1 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
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.
12 static u32 ipvlan_jhash_secret __read_mostly;
14 void ipvlan_init_secret(void)
16 net_get_random_once(&ipvlan_jhash_secret, sizeof(ipvlan_jhash_secret));
19 static void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
20 unsigned int len, bool success, bool mcast)
25 if (likely(success)) {
26 struct ipvl_pcpu_stats *pcptr;
28 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
29 u64_stats_update_begin(&pcptr->syncp);
31 pcptr->rx_bytes += len;
34 u64_stats_update_end(&pcptr->syncp);
36 this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
40 static u8 ipvlan_get_v6_hash(const void *iaddr)
42 const struct in6_addr *ip6_addr = iaddr;
44 return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
48 static u8 ipvlan_get_v4_hash(const void *iaddr)
50 const struct in_addr *ip4_addr = iaddr;
52 return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
56 struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
57 const void *iaddr, bool is_v6)
59 struct ipvl_addr *addr;
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))
68 else if (!is_v6 && addr->atype == IPVL_IPV4 &&
69 addr->ip4addr.s_addr ==
70 ((struct in_addr *)iaddr)->s_addr)
76 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
78 struct ipvl_port *port = ipvlan->port;
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]);
88 void ipvlan_ht_addr_del(struct ipvl_addr *addr, bool sync)
90 hlist_del_init_rcu(&addr->hlnode);
95 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
96 const void *iaddr, bool is_v6)
98 struct ipvl_addr *addr;
100 list_for_each_entry(addr, &ipvlan->addrs, anode) {
101 if ((is_v6 && addr->atype == IPVL_IPV6 &&
102 ipv6_addr_equal(&addr->ip6addr, iaddr)) ||
103 (!is_v6 && addr->atype == IPVL_IPV4 &&
104 addr->ip4addr.s_addr == ((struct in_addr *)iaddr)->s_addr))
110 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
112 struct ipvl_dev *ipvlan;
116 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
117 if (ipvlan_find_addr(ipvlan, iaddr, is_v6))
123 static void *ipvlan_get_L3_hdr(struct sk_buff *skb, int *type)
127 switch (skb->protocol) {
128 case htons(ETH_P_ARP): {
131 if (unlikely(!pskb_may_pull(skb, sizeof(*arph))))
139 case htons(ETH_P_IP): {
143 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
147 pktlen = ntohs(ip4h->tot_len);
148 if (ip4h->ihl < 5 || ip4h->version != 4)
150 if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
157 case htons(ETH_P_IPV6): {
158 struct ipv6hdr *ip6h;
160 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
163 ip6h = ipv6_hdr(skb);
164 if (ip6h->version != 6)
169 /* Only Neighbour Solicitation pkts need different treatment */
170 if (ipv6_addr_any(&ip6h->saddr) &&
171 ip6h->nexthdr == NEXTHDR_ICMP) {
184 unsigned int ipvlan_mac_hash(const unsigned char *addr)
186 u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
187 ipvlan_jhash_secret);
189 return hash & IPVLAN_MAC_FILTER_MASK;
192 void ipvlan_process_multicast(struct work_struct *work)
194 struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
196 struct ipvl_dev *ipvlan;
197 struct sk_buff *skb, *nskb;
198 struct sk_buff_head list;
200 unsigned int mac_hash;
205 __skb_queue_head_init(&list);
207 spin_lock_bh(&port->backlog.lock);
208 skb_queue_splice_tail_init(&port->backlog, &list);
209 spin_unlock_bh(&port->backlog.lock);
211 while ((skb = __skb_dequeue(&list)) != NULL) {
213 hlocal = ether_addr_equal(ethh->h_source, port->dev->dev_addr);
214 mac_hash = ipvlan_mac_hash(ethh->h_dest);
216 if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
217 pkt_type = PACKET_BROADCAST;
219 pkt_type = PACKET_MULTICAST;
223 list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
