3 * Distributed under the terms of the GNU GPL version 2.
4 * Copyright (c) 2008 Nicira Networks
7 #include <linux/netdevice.h>
8 #include <linux/netfilter.h>
9 #include <linux/netfilter_ipv4.h>
12 #include <linux/icmp.h>
13 #include <linux/if_ether.h>
15 #include <net/route.h>
19 #include "nx_act_snat.h"
22 /* We need these fake structures to make netfilter happy --
23 * lots of places assume that skb->dst != NULL, which isn't
24 * all that unreasonable.
26 * Currently, we fill in the PMTU entry because netfilter
27 * refragmentation needs it, and the rt_flags entry because
28 * ipt_REJECT needs it. Future netfilter modules might
29 * require us to fill additional fields. */
30 static struct net_device __fake_net_device = {
31 .hard_header_len = ETH_HLEN
34 static struct rtable __fake_rtable = {
37 .__refcnt = ATOMIC_INIT(1),
38 .dev = &__fake_net_device,
39 .path = &__fake_rtable.u.dst,
40 .metrics = {[RTAX_MTU - 1] = 1500},
47 /* Define ARP for IP since the Linux headers don't do it cleanly. */
54 uint8_t ar_sha[ETH_ALEN];
56 uint8_t ar_tha[ETH_ALEN];
58 } __attribute__((packed));
59 OFP_ASSERT(sizeof(struct ip_arphdr) == 28);
62 /* Push the Ethernet header back on and tranmit the packet. */
64 dp_xmit_skb_push(struct sk_buff *skb)
66 skb_push(skb, ETH_HLEN);
67 return dp_xmit_skb(skb);
70 /* Perform maintainence related to a SNAT'd interface. Currently, this only
71 * checks whether MAC->IP bindings have expired.
73 * Called with the RCU read lock */
75 snat_maint(struct net_bridge_port *p)
78 struct snat_mapping *m, *n;
80 unsigned long timeout;
82 spin_lock_irqsave(&p->lock, flags);
87 timeout = sc->mac_timeout * HZ;
89 list_for_each_entry_safe (m, n, &sc->mappings, node) {
90 if (time_after(jiffies, m->used + timeout)) {
97 spin_unlock_irqrestore(&p->lock, flags);
100 /* When the packet is bound for a local interface, strip off the fake
103 void snat_local_in(struct sk_buff *skb)
105 if (skb->dst == (struct dst_entry *)&__fake_rtable) {
106 dst_release(skb->dst);
111 /* Check whether destination IP's address is in the IP->MAC mappings.
112 * If it is, then overwrite the destination MAC with the value from the
115 * Returns -1 if there is a problem, otherwise 0. */
117 dnat_mac(struct net_bridge_port *p, struct sk_buff *skb)
119 struct snat_conf *sc = p->snat;
120 struct iphdr *iph = ip_hdr(skb);
121 struct ethhdr *eh = eth_hdr(skb);
122 struct snat_mapping *m;
124 if (skb->protocol != htons(ETH_P_IP))
127 list_for_each_entry (m, &sc->mappings, node) {
128 if (m->ip_addr == iph->daddr){
130 if (!make_writable(&skb)) {
132 printk("make_writable failed\n");
136 memcpy(eh->h_dest, m->hw_addr, ETH_ALEN);
145 __snat_this_address(struct snat_conf *sc, u32 ip_addr)
148 u32 h_ip_addr = ntohl(ip_addr);
149 return (h_ip_addr >= sc->ip_addr_start &&
150 h_ip_addr <= sc->ip_addr_end);
156 snat_this_address(struct net_bridge_port *p, u32 ip_addr)
158 unsigned long int flags;
161 spin_lock_irqsave(&p->lock, flags);
162 retval = __snat_this_address(p->snat, ip_addr);
163 spin_unlock_irqrestore(&p->lock, flags);
169 snat_pre_route_finish(struct sk_buff *skb)
171 struct net_bridge_port *p = skb->dev->br_port;
172 struct snat_conf *sc;
173 struct iphdr *iph = ip_hdr(skb);
176 skb->dst = (struct dst_entry *)&__fake_rtable;
179 /* Don't process packets that were not translated due to NAT */
180 spin_lock_irqsave(&p->lock, flags);
182 if (__snat_this_address(sc, iph->daddr)) {
183 spin_unlock_irqrestore(&p->lock, flags);
187 /* If SNAT is configured for this input device, check the IP->MAC
188 * mappings to see if we should update the destination MAC. */
190 dnat_mac(skb->dev->br_port, skb);
192 spin_unlock_irqrestore(&p->lock, flags);
197 /* Checks whether 'skb' is an ARP request for an SNAT'd interface. If
198 * so, it will generate a response.
