1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
7 static u32 ipvlan_jhash_secret __read_mostly
;
9 void ipvlan_init_secret(void)
11 net_get_random_once(&ipvlan_jhash_secret
, sizeof(ipvlan_jhash_secret
));
14 void ipvlan_count_rx(const struct ipvl_dev
*ipvlan
,
15 unsigned int len
, bool success
, bool mcast
)
17 if (likely(success
)) {
18 struct ipvl_pcpu_stats
*pcptr
;
20 pcptr
= this_cpu_ptr(ipvlan
->pcpu_stats
);
21 u64_stats_update_begin(&pcptr
->syncp
);
23 pcptr
->rx_bytes
+= len
;
26 u64_stats_update_end(&pcptr
->syncp
);
28 this_cpu_inc(ipvlan
->pcpu_stats
->rx_errs
);
31 EXPORT_SYMBOL_GPL(ipvlan_count_rx
);
33 #if IS_ENABLED(CONFIG_IPV6)
34 static u8
ipvlan_get_v6_hash(const void *iaddr
)
36 const struct in6_addr
*ip6_addr
= iaddr
;
38 return __ipv6_addr_jhash(ip6_addr
, ipvlan_jhash_secret
) &
42 static u8
ipvlan_get_v6_hash(const void *iaddr
)
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 bool addr_equal(bool is_v6
, struct ipvl_addr
*addr
, const void *iaddr
)
58 if (!is_v6
&& addr
->atype
== IPVL_IPV4
) {
59 struct in_addr
*i4addr
= (struct in_addr
*)iaddr
;
61 return addr
->ip4addr
.s_addr
== i4addr
->s_addr
;
62 #if IS_ENABLED(CONFIG_IPV6)
63 } else if (is_v6
&& addr
->atype
== IPVL_IPV6
) {
64 struct in6_addr
*i6addr
= (struct in6_addr
*)iaddr
;
66 return ipv6_addr_equal(&addr
->ip6addr
, i6addr
);
73 static struct ipvl_addr
*ipvlan_ht_addr_lookup(const struct ipvl_port
*port
,
74 const void *iaddr
, bool is_v6
)
76 struct ipvl_addr
*addr
;
79 hash
= is_v6
? ipvlan_get_v6_hash(iaddr
) :
80 ipvlan_get_v4_hash(iaddr
);
81 hlist_for_each_entry_rcu(addr
, &port
->hlhead
[hash
], hlnode
)
82 if (addr_equal(is_v6
, addr
, iaddr
))
87 void ipvlan_ht_addr_add(struct ipvl_dev
*ipvlan
, struct ipvl_addr
*addr
)
89 struct ipvl_port
*port
= ipvlan
->port
;
92 hash
= (addr
->atype
== IPVL_IPV6
) ?
93 ipvlan_get_v6_hash(&addr
->ip6addr
) :
94 ipvlan_get_v4_hash(&addr
->ip4addr
);
95 if (hlist_unhashed(&addr
->hlnode
))
96 hlist_add_head_rcu(&addr
->hlnode
, &port
->hlhead
[hash
]);
99 void ipvlan_ht_addr_del(struct ipvl_addr
*addr
)
101 hlist_del_init_rcu(&addr
->hlnode
);
104 struct ipvl_addr
*ipvlan_find_addr(const struct ipvl_dev
*ipvlan
,
105 const void *iaddr
, bool is_v6
)
107 struct ipvl_addr
*addr
, *ret
= NULL
;
110 list_for_each_entry_rcu(addr
, &ipvlan
->addrs
, anode
) {
111 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_rcu(ipvlan
, &port
->ipvlans
, pnode
) {
127 if (ipvlan_find_addr(ipvlan
, iaddr
, is_v6
)) {
136 void *ipvlan_get_L3_hdr(struct ipvl_port
*port
, struct sk_buff
*skb
, int *type
)
140 switch (skb
->protocol
) {
141 case htons(ETH_P_ARP
): {
144 if (unlikely(!pskb_may_pull(skb
, arp_hdr_len(port
->dev
))))
152 case htons(ETH_P_IP
): {
156 if (unlikely(!pskb_may_pull(skb
, sizeof(*ip4h
))))
160 pktlen
= ntohs(ip4h
->tot_len
);
161 if (ip4h
->ihl
< 5 || ip4h
->version
!= 4)
163 if (skb
->len
< pktlen
|| pktlen
< (ip4h
->ihl
* 4))
170 #if IS_ENABLED(CONFIG_IPV6)
