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 #if IS_ENABLED(CONFIG_IPV6)
39 static u8 ipvlan_get_v6_hash(const void *iaddr)
41 const struct in6_addr *ip6_addr = iaddr;
43 return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
47 static u8 ipvlan_get_v6_hash(const void *iaddr)
53 static u8 ipvlan_get_v4_hash(const void *iaddr)
55 const struct in_addr *ip4_addr = iaddr;
57 return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
61 static bool addr_equal(bool is_v6, struct ipvl_addr *addr, const void *iaddr)
63 if (!is_v6 && addr->atype == IPVL_IPV4) {
64 struct in_addr *i4addr = (struct in_addr *)iaddr;
66 return addr->ip4addr.s_addr == i4addr->s_addr;
67 #if IS_ENABLED(CONFIG_IPV6)
68 } else if (is_v6 && addr->atype == IPVL_IPV6) {
69 struct in6_addr *i6addr = (struct in6_addr *)iaddr;
71 return ipv6_addr_equal(&addr->ip6addr, i6addr);
78 static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
79 const void *iaddr, bool is_v6)
81 struct ipvl_addr *addr;
84 hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
85 ipvlan_get_v4_hash(iaddr);
86 hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode)
87 if (addr_equal(is_v6, addr, iaddr))
92 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
94 struct ipvl_port *port = ipvlan->port;
97 hash = (addr->atype == IPVL_IPV6) ?
98 ipvlan_get_v6_hash(&addr->ip6addr) :
99 ipvlan_get_v4_hash(&addr->ip4addr);
100 if (hlist_unhashed(&addr->hlnode))
101 hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
104 void ipvlan_ht_addr_del(struct ipvl_addr *addr)
106 hlist_del_init_rcu(&addr->hlnode);
109 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
110 const void *iaddr, bool is_v6)
112 struct ipvl_addr *addr;
114 list_for_each_entry(addr, &ipvlan->addrs, anode)
115 if (addr_equal(is_v6, addr, iaddr))
120 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
122 struct ipvl_dev *ipvlan;
126 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
127 if (ipvlan_find_addr(ipvlan, iaddr, is_v6))
133 static void *ipvlan_get_L3_hdr(struct ipvl_port *port, struct sk_buff *skb, int *type)
137 switch (skb->protocol) {
138 case htons(ETH_P_ARP): {
141 if (unlikely(!pskb_may_pull(skb, arp_hdr_len(port->dev))))
149 case htons(ETH_P_IP): {
153 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
157 pktlen = ntohs(ip4h->tot_len);
158 if (ip4h->ihl < 5 || ip4h->version != 4)
160 if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
167 #if IS_ENABLED(CONFIG_IPV6)
168 case htons(ETH_P_IPV6): {
169 struct ipv6hdr *ip6h;
171 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
174 ip6h = ipv6_hdr(skb);
175 if (ip6h->version != 6)
180 /* Only Neighbour Solicitation pkts need different treatment */
181 if (ipv6_addr_any(&ip6h->saddr) &&
182 ip6h->nexthdr == NEXTHDR_ICMP) {
183 struct icmp6hdr *icmph;
185 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph))))
188 ip6h = ipv6_hdr(skb);
189 icmph = (struct icmp6hdr *)(ip6h + 1);
191 if (icmph->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
192 /* Need to access the ipv6 address in body */
193 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph)
194 + sizeof(struct in6_addr))))
197 ip6h = ipv6_hdr(skb);
198 icmph = (struct icmp6hdr *)(ip6h + 1);
214 unsigned int ipvlan_mac_hash(const unsigned char *addr)
216 u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
217 ipvlan_jhash_secret);
219 return hash & IPVLAN_MAC_FILTER_MASK;
222 void ipvlan_process_multicast(struct work_struct *work)
224 struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
226 struct ipvl_dev *ipvlan;
227 struct sk_buff *skb, *nskb;
228 struct sk_buff_head list;
230 unsigned int mac_hash;
235 __skb_queue_head_init(&list);
237 spin_lock_bh(&port->backlog.lock);
