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 void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
20 unsigned int len, bool success, bool mcast)
22 if (likely(success)) {
23 struct ipvl_pcpu_stats *pcptr;
25 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
26 u64_stats_update_begin(&pcptr->syncp);
28 pcptr->rx_bytes += len;
31 u64_stats_update_end(&pcptr->syncp);
33 this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
36 EXPORT_SYMBOL_GPL(ipvlan_count_rx);
38 static u8 ipvlan_get_v6_hash(const void *iaddr)
40 const struct in6_addr *ip6_addr = iaddr;
42 return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
46 static u8 ipvlan_get_v4_hash(const void *iaddr)
48 const struct in_addr *ip4_addr = iaddr;
50 return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
54 static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
55 const void *iaddr, bool is_v6)
57 struct ipvl_addr *addr;
60 hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
61 ipvlan_get_v4_hash(iaddr);
62 hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode) {
63 if (is_v6 && addr->atype == IPVL_IPV6 &&
64 ipv6_addr_equal(&addr->ip6addr, iaddr))
66 else if (!is_v6 && addr->atype == IPVL_IPV4 &&
67 addr->ip4addr.s_addr ==
68 ((struct in_addr *)iaddr)->s_addr)
74 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
76 struct ipvl_port *port = ipvlan->port;
79 hash = (addr->atype == IPVL_IPV6) ?
80 ipvlan_get_v6_hash(&addr->ip6addr) :
81 ipvlan_get_v4_hash(&addr->ip4addr);
82 if (hlist_unhashed(&addr->hlnode))
83 hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
86 void ipvlan_ht_addr_del(struct ipvl_addr *addr)
88 hlist_del_init_rcu(&addr->hlnode);
91 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
92 const void *iaddr, bool is_v6)
94 struct ipvl_addr *addr;
96 list_for_each_entry(addr, &ipvlan->addrs, anode) {
97 if ((is_v6 && addr->atype == IPVL_IPV6 &&
98 ipv6_addr_equal(&addr->ip6addr, iaddr)) ||
99 (!is_v6 && addr->atype == IPVL_IPV4 &&
100 addr->ip4addr.s_addr == ((struct in_addr *)iaddr)->s_addr))
106 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
108 struct ipvl_dev *ipvlan;
112 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
113 if (ipvlan_find_addr(ipvlan, iaddr, is_v6))
119 static void *ipvlan_get_L3_hdr(struct sk_buff *skb, int *type)
123 switch (skb->protocol) {
124 case htons(ETH_P_ARP): {
127 if (unlikely(!pskb_may_pull(skb, sizeof(*arph))))
135 case htons(ETH_P_IP): {
139 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
143 pktlen = ntohs(ip4h->tot_len);
144 if (ip4h->ihl < 5 || ip4h->version != 4)
146 if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
153 case htons(ETH_P_IPV6): {
154 struct ipv6hdr *ip6h;
156 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
159 ip6h = ipv6_hdr(skb);
160 if (ip6h->version != 6)
165 /* Only Neighbour Solicitation pkts need different treatment */
166 if (ipv6_addr_any(&ip6h->saddr) &&
167 ip6h->nexthdr == NEXTHDR_ICMP) {
180 unsigned int ipvlan_mac_hash(const unsigned char *addr)
182 u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
183 ipvlan_jhash_secret);
185 return hash & IPVLAN_MAC_FILTER_MASK;
188 void ipvlan_process_multicast(struct work_struct *work)
190 struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
192 struct ipvl_dev *ipvlan;
193 struct sk_buff *skb, *nskb;
194 struct sk_buff_head list;
196 unsigned int mac_hash;
201 __skb_queue_head_init(&list);
203 spin_lock_bh(&port->backlog.lock);
204 skb_queue_splice_tail_init(&port->backlog, &list);
205 spin_unlock_bh(&port->backlog.lock);
207 while ((skb = __skb_dequeue(&list)) != NULL) {