200 * Returns 0 if the packet was not handled. Otherwise, -1 is returned
201 * and the caller is responsible for freeing 'skb'. */
203 handle_arp_snat(struct sk_buff *skb)
205 struct net_bridge_port *p = skb->dev->br_port;
206 struct ip_arphdr *ah = (struct ip_arphdr *)arp_hdr(skb);
208 if (!pskb_may_pull(skb, sizeof *ah))
211 if ((ah->ar_op != htons(ARPOP_REQUEST))
212 || ah->ar_hln != ETH_ALEN
213 || ah->ar_pro != htons(ETH_P_IP)
217 /* We're only interested in addresses we rewrite. */
218 if (!snat_this_address(p, ah->ar_tip)) {
222 arp_send(ARPOP_REPLY, ETH_P_ARP, ah->ar_sip, skb->dev, ah->ar_tip,
223 ah->ar_sha, p->dp->netdev->dev_addr, ah->ar_sha);
228 /* Checks whether 'skb' is a ping request for an SNAT'd interface. If
229 * so, it will generate a response.
231 * Returns 0 if the packet was not handled. Otherwise, -1 is returned
232 * and the caller is responsible for freeing 'skb'. */
234 handle_icmp_snat(struct sk_buff *skb)
236 struct net_bridge_port *p = skb->dev->br_port;
238 struct iphdr *iph = ip_hdr(skb);
239 struct icmphdr *icmph;
240 unsigned int datalen;
241 uint8_t tmp_eth[ETH_ALEN];
243 struct sk_buff *nskb;
245 /* We're only interested in addresses we rewrite. */
246 if (!snat_this_address(p, iph->daddr)) {
250 icmph = (struct icmphdr *) ((u_int32_t *)iph + iph->ihl);
251 datalen = skb->len - iph->ihl * 4;
253 /* Drop fragments and packets not long enough to hold the ICMP
255 if (((ntohs(iph->frag_off) & IP_OFFSET) != 0) || datalen < 4)
258 /* We only respond to echo requests to our address. Continue
259 * processing replies and other ICMP messages since they may be
260 * intended for NAT'd hosts. */
261 if (icmph->type != ICMP_ECHO)
264 /* Send an echo reply in response */
265 nskb = skb_copy(skb, GFP_ATOMIC);
268 printk("skb copy failed for icmp reply\n");
274 icmph = (struct icmphdr *) ((u_int32_t *)iph + iph->ihl);
277 iph->daddr = iph->saddr;
280 memcpy(tmp_eth, eh->h_dest, ETH_ALEN);
281 memcpy(eh->h_dest, eh->h_source, ETH_ALEN);
282 memcpy(eh->h_source, tmp_eth, ETH_ALEN);
284 icmph->type = ICMP_ECHOREPLY;
286 dp_xmit_skb_push(nskb);
291 /* Check if any SNAT maintenance needs to be done on 'skb' before it's
292 * checked against the datapath's tables. This includes DNAT
293 * modification based on prior SNAT action and responding to ARP and
294 * echo requests for the SNAT interface.
296 * Returns 0 if 'skb' should continue to be processed by the caller.
297 * Returns -1 if the packet was handled, and the caller should free
301 snat_pre_route(struct sk_buff *skb)
306 if (skb->protocol == htons(ETH_P_ARP))
307 return handle_arp_snat(skb);
308 else if (skb->protocol != htons(ETH_P_IP))
311 if (!pskb_may_pull(skb, sizeof *iph))
315 if (iph->ihl < 5 || iph->version != 4)
318 if (!pskb_may_pull(skb, iph->ihl*4))
321 /* Check if we need to echo reply for this address */
322 if ((iph->protocol == IPPROTO_ICMP) && (handle_icmp_snat(skb)))
325 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
328 len = ntohs(iph->tot_len);
329 if ((skb->len < len) || len < (iph->ihl*4))
332 if (pskb_trim_rcsum(skb, len))
335 return NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
336 snat_pre_route_finish);
344 snat_skb_finish(struct sk_buff *skb)
346 NF_HOOK(PF_INET, NF_INET_POST_ROUTING, skb, NULL, skb->dev,
352 /* Update the MAC->IP mappings for the private side of the SNAT'd
355 update_mapping(struct net_bridge_port *p, const struct sk_buff *skb)
358 struct snat_conf *sc;
359 const struct iphdr *iph = ip_hdr(skb);
360 const struct ethhdr *eh = eth_hdr(skb);
361 struct snat_mapping *m;
363 spin_lock_irqsave(&p->lock, flags);
368 list_for_each_entry (m, &sc->mappings, node) {
369 if (m->ip_addr == iph->saddr){
370 memcpy(m->hw_addr, eh->h_source, ETH_ALEN);
376 m = kmalloc(sizeof *m, GFP_ATOMIC);
379 m->ip_addr = iph->saddr;
380 memcpy(m->hw_addr, eh->h_source, ETH_ALEN);
383 list_add(&m->node, &sc->mappings);
386 spin_unlock_irqrestore(&p->lock, flags);
389 /* Perform SNAT modification on 'skb' and send out 'out_port'. If the
390 * port was not configured for SNAT, it will be sent through the interface
391 * unmodified. 'skb' is not consumed, so caller will need to free it.
