2 * VXLAN: Virtual eXtensible Local Area Network
4 * Copyright (c) 2012-2013 Vyatta Inc.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16 #include <linux/kernel.h>
17 #include <linux/types.h>
18 #include <linux/module.h>
19 #include <linux/errno.h>
20 #include <linux/slab.h>
21 #include <linux/skbuff.h>
22 #include <linux/rculist.h>
23 #include <linux/netdevice.h>
26 #include <linux/udp.h>
27 #include <linux/igmp.h>
28 #include <linux/etherdevice.h>
29 #include <linux/if_ether.h>
30 #include <linux/hash.h>
31 #include <linux/ethtool.h>
33 #include <net/ndisc.h>
35 #include <net/ip_tunnels.h>
38 #include <net/rtnetlink.h>
39 #include <net/route.h>
40 #include <net/dsfield.h>
41 #include <net/inet_ecn.h>
42 #include <net/net_namespace.h>
43 #include <net/netns/generic.h>
45 #define VXLAN_VERSION "0.1"
47 #define PORT_HASH_BITS 8
48 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
49 #define VNI_HASH_BITS 10
50 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
51 #define FDB_HASH_BITS 8
52 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
53 #define FDB_AGE_DEFAULT 300 /* 5 min */
54 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
56 #define VXLAN_N_VID (1u << 24)
57 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
58 /* IP header + UDP + VXLAN + Ethernet header */
59 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
61 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
63 /* VXLAN protocol header */
69 /* UDP port for VXLAN traffic.
70 * The IANA assigned port is 4789, but the Linux default is 8472
71 * for compatability with early adopters.
73 static unsigned short vxlan_port __read_mostly
= 8472;
74 module_param_named(udp_port
, vxlan_port
, ushort
, 0444);
75 MODULE_PARM_DESC(udp_port
, "Destination UDP port");
77 static bool log_ecn_error
= true;
78 module_param(log_ecn_error
, bool, 0644);
79 MODULE_PARM_DESC(log_ecn_error
, "Log packets received with corrupted ECN");
81 static unsigned int vxlan_net_id
;
83 /* per UDP socket information */
85 struct hlist_node hlist
;
87 struct work_struct del_work
;
90 struct hlist_head vni_list
[VNI_HASH_SIZE
];
93 /* per-network namespace private data for this module */
95 struct list_head vxlan_list
;
96 struct hlist_head sock_list
[PORT_HASH_SIZE
];
105 struct list_head list
;
108 /* Forwarding table entry */
110 struct hlist_node hlist
; /* linked list of entries */
112 unsigned long updated
; /* jiffies */
114 struct list_head remotes
;
115 u16 state
; /* see ndm_state */
116 u8 flags
; /* see ndm_flags */
117 u8 eth_addr
[ETH_ALEN
];
120 /* Pseudo network device */
122 struct hlist_node hlist
; /* vni hash table */
123 struct list_head next
; /* vxlan's per namespace list */
124 struct vxlan_sock
*vn_sock
; /* listening socket */
125 struct net_device
*dev
;
126 struct vxlan_rdst default_dst
; /* default destination */
127 __be32 saddr
; /* source address */
129 __u16 port_min
; /* source port range */
131 __u8 tos
; /* TOS override */
133 u32 flags
; /* VXLAN_F_* below */
135 struct work_struct sock_work
;
136 struct work_struct igmp_work
;
138 unsigned long age_interval
;
139 struct timer_list age_timer
;
140 spinlock_t hash_lock
;
141 unsigned int addrcnt
;
142 unsigned int addrmax
;
144 struct hlist_head fdb_head
[FDB_HASH_SIZE
];
147 #define VXLAN_F_LEARN 0x01
148 #define VXLAN_F_PROXY 0x02
149 #define VXLAN_F_RSC 0x04
150 #define VXLAN_F_L2MISS 0x08
151 #define VXLAN_F_L3MISS 0x10
153 /* salt for hash table */
154 static u32 vxlan_salt __read_mostly
;
155 static struct workqueue_struct
*vxlan_wq
;
157 static void vxlan_sock_work(struct work_struct
*work
);
159 /* Virtual Network hash table head */
160 static inline struct hlist_head
*vni_head(struct vxlan_sock
*vs
, u32 id
)
162 return &vs
->vni_list
[hash_32(id
, VNI_HASH_BITS
)];
165 /* Socket hash table head */
166 static inline struct hlist_head
*vs_head(struct net
*net
, __be16 port
)
168 struct vxlan_net
*vn
= net_generic(net
, vxlan_net_id
);
170 return &vn
->sock_list
[hash_32(ntohs(port
), PORT_HASH_BITS
)];
173 /* First remote destination for a forwarding entry.
174 * Guaranteed to be non-NULL because remotes are never deleted.
176 static inline struct vxlan_rdst
*first_remote(struct vxlan_fdb
*fdb
)
178 return list_first_or_null_rcu(&fdb
->remotes
, struct vxlan_rdst
, list
);
181 /* Find VXLAN socket based on network namespace and UDP port */
182 static struct vxlan_sock
*vxlan_find_port(struct net
*net
, __be16 port
)
184 struct vxlan_sock
*vs
;
186 hlist_for_each_entry_rcu(vs
, vs_head(net
, port
), hlist
) {
187 if (inet_sk(vs
->sock
->sk
)->inet_sport
== port
)
193 /* Look up VNI in a per net namespace table */
194 static struct vxlan_dev
*vxlan_find_vni(struct net
*net
, u32 id
, __be16 port
)
196 struct vxlan_sock
*vs
;
197 struct vxlan_dev
*vxlan
;
199 vs
= vxlan_find_port(net
, port
);
203 hlist_for_each_entry_rcu(vxlan
, vni_head(vs
, id
), hlist
) {
204 if (vxlan
->default_dst
.remote_vni
== id
)
211 /* Fill in neighbour message in skbuff. */
212 static int vxlan_fdb_info(struct sk_buff
*skb
, struct vxlan_dev
*vxlan
,
213 const struct vxlan_fdb
*fdb
,
214 u32 portid
, u32 seq
, int type
, unsigned int flags
,
215 const struct vxlan_rdst
*rdst
)
217 unsigned long now
= jiffies
;
218 struct nda_cacheinfo ci
;
219 struct nlmsghdr
*nlh
;
221 bool send_ip
, send_eth
;
223 nlh
= nlmsg_put(skb
, portid
, seq
, type
, sizeof(*ndm
), flags
);
227 ndm
= nlmsg_data(nlh
);
228 memset(ndm
, 0, sizeof(*ndm
));
230 send_eth
= send_ip
= true;
232 if (type
== RTM_GETNEIGH
) {
233 ndm
->ndm_family
= AF_INET
;
234 send_ip
= rdst
->remote_ip
!= htonl(INADDR_ANY
);
235 send_eth
= !is_zero_ether_addr(fdb
->eth_addr
);
237 ndm
->ndm_family
= AF_BRIDGE
;
238 ndm
->ndm_state
= fdb
->state
;
239 ndm
->ndm_ifindex
= vxlan
->dev
->ifindex
;
240 ndm
->ndm_flags
= fdb
->flags
;
241 ndm
->ndm_type
= NDA_DST
;
243 if (send_eth
&& nla_put(skb
, NDA_LLADDR
, ETH_ALEN
, &fdb
->eth_addr
