]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * ip_vs_xmit.c: various packet transmitters for IPVS | |
3 | * | |
1da177e4 LT |
4 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> |
5 | * Julian Anastasov <ja@ssi.bg> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License | |
9 | * as published by the Free Software Foundation; either version | |
10 | * 2 of the License, or (at your option) any later version. | |
11 | * | |
12 | * Changes: | |
13 | * | |
14 | */ | |
15 | ||
16 | #include <linux/kernel.h> | |
1da177e4 | 17 | #include <linux/tcp.h> /* for tcphdr */ |
c439cb2e | 18 | #include <net/ip.h> |
1da177e4 LT |
19 | #include <net/tcp.h> /* for csum_tcpudp_magic */ |
20 | #include <net/udp.h> | |
21 | #include <net/icmp.h> /* for icmp_send */ | |
22 | #include <net/route.h> /* for ip_route_output */ | |
38cdcc9a JV |
23 | #include <net/ipv6.h> |
24 | #include <net/ip6_route.h> | |
25 | #include <linux/icmpv6.h> | |
1da177e4 LT |
26 | #include <linux/netfilter.h> |
27 | #include <linux/netfilter_ipv4.h> | |
28 | ||
29 | #include <net/ip_vs.h> | |
30 | ||
31 | ||
32 | /* | |
33 | * Destination cache to speed up outgoing route lookup | |
34 | */ | |
35 | static inline void | |
36 | __ip_vs_dst_set(struct ip_vs_dest *dest, u32 rtos, struct dst_entry *dst) | |
37 | { | |
38 | struct dst_entry *old_dst; | |
39 | ||
40 | old_dst = dest->dst_cache; | |
41 | dest->dst_cache = dst; | |
42 | dest->dst_rtos = rtos; | |
43 | dst_release(old_dst); | |
44 | } | |
45 | ||
46 | static inline struct dst_entry * | |
47 | __ip_vs_dst_check(struct ip_vs_dest *dest, u32 rtos, u32 cookie) | |
48 | { | |
49 | struct dst_entry *dst = dest->dst_cache; | |
50 | ||
51 | if (!dst) | |
52 | return NULL; | |
38cdcc9a JV |
53 | if ((dst->obsolete |
54 | || (dest->af == AF_INET && rtos != dest->dst_rtos)) && | |
1da177e4 LT |
55 | dst->ops->check(dst, cookie) == NULL) { |
56 | dest->dst_cache = NULL; | |
57 | dst_release(dst); | |
58 | return NULL; | |
59 | } | |
60 | dst_hold(dst); | |
61 | return dst; | |
62 | } | |
63 | ||
ad1b30b1 | 64 | static struct rtable * |
1da177e4 LT |
65 | __ip_vs_get_out_rt(struct ip_vs_conn *cp, u32 rtos) |
66 | { | |
67 | struct rtable *rt; /* Route to the other host */ | |
68 | struct ip_vs_dest *dest = cp->dest; | |
69 | ||
70 | if (dest) { | |
71 | spin_lock(&dest->dst_lock); | |
72 | if (!(rt = (struct rtable *) | |
73 | __ip_vs_dst_check(dest, rtos, 0))) { | |
74 | struct flowi fl = { | |
75 | .oif = 0, | |
76 | .nl_u = { | |
77 | .ip4_u = { | |
e7ade46a | 78 | .daddr = dest->addr.ip, |
1da177e4 LT |
79 | .saddr = 0, |
80 | .tos = rtos, } }, | |
81 | }; | |
82 | ||
f206351a | 83 | if (ip_route_output_key(&init_net, &rt, &fl)) { |
1da177e4 | 84 | spin_unlock(&dest->dst_lock); |
14d5e834 HH |
85 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", |
86 | &dest->addr.ip); | |
1da177e4 LT |
87 | return NULL; |
88 | } | |
89 | __ip_vs_dst_set(dest, rtos, dst_clone(&rt->u.dst)); | |
14d5e834 HH |
90 | IP_VS_DBG(10, "new dst %pI4, refcnt=%d, rtos=%X\n", |
91 | &dest->addr.ip, | |
1da177e4 LT |
92 | atomic_read(&rt->u.dst.__refcnt), rtos); |
93 | } | |
94 | spin_unlock(&dest->dst_lock); | |
95 | } else { | |
96 | struct flowi fl = { | |
97 | .oif = 0, | |
98 | .nl_u = { | |
99 | .ip4_u = { | |
e7ade46a | 100 | .daddr = cp->daddr.ip, |
1da177e4 LT |
101 | .saddr = 0, |
102 | .tos = rtos, } }, | |
103 | }; | |
104 | ||
f206351a | 105 | if (ip_route_output_key(&init_net, &rt, &fl)) { |
14d5e834 HH |
106 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", |
107 | &cp->daddr.ip); | |
1da177e4 LT |
108 | return NULL; |
109 | } | |
110 | } | |
111 | ||
112 | return rt; | |
113 | } | |
114 | ||
38cdcc9a JV |
115 | #ifdef CONFIG_IP_VS_IPV6 |
116 | static struct rt6_info * | |
117 | __ip_vs_get_out_rt_v6(struct ip_vs_conn *cp) | |
118 | { | |
119 | struct rt6_info *rt; /* Route to the other host */ | |
