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 static 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)
90 hlist_del_init_rcu(&addr->hlnode);
93 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
94 const void *iaddr, bool is_v6)
96 struct ipvl_addr *addr;
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))
108 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
110 struct ipvl_dev *ipvlan;
114 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
115 if (ipvlan_find_addr(ipvlan, iaddr, is_v6))
121 static void *ipvlan_get_L3_hdr(struct sk_buff *skb, int *type)
125 switch (skb->protocol) {
126 case htons(ETH_P_ARP): {
129 if (unlikely(!pskb_may_pull(skb, sizeof(*arph))))
137 case htons(ETH_P_IP): {
141 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
145 pktlen = ntohs(ip4h->tot_len);
146 if (ip4h->ihl < 5 || ip4h->version != 4)
148 if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
155 case htons(ETH_P_IPV6): {
156 struct ipv6hdr *ip6h;
158 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
161 ip6h = ipv6_hdr(skb);
162 if (ip6h->version != 6)
167 /* Only Neighbour Solicitation pkts need different treatment */
168 if (ipv6_addr_any(&ip6h->saddr) &&
169 ip6h->nexthdr == NEXTHDR_ICMP) {
182 unsigned int ipvlan_mac_hash(const unsigned char *addr)
184 u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
185 ipvlan_jhash_secret);
187 return hash & IPVLAN_MAC_FILTER_MASK;
190 void ipvlan_process_multicast(struct work_struct *work)
192 struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
194 struct ipvl_dev *ipvlan;
195 struct sk_buff *skb, *nskb;
196 struct sk_buff_head list;
198 unsigned int mac_hash;
203 __skb_queue_head_init(&list);
205 spin_lock_bh(&port->backlog.lock);
206 skb_queue_splice_tail_init(&port->backlog, &list);
207 spin_unlock_bh(&port->backlog.lock);
209 while ((skb = __skb_dequeue(&list)) != NULL) {
210 struct net_device *dev = skb->dev;
211 bool consumed = false;
214 tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
215 mac_hash = ipvlan_mac_hash(ethh->h_dest);
217 if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
218 pkt_type = PACKET_BROADCAST;
220 pkt_type = PACKET_MULTICAST;
223 list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
224 if (tx_pkt && (ipvlan->dev == skb->dev))
226 if (!test_bit(mac_hash, ipvlan->mac_filters))
228 if (!(ipvlan->dev->flags & IFF_UP))
231 len = skb->len + ETH_HLEN;
232 nskb = skb_clone(skb, GFP_ATOMIC);
236 nskb->pkt_type = pkt_type;
237 nskb->dev = ipvlan->dev;
239 ret = dev_forward_skb(ipvlan->dev, nskb);
241 ret = netif_rx(nskb);
243 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;
264 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
269 xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
271 skb_scrub_packet(skb, xnet);
276 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
279 struct ipvl_dev *ipvlan = addr->master;
280 struct net_device *dev = ipvlan->dev;
282 rx_handler_result_t ret = RX_HANDLER_CONSUMED;
283 bool success = false;
284 struct sk_buff *skb = *pskb;
286 len = skb->len + ETH_HLEN;
287 /* Only packets exchanged between two local slaves need to have
288 * device-up check as well as skb-share check.
291 if (unlikely(!(dev->flags & IFF_UP))) {
296 skb = skb_share_check(skb, GFP_ATOMIC);
302 ipvlan_skb_crossing_ns(skb, dev);
305 skb->pkt_type = PACKET_HOST;
306 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
309 ret = RX_HANDLER_ANOTHER;
314 ipvlan_count_rx(ipvlan, len, success, false);
318 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
319 void *lyr3h, int addr_type,
322 struct ipvl_addr *addr = NULL;
324 if (addr_type == IPVL_IPV6) {
325 struct ipv6hdr *ip6h;
326 struct in6_addr *i6addr;
328 ip6h = (struct ipv6hdr *)lyr3h;
329 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
330 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
331 } else if (addr_type == IPVL_ICMPV6) {
333 struct in6_addr *i6addr;
335 /* Make sure that the NeighborSolicitation ICMPv6 packets
336 * are handled to avoid DAD issue.
