746			 * straight away to avoid sending duplicates later
747			 * when work-queue processes this frame. This is
748			 * achieved by returning RX_HANDLER_PASS.
749			 */
750			if (nskb) {
751				ipvlan_skb_crossing_ns(nskb, NULL);
752				ipvlan_multicast_enqueue(port, nskb, false);
753			}
754		}
755	} else {
756		/* Perform like l3 mode for non-multicast packet */
757		ret = ipvlan_handle_mode_l3(pskb, port);
758	}
759
760	return ret;
761}
762
763rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
764{
765	struct sk_buff *skb = *pskb;
766	struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
767
768	if (!port)
769		return RX_HANDLER_PASS;
770
771	switch (port->mode) {
772	case IPVLAN_MODE_L2:
773		return ipvlan_handle_mode_l2(pskb, port);
774	case IPVLAN_MODE_L3:
775		return ipvlan_handle_mode_l3(pskb, port);
776#ifdef CONFIG_IPVLAN_L3S
777	case IPVLAN_MODE_L3S:
778		return RX_HANDLER_PASS;
779#endif
780	}
781
782	/* Should not reach here */
783	WARN_ONCE(true, "%s called for mode = [%x]\n", __func__, port->mode);
784	kfree_skb(skb);
785	return RX_HANDLER_CONSUMED;
786}