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 compatibility 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 int vxlan_net_id
;
83 static const u8 all_zeros_mac
[ETH_ALEN
];
85 /* per UDP socket information */
87 struct hlist_node hlist
;
89 struct work_struct del_work
;
92 struct hlist_head vni_list
[VNI_HASH_SIZE
];
95 /* per-network namespace private data for this module */
97 struct list_head vxlan_list
;
98 struct hlist_head sock_list
[PORT_HASH_SIZE
];
107 struct list_head list
;
111 /* Forwarding table entry */
113 struct hlist_node hlist
; /* linked list of entries */
115 unsigned long updated
; /* jiffies */
117 struct list_head remotes
;
118 u16 state
; /* see ndm_state */
119 u8 flags
; /* see ndm_flags */
120 u8 eth_addr
[ETH_ALEN
];
123 /* Pseudo network device */
125 struct hlist_node hlist
; /* vni hash table */
126 struct list_head next
; /* vxlan's per namespace list */
127 struct vxlan_sock
*vn_sock
; /* listening socket */
128 struct net_device
*dev
;
129 struct vxlan_rdst default_dst
; /* default destination */
130 __be32 saddr
; /* source address */
132 __u16 port_min
; /* source port range */
134 __u8 tos
; /* TOS override */
136 u32 flags
; /* VXLAN_F_* below */
138 struct work_struct sock_work
;
139 struct work_struct igmp_join
;
140 struct work_struct igmp_leave
;
142 unsigned long age_interval
;
143 struct timer_list age_timer
;
144 spinlock_t hash_lock
;
145 unsigned int addrcnt
;
146 unsigned int addrmax
;
148 struct hlist_head fdb_head
[FDB_HASH_SIZE
];
151 #define VXLAN_F_LEARN 0x01
152 #define VXLAN_F_PROXY 0x02
153 #define VXLAN_F_RSC 0x04
154 #define VXLAN_F_L2MISS 0x08
155 #define VXLAN_F_L3MISS 0x10
157 /* salt for hash table */
158 static u32 vxlan_salt __read_mostly
;
159 static struct workqueue_struct
*vxlan_wq
;
161 static void vxlan_sock_work(struct work_struct
*work
);
163 /* Virtual Network hash table head */
164 static inline struct hlist_head
*vni_head(struct vxlan_sock
*vs
, u32 id
)
166 return &vs
->vni_list
[hash_32(id
, VNI_HASH_BITS
)];
169 /* Socket hash table head */
170 static inline struct hlist_head
*vs_head(struct net
*net
, __be16 port
)
172 struct vxlan_net
*vn
= net_generic(net
, vxlan_net_id
);
174 return &vn
->sock_list
[hash_32(ntohs(port
), PORT_HASH_BITS
)];
177 /* First remote destination for a forwarding entry.
178 * Guaranteed to be non-NULL because remotes are never deleted.
180 static inline struct vxlan_rdst
*first_remote(struct vxlan_fdb
*fdb
)
182 return list_first_or_null_rcu(&fdb
->remotes
, struct vxlan_rdst
, list
);
185 /* Find VXLAN socket based on network namespace and UDP port */
186 static struct vxlan_sock
*vxlan_find_port(struct net
*net
, __be16 port
)
188 struct vxlan_sock
*vs
;
190 hlist_for_each_entry_rcu(vs
, vs_head(net
, port
), hlist
) {
191 if (inet_sk(vs
->sock
->sk
)->inet_sport
== port
)
197 /* Look up VNI in a per net namespace table */
198 static struct vxlan_dev
*vxlan_find_vni(struct net
*net
, u32 id
, __be16 port
)
200 struct vxlan_sock
*vs
;
201 struct vxlan_dev
*vxlan
;
203 vs
= vxlan_find_port(net
, port
);
207 hlist_for_each_entry_rcu(vxlan
, vni_head(vs
, id
), hlist
) {
208 if (vxlan
->default_dst
.remote_vni
== id
)
215 /* Fill in neighbour message in skbuff. */
216 static int vxlan_fdb_info(struct sk_buff
*skb
, struct vxlan_dev
*vxlan
,
217 const struct vxlan_fdb
*fdb
,
218 u32 portid
, u32 seq
, int type
, unsigned int flags
,
219 const struct vxlan_rdst
*rdst
)
221 unsigned long now
= jiffies
;
222 struct nda_cacheinfo ci
;
223 struct nlmsghdr
*nlh
;
225 bool send_ip
, send_eth
;
227 nlh
= nlmsg_put(skb
, portid
, seq
, type
, sizeof(*ndm
), flags
);
231 ndm
= nlmsg_data(nlh
);
232 memset(ndm
, 0, sizeof(*ndm
));
234 send_eth
= send_ip
= true;
236 if (type
== RTM_GETNEIGH
) {
237 ndm
->ndm_family
= AF_INET
;
238 send_ip
= rdst
->remote_ip
!= htonl(INADDR_ANY
);
239 send_eth
= !is_zero_ether_addr(fdb
->eth_addr
);
241 ndm
->ndm_family
= AF_BRIDGE
;
242 ndm
->ndm_state
= fdb
->state
;
243 ndm
->ndm_ifindex
= vxlan
->dev
->ifindex
;
244 ndm
->ndm_flags
= fdb
->flags
;
245 ndm
->ndm_type
= NDA_DST
;
247 if (send_eth
&& nla_put(skb
, NDA_LLADDR
, ETH_ALEN
, &fdb
->eth_addr
))
248 goto nla_put_failure
;
250 if (send_ip
&& nla_put_be32(skb
, NDA_DST
, rdst
->remote_ip
))
251 goto nla_put_failure
;
253 if (rdst
->remote_port
&& rdst
->remote_port
!= vxlan
->dst_port
&&
254 nla_put_be16(skb
, NDA_PORT
, rdst
->remote_port
))
255 goto nla_put_failure
;
256 if (rdst
->remote_vni
!= vxlan
->default_dst
.remote_vni
&&
257 nla_put_u32(skb
, NDA_VNI
, rdst
->remote_vni
))
258 goto nla_put_failure
;
259 if (rdst
->remote_ifindex
&&
260 nla_put_u32(skb
, NDA_IFINDEX
, rdst
->remote_ifindex
))
261 goto nla_put_failure
;
263 ci
.ndm_used
= jiffies_to_clock_t(now
- fdb
->used
);
264 ci
.ndm_confirmed
= 0;
265 ci
.ndm_updated
= jiffies_to_clock_t(now
- fdb
->updated
);
268 if (nla_put(skb
, NDA_CACHEINFO
, sizeof(ci
), &ci
))
269 goto nla_put_failure
;
271 return nlmsg_end(skb
, nlh
);
274 nlmsg_cancel(skb
, nlh
);
278 static inline size_t vxlan_nlmsg_size(void)
280 return NLMSG_ALIGN(sizeof(struct ndmsg
))
281 + nla_total_size(ETH_ALEN
) /* NDA_LLADDR */
282 + nla_total_size(sizeof(__be32
)) /* NDA_DST */
283 + nla_total_size(sizeof(__be16
)) /* NDA_PORT */
284 + nla_total_size(sizeof(__be32
)) /* NDA_VNI */
285 + nla_total_size(sizeof(__u32
)) /* NDA_IFINDEX */
286 + nla_total_size(sizeof(struct nda_cacheinfo
));
289 static void vxlan_fdb_notify(struct vxlan_dev
*vxlan
,
290 struct vxlan_fdb
*fdb
, int type
)
292 struct net
*net
= dev_net(vxlan
->dev
);
296 skb
= nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC
);
300 err
= vxlan_fdb_info(skb
, vxlan
, fdb
, 0, 0, type
, 0, first_remote(fdb
));
302 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
303 WARN_ON(err
== -EMSGSIZE
);
308 rtnl_notify(skb
, net
, 0, RTNLGRP_NEIGH
, NULL
, GFP_ATOMIC
);
312 rtnl_set_sk_err(net
, RTNLGRP_NEIGH
, err
);
315 static void vxlan_ip_miss(struct net_device
*dev
, __be32 ipa
)
317 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
318 struct vxlan_fdb f
= {
321 struct vxlan_rdst remote
= {
322 .remote_ip
= ipa
, /* goes to NDA_DST */
323 .remote_vni
