1/* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
2 *
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of the GNU General Public License as
5 * published by the Free Software Foundation; either version 2 of
6 * the License, or (at your option) any later version.
7 *
8 */
9
10#include "ipvlan.h"
11
12static u32 ipvlan_jhash_secret __read_mostly;
13
14void ipvlan_init_secret(void)
15{
16	net_get_random_once(&ipvlan_jhash_secret, sizeof(ipvlan_jhash_secret));
17}
18
19static void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
20			    unsigned int len, bool success, bool mcast)
21{
22	if (!ipvlan)
23		return;
24
25	if (likely(success)) {
26		struct ipvl_pcpu_stats *pcptr;
27
28		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
29		u64_stats_update_begin(&pcptr->syncp);
30		pcptr->rx_pkts++;
31		pcptr->rx_bytes += len;
32		if (mcast)
33			pcptr->rx_mcast++;
34		u64_stats_update_end(&pcptr->syncp);
35	} else {
36		this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
37	}
38}
39
40static u8 ipvlan_get_v6_hash(const void *iaddr)
41{
42	const struct in6_addr *ip6_addr = iaddr;
43
44	return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
45	       IPVLAN_HASH_MASK;
46}
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
56struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
57					const void *iaddr, bool is_v6)
58{
59	struct ipvl_addr *addr;
60	u8 hash;
61
62	hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
63	       ipvlan_get_v4_hash(iaddr);
64	hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode) {
65		if (is_v6 && addr->atype == IPVL_IPV6 &&
66		    ipv6_addr_equal(&addr->ip6addr, iaddr))
67			return addr;
68		else if (!is_v6 && addr->atype == IPVL_IPV4 &&
69			 addr->ip4addr.s_addr ==
70				((struct in_addr *)iaddr)->s_addr)
71			return addr;
72	}
73	return NULL;
74}
75
76void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
77{
78	struct ipvl_port *port = ipvlan->port;
79	u8 hash;
80
81	hash = (addr->atype == IPVL_IPV6) ?
82	       ipvlan_get_v6_hash(&addr->ip6addr) :
83	       ipvlan_get_v4_hash(&addr->ip4addr);
84	if (hlist_unhashed(&addr->hlnode))
85		hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
86}
87
88void ipvlan_ht_addr_del(struct ipvl_addr *addr)
89{
90	hlist_del_init_rcu(&addr->hlnode);
91}
92
93struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
94				   const void *iaddr, bool is_v6)
95{
96	struct ipvl_addr *addr;
97
98	list_for_each_entry(addr, &ipvlan->addrs, anode) {
99		if ((is_v6 && addr->atype == IPVL_IPV6 &&
100		    ipv6_addr_equal(&addr->ip6addr, iaddr)) ||
101		    (!is_v6 && addr->atype == IPVL_IPV4 &&
102		    addr->ip4addr.s_addr == ((struct in_addr *)iaddr)->s_addr))
103			return addr;
104	}
105	return NULL;
106}
107
108bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
109{
110	struct ipvl_dev *ipvlan;
111
112	ASSERT_RTNL();
113
114	list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
115		if (ipvlan_find_addr(ipvlan, iaddr, is_v6))
116			return true;
117	}
118	return false;
119}
120
121static void *ipvlan_get_L3_hdr(struct sk_buff *skb, int *type)
122{
123	void *lyr3h = NULL;
124
125	switch (skb->protocol) {
126	case htons(ETH_P_ARP): {
127		struct arphdr *arph;
128
129		if (unlikely(!pskb_may_pull(skb, sizeof(*arph))))
130			return NULL;
131
132		arph = arp_hdr(skb);
133		*type = IPVL_ARP;
134		lyr3h = arph;
135		break;
136	}
137	case htons(ETH_P_IP): {
138		u32 pktlen;
139		struct iphdr *ip4h;
140
141		if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
142			return NULL;
143
144		ip4h = ip_hdr(skb);
145		pktlen = ntohs(ip4h->tot_len);
146		if (ip4h->ihl < 5 || ip4h->version != 4)
147			return NULL;
148		if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
149			return NULL;
150
151		*type = IPVL_IPV4;
152		lyr3h = ip4h;
153		break;
154	}
155	case htons(ETH_P_IPV6): {
156		struct ipv6hdr *ip6h;
157
158		if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
