1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The IP forwarding functionality.
7 *
8 * Authors: see ip.c
9 *
10 * Fixes:
11 * Many : Split from ip.c , see ip_input.c for history.
12 */
13
14 #include <linux/config.h>
15 #include <linux/types.h>
16 #include <linux/mm.h>
17 #include <linux/sched.h>
18 #include <linux/skbuff.h>
19 #include <linux/ip.h>
20 #include <linux/icmp.h>
21 #include <linux/netdevice.h>
22 #include <net/sock.h>
23 #include <net/ip.h>
24 #include <net/icmp.h>
25 #include <linux/tcp.h>
26 #include <linux/udp.h>
27 #include <linux/firewall.h>
28 #include <linux/ip_fw.h>
29 #include <net/checksum.h>
30 #include <linux/route.h>
31 #include <net/route.h>
32
33 #ifdef CONFIG_IP_FORWARD
34 #ifdef CONFIG_IP_MROUTE
35
36 /*
37 * Encapsulate a packet by attaching a valid IPIP header to it.
38 * This avoids tunnel drivers and other mess and gives us the speed so
39 * important for multicast video.
40 */
41
42 static void ip_encap(struct sk_buff *skb, int len, struct device *out, __u32 daddr)
/* ![[previous]](../icons/n_left.png)
![[next]](../icons/right.png)
![[first]](../icons/n_first.png)
![[last]](../icons/last.png)
![[top]](../icons/top.png)
![[bottom]](../icons/bottom.png)
![[index]](../icons/index.png)
*/
43 {
44 /*
45 * There is space for the IPIP header and MAC left.
46 *
47 * Firstly push down and install the IPIP header.
48 */
49 struct iphdr *iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr));
50 if(len>65515)
51 len=65515;
52 iph->version = 4;
53 iph->tos = skb->ip_hdr->tos;
54 iph->ttl = skb->ip_hdr->ttl;
55 iph->frag_off = 0;
56 iph->daddr = daddr;
57 iph->saddr = out->pa_addr;
58 iph->protocol = IPPROTO_IPIP;
59 iph->ihl = 5;
60 iph->tot_len = htons(skb->len);
61 iph->id = htons(ip_id_count++);
62 ip_send_check(iph);
63
64 skb->dev = out;
65 skb->arp = 1;
66 skb->raddr=daddr;
67 /*
68 * Now add the physical header (driver will push it down).
69 */
70 if (out->hard_header && out->hard_header(skb, out, ETH_P_IP, NULL, NULL, len)<0)
71 skb->arp=0;
72 /*
73 * Read to queue for transmission.
74 */
75 }
76
77 #endif
78
79 /*
80 * Forward an IP datagram to its next destination.
81 */
82
83 int ip_forward(struct sk_buff *skb, struct device *dev, int is_frag,
/* ![[previous]](../icons/left.png)
![[next]](../icons/n_right.png)
![[first]](../icons/first.png)
![[last]](../icons/n_last.png)
![[top]](../icons/top.png)
![[bottom]](../icons/bottom.png)
![[index]](../icons/index.png)
*/
84 __u32 target_addr)
85 {
86 struct device *dev2; /* Output device */
87 struct iphdr *iph; /* Our header */
88 struct sk_buff *skb2; /* Output packet */
89 struct rtable *rt; /* Route we use */
90 unsigned char *ptr; /* Data pointer */
91 unsigned long raddr; /* Router IP address */
92 struct options * opt = (struct options*)skb->proto_priv;
93 struct hh_cache *hh = NULL;
94 int encap = 0; /* Encap length */
95 #ifdef CONFIG_FIREWALL
96 int fw_res = 0; /* Forwarding result */
97 #ifdef CONFIG_IP_MASQUERADE
98 struct sk_buff *skb_in = skb; /* So we can remember if the masquerader did some swaps */
99 #endif
100
101 /*
102 * See if we are allowed to forward this.
103 * Note: demasqueraded fragments are always 'back'warded.
104 */
105
106
107 if(!(is_frag&4))
108 {
109 fw_res=call_fw_firewall(PF_INET, skb, skb->h.iph);
110 switch (fw_res) {
111 case FW_ACCEPT:
112 case FW_MASQUERADE:
113 break;
114 case FW_REJECT:
115 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, dev);
116 /* fall thru */
117 default:
118 return -1;
119 }
120 }
121 #endif
122 /*
123 * According to the RFC, we must first decrease the TTL field. If
124 * that reaches zero, we must reply an ICMP control message telling
125 * that the packet's lifetime expired.
126 *
127 * Exception:
128 * We may not generate an ICMP for an ICMP. icmp_send does the
129 * enforcement of this so we can forget it here. It is however
130 * sometimes VERY important.
131 */
132
133 iph = skb->h.iph;
134 iph->ttl--;
135
136 /*
137 * Re-compute the IP header checksum.