208 struct net_device *dev = skb->dev;
209 bool consumed = false;
212 tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
213 mac_hash = ipvlan_mac_hash(ethh->h_dest);
215 if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
216 pkt_type = PACKET_BROADCAST;
218 pkt_type = PACKET_MULTICAST;
221 list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
222 if (tx_pkt && (ipvlan->dev == skb->dev))
224 if (!test_bit(mac_hash, ipvlan->mac_filters))
226 if (!(ipvlan->dev->flags & IFF_UP))
229 len = skb->len + ETH_HLEN;
230 nskb = skb_clone(skb, GFP_ATOMIC);
234 nskb->pkt_type = pkt_type;
235 nskb->dev = ipvlan->dev;
237 ret = dev_forward_skb(ipvlan->dev, nskb);
239 ret = netif_rx(nskb);
241 ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
247 /* If the packet originated here, send it out. */
248 skb->dev = port->dev;
249 skb->pkt_type = pkt_type;
262 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
267 xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
269 skb_scrub_packet(skb, xnet);
274 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
277 struct ipvl_dev *ipvlan = addr->master;
278 struct net_device *dev = ipvlan->dev;
280 rx_handler_result_t ret = RX_HANDLER_CONSUMED;
281 bool success = false;
282 struct sk_buff *skb = *pskb;
284 len = skb->len + ETH_HLEN;
285 /* Only packets exchanged between two local slaves need to have
286 * device-up check as well as skb-share check.
289 if (unlikely(!(dev->flags & IFF_UP))) {
294 skb = skb_share_check(skb, GFP_ATOMIC);
300 ipvlan_skb_crossing_ns(skb, dev);
303 skb->pkt_type = PACKET_HOST;
304 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
307 ret = RX_HANDLER_ANOTHER;
312 ipvlan_count_rx(ipvlan, len, success, false);
316 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
317 void *lyr3h, int addr_type,
320 struct ipvl_addr *addr = NULL;
322 if (addr_type == IPVL_IPV6) {
323 struct ipv6hdr *ip6h;
324 struct in6_addr *i6addr;
326 ip6h = (struct ipv6hdr *)lyr3h;
327 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
328 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
329 } else if (addr_type == IPVL_ICMPV6) {
331 struct in6_addr *i6addr;
333 /* Make sure that the NeighborSolicitation ICMPv6 packets
334 * are handled to avoid DAD issue.
336 ndmh = (struct nd_msg *)lyr3h;
337 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
338 i6addr = &ndmh->target;
339 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
341 } else if (addr_type == IPVL_IPV4) {
345 ip4h = (struct iphdr *)lyr3h;
346 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
347 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
348 } else if (addr_type == IPVL_ARP) {
350 unsigned char *arp_ptr;
353 arph = (struct arphdr *)lyr3h;
354 arp_ptr = (unsigned char *)(arph + 1);
356 arp_ptr += (2 * port->dev->addr_len) + 4;
358 arp_ptr += port->dev->addr_len;
360 memcpy(&dip, arp_ptr, 4);
361 addr = ipvlan_ht_addr_lookup(port, &dip, false);
367 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
369 const struct iphdr *ip4h = ip_hdr(skb);
370 struct net_device *dev = skb->dev;
371 struct net *net = dev_net(dev);
373 int err, ret = NET_XMIT_DROP;
374 struct flowi4 fl4 = {
375 .flowi4_oif = dev->ifindex,
376 .flowi4_tos = RT_TOS(ip4h->tos),
377 .flowi4_flags = FLOWI_FLAG_ANYSRC,
378 .daddr = ip4h->daddr,
379 .saddr = ip4h->saddr,
382 rt = ip_route_output_flow(net, &fl4, NULL);
386 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
390 skb_dst_set(skb, &rt->dst);
391 err = ip_local_out(net, skb->sk, skb);
392 if (unlikely(net_xmit_eval(err)))
393 dev->stats.tx_errors++;