171 case htons(ETH_P_IPV6
): {
172 struct ipv6hdr
*ip6h
;
174 if (unlikely(!pskb_may_pull(skb
, sizeof(*ip6h
))))
177 ip6h
= ipv6_hdr(skb
);
178 if (ip6h
->version
!= 6)
183 /* Only Neighbour Solicitation pkts need different treatment */
184 if (ipv6_addr_any(&ip6h
->saddr
) &&
185 ip6h
->nexthdr
== NEXTHDR_ICMP
) {
186 struct icmp6hdr
*icmph
;
188 if (unlikely(!pskb_may_pull(skb
, sizeof(*ip6h
) + sizeof(*icmph
))))
191 ip6h
= ipv6_hdr(skb
);
192 icmph
= (struct icmp6hdr
*)(ip6h
+ 1);
194 if (icmph
->icmp6_type
== NDISC_NEIGHBOUR_SOLICITATION
) {
195 /* Need to access the ipv6 address in body */
196 if (unlikely(!pskb_may_pull(skb
, sizeof(*ip6h
) + sizeof(*icmph
)
197 + sizeof(struct in6_addr
))))
200 ip6h
= ipv6_hdr(skb
);
201 icmph
= (struct icmp6hdr
*)(ip6h
+ 1);
217 unsigned int ipvlan_mac_hash(const unsigned char *addr
)
219 u32 hash
= jhash_1word(__get_unaligned_cpu32(addr
+2),
220 ipvlan_jhash_secret
);
222 return hash
& IPVLAN_MAC_FILTER_MASK
;
225 void ipvlan_process_multicast(struct work_struct
*work
)
227 struct ipvl_port
*port
= container_of(work
, struct ipvl_port
, wq
);
229 struct ipvl_dev
*ipvlan
;
230 struct sk_buff
*skb
, *nskb
;
231 struct sk_buff_head list
;
233 unsigned int mac_hash
;
238 __skb_queue_head_init(&list
);
240 spin_lock_bh(&port
->backlog
.lock
);
241 skb_queue_splice_tail_init(&port
->backlog
, &list
);
242 spin_unlock_bh(&port
->backlog
.lock
);
244 while ((skb
= __skb_dequeue(&list
)) != NULL
) {
245 struct net_device
*dev
= skb
->dev
;
246 bool consumed
= false;
249 tx_pkt
= IPVL_SKB_CB(skb
)->tx_pkt
;
250 mac_hash
= ipvlan_mac_hash(ethh
->h_dest
);
252 if (ether_addr_equal(ethh
->h_dest
, port
->dev
->broadcast
))
253 pkt_type
= PACKET_BROADCAST
;
255 pkt_type
= PACKET_MULTICAST
;
258 list_for_each_entry_rcu(ipvlan
, &port
->ipvlans
, pnode
) {
259 if (tx_pkt
&& (ipvlan
->dev
== skb
->dev
))
261 if (!test_bit(mac_hash
, ipvlan
->mac_filters
))
263 if (!(ipvlan
->dev
->flags
& IFF_UP
))
266 len
= skb
->len
+ ETH_HLEN
;
267 nskb
= skb_clone(skb
, GFP_ATOMIC
);
271 nskb
->pkt_type
= pkt_type
;
272 nskb
->dev
= ipvlan
->dev
;
274 ret
= dev_forward_skb(ipvlan
->dev
, nskb
);
276 ret
= netif_rx(nskb
);
278 ipvlan_count_rx(ipvlan
, len
, ret
== NET_RX_SUCCESS
, true);
284 /* If the packet originated here, send it out. */
285 skb
->dev
= port
->dev
;
286 skb
->pkt_type
= pkt_type
;
300 static void ipvlan_skb_crossing_ns(struct sk_buff
*skb
, struct net_device
*dev
)
305 xnet
= !net_eq(dev_net(skb
->dev
), dev_net(dev
));
307 skb_scrub_packet(skb
, xnet
);
312 static int ipvlan_rcv_frame(struct ipvl_addr
*addr
, struct sk_buff
**pskb
,
315 struct ipvl_dev
*ipvlan
= addr
->master
;
316 struct net_device
*dev
= ipvlan
->dev
;
318 rx_handler_result_t ret
= RX_HANDLER_CONSUMED
;
319 bool success
= false;
320 struct sk_buff
*skb
= *pskb
;
322 len
= skb
->len
+ ETH_HLEN
;
323 /* Only packets exchanged between two local slaves need to have
324 * device-up check as well as skb-share check.