))
244 goto nla_put_failure
;
246 if (send_ip
&& nla_put_be32(skb
, NDA_DST
, rdst
->remote_ip
))
247 goto nla_put_failure
;
249 if (rdst
->remote_port
&& rdst
->remote_port
!= vxlan
->dst_port
&&
250 nla_put_be16(skb
, NDA_PORT
, rdst
->remote_port
))
251 goto nla_put_failure
;
252 if (rdst
->remote_vni
!= vxlan
->default_dst
.remote_vni
&&
253 nla_put_be32(skb
, NDA_VNI
, rdst
->remote_vni
))
254 goto nla_put_failure
;
255 if (rdst
->remote_ifindex
&&
256 nla_put_u32(skb
, NDA_IFINDEX
, rdst
->remote_ifindex
))
257 goto nla_put_failure
;
259 ci
.ndm_used
= jiffies_to_clock_t(now
- fdb
->used
);
260 ci
.ndm_confirmed
= 0;
261 ci
.ndm_updated
= jiffies_to_clock_t(now
- fdb
->updated
);
264 if (nla_put(skb
, NDA_CACHEINFO
, sizeof(ci
), &ci
))
265 goto nla_put_failure
;
267 return nlmsg_end(skb
, nlh
);
270 nlmsg_cancel(skb
, nlh
);
274 static inline size_t vxlan_nlmsg_size(void)
276 return NLMSG_ALIGN(sizeof(struct ndmsg
))
277 + nla_total_size(ETH_ALEN
) /* NDA_LLADDR */
278 + nla_total_size(sizeof(__be32
)) /* NDA_DST */
279 + nla_total_size(sizeof(__be16
)) /* NDA_PORT */
280 + nla_total_size(sizeof(__be32
)) /* NDA_VNI */
281 + nla_total_size(sizeof(__u32
)) /* NDA_IFINDEX */
282 + nla_total_size(sizeof(struct nda_cacheinfo
));
285 static void vxlan_fdb_notify(struct vxlan_dev
*vxlan
,
286 struct vxlan_fdb
*fdb
, int type
)
288 struct net
*net
= dev_net(vxlan
->dev
);
292 skb
= nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC
);
296 err
= vxlan_fdb_info(skb
, vxlan
, fdb
, 0, 0, type
, 0, first_remote(fdb
));
298 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
299 WARN_ON(err
== -EMSGSIZE
);
304 rtnl_notify(skb
, net
, 0, RTNLGRP_NEIGH
, NULL
, GFP_ATOMIC
);
308 rtnl_set_sk_err(net
, RTNLGRP_NEIGH
, err
);
311 static void vxlan_ip_miss(struct net_device
*dev
, __be32 ipa
)
313 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
315 struct vxlan_rdst remote
;
317 memset(&f
, 0, sizeof f
);
320 remote
.remote_ip
= ipa
; /* goes to NDA_DST */
321 remote
.remote_vni
= VXLAN_N_VID
;
323 INIT_LIST_HEAD(&f
.remotes
);
324 list_add_rcu(&remote
.list
, &f
.remotes
);
326 vxlan_fdb_notify(vxlan
, &f
, RTM_GETNEIGH
);
329 static void vxlan_fdb_miss(struct vxlan_dev
*vxlan
, const u8 eth_addr
[ETH_ALEN
])
333 memset(&f
, 0, sizeof f
);
334 INIT_LIST_HEAD(&f
.remotes
);
336 memcpy(f
.eth_addr
, eth_addr
, ETH_ALEN
);
338 vxlan_fdb_notify(vxlan
, &f
, RTM_GETNEIGH
);
341 /* Hash Ethernet address */
342 static u32
eth_hash(const unsigned char *addr
)
344 u64 value
= get_unaligned((u64
*)addr
);
346 /* only want 6 bytes */
352 return hash_64(value
, FDB_HASH_BITS
);
355 /* Hash chain to use given mac address */
356 static inline struct hlist_head
*vxlan_fdb_head(struct vxlan_dev
*vxlan
,
359 return &vxlan
->fdb_head
[eth_hash(mac
)];
362 /* Look up Ethernet address in forwarding table */
363 static struct vxlan_fdb
*__vxlan_find_mac(struct vxlan_dev
*vxlan
,
367 struct hlist_head
*head
= vxlan_fdb_head(vxlan
, mac
);
370 hlist_for_each_entry_rcu(f
, head
, hlist
) {
371 if (compare_ether_addr(mac
, f
->eth_addr
) == 0)
378 static struct vxlan_fdb
*vxlan_find_mac(struct vxlan_dev
*vxlan
,
383 f
= __vxlan_find_mac(vxlan
, mac
);
390 /* Add/update destinations for multicast */
391 static int vxlan_fdb_append(struct vxlan_fdb
*f
,
392 __be32 ip
, __be16 port
, __u32 vni
, __u32 ifindex
)
394 struct vxlan_rdst
*rd
;
396 /* protected by vxlan->hash_lock */
397 list_for_each_entry(rd
, &f
->remotes
, list
) {
398 if (rd
->remote_ip
== ip
&&
399 rd
->remote_port
== port
&&
400 rd
->remote_vni
== vni
&&
401 rd
->remote_ifindex
== ifindex
)
405 rd
= kmalloc(sizeof(*rd
), GFP_ATOMIC
);
409 rd
->remote_port
= port
;
410 rd
->remote_vni
= vni
;
411 rd
->remote_ifindex
= ifindex
;
413 list_add_tail_rcu(&rd
->list
, &f
->remotes
);
418 /* Add new entry to forwarding table -- assumes lock held */
419 static int vxlan_fdb_create(struct vxlan_dev
*vxlan
,
420 const u8
*mac
, __be32 ip
,
421 __u16 state
, __u16 flags
,
422 __be16 port
, __u32 vni
, __u32 ifindex
,
428 f
= __vxlan_find_mac(vxlan
, mac
);
430 if (flags
& NLM_F_EXCL
) {
431 netdev_dbg(vxlan
->dev
,
432 "lost race to create %pM\n", mac
);
435 if (f
->state
!= state
) {
437 f
->updated
= jiffies
;
440 if (f
->flags
!= ndm_flags
) {
441 f
->flags
= ndm_flags
;
442 f
->updated
= jiffies
;
445 if ((flags
& NLM_F_APPEND
) &&
446 is_multicast_ether_addr(f
->eth_addr
)) {
447 int rc
= vxlan_fdb_append(f
, ip
, port
, vni
, ifindex
);
454 if (!(flags
& NLM_F_CREATE
))
457 if (vxlan
->addrmax
&& vxlan
->addrcnt
>= vxlan
->addrmax
)
460 netdev_dbg(vxlan
->dev
, "add %pM -> %pI4\n", mac
, &ip
);
461 f
= kmalloc(sizeof(*f
), GFP_ATOMIC
);
467 f
->flags
= ndm_flags
;
468 f
->updated
= f
->used
= jiffies
;
469 INIT_LIST_HEAD(&f
->remotes
);
470 memcpy(f
->eth_addr
, mac
, ETH_ALEN
);
472 vxlan_fdb_append(f
, ip
, port
, vni
, ifindex
);
475 hlist_add_head_rcu(&f
->hlist
,
476 vxlan_fdb_head(vxlan
, mac
));
480 vxlan_fdb_notify(vxlan
, f
, RTM_NEWNEIGH
);
485 static void vxlan_fdb_free(struct rcu_head
*head
)
487 struct vxlan_fdb
*f
= container_of(head
, struct vxlan_fdb
, rcu
);
488 struct vxlan_rdst
*rd
, *nd
;
490 list_for_each_entry_safe(rd
, nd
, &f
->remotes
, list
)
495 static void vxlan_fdb_destroy(struct vxlan_dev
*vxlan
, struct vxlan_fdb
*f
)
497 netdev_dbg(vxlan
->dev
,
498 "delete %pM\n", f
->eth_addr
);
501 vxlan_fdb_notify(vxlan
, f
, RTM_DELNEIGH
);
503 hlist_del_rcu(&f
->hlist
);
504 call_rcu(&f
->rcu
, vxlan_fdb_free
);
507 /* Add static entry (via netlink) */
508 static int vxlan_fdb_add(struct ndmsg
*ndm
, struct nlattr
*tb
[],
509 struct net_device
*dev
,
510 const unsigned char *addr
, u16 flags
)
512 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
513 struct net
*net
= dev_net(vxlan
->dev
);
519 if (!(ndm
->ndm_state
& (NUD_PERMANENT
|NUD_REACHABLE
))) {
520 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
525 if (tb
[NDA_DST
] == NULL
)