= VXLAN_N_VID
,
326 INIT_LIST_HEAD(&f
.remotes
);
327 list_add_rcu(&remote
.list
, &f
.remotes
);
329 vxlan_fdb_notify(vxlan
, &f
, RTM_GETNEIGH
);
332 static void vxlan_fdb_miss(struct vxlan_dev
*vxlan
, const u8 eth_addr
[ETH_ALEN
])
334 struct vxlan_fdb f
= {
338 INIT_LIST_HEAD(&f
.remotes
);
339 memcpy(f
.eth_addr
, eth_addr
, ETH_ALEN
);
341 vxlan_fdb_notify(vxlan
, &f
, RTM_GETNEIGH
);
344 /* Hash Ethernet address */
345 static u32
eth_hash(const unsigned char *addr
)
347 u64 value
= get_unaligned((u64
*)addr
);
349 /* only want 6 bytes */
355 return hash_64(value
, FDB_HASH_BITS
);
358 /* Hash chain to use given mac address */
359 static inline struct hlist_head
*vxlan_fdb_head(struct vxlan_dev
*vxlan
,
362 return &vxlan
->fdb_head
[eth_hash(mac
)];
365 /* Look up Ethernet address in forwarding table */
366 static struct vxlan_fdb
*__vxlan_find_mac(struct vxlan_dev
*vxlan
,
370 struct hlist_head
*head
= vxlan_fdb_head(vxlan
, mac
);
373 hlist_for_each_entry_rcu(f
, head
, hlist
) {
374 if (compare_ether_addr(mac
, f
->eth_addr
) == 0)
381 static struct vxlan_fdb
*vxlan_find_mac(struct vxlan_dev
*vxlan
,
386 f
= __vxlan_find_mac(vxlan
, mac
);
393 /* caller should hold vxlan->hash_lock */
394 static struct vxlan_rdst
*vxlan_fdb_find_rdst(struct vxlan_fdb
*f
,
395 __be32 ip
, __be16 port
,
396 __u32 vni
, __u32 ifindex
)
398 struct vxlan_rdst
*rd
;
400 list_for_each_entry(rd
, &f
->remotes
, list
) {
401 if (rd
->remote_ip
== ip
&&
402 rd
->remote_port
== port
&&
403 rd
->remote_vni
== vni
&&
404 rd
->remote_ifindex
== ifindex
)
411 /* Replace destination of unicast mac */
412 static int vxlan_fdb_replace(struct vxlan_fdb
*f
,
413 __be32 ip
, __be16 port
, __u32 vni
, __u32 ifindex
)
415 struct vxlan_rdst
*rd
;
417 rd
= vxlan_fdb_find_rdst(f
, ip
, port
, vni
, ifindex
);
421 rd
= list_first_entry_or_null(&f
->remotes
, struct vxlan_rdst
, list
);
425 rd
->remote_port
= port
;
426 rd
->remote_vni
= vni
;
427 rd
->remote_ifindex
= ifindex
;
431 /* Add/update destinations for multicast */
432 static int vxlan_fdb_append(struct vxlan_fdb
*f
,
433 __be32 ip
, __be16 port
, __u32 vni
, __u32 ifindex
)
435 struct vxlan_rdst
*rd
;
437 rd
= vxlan_fdb_find_rdst(f
, ip
, port
, vni
, ifindex
);
441 rd
= kmalloc(sizeof(*rd
), GFP_ATOMIC
);
445 rd
->remote_port
= port
;
446 rd
->remote_vni
= vni
;
447 rd
->remote_ifindex
= ifindex
;
449 list_add_tail_rcu(&rd
->list
, &f
->remotes
);
454 /* Add new entry to forwarding table -- assumes lock held */
455 static int vxlan_fdb_create(struct vxlan_dev
*vxlan
,
456 const u8
*mac
, __be32 ip
,
457 __u16 state
, __u16 flags
,
458 __be16 port
, __u32 vni
, __u32 ifindex
,
464 f
= __vxlan_find_mac(vxlan
, mac
);
466 if (flags
& NLM_F_EXCL
) {
467 netdev_dbg(vxlan
->dev
,
468 "lost race to create %pM\n", mac
);
471 if (f
->state
!= state
) {
473 f
->updated
= jiffies
;
476 if (f
->flags
!= ndm_flags
) {
477 f
->flags
= ndm_flags
;
478 f
->updated
= jiffies
;
481 if ((flags
& NLM_F_REPLACE
)) {
482 /* Only change unicasts */
483 if (!(is_multicast_ether_addr(f
->eth_addr
) ||
484 is_zero_ether_addr(f
->eth_addr
))) {
485 int rc
= vxlan_fdb_replace(f
, ip
, port
, vni
,
494 if ((flags
& NLM_F_APPEND
) &&
495 (is_multicast_ether_addr(f
->eth_addr
) ||
496 is_zero_ether_addr(f
->eth_addr
))) {
497 int rc
= vxlan_fdb_append(f
, ip
, port
, vni
, ifindex
);
504 if (!(flags
& NLM_F_CREATE
))
507 if (vxlan
->addrmax
&& vxlan
->addrcnt
>= vxlan
->addrmax
)
510 /* Disallow replace to add a multicast entry */
511 if ((flags
& NLM_F_REPLACE
) &&
512 (is_multicast_ether_addr(mac
) || is_zero_ether_addr(mac
)))
515 netdev_dbg(vxlan
->dev
, "add %pM -> %pI4\n", mac
, &ip
);
516 f
= kmalloc(sizeof(*f
), GFP_ATOMIC
);
522 f
->flags
= ndm_flags
;
523 f
->updated
= f
->used
= jiffies
;
524 INIT_LIST_HEAD(&f
->remotes
);
525 memcpy(f
->eth_addr
, mac
, ETH_ALEN
);
527 vxlan_fdb_append(f
, ip
, port
, vni
, ifindex
);
530 hlist_add_head_rcu(&f
->hlist
,
531 vxlan_fdb_head(vxlan
, mac
));
535 vxlan_fdb_notify(vxlan
, f
, RTM_NEWNEIGH
);
540 static void vxlan_fdb_free_rdst(struct rcu_head
*head
)
542 struct vxlan_rdst
*rd
= container_of(head
, struct vxlan_rdst
, rcu
);
546 static void vxlan_fdb_free(struct rcu_head
*head
)
548 struct vxlan_fdb
*f
= container_of(head
, struct vxlan_fdb
, rcu
);
549 struct vxlan_rdst
*rd
, *nd
;
551 list_for_each_entry_safe(rd
, nd
, &f
->remotes
, list
)
556 static void vxlan_fdb_destroy(struct vxlan_dev
*vxlan
, struct vxlan_fdb
*f
)
558 netdev_dbg(vxlan
->dev
,
559 "delete %pM\n", f
->eth_addr
);
562 vxlan_fdb_notify(vxlan
, f
, RTM_DELNEIGH
);
564 hlist_del_rcu(&f
->hlist
);
565 call_rcu(&f
->rcu
, vxlan_fdb_free
);
568 static int vxlan_fdb_parse(struct nlattr
*tb
[], struct vxlan_dev
*vxlan
,
569 __be32
*ip
, __be16
*port
, u32
*vni
, u32
*ifindex
)
571 struct net
*net
= dev_net(vxlan
->dev
);
574 if (nla_len(tb
[NDA_DST
]) != sizeof(__be32
))
575 return -EAFNOSUPPORT
;
577 *ip
= nla_get_be32(tb
[NDA_DST
]);
579 *ip
= htonl(INADDR_ANY
);
583 if (nla_len(tb
[NDA_PORT
]) != sizeof(__be16
))
585 *port
= nla_get_be16(tb
[NDA_PORT
]);
587 *port
= vxlan
->dst_port
;
591 if (nla_len(tb
[NDA_VNI
]) != sizeof(u32
))
593 *vni
= nla_get_u32(tb
[NDA_VNI
]);
595 *vni
= vxlan
->default_dst
.remote_vni
;
598 if (tb
[NDA_IFINDEX
]) {
599 struct net_device
*tdev
;
601 if (nla_len(tb
[NDA_IFINDEX
]) != sizeof(u32
))
603 *ifindex
= nla_get_u32(tb
[NDA_IFINDEX
]);
604 tdev
= dev_get_by_index(net
, *ifindex
);
606 return -EADDRNOTAVAIL
;
615 /* Add static entry (via netlink) */
616 static int vxlan_fdb_add(struct ndmsg
*ndm
, struct nlattr
*tb
[],
617 struct net_device
*dev
,
618 const unsigned char *addr
, u16 flags
)
620 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
621 /* struct net *net = dev_net(vxlan->dev); */
627 if (!(ndm
->ndm_state
& (NUD_PERMANENT
|NUD_REACHABLE
))) {
628 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
633 if (tb
[NDA_DST
] == NULL
)
636 err
= vxlan_fdb_parse(tb
, vxlan
, &ip
, &port
, &vni
, &ifindex
);
640 spin_lock_bh(&vxlan
->hash_lock
);
641 err
= vxlan_fdb_create(vxlan
, addr
, ip
, ndm
->ndm_state
, flags
,
642 port
, vni
, ifindex
, ndm
->ndm_flags
);
643 spin_unlock_bh(&vxlan