120 | struct ip_vs_dest *dest = cp->dest; | |
121 | ||
122 | if (dest) { | |
123 | spin_lock(&dest->dst_lock); | |
124 | rt = (struct rt6_info *)__ip_vs_dst_check(dest, 0, 0); | |
125 | if (!rt) { | |
126 | struct flowi fl = { | |
127 | .oif = 0, | |
128 | .nl_u = { | |
129 | .ip6_u = { | |
130 | .daddr = dest->addr.in6, | |
131 | .saddr = { | |
132 | .s6_addr32 = | |
133 | { 0, 0, 0, 0 }, | |
134 | }, | |
135 | }, | |
136 | }, | |
137 | }; | |
138 | ||
139 | rt = (struct rt6_info *)ip6_route_output(&init_net, | |
140 | NULL, &fl); | |
141 | if (!rt) { | |
142 | spin_unlock(&dest->dst_lock); | |
5b095d98 | 143 | IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", |
38ff4fa4 | 144 | &dest->addr.in6); |
38cdcc9a JV |
145 | return NULL; |
146 | } | |
147 | __ip_vs_dst_set(dest, 0, dst_clone(&rt->u.dst)); | |
5b095d98 | 148 | IP_VS_DBG(10, "new dst %pI6, refcnt=%d\n", |
38ff4fa4 | 149 | &dest->addr.in6, |
38cdcc9a JV |
150 | atomic_read(&rt->u.dst.__refcnt)); |
151 | } | |
152 | spin_unlock(&dest->dst_lock); | |
153 | } else { | |
154 | struct flowi fl = { | |
155 | .oif = 0, | |
156 | .nl_u = { | |
157 | .ip6_u = { | |
158 | .daddr = cp->daddr.in6, | |
159 | .saddr = { | |
160 | .s6_addr32 = { 0, 0, 0, 0 }, | |
161 | }, | |
162 | }, | |
163 | }, | |
164 | }; | |
165 | ||
166 | rt = (struct rt6_info *)ip6_route_output(&init_net, NULL, &fl); | |
167 | if (!rt) { | |
5b095d98 | 168 | IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", |
38ff4fa4 | 169 | &cp->daddr.in6); |
38cdcc9a JV |
170 | return NULL; |
171 | } | |
172 | } | |
173 | ||
174 | return rt; | |
175 | } | |
176 | #endif | |
177 | ||
1da177e4 LT |
178 | |
179 | /* | |
180 | * Release dest->dst_cache before a dest is removed | |
181 | */ | |
182 | void | |
183 | ip_vs_dst_reset(struct ip_vs_dest *dest) | |
184 | { | |
185 | struct dst_entry *old_dst; | |
186 | ||
187 | old_dst = dest->dst_cache; | |
188 | dest->dst_cache = NULL; | |
189 | dst_release(old_dst); | |
190 | } | |
191 | ||
38cdcc9a | 192 | #define IP_VS_XMIT(pf, skb, rt) \ |
1da177e4 | 193 | do { \ |
6869c4d8 | 194 | (skb)->ipvs_property = 1; \ |
ccc7911f | 195 | skb_forward_csum(skb); \ |
38cdcc9a | 196 | NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \ |
1da177e4 LT |
197 | (rt)->u.dst.dev, dst_output); \ |
198 | } while (0) | |
199 | ||
200 | ||
201 | /* | |
202 | * NULL transmitter (do nothing except return NF_ACCEPT) | |
203 | */ | |
204 | int | |
205 | ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
206 | struct ip_vs_protocol *pp) | |
207 | { | |
208 | /* we do not touch skb and do not need pskb ptr */ | |
209 | return NF_ACCEPT; | |
210 | } | |
211 | ||
212 | ||
213 | /* | |
214 | * Bypass transmitter | |
215 | * Let packets bypass the destination when the destination is not | |
216 | * available, it may be only used in transparent cache cluster. | |
217 | */ | |
218 | int | |
219 | ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
220 | struct ip_vs_protocol *pp) | |
221 | { | |
222 | struct rtable *rt; /* Route to the other host */ | |
eddc9ec5 | 223 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
224 | u8 tos = iph->tos; |
225 | int mtu; | |
226 | struct flowi fl = { | |
227 | .oif = 0, | |
228 | .nl_u = { | |
229 | .ip4_u = { | |
230 | .daddr = iph->daddr, | |
231 | .saddr = 0, | |
232 | .tos = RT_TOS(tos), } }, | |
233 | }; | |
234 | ||
235 | EnterFunction(10); | |
236 | ||
f206351a | 237 | if (ip_route_output_key(&init_net, &rt, &fl)) { |
14d5e834 HH |
238 | IP_VS_DBG_RL("ip_vs_bypass_xmit(): ip_route_output error, dest: %pI4\n", |
239 | &iph->daddr); | |
1da177e4 LT |
240 | goto tx_error_icmp; |
241 | } | |
242 | ||
243 | /* MTU checking */ | |
244 | mtu = dst_mtu(&rt->u.dst); | |
4412ec49 | 245 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) { |
1da177e4 LT |
246 | ip_rt_put(rt); |
247 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
