xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_core.c (revision d5e7cafd)
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 
12 static u32 ipvlan_jhash_secret __read_mostly;
13 
14 void ipvlan_init_secret(void)
15 {
16 	net_get_random_once(&ipvlan_jhash_secret, sizeof(ipvlan_jhash_secret));
17 }
18 
19 static 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 
40 static 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 
48 static 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 
56 struct 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 
76 void 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 	hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
85 }
86 
87 void ipvlan_ht_addr_del(struct ipvl_addr *addr, bool sync)
88 {
89 	hlist_del_rcu(&addr->hlnode);
90 	if (sync)
91 		synchronize_rcu();
92 }
93 
94 bool ipvlan_addr_busy(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
95 {
96 	struct ipvl_port *port = ipvlan->port;
97 	struct ipvl_addr *addr;
98 
99 	list_for_each_entry(addr, &ipvlan->addrs, anode) {
100 		if ((is_v6 && addr->atype == IPVL_IPV6 &&
101 		    ipv6_addr_equal(&addr->ip6addr, iaddr)) ||
102 		    (!is_v6 && addr->atype == IPVL_IPV4 &&
103 		    addr->ip4addr.s_addr == ((struct in_addr *)iaddr)->s_addr))
104 			return true;
105 	}
106 
107 	if (ipvlan_ht_addr_lookup(port, iaddr, is_v6))
108 		return true;
109 
110 	return false;
111 }
112 
113 static void *ipvlan_get_L3_hdr(struct sk_buff *skb, int *type)
114 {
115 	void *lyr3h = NULL;
116 
117 	switch (skb->protocol) {
118 	case htons(ETH_P_ARP): {
119 		struct arphdr *arph;
120 
121 		if (unlikely(!pskb_may_pull(skb, sizeof(*arph))))
122 			return NULL;
123 
124 		arph = arp_hdr(skb);
125 		*type = IPVL_ARP;
126 		lyr3h = arph;
127 		break;
128 	}
129 	case htons(ETH_P_IP): {
130 		u32 pktlen;
131 		struct iphdr *ip4h;
132 
133 		if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
134 			return NULL;
135 
136 		ip4h = ip_hdr(skb);
137 		pktlen = ntohs(ip4h->tot_len);
138 		if (ip4h->ihl < 5 || ip4h->version != 4)
139 			return NULL;
140 		if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
141 			return NULL;
142 
143 		*type = IPVL_IPV4;
144 		lyr3h = ip4h;
145 		break;
146 	}
147 	case htons(ETH_P_IPV6): {
148 		struct ipv6hdr *ip6h;
149 
150 		if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
151 			return NULL;
152 
153 		ip6h = ipv6_hdr(skb);
154 		if (ip6h->version != 6)
155 			return NULL;
156 
157 		*type = IPVL_IPV6;
158 		lyr3h = ip6h;
159 		/* Only Neighbour Solicitation pkts need different treatment */
160 		if (ipv6_addr_any(&ip6h->saddr) &&
161 		    ip6h->nexthdr == NEXTHDR_ICMP) {
162 			*type = IPVL_ICMPV6;
163 			lyr3h = ip6h + 1;
164 		}
165 		break;
166 	}
167 	default:
168 		return NULL;
169 	}
170 
171 	return lyr3h;
172 }
173 
174 unsigned int ipvlan_mac_hash(const unsigned char *addr)
175 {
176 	u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
177 			       ipvlan_jhash_secret);
178 
179 	return hash & IPVLAN_MAC_FILTER_MASK;
180 }
181 
182 static void ipvlan_multicast_frame(struct ipvl_port *port, struct sk_buff *skb,
183 				   const struct ipvl_dev *in_dev, bool local)
184 {
185 	struct ethhdr *eth = eth_hdr(skb);
186 	struct ipvl_dev *ipvlan;
187 	struct sk_buff *nskb;
188 	unsigned int len;
189 	unsigned int mac_hash;
190 	int ret;
191 
192 	if (skb->protocol == htons(ETH_P_PAUSE))
193 		return;
194 
195 	list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
196 		if (local && (ipvlan == in_dev))
197 			continue;
198 
199 		mac_hash = ipvlan_mac_hash(eth->h_dest);
200 		if (!test_bit(mac_hash, ipvlan->mac_filters))
201 			continue;
202 
203 		ret = NET_RX_DROP;
204 		len = skb->len + ETH_HLEN;
205 		nskb = skb_clone(skb, GFP_ATOMIC);
206 		if (!nskb)
207 			goto mcast_acct;
208 
209 		if (ether_addr_equal(eth->h_dest, ipvlan->phy_dev->broadcast))
210 			nskb->pkt_type = PACKET_BROADCAST;
211 		else
212 			nskb->pkt_type = PACKET_MULTICAST;
213 
214 		nskb->dev = ipvlan->dev;
215 		if (local)
216 			ret = dev_forward_skb(ipvlan->dev, nskb);
217 		else
218 			ret = netif_rx(nskb);
219 mcast_acct:
220 		ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
221 	}
222 
223 	/* Locally generated? ...Forward a copy to the main-device as
224 	 * well. On the RX side we'll ignore it (wont give it to any
225 	 * of the virtual devices.