327 if (unlikely(!(dev
->flags
& IFF_UP
))) {
332 skb
= skb_share_check(skb
, GFP_ATOMIC
);
340 skb
->pkt_type
= PACKET_HOST
;
341 if (dev_forward_skb(ipvlan
->dev
, skb
) == NET_RX_SUCCESS
)
345 ret
= RX_HANDLER_ANOTHER
;
350 ipvlan_count_rx(ipvlan
, len
, success
, false);
354 struct ipvl_addr
*ipvlan_addr_lookup(struct ipvl_port
*port
, void *lyr3h
,
355 int addr_type
, bool use_dest
)
357 struct ipvl_addr
*addr
= NULL
;
360 #if IS_ENABLED(CONFIG_IPV6)
362 struct ipv6hdr
*ip6h
;
363 struct in6_addr
*i6addr
;
365 ip6h
= (struct ipv6hdr
*)lyr3h
;
366 i6addr
= use_dest
? &ip6h
->daddr
: &ip6h
->saddr
;
367 addr
= ipvlan_ht_addr_lookup(port
, i6addr
, true);
372 struct in6_addr
*i6addr
;
374 /* Make sure that the NeighborSolicitation ICMPv6 packets
375 * are handled to avoid DAD issue.
377 ndmh
= (struct nd_msg
*)lyr3h
;
378 if (ndmh
->icmph
.icmp6_type
== NDISC_NEIGHBOUR_SOLICITATION
) {
379 i6addr
= &ndmh
->target
;
380 addr
= ipvlan_ht_addr_lookup(port
, i6addr
, true);
389 ip4h
= (struct iphdr
*)lyr3h
;
390 i4addr
= use_dest
? &ip4h
->daddr
: &ip4h
->saddr
;
391 addr
= ipvlan_ht_addr_lookup(port
, i4addr
, false);
396 unsigned char *arp_ptr
;
399 arph
= (struct arphdr
*)lyr3h
;
400 arp_ptr
= (unsigned char *)(arph
+ 1);
402 arp_ptr
+= (2 * port
->dev
->addr_len
) + 4;
404 arp_ptr
+= port
->dev
->addr_len
;
406 memcpy(&dip
, arp_ptr
, 4);
407 addr
= ipvlan_ht_addr_lookup(port
, &dip
, false);
415 static int ipvlan_process_v4_outbound(struct sk_buff
*skb
)
417 const struct iphdr
*ip4h
= ip_hdr(skb
);
418 struct net_device
*dev
= skb
->dev
;
419 struct net
*net
= dev_net(dev
);
421 int err
, ret
= NET_XMIT_DROP
;
422 struct flowi4 fl4
= {
423 .flowi4_oif
= dev
->ifindex
,
424 .flowi4_tos
= RT_TOS(ip4h
->tos
),
425 .flowi4_flags
= FLOWI_FLAG_ANYSRC
,
426 .flowi4_mark
= skb
->mark
,
427 .daddr
= ip4h
->daddr
,
428 .saddr
= ip4h
->saddr
,
431 rt
= ip_route_output_flow(net
, &fl4
, NULL
);
435 if (rt
->rt_type
!= RTN_UNICAST
&& rt
->rt_type
!= RTN_LOCAL
) {
439 skb_dst_set(skb
, &rt
->dst
);
440 err
= ip_local_out(net
, skb
->sk
, skb
);
441 if (unlikely(net_xmit_eval(err
)))
442 dev
->stats
.tx_errors
++;
444 ret
= NET_XMIT_SUCCESS
;
447 dev
->stats
.tx_errors
++;
453 #if IS_ENABLED(CONFIG_IPV6)
454 static int ipvlan_process_v6_outbound(struct sk_buff
*skb
)
456 const struct ipv6hdr
*ip6h
= ipv6_hdr(skb
);
457 struct net_device
*dev
= skb
->dev
;
458 struct net
*net
= dev_net(dev
);
459 struct dst_entry
*dst
;
460 int err
, ret
= NET_XMIT_DROP
;
461 struct flowi6 fl6
= {
462 .flowi6_oif
= dev
->ifindex
,
463 .daddr
= ip6h
->daddr
,
464 .saddr
= ip6h
->saddr
,
465 .flowi6_flags
= FLOWI_FLAG_ANYSRC
,
466 .flowlabel
= ip6_flowinfo(ip6h
),
467 .flowi6_mark
= skb
->mark
,
468 .flowi6_proto
= ip6h
->nexthdr
,
471 dst
= ip6_route_output(net
, NULL
, &fl6
);
477 skb_dst_set(skb
, dst
);
478 err
= ip6_local_out(net
, skb
->sk
, skb
);
479 if (unlikely(net_xmit_eval(err
)))
480 dev
->stats
.tx_errors
++;
482 ret
= NET_XMIT_SUCCESS
;
485 dev
->stats
.tx_errors
++;
491 static int ipvlan_process_v6_outbound(struct sk_buff
*skb
)
493 return NET_XMIT_DROP
;
497 static int ipvlan_process_outbound(struct sk_buff
*skb
)
499 struct ethhdr
*ethh
= eth_hdr(skb
);
500 int ret
= NET_XMIT_DROP
;
502 /* The ipvlan is a pseudo-L2 device, so the packets that we receive
503 * will have L2; which need to discarded and processed further
504 * in the net-ns of the main-device.