338 ndmh = (struct nd_msg *)lyr3h;
339 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
340 i6addr = &ndmh->target;
341 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
343 } else if (addr_type == IPVL_IPV4) {
347 ip4h = (struct iphdr *)lyr3h;
348 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
349 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
350 } else if (addr_type == IPVL_ARP) {
352 unsigned char *arp_ptr;
355 arph = (struct arphdr *)lyr3h;
356 arp_ptr = (unsigned char *)(arph + 1);
358 arp_ptr += (2 * port->dev->addr_len) + 4;
360 arp_ptr += port->dev->addr_len;
362 memcpy(&dip, arp_ptr, 4);
363 addr = ipvlan_ht_addr_lookup(port, &dip, false);
369 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
371 const struct iphdr *ip4h = ip_hdr(skb);
372 struct net_device *dev = skb->dev;
373 struct net *net = dev_net(dev);
375 int err, ret = NET_XMIT_DROP;
376 struct flowi4 fl4 = {
377 .flowi4_oif = dev->ifindex,
378 .flowi4_tos = RT_TOS(ip4h->tos),
379 .flowi4_flags = FLOWI_FLAG_ANYSRC,
380 .daddr = ip4h->daddr,
381 .saddr = ip4h->saddr,
384 rt = ip_route_output_flow(net, &fl4, NULL);
388 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
392 skb_dst_set(skb, &rt->dst);
393 err = ip_local_out(net, skb->sk, skb);
394 if (unlikely(net_xmit_eval(err)))
395 dev->stats.tx_errors++;
397 ret = NET_XMIT_SUCCESS;
400 dev->stats.tx_errors++;
406 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
408 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
409 struct net_device *dev = skb->dev;
410 struct net *net = dev_net(dev);
411 struct dst_entry *dst;
412 int err, ret = NET_XMIT_DROP;
413 struct flowi6 fl6 = {
414 .flowi6_iif = dev->ifindex,
415 .daddr = ip6h->daddr,
416 .saddr = ip6h->saddr,
417 .flowi6_flags = FLOWI_FLAG_ANYSRC,
418 .flowlabel = ip6_flowinfo(ip6h),
419 .flowi6_mark = skb->mark,
420 .flowi6_proto = ip6h->nexthdr,
423 dst = ip6_route_output(net, NULL, &fl6);
429 skb_dst_set(skb, dst);
430 err = ip6_local_out(net, skb->sk, skb);
431 if (unlikely(net_xmit_eval(err)))
432 dev->stats.tx_errors++;
434 ret = NET_XMIT_SUCCESS;
437 dev->stats.tx_errors++;
443 static int ipvlan_process_outbound(struct sk_buff *skb)
445 struct ethhdr *ethh = eth_hdr(skb);
446 int ret = NET_XMIT_DROP;
448 /* In this mode we dont care about multicast and broadcast traffic */
449 if (is_multicast_ether_addr(ethh->h_dest)) {
450 pr_warn_ratelimited("Dropped {multi|broad}cast of type= [%x]\n",
451 ntohs(skb->protocol));
456 /* The ipvlan is a pseudo-L2 device, so the packets that we receive
457 * will have L2; which need to discarded and processed further
458 * in the net-ns of the main-device.
460 if (skb_mac_header_was_set(skb)) {
461 skb_pull(skb, sizeof(*ethh));
462 skb->mac_header = (typeof(skb->mac_header))~0U;
463 skb_reset_network_header(skb);
466 if (skb->protocol == htons(ETH_P_IPV6))
467 ret = ipvlan_process_v6_outbound(skb);
468 else if (skb->protocol == htons(ETH_P_IP))
469 ret = ipvlan_process_v4_outbound(skb);
471 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
472 ntohs(skb->protocol));
479 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
480 struct sk_buff *skb, bool tx_pkt)
482 if (skb->protocol == htons(ETH_P_PAUSE)) {
487 /* Record that the deferred packet is from TX or RX path. By
488 * looking at mac-addresses on packet will lead to erronus decisions.
489 * (This would be true for a loopback-mode on master device or a
490 * hair-pin mode of the switch.)
492 IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
494 spin_lock(&port->backlog.lock);
495 if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
498 __skb_queue_tail(&port->backlog, skb);
499 spin_unlock(&port->backlog.lock);
500 schedule_work(&port->wq);
502 spin_unlock(&port->backlog.lock);
503 atomic_long_inc(&skb->dev->rx_dropped);
508 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
510 const struct ipvl_dev *ipvlan = netdev_priv(dev);
512 struct ipvl_addr *addr;
515 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
519 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
521 return ipvlan_rcv_frame(addr, &skb, true);
524 ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
525 return ipvlan_process_outbound(skb);
528 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
530 const struct ipvl_dev *ipvlan = netdev_priv(dev);
531 struct ethhdr *eth = eth_hdr(skb);
532 struct ipvl_addr *addr;
536 if (ether_addr_equal(eth->h_dest, eth->h_source)) {
537 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
539 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
541 return ipvlan_rcv_frame(addr, &skb, true);
543 skb = skb_share_check(skb, GFP_ATOMIC);
545 return NET_XMIT_DROP;
547 /* Packet definitely does not belong to any of the
548 * virtual devices, but the dest is local. So forward
549 * the skb for the main-dev. At the RX side we just return
550 * RX_PASS for it to be processed further on the stack.