226 	 */
227 	if (local) {
228 		nskb = skb_clone(skb, GFP_ATOMIC);
229 		if (nskb) {
230 			if (ether_addr_equal(eth->h_dest, port->dev->broadcast))
231 				nskb->pkt_type = PACKET_BROADCAST;
232 			else
233 				nskb->pkt_type = PACKET_MULTICAST;
234 
235 			dev_forward_skb(port->dev, nskb);
236 		}
237 	}
238 }
239 
240 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff *skb,
241 			    bool local)
242 {
243 	struct ipvl_dev *ipvlan = addr->master;
244 	struct net_device *dev = ipvlan->dev;
245 	unsigned int len;
246 	rx_handler_result_t ret = RX_HANDLER_CONSUMED;
247 	bool success = false;
248 
249 	len = skb->len + ETH_HLEN;
250 	if (unlikely(!(dev->flags & IFF_UP))) {
251 		kfree_skb(skb);
252 		goto out;
253 	}
254 
255 	skb = skb_share_check(skb, GFP_ATOMIC);
256 	if (!skb)
257 		goto out;
258 
259 	skb->dev = dev;
260 	skb->pkt_type = PACKET_HOST;
261 
262 	if (local) {
263 		if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
264 			success = true;
265 	} else {
266 		ret = RX_HANDLER_ANOTHER;
267 		success = true;
268 	}
269 
270 out:
271 	ipvlan_count_rx(ipvlan, len, success, false);
272 	return ret;
273 }
274 
275 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
276 					    void *lyr3h, int addr_type,
277 					    bool use_dest)
278 {
279 	struct ipvl_addr *addr = NULL;
280 
281 	if (addr_type == IPVL_IPV6) {
282 		struct ipv6hdr *ip6h;
283 		struct in6_addr *i6addr;
284 
285 		ip6h = (struct ipv6hdr *)lyr3h;
286 		i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
287 		addr = ipvlan_ht_addr_lookup(port, i6addr, true);
288 	} else if (addr_type == IPVL_ICMPV6) {
289 		struct nd_msg *ndmh;
290 		struct in6_addr *i6addr;
291 
292 		/* Make sure that the NeighborSolicitation ICMPv6 packets
293 		 * are handled to avoid DAD issue.
294 		 */
295 		ndmh = (struct nd_msg *)lyr3h;
296 		if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
297 			i6addr = &ndmh->target;
298 			addr = ipvlan_ht_addr_lookup(port, i6addr, true);
299 		}
300 	} else if (addr_type == IPVL_IPV4) {
301 		struct iphdr *ip4h;
302 		__be32 *i4addr;
303 
304 		ip4h = (struct iphdr *)lyr3h;
305 		i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
306 		addr = ipvlan_ht_addr_lookup(port, i4addr, false);
307 	} else if (addr_type == IPVL_ARP) {
308 		struct arphdr *arph;
309 		unsigned char *arp_ptr;
310 		__be32 dip;
311 
312 		arph = (struct arphdr *)lyr3h;
313 		arp_ptr = (unsigned char *)(arph + 1);
314 		if (use_dest)
315 			arp_ptr += (2 * port->dev->addr_len) + 4;
316 		else
317 			arp_ptr += port->dev->addr_len;
318 
319 		memcpy(&dip, arp_ptr, 4);
320 		addr = ipvlan_ht_addr_lookup(port, &dip, false);
321 	}
322 
323 	return addr;
324 }
325 
326 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
327 {
328 	const struct iphdr *ip4h = ip_hdr(skb);
329 	struct net_device *dev = skb->dev;
330 	struct rtable *rt;
331 	int err, ret = NET_XMIT_DROP;
332 	struct flowi4 fl4 = {
333 		.flowi4_oif = dev->iflink,
334 		.flowi4_tos = RT_TOS(ip4h->tos),
335 		.flowi4_flags = FLOWI_FLAG_ANYSRC,
336 		.daddr = ip4h->daddr,
337 		.saddr = ip4h->saddr,
338 	};
339 
340 	rt = ip_route_output_flow(dev_net(dev), &fl4, NULL);
341 	if (IS_ERR(rt))
342 		goto err;
343 
344 	if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
345 		ip_rt_put(rt);
346 		goto err;
347 	}
348 	skb_dst_drop(skb);
349 	skb_dst_set(skb, &rt->dst);
350 	err = ip_local_out(skb);
351 	if (unlikely(net_xmit_eval(err)))
352 		dev->stats.tx_errors++;
353 	else
354 		ret = NET_XMIT_SUCCESS;
355 	goto out;
356 err:
357 	dev->stats.tx_errors++;