->hash_lock
);
648 /* Delete entry (via netlink) */
649 static int vxlan_fdb_delete(struct ndmsg
*ndm
, struct nlattr
*tb
[],
650 struct net_device
*dev
,
651 const unsigned char *addr
)
653 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
655 struct vxlan_rdst
*rd
= NULL
;
661 err
= vxlan_fdb_parse(tb
, vxlan
, &ip
, &port
, &vni
, &ifindex
);
667 spin_lock_bh(&vxlan
->hash_lock
);
668 f
= vxlan_find_mac(vxlan
, addr
);
672 if (ip
!= htonl(INADDR_ANY
)) {
673 rd
= vxlan_fdb_find_rdst(f
, ip
, port
, vni
, ifindex
);
680 /* remove a destination if it's not the only one on the list,
681 * otherwise destroy the fdb entry
683 if (rd
&& !list_is_singular(&f
->remotes
)) {
684 list_del_rcu(&rd
->list
);
685 call_rcu(&rd
->rcu
, vxlan_fdb_free_rdst
);
689 vxlan_fdb_destroy(vxlan
, f
);
692 spin_unlock_bh(&vxlan
->hash_lock
);
697 /* Dump forwarding table */
698 static int vxlan_fdb_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
,
699 struct net_device
*dev
, int idx
)
701 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
704 for (h
= 0; h
< FDB_HASH_SIZE
; ++h
) {
708 hlist_for_each_entry_rcu(f
, &vxlan
->fdb_head
[h
], hlist
) {
709 struct vxlan_rdst
*rd
;
711 if (idx
< cb
->args
[0])
714 list_for_each_entry_rcu(rd
, &f
->remotes
, list
) {
715 err
= vxlan_fdb_info(skb
, vxlan
, f
,
716 NETLINK_CB(cb
->skb
).portid
,
731 /* Watch incoming packets to learn mapping between Ethernet address
732 * and Tunnel endpoint.
733 * Return true if packet is bogus and should be droppped.
735 static bool vxlan_snoop(struct net_device
*dev
,
736 __be32 src_ip
, const u8
*src_mac
)
738 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
741 f
= vxlan_find_mac(vxlan
, src_mac
);
743 struct vxlan_rdst
*rdst
= first_remote(f
);
745 if (likely(rdst
->remote_ip
== src_ip
))
748 /* Don't migrate static entries, drop packets */
749 if (f
->state
& NUD_NOARP
)
754 "%pM migrated from %pI4 to %pI4\n",
755 src_mac
, &rdst
->remote_ip
, &src_ip
);
757 rdst
->remote_ip
= src_ip
;
758 f
->updated
= jiffies
;
759 vxlan_fdb_notify(vxlan
, f
, RTM_NEWNEIGH
);
761 /* learned new entry */
762 spin_lock(&vxlan
->hash_lock
);
764 /* close off race between vxlan_flush and incoming packets */
765 if (netif_running(dev
))
766 vxlan_fdb_create(vxlan
, src_mac
, src_ip
,
768 NLM_F_EXCL
|NLM_F_CREATE
,
770 vxlan
->default_dst
.remote_vni
,
772 spin_unlock(&vxlan
->hash_lock
);
778 /* See if multicast group is already in use by other ID */
779 static bool vxlan_group_used(struct vxlan_net
*vn
, __be32 remote_ip
)
781 struct vxlan_dev
*vxlan
;
783 list_for_each_entry(vxlan
, &vn
->vxlan_list
, next
) {
784 if (!netif_running(vxlan
->dev
))
787 if (vxlan
->default_dst
.remote_ip
== remote_ip
)
794 static void vxlan_sock_hold(struct vxlan_sock
*vs
)
796 atomic_inc(&vs
->refcnt
);
799 static void vxlan_sock_release(struct vxlan_net
*vn
, struct vxlan_sock
*vs
)
801 if (!atomic_dec_and_test(&vs
->refcnt
))
804 spin_lock(&vn
->sock_lock
);
805 hlist_del_rcu(&vs
->hlist
);
806 spin_unlock(&vn
->sock_lock
);
808 queue_work(vxlan_wq
, &vs
->del_work
);
811 /* Callback to update multicast group membership when first VNI on
812 * multicast asddress is brought up
813 * Done as workqueue because ip_mc_join_group acquires RTNL.
815 static void vxlan_igmp_join(struct work_struct
*work
)
817 struct vxlan_dev
*vxlan
= container_of(work
, struct vxlan_dev
, igmp_join
);
818 struct vxlan_net
*vn
= net_generic(dev_net(vxlan
->dev
), vxlan_net_id
);
819 struct vxlan_sock
*vs
= vxlan
->vn_sock
;
820 struct sock
*sk
= vs
->sock
->sk
;
821 struct ip_mreqn mreq
= {
822 .imr_multiaddr
.s_addr
= vxlan
->default_dst
.remote_ip
,
823 .imr_ifindex
= vxlan
->default_dst
.remote_ifindex
,
827 ip_mc_join_group(sk
, &mreq
);
830 vxlan_sock_release(vn
, vs
);
834 /* Inverse of vxlan_igmp_join when last VNI is brought down */
835 static void vxlan_igmp_leave(struct work_struct
*work
)
837 struct vxlan_dev
*vxlan
= container_of(work
, struct vxlan_dev
, igmp_leave
);
838 struct vxlan_net
*vn
= net_generic(dev_net(vxlan
->dev
), vxlan_net_id
);
839 struct vxlan_sock
*vs
= vxlan
->vn_sock
;
840 struct sock
*sk
= vs
->sock
->sk
;
841 struct ip_mreqn mreq
= {
842 .imr_multiaddr
.s_addr
= vxlan
->default_dst
.remote_ip
,
843 .imr_ifindex
= vxlan
->default_dst
.remote_ifindex
,
847 ip_mc_leave_group(sk
, &mreq
);
850 vxlan_sock_release(vn
, vs
);
854 /* Callback from net/ipv4/udp.c to receive packets */
855 static int vxlan_udp_encap_recv(struct sock
*sk
, struct sk_buff
*skb
)
858 struct vxlanhdr
*vxh
;
859 struct vxlan_dev
*vxlan
;
860 struct pcpu_tstats
*stats
;
865 /* pop off outer UDP header */
866 __skb_pull(skb
, sizeof(struct udphdr
));
868 /* Need Vxlan and inner Ethernet header to be present */
869 if (!pskb_may_pull(skb
, sizeof(struct vxlanhdr
)))
872 /* Drop packets with reserved bits set */
873 vxh
= (struct vxlanhdr
*) skb
->data
;
874 if (vxh
->vx_flags
!= htonl(VXLAN_FLAGS
) ||
875 (vxh
->vx_vni
& htonl(0xff))) {
876 netdev_dbg(skb
->dev
, "invalid vxlan flags=%#x vni=%#x\n",
877 ntohl(vxh
->vx_flags
), ntohl(vxh
->vx_vni
));
881 __skb_pull(skb
, sizeof(struct vxlanhdr
));
883 /* Is this VNI defined? */
884 vni
= ntohl(vxh
->vx_vni
) >> 8;
885 port
= inet_sk(sk
)->inet_sport
;
886 vxlan
= vxlan_find_vni(sock_net(sk
), vni
, port
);
888 netdev_dbg(skb
->dev
, "unknown vni %d port %u\n",
893 if (!pskb_may_pull(skb
, ETH_HLEN
)) {
894 vxlan
->dev
->stats
.rx_length_errors
++;
895 vxlan
->dev
->stats
.rx_errors
++;
899 skb_reset_mac_header(skb
);
901 /* Re-examine inner Ethernet packet */
903 skb
->protocol
= eth_type_trans(skb
, vxlan
->dev
);
905 /* Ignore packet loops (and multicast echo) */
906 if (compare_ether_addr(eth_hdr(skb
)->h_source
,
907 vxlan
->dev
->dev_addr
) == 0)
910 if ((vxlan
->flags
& VXLAN_F_LEARN
) &&
911 vxlan_snoop(skb
->dev
, oip
->saddr
, eth_hdr(skb
)->h_source
))
914 __skb_tunnel_rx(skb
, vxlan
->dev
);
915 skb_reset_network_header(skb
);
917 /* If the NIC driver gave us an encapsulated packet with
918 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
919 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
920 * for us. Otherwise force the upper layers to verify it.