506 if (skb_mac_header_was_set(skb
)) {
507 /* In this mode we dont care about
508 * multicast and broadcast traffic */
509 if (is_multicast_ether_addr(ethh
->h_dest
)) {
510 pr_debug_ratelimited(
511 "Dropped {multi|broad}cast of type=[%x]\n",
512 ntohs(skb
->protocol
));
517 skb_pull(skb
, sizeof(*ethh
));
518 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
519 skb_reset_network_header(skb
);
522 if (skb
->protocol
== htons(ETH_P_IPV6
))
523 ret
= ipvlan_process_v6_outbound(skb
);
524 else if (skb
->protocol
== htons(ETH_P_IP
))
525 ret
= ipvlan_process_v4_outbound(skb
);
527 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
528 ntohs(skb
->protocol
));
535 static void ipvlan_multicast_enqueue(struct ipvl_port
*port
,
536 struct sk_buff
*skb
, bool tx_pkt
)
538 if (skb
->protocol
== htons(ETH_P_PAUSE
)) {
543 /* Record that the deferred packet is from TX or RX path. By
544 * looking at mac-addresses on packet will lead to erronus decisions.
545 * (This would be true for a loopback-mode on master device or a
546 * hair-pin mode of the switch.)
548 IPVL_SKB_CB(skb
)->tx_pkt
= tx_pkt
;
550 spin_lock(&port
->backlog
.lock
);
551 if (skb_queue_len(&port
->backlog
) < IPVLAN_QBACKLOG_LIMIT
) {
554 __skb_queue_tail(&port
->backlog
, skb
);
555 spin_unlock(&port
->backlog
.lock
);
556 schedule_work(&port
->wq
);
558 spin_unlock(&port
->backlog
.lock
);
559 atomic_long_inc(&skb
->dev
->rx_dropped
);
564 static int ipvlan_xmit_mode_l3(struct sk_buff
*skb
, struct net_device
*dev
)
566 const struct ipvl_dev
*ipvlan
= netdev_priv(dev
);
568 struct ipvl_addr
*addr
;
571 lyr3h
= ipvlan_get_L3_hdr(ipvlan
->port
, skb
, &addr_type
);
575 if (!ipvlan_is_vepa(ipvlan
->port
)) {
576 addr
= ipvlan_addr_lookup(ipvlan
->port
, lyr3h
, addr_type
, true);
578 if (ipvlan_is_private(ipvlan
->port
)) {
580 return NET_XMIT_DROP
;
582 return ipvlan_rcv_frame(addr
, &skb
, true);
586 ipvlan_skb_crossing_ns(skb
, ipvlan
->phy_dev
);
587 return ipvlan_process_outbound(skb
);
590 static int ipvlan_xmit_mode_l2(struct sk_buff
*skb
, struct net_device
*dev
)
592 const struct ipvl_dev
*ipvlan
= netdev_priv(dev
);
593 struct ethhdr
*eth
= eth_hdr(skb
);
594 struct ipvl_addr
*addr
;
598 if (!ipvlan_is_vepa(ipvlan
->port
) &&
599 ether_addr_equal(eth
->h_dest
, eth
->h_source
)) {
600 lyr3h
= ipvlan_get_L3_hdr(ipvlan
->port
, skb
, &addr_type
);
602 addr
= ipvlan_addr_lookup(ipvlan
->port
, lyr3h
, addr_type
, true);
604 if (ipvlan_is_private(ipvlan
->port
)) {
606 return NET_XMIT_DROP
;
608 return ipvlan_rcv_frame(addr
, &skb
, true);
611 skb
= skb_share_check(skb
, GFP_ATOMIC
);
613 return NET_XMIT_DROP
;
615 /* Packet definitely does not belong to any of the
616 * virtual devices, but the dest is local. So forward
617 * the skb for the main-dev. At the RX side we just return
618 * RX_PASS for it to be processed further on the stack.