238 skb_queue_splice_tail_init(&port->backlog, &list);
239 spin_unlock_bh(&port->backlog.lock);
241 while ((skb = __skb_dequeue(&list)) != NULL) {
242 struct net_device *dev = skb->dev;
243 bool consumed = false;
246 tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
247 mac_hash = ipvlan_mac_hash(ethh->h_dest);
249 if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
250 pkt_type = PACKET_BROADCAST;
252 pkt_type = PACKET_MULTICAST;
255 list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
256 if (tx_pkt && (ipvlan->dev == skb->dev))
258 if (!test_bit(mac_hash, ipvlan->mac_filters))
260 if (!(ipvlan->dev->flags & IFF_UP))
263 len = skb->len + ETH_HLEN;
264 nskb = skb_clone(skb, GFP_ATOMIC);
268 nskb->pkt_type = pkt_type;
269 nskb->dev = ipvlan->dev;
271 ret = dev_forward_skb(ipvlan->dev, nskb);
273 ret = netif_rx(nskb);
275 ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
281 /* If the packet originated here, send it out. */
282 skb->dev = port->dev;
283 skb->pkt_type = pkt_type;
296 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
301 xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
303 skb_scrub_packet(skb, xnet);
308 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
311 struct ipvl_dev *ipvlan = addr->master;
312 struct net_device *dev = ipvlan->dev;
314 rx_handler_result_t ret = RX_HANDLER_CONSUMED;
315 bool success = false;
316 struct sk_buff *skb = *pskb;
318 len = skb->len + ETH_HLEN;
319 /* Only packets exchanged between two local slaves need to have
320 * device-up check as well as skb-share check.
323 if (unlikely(!(dev->flags & IFF_UP))) {
328 skb = skb_share_check(skb, GFP_ATOMIC);
336 skb->pkt_type = PACKET_HOST;
337 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
341 ret = RX_HANDLER_ANOTHER;
346 ipvlan_count_rx(ipvlan, len, success, false);
350 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
351 void *lyr3h, int addr_type,
354 struct ipvl_addr *addr = NULL;
357 #if IS_ENABLED(CONFIG_IPV6)
359 struct ipv6hdr *ip6h;
360 struct in6_addr *i6addr;
362 ip6h = (struct ipv6hdr *)lyr3h;
363 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
364 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
369 struct in6_addr *i6addr;
371 /* Make sure that the NeighborSolicitation ICMPv6 packets
372 * are handled to avoid DAD issue.
374 ndmh = (struct nd_msg *)lyr3h;
375 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
376 i6addr = &ndmh->target;
377 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
386 ip4h = (struct iphdr *)lyr3h;
387 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
388 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
393 unsigned char *arp_ptr;
396 arph = (struct arphdr *)lyr3h;
397 arp_ptr = (unsigned char *)(arph + 1);
399 arp_ptr += (2 * port->dev->addr_len) + 4;
401 arp_ptr += port->dev->addr_len;
403 memcpy(&dip, arp_ptr, 4);
404 addr = ipvlan_ht_addr_lookup(port, &dip, false);
412 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
414 const struct iphdr *ip4h = ip_hdr(skb);
415 struct net_device *dev = skb->dev;
416 struct net *net = dev_net(dev);
418 int err, ret = NET_XMIT_DROP;
419 struct flowi4 fl4 = {
420 .flowi4_oif = dev->ifindex,
421 .flowi4_tos = RT_TOS(ip4h->tos),
422 .flowi4_flags = FLOWI_FLAG_ANYSRC,
423 .flowi4_mark = skb->mark,
424 .daddr = ip4h->daddr,
425 .saddr = ip4h->saddr,
428 rt = ip_route_output_flow(net, &fl4, NULL);
432 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
436 skb_dst_set(skb, &rt->dst);
437 err = ip_local_out(net, skb->sk, skb);
438 if (unlikely(net_xmit_eval(err)))
439 dev->stats.tx_errors++;
441 ret = NET_XMIT_SUCCESS;
444 dev->stats.tx_errors++;
450 #if IS_ENABLED(CONFIG_IPV6)
451 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