922 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
|| !skb
->encapsulation
||
923 !(vxlan
->dev
->features
& NETIF_F_RXCSUM
))
924 skb
->ip_summed
= CHECKSUM_NONE
;
926 skb
->encapsulation
= 0;
928 err
= IP_ECN_decapsulate(oip
, skb
);
931 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
932 &oip
->saddr
, oip
->tos
);
934 ++vxlan
->dev
->stats
.rx_frame_errors
;
935 ++vxlan
->dev
->stats
.rx_errors
;
940 stats
= this_cpu_ptr(vxlan
->dev
->tstats
);
941 u64_stats_update_begin(&stats
->syncp
);
943 stats
->rx_bytes
+= skb
->len
;
944 u64_stats_update_end(&stats
->syncp
);
950 /* Put UDP header back */
951 __skb_push(skb
, sizeof(struct udphdr
));
955 /* Consume bad packet */
960 static int arp_reduce(struct net_device
*dev
, struct sk_buff
*skb
)
962 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
968 if (dev
->flags
& IFF_NOARP
)
971 if (!pskb_may_pull(skb
, arp_hdr_len(dev
))) {
972 dev
->stats
.tx_dropped
++;
977 if ((parp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
978 parp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
979 parp
->ar_pro
!= htons(ETH_P_IP
) ||
980 parp
->ar_op
!= htons(ARPOP_REQUEST
) ||
981 parp
->ar_hln
!= dev
->addr_len
||
984 arpptr
= (u8
*)parp
+ sizeof(struct arphdr
);
986 arpptr
+= dev
->addr_len
; /* sha */
987 memcpy(&sip
, arpptr
, sizeof(sip
));
988 arpptr
+= sizeof(sip
);
989 arpptr
+= dev
->addr_len
; /* tha */
990 memcpy(&tip
, arpptr
, sizeof(tip
));
992 if (ipv4_is_loopback(tip
) ||
993 ipv4_is_multicast(tip
))
996 n
= neigh_lookup(&arp_tbl
, &tip
, dev
);
1000 struct sk_buff
*reply
;
1002 if (!(n
->nud_state
& NUD_CONNECTED
)) {
1007 f
= vxlan_find_mac(vxlan
, n
->ha
);
1008 if (f
&& first_remote(f
)->remote_ip
== htonl(INADDR_ANY
)) {
1009 /* bridge-local neighbor */
1014 reply
= arp_create(ARPOP_REPLY
, ETH_P_ARP
, sip
, dev
, tip
, sha
,
1019 skb_reset_mac_header(reply
);
1020 __skb_pull(reply
, skb_network_offset(reply
));
1021 reply
->ip_summed
= CHECKSUM_UNNECESSARY
;
1022 reply
->pkt_type
= PACKET_HOST
;
1024 if (netif_rx_ni(reply
) == NET_RX_DROP
)
1025 dev
->stats
.rx_dropped
++;
1026 } else if (vxlan
->flags
& VXLAN_F_L3MISS
)
1027 vxlan_ip_miss(dev
, tip
);
1030 return NETDEV_TX_OK
;
1033 static bool route_shortcircuit(struct net_device
*dev
, struct sk_buff
*skb
)
1035 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1036 struct neighbour
*n
;
1039 if (is_multicast_ether_addr(eth_hdr(skb
)->h_dest
))
1043 switch (ntohs(eth_hdr(skb
)->h_proto
)) {
1045 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
1048 n
= neigh_lookup(&arp_tbl
, &pip
->daddr
, dev
);
1057 diff
= compare_ether_addr(eth_hdr(skb
)->h_dest
, n
->ha
) != 0;
1059 memcpy(eth_hdr(skb
)->h_source
, eth_hdr(skb
)->h_dest
,
1061 memcpy(eth_hdr(skb
)->h_dest
, n
->ha
, dev
->addr_len
);
1065 } else if (vxlan
->flags
& VXLAN_F_L3MISS
)
1066 vxlan_ip_miss(dev
, pip
->daddr
);
1070 static void vxlan_sock_put(struct sk_buff
*skb
)
1075 /* On transmit, associate with the tunnel socket */
1076 static void vxlan_set_owner(struct net_device
*dev
, struct sk_buff
*skb
)
1078 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1079 struct sock
*sk
= vxlan
->vn_sock
->sock
->sk
;
1084 skb
->destructor
= vxlan_sock_put
;
1087 /* Compute source port for outgoing packet
1088 * first choice to use L4 flow hash since it will spread
1089 * better and maybe available from hardware
1090 * secondary choice is to use jhash on the Ethernet header
1092 static __be16
vxlan_src_port(const struct vxlan_dev
*vxlan
, struct sk_buff
*skb
)
1094 unsigned int range
= (vxlan
->port_max
- vxlan
->port_min
) + 1;
1097 hash
= skb_get_rxhash(skb
);
1099 hash
= jhash(skb
->data
, 2 * ETH_ALEN
,
1100 (__force u32
) skb
->protocol
);
1102 return htons((((u64
) hash
* range
) >> 32) + vxlan
->port_min
);
1105 static int handle_offloads(struct sk_buff
*skb
)
1107 if (skb_is_gso(skb
)) {
1108 int err
= skb_unclone(skb
, GFP_ATOMIC
);
1112 skb_shinfo(skb
)->gso_type
|= SKB_GSO_UDP_TUNNEL
;
1113 } else if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
1114 skb
->ip_summed
= CHECKSUM_NONE
;
1119 /* Bypass encapsulation if the destination is local */
1120 static void vxlan_encap_bypass(struct sk_buff
*skb
, struct vxlan_dev
*src_vxlan
,
1121 struct vxlan_dev
*dst_vxlan
)
1123 struct pcpu_tstats
*tx_stats
= this_cpu_ptr(src_vxlan
->dev
->tstats
);
1124 struct pcpu_tstats
*rx_stats
= this_cpu_ptr(dst_vxlan
->dev
->tstats
);
1126 skb
->pkt_type
= PACKET_HOST
;
1127 skb
->encapsulation
= 0;
1128 skb
->dev
= dst_vxlan
->dev
;
1129 __skb_pull(skb
, skb_network_offset(skb
));
1131 if (dst_vxlan
->flags
& VXLAN_F_LEARN
)
1132 vxlan_snoop(skb
->dev
, htonl(INADDR_LOOPBACK
),
1133 eth_hdr(skb
)->h_source
);
1135 u64_stats_update_begin(&tx_stats
->syncp
);
1136 tx_stats
->tx_packets
++;
1137 tx_stats
->tx_bytes
+= skb
->len
;
1138 u64_stats_update_end(&tx_stats
->syncp
);
1140 if (netif_rx(skb
) == NET_RX_SUCCESS
) {
1141 u64_stats_update_begin(&rx_stats
->syncp
);
1142 rx_stats
->rx_packets
++;
1143 rx_stats
->rx_bytes
+= skb
->len
;
1144 u64_stats_update_end(&rx_stats
->syncp
);
1146 skb
->dev
->stats
.rx_dropped
++;
1150 static void vxlan_xmit_one(struct sk_buff
*skb
, struct net_device
*dev
,
1151 struct vxlan_rdst
*rdst
, bool did_rsc
)
1153 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1155 const struct iphdr
*old_iph
;
1156 struct vxlanhdr
*vxh
;
1160 __be16 src_port
, dst_port
;
1166 dst_port
= rdst
->remote_port
? rdst
->remote_port
: vxlan
->dst_port
;
1167 vni
= rdst
->remote_vni