248 | IP_VS_DBG_RL("ip_vs_bypass_xmit(): frag needed\n"); | |
249 | goto tx_error; | |
250 | } | |
251 | ||
252 | /* | |
253 | * Call ip_send_check because we are not sure it is called | |
254 | * after ip_defrag. Is copy-on-write needed? | |
255 | */ | |
256 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
257 | ip_rt_put(rt); | |
258 | return NF_STOLEN; | |
259 | } | |
eddc9ec5 | 260 | ip_send_check(ip_hdr(skb)); |
1da177e4 LT |
261 | |
262 | /* drop old route */ | |
adf30907 ED |
263 | skb_dst_drop(skb); |
264 | skb_dst_set(skb, &rt->u.dst); | |
1da177e4 LT |
265 | |
266 | /* Another hack: avoid icmp_send in ip_fragment */ | |
267 | skb->local_df = 1; | |
268 | ||
38cdcc9a | 269 | IP_VS_XMIT(PF_INET, skb, rt); |
1da177e4 LT |
270 | |
271 | LeaveFunction(10); | |
272 | return NF_STOLEN; | |
273 | ||
274 | tx_error_icmp: | |
275 | dst_link_failure(skb); | |
276 | tx_error: | |
277 | kfree_skb(skb); | |
278 | LeaveFunction(10); | |
279 | return NF_STOLEN; | |
280 | } | |
281 | ||
b3cdd2a7 JV |
282 | #ifdef CONFIG_IP_VS_IPV6 |
283 | int | |
284 | ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
285 | struct ip_vs_protocol *pp) | |
286 | { | |
287 | struct rt6_info *rt; /* Route to the other host */ | |
288 | struct ipv6hdr *iph = ipv6_hdr(skb); | |
289 | int mtu; | |
290 | struct flowi fl = { | |
291 | .oif = 0, | |
292 | .nl_u = { | |
293 | .ip6_u = { | |
294 | .daddr = iph->daddr, | |
295 | .saddr = { .s6_addr32 = {0, 0, 0, 0} }, } }, | |
296 | }; | |
297 | ||
298 | EnterFunction(10); | |
299 | ||
300 | rt = (struct rt6_info *)ip6_route_output(&init_net, NULL, &fl); | |
301 | if (!rt) { | |
5b095d98 | 302 | IP_VS_DBG_RL("ip_vs_bypass_xmit_v6(): ip6_route_output error, dest: %pI6\n", |
38ff4fa4 | 303 | &iph->daddr); |
b3cdd2a7 JV |
304 | goto tx_error_icmp; |
305 | } | |
306 | ||
307 | /* MTU checking */ | |
308 | mtu = dst_mtu(&rt->u.dst); | |
309 | if (skb->len > mtu) { | |
310 | dst_release(&rt->u.dst); | |
311 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); | |
312 | IP_VS_DBG_RL("ip_vs_bypass_xmit_v6(): frag needed\n"); | |
313 | goto tx_error; | |
314 | } | |
315 | ||
316 | /* | |
317 | * Call ip_send_check because we are not sure it is called | |
318 | * after ip_defrag. Is copy-on-write needed? | |
319 | */ | |
320 | skb = skb_share_check(skb, GFP_ATOMIC); | |
321 | if (unlikely(skb == NULL)) { | |
322 | dst_release(&rt->u.dst); | |
323 | return NF_STOLEN; | |
324 | } | |
325 | ||
326 | /* drop old route */ | |
adf30907 ED |
327 | skb_dst_drop(skb); |
328 | skb_dst_set(skb, &rt->u.dst); | |
b3cdd2a7 JV |
329 | |
330 | /* Another hack: avoid icmp_send in ip_fragment */ | |
331 | skb->local_df = 1; | |
332 | ||
333 | IP_VS_XMIT(PF_INET6, skb, rt); | |
334 | ||
335 | LeaveFunction(10); | |
336 | return NF_STOLEN; | |
337 | ||
338 | tx_error_icmp: | |
339 | dst_link_failure(skb); | |
340 | tx_error: | |
341 | kfree_skb(skb); | |
342 | LeaveFunction(10); | |
343 | return NF_STOLEN; | |
344 | } | |
345 | #endif | |
1da177e4 LT |
346 | |
347 | /* | |
348 | * NAT transmitter (only for outside-to-inside nat forwarding) | |
349 | * Not used for related ICMP | |
350 | */ | |
351 | int | |
352 | ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
353 | struct ip_vs_protocol *pp) | |
354 | { | |
355 | struct rtable *rt; /* Route to the other host */ | |
356 | int mtu; | |
eddc9ec5 | 357 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
358 | |
359 | EnterFunction(10); | |
360 | ||
361 | /* check if it is a connection of no-client-port */ | |
362 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
014d730d | 363 | __be16 _pt, *p; |
1da177e4 LT |
364 | p = skb_header_pointer(skb, iph->ihl*4, sizeof(_pt), &_pt); |
365 | if (p == NULL) | |
366 | goto tx_error; | |
367 | ip_vs_conn_fill_cport(cp, *p); | |
368 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