358 	kfree_skb(skb);
359 out:
360 	return ret;
361 }
362 
363 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
364 {
365 	const struct ipv6hdr *ip6h = ipv6_hdr(skb);
366 	struct net_device *dev = skb->dev;
367 	struct dst_entry *dst;
368 	int err, ret = NET_XMIT_DROP;
369 	struct flowi6 fl6 = {
370 		.flowi6_iif = skb->dev->ifindex,
371 		.daddr = ip6h->daddr,
372 		.saddr = ip6h->saddr,
373 		.flowi6_flags = FLOWI_FLAG_ANYSRC,
374 		.flowlabel = ip6_flowinfo(ip6h),
375 		.flowi6_mark = skb->mark,
376 		.flowi6_proto = ip6h->nexthdr,
377 	};
378 
379 	dst = ip6_route_output(dev_net(dev), NULL, &fl6);
380 	if (dst->error) {
381 		ret = dst->error;
382 		dst_release(dst);
383 		goto err;
384 	}
385 	skb_dst_drop(skb);
386 	skb_dst_set(skb, dst);
387 	err = ip6_local_out(skb);
388 	if (unlikely(net_xmit_eval(err)))
389 		dev->stats.tx_errors++;
390 	else
391 		ret = NET_XMIT_SUCCESS;
392 	goto out;
393 err:
394 	dev->stats.tx_errors++;
395 	kfree_skb(skb);
396 out:
397 	return ret;
398 }
399 
400 static int ipvlan_process_outbound(struct sk_buff *skb,
401 				   const struct ipvl_dev *ipvlan)
402 {
403 	struct ethhdr *ethh = eth_hdr(skb);
404 	int ret = NET_XMIT_DROP;
405 
406 	/* In this mode we dont care about multicast and broadcast traffic */
407 	if (is_multicast_ether_addr(ethh->h_dest)) {
408 		pr_warn_ratelimited("Dropped {multi|broad}cast of type= [%x]\n",
409 				    ntohs(skb->protocol));
410 		kfree_skb(skb);
411 		goto out;
412 	}
413 
414 	/* The ipvlan is a pseudo-L2 device, so the packets that we receive
415 	 * will have L2; which need to discarded and processed further
416 	 * in the net-ns of the main-device.
417 	 */
418 	if (skb_mac_header_was_set(skb)) {
419 		skb_pull(skb, sizeof(*ethh));
420 		skb->mac_header = (typeof(skb->mac_header))~0U;
421 		skb_reset_network_header(skb);
422 	}
423 
424 	if (skb->protocol == htons(ETH_P_IPV6))
425 		ret = ipvlan_process_v6_outbound(skb);
426 	else if (skb->protocol == htons(ETH_P_IP))
427 		ret = ipvlan_process_v4_outbound(skb);
428 	else {
429 		pr_warn_ratelimited("Dropped outbound packet type=%x\n",
430 				    ntohs(skb->protocol));
431 		kfree_skb(skb);
432 	}
433 out:
434 	return ret;
435 }
436 
437 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
438 {
439 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
440 	void *lyr3h;
441 	struct ipvl_addr *addr;
442 	int addr_type;
443 
444 	lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
445 	if (!lyr3h)
446 		goto out;
447 
448 	addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
449 	if (addr)
450 		return ipvlan_rcv_frame(addr, skb, true);
451 
452 out:
453 	skb->dev = ipvlan->phy_dev;
454 	return ipvlan_process_outbound(skb, ipvlan);
455 }
456 
457 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
458 {
459 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
460 	struct ethhdr *eth = eth_hdr(skb);
461 	struct ipvl_addr *addr;
462 	void *lyr3h;
463 	int addr_type;
464 
465 	if (ether_addr_equal(eth->h_dest, eth->h_source)) {
466 		lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
467 		if (lyr3h) {
468 			addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
469 			if (addr)
470 				return ipvlan_rcv_frame(addr, skb, true);
471 		}
472 		skb = skb_share_check(skb, GFP_ATOMIC);
473 		if (!skb)
474 			return NET_XMIT_DROP;
475 
476 		/* Packet definitely does not belong to any of the
477 		 * virtual devices, but the dest is local. So forward
478 		 * the skb for the main-dev. At the RX side we just return
479 		 * RX_PASS for it to be processed further on the stack.