528 if (nla_len(tb
[NDA_DST
]) != sizeof(__be32
))
529 return -EAFNOSUPPORT
;
531 ip
= nla_get_be32(tb
[NDA_DST
]);
534 if (nla_len(tb
[NDA_PORT
]) != sizeof(__be16
))
536 port
= nla_get_be16(tb
[NDA_PORT
]);
538 port
= vxlan
->dst_port
;
541 if (nla_len(tb
[NDA_VNI
]) != sizeof(u32
))
543 vni
= nla_get_u32(tb
[NDA_VNI
]);
545 vni
= vxlan
->default_dst
.remote_vni
;
547 if (tb
[NDA_IFINDEX
]) {
548 struct net_device
*tdev
;
550 if (nla_len(tb
[NDA_IFINDEX
]) != sizeof(u32
))
552 ifindex
= nla_get_u32(tb
[NDA_IFINDEX
]);
553 tdev
= dev_get_by_index(net
, ifindex
);
555 return -EADDRNOTAVAIL
;
560 spin_lock_bh(&vxlan
->hash_lock
);
561 err
= vxlan_fdb_create(vxlan
, addr
, ip
, ndm
->ndm_state
, flags
,
562 port
, vni
, ifindex
, ndm
->ndm_flags
);
563 spin_unlock_bh(&vxlan
->hash_lock
);
568 /* Delete entry (via netlink) */
569 static int vxlan_fdb_delete(struct ndmsg
*ndm
, struct nlattr
*tb
[],
570 struct net_device
*dev
,
571 const unsigned char *addr
)
573 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
577 spin_lock_bh(&vxlan
->hash_lock
);
578 f
= vxlan_find_mac(vxlan
, addr
);
580 vxlan_fdb_destroy(vxlan
, f
);
583 spin_unlock_bh(&vxlan
->hash_lock
);
588 /* Dump forwarding table */
589 static int vxlan_fdb_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
,
590 struct net_device
*dev
, int idx
)
592 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
595 for (h
= 0; h
< FDB_HASH_SIZE
; ++h
) {
599 hlist_for_each_entry_rcu(f
, &vxlan
->fdb_head
[h
], hlist
) {
600 struct vxlan_rdst
*rd
;
602 if (idx
< cb
->args
[0])
605 list_for_each_entry_rcu(rd
, &f
->remotes
, list
) {
606 err
= vxlan_fdb_info(skb
, vxlan
, f
,
607 NETLINK_CB(cb
->skb
).portid
,
622 /* Watch incoming packets to learn mapping between Ethernet address
623 * and Tunnel endpoint.
624 * Return true if packet is bogus and should be droppped.
626 static bool vxlan_snoop(struct net_device
*dev
,
627 __be32 src_ip
, const u8
*src_mac
)
629 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
632 f
= vxlan_find_mac(vxlan
, src_mac
);
634 struct vxlan_rdst
*rdst
= first_remote(f
);
636 if (likely(rdst
->remote_ip
== src_ip
))
639 /* Don't migrate static entries, drop packets */
640 if (f
->state
& NUD_NOARP
)
645 "%pM migrated from %pI4 to %pI4\n",
646 src_mac
, &rdst
->remote_ip
, &src_ip
);
648 rdst
->remote_ip
= src_ip
;
649 f
->updated
= jiffies
;
650 vxlan_fdb_notify(vxlan
, f
, RTM_NEWNEIGH
);
652 /* learned new entry */
653 spin_lock(&vxlan
->hash_lock
);
655 /* close off race between vxlan_flush and incoming packets */
656 if (netif_running(dev
))
657 vxlan_fdb_create(vxlan
, src_mac
, src_ip
,
659 NLM_F_EXCL
|NLM_F_CREATE
,
661 vxlan
->default_dst
.remote_vni
,
663 spin_unlock(&vxlan
->hash_lock
);
670 /* See if multicast group is already in use by other ID */
671 static bool vxlan_group_used(struct vxlan_net
*vn
, __be32 remote_ip
)
673 struct vxlan_dev
*vxlan
;
675 list_for_each_entry(vxlan
, &vn
->vxlan_list
, next
) {
676 if (!netif_running(vxlan
->dev
))
679 if (vxlan
->default_dst
.remote_ip
== remote_ip
)
686 static void vxlan_sock_hold(struct vxlan_sock
*vs
)
688 atomic_inc(&vs
->refcnt
);
691 static void vxlan_sock_release(struct vxlan_net
*vn
, struct vxlan_sock
*vs
)
693 if (!atomic_dec_and_test(&vs
->refcnt
))
696 spin_lock(&vn
->sock_lock
);
697 hlist_del_rcu(&vs
->hlist
);
698 spin_unlock(&vn
->sock_lock
);
700 queue_work(vxlan_wq
, &vs
->del_work
);
703 /* Callback to update multicast group membership.
704 * Scheduled when vxlan goes up/down.
706 static void vxlan_igmp_work(struct work_struct
*work
)
708 struct vxlan_dev
*vxlan
= container_of(work
, struct vxlan_dev
, igmp_work
);
709 struct vxlan_net
*vn
= net_generic(dev_net(vxlan
->dev
), vxlan_net_id
);
710 struct vxlan_sock
*vs
= vxlan
->vn_sock
;
711 struct sock
*sk
= vs
->sock
->sk
;
712 struct ip_mreqn mreq
= {
713 .imr_multiaddr
.s_addr
= vxlan
->default_dst
.remote_ip
,
714 .imr_ifindex
= vxlan
->default_dst
.remote_ifindex
,
718 if (vxlan_group_used(vn
, vxlan
->default_dst
.remote_ip
))
719 ip_mc_join_group(sk
, &mreq
);
721 ip_mc_leave_group(sk
, &mreq
);
724 vxlan_sock_release(vn
, vs
);
728 /* Callback from net/ipv4/udp.c to receive packets */
729 static int vxlan_udp_encap_recv(struct sock
*sk
, struct sk_buff
*skb
)
732 struct vxlanhdr
*vxh
;
733 struct vxlan_dev
*vxlan
;
734 struct pcpu_tstats
*stats
;
739 /* pop off outer UDP header */
740 __skb_pull(skb
, sizeof(struct udphdr
));
742 /* Need Vxlan and inner Ethernet header to be present */
743 if (!pskb_may_pull(skb
, sizeof(struct vxlanhdr
)))
746 /* Drop packets with reserved bits set */
747 vxh
= (struct vxlanhdr
*) skb
->data
;
748 if (vxh
->vx_flags
!= htonl(VXLAN_FLAGS
) ||
749 (vxh
->vx_vni
& htonl(0xff))) {
750 netdev_dbg(skb
->dev
, "invalid vxlan flags=%#x vni=%#x\n",
751 ntohl(vxh
->vx_flags
), ntohl(vxh
->vx_vni
));
755 __skb_pull(skb
, sizeof(struct vxlanhdr
));
757 /* Is this VNI defined? */
758 vni
= ntohl(vxh
->vx_vni
) >> 8;
759 port
= inet_sk(sk
)->inet_sport
;
760 vxlan
= vxlan_find_vni(sock_net(sk
), vni
, port
);
762 netdev_dbg(skb
->dev
, "unknown vni %d port %u\n",
767 if (!pskb_may_pull(skb
, ETH_HLEN
)) {
768 vxlan
->dev
->stats
.rx_length_errors
++;
769 vxlan
->dev
->stats
.rx_errors
++;
773 skb_reset_mac_header(skb
);
775 /* Re-examine inner Ethernet packet */
777 skb
->protocol
= eth_type_trans(skb
, vxlan
->dev
);
779 /* Ignore packet loops (and multicast echo) */
780 if (compare_ether_addr(eth_hdr(skb
)->h_source
,
781 vxlan
->dev
->dev_addr
) == 0)
784 if ((vxlan
->flags
& VXLAN_F_LEARN
) &&
785 vxlan_snoop(skb
->dev
, oip
->saddr
, eth_hdr(skb
)->h_source
))
788 __skb_tunnel_rx(skb
, vxlan
->dev
);
789 skb_reset_network_header(skb
);
791 /* If the NIC driver gave us an encapsulated packet with
792 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
793 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
794 * for us. Otherwise force the upper layers to verify it.