369 | } | |
370 | ||
371 | if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(iph->tos)))) | |
372 | goto tx_error_icmp; | |
373 | ||
374 | /* MTU checking */ | |
375 | mtu = dst_mtu(&rt->u.dst); | |
4412ec49 | 376 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) { |
1da177e4 LT |
377 | ip_rt_put(rt); |
378 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
379 | IP_VS_DBG_RL_PKT(0, pp, skb, 0, "ip_vs_nat_xmit(): frag needed for"); | |
380 | goto tx_error; | |
381 | } | |
382 | ||
383 | /* copy-on-write the packet before mangling it */ | |
af1e1cf0 | 384 | if (!skb_make_writable(skb, sizeof(struct iphdr))) |
1da177e4 LT |
385 | goto tx_error_put; |
386 | ||
387 | if (skb_cow(skb, rt->u.dst.dev->hard_header_len)) | |
388 | goto tx_error_put; | |
389 | ||
390 | /* drop old route */ | |
adf30907 ED |
391 | skb_dst_drop(skb); |
392 | skb_dst_set(skb, &rt->u.dst); | |
1da177e4 LT |
393 | |
394 | /* mangle the packet */ | |
3db05fea | 395 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) |
1da177e4 | 396 | goto tx_error; |
e7ade46a | 397 | ip_hdr(skb)->daddr = cp->daddr.ip; |
eddc9ec5 | 398 | ip_send_check(ip_hdr(skb)); |
1da177e4 LT |
399 | |
400 | IP_VS_DBG_PKT(10, pp, skb, 0, "After DNAT"); | |
401 | ||
402 | /* FIXME: when application helper enlarges the packet and the length | |
403 | is larger than the MTU of outgoing device, there will be still | |
404 | MTU problem. */ | |
405 | ||
406 | /* Another hack: avoid icmp_send in ip_fragment */ | |
407 | skb->local_df = 1; | |
408 | ||
38cdcc9a | 409 | IP_VS_XMIT(PF_INET, skb, rt); |
1da177e4 LT |
410 | |
411 | LeaveFunction(10); | |
412 | return NF_STOLEN; | |
413 | ||
414 | tx_error_icmp: | |
415 | dst_link_failure(skb); | |
416 | tx_error: | |
417 | LeaveFunction(10); | |
418 | kfree_skb(skb); | |
419 | return NF_STOLEN; | |
420 | tx_error_put: | |
421 | ip_rt_put(rt); | |
422 | goto tx_error; | |
423 | } | |
424 | ||
b3cdd2a7 JV |
425 | #ifdef CONFIG_IP_VS_IPV6 |
426 | int | |
427 | ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
428 | struct ip_vs_protocol *pp) | |
429 | { | |
430 | struct rt6_info *rt; /* Route to the other host */ | |
431 | int mtu; | |
432 | ||
433 | EnterFunction(10); | |
434 | ||
435 | /* check if it is a connection of no-client-port */ | |
436 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
437 | __be16 _pt, *p; | |
438 | p = skb_header_pointer(skb, sizeof(struct ipv6hdr), | |
439 | sizeof(_pt), &_pt); | |
440 | if (p == NULL) | |
441 | goto tx_error; | |
442 | ip_vs_conn_fill_cport(cp, *p); | |
443 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
444 | } | |
445 | ||
446 | rt = __ip_vs_get_out_rt_v6(cp); | |
447 | if (!rt) | |
448 | goto tx_error_icmp; | |
449 | ||
450 | /* MTU checking */ | |
451 | mtu = dst_mtu(&rt->u.dst); | |
452 | if (skb->len > mtu) { | |
453 | dst_release(&rt->u.dst); | |
454 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); | |
455 | IP_VS_DBG_RL_PKT(0, pp, skb, 0, | |
456 | "ip_vs_nat_xmit_v6(): frag needed for"); | |
457 | goto tx_error; | |
458 | } | |
459 | ||
460 | /* copy-on-write the packet before mangling it */ | |
461 | if (!skb_make_writable(skb, sizeof(struct ipv6hdr))) | |
462 | goto tx_error_put; | |
463 | ||
464 | if (skb_cow(skb, rt->u.dst.dev->hard_header_len)) | |
465 | goto tx_error_put; | |
466 | ||
467 | /* drop old route */ | |
adf30907 ED |
468 | skb_dst_drop(skb); |
469 | skb_dst_set(skb, &rt->u.dst); | |
b3cdd2a7 JV |
470 | |
471 | /* mangle the packet */ | |
472 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) | |
473 | goto tx_error; | |
474 | ipv6_hdr(skb)->daddr = cp->daddr.in6; | |
475 | ||
476 | IP_VS_DBG_PKT(10, pp, skb, 0, "After DNAT"); | |
477 | ||
478 | /* FIXME: when application helper enlarges the packet and the length | |
479 | is larger than the MTU of outgoing device, there will be still | |
480 | MTU problem. */ | |