159			return NULL;
160
161		ip6h = ipv6_hdr(skb);
162		if (ip6h->version != 6)
163			return NULL;
164
165		*type = IPVL_IPV6;
166		lyr3h = ip6h;
167		/* Only Neighbour Solicitation pkts need different treatment */
168		if (ipv6_addr_any(&ip6h->saddr) &&
169		    ip6h->nexthdr == NEXTHDR_ICMP) {
170			*type = IPVL_ICMPV6;
171			lyr3h = ip6h + 1;
172		}
173		break;
174	}
175	default:
176		return NULL;
177	}
178
179	return lyr3h;
180}
181
182unsigned int ipvlan_mac_hash(const unsigned char *addr)
183{
184	u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
185			       ipvlan_jhash_secret);
186
187	return hash & IPVLAN_MAC_FILTER_MASK;
188}
189
190void ipvlan_process_multicast(struct work_struct *work)
191{
192	struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
193	struct ethhdr *ethh;
194	struct ipvl_dev *ipvlan;
195	struct sk_buff *skb, *nskb;
196	struct sk_buff_head list;
197	unsigned int len;
198	unsigned int mac_hash;
199	int ret;
200	u8 pkt_type;
201	bool hlocal, dlocal;
202
203	__skb_queue_head_init(&list);
204
205	spin_lock_bh(&port->backlog.lock);
206	skb_queue_splice_tail_init(&port->backlog, &list);
207	spin_unlock_bh(&port->backlog.lock);
208
209	while ((skb = __skb_dequeue(&list)) != NULL) {
210		ethh = eth_hdr(skb);
211		hlocal = ether_addr_equal(ethh->h_source, port->dev->dev_addr);
212		mac_hash = ipvlan_mac_hash(ethh->h_dest);
213
214		if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
215			pkt_type = PACKET_BROADCAST;
216		else
217			pkt_type = PACKET_MULTICAST;
218
219		dlocal = false;
220		rcu_read_lock();
221		list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
222			if (hlocal && (ipvlan->dev == skb->dev)) {
223				dlocal = true;
224				continue;
225			}
226			if (!test_bit(mac_hash, ipvlan->mac_filters))
227				continue;
228
229			ret = NET_RX_DROP;
230			len = skb->len + ETH_HLEN;
231			nskb = skb_clone(skb, GFP_ATOMIC);
232			if (!nskb)
233				goto acct;
234
235			nskb->pkt_type = pkt_type;
236			nskb->dev = ipvlan->dev;
237			if (hlocal)
238				ret = dev_forward_skb(ipvlan->dev, nskb);
239			else
240				ret = netif_rx(nskb);
241acct:
242			ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
243		}
244		rcu_read_unlock();
245
246		if (dlocal) {
247			/* If the packet originated here, send it out. */
248			skb->dev = port->dev;
249			skb->pkt_type = pkt_type;
250			dev_queue_xmit(skb);
251		} else {
252			kfree_skb(skb);
253		}
254	}
255}
256
257static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
258			    bool local)
259{
260	struct ipvl_dev *ipvlan = addr->master;
261	struct net_device *dev = ipvlan->dev;
262	unsigned int len;
263	rx_handler_result_t ret = RX_HANDLER_CONSUMED;
264	bool success = false;
265	struct sk_buff *skb = *pskb;
266
267	len = skb->len + ETH_HLEN;
268	if (unlikely(!(dev->flags & IFF_UP))) {
269		kfree_skb(skb);
270		goto out;
271	}
272
273	skb = skb_share_check(skb, GFP_ATOMIC);
274	if (!skb)
275		goto out;
276
277	*pskb = skb;
278	skb->dev = dev;
279	skb->pkt_type = PACKET_HOST;
280
281	if (local) {
282		if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
283			success = true;
284	} else {
285		ret = RX_HANDLER_ANOTHER;
286		success = true;
287	}
288
289out:
290	ipvlan_count_rx(ipvlan, len, success, false);
291	return ret;
292}
293
294static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
295					    void *lyr3h, int addr_type,
296					    bool use_dest)
297{
298	struct ipvl_addr *addr = NULL;
299
300	if (addr_type == IPVL_IPV6) {
301		struct ipv6hdr *ip6h;
302		struct in6_addr *i6addr;
303
304		ip6h = (struct ipv6hdr *)lyr3h;
305		i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
306		addr = ipvlan_ht_addr_lookup(port, i6addr, true);
307	} else if (addr_type == IPVL_ICMPV6) {
308		struct nd_msg *ndmh;
309		struct in6_addr *i6addr;
310
311		/* Make sure that the NeighborSolicitation ICMPv6 packets
312		 * are handled to avoid DAD issue.