796 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
|| !skb
->encapsulation
||
797 !(vxlan
->dev
->features
& NETIF_F_RXCSUM
))
798 skb
->ip_summed
= CHECKSUM_NONE
;
800 skb
->encapsulation
= 0;
802 err
= IP_ECN_decapsulate(oip
, skb
);
805 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
806 &oip
->saddr
, oip
->tos
);
808 ++vxlan
->dev
->stats
.rx_frame_errors
;
809 ++vxlan
->dev
->stats
.rx_errors
;
814 stats
= this_cpu_ptr(vxlan
->dev
->tstats
);
815 u64_stats_update_begin(&stats
->syncp
);
817 stats
->rx_bytes
+= skb
->len
;
818 u64_stats_update_end(&stats
->syncp
);
824 /* Put UDP header back */
825 __skb_push(skb
, sizeof(struct udphdr
));
829 /* Consume bad packet */
834 static int arp_reduce(struct net_device
*dev
, struct sk_buff
*skb
)
836 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
842 if (dev
->flags
& IFF_NOARP
)
845 if (!pskb_may_pull(skb
, arp_hdr_len(dev
))) {
846 dev
->stats
.tx_dropped
++;
851 if ((parp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
852 parp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
853 parp
->ar_pro
!= htons(ETH_P_IP
) ||
854 parp
->ar_op
!= htons(ARPOP_REQUEST
) ||
855 parp
->ar_hln
!= dev
->addr_len
||
858 arpptr
= (u8
*)parp
+ sizeof(struct arphdr
);
860 arpptr
+= dev
->addr_len
; /* sha */
861 memcpy(&sip
, arpptr
, sizeof(sip
));
862 arpptr
+= sizeof(sip
);
863 arpptr
+= dev
->addr_len
; /* tha */
864 memcpy(&tip
, arpptr
, sizeof(tip
));
866 if (ipv4_is_loopback(tip
) ||
867 ipv4_is_multicast(tip
))
870 n
= neigh_lookup(&arp_tbl
, &tip
, dev
);
874 struct sk_buff
*reply
;
876 if (!(n
->nud_state
& NUD_CONNECTED
)) {
881 f
= vxlan_find_mac(vxlan
, n
->ha
);
882 if (f
&& first_remote(f
)->remote_ip
== htonl(INADDR_ANY
)) {
883 /* bridge-local neighbor */
888 reply
= arp_create(ARPOP_REPLY
, ETH_P_ARP
, sip
, dev
, tip
, sha
,
893 skb_reset_mac_header(reply
);
894 __skb_pull(reply
, skb_network_offset(reply
));
895 reply
->ip_summed
= CHECKSUM_UNNECESSARY
;
896 reply
->pkt_type
= PACKET_HOST
;
898 if (netif_rx_ni(reply
) == NET_RX_DROP
)
899 dev
->stats
.rx_dropped
++;
900 } else if (vxlan
->flags
& VXLAN_F_L3MISS
)
901 vxlan_ip_miss(dev
, tip
);
907 static bool route_shortcircuit(struct net_device
*dev
, struct sk_buff
*skb
)
909 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
913 if (is_multicast_ether_addr(eth_hdr(skb
)->h_dest
))
917 switch (ntohs(eth_hdr(skb
)->h_proto
)) {
919 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
922 n
= neigh_lookup(&arp_tbl
, &pip
->daddr
, dev
);
931 diff
= compare_ether_addr(eth_hdr(skb
)->h_dest
, n
->ha
) != 0;
933 memcpy(eth_hdr(skb
)->h_source
, eth_hdr(skb
)->h_dest
,
935 memcpy(eth_hdr(skb
)->h_dest
, n
->ha
, dev
->addr_len
);
939 } else if (vxlan
->flags
& VXLAN_F_L3MISS
)
940 vxlan_ip_miss(dev
, pip
->daddr
);
944 static void vxlan_sock_put(struct sk_buff
*skb
)
949 /* On transmit, associate with the tunnel socket */
950 static void vxlan_set_owner(struct net_device
*dev
, struct sk_buff
*skb
)
952 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
953 struct sock
*sk
= vxlan
->vn_sock
->sock
->sk
;
958 skb
->destructor
= vxlan_sock_put
;
961 /* Compute source port for outgoing packet
962 * first choice to use L4 flow hash since it will spread
963 * better and maybe available from hardware
964 * secondary choice is to use jhash on the Ethernet header
966 static __be16
vxlan_src_port(const struct vxlan_dev
*vxlan
, struct sk_buff
*skb
)
968 unsigned int range
= (vxlan
->port_max
- vxlan
->port_min
) + 1;
971 hash
= skb_get_rxhash(skb
);
973 hash
= jhash(skb
->data
, 2 * ETH_ALEN
,
974 (__force u32
) skb
->protocol
);
976 return htons((((u64
) hash
* range
) >> 32) + vxlan
->port_min
);
979 static int handle_offloads(struct sk_buff
*skb
)
981 if (skb_is_gso(skb
)) {
982 int err
= skb_unclone(skb
, GFP_ATOMIC
);
986 skb_shinfo(skb
)->gso_type
|= SKB_GSO_UDP_TUNNEL
;
987 } else if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
988 skb
->ip_summed
= CHECKSUM_NONE
;
993 /* Bypass encapsulation if the destination is local */
994 static void vxlan_encap_bypass(struct sk_buff
*skb
, struct vxlan_dev
*src_vxlan
,
995 struct vxlan_dev
*dst_vxlan
)
997 struct pcpu_tstats
*tx_stats
= this_cpu_ptr(src_vxlan
->dev
->tstats
);
998 struct pcpu_tstats
*rx_stats
= this_cpu_ptr(dst_vxlan
->dev
->tstats
);
1000 skb
->pkt_type
= PACKET_HOST
;
1001 skb
->encapsulation
= 0;
1002 skb
->dev
= dst_vxlan
->dev
;
1003 __skb_pull(skb
, skb_network_offset(skb
));
1005 if (dst_vxlan
->flags
& VXLAN_F_LEARN
)
1006 vxlan_snoop(skb
->dev
, htonl(INADDR_LOOPBACK
),
1007 eth_hdr(skb
)->h_source
);
1009 u64_stats_update_begin(&tx_stats
->syncp
);
1010 tx_stats
->tx_packets
++;
1011 tx_stats
->tx_bytes
+= skb
->len
;
1012 u64_stats_update_end(&tx_stats
->syncp
);
1014 if (netif_rx(skb
) == NET_RX_SUCCESS
) {
1015 u64_stats_update_begin(&rx_stats
->syncp
);
1016 rx_stats
->rx_packets
++;
1017 rx_stats
->rx_bytes
+= skb
->len
;
1018 u64_stats_update_end(&rx_stats
->syncp
);
1020 skb
->dev
->stats
.rx_dropped
++;
1024 static void vxlan_xmit_one(struct sk_buff
*skb
, struct net_device
*dev
,
1025 struct vxlan_rdst
*rdst
, bool did_rsc
)
1027 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1029 const struct iphdr
*old_iph
;
1030 struct vxlanhdr
*vxh