481 | ||
482 | /* Another hack: avoid icmp_send in ip_fragment */ | |
483 | skb->local_df = 1; | |
484 | ||
485 | IP_VS_XMIT(PF_INET6, skb, rt); | |
486 | ||
487 | LeaveFunction(10); | |
488 | return NF_STOLEN; | |
489 | ||
490 | tx_error_icmp: | |
491 | dst_link_failure(skb); | |
492 | tx_error: | |
493 | LeaveFunction(10); | |
494 | kfree_skb(skb); | |
495 | return NF_STOLEN; | |
496 | tx_error_put: | |
497 | dst_release(&rt->u.dst); | |
498 | goto tx_error; | |
499 | } | |
500 | #endif | |
501 | ||
1da177e4 LT |
502 | |
503 | /* | |
504 | * IP Tunneling transmitter | |
505 | * | |
506 | * This function encapsulates the packet in a new IP packet, its | |
507 | * destination will be set to cp->daddr. Most code of this function | |
508 | * is taken from ipip.c. | |
509 | * | |
510 | * It is used in VS/TUN cluster. The load balancer selects a real | |
511 | * server from a cluster based on a scheduling algorithm, | |
512 | * encapsulates the request packet and forwards it to the selected | |
513 | * server. For example, all real servers are configured with | |
514 | * "ifconfig tunl0 <Virtual IP Address> up". When the server receives | |
515 | * the encapsulated packet, it will decapsulate the packet, processe | |
516 | * the request and return the response packets directly to the client | |
517 | * without passing the load balancer. This can greatly increase the | |
518 | * scalability of virtual server. | |
519 | * | |
520 | * Used for ANY protocol | |
521 | */ | |
522 | int | |
523 | ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
524 | struct ip_vs_protocol *pp) | |
525 | { | |
526 | struct rtable *rt; /* Route to the other host */ | |
527 | struct net_device *tdev; /* Device to other host */ | |
eddc9ec5 | 528 | struct iphdr *old_iph = ip_hdr(skb); |
1da177e4 | 529 | u8 tos = old_iph->tos; |
76ab608d | 530 | __be16 df = old_iph->frag_off; |
2e07fa9c | 531 | sk_buff_data_t old_transport_header = skb->transport_header; |
1da177e4 | 532 | struct iphdr *iph; /* Our new IP header */ |
c2636b4d | 533 | unsigned int max_headroom; /* The extra header space needed */ |
1da177e4 LT |
534 | int mtu; |
535 | ||
536 | EnterFunction(10); | |
537 | ||
4412ec49 | 538 | if (skb->protocol != htons(ETH_P_IP)) { |
1da177e4 LT |
539 | IP_VS_DBG_RL("ip_vs_tunnel_xmit(): protocol error, " |
540 | "ETH_P_IP: %d, skb protocol: %d\n", | |
4412ec49 | 541 | htons(ETH_P_IP), skb->protocol); |
1da177e4 LT |
542 | goto tx_error; |
543 | } | |
544 | ||
545 | if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(tos)))) | |
546 | goto tx_error_icmp; | |
547 | ||
548 | tdev = rt->u.dst.dev; | |
549 | ||
550 | mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr); | |
551 | if (mtu < 68) { | |
552 | ip_rt_put(rt); | |
553 | IP_VS_DBG_RL("ip_vs_tunnel_xmit(): mtu less than 68\n"); | |
554 | goto tx_error; | |
555 | } | |
adf30907 ED |
556 | if (skb_dst(skb)) |
557 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
1da177e4 | 558 | |
4412ec49 | 559 | df |= (old_iph->frag_off & htons(IP_DF)); |
1da177e4 | 560 | |
4412ec49 | 561 | if ((old_iph->frag_off & htons(IP_DF)) |
1da177e4 LT |
562 | && mtu < ntohs(old_iph->tot_len)) { |
563 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
564 | ip_rt_put(rt); | |
565 | IP_VS_DBG_RL("ip_vs_tunnel_xmit(): frag needed\n"); | |
566 | goto tx_error; | |
567 | } | |
568 | ||
569 | /* | |
570 | * Okay, now see if we can stuff it in the buffer as-is. | |
571 | */ | |
572 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr); | |
573 | ||
574 | if (skb_headroom(skb) < max_headroom | |
575 | || skb_cloned(skb) || skb_shared(skb)) { | |
576 | struct sk_buff *new_skb = | |
577 | skb_realloc_headroom(skb, max_headroom); | |
578 | if (!new_skb) { | |
579 | ip_rt_put(rt); | |
580 | kfree_skb(skb); | |
581 | IP_VS_ERR_RL("ip_vs_tunnel_xmit(): no memory\n"); | |
582 | return NF_STOLEN; | |