453 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
454 struct net_device *dev = skb->dev;
455 struct net *net = dev_net(dev);
456 struct dst_entry *dst;
457 int err, ret = NET_XMIT_DROP;
458 struct flowi6 fl6 = {
459 .flowi6_oif = dev->ifindex,
460 .daddr = ip6h->daddr,
461 .saddr = ip6h->saddr,
462 .flowi6_flags = FLOWI_FLAG_ANYSRC,
463 .flowlabel = ip6_flowinfo(ip6h),
464 .flowi6_mark = skb->mark,
465 .flowi6_proto = ip6h->nexthdr,
468 dst = ip6_route_output(net, NULL, &fl6);
474 skb_dst_set(skb, dst);
475 err = ip6_local_out(net, skb->sk, skb);
476 if (unlikely(net_xmit_eval(err)))
477 dev->stats.tx_errors++;
479 ret = NET_XMIT_SUCCESS;
482 dev->stats.tx_errors++;
488 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
490 return NET_XMIT_DROP;
494 static int ipvlan_process_outbound(struct sk_buff *skb)
496 struct ethhdr *ethh = eth_hdr(skb);
497 int ret = NET_XMIT_DROP;
499 /* In this mode we dont care about multicast and broadcast traffic */
500 if (is_multicast_ether_addr(ethh->h_dest)) {
501 pr_debug_ratelimited("Dropped {multi|broad}cast of type=[%x]\n",
502 ntohs(skb->protocol));
507 /* The ipvlan is a pseudo-L2 device, so the packets that we receive
508 * will have L2; which need to discarded and processed further
509 * in the net-ns of the main-device.
511 if (skb_mac_header_was_set(skb)) {
512 skb_pull(skb, sizeof(*ethh));
513 skb->mac_header = (typeof(skb->mac_header))~0U;
514 skb_reset_network_header(skb);
517 if (skb->protocol == htons(ETH_P_IPV6))
518 ret = ipvlan_process_v6_outbound(skb);
519 else if (skb->protocol == htons(ETH_P_IP))
520 ret = ipvlan_process_v4_outbound(skb);
522 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
523 ntohs(skb->protocol));
530 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
531 struct sk_buff *skb, bool tx_pkt)
533 if (skb->protocol == htons(ETH_P_PAUSE)) {
538 /* Record that the deferred packet is from TX or RX path. By
539 * looking at mac-addresses on packet will lead to erronus decisions.
540 * (This would be true for a loopback-mode on master device or a
541 * hair-pin mode of the switch.)
543 IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
545 spin_lock(&port->backlog.lock);
546 if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
549 __skb_queue_tail(&port->backlog, skb);
550 spin_unlock(&port->backlog.lock);
551 schedule_work(&port->wq);
553 spin_unlock(&port->backlog.lock);
554 atomic_long_inc(&skb->dev->rx_dropped);
559 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
561 const struct ipvl_dev *ipvlan = netdev_priv(dev);
563 struct ipvl_addr *addr;
566 lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
570 if (!ipvlan_is_vepa(ipvlan->port)) {
571 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
573 if (ipvlan_is_private(ipvlan->port)) {
575 return NET_XMIT_DROP;
577 return ipvlan_rcv_frame(addr, &skb, true);
581 ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
582 return ipvlan_process_outbound(skb);
585 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
587 const struct ipvl_dev *ipvlan = netdev_priv(dev);
588 struct ethhdr *eth = eth_hdr(skb);
589 struct ipvl_addr *addr;
593 if (!ipvlan_is_vepa(ipvlan->port) &&
594 ether_addr_equal(eth->h_dest, eth->h_source)) {
595 lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
597 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
599 if (ipvlan_is_private(ipvlan->port)) {
601 return NET_XMIT_DROP;
603 return ipvlan_rcv_frame(addr, &skb, true);
606 skb = skb_share_check(skb, GFP_ATOMIC);
608 return NET_XMIT_DROP;
610 /* Packet definitely does not belong to any of the
611 * virtual devices, but the dest is local. So forward
612 * the skb for the main-dev. At the RX side we just return
613 * RX_PASS for it to be processed further on the stack.