;
1034 __be16 src_port
, dst_port
;
1040 dst_port
= rdst
->remote_port
? rdst
->remote_port
: vxlan
->dst_port
;
1041 vni
= rdst
->remote_vni
;
1042 dst
= rdst
->remote_ip
;
1046 /* short-circuited back to local bridge */
1047 vxlan_encap_bypass(skb
, vxlan
, vxlan
);
1053 if (!skb
->encapsulation
) {
1054 skb_reset_inner_headers(skb
);
1055 skb
->encapsulation
= 1;
1058 /* Need space for new headers (invalidates iph ptr) */
1059 if (skb_cow_head(skb
, VXLAN_HEADROOM
))
1062 old_iph
= ip_hdr(skb
);
1065 if (!ttl
&& IN_MULTICAST(ntohl(dst
)))
1070 tos
= ip_tunnel_get_dsfield(old_iph
, skb
);
1072 src_port
= vxlan_src_port(vxlan
, skb
);
1074 memset(&fl4
, 0, sizeof(fl4
));
1075 fl4
.flowi4_oif
= rdst
->remote_ifindex
;
1076 fl4
.flowi4_tos
= RT_TOS(tos
);
1078 fl4
.saddr
= vxlan
->saddr
;
1080 rt
= ip_route_output_key(dev_net(dev
), &fl4
);
1082 netdev_dbg(dev
, "no route to %pI4\n", &dst
);
1083 dev
->stats
.tx_carrier_errors
++;
1087 if (rt
->dst
.dev
== dev
) {
1088 netdev_dbg(dev
, "circular route to %pI4\n", &dst
);
1090 dev
->stats
.collisions
++;
1094 /* Bypass encapsulation if the destination is local */
1095 if (rt
->rt_flags
& RTCF_LOCAL
&&
1096 !(rt
->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))) {
1097 struct vxlan_dev
*dst_vxlan
;
1100 dst_vxlan
= vxlan_find_vni(dev_net(dev
), vni
, dst_port
);
1103 vxlan_encap_bypass(skb
, vxlan
, dst_vxlan
);
1106 vxh
= (struct vxlanhdr
*) __skb_push(skb
, sizeof(*vxh
));
1107 vxh
->vx_flags
= htonl(VXLAN_FLAGS
);
1108 vxh
->vx_vni
= htonl(vni
<< 8);
1110 __skb_push(skb
, sizeof(*uh
));
1111 skb_reset_transport_header(skb
);
1114 uh
->dest
= dst_port
;
1115 uh
->source
= src_port
;
1117 uh
->len
= htons(skb
->len
);
1120 vxlan_set_owner(dev
, skb
);
1122 if (handle_offloads(skb
))
1125 tos
= ip_tunnel_ecn_encap(tos
, old_iph
, skb
);
1126 ttl
= ttl
? : ip4_dst_hoplimit(&rt
->dst
);
1128 err
= iptunnel_xmit(dev_net(dev
), rt
, skb
, fl4
.saddr
, dst
,
1129 IPPROTO_UDP
, tos
, ttl
, df
);
1130 iptunnel_xmit_stats(err
, &dev
->stats
, dev
->tstats
);
1135 dev
->stats
.tx_dropped
++;
1139 dev
->stats
.tx_errors
++;
1144 /* Transmit local packets over Vxlan
1146 * Outer IP header inherits ECN and DF from inner header.
1147 * Outer UDP destination is the VXLAN assigned port.
1148 * source port is based on hash of flow
1150 static netdev_tx_t
vxlan_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1152 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1154 bool did_rsc
= false;
1155 struct vxlan_rdst
*rdst0
, *rdst
;
1156 struct vxlan_fdb
*f
;
1158 skb_reset_mac_header(skb
);
1161 if ((vxlan
->flags
& VXLAN_F_PROXY
) && ntohs(eth
->h_proto
) == ETH_P_ARP
)
1162 return arp_reduce(dev
, skb
);
1164 f
= vxlan_find_mac(vxlan
, eth
->h_dest
);
1167 if (f
&& (f
->flags
& NTF_ROUTER
) && (vxlan
->flags
& VXLAN_F_RSC
) &&
1168 ntohs(eth
->h_proto
) == ETH_P_IP
) {
1169 did_rsc
= route_shortcircuit(dev
, skb
);
1171 f
= vxlan_find_mac(vxlan
, eth
->h_dest
);
1175 rdst0
= &vxlan
->default_dst
;
1177 if (rdst0
->remote_ip
== htonl(INADDR_ANY
) &&
1178 (vxlan
->flags
& VXLAN_F_L2MISS
) &&
1179 !is_multicast_ether_addr(eth
->h_dest
))
1180 vxlan_fdb_miss(vxlan
, eth
->h_dest
);
1182 rdst
= rdst0
= first_remote(f
);
1184 /* if there are multiple destinations, send copies */
1185 list_for_each_entry_continue_rcu(rdst
, &f
->remotes
, list
) {
1186 struct sk_buff
*skb1
;
1188 skb1
= skb_clone(skb
, GFP_ATOMIC
);
1190 vxlan_xmit_one(skb1
, dev
, rdst
, did_rsc
);
1194 vxlan_xmit_one(skb
, dev
, rdst0
, did_rsc
);
1195 return NETDEV_TX_OK
;
1198 /* Walk the forwarding table and purge stale entries */
1199 static void vxlan_cleanup(unsigned long arg
)
1201 struct vxlan_dev
*vxlan
= (struct vxlan_dev
*) arg
;
1202 unsigned long next_timer
= jiffies
+ FDB_AGE_INTERVAL
;
1205 if (!netif_running(vxlan
->dev
))
1208 spin_lock_bh(&vxlan
->hash_lock
);
1209 for (h
= 0; h
< FDB_HASH_SIZE
; ++h
) {
1210 struct hlist_node
*p
, *n
;
1211 hlist_for_each_safe(p
, n
, &vxlan
->fdb_head
[h
]) {
1213 = container_of(p
, struct vxlan_fdb
, hlist
);
1214 unsigned long timeout
;
1216 if (f
->state
& NUD_PERMANENT
)
1219 timeout
= f
->used
+ vxlan
->age_interval
* HZ
;
1220 if (time_before_eq(timeout
, jiffies
)) {
1221 netdev_dbg(vxlan
->dev
,
1222 "garbage collect %pM\n",
1224 f
->state
= NUD_STALE
;
1225 vxlan_fdb_destroy(vxlan
, f
);
1226 } else if (time_before(timeout
, next_timer
))
1227 next_timer
= timeout
;
1230 spin_unlock_bh(&vxlan
->hash_lock
);
1232 mod_timer(&vxlan
->age_timer
, next_timer
);
1235 /* Setup stats when device is created */
1236 static int vxlan_init(struct net_device
*dev
)
1238 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1239 struct vxlan_net
*vn
= net_generic(dev_net(dev
), vxlan_net_id
);
1240 struct vxlan_sock
*vs
;
1241 __u32 vni
= vxlan
->default_dst
.remote_vni
;
1243 dev
->tstats
= alloc_percpu(struct pcpu_tstats
);
1247 spin_lock(&vn
->sock_lock
);
1248 vs
= vxlan_find_port(dev_net(dev
), vxlan
->dst_port
);
1250 /* If we have a socket with same port already, reuse it */
1251 atomic_inc(&vs
->refcnt
);
1252 vxlan
->vn_sock
= vs
;
1253 hlist_add_head_rcu(&vxlan
->hlist
, vni_head(vs
, vni
));
1255 /* otherwise make new socket outside of RTNL */
1257 queue_work(vxlan_wq
, &vxlan
->sock_work