313		 */
314		ndmh = (struct nd_msg *)lyr3h;
315		if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
316			i6addr = &ndmh->target;
317			addr = ipvlan_ht_addr_lookup(port, i6addr, true);
318		}
319	} else if (addr_type == IPVL_IPV4) {
320		struct iphdr *ip4h;
321		__be32 *i4addr;
322
323		ip4h = (struct iphdr *)lyr3h;
324		i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
325		addr = ipvlan_ht_addr_lookup(port, i4addr, false);
326	} else if (addr_type == IPVL_ARP) {
327		struct arphdr *arph;
328		unsigned char *arp_ptr;
329		__be32 dip;
330
331		arph = (struct arphdr *)lyr3h;
332		arp_ptr = (unsigned char *)(arph + 1);
333		if (use_dest)
334			arp_ptr += (2 * port->dev->addr_len) + 4;
335		else
336			arp_ptr += port->dev->addr_len;
337
338		memcpy(&dip, arp_ptr, 4);
339		addr = ipvlan_ht_addr_lookup(port, &dip, false);
340	}
341
342	return addr;
343}
344
345static int ipvlan_process_v4_outbound(struct sk_buff *skb)
346{
347	const struct iphdr *ip4h = ip_hdr(skb);
348	struct net_device *dev = skb->dev;
349	struct net *net = dev_net(dev);
350	struct rtable *rt;
351	int err, ret = NET_XMIT_DROP;
352	struct flowi4 fl4 = {
353		.flowi4_oif = dev->ifindex,
354		.flowi4_tos = RT_TOS(ip4h->tos),
355		.flowi4_flags = FLOWI_FLAG_ANYSRC,
356		.daddr = ip4h->daddr,
357		.saddr = ip4h->saddr,
358	};
359
360	rt = ip_route_output_flow(net, &fl4, NULL);
361	if (IS_ERR(rt))
362		goto err;
363
364	if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
365		ip_rt_put(rt);
366		goto err;
367	}
368	skb_dst_drop(skb);
369	skb_dst_set(skb, &rt->dst);
370	err = ip_local_out(net, skb->sk, skb);
371	if (unlikely(net_xmit_eval(err)))
372		dev->stats.tx_errors++;
373	else
374		ret = NET_XMIT_SUCCESS;
375	goto out;
376err:
377	dev->stats.tx_errors++;
378	kfree_skb(skb);
379out:
380	return ret;
381}
382
383static int ipvlan_process_v6_outbound(struct sk_buff *skb)
384{
385	const struct ipv6hdr *ip6h = ipv6_hdr(skb);
386	struct net_device *dev = skb->dev;
387	struct net *net = dev_net(dev);
388	struct dst_entry *dst;
389	int err, ret = NET_XMIT_DROP;
390	struct flowi6 fl6 = {
391		.flowi6_iif = dev->ifindex,
392		.daddr = ip6h->daddr,
393		.saddr = ip6h->saddr,
394		.flowi6_flags = FLOWI_FLAG_ANYSRC,
395		.flowlabel = ip6_flowinfo(ip6h),
396		.flowi6_mark = skb->mark,
397		.flowi6_proto = ip6h->nexthdr,
398	};
399
400	dst = ip6_route_output(net, NULL, &fl6);
401	if (dst->error) {
402		ret = dst->error;
403		dst_release(dst);
404		goto err;
405	}
406	skb_dst_drop(skb);
407	skb_dst_set(skb, dst);
408	err = ip6_local_out(net, skb->sk, skb);
409	if (unlikely(net_xmit_eval(err)))
410		dev->stats.tx_errors++;
411	else
412		ret = NET_XMIT_SUCCESS;
413	goto out;
414err:
415	dev->stats.tx_errors++;
416	kfree_skb(skb);
417out:
418	return ret;
419}
420
421static int ipvlan_process_outbound(struct sk_buff *skb,
422				   const struct ipvl_dev *ipvlan)
423{
424	struct ethhdr *ethh = eth_hdr(skb);
