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_pre_route_finish(struct sk_buff *skb)
147 struct net_bridge_port *p = skb->dev->br_port;
148 struct snat_conf *sc;
149 struct iphdr *iph = ip_hdr(skb);
153 skb->dst = (struct dst_entry *)&__fake_rtable;
156 /* Don't process packets that were not translated due to NAT */
157 spin_lock_irqsave(&p->lock, flags);
159 ip_addr = ntohl(iph->daddr);
160 if (sc && (ip_addr >= sc->ip_addr_start)
161 && (ip_addr <= sc->ip_addr_end)) {
162 spin_unlock_irqrestore(&p->lock, flags);
166 /* If SNAT is configured for this input device, check the IP->MAC
167 * mappings to see if we should update the destination MAC. */
169 dnat_mac(skb->dev->br_port, skb);
171 spin_unlock_irqrestore(&p->lock, flags);
176 /* Checks whether 'skb' is an ARP request for an SNAT'd interface. If
177 * so, it will generate a response.
179 * Returns 0 if the packet was not handled. Otherwise, -1 is returned
180 * and the caller is responsible for freeing 'skb'. */
182 handle_arp_snat(struct sk_buff *skb)
184 struct net_bridge_port *p = skb->dev->br_port;
185 struct ip_arphdr *ah = (struct ip_arphdr *)arp_hdr(skb);
188 struct snat_conf *sc;
190 if ((ah->ar_op != htons(ARPOP_REQUEST))
191 || ah->ar_hln != ETH_ALEN
192 || ah->ar_pro != htons(ETH_P_IP)
196 ip_addr = ntohl(ah->ar_tip);
197 spin_lock_irqsave(&p->lock, flags);
200 /* We're only interested in addresses we rewrite. */
201 if (!sc || (sc && ((ip_addr < sc->ip_addr_start)
202 || (ip_addr > sc->ip_addr_end)))) {
203 spin_unlock_irqrestore(&p->lock, flags);
206 spin_unlock_irqrestore(&p->lock, flags);
208 arp_send(ARPOP_REPLY, ETH_P_ARP, ah->ar_sip, skb->dev, ah->ar_tip,
209 ah->ar_sha, p->dp->netdev->dev_addr, ah->ar_sha);
214 /* Checks whether 'skb' is a ping request for an SNAT'd interface. If
215 * so, it will generate a response.
217 * Returns 0 if the packet was not handled. Otherwise, -1 is returned
218 * and the caller is responsible for freeing 'skb'. */
220 handle_icmp_snat(struct sk_buff *skb)
222 struct net_bridge_port *p = skb->dev->br_port;
223 struct snat_conf *sc;
225 struct iphdr *iph = ip_hdr(skb);
227 struct icmphdr *icmph;
228 unsigned int datalen;
229 uint8_t tmp_eth[ETH_ALEN];
231 struct sk_buff *nskb;
235 ip_addr = ntohl(iph->daddr);
236 spin_lock_irqsave(&p->lock, flags);
239 /* We're only interested in addresses we rewrite. */
240 if (!sc || (sc && ((ip_addr < sc->ip_addr_start)
241 || (ip_addr > sc->ip_addr_end)))) {
242 spin_unlock_irqrestore(&p->lock, flags);
245 spin_unlock_irqrestore(&p->lock, flags);
247 icmph = (struct icmphdr *) ((u_int32_t *)iph + iph->ihl);
248 datalen = skb->len - iph->ihl * 4;
250 /* Drop fragments and packets not long enough to hold the ICMP
252 if (((ntohs(iph->frag_off) & IP_OFFSET) != 0) || datalen < 4)
255 /* We only respond to echo requests to our address. Continue
256 * processing replies and other ICMP messages since they may be
257 * intended for NAT'd hosts. */
258 if (icmph->type != ICMP_ECHO)
261 /* Send an echo reply in response */
262 nskb = skb_copy(skb, GFP_ATOMIC);
265 printk("skb copy failed for icmp reply\n");
271 icmph = (struct icmphdr *) ((u_int32_t *)iph + iph->ihl);
274 iph->daddr = iph->saddr;
277 memcpy(tmp_eth, eh->h_dest, ETH_ALEN);
278 memcpy(eh->h_dest, eh->h_source, ETH_ALEN);
279 memcpy(eh->h_source, tmp_eth, ETH_ALEN);
281 icmph->type = ICMP_ECHOREPLY;
283 dp_xmit_skb_push(nskb);
288 /* Check if any SNAT maintenance needs to be done on 'skb' before it's
289 * checked against the datapath's tables. This includes DNAT
290 * modification based on prior SNAT action and responding to ARP and
291 * echo requests for the SNAT interface.
293 * Returns 0 if 'skb' should continue to be processed by the caller.