);
1259 spin_unlock(&vn
->sock_lock
);
1264 static void vxlan_uninit(struct net_device
*dev
)
1266 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1267 struct vxlan_net
*vn
= net_generic(dev_net(dev
), vxlan_net_id
);
1268 struct vxlan_sock
*vs
= vxlan
->vn_sock
;
1271 vxlan_sock_release(vn
, vs
);
1272 free_percpu(dev
->tstats
);
1275 /* Start ageing timer and join group when device is brought up */
1276 static int vxlan_open(struct net_device
*dev
)
1278 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1279 struct vxlan_sock
*vs
= vxlan
->vn_sock
;
1281 /* socket hasn't been created */
1285 if (IN_MULTICAST(ntohl(vxlan
->default_dst
.remote_ip
))) {
1286 vxlan_sock_hold(vs
);
1288 queue_work(vxlan_wq
, &vxlan
->igmp_work
);
1291 if (vxlan
->age_interval
)
1292 mod_timer(&vxlan
->age_timer
, jiffies
+ FDB_AGE_INTERVAL
);
1297 /* Purge the forwarding table */
1298 static void vxlan_flush(struct vxlan_dev
*vxlan
)
1302 spin_lock_bh(&vxlan
->hash_lock
);
1303 for (h
= 0; h
< FDB_HASH_SIZE
; ++h
) {
1304 struct hlist_node
*p
, *n
;
1305 hlist_for_each_safe(p
, n
, &vxlan
->fdb_head
[h
]) {
1307 = container_of(p
, struct vxlan_fdb
, hlist
);
1308 vxlan_fdb_destroy(vxlan
, f
);
1311 spin_unlock_bh(&vxlan
->hash_lock
);
1314 /* Cleanup timer and forwarding table on shutdown */
1315 static int vxlan_stop(struct net_device
*dev
)
1317 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1318 struct vxlan_sock
*vs
= vxlan
->vn_sock
;
1320 if (vs
&& IN_MULTICAST(ntohl(vxlan
->default_dst
.remote_ip
))) {
1321 vxlan_sock_hold(vs
);
1323 queue_work(vxlan_wq
, &vxlan
->igmp_work
);
1326 del_timer_sync(&vxlan
->age_timer
);
1333 /* Stub, nothing needs to be done. */
1334 static void vxlan_set_multicast_list(struct net_device
*dev
)
1338 static const struct net_device_ops vxlan_netdev_ops
= {
1339 .ndo_init
= vxlan_init
,
1340 .ndo_uninit
= vxlan_uninit
,
1341 .ndo_open
= vxlan_open
,
1342 .ndo_stop
= vxlan_stop
,
1343 .ndo_start_xmit
= vxlan_xmit
,
1344 .ndo_get_stats64
= ip_tunnel_get_stats64
,
1345 .ndo_set_rx_mode
= vxlan_set_multicast_list
,
1346 .ndo_change_mtu
= eth_change_mtu
,
1347 .ndo_validate_addr
= eth_validate_addr
,
1348 .ndo_set_mac_address
= eth_mac_addr
,
1349 .ndo_fdb_add
= vxlan_fdb_add
,
1350 .ndo_fdb_del
= vxlan_fdb_delete
,
1351 .ndo_fdb_dump
= vxlan_fdb_dump
,
1354 /* Info for udev, that this is a virtual tunnel endpoint */
1355 static struct device_type vxlan_type
= {
1359 /* Initialize the device structure. */
1360 static void vxlan_setup(struct net_device
*dev
)
1362 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1366 eth_hw_addr_random(dev
);
1368 dev
->hard_header_len
= ETH_HLEN
+ VXLAN_HEADROOM
;
1370 dev
->netdev_ops
= &vxlan_netdev_ops
;
1371 dev
->destructor
= free_netdev
;
1372 SET_NETDEV_DEVTYPE(dev
, &vxlan_type
);
1374 dev
->tx_queue_len
= 0;
1375 dev
->features
|= NETIF_F_LLTX
;
1376 dev
->features
|= NETIF_F_NETNS_LOCAL
;
1377 dev
->features
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1378 dev
->features
|= NETIF_F_RXCSUM
;
1379 dev
->features
|= NETIF_F_GSO_SOFTWARE
;
1381 dev
->hw_features
|= NETIF_F_SG
| NETIF_F_HW_CSUM
| NETIF_F_RXCSUM
;
1382 dev
->hw_features
|= NETIF_F_GSO_SOFTWARE
;
1383 dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
1384 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1386 INIT_LIST_HEAD(&vxlan
->next
);
1387 spin_lock_init(&vxlan
->hash_lock
);
1388 INIT_WORK(&vxlan
->igmp_work
, vxlan_igmp_work
);
1389 INIT_WORK(&vxlan
->sock_work
, vxlan_sock_work
);
1391 init_timer_deferrable(&vxlan
->age_timer
);
1392 vxlan
->age_timer
.function
= vxlan_cleanup
;
1393 vxlan
->age_timer
.data
= (unsigned long) vxlan
;
1395 inet_get_local_port_range(&low
, &high
);
1396 vxlan
->port_min
= low
;
1397 vxlan
->port_max
= high
;
1398 vxlan
->dst_port
= htons(vxlan_port
);
1402 for (h
= 0; h
< FDB_HASH_SIZE
; ++h
)
1403 INIT_HLIST_HEAD(&vxlan
->fdb_head
[h
]);
1406 static const struct nla_policy vxlan_policy
[IFLA_VXLAN_MAX
+ 1] = {
1407 [IFLA_VXLAN_ID
] = { .type
= NLA_U32
},
1408 [IFLA_VXLAN_GROUP
] = { .len
= FIELD_SIZEOF(struct iphdr
, daddr
) },
1409 [IFLA_VXLAN_LINK
] = { .type
= NLA_U32
},
1410 [IFLA_VXLAN_LOCAL
] = { .len
= FIELD_SIZEOF(struct iphdr
, saddr
) },
1411 [IFLA_VXLAN_TOS
] = { .type
= NLA_U8
},
1412 [IFLA_VXLAN_TTL
] = { .type
= NLA_U8
},
1413 [IFLA_VXLAN_LEARNING
] = { .type
= NLA_U8
},
1414 [IFLA_VXLAN_AGEING
] = { .type
= NLA_U32
},
1415 [IFLA_VXLAN_LIMIT
] = { .type
= NLA_U32
},
1416 [IFLA_VXLAN_PORT_RANGE
] = { .len
= sizeof(struct ifla_vxlan_port_range
) },
1417 [IFLA_VXLAN_PROXY
] = { .type
= NLA_U8
},
1418 [IFLA_VXLAN_RSC
] = { .type
= NLA_U8
},
1419 [IFLA_VXLAN_L2MISS
] = { .type
= NLA_U8
},
1420 [IFLA_VXLAN_L3MISS
] = { .type
= NLA_U8
},
1421 [IFLA_VXLAN_PORT
] = { .type
= NLA_U16
},
1424 static int vxlan_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1426 if (tb
[IFLA_ADDRESS
]) {
1427 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
) {
1428 pr_debug("invalid link address (not ethernet)\n");
1432 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
]))) {
1433 pr_debug("invalid all zero ethernet address\n");
1434 return -EADDRNOTAVAIL
;
1441 if (data
[IFLA_VXLAN_ID
]) {
1442 __u32 id
= nla_get_u32(data
[IFLA_VXLAN_ID
]);
1443 if (id
>= VXLAN_VID_MASK