395 ret = NET_XMIT_SUCCESS;
398 dev->stats.tx_errors++;
404 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
406 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
407 struct net_device *dev = skb->dev;
408 struct net *net = dev_net(dev);
409 struct dst_entry *dst;
410 int err, ret = NET_XMIT_DROP;
411 struct flowi6 fl6 = {
412 .flowi6_iif = dev->ifindex,
413 .daddr = ip6h->daddr,
414 .saddr = ip6h->saddr,
415 .flowi6_flags = FLOWI_FLAG_ANYSRC,
416 .flowlabel = ip6_flowinfo(ip6h),
417 .flowi6_mark = skb->mark,
418 .flowi6_proto = ip6h->nexthdr,
421 dst = ip6_route_output(net, NULL, &fl6);
427 skb_dst_set(skb, dst);
428 err = ip6_local_out(net, skb->sk, skb);
429 if (unlikely(net_xmit_eval(err)))
430 dev->stats.tx_errors++;
432 ret = NET_XMIT_SUCCESS;
435 dev->stats.tx_errors++;
441 static int ipvlan_process_outbound(struct sk_buff *skb)
443 struct ethhdr *ethh = eth_hdr(skb);
444 int ret = NET_XMIT_DROP;
446 /* In this mode we dont care about multicast and broadcast traffic */
447 if (is_multicast_ether_addr(ethh->h_dest)) {
448 pr_warn_ratelimited("Dropped {multi|broad}cast of type= [%x]\n",
449 ntohs(skb->protocol));
454 /* The ipvlan is a pseudo-L2 device, so the packets that we receive
455 * will have L2; which need to discarded and processed further
456 * in the net-ns of the main-device.
458 if (skb_mac_header_was_set(skb)) {
459 skb_pull(skb, sizeof(*ethh));
460 skb->mac_header = (typeof(skb->mac_header))~0U;
461 skb_reset_network_header(skb);
464 if (skb->protocol == htons(ETH_P_IPV6))
465 ret = ipvlan_process_v6_outbound(skb);
466 else if (skb->protocol == htons(ETH_P_IP))
467 ret = ipvlan_process_v4_outbound(skb);
469 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
470 ntohs(skb->protocol));
477 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
478 struct sk_buff *skb, bool tx_pkt)
480 if (skb->protocol == htons(ETH_P_PAUSE)) {
485 /* Record that the deferred packet is from TX or RX path. By
486 * looking at mac-addresses on packet will lead to erronus decisions.
487 * (This would be true for a loopback-mode on master device or a
488 * hair-pin mode of the switch.)
490 IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
492 spin_lock(&port->backlog.lock);
493 if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
496 __skb_queue_tail(&port->backlog, skb);
497 spin_unlock(&port->backlog.lock);
498 schedule_work(&port->wq);
500 spin_unlock(&port->backlog.lock);
501 atomic_long_inc(&skb->dev->rx_dropped);
506 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
508 const struct ipvl_dev *ipvlan = netdev_priv(dev);
510 struct ipvl_addr *addr;
513 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
517 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
519 return ipvlan_rcv_frame(addr, &skb, true);
522 ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
523 return ipvlan_process_outbound(skb);
526 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
528 const struct ipvl_dev *ipvlan = netdev_priv(dev);
529 struct ethhdr *eth = eth_hdr(skb);
530 struct ipvl_addr *addr;
534 if (ether_addr_equal(eth->h_dest, eth->h_source)) {
535 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
537 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
539 return ipvlan_rcv_frame(addr, &skb, true);
541 skb = skb_share_check(skb, GFP_ATOMIC);
543 return NET_XMIT_DROP;
545 /* Packet definitely does not belong to any of the
546 * virtual devices, but the dest is local. So forward
547 * the skb for the main-dev. At the RX side we just return
548 * RX_PASS for it to be processed further on the stack.