620 return dev_forward_skb(ipvlan
->phy_dev
, skb
);
622 } else if (is_multicast_ether_addr(eth
->h_dest
)) {
623 ipvlan_skb_crossing_ns(skb
, NULL
);
624 ipvlan_multicast_enqueue(ipvlan
->port
, skb
, true);
625 return NET_XMIT_SUCCESS
;
628 skb
->dev
= ipvlan
->phy_dev
;
629 return dev_queue_xmit(skb
);
632 int ipvlan_queue_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
634 struct ipvl_dev
*ipvlan
= netdev_priv(dev
);
635 struct ipvl_port
*port
= ipvlan_port_get_rcu_bh(ipvlan
->phy_dev
);
640 if (unlikely(!pskb_may_pull(skb
, sizeof(struct ethhdr
))))
645 return ipvlan_xmit_mode_l2(skb
, dev
);
647 #ifdef CONFIG_IPVLAN_L3S
648 case IPVLAN_MODE_L3S
:
650 return ipvlan_xmit_mode_l3(skb
, dev
);
653 /* Should not reach here */
654 WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
658 return NET_XMIT_DROP
;
661 static bool ipvlan_external_frame(struct sk_buff
*skb
, struct ipvl_port
*port
)
663 struct ethhdr
*eth
= eth_hdr(skb
);
664 struct ipvl_addr
*addr
;
668 if (ether_addr_equal(eth
->h_source
, skb
->dev
->dev_addr
)) {
669 lyr3h
= ipvlan_get_L3_hdr(port
, skb
, &addr_type
);
673 addr
= ipvlan_addr_lookup(port
, lyr3h
, addr_type
, false);
681 static rx_handler_result_t
ipvlan_handle_mode_l3(struct sk_buff
**pskb
,
682 struct ipvl_port
*port
)
686 struct ipvl_addr
*addr
;
687 struct sk_buff
*skb
= *pskb
;
688 rx_handler_result_t ret
= RX_HANDLER_PASS
;
690 lyr3h
= ipvlan_get_L3_hdr(port
, skb
, &addr_type
);
694 addr
= ipvlan_addr_lookup(port
, lyr3h
, addr_type
, true);
696 ret
= ipvlan_rcv_frame(addr
, pskb
, false);
702 static rx_handler_result_t
ipvlan_handle_mode_l2(struct sk_buff
**pskb
,
703 struct ipvl_port
*port
)
705 struct sk_buff
*skb
= *pskb
;
706 struct ethhdr
*eth
= eth_hdr(skb
);
707 rx_handler_result_t ret
= RX_HANDLER_PASS
;
709 if (is_multicast_ether_addr(eth
->h_dest
)) {
710 if (ipvlan_external_frame(skb
, port
)) {
711 struct sk_buff
*nskb
= skb_clone(skb
, GFP_ATOMIC
);
713 /* External frames are queued for device local
714 * distribution, but a copy is given to master
715 * straight away to avoid sending duplicates later
716 * when work-queue processes this frame. This is
717 * achieved by returning RX_HANDLER_PASS.
720 ipvlan_skb_crossing_ns(nskb
, NULL
);
721 ipvlan_multicast_enqueue(port
, nskb
, false);
725 /* Perform like l3 mode for non-multicast packet */
726 ret
= ipvlan_handle_mode_l3(pskb
, port
);
732 rx_handler_result_t
ipvlan_handle_frame(struct sk_buff
**pskb
)
734 struct sk_buff
*skb
= *pskb
;
735 struct ipvl_port
*port
= ipvlan_port_get_rcu(skb
->dev
);
738 return RX_HANDLER_PASS
;
740 switch (port
->mode
) {
742 return ipvlan_handle_mode_l2(pskb
, port
);
744 return ipvlan_handle_mode_l3(pskb
, port
);
745 #ifdef CONFIG_IPVLAN_L3S
746 case IPVLAN_MODE_L3S
:
747 return RX_HANDLER_PASS
;
751 /* Should not reach here */
752 WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
755 return RX_HANDLER_CONSUMED
;