)
1447 if (data
[IFLA_VXLAN_PORT_RANGE
]) {
1448 const struct ifla_vxlan_port_range
*p
1449 = nla_data(data
[IFLA_VXLAN_PORT_RANGE
]);
1451 if (ntohs(p
->high
) < ntohs(p
->low
)) {
1452 pr_debug("port range %u .. %u not valid\n",
1453 ntohs(p
->low
), ntohs(p
->high
));
1461 static void vxlan_get_drvinfo(struct net_device
*netdev
,
1462 struct ethtool_drvinfo
*drvinfo
)
1464 strlcpy(drvinfo
->version
, VXLAN_VERSION
, sizeof(drvinfo
->version
));
1465 strlcpy(drvinfo
->driver
, "vxlan", sizeof(drvinfo
->driver
));
1468 static const struct ethtool_ops vxlan_ethtool_ops
= {
1469 .get_drvinfo
= vxlan_get_drvinfo
,
1470 .get_link
= ethtool_op_get_link
,
1473 static void vxlan_del_work(struct work_struct
*work
)
1475 struct vxlan_sock
*vs
= container_of(work
, struct vxlan_sock
, del_work
);
1477 sk_release_kernel(vs
->sock
->sk
);
1481 static struct vxlan_sock
*vxlan_socket_create(struct net
*net
, __be16 port
)
1483 struct vxlan_sock
*vs
;
1485 struct sockaddr_in vxlan_addr
= {
1486 .sin_family
= AF_INET
,
1487 .sin_addr
.s_addr
= htonl(INADDR_ANY
),
1492 vs
= kmalloc(sizeof(*vs
), GFP_KERNEL
);
1494 return ERR_PTR(-ENOMEM
);
1496 for (h
= 0; h
< VNI_HASH_SIZE
; ++h
)
1497 INIT_HLIST_HEAD(&vs
->vni_list
[h
]);
1499 INIT_WORK(&vs
->del_work
, vxlan_del_work
);
1501 /* Create UDP socket for encapsulation receive. */
1502 rc
= sock_create_kern(AF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &vs
->sock
);
1504 pr_debug("UDP socket create failed\n");
1509 /* Put in proper namespace */
1511 sk_change_net(sk
, net
);
1513 vxlan_addr
.sin_port
= port
;
1515 rc
= kernel_bind(vs
->sock
, (struct sockaddr
*) &vxlan_addr
,
1516 sizeof(vxlan_addr
));
1518 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1519 &vxlan_addr
.sin_addr
, ntohs(vxlan_addr
.sin_port
), rc
);
1520 sk_release_kernel(sk
);
1525 /* Disable multicast loopback */
1526 inet_sk(sk
)->mc_loop
= 0;
1528 /* Mark socket as an encapsulation socket. */
1529 udp_sk(sk
)->encap_type
= 1;
1530 udp_sk(sk
)->encap_rcv
= vxlan_udp_encap_recv
;
1532 atomic_set(&vs
->refcnt
, 1);
1537 /* Scheduled at device creation to bind to a socket */
1538 static void vxlan_sock_work(struct work_struct
*work
)
1540 struct vxlan_dev
*vxlan
1541 = container_of(work
, struct vxlan_dev
, sock_work
);
1542 struct net_device
*dev
= vxlan
->dev
;
1543 struct net
*net
= dev_net(dev
);
1544 __u32 vni
= vxlan
->default_dst
.remote_vni
;
1545 __be16 port
= vxlan
->dst_port
;
1546 struct vxlan_net
*vn
= net_generic(net
, vxlan_net_id
);
1547 struct vxlan_sock
*nvs
, *ovs
;
1549 nvs
= vxlan_socket_create(net
, port
);
1551 netdev_err(vxlan
->dev
, "Can not create UDP socket, %ld\n",
1556 spin_lock(&vn
->sock_lock
);
1557 /* Look again to see if can reuse socket */
1558 ovs
= vxlan_find_port(net
, port
);
1560 atomic_inc(&ovs
->refcnt
);
1561 vxlan
->vn_sock
= ovs
;
1562 hlist_add_head_rcu(&vxlan
->hlist
, vni_head(ovs
, vni
));
1563 spin_unlock(&vn
->sock_lock
);
1565 sk_release_kernel(nvs
->sock
->sk
);
1568 vxlan
->vn_sock
= nvs
;
1569 hlist_add_head_rcu(&nvs
->hlist
, vs_head(net
, port
));
1570 hlist_add_head_rcu(&vxlan
->hlist
, vni_head(nvs
, vni
));
1571 spin_unlock(&vn
->sock_lock
);
1577 static int vxlan_newlink(struct net
*net
, struct net_device
*dev
,
1578 struct nlattr
*tb
[], struct nlattr
*data
[])
1580 struct vxlan_net
*vn
= net_generic(net
, vxlan_net_id
);
1581 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1582 struct vxlan_rdst
*dst
= &vxlan
->default_dst
;
1586 if (!data
[IFLA_VXLAN_ID
])
1589 vni
= nla_get_u32(data
[IFLA_VXLAN_ID
]);
1590 dst
->remote_vni
= vni
;
1592 if (data
[IFLA_VXLAN_GROUP
])
1593 dst
->remote_ip
= nla_get_be32(data
[IFLA_VXLAN_GROUP
]);
1595 if (data
[IFLA_VXLAN_LOCAL
])
1596 vxlan
->saddr
= nla_get_be32(data
[IFLA_VXLAN_LOCAL
]);
1598 if (data
[IFLA_VXLAN_LINK
] &&
1599 (dst
->remote_ifindex
= nla_get_u32(data
[IFLA_VXLAN_LINK
]))) {
1600 struct net_device
*lowerdev
1601 = __dev_get_by_index(net
, dst
->remote_ifindex
);
1604 pr_info("ifindex %d does not exist\n", dst
->remote_ifindex
);
1609 dev
->mtu
= lowerdev
->mtu
- VXLAN_HEADROOM
;
1611 /* update header length based on lower device */
1612 dev
->hard_header_len
= lowerdev
->hard_header_len
+
1616 if (data
[IFLA_VXLAN_TOS
])
1617 vxlan
->tos
= nla_get_u8(data
[IFLA_VXLAN_TOS
]);
1619 if (data
[IFLA_VXLAN_TTL
])
1620 vxlan
->ttl
= nla_get_u8(data
[IFLA_VXLAN_TTL
]);
1622 if (!data
[IFLA_VXLAN_LEARNING
] || nla_get_u8(data
[IFLA_VXLAN_LEARNING
]))
1623 vxlan
->flags
|= VXLAN_F_LEARN
;
1625 if (data
[IFLA_VXLAN_AGEING
])
1626 vxlan
->age_interval
= nla_get_u32(data
[IFLA_VXLAN_AGEING
]);
1628 vxlan
->age_interval
= FDB_AGE_DEFAULT
;
1630 if (data
[IFLA_VXLAN_PROXY
] && nla_get_u8(data
[IFLA_VXLAN_PROXY
]))
1631 vxlan
->flags
|= VXLAN_F_PROXY
;
1633 if (data
[IFLA_VXLAN_RSC
] && nla_get_u8(data
[IFLA_VXLAN_RSC
]))
1634 vxlan
->flags
|= VXLAN_F_RSC
;
1636 if (data
[IFLA_VXLAN_L2MISS
] && nla_get_u8(data
[IFLA_VXLAN_L2MISS
]))
1637 vxlan
->flags
|= VXLAN_F_L2MISS
;
1639 if (data
[IFLA_VXLAN_L3MISS
] && nla_get_u8(data
[IFLA_VXLAN_L3MISS
]))
1640 vxlan
->flags
|= VXLAN_F_L3MISS
;
1642 if (data
[IFLA_VXLAN_LIMIT
])
1643 vxlan
->addrmax
= nla_get_u32(data
[IFLA_VXLAN_LIMIT
]);
1645 if (data
[IFLA_VXLAN_PORT_RANGE
]) {