138 * This is inefficient. We know what has happened to the header
139 * and could thus adjust the checksum as Phil Karn does in KA9Q
140 */
141
142 iph->check = ntohs(iph->check) + 0x0100;
143 if ((iph->check & 0xFF00) == 0)
144 iph->check++; /* carry overflow */
145 iph->check = htons(iph->check);
146
147 if (iph->ttl <= 0)
148 {
149 /* Tell the sender its packet died... */
150 icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0, dev);
151 return -1;
152 }
153
154 #ifdef CONFIG_IP_MROUTE
155 if(!(is_frag&8))
156 {
157 #endif
158 /*
159 * OK, the packet is still valid. Fetch its destination address,
160 * and give it to the IP sender for further processing.
161 */
162
163 rt = ip_rt_route(target_addr, 0);
164
165 if (rt == NULL)
166 {
167 /*
168 * Tell the sender its packet cannot be delivered. Again
169 * ICMP is screened later.
170 */
171 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_NET_UNREACH, 0, dev);
172 return -1;
173 }
174
175
176 /*
177 * Gosh. Not only is the packet valid; we even know how to
178 * forward it onto its final destination. Can we say this
179 * is being plain lucky?
180 * If the router told us that there is no GW, use the dest.
181 * IP address itself- we seem to be connected directly...
182 */
183
184 raddr = rt->rt_gateway;
185
186 if (opt->is_strictroute && (rt->rt_flags & RTF_GATEWAY)) {
187 /*
188 * Strict routing permits no gatewaying
189 */
190
191 ip_rt_put(rt);
192 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_SR_FAILED, 0, dev);
193 return -1;
194 }
195
196 /*
197 * Having picked a route we can now send the frame out.
198 */
199
200 dev2 = rt->rt_dev;
201 hh = rt->rt_hh;
202 /*
203 * In IP you never have to forward a frame on the interface that it
204 * arrived upon. We now generate an ICMP HOST REDIRECT giving the route
205 * we calculated.
206 */
207 #ifndef CONFIG_IP_NO_ICMP_REDIRECT
208 if (dev == dev2 && !((iph->saddr^iph->daddr)&dev->pa_mask) &&
209 (rt->rt_flags&RTF_MODIFIED) && !opt->srr)
210 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, raddr, dev);
211 #endif
212 #ifdef CONFIG_IP_MROUTE
213 }
214 else
215 {
216 /*
217 * Multicast route forward. Routing is already done
218 */
219 dev2=skb->dev;
220 raddr=skb->raddr;
221 if(is_frag&16) /* VIFF_TUNNEL mode */
222 encap=20;
223 rt=NULL;
224 }
225 #endif
226
227
228 /*
229 * We now may allocate a new buffer, and copy the datagram into it.
230 * If the indicated interface is up and running, kick it.
231 */
232
233 if (dev2->flags & IFF_UP)
234 {
235 #ifdef CONFIG_IP_MASQUERADE
236 /*
237 * If this fragment needs masquerading, make it so...
238 * (Dont masquerade de-masqueraded fragments)
239 */
240 if (!(is_frag&4) && fw_res==FW_MASQUERADE)
241 ip_fw_masquerade(&skb, dev2);
242 #endif
243 IS_SKB(skb);
244
245 if (skb->len+encap > dev2->mtu && (ntohs(iph->frag_off) & IP_DF)) {
246 ip_statistics.IpFragFails++;
247 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(dev2->mtu), dev);
248 ip_rt_put(rt);
249 return -1;
250 }
251
252 #ifdef CONFIG_IP_MROUTE
253 if(skb_headroom(skb)-encap<dev2->hard_header_len)
254 {
255 skb2 = alloc_skb(dev2->hard_header_len + skb->len + encap + 15, GFP_ATOMIC);
256 #else
257 if(skb_headroom(skb)<dev2->hard_header_len)
258 {
259 skb2 = alloc_skb(dev2->hard_header_len + skb->len + 15, GFP_ATOMIC);
260 #endif
261 /*
262 * This is rare and since IP is tolerant of network failures
263 * quite harmless.
264 */
265
266 if (skb2 == NULL)
267 {
268 NETDEBUG(printk("\nIP: No memory available for IP forward\n"));
269 ip_rt_put(rt);
270 return -1;
271 }
272
273 IS_SKB(skb2);
274 /*
275 * Add the physical headers.
276 */
277 skb2->protocol=htons(ETH_P_IP);
278 #ifdef CONFIG_IP_MROUTE
279 if(is_frag&16)
280 {
281 skb_reserve(skb,(encap+dev->hard_header_len+15)&~15); /* 16 byte aligned IP headers are good */
282 ip_encap(skb2,skb->len, dev2, raddr);
283 }
284 else
285 #endif
286 ip_send(rt,skb2,raddr,skb->len,dev2,dev2->pa_addr);
287
288 /*
289 * We have to copy the bytes over as the new header wouldn't fit
290 * the old buffer. This should be very rare.
291 */
292
293 ptr = skb_put(skb2,skb->len);
294 skb2->free = 1;
295 skb2->h.raw = ptr;
296
297 /*
298 * Copy the packet data into the new buffer.