;
1168 dst
= rdst
->remote_ip
;
1172 /* short-circuited back to local bridge */
1173 vxlan_encap_bypass(skb
, vxlan
, vxlan
);
1179 if (!skb
->encapsulation
) {
1180 skb_reset_inner_headers(skb
);
1181 skb
->encapsulation
= 1;
1184 /* Need space for new headers (invalidates iph ptr) */
1185 if (skb_cow_head(skb
, VXLAN_HEADROOM
))
1188 old_iph
= ip_hdr(skb
);
1191 if (!ttl
&& IN_MULTICAST(ntohl(dst
)))
1196 tos
= ip_tunnel_get_dsfield(old_iph
, skb
);
1198 src_port
= vxlan_src_port(vxlan
, skb
);
1200 memset(&fl4
, 0, sizeof(fl4
));
1201 fl4
.flowi4_oif
= rdst
->remote_ifindex
;
1202 fl4
.flowi4_tos
= RT_TOS(tos
);
1204 fl4
.saddr
= vxlan
->saddr
;
1206 rt
= ip_route_output_key(dev_net(dev
), &fl4
);
1208 netdev_dbg(dev
, "no route to %pI4\n", &dst
);
1209 dev
->stats
.tx_carrier_errors
++;
1213 if (rt
->dst
.dev
== dev
) {
1214 netdev_dbg(dev
, "circular route to %pI4\n", &dst
);
1216 dev
->stats
.collisions
++;
1220 /* Bypass encapsulation if the destination is local */
1221 if (rt
->rt_flags
& RTCF_LOCAL
&&
1222 !(rt
->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))) {
1223 struct vxlan_dev
*dst_vxlan
;
1226 dst_vxlan
= vxlan_find_vni(dev_net(dev
), vni
, dst_port
);
1229 vxlan_encap_bypass(skb
, vxlan
, dst_vxlan
);
1232 vxh
= (struct vxlanhdr
*) __skb_push(skb
, sizeof(*vxh
));
1233 vxh
->vx_flags
= htonl(VXLAN_FLAGS
);
1234 vxh
->vx_vni
= htonl(vni
<< 8);
1236 __skb_push(skb
, sizeof(*uh
));
1237 skb_reset_transport_header(skb
);
1240 uh
->dest
= dst_port
;
1241 uh
->source
= src_port
;
1243 uh
->len
= htons(skb
->len
);
1246 vxlan_set_owner(dev
, skb
);
1248 if (handle_offloads(skb
))
1251 tos
= ip_tunnel_ecn_encap(tos
, old_iph
, skb
);
1252 ttl
= ttl
? : ip4_dst_hoplimit(&rt
->dst
);
1254 err
= iptunnel_xmit(dev_net(dev
), rt
, skb
, fl4
.saddr
, dst
,
1255 IPPROTO_UDP
, tos
, ttl
, df
);
1256 iptunnel_xmit_stats(err
, &dev
->stats
, dev
->tstats
);
1261 dev
->stats
.tx_dropped
++;
1265 dev
->stats
.tx_errors
++;
1270 /* Transmit local packets over Vxlan
1272 * Outer IP header inherits ECN and DF from inner header.
1273 * Outer UDP destination is the VXLAN assigned port.
1274 * source port is based on hash of flow
1276 static netdev_tx_t
vxlan_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1278 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1280 bool did_rsc
= false;
1281 struct vxlan_rdst
*rdst
;
1282 struct vxlan_fdb
*f
;
1284 skb_reset_mac_header(skb
);
1287 if ((vxlan
->flags
& VXLAN_F_PROXY
) && ntohs(eth
->h_proto
) == ETH_P_ARP
)
1288 return arp_reduce(dev
, skb
);
1290 f
= vxlan_find_mac(vxlan
, eth
->h_dest
);
1293 if (f
&& (f
->flags
& NTF_ROUTER
) && (vxlan
->flags
& VXLAN_F_RSC
) &&
1294 ntohs(eth
->h_proto
) == ETH_P_IP
) {
1295 did_rsc
= route_shortcircuit(dev
, skb
);
1297 f
= vxlan_find_mac(vxlan
, eth
->h_dest
);
1301 f
= vxlan_find_mac(vxlan
, all_zeros_mac
);
1303 if ((vxlan
->flags
& VXLAN_F_L2MISS
) &&
1304 !is_multicast_ether_addr(eth
->h_dest
))
1305 vxlan_fdb_miss(vxlan
, eth
->h_dest
);
1307 dev
->stats
.tx_dropped
++;
1309 return NETDEV_TX_OK
;
1313 list_for_each_entry_rcu(rdst
, &f
->remotes
, list
) {
1314 struct sk_buff
*skb1
;
1316 skb1
= skb_clone(skb
, GFP_ATOMIC
);
1318 vxlan_xmit_one(skb1
, dev
, rdst
, did_rsc
);
1322 return NETDEV_TX_OK
;
1325 /* Walk the forwarding table and purge stale entries */
1326 static void vxlan_cleanup(unsigned long arg
)
1328 struct vxlan_dev
*vxlan
= (struct vxlan_dev
*) arg
;
1329 unsigned long next_timer
= jiffies
+ FDB_AGE_INTERVAL
;
1332 if (!netif_running(vxlan
->dev
))
1335 spin_lock_bh(&vxlan
->hash_lock
);
1336 for (h
= 0; h
< FDB_HASH_SIZE
; ++h
) {
1337 struct hlist_node
*p
, *n
;
1338 hlist_for_each_safe(p
, n
, &vxlan
->fdb_head
[h
]) {
1340 = container_of(p
, struct vxlan_fdb
, hlist
);
1341 unsigned long timeout
;
1343 if (f
->state
& NUD_PERMANENT
)
1346 timeout
= f
->used
+ vxlan
->age_interval
* HZ
;
1347 if (time_before_eq(timeout
, jiffies
)) {
1348 netdev_dbg(vxlan
->dev
,
1349 "garbage collect %pM\n",
1351 f
->state
= NUD_STALE
;
1352 vxlan_fdb_destroy(vxlan
, f
);
1353 } else if (time_before(timeout
, next_timer
))
1354 next_timer
= timeout
;
1357 spin_unlock_bh(&vxlan
->hash_lock
);
1359 mod_timer(&vxlan
->age_timer
, next_timer
);
1362 /* Setup stats when device is created */
1363 static int vxlan_init(struct net_device
*dev
)
1365 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1366 struct vxlan_net
*vn
= net_generic(dev_net(dev
), vxlan_net_id
);
1367 struct vxlan_sock
*vs
;
1368 __u32 vni
= vxlan
->default_dst
.remote_vni
;
1370 dev
->tstats
= alloc_percpu(struct pcpu_tstats
);
1374 spin_lock(&vn
->sock_lock
);
1375 vs
= vxlan_find_port(dev_net(dev
), vxlan
->dst_port
);
1377 /* If we have a socket with same port already, reuse it */
1378 atomic_inc(&vs
->refcnt
);
1379 vxlan
->vn_sock
= vs
;
1380 hlist_add_head_rcu(&vxlan
->hlist
, vni_head(vs
, vni
));
1382 /* otherwise make new socket outside of RTNL */
1384 queue_work(vxlan_wq
, &vxlan
->sock_work
);
1386 spin_unlock(&vn
->sock_lock
);
1391 static void vxlan_fdb_delete_default(struct vxlan_dev
*vxlan
)
1393 struct vxlan_fdb
*f
;
1395 spin_lock_bh(&vxlan
->hash_lock
);
1396 f
= __vxlan_find_mac(vxlan
, all_zeros_mac
);
1398 vxlan_fdb_destroy(vxlan
, f
);
1399 spin_unlock_bh(&vxlan
->hash_lock
);
1402 static void vxlan_uninit(struct net_device
*dev
)
1404 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1405 struct vxlan_net
*vn
= net_generic(dev_net(dev
), vxlan_net_id
);
1406 struct vxlan_sock
*vs
= vxlan
->vn_sock