294 * Returns -1 if the packet was handled, and the caller should free
298 snat_pre_route(struct sk_buff *skb)
303 if (skb->protocol == htons(ETH_P_ARP))
304 return handle_arp_snat(skb);
305 else if (skb->protocol != htons(ETH_P_IP))
309 if (iph->ihl < 5 || iph->version != 4)
312 if (!pskb_may_pull(skb, iph->ihl*4))
315 /* Check if we need to echo reply for this address */
316 if ((iph->protocol == IPPROTO_ICMP) && (handle_icmp_snat(skb)))
319 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
322 len = ntohs(iph->tot_len);
323 if ((skb->len < len) || len < (iph->ihl*4))
326 if (pskb_trim_rcsum(skb, len))
329 return NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
330 snat_pre_route_finish);
338 snat_skb_finish(struct sk_buff *skb)
340 NF_HOOK(PF_INET, NF_INET_POST_ROUTING, skb, NULL, skb->dev,
346 /* Update the MAC->IP mappings for the private side of the SNAT'd
349 update_mapping(struct net_bridge_port *p, const struct sk_buff *skb)
352 struct snat_conf *sc;
353 const struct iphdr *iph = ip_hdr(skb);
354 const struct ethhdr *eh = eth_hdr(skb);
355 struct snat_mapping *m;
357 spin_lock_irqsave(&p->lock, flags);
362 list_for_each_entry (m, &sc->mappings, node) {
363 if (m->ip_addr == iph->saddr){
364 if (memcmp(m->hw_addr, eh->h_source, ETH_ALEN)) {
365 memcpy(m->hw_addr, eh->h_source, ETH_ALEN);
372 m = kmalloc(sizeof *m, GFP_ATOMIC);
375 m->ip_addr = iph->saddr;
376 memcpy(m->hw_addr, eh->h_source, ETH_ALEN);
379 list_add(&m->node, &sc->mappings);
382 spin_unlock_irqrestore(&p->lock, flags);
385 /* Perform SNAT modification on 'skb' and send out 'out_port'. If the
386 * port was not configured for SNAT, it will be sent through the interface
387 * unmodified. 'skb' is not consumed, so caller will need to free it.
390 snat_skb(struct datapath *dp, const struct sk_buff *skb, int out_port)
392 struct net_bridge_port *p = dp->ports[out_port];
393 struct sk_buff *nskb;
398 nskb = skb_copy(skb, GFP_ATOMIC);
404 /* We only SNAT IP, so just send it on its way if not */
405 if (skb->protocol != htons(ETH_P_IP)) {
410 /* Set the source MAC to the OF interface */
411 memcpy(eth_hdr(nskb)->h_source, dp->netdev->dev_addr, ETH_ALEN);
413 update_mapping(p, skb);
415 /* Take the Ethernet header back off for netfilter hooks. */
416 skb_pull(nskb, ETH_HLEN);
418 NF_HOOK(PF_INET, NF_INET_FORWARD, nskb, skb->dev, nskb->dev,
422 /* Remove SNAT configuration on port 'p'.
424 * NB: The caller must hold the port's spinlock. */
426 snat_free_conf(struct net_bridge_port *p)
428 struct snat_conf *sc = p->snat;
433 /* Free existing mapping entries */
434 while (!list_empty(&sc->mappings)) {
435 struct snat_mapping *m = list_entry(sc->mappings.next,
436 struct snat_mapping, node);
447 /* Remove SNAT configuration from an interface. */
449 snat_del_port(struct datapath *dp, uint16_t port)
452 struct net_bridge_port *p = dp->ports[port];
456 printk("Attempt to remove snat on non-existent port: %d\n", port);
460 spin_lock_irqsave(&p->lock, flags);
461 if (snat_free_conf(p)) {
462 /* SNAT not configured on this port */
463 spin_unlock_irqrestore(&p->lock, flags);
465 printk("Attempt to remove snat on non-snat port: %d\n", port);
469 spin_unlock_irqrestore(&p->lock, flags);
474 /* Add SNAT configuration to an interface. */
476 snat_add_port(struct datapath *dp, uint16_t port,
477 uint32_t ip_addr_start, uint32_t ip_addr_end,
478 uint16_t mac_timeout)
481 struct net_bridge_port *p = dp->ports[port];
482 struct snat_conf *sc;
485 if (mac_timeout == 0)
486 mac_timeout = MAC_TIMEOUT_DEFAULT;
490 printk("Attempt to add snat on non-existent port: %d\n", port);
494 /* If SNAT is already configured on the port, check whether the same
495 * IP addresses are used. If so, just update the mac timeout
496 * configuration. Otherwise, drop all SNAT configuration and
498 spin_lock_irqsave(&p->lock, flags);
500 if ((p->snat->ip_addr_start == ip_addr_start)
501 && (p->snat->ip_addr_end == ip_addr_end)) {
502 p->snat->mac_timeout = mac_timeout;
503 spin_unlock_irqrestore(&p->lock, flags);
507 /* Free the existing configuration and mappings. */
511 sc = kzalloc(sizeof *sc, GFP_ATOMIC);
513 spin_unlock_irqrestore(&p->lock, flags);
517 sc->ip_addr_start = ip_addr_start;
518 sc->ip_addr_end = ip_addr_end;
519 sc->mac_timeout = mac_timeout;
520 INIT_LIST_HEAD(&sc->mappings);
523 spin_unlock_irqrestore(&p->lock, flags);
528 /* Handle a SNAT configuration message.
530 * Returns 0 if no problems are found. Otherwise, a negative errno. */
532 snat_mod_config(struct datapath *dp, const struct nx_act_config *nac)
534 int n_entries = (ntohs(nac->header.header.length) - sizeof *nac)
535 / sizeof (struct nx_snat_config);
539 for (i=0; i<n_entries; i++) {
540 const struct nx_snat_config *sc = &nac->snat[i];
541 uint16_t port = ntohs(sc->port);
544 if (sc->command == NXSC_ADD)
545 r = snat_add_port(dp, port,
546 ntohl(sc->ip_addr_start), ntohl(sc->ip_addr_end),
547 ntohs(sc->mac_timeout));
549 r = snat_del_port(dp, port);