1646 const struct ifla_vxlan_port_range
*p
1647 = nla_data(data
[IFLA_VXLAN_PORT_RANGE
]);
1648 vxlan
->port_min
= ntohs(p
->low
);
1649 vxlan
->port_max
= ntohs(p
->high
);
1652 if (data
[IFLA_VXLAN_PORT
])
1653 vxlan
->dst_port
= nla_get_be16(data
[IFLA_VXLAN_PORT
]);
1655 if (vxlan_find_vni(net
, vni
, vxlan
->dst_port
)) {
1656 pr_info("duplicate VNI %u\n", vni
);
1660 SET_ETHTOOL_OPS(dev
, &vxlan_ethtool_ops
);
1662 err
= register_netdevice(dev
);
1666 list_add(&vxlan
->next
, &vn
->vxlan_list
);
1671 static void vxlan_dellink(struct net_device
*dev
, struct list_head
*head
)
1673 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1675 hlist_del_rcu(&vxlan
->hlist
);
1676 list_del(&vxlan
->next
);
1677 unregister_netdevice_queue(dev
, head
);
1680 static size_t vxlan_get_size(const struct net_device
*dev
)
1683 return nla_total_size(sizeof(__u32
)) + /* IFLA_VXLAN_ID */
1684 nla_total_size(sizeof(__be32
)) +/* IFLA_VXLAN_GROUP */
1685 nla_total_size(sizeof(__u32
)) + /* IFLA_VXLAN_LINK */
1686 nla_total_size(sizeof(__be32
))+ /* IFLA_VXLAN_LOCAL */
1687 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_TTL */
1688 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_TOS */
1689 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_LEARNING */
1690 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_PROXY */
1691 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_RSC */
1692 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_L2MISS */
1693 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_L3MISS */
1694 nla_total_size(sizeof(__u32
)) + /* IFLA_VXLAN_AGEING */
1695 nla_total_size(sizeof(__u32
)) + /* IFLA_VXLAN_LIMIT */
1696 nla_total_size(sizeof(struct ifla_vxlan_port_range
)) +
1697 nla_total_size(sizeof(__be16
))+ /* IFLA_VXLAN_PORT */
1701 static int vxlan_fill_info(struct sk_buff
*skb
, const struct net_device
*dev
)
1703 const struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1704 const struct vxlan_rdst
*dst
= &vxlan
->default_dst
;
1705 struct ifla_vxlan_port_range ports
= {
1706 .low
= htons(vxlan
->port_min
),
1707 .high
= htons(vxlan
->port_max
),
1710 if (nla_put_u32(skb
, IFLA_VXLAN_ID
, dst
->remote_vni
))
1711 goto nla_put_failure
;
1713 if (dst
->remote_ip
&& nla_put_be32(skb
, IFLA_VXLAN_GROUP
, dst
->remote_ip
))
1714 goto nla_put_failure
;
1716 if (dst
->remote_ifindex
&& nla_put_u32(skb
, IFLA_VXLAN_LINK
, dst
->remote_ifindex
))
1717 goto nla_put_failure
;
1719 if (vxlan
->saddr
&& nla_put_be32(skb
, IFLA_VXLAN_LOCAL
, vxlan
->saddr
))
1720 goto nla_put_failure
;
1722 if (nla_put_u8(skb
, IFLA_VXLAN_TTL
, vxlan
->ttl
) ||
1723 nla_put_u8(skb
, IFLA_VXLAN_TOS
, vxlan
->tos
) ||
1724 nla_put_u8(skb
, IFLA_VXLAN_LEARNING
,
1725 !!(vxlan
->flags
& VXLAN_F_LEARN
)) ||
1726 nla_put_u8(skb
, IFLA_VXLAN_PROXY
,
1727 !!(vxlan
->flags
& VXLAN_F_PROXY
)) ||
1728 nla_put_u8(skb
, IFLA_VXLAN_RSC
, !!(vxlan
->flags
& VXLAN_F_RSC
)) ||
1729 nla_put_u8(skb
, IFLA_VXLAN_L2MISS
,
1730 !!(vxlan
->flags
& VXLAN_F_L2MISS
)) ||
1731 nla_put_u8(skb
, IFLA_VXLAN_L3MISS
,
1732 !!(vxlan
->flags
& VXLAN_F_L3MISS
)) ||
1733 nla_put_u32(skb
, IFLA_VXLAN_AGEING
, vxlan
->age_interval
) ||
1734 nla_put_u32(skb
, IFLA_VXLAN_LIMIT
, vxlan
->addrmax
) ||
1735 nla_put_be16(skb
, IFLA_VXLAN_PORT
, vxlan
->dst_port
))
1736 goto nla_put_failure
;
1738 if (nla_put(skb
, IFLA_VXLAN_PORT_RANGE
, sizeof(ports
), &ports
))
1739 goto nla_put_failure
;
1747 static struct rtnl_link_ops vxlan_link_ops __read_mostly
= {
1749 .maxtype
= IFLA_VXLAN_MAX
,
1750 .policy
= vxlan_policy
,
1751 .priv_size
= sizeof(struct vxlan_dev
),
1752 .setup
= vxlan_setup
,
1753 .validate
= vxlan_validate
,
1754 .newlink
= vxlan_newlink
,
1755 .dellink
= vxlan_dellink
,
1756 .get_size
= vxlan_get_size
,
1757 .fill_info
= vxlan_fill_info
,
1760 static __net_init
int vxlan_init_net(struct net
*net
)
1762 struct vxlan_net
*vn
= net_generic(net
, vxlan_net_id
);
1765 INIT_LIST_HEAD(&vn
->vxlan_list
);
1766 spin_lock_init(&vn
->sock_lock
);
1768 for (h
= 0; h
< PORT_HASH_SIZE
; ++h
)
1769 INIT_HLIST_HEAD(&vn
->sock_list
[h
]);
1774 static __net_exit
void vxlan_exit_net(struct net
*net
)
1776 struct vxlan_net
*vn
= net_generic(net
, vxlan_net_id
);
1777 struct vxlan_dev
*vxlan
;
1780 list_for_each_entry(vxlan
, &vn
->vxlan_list
, next
)
1781 dev_close(vxlan
->dev
);
1785 static struct pernet_operations vxlan_net_ops
= {
1786 .init
= vxlan_init_net
,
1787 .exit
= vxlan_exit_net
,
1788 .id
= &vxlan_net_id
,
1789 .size
= sizeof(struct vxlan_net
),
1792 static int __init
vxlan_init_module(void)
1796 vxlan_wq
= alloc_workqueue("vxlan", 0, 0);
1800 get_random_bytes(&vxlan_salt
, sizeof(vxlan_salt
));
1802 rc
= register_pernet_device(&vxlan_net_ops
);
1806 rc
= rtnl_link_register(&vxlan_link_ops
);
1813 unregister_pernet_device(&vxlan_net_ops
);
1815 destroy_workqueue(vxlan_wq
);
1818 late_initcall(vxlan_init_module
);
1820 static void __exit
vxlan_cleanup_module(void)
1822 unregister_pernet_device(&vxlan_net_ops
);
1823 rtnl_link_unregister(&vxlan_link_ops
);
1824 destroy_workqueue(vxlan_wq
);
1827 module_exit(vxlan_cleanup_module
);
1829 MODULE_LICENSE("GPL");
1830 MODULE_VERSION(VXLAN_VERSION
);
1831 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
1832 MODULE_ALIAS_RTNL_LINK("vxlan");