224 if (hlocal && (ipvlan->dev == skb->dev)) {
228 if (!test_bit(mac_hash, ipvlan->mac_filters))
232 len = skb->len + ETH_HLEN;
233 nskb = skb_clone(skb, GFP_ATOMIC);
237 nskb->pkt_type = pkt_type;
238 nskb->dev = ipvlan->dev;
240 ret = dev_forward_skb(ipvlan->dev, nskb);
242 ret = netif_rx(nskb);
244 ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
249 /* If the packet originated here, send it out. */
250 skb->dev = port->dev;
251 skb->pkt_type = pkt_type;
259 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff *skb,
262 struct ipvl_dev *ipvlan = addr->master;
263 struct net_device *dev = ipvlan->dev;
265 rx_handler_result_t ret = RX_HANDLER_CONSUMED;
266 bool success = false;
268 len = skb->len + ETH_HLEN;
269 if (unlikely(!(dev->flags & IFF_UP))) {
274 skb = skb_share_check(skb, GFP_ATOMIC);
279 skb->pkt_type = PACKET_HOST;
282 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
285 ret = RX_HANDLER_ANOTHER;
290 ipvlan_count_rx(ipvlan, len, success, false);
294 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
295 void *lyr3h, int addr_type,
298 struct ipvl_addr *addr = NULL;
300 if (addr_type == IPVL_IPV6) {
301 struct ipv6hdr *ip6h;
302 struct in6_addr *i6addr;
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) {
309 struct in6_addr *i6addr;
311 /* Make sure that the NeighborSolicitation ICMPv6 packets
312 * are handled to avoid DAD issue.
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);
319 } else if (addr_type == IPVL_IPV4) {
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) {
328 unsigned char *arp_ptr;
331 arph = (struct arphdr *)lyr3h;
332 arp_ptr = (unsigned char *)(arph + 1);
334 arp_ptr += (2 * port->dev->addr_len) + 4;
336 arp_ptr += port->dev->addr_len;
338 memcpy(&dip, arp_ptr, 4);
339 addr = ipvlan_ht_addr_lookup(port, &dip, false);
345 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
347 const struct iphdr *ip4h = ip_hdr(skb);
348 struct net_device *dev = skb->dev;
350 int err, ret = NET_XMIT_DROP;
351 struct flowi4 fl4 = {
352 .flowi4_oif = dev_get_iflink(dev),
353 .flowi4_tos = RT_TOS(ip4h->tos),
354 .flowi4_flags = FLOWI_FLAG_ANYSRC,
355 .daddr = ip4h->daddr,
356 .saddr = ip4h->saddr,
359 rt = ip_route_output_flow(dev_net(dev), &fl4, NULL);
363 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
368 skb_dst_set(skb, &rt->dst);
369 err = ip_local_out(skb);
370 if (unlikely(net_xmit_eval(err)))
371 dev->stats.tx_errors++;
373 ret = NET_XMIT_SUCCESS;
376 dev->stats.tx_errors++;
382 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
384 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
385 struct net_device *dev = skb->dev;
386 struct dst_entry *dst;
387 int err, ret = NET_XMIT_DROP;
388 struct flowi6 fl6 = {
389 .flowi6_iif = skb->dev->ifindex,
390 .daddr = ip6h->daddr,
391 .saddr = ip6h->saddr,
392 .flowi6_flags = FLOWI_FLAG_ANYSRC,
393 .flowlabel = ip6_flowinfo(ip6h),
394 .flowi6_mark = skb->mark,
395 .flowi6_proto = ip6h->nexthdr,
398 dst = ip6_route_output(dev_net(dev), NULL, &fl6);
405 skb_dst_set(skb, dst);
406 err = ip6_local_out(skb);
407 if (unlikely(net_xmit_eval(err)))
408 dev->stats.tx_errors++;
410 ret = NET_XMIT_SUCCESS;
413 dev->stats.tx_errors++;
419 static int ipvlan_process_outbound(struct sk_buff *skb,
420 const struct ipvl_dev *ipvlan)
422 struct ethhdr *ethh = eth_hdr(skb);
423 int ret = NET_XMIT_DROP;
425 /* In this mode we dont care about multicast and broadcast traffic */
426 if (is_multicast_ether_addr(ethh->h_dest)) {
427 pr_warn_ratelimited("Dropped {multi|broad}cast of type= [%x]\n",
428 ntohs(skb->protocol));
433 /* The ipvlan is a pseudo-L2 device, so the packets that we receive
434 * will have L2; which need to discarded and processed further
435 * in the net-ns of the main-device.