583 | } | |
584 | kfree_skb(skb); | |
585 | skb = new_skb; | |
eddc9ec5 | 586 | old_iph = ip_hdr(skb); |
1da177e4 LT |
587 | } |
588 | ||
b0e380b1 | 589 | skb->transport_header = old_transport_header; |
1da177e4 LT |
590 | |
591 | /* fix old IP header checksum */ | |
592 | ip_send_check(old_iph); | |
593 | ||
e2d1bca7 ACM |
594 | skb_push(skb, sizeof(struct iphdr)); |
595 | skb_reset_network_header(skb); | |
1da177e4 LT |
596 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); |
597 | ||
598 | /* drop old route */ | |
adf30907 ED |
599 | skb_dst_drop(skb); |
600 | skb_dst_set(skb, &rt->u.dst); | |
1da177e4 LT |
601 | |
602 | /* | |
603 | * Push down and install the IPIP header. | |
604 | */ | |
eddc9ec5 | 605 | iph = ip_hdr(skb); |
1da177e4 LT |
606 | iph->version = 4; |
607 | iph->ihl = sizeof(struct iphdr)>>2; | |
608 | iph->frag_off = df; | |
609 | iph->protocol = IPPROTO_IPIP; | |
610 | iph->tos = tos; | |
611 | iph->daddr = rt->rt_dst; | |
612 | iph->saddr = rt->rt_src; | |
613 | iph->ttl = old_iph->ttl; | |
1da177e4 | 614 | ip_select_ident(iph, &rt->u.dst, NULL); |
1da177e4 LT |
615 | |
616 | /* Another hack: avoid icmp_send in ip_fragment */ | |
617 | skb->local_df = 1; | |
618 | ||
c439cb2e | 619 | ip_local_out(skb); |
1da177e4 LT |
620 | |
621 | LeaveFunction(10); | |
622 | ||
623 | return NF_STOLEN; | |
624 | ||
625 | tx_error_icmp: | |
626 | dst_link_failure(skb); | |
627 | tx_error: | |
628 | kfree_skb(skb); | |
629 | LeaveFunction(10); | |
630 | return NF_STOLEN; | |
631 | } | |
632 | ||
b3cdd2a7 JV |
633 | #ifdef CONFIG_IP_VS_IPV6 |
634 | int | |
635 | ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
636 | struct ip_vs_protocol *pp) | |
637 | { | |
638 | struct rt6_info *rt; /* Route to the other host */ | |
639 | struct net_device *tdev; /* Device to other host */ | |
640 | struct ipv6hdr *old_iph = ipv6_hdr(skb); | |
641 | sk_buff_data_t old_transport_header = skb->transport_header; | |
642 | struct ipv6hdr *iph; /* Our new IP header */ | |
643 | unsigned int max_headroom; /* The extra header space needed */ | |
644 | int mtu; | |
645 | ||
646 | EnterFunction(10); | |
647 | ||
648 | if (skb->protocol != htons(ETH_P_IPV6)) { | |
649 | IP_VS_DBG_RL("ip_vs_tunnel_xmit_v6(): protocol error, " | |
650 | "ETH_P_IPV6: %d, skb protocol: %d\n", | |
651 | htons(ETH_P_IPV6), skb->protocol); | |
652 | goto tx_error; | |
653 | } | |
654 | ||
655 | rt = __ip_vs_get_out_rt_v6(cp); | |
656 | if (!rt) | |
657 | goto tx_error_icmp; | |
658 | ||
659 | tdev = rt->u.dst.dev; | |
660 | ||
661 | mtu = dst_mtu(&rt->u.dst) - sizeof(struct ipv6hdr); | |
662 | /* TODO IPv6: do we need this check in IPv6? */ | |
663 | if (mtu < 1280) { | |
664 | dst_release(&rt->u.dst); | |
665 | IP_VS_DBG_RL("ip_vs_tunnel_xmit_v6(): mtu less than 1280\n"); | |
666 | goto tx_error; | |
667 | } | |
adf30907 ED |
668 | if (skb_dst(skb)) |
669 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
b3cdd2a7 JV |
670 | |
671 | if (mtu < ntohs(old_iph->payload_len) + sizeof(struct ipv6hdr)) { | |
672 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); | |
673 | dst_release(&rt->u.dst); | |
674 | IP_VS_DBG_RL("ip_vs_tunnel_xmit_v6(): frag needed\n"); | |
675 | goto tx_error; | |
676 | } | |
677 | ||
678 | /* | |
679 | * Okay, now see if we can stuff it in the buffer as-is. | |
680 | */ | |
681 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr); | |
682 | ||
683 | if (skb_headroom(skb) < max_headroom | |
684 | || skb_cloned(skb) || skb_shared(skb)) { | |
685 | struct sk_buff *new_skb = | |
686 | skb_realloc_headroom(skb, max_headroom); | |
687 | if (!new_skb) { | |
688 | dst_release(&rt->u.dst); | |
689 | kfree_skb(skb); | |
690 | IP_VS_ERR_RL("ip_vs_tunnel_xmit_v6(): no memory\n"); | |
691 | return NF_STOLEN; | |
692 | } | |
693 | kfree_skb(skb); | |
694 | skb = new_skb; | |
695 | old_iph = ipv6_hdr(skb); | |
696 | } | |