480 		 */
481 		return dev_forward_skb(ipvlan->phy_dev, skb);
482 
483 	} else if (is_multicast_ether_addr(eth->h_dest)) {
484 		u8 ip_summed = skb->ip_summed;
485 
486 		skb->ip_summed = CHECKSUM_UNNECESSARY;
487 		ipvlan_multicast_frame(ipvlan->port, skb, ipvlan, true);
488 		skb->ip_summed = ip_summed;
489 	}
490 
491 	skb->dev = ipvlan->phy_dev;
492 	return dev_queue_xmit(skb);
493 }
494 
495 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
496 {
497 	struct ipvl_dev *ipvlan = netdev_priv(dev);
498 	struct ipvl_port *port = ipvlan_port_get_rcu(ipvlan->phy_dev);
499 
500 	if (!port)
501 		goto out;
502 
503 	if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
504 		goto out;
505 
506 	switch(port->mode) {
507 	case IPVLAN_MODE_L2:
508 		return ipvlan_xmit_mode_l2(skb, dev);
509 	case IPVLAN_MODE_L3:
510 		return ipvlan_xmit_mode_l3(skb, dev);
511 	}
512 
513 	/* Should not reach here */
514 	WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
515 			  port->mode);
516 out:
517 	kfree_skb(skb);
518 	return NET_XMIT_DROP;
519 }
520 
521 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
522 {
523 	struct ethhdr *eth = eth_hdr(skb);
524 	struct ipvl_addr *addr;
525 	void *lyr3h;
526 	int addr_type;
527 
528 	if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
529 		lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
530 		if (!lyr3h)
531 			return true;
532 
533 		addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
534 		if (addr)
535 			return false;
536 	}
537 
538 	return true;
539 }
540 
541 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
542 						 struct ipvl_port *port)
543 {
544 	void *lyr3h;
545 	int addr_type;
546 	struct ipvl_addr *addr;
547 	struct sk_buff *skb = *pskb;
548 	rx_handler_result_t ret = RX_HANDLER_PASS;
549 
550 	lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
551 	if (!lyr3h)
552 		goto out;
553 
554 	addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
555 	if (addr)
556 		ret = ipvlan_rcv_frame(addr, skb, false);
557 
558 out:
559 	return ret;
560 }
561 
562 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
563 						 struct ipvl_port *port)
564 {
565 	struct sk_buff *skb = *pskb;
566 	struct ethhdr *eth = eth_hdr(skb);
567 	rx_handler_result_t ret = RX_HANDLER_PASS;
568 	void *lyr3h;
569 	int addr_type;
570 
571 	if (is_multicast_ether_addr(eth->h_dest)) {
572 		if (ipvlan_external_frame(skb, port))
573 			ipvlan_multicast_frame(port, skb, NULL, false);
574 	} else {
575 		struct ipvl_addr *addr;
576 
577 		lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
578 		if (!lyr3h)
579 			return ret;
580 
581 		addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
582 		if (addr)
583 			ret = ipvlan_rcv_frame(addr, skb, false);
584 	}
585 
586 	return ret;
587 }
588 
589 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
590 {
591 	struct sk_buff *skb = *pskb;
592 	struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
593 
594 	if (!port)
595 		return RX_HANDLER_PASS;
596 
597 	switch (port->mode) {
598 	case IPVLAN_MODE_L2:
599 		return ipvlan_handle_mode_l2(pskb, port);
600 	case IPVLAN_MODE_L3:
601 		return ipvlan_handle_mode_l3(pskb, port);
602 	}
603 
604 	/* Should not reach here */
605 	WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
606 			  port->mode);
607 	kfree_skb(skb);
608 	return NET_RX_DROP;
609 }
610