437 if (skb_mac_header_was_set(skb)) {
438 skb_pull(skb, sizeof(*ethh));
439 skb->mac_header = (typeof(skb->mac_header))~0U;
440 skb_reset_network_header(skb);
443 if (skb->protocol == htons(ETH_P_IPV6))
444 ret = ipvlan_process_v6_outbound(skb);
445 else if (skb->protocol == htons(ETH_P_IP))
446 ret = ipvlan_process_v4_outbound(skb);
448 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
449 ntohs(skb->protocol));
456 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
459 if (skb->protocol == htons(ETH_P_PAUSE)) {
464 spin_lock(&port->backlog.lock);
465 if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
466 __skb_queue_tail(&port->backlog, skb);
467 spin_unlock(&port->backlog.lock);
468 schedule_work(&port->wq);
470 spin_unlock(&port->backlog.lock);
471 atomic_long_inc(&skb->dev->rx_dropped);
476 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
478 const struct ipvl_dev *ipvlan = netdev_priv(dev);
480 struct ipvl_addr *addr;
483 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
487 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
489 return ipvlan_rcv_frame(addr, skb, true);
492 skb->dev = ipvlan->phy_dev;
493 return ipvlan_process_outbound(skb, ipvlan);
496 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
498 const struct ipvl_dev *ipvlan = netdev_priv(dev);
499 struct ethhdr *eth = eth_hdr(skb);
500 struct ipvl_addr *addr;
504 if (ether_addr_equal(eth->h_dest, eth->h_source)) {
505 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
507 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
509 return ipvlan_rcv_frame(addr, skb, true);
511 skb = skb_share_check(skb, GFP_ATOMIC);
513 return NET_XMIT_DROP;
515 /* Packet definitely does not belong to any of the
516 * virtual devices, but the dest is local. So forward
517 * the skb for the main-dev. At the RX side we just return
518 * RX_PASS for it to be processed further on the stack.
520 return dev_forward_skb(ipvlan->phy_dev, skb);
522 } else if (is_multicast_ether_addr(eth->h_dest)) {
523 ipvlan_multicast_enqueue(ipvlan->port, skb);
524 return NET_XMIT_SUCCESS;
527 skb->dev = ipvlan->phy_dev;
528 return dev_queue_xmit(skb);
531 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
533 struct ipvl_dev *ipvlan = netdev_priv(dev);
534 struct ipvl_port *port = ipvlan_port_get_rcu(ipvlan->phy_dev);
539 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
544 return ipvlan_xmit_mode_l2(skb, dev);
546 return ipvlan_xmit_mode_l3(skb, dev);
549 /* Should not reach here */
550 WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
554 return NET_XMIT_DROP;
557 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
559 struct ethhdr *eth = eth_hdr(skb);
560 struct ipvl_addr *addr;
564 if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
565 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
569 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
577 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
578 struct ipvl_port *port)
582 struct ipvl_addr *addr;
583 struct sk_buff *skb = *pskb;
584 rx_handler_result_t ret = RX_HANDLER_PASS;
586 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
590 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
592 ret = ipvlan_rcv_frame(addr, skb, false);
598 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
599 struct ipvl_port *port)
601 struct sk_buff *skb = *pskb;
602 struct ethhdr *eth = eth_hdr(skb);
603 rx_handler_result_t ret = RX_HANDLER_PASS;
607 if (is_multicast_ether_addr(eth->h_dest)) {
608 if (ipvlan_external_frame(skb, port)) {
609 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
611 /* External frames are queued for device local
612 * distribution, but a copy is given to master
613 * straight away to avoid sending duplicates later
614 * when work-queue processes this frame. This is
615 * achieved by returning RX_HANDLER_PASS.
618 ipvlan_multicast_enqueue(port, nskb);
621 struct ipvl_addr *addr;
623 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
627 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
629 ret = ipvlan_rcv_frame(addr, skb, false);
635 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
637 struct sk_buff *skb = *pskb;
638 struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
641 return RX_HANDLER_PASS;
643 switch (port->mode) {
645 return ipvlan_handle_mode_l2(pskb, port);
647 return ipvlan_handle_mode_l3(pskb, port);
650 /* Should not reach here */
651 WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",