697 | ||
698 | skb->transport_header = old_transport_header; | |
699 | ||
700 | skb_push(skb, sizeof(struct ipv6hdr)); | |
701 | skb_reset_network_header(skb); | |
702 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); | |
703 | ||
704 | /* drop old route */ | |
adf30907 ED |
705 | skb_dst_drop(skb); |
706 | skb_dst_set(skb, &rt->u.dst); | |
b3cdd2a7 JV |
707 | |
708 | /* | |
709 | * Push down and install the IPIP header. | |
710 | */ | |
711 | iph = ipv6_hdr(skb); | |
712 | iph->version = 6; | |
713 | iph->nexthdr = IPPROTO_IPV6; | |
b7b45f47 HH |
714 | iph->payload_len = old_iph->payload_len; |
715 | be16_add_cpu(&iph->payload_len, sizeof(*old_iph)); | |
b3cdd2a7 JV |
716 | iph->priority = old_iph->priority; |
717 | memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl)); | |
718 | iph->daddr = rt->rt6i_dst.addr; | |
719 | iph->saddr = cp->vaddr.in6; /* rt->rt6i_src.addr; */ | |
720 | iph->hop_limit = old_iph->hop_limit; | |
721 | ||
722 | /* Another hack: avoid icmp_send in ip_fragment */ | |
723 | skb->local_df = 1; | |
724 | ||
725 | ip6_local_out(skb); | |
726 | ||
727 | LeaveFunction(10); | |
728 | ||
729 | return NF_STOLEN; | |
730 | ||
731 | tx_error_icmp: | |
732 | dst_link_failure(skb); | |
733 | tx_error: | |
734 | kfree_skb(skb); | |
735 | LeaveFunction(10); | |
736 | return NF_STOLEN; | |
737 | } | |
738 | #endif | |
739 | ||
1da177e4 LT |
740 | |
741 | /* | |
742 | * Direct Routing transmitter | |
743 | * Used for ANY protocol | |
744 | */ | |
745 | int | |
746 | ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
747 | struct ip_vs_protocol *pp) | |
748 | { | |
749 | struct rtable *rt; /* Route to the other host */ | |
eddc9ec5 | 750 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
751 | int mtu; |
752 | ||
753 | EnterFunction(10); | |
754 | ||
755 | if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(iph->tos)))) | |
756 | goto tx_error_icmp; | |
757 | ||
758 | /* MTU checking */ | |
759 | mtu = dst_mtu(&rt->u.dst); | |
4412ec49 | 760 | if ((iph->frag_off & htons(IP_DF)) && skb->len > mtu) { |
1da177e4 LT |
761 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); |
762 | ip_rt_put(rt); | |
763 | IP_VS_DBG_RL("ip_vs_dr_xmit(): frag needed\n"); | |
764 | goto tx_error; | |
765 | } | |
766 | ||
767 | /* | |
768 | * Call ip_send_check because we are not sure it is called | |
769 | * after ip_defrag. Is copy-on-write needed? | |
770 | */ | |
771 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
772 | ip_rt_put(rt); | |
773 | return NF_STOLEN; | |
774 | } | |
eddc9ec5 | 775 | ip_send_check(ip_hdr(skb)); |
1da177e4 LT |
776 | |
777 | /* drop old route */ | |
adf30907 ED |
778 | skb_dst_drop(skb); |
779 | skb_dst_set(skb, &rt->u.dst); | |
1da177e4 LT |
780 | |
781 | /* Another hack: avoid icmp_send in ip_fragment */ | |
782 | skb->local_df = 1; | |
783 | ||
38cdcc9a | 784 | IP_VS_XMIT(PF_INET, skb, rt); |
1da177e4 LT |
785 | |
786 | LeaveFunction(10); | |
787 | return NF_STOLEN; | |
788 | ||
789 | tx_error_icmp: | |
790 | dst_link_failure(skb); | |
791 | tx_error: | |
792 | kfree_skb(skb); | |
793 | LeaveFunction(10); | |
794 | return NF_STOLEN; | |
795 | } | |
796 | ||
b3cdd2a7 JV |
797 | #ifdef CONFIG_IP_VS_IPV6 |
798 | int | |
799 | ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
800 | struct ip_vs_protocol *pp) | |
801 | { | |
802 | struct rt6_info *rt; /* Route to the other host */ | |
803 | int mtu; | |
804 | ||
805 | EnterFunction(10); | |
806 | ||
807 | rt = __ip_vs_get_out_rt_v6(cp); | |
808 | if (!rt) | |
809 | goto tx_error_icmp; | |
810 | ||
811 | /* MTU checking */ | |
812 | mtu = dst_mtu(&rt->u.dst); | |
813 | if (skb->len > mtu) { | |
814 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); | |
815 | dst_release(&rt->u.dst); | |
816 | IP_VS_DBG_RL("ip_vs_dr_xmit_v6(): frag needed\n"); | |
817 | goto tx_error; | |
818 | } | |
819 | ||
820 | /* | |