552 return dev_forward_skb(ipvlan->phy_dev, skb);
554 } else if (is_multicast_ether_addr(eth->h_dest)) {
555 ipvlan_skb_crossing_ns(skb, NULL);
556 ipvlan_multicast_enqueue(ipvlan->port, skb, true);
557 return NET_XMIT_SUCCESS;
560 ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
561 return dev_queue_xmit(skb);
564 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
566 struct ipvl_dev *ipvlan = netdev_priv(dev);
567 struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
572 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
577 return ipvlan_xmit_mode_l2(skb, dev);
579 case IPVLAN_MODE_L3S:
580 return ipvlan_xmit_mode_l3(skb, dev);
583 /* Should not reach here */
584 WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
588 return NET_XMIT_DROP;
591 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
593 struct ethhdr *eth = eth_hdr(skb);
594 struct ipvl_addr *addr;
598 if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
599 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
603 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
611 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
612 struct ipvl_port *port)
616 struct ipvl_addr *addr;
617 struct sk_buff *skb = *pskb;
618 rx_handler_result_t ret = RX_HANDLER_PASS;
620 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
624 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
626 ret = ipvlan_rcv_frame(addr, pskb, false);
632 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
633 struct ipvl_port *port)
635 struct sk_buff *skb = *pskb;
636 struct ethhdr *eth = eth_hdr(skb);
637 rx_handler_result_t ret = RX_HANDLER_PASS;
641 if (is_multicast_ether_addr(eth->h_dest)) {
642 if (ipvlan_external_frame(skb, port)) {
643 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
645 /* External frames are queued for device local
646 * distribution, but a copy is given to master
647 * straight away to avoid sending duplicates later
648 * when work-queue processes this frame. This is
649 * achieved by returning RX_HANDLER_PASS.
652 ipvlan_skb_crossing_ns(nskb, NULL);
653 ipvlan_multicast_enqueue(port, nskb, false);
657 struct ipvl_addr *addr;
659 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
663 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
665 ret = ipvlan_rcv_frame(addr, pskb, false);
671 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
673 struct sk_buff *skb = *pskb;
674 struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
677 return RX_HANDLER_PASS;
679 switch (port->mode) {
681 return ipvlan_handle_mode_l2(pskb, port);
683 return ipvlan_handle_mode_l3(pskb, port);
684 case IPVLAN_MODE_L3S:
685 return RX_HANDLER_PASS;
688 /* Should not reach here */
689 WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
692 return RX_HANDLER_CONSUMED;
695 static struct ipvl_addr *ipvlan_skb_to_addr(struct sk_buff *skb,
696 struct net_device *dev)
698 struct ipvl_addr *addr = NULL;
699 struct ipvl_port *port;
703 if (!dev || !netif_is_ipvlan_port(dev))
706 port = ipvlan_port_get_rcu(dev);
707 if (!port || port->mode != IPVLAN_MODE_L3S)
710 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
714 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
719 struct sk_buff *ipvlan_l3_rcv(struct net_device *dev, struct sk_buff *skb,
722 struct ipvl_addr *addr;
723 struct net_device *sdev;
725 addr = ipvlan_skb_to_addr(skb, dev);
729 sdev = addr->master->dev;
734 struct iphdr *ip4h = ip_hdr(skb);
736 err = ip_route_input_noref(skb, ip4h->daddr, ip4h->saddr,
744 struct dst_entry *dst;
745 struct ipv6hdr *ip6h = ipv6_hdr(skb);
746 int flags = RT6_LOOKUP_F_HAS_SADDR;
747 struct flowi6 fl6 = {
748 .flowi6_iif = sdev->ifindex,
749 .daddr = ip6h->daddr,
750 .saddr = ip6h->saddr,
751 .flowlabel = ip6_flowinfo(ip6h),
752 .flowi6_mark = skb->mark,
753 .flowi6_proto = ip6h->nexthdr,
757 dst = ip6_route_input_lookup(dev_net(sdev), sdev, &fl6, flags);
758 skb_dst_set(skb, dst);
769 unsigned int ipvlan_nf_input(void *priv, struct sk_buff *skb,
770 const struct nf_hook_state *state)
772 struct ipvl_addr *addr;
775 addr = ipvlan_skb_to_addr(skb, skb->dev);
779 skb->dev = addr->master->dev;
780 len = skb->len + ETH_HLEN;
781 ipvlan_count_rx(addr->master, len, true, false);