299 */
300 memcpy(ptr, skb->h.raw, skb->len);
301 memcpy(skb2->proto_priv, skb->proto_priv, sizeof(skb->proto_priv));
302 iph = skb2->ip_hdr = skb2->h.iph;
303 }
304 else
305 {
306 /*
307 * Build a new MAC header.
308 */
309
310 skb2 = skb;
311 skb2->dev=dev2;
312 #ifdef CONFIG_IP_MROUTE
313 if(is_frag&16)
314 ip_encap(skb,skb->len, dev2, raddr);
315 else
316 {
317 #endif
318 skb->arp=1;
319 skb->raddr=raddr;
320 if (hh)
321 {
322 memcpy(skb_push(skb, dev2->hard_header_len), hh->hh_data, dev2->hard_header_len);
323 if (!hh->hh_uptodate)
324 {
325 #if RT_CACHE_DEBUG >= 2
326 printk("ip_forward: hh miss %08x via %08x\n", target_addr, rt->rt_gateway);
327 #endif
328 skb->arp = 0;
329 }
330 }
331 else if (dev2->hard_header)
332 {
333 if(dev2->hard_header(skb, dev2, ETH_P_IP, NULL, NULL, skb->len)<0)
334 skb->arp=0;
335 }
336 #ifdef CONFIG_IP_MROUTE
337 }
338 #endif
339 ip_statistics.IpForwDatagrams++;
340 }
341
342 if (opt->optlen)
343 {
344 unsigned char * optptr;
345 if (opt->rr_needaddr)
346 {
347 optptr = (unsigned char *)iph + opt->rr;
348 memcpy(&optptr[optptr[2]-5], &dev2->pa_addr, 4);
349 opt->is_changed = 1;
350 }
351 if (opt->srr_is_hit)
352 {
353 int srrptr, srrspace;
354
355 optptr = (unsigned char *)iph + opt->srr;
356
357 for ( srrptr=optptr[2], srrspace = optptr[1];
358 srrptr <= srrspace;
359 srrptr += 4
360 )
361 {
362 if (srrptr + 3 > srrspace)
363 break;
364 if (memcmp(&target_addr, &optptr[srrptr-1], 4) == 0)
365 break;
366 }
367 if (srrptr + 3 <= srrspace)
368 {
369 opt->is_changed = 1;
370 memcpy(&optptr[srrptr-1], &dev2->pa_addr, 4);
371 iph->daddr = target_addr;
372 optptr[2] = srrptr+4;
373 }
374 else
375 printk("ip_forward(): Argh! Destination lost!\n");
376 }
377 if (opt->ts_needaddr)
378 {
379 optptr = (unsigned char *)iph + opt->ts;
380 memcpy(&optptr[optptr[2]-9], &dev2->pa_addr, 4);
381 opt->is_changed = 1;
382 }
383 if (opt->is_changed)
384 {
385 opt->is_changed = 0;
386 ip_send_check(iph);
387 }
388 }
389 /*
390 * ANK: this is point of "no return", we cannot send an ICMP,
391 * because we changed SRR option.
392 */
393
394 /*
395 * See if it needs fragmenting. Note in ip_rcv we tagged
396 * the fragment type. This must be right so that
397 * the fragmenter does the right thing.
398 */
399
400 if(skb2->len > dev2->mtu + dev2->hard_header_len)
401 {
402 ip_fragment(NULL,skb2,dev2, is_frag);
403 kfree_skb(skb2,FREE_WRITE);
404 }
405 else
406 {
407 #ifdef CONFIG_IP_ACCT
408 /*
409 * Count mapping we shortcut
410 */
411
412 ip_fw_chk(iph,dev,ip_acct_chain,IP_FW_F_ACCEPT,1);
413 #endif
414
415 /*
416 * Map service types to priority. We lie about
417 * throughput being low priority, but it's a good
418 * choice to help improve general usage.
419 */
420 if(iph->tos & IPTOS_LOWDELAY)
421 dev_queue_xmit(skb2, dev2, SOPRI_INTERACTIVE);
422 else if(iph->tos & IPTOS_THROUGHPUT)
423 dev_queue_xmit(skb2, dev2, SOPRI_BACKGROUND);
424 else
425 dev_queue_xmit(skb2, dev2, SOPRI_NORMAL);
426 }
427 }
428 else
429 {
430 ip_rt_put(rt);
431 return -1;
432 }
433 ip_rt_put(rt);
434
435 /*
436 * Tell the caller if their buffer is free.
437 */
438
439 if(skb==skb2)
440 return 0;
441
442 #ifdef CONFIG_IP_MASQUERADE
443 /*
444 * The original is free. Free our copy and
445 * tell the caller not to free.
446 */
447 if(skb!=skb_in)
448 {
449 kfree_skb(skb_in, FREE_WRITE);
450 return 0;
451 }
452 #endif
453 return 1;
454 }
455
456
457 #endif