550 return dev_forward_skb(ipvlan->phy_dev, skb);
552 } else if (is_multicast_ether_addr(eth->h_dest)) {
553 ipvlan_skb_crossing_ns(skb, NULL);
554 ipvlan_multicast_enqueue(ipvlan->port, skb, true);
555 return NET_XMIT_SUCCESS;
558 ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
559 return dev_queue_xmit(skb);
562 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
564 struct ipvl_dev *ipvlan = netdev_priv(dev);
565 struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
570 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
575 return ipvlan_xmit_mode_l2(skb, dev);
577 case IPVLAN_MODE_L3S:
578 return ipvlan_xmit_mode_l3(skb, dev);
581 /* Should not reach here */
582 WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
586 return NET_XMIT_DROP;
589 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
591 struct ethhdr *eth = eth_hdr(skb);
592 struct ipvl_addr *addr;
596 if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
597 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
601 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
609 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
610 struct ipvl_port *port)
614 struct ipvl_addr *addr;
615 struct sk_buff *skb = *pskb;
616 rx_handler_result_t ret = RX_HANDLER_PASS;
618 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
622 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
624 ret = ipvlan_rcv_frame(addr, pskb, false);
630 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
631 struct ipvl_port *port)
633 struct sk_buff *skb = *pskb;
634 struct ethhdr *eth = eth_hdr(skb);
635 rx_handler_result_t ret = RX_HANDLER_PASS;
639 if (is_multicast_ether_addr(eth->h_dest)) {
640 if (ipvlan_external_frame(skb, port)) {
641 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
643 /* External frames are queued for device local
644 * distribution, but a copy is given to master
645 * straight away to avoid sending duplicates later
646 * when work-queue processes this frame. This is
647 * achieved by returning RX_HANDLER_PASS.
650 ipvlan_skb_crossing_ns(nskb, NULL);
651 ipvlan_multicast_enqueue(port, nskb, false);
655 struct ipvl_addr *addr;
657 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
661 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
663 ret = ipvlan_rcv_frame(addr, pskb, false);
669 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
671 struct sk_buff *skb = *pskb;
672 struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
675 return RX_HANDLER_PASS;
677 switch (port->mode) {
679 return ipvlan_handle_mode_l2(pskb, port);
681 return ipvlan_handle_mode_l3(pskb, port);
682 case IPVLAN_MODE_L3S:
683 return RX_HANDLER_PASS;
686 /* Should not reach here */
687 WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
690 return RX_HANDLER_CONSUMED;
693 static struct ipvl_addr *ipvlan_skb_to_addr(struct sk_buff *skb,
694 struct net_device *dev)
696 struct ipvl_addr *addr = NULL;
697 struct ipvl_port *port;
701 if (!dev || !netif_is_ipvlan_port(dev))
704 port = ipvlan_port_get_rcu(dev);
705 if (!port || port->mode != IPVLAN_MODE_L3S)
708 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
712 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
717 struct sk_buff *ipvlan_l3_rcv(struct net_device *dev, struct sk_buff *skb,
720 struct ipvl_addr *addr;
721 struct net_device *sdev;
723 addr = ipvlan_skb_to_addr(skb, dev);
727 sdev = addr->master->dev;
732 struct iphdr *ip4h = ip_hdr(skb);
734 err = ip_route_input_noref(skb, ip4h->daddr, ip4h->saddr,
742 struct dst_entry *dst;
743 struct ipv6hdr *ip6h = ipv6_hdr(skb);
744 int flags = RT6_LOOKUP_F_HAS_SADDR;
745 struct flowi6 fl6 = {
746 .flowi6_iif = sdev->ifindex,
747 .daddr = ip6h->daddr,
748 .saddr = ip6h->saddr,
749 .flowlabel = ip6_flowinfo(ip6h),
750 .flowi6_mark = skb->mark,
751 .flowi6_proto = ip6h->nexthdr,
755 dst = ip6_route_input_lookup(dev_net(sdev), sdev, &fl6, flags);
756 skb_dst_set(skb, dst);
767 unsigned int ipvlan_nf_input(void *priv, struct sk_buff *skb,
768 const struct nf_hook_state *state)
770 struct ipvl_addr *addr;
773 addr = ipvlan_skb_to_addr(skb, skb->dev);
777 skb->dev = addr->master->dev;
778 len = skb->len + ETH_HLEN;
779 ipvlan_count_rx(addr->master, len, true, false);