394 snat_skb(struct datapath *dp, const struct sk_buff *skb, int out_port)
396 struct net_bridge_port *p = dp->ports[out_port];
397 struct sk_buff *nskb;
402 /* FIXME: Expensive. Just need to skb_clone() here?
403 * (However, the skb_copy() does linearize and ensure that the headers
404 * are accessible.) */
405 nskb = skb_copy(skb, GFP_ATOMIC);
411 /* We only SNAT IP, so just send it on its way if not */
412 if (skb->protocol != htons(ETH_P_IP)) {
417 /* Set the source MAC to the OF interface */
418 memcpy(eth_hdr(nskb)->h_source, dp->netdev->dev_addr, ETH_ALEN);
420 update_mapping(p, skb);
422 /* Take the Ethernet header back off for netfilter hooks. */
423 skb_pull(nskb, ETH_HLEN);
425 NF_HOOK(PF_INET, NF_INET_FORWARD, nskb, skb->dev, nskb->dev,
429 /* Remove SNAT configuration on port 'p'.
431 * NB: The caller must hold the port's spinlock. */
433 snat_free_conf(struct net_bridge_port *p)
435 struct snat_conf *sc = p->snat;
440 /* Free existing mapping entries */
441 while (!list_empty(&sc->mappings)) {
442 struct snat_mapping *m = list_entry(sc->mappings.next,
443 struct snat_mapping, node);
454 /* Remove SNAT configuration from an interface. */
456 snat_del_port(struct datapath *dp, uint16_t port)
459 struct net_bridge_port *p = dp->ports[port];
463 printk("Attempt to remove snat on non-existent port: %d\n", port);
467 spin_lock_irqsave(&p->lock, flags);
468 if (snat_free_conf(p)) {
469 /* SNAT not configured on this port */
470 spin_unlock_irqrestore(&p->lock, flags);
472 printk("Attempt to remove snat on non-snat port: %d\n", port);
476 spin_unlock_irqrestore(&p->lock, flags);
481 /* Add SNAT configuration to an interface. */
483 snat_add_port(struct datapath *dp, uint16_t port,
484 uint32_t ip_addr_start, uint32_t ip_addr_end,
485 uint16_t mac_timeout)
488 struct net_bridge_port *p = dp->ports[port];
489 struct snat_conf *sc;
492 if (mac_timeout == 0)
493 mac_timeout = MAC_TIMEOUT_DEFAULT;
497 printk("Attempt to add snat on non-existent port: %d\n", port);
501 /* If SNAT is already configured on the port, check whether the same
502 * IP addresses are used. If so, just update the mac timeout
503 * configuration. Otherwise, drop all SNAT configuration and
505 spin_lock_irqsave(&p->lock, flags);
507 if ((p->snat->ip_addr_start == ip_addr_start)
508 && (p->snat->ip_addr_end == ip_addr_end)) {
509 p->snat->mac_timeout = mac_timeout;
510 spin_unlock_irqrestore(&p->lock, flags);
514 /* Free the existing configuration and mappings. */
518 sc = kzalloc(sizeof *sc, GFP_ATOMIC);
520 spin_unlock_irqrestore(&p->lock, flags);
524 sc->ip_addr_start = ip_addr_start;
525 sc->ip_addr_end = ip_addr_end;
526 sc->mac_timeout = mac_timeout;
527 INIT_LIST_HEAD(&sc->mappings);
530 spin_unlock_irqrestore(&p->lock, flags);
535 /* Handle a SNAT configuration message.
537 * Returns 0 if no problems are found. Otherwise, a negative errno. */
539 snat_mod_config(struct datapath *dp, const struct nx_act_config *nac)
541 int n_entries = (ntohs(nac->header.header.length) - sizeof *nac)
542 / sizeof (struct nx_snat_config);
546 for (i=0; i<n_entries; i++) {
547 const struct nx_snat_config *sc = &nac->snat[i];
548 uint16_t port = ntohs(sc->port);
551 if (sc->command == NXSC_ADD)
552 r = snat_add_port(dp, port,
553 ntohl(sc->ip_addr_start), ntohl(sc->ip_addr_end),
554 ntohs(sc->mac_timeout));
556 r = snat_del_port(dp, port);