;
1408 vxlan_fdb_delete_default(vxlan
);
1411 vxlan_sock_release(vn
, vs
);
1412 free_percpu(dev
->tstats
);
1415 /* Start ageing timer and join group when device is brought up */
1416 static int vxlan_open(struct net_device
*dev
)
1418 struct vxlan_net
*vn
= net_generic(dev_net(dev
), vxlan_net_id
);
1419 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1420 struct vxlan_sock
*vs
= vxlan
->vn_sock
;
1422 /* socket hasn't been created */
1426 if (IN_MULTICAST(ntohl(vxlan
->default_dst
.remote_ip
)) &&
1427 ! vxlan_group_used(vn
, vxlan
->default_dst
.remote_ip
)) {
1428 vxlan_sock_hold(vs
);
1430 queue_work(vxlan_wq
, &vxlan
->igmp_join
);
1433 if (vxlan
->age_interval
)
1434 mod_timer(&vxlan
->age_timer
, jiffies
+ FDB_AGE_INTERVAL
);
1439 /* Purge the forwarding table */
1440 static void vxlan_flush(struct vxlan_dev
*vxlan
)
1444 spin_lock_bh(&vxlan
->hash_lock
);
1445 for (h
= 0; h
< FDB_HASH_SIZE
; ++h
) {
1446 struct hlist_node
*p
, *n
;
1447 hlist_for_each_safe(p
, n
, &vxlan
->fdb_head
[h
]) {
1449 = container_of(p
, struct vxlan_fdb
, hlist
);
1450 /* the all_zeros_mac entry is deleted at vxlan_uninit */
1451 if (!is_zero_ether_addr(f
->eth_addr
))
1452 vxlan_fdb_destroy(vxlan
, f
);
1455 spin_unlock_bh(&vxlan
->hash_lock
);
1458 /* Cleanup timer and forwarding table on shutdown */
1459 static int vxlan_stop(struct net_device
*dev
)
1461 struct vxlan_net
*vn
= net_generic(dev_net(dev
), vxlan_net_id
);
1462 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1463 struct vxlan_sock
*vs
= vxlan
->vn_sock
;
1465 if (vs
&& IN_MULTICAST(ntohl(vxlan
->default_dst
.remote_ip
)) &&
1466 ! vxlan_group_used(vn
, vxlan
->default_dst
.remote_ip
)) {
1467 vxlan_sock_hold(vs
);
1469 queue_work(vxlan_wq
, &vxlan
->igmp_leave
);
1472 del_timer_sync(&vxlan
->age_timer
);
1479 /* Stub, nothing needs to be done. */
1480 static void vxlan_set_multicast_list(struct net_device
*dev
)
1484 static const struct net_device_ops vxlan_netdev_ops
= {
1485 .ndo_init
= vxlan_init
,
1486 .ndo_uninit
= vxlan_uninit
,
1487 .ndo_open
= vxlan_open
,
1488 .ndo_stop
= vxlan_stop
,
1489 .ndo_start_xmit
= vxlan_xmit
,
1490 .ndo_get_stats64
= ip_tunnel_get_stats64
,
1491 .ndo_set_rx_mode
= vxlan_set_multicast_list
,
1492 .ndo_change_mtu
= eth_change_mtu
,
1493 .ndo_validate_addr
= eth_validate_addr
,
1494 .ndo_set_mac_address
= eth_mac_addr
,
1495 .ndo_fdb_add
= vxlan_fdb_add
,
1496 .ndo_fdb_del
= vxlan_fdb_delete
,
1497 .ndo_fdb_dump
= vxlan_fdb_dump
,
1500 /* Info for udev, that this is a virtual tunnel endpoint */
1501 static struct device_type vxlan_type
= {
1505 /* Initialize the device structure. */
1506 static void vxlan_setup(struct net_device
*dev
)
1508 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1512 eth_hw_addr_random(dev
);
1514 dev
->hard_header_len
= ETH_HLEN
+ VXLAN_HEADROOM
;
1516 dev
->netdev_ops
= &vxlan_netdev_ops
;
1517 dev
->destructor
= free_netdev
;
1518 SET_NETDEV_DEVTYPE(dev
, &vxlan_type
);
1520 dev
->tx_queue_len
= 0;
1521 dev
->features
|= NETIF_F_LLTX
;
1522 dev
->features
|= NETIF_F_NETNS_LOCAL
;
1523 dev
->features
|= NETIF_F_SG
| NETIF_F_HW_CSUM
;
1524 dev
->features
|= NETIF_F_RXCSUM
;
1525 dev
->features
|= NETIF_F_GSO_SOFTWARE
;
1527 dev
->hw_features
|= NETIF_F_SG
| NETIF_F_HW_CSUM
| NETIF_F_RXCSUM
;
1528 dev
->hw_features
|= NETIF_F_GSO_SOFTWARE
;
1529 dev
->priv_flags
&= ~IFF_XMIT_DST_RELEASE
;
1530 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1532 INIT_LIST_HEAD(&vxlan
->next
);
1533 spin_lock_init(&vxlan
->hash_lock
);
1534 INIT_WORK(&vxlan
->igmp_join
, vxlan_igmp_join
);
1535 INIT_WORK(&vxlan
->igmp_leave
, vxlan_igmp_leave
);
1536 INIT_WORK(&vxlan
->sock_work
, vxlan_sock_work
);
1538 init_timer_deferrable(&vxlan
->age_timer
);
1539 vxlan
->age_timer
.function
= vxlan_cleanup
;
1540 vxlan
->age_timer
.data
= (unsigned long) vxlan
;
1542 inet_get_local_port_range(&low
, &high
);
1543 vxlan
->port_min
= low
;
1544 vxlan
->port_max
= high
;
1545 vxlan
->dst_port
= htons(vxlan_port
);
1549 for (h
= 0; h
< FDB_HASH_SIZE
; ++h
)
1550 INIT_HLIST_HEAD(&vxlan
->fdb_head
[h
]);
1553 static const struct nla_policy vxlan_policy
[IFLA_VXLAN_MAX
+ 1] = {
1554 [IFLA_VXLAN_ID
] = { .type
= NLA_U32
},
1555 [IFLA_VXLAN_GROUP
] = { .len
= FIELD_SIZEOF(struct iphdr
, daddr
) },
1556 [IFLA_VXLAN_LINK
] = { .type
= NLA_U32
},
1557 [IFLA_VXLAN_LOCAL
] = { .len
= FIELD_SIZEOF(struct iphdr
, saddr
) },
1558 [IFLA_VXLAN_TOS
] = { .type
= NLA_U8
},
1559 [IFLA_VXLAN_TTL
] = { .type
= NLA_U8
},
1560 [IFLA_VXLAN_LEARNING
] = { .type
= NLA_U8
},
1561 [IFLA_VXLAN_AGEING
] = { .type
= NLA_U32
},
1562 [IFLA_VXLAN_LIMIT
] = { .type
= NLA_U32
},
1563 [IFLA_VXLAN_PORT_RANGE
] = { .len
= sizeof(struct ifla_vxlan_port_range
) },
1564 [IFLA_VXLAN_PROXY
] = { .type
= NLA_U8
},
1565 [IFLA_VXLAN_RSC
] = { .type
= NLA_U8
},
1566 [IFLA_VXLAN_L2MISS
] = { .type
= NLA_U8
},
1567 [IFLA_VXLAN_L3MISS
] = { .type
= NLA_U8
},
1568 [IFLA_VXLAN_PORT
] = { .type
= NLA_U16
},
1571 static int vxlan_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1573 if (tb
[IFLA_ADDRESS
]) {
1574 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
) {
1575 pr_debug("invalid link address (not ethernet)\n");
1579 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
]))) {
1580 pr_debug("invalid all zero ethernet address\n");
1581 return -EADDRNOTAVAIL
;
1588 if (data
[IFLA_VXLAN_ID
]) {
1589 __u32 id
= nla_get_u32(data
[IFLA_VXLAN_ID
]);
1590 if (id
>= VXLAN_VID_MASK
)
1594 if (data
[IFLA_VXLAN_PORT_RANGE
]) {