425	int ret = NET_XMIT_DROP;
426
427	/* In this mode we dont care about multicast and broadcast traffic */
428	if (is_multicast_ether_addr(ethh->h_dest)) {
429		pr_warn_ratelimited("Dropped {multi|broad}cast of type= [%x]\n",
430				    ntohs(skb->protocol));
431		kfree_skb(skb);
432		goto out;
433	}
434
435	/* The ipvlan is a pseudo-L2 device, so the packets that we receive
436	 * will have L2; which need to discarded and processed further
437	 * in the net-ns of the main-device.
438	 */
439	if (skb_mac_header_was_set(skb)) {
440		skb_pull(skb, sizeof(*ethh));
441		skb->mac_header = (typeof(skb->mac_header))~0U;
442		skb_reset_network_header(skb);
443	}
444
445	if (skb->protocol == htons(ETH_P_IPV6))
446		ret = ipvlan_process_v6_outbound(skb);
447	else if (skb->protocol == htons(ETH_P_IP))
448		ret = ipvlan_process_v4_outbound(skb);
449	else {
450		pr_warn_ratelimited("Dropped outbound packet type=%x\n",
451				    ntohs(skb->protocol));
452		kfree_skb(skb);
453	}
454out:
455	return ret;
456}
457
458static void ipvlan_multicast_enqueue(struct ipvl_port *port,
459				     struct sk_buff *skb)
460{
461	if (skb->protocol == htons(ETH_P_PAUSE)) {
462		kfree_skb(skb);
463		return;
464	}
465
466	spin_lock(&port->backlog.lock);
467	if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
468		__skb_queue_tail(&port->backlog, skb);
469		spin_unlock(&port->backlog.lock);
470		schedule_work(&port->wq);
471	} else {
472		spin_unlock(&port->backlog.lock);
473		atomic_long_inc(&skb->dev->rx_dropped);
474		kfree_skb(skb);
475	}
476}
477
478static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
479{
480	const struct ipvl_dev *ipvlan = netdev_priv(dev);
481	void *lyr3h;
482	struct ipvl_addr *addr;
483	int addr_type;
484
485	lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
486	if (!lyr3h)
487		goto out;
488
489	addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
490	if (addr)
491		return ipvlan_rcv_frame(addr, &skb, true);
492
493out:
494	skb->dev = ipvlan->phy_dev;
495	return ipvlan_process_outbound(skb, ipvlan);
496}
497
498static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
499{
500	const struct ipvl_dev *ipvlan = netdev_priv(dev);
501	struct ethhdr *eth = eth_hdr(skb);
502	struct ipvl_addr *addr;
503	void *lyr3h;
504	int addr_type;
505
506	if (ether_addr_equal(eth->h_dest, eth->h_source)) {
507		lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
508		if (lyr3h) {
509			addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
510			if (addr)
511				return ipvlan_rcv_frame(addr, &skb, true);
512		}
513		skb = skb_share_check(skb, GFP_ATOMIC);
514		if (!skb)
515			return NET_XMIT_DROP;
516
517		/* Packet definitely does not belong to any of the
518		 * virtual devices, but the dest is local. So forward
519		 * the skb for the main-dev. At the RX side we just return
520		 * RX_PASS for it to be processed further on the stack.