615 return dev_forward_skb(ipvlan->phy_dev, skb);
617 } else if (is_multicast_ether_addr(eth->h_dest)) {
618 ipvlan_skb_crossing_ns(skb, NULL);
619 ipvlan_multicast_enqueue(ipvlan->port, skb, true);
620 return NET_XMIT_SUCCESS;
623 skb->dev = ipvlan->phy_dev;
624 return dev_queue_xmit(skb);
627 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
629 struct ipvl_dev *ipvlan = netdev_priv(dev);
630 struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
635 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
640 return ipvlan_xmit_mode_l2(skb, dev);
642 case IPVLAN_MODE_L3S:
643 return ipvlan_xmit_mode_l3(skb, dev);
646 /* Should not reach here */
647 WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
651 return NET_XMIT_DROP;
654 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
656 struct ethhdr *eth = eth_hdr(skb);
657 struct ipvl_addr *addr;
661 if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
662 lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
666 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
674 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
675 struct ipvl_port *port)
679 struct ipvl_addr *addr;
680 struct sk_buff *skb = *pskb;
681 rx_handler_result_t ret = RX_HANDLER_PASS;
683 lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
687 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
689 ret = ipvlan_rcv_frame(addr, pskb, false);
695 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
696 struct ipvl_port *port)
698 struct sk_buff *skb = *pskb;
699 struct ethhdr *eth = eth_hdr(skb);
700 rx_handler_result_t ret = RX_HANDLER_PASS;
702 if (is_multicast_ether_addr(eth->h_dest)) {
703 if (ipvlan_external_frame(skb, port)) {
704 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
706 /* External frames are queued for device local
707 * distribution, but a copy is given to master
708 * straight away to avoid sending duplicates later
709 * when work-queue processes this frame. This is
710 * achieved by returning RX_HANDLER_PASS.
713 ipvlan_skb_crossing_ns(nskb, NULL);
714 ipvlan_multicast_enqueue(port, nskb, false);
718 /* Perform like l3 mode for non-multicast packet */
719 ret = ipvlan_handle_mode_l3(pskb, port);
725 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
727 struct sk_buff *skb = *pskb;
728 struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
731 return RX_HANDLER_PASS;
733 switch (port->mode) {
735 return ipvlan_handle_mode_l2(pskb, port);
737 return ipvlan_handle_mode_l3(pskb, port);
738 case IPVLAN_MODE_L3S:
739 return RX_HANDLER_PASS;
742 /* Should not reach here */
743 WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
746 return RX_HANDLER_CONSUMED;
749 static struct ipvl_addr *ipvlan_skb_to_addr(struct sk_buff *skb,
750 struct net_device *dev)
752 struct ipvl_addr *addr = NULL;
753 struct ipvl_port *port;
757 if (!dev || !netif_is_ipvlan_port(dev))
760 port = ipvlan_port_get_rcu(dev);
761 if (!port || port->mode != IPVLAN_MODE_L3S)
764 lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
768 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
773 struct sk_buff *ipvlan_l3_rcv(struct net_device *dev, struct sk_buff *skb,
776 struct ipvl_addr *addr;
777 struct net_device *sdev;
779 addr = ipvlan_skb_to_addr(skb, dev);
783 sdev = addr->master->dev;
788 struct iphdr *ip4h = ip_hdr(skb);
790 err = ip_route_input_noref(skb, ip4h->daddr, ip4h->saddr,
796 #if IS_ENABLED(CONFIG_IPV6)
799 struct dst_entry *dst;
800 struct ipv6hdr *ip6h = ipv6_hdr(skb);
801 int flags = RT6_LOOKUP_F_HAS_SADDR;
802 struct flowi6 fl6 = {
803 .flowi6_iif = sdev->ifindex,
804 .daddr = ip6h->daddr,
805 .saddr = ip6h->saddr,
806 .flowlabel = ip6_flowinfo(ip6h),
807 .flowi6_mark = skb->mark,
808 .flowi6_proto = ip6h->nexthdr,
812 dst = ip6_route_input_lookup(dev_net(sdev), sdev, &fl6, flags);
813 skb_dst_set(skb, dst);
825 unsigned int ipvlan_nf_input(void *priv, struct sk_buff *skb,
826 const struct nf_hook_state *state)
828 struct ipvl_addr *addr;
831 addr = ipvlan_skb_to_addr(skb, skb->dev);
835 skb->dev = addr->master->dev;
836 len = skb->len + ETH_HLEN;
837 ipvlan_count_rx(addr->master, len, true, false);