1595 const struct ifla_vxlan_port_range
*p
1596 = nla_data(data
[IFLA_VXLAN_PORT_RANGE
]);
1598 if (ntohs(p
->high
) < ntohs(p
->low
)) {
1599 pr_debug("port range %u .. %u not valid\n",
1600 ntohs(p
->low
), ntohs(p
->high
));
1608 static void vxlan_get_drvinfo(struct net_device
*netdev
,
1609 struct ethtool_drvinfo
*drvinfo
)
1611 strlcpy(drvinfo
->version
, VXLAN_VERSION
, sizeof(drvinfo
->version
));
1612 strlcpy(drvinfo
->driver
, "vxlan", sizeof(drvinfo
->driver
));
1615 static const struct ethtool_ops vxlan_ethtool_ops
= {
1616 .get_drvinfo
= vxlan_get_drvinfo
,
1617 .get_link
= ethtool_op_get_link
,
1620 static void vxlan_del_work(struct work_struct
*work
)
1622 struct vxlan_sock
*vs
= container_of(work
, struct vxlan_sock
, del_work
);
1624 sk_release_kernel(vs
->sock
->sk
);
1628 static struct vxlan_sock
*vxlan_socket_create(struct net
*net
, __be16 port
)
1630 struct vxlan_sock
*vs
;
1632 struct sockaddr_in vxlan_addr
= {
1633 .sin_family
= AF_INET
,
1634 .sin_addr
.s_addr
= htonl(INADDR_ANY
),
1640 vs
= kmalloc(sizeof(*vs
), GFP_KERNEL
);
1642 return ERR_PTR(-ENOMEM
);
1644 for (h
= 0; h
< VNI_HASH_SIZE
; ++h
)
1645 INIT_HLIST_HEAD(&vs
->vni_list
[h
]);
1647 INIT_WORK(&vs
->del_work
, vxlan_del_work
);
1649 /* Create UDP socket for encapsulation receive. */
1650 rc
= sock_create_kern(AF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &vs
->sock
);
1652 pr_debug("UDP socket create failed\n");
1657 /* Put in proper namespace */
1659 sk_change_net(sk
, net
);
1661 rc
= kernel_bind(vs
->sock
, (struct sockaddr
*) &vxlan_addr
,
1662 sizeof(vxlan_addr
));
1664 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1665 &vxlan_addr
.sin_addr
, ntohs(vxlan_addr
.sin_port
), rc
);
1666 sk_release_kernel(sk
);
1671 /* Disable multicast loopback */
1672 inet_sk(sk
)->mc_loop
= 0;
1674 /* Mark socket as an encapsulation socket. */
1675 udp_sk(sk
)->encap_type
= 1;
1676 udp_sk(sk
)->encap_rcv
= vxlan_udp_encap_recv
;
1678 atomic_set(&vs
->refcnt
, 1);
1683 /* Scheduled at device creation to bind to a socket */
1684 static void vxlan_sock_work(struct work_struct
*work
)
1686 struct vxlan_dev
*vxlan
1687 = container_of(work
, struct vxlan_dev
, sock_work
);
1688 struct net_device
*dev
= vxlan
->dev
;
1689 struct net
*net
= dev_net(dev
);
1690 __u32 vni
= vxlan
->default_dst
.remote_vni
;
1691 __be16 port
= vxlan
->dst_port
;
1692 struct vxlan_net
*vn
= net_generic(net
, vxlan_net_id
);
1693 struct vxlan_sock
*nvs
, *ovs
;
1695 nvs
= vxlan_socket_create(net
, port
);
1697 netdev_err(vxlan
->dev
, "Can not create UDP socket, %ld\n",
1702 spin_lock(&vn
->sock_lock
);
1703 /* Look again to see if can reuse socket */
1704 ovs
= vxlan_find_port(net
, port
);
1706 atomic_inc(&ovs
->refcnt
);
1707 vxlan
->vn_sock
= ovs
;
1708 hlist_add_head_rcu(&vxlan
->hlist
, vni_head(ovs
, vni
));
1709 spin_unlock(&vn
->sock_lock
);
1711 sk_release_kernel(nvs
->sock
->sk
);
1714 vxlan
->vn_sock
= nvs
;
1715 hlist_add_head_rcu(&nvs
->hlist
, vs_head(net
, port
));
1716 hlist_add_head_rcu(&vxlan
->hlist
, vni_head(nvs
, vni
));
1717 spin_unlock(&vn
->sock_lock
);
1723 static int vxlan_newlink(struct net
*net
, struct net_device
*dev
,
1724 struct nlattr
*tb
[], struct nlattr
*data
[])
1726 struct vxlan_net
*vn
= net_generic(net
, vxlan_net_id
);
1727 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1728 struct vxlan_rdst
*dst
= &vxlan
->default_dst
;
1732 if (!data
[IFLA_VXLAN_ID
])
1735 vni
= nla_get_u32(data
[IFLA_VXLAN_ID
]);
1736 dst
->remote_vni
= vni
;
1738 if (data
[IFLA_VXLAN_GROUP
])
1739 dst
->remote_ip
= nla_get_be32(data
[IFLA_VXLAN_GROUP
]);
1741 if (data
[IFLA_VXLAN_LOCAL
])
1742 vxlan
->saddr
= nla_get_be32(data
[IFLA_VXLAN_LOCAL
]);
1744 if (data
[IFLA_VXLAN_LINK
] &&
1745 (dst
->remote_ifindex
= nla_get_u32(data
[IFLA_VXLAN_LINK
]))) {
1746 struct net_device
*lowerdev
1747 = __dev_get_by_index(net
, dst
->remote_ifindex
);
1750 pr_info("ifindex %d does not exist\n", dst
->remote_ifindex
);
1755 dev
->mtu
= lowerdev
->mtu
- VXLAN_HEADROOM
;
1757 /* update header length based on lower device */
1758 dev
->hard_header_len
= lowerdev
->hard_header_len
+
1762 if (data
[IFLA_VXLAN_TOS
])
1763 vxlan
->tos
= nla_get_u8(data
[IFLA_VXLAN_TOS
]);
1765 if (data
[IFLA_VXLAN_TTL
])
1766 vxlan
->ttl
= nla_get_u8(data
[IFLA_VXLAN_TTL
]);
1768 if (!data
[IFLA_VXLAN_LEARNING
] || nla_get_u8(data
[IFLA_VXLAN_LEARNING
]))
1769 vxlan
->flags
|= VXLAN_F_LEARN
;
1771 if (data
[IFLA_VXLAN_AGEING
])
1772 vxlan
->age_interval
= nla_get_u32(data
[IFLA_VXLAN_AGEING
]);
1774 vxlan
->age_interval
= FDB_AGE_DEFAULT
;
1776 if (data
[IFLA_VXLAN_PROXY
] && nla_get_u8(data
[IFLA_VXLAN_PROXY
]))
1777 vxlan
->flags
|= VXLAN_F_PROXY
;
1779 if (data
[IFLA_VXLAN_RSC
] && nla_get_u8(data
[IFLA_VXLAN_RSC
]))
1780 vxlan
->flags
|= VXLAN_F_RSC
;
1782 if (data
[IFLA_VXLAN_L2MISS
] && nla_get_u8(data
[IFLA_VXLAN_L2MISS
]))
1783 vxlan
->flags
|= VXLAN_F_L2MISS
;
1785 if (data
[IFLA_VXLAN_L3MISS
] && nla_get_u8(data
[IFLA_VXLAN_L3MISS
]))
1786 vxlan
->flags
|= VXLAN_F_L3MISS
;
1788 if (data
[IFLA_VXLAN_LIMIT
])
1789 vxlan
->addrmax
= nla_get_u32(data
[IFLA_VXLAN_LIMIT
]);
1791 if (data
[IFLA_VXLAN_PORT_RANGE
]) {
1792 const struct ifla_vxlan_port_range
*p
1793 = nla_data(data
[IFLA_VXLAN_PORT_RANGE
]);
1794 vxlan
->port_min
= ntohs(p
->low
);
1795 vxlan
->port_max
= ntohs(p
->high
);
1798 if (data
[IFLA_VXLAN_PORT
])
1799 vxlan
->dst_port
= nla_get_be16(data
[IFLA_VXLAN_PORT
]);
1801 if (vxlan_find_vni(net
, vni
, vxlan
->dst_port
)) {
1802 pr_info("duplicate VNI %u\n", vni