521		 */
522		return dev_forward_skb(ipvlan->phy_dev, skb);
523
524	} else if (is_multicast_ether_addr(eth->h_dest)) {
525		ipvlan_multicast_enqueue(ipvlan->port, skb);
526		return NET_XMIT_SUCCESS;
527	}
528
529	skb->dev = ipvlan->phy_dev;
530	return dev_queue_xmit(skb);
531}
532
533int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
534{
535	struct ipvl_dev *ipvlan = netdev_priv(dev);
536	struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
537
538	if (!port)
539		goto out;
540
541	if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
542		goto out;
543
544	switch(port->mode) {
545	case IPVLAN_MODE_L2:
546		return ipvlan_xmit_mode_l2(skb, dev);
547	case IPVLAN_MODE_L3:
548		return ipvlan_xmit_mode_l3(skb, dev);
549	}
550
551	/* Should not reach here */
552	WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
553			  port->mode);
554out:
555	kfree_skb(skb);
556	return NET_XMIT_DROP;
557}
558
559static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
560{
561	struct ethhdr *eth = eth_hdr(skb);
562	struct ipvl_addr *addr;
563	void *lyr3h;
564	int addr_type;
565
566	if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
567		lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
568		if (!lyr3h)
569			return true;
570
571		addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
572		if (addr)
573			return false;
574	}
575
576	return true;
577}
578
579static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
580						 struct ipvl_port *port)
581{
582	void *lyr3h;
583	int addr_type;
584	struct ipvl_addr *addr;
585	struct sk_buff *skb = *pskb;
586	rx_handler_result_t ret = RX_HANDLER_PASS;
587
588	lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
589	if (!lyr3h)
590		goto out;
591
592	addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
593	if (addr)
594		ret = ipvlan_rcv_frame(addr, pskb, false);
595
596out:
597	return ret;
598}
599
600static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
601						 struct ipvl_port *port)
602{
603	struct sk_buff *skb = *pskb;
604	struct ethhdr *eth = eth_hdr(skb);
605	rx_handler_result_t ret = RX_HANDLER_PASS;
606	void *lyr3h;
607	int addr_type;
608
609	if (is_multicast_ether_addr(eth->h_dest)) {
610		if (ipvlan_external_frame(skb, port)) {
611			struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
612
613			/* External frames are queued for device local
614			 * distribution, but a copy is given to master
615			 * straight away to avoid sending duplicates later
616			 * when work-queue processes this frame. This is
617			 * achieved by returning RX_HANDLER_PASS.
618			 */
619			if (nskb)
620				ipvlan_multicast_enqueue(port, nskb);
621		}
622	} else {
623		struct ipvl_addr *addr;
624
625		lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
626		if (!lyr3h)
627			return ret;
628
629		addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
630		if (addr)
631			ret = ipvlan_rcv_frame(addr, pskb, false);
632	}
633
634	return ret;
635}
636
637rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
638{
639	struct sk_buff *skb = *pskb;
640	struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
641
642	if (!port)
643		return RX_HANDLER_PASS;
644
645	switch (port->mode) {
646	case IPVLAN_MODE_L2:
647		return ipvlan_handle_mode_l2(pskb, port);
648	case IPVLAN_MODE_L3:
649		return ipvlan_handle_mode_l3(pskb, port);
650	}
651
652	/* Should not reach here */
653	WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
654			  port->mode);
655	kfree_skb(skb);
656	return RX_HANDLER_CONSUMED;
657}
658