);
1806 SET_ETHTOOL_OPS(dev
, &vxlan_ethtool_ops
);
1808 /* create an fdb entry for default destination */
1809 err
= vxlan_fdb_create(vxlan
, all_zeros_mac
,
1810 vxlan
->default_dst
.remote_ip
,
1811 NUD_REACHABLE
|NUD_PERMANENT
,
1812 NLM_F_EXCL
|NLM_F_CREATE
,
1813 vxlan
->dst_port
, vxlan
->default_dst
.remote_vni
,
1814 vxlan
->default_dst
.remote_ifindex
, NTF_SELF
);
1818 err
= register_netdevice(dev
);
1820 vxlan_fdb_delete_default(vxlan
);
1824 list_add(&vxlan
->next
, &vn
->vxlan_list
);
1829 static void vxlan_dellink(struct net_device
*dev
, struct list_head
*head
)
1831 struct vxlan_net
*vn
= net_generic(dev_net(dev
), vxlan_net_id
);
1832 struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1834 flush_workqueue(vxlan_wq
);
1836 spin_lock(&vn
->sock_lock
);
1837 hlist_del_rcu(&vxlan
->hlist
);
1838 spin_unlock(&vn
->sock_lock
);
1840 list_del(&vxlan
->next
);
1841 unregister_netdevice_queue(dev
, head
);
1844 static size_t vxlan_get_size(const struct net_device
*dev
)
1847 return nla_total_size(sizeof(__u32
)) + /* IFLA_VXLAN_ID */
1848 nla_total_size(sizeof(__be32
)) +/* IFLA_VXLAN_GROUP */
1849 nla_total_size(sizeof(__u32
)) + /* IFLA_VXLAN_LINK */
1850 nla_total_size(sizeof(__be32
))+ /* IFLA_VXLAN_LOCAL */
1851 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_TTL */
1852 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_TOS */
1853 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_LEARNING */
1854 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_PROXY */
1855 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_RSC */
1856 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_L2MISS */
1857 nla_total_size(sizeof(__u8
)) + /* IFLA_VXLAN_L3MISS */
1858 nla_total_size(sizeof(__u32
)) + /* IFLA_VXLAN_AGEING */
1859 nla_total_size(sizeof(__u32
)) + /* IFLA_VXLAN_LIMIT */
1860 nla_total_size(sizeof(struct ifla_vxlan_port_range
)) +
1861 nla_total_size(sizeof(__be16
))+ /* IFLA_VXLAN_PORT */
1865 static int vxlan_fill_info(struct sk_buff
*skb
, const struct net_device
*dev
)
1867 const struct vxlan_dev
*vxlan
= netdev_priv(dev
);
1868 const struct vxlan_rdst
*dst
= &vxlan
->default_dst
;
1869 struct ifla_vxlan_port_range ports
= {
1870 .low
= htons(vxlan
->port_min
),
1871 .high
= htons(vxlan
->port_max
),
1874 if (nla_put_u32(skb
, IFLA_VXLAN_ID
, dst
->remote_vni
))
1875 goto nla_put_failure
;
1877 if (dst
->remote_ip
&& nla_put_be32(skb
, IFLA_VXLAN_GROUP
, dst
->remote_ip
))
1878 goto nla_put_failure
;
1880 if (dst
->remote_ifindex
&& nla_put_u32(skb
, IFLA_VXLAN_LINK
, dst
->remote_ifindex
))
1881 goto nla_put_failure
;
1883 if (vxlan
->saddr
&& nla_put_be32(skb
, IFLA_VXLAN_LOCAL
, vxlan
->saddr
))
1884 goto nla_put_failure
;
1886 if (nla_put_u8(skb
, IFLA_VXLAN_TTL
, vxlan
->ttl
) ||
1887 nla_put_u8(skb
, IFLA_VXLAN_TOS
, vxlan
->tos
) ||
1888 nla_put_u8(skb
, IFLA_VXLAN_LEARNING
,
1889 !!(vxlan
->flags
& VXLAN_F_LEARN
)) ||
1890 nla_put_u8(skb
, IFLA_VXLAN_PROXY
,
1891 !!(vxlan
->flags
& VXLAN_F_PROXY
)) ||
1892 nla_put_u8(skb
, IFLA_VXLAN_RSC
, !!(vxlan
->flags
& VXLAN_F_RSC
)) ||
1893 nla_put_u8(skb
, IFLA_VXLAN_L2MISS
,
1894 !!(vxlan
->flags
& VXLAN_F_L2MISS
)) ||
1895 nla_put_u8(skb
, IFLA_VXLAN_L3MISS
,
1896 !!(vxlan
->flags
& VXLAN_F_L3MISS
)) ||
1897 nla_put_u32(skb
, IFLA_VXLAN_AGEING
, vxlan
->age_interval
) ||
1898 nla_put_u32(skb
, IFLA_VXLAN_LIMIT
, vxlan
->addrmax
) ||
1899 nla_put_be16(skb
, IFLA_VXLAN_PORT
, vxlan
->dst_port
))
1900 goto nla_put_failure
;
1902 if (nla_put(skb
, IFLA_VXLAN_PORT_RANGE
, sizeof(ports
), &ports
))
1903 goto nla_put_failure
;
1911 static struct rtnl_link_ops vxlan_link_ops __read_mostly
= {
1913 .maxtype
= IFLA_VXLAN_MAX
,
1914 .policy
= vxlan_policy
,
1915 .priv_size
= sizeof(struct vxlan_dev
),
1916 .setup
= vxlan_setup
,
1917 .validate
= vxlan_validate
,
1918 .newlink
= vxlan_newlink
,
1919 .dellink
= vxlan_dellink
,
1920 .get_size
= vxlan_get_size
,
1921 .fill_info
= vxlan_fill_info
,
1924 static __net_init
int vxlan_init_net(struct net
*net
)
1926 struct vxlan_net
*vn
= net_generic(net
, vxlan_net_id
);
1929 INIT_LIST_HEAD(&vn
->vxlan_list
);
1930 spin_lock_init(&vn
->sock_lock
);
1932 for (h
= 0; h
< PORT_HASH_SIZE
; ++h
)
1933 INIT_HLIST_HEAD(&vn
->sock_list
[h
]);
1938 static __net_exit
void vxlan_exit_net(struct net
*net
)
1940 struct vxlan_net
*vn
= net_generic(net
, vxlan_net_id
);
1941 struct vxlan_dev
*vxlan
;
1945 list_for_each_entry(vxlan
, &vn
->vxlan_list
, next
)
1946 unregister_netdevice_queue(vxlan
->dev
, &list
);
1947 unregister_netdevice_many(&list
);
1951 static struct pernet_operations vxlan_net_ops
= {
1952 .init
= vxlan_init_net
,
1953 .exit
= vxlan_exit_net
,
1954 .id
= &vxlan_net_id
,
1955 .size
= sizeof(struct vxlan_net
),
1958 static int __init
vxlan_init_module(void)
1962 vxlan_wq
= alloc_workqueue("vxlan", 0, 0);
1966 get_random_bytes(&vxlan_salt
, sizeof(vxlan_salt
));
1968 rc
= register_pernet_device(&vxlan_net_ops
);
1972 rc
= rtnl_link_register(&vxlan_link_ops
);
1979 unregister_pernet_device(&vxlan_net_ops
);
1981 destroy_workqueue(vxlan_wq
);
1984 late_initcall(vxlan_init_module
);
1986 static void __exit
vxlan_cleanup_module(void)
1988 rtnl_link_unregister(&vxlan_link_ops
);
1989 destroy_workqueue(vxlan_wq
);
1990 unregister_pernet_device(&vxlan_net_ops
);
1993 module_exit(vxlan_cleanup_module
);
1995 MODULE_LICENSE("GPL");
1996 MODULE_VERSION(VXLAN_VERSION
);
1997 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
1998 MODULE_ALIAS_RTNL_LINK("vxlan");