821 | * Call ip_send_check because we are not sure it is called | |
822 | * after ip_defrag. Is copy-on-write needed? | |
823 | */ | |
824 | skb = skb_share_check(skb, GFP_ATOMIC); | |
825 | if (unlikely(skb == NULL)) { | |
826 | dst_release(&rt->u.dst); | |
827 | return NF_STOLEN; | |
828 | } | |
829 | ||
830 | /* drop old route */ | |
adf30907 ED |
831 | skb_dst_drop(skb); |
832 | skb_dst_set(skb, &rt->u.dst); | |
b3cdd2a7 JV |
833 | |
834 | /* Another hack: avoid icmp_send in ip_fragment */ | |
835 | skb->local_df = 1; | |
836 | ||
837 | IP_VS_XMIT(PF_INET6, skb, rt); | |
838 | ||
839 | LeaveFunction(10); | |
840 | return NF_STOLEN; | |
841 | ||
842 | tx_error_icmp: | |
843 | dst_link_failure(skb); | |
844 | tx_error: | |
845 | kfree_skb(skb); | |
846 | LeaveFunction(10); | |
847 | return NF_STOLEN; | |
848 | } | |
849 | #endif | |
850 | ||
1da177e4 LT |
851 | |
852 | /* | |
853 | * ICMP packet transmitter | |
854 | * called by the ip_vs_in_icmp | |
855 | */ | |
856 | int | |
857 | ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
858 | struct ip_vs_protocol *pp, int offset) | |
859 | { | |
860 | struct rtable *rt; /* Route to the other host */ | |
861 | int mtu; | |
862 | int rc; | |
863 | ||
864 | EnterFunction(10); | |
865 | ||
866 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
867 | forwarded directly here, because there is no need to | |
868 | translate address/port back */ | |
869 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
870 | if (cp->packet_xmit) | |
871 | rc = cp->packet_xmit(skb, cp, pp); | |
872 | else | |
873 | rc = NF_ACCEPT; | |
874 | /* do not touch skb anymore */ | |
875 | atomic_inc(&cp->in_pkts); | |
1da177e4 LT |
876 | goto out; |
877 | } | |
878 | ||
879 | /* | |
880 | * mangle and send the packet here (only for VS/NAT) | |
881 | */ | |
882 | ||
eddc9ec5 | 883 | if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(ip_hdr(skb)->tos)))) |
1da177e4 LT |
884 | goto tx_error_icmp; |
885 | ||
886 | /* MTU checking */ | |
887 | mtu = dst_mtu(&rt->u.dst); | |
eddc9ec5 | 888 | if ((skb->len > mtu) && (ip_hdr(skb)->frag_off & htons(IP_DF))) { |
1da177e4 LT |
889 | ip_rt_put(rt); |
890 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); | |
891 | IP_VS_DBG_RL("ip_vs_in_icmp(): frag needed\n"); | |
892 | goto tx_error; | |
893 | } | |
894 | ||
895 | /* copy-on-write the packet before mangling it */ | |
af1e1cf0 | 896 | if (!skb_make_writable(skb, offset)) |
1da177e4 LT |
897 | goto tx_error_put; |
898 | ||
899 | if (skb_cow(skb, rt->u.dst.dev->hard_header_len)) | |
900 | goto tx_error_put; | |
901 | ||
902 | /* drop the old route when skb is not shared */ | |
adf30907 ED |
903 | skb_dst_drop(skb); |
904 | skb_dst_set(skb, &rt->u.dst); | |
1da177e4 LT |
905 | |
906 | ip_vs_nat_icmp(skb, pp, cp, 0); | |
907 | ||
908 | /* Another hack: avoid icmp_send in ip_fragment */ | |
909 | skb->local_df = 1; | |
910 | ||
38cdcc9a | 911 | IP_VS_XMIT(PF_INET, skb, rt); |
1da177e4 LT |
912 | |
913 | rc = NF_STOLEN; | |
914 | goto out; | |
915 | ||
916 | tx_error_icmp: | |
917 | dst_link_failure(skb); | |
918 | tx_error: | |
919 | dev_kfree_skb(skb); | |
920 | rc = NF_STOLEN; | |
921 | out: | |
922 | LeaveFunction(10); | |
923 | return rc; | |
924 | tx_error_put: | |
925 | ip_rt_put(rt); | |
926 | goto tx_error; | |
927 | } | |
b3cdd2a7 JV |
928 | |
929 | #ifdef CONFIG_IP_VS_IPV6 | |
930 | int | |
931 | ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
932 | struct ip_vs_protocol *pp, int offset) | |
933 | { | |
934 | struct rt6_info *rt; /* Route to the other host */ | |
935 | int mtu; | |
936 | int rc; | |
937 | ||
938 | EnterFunction(10); | |
939 | ||
940 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
941 | forwarded directly here, because there is no need to | |
942 | translate address/port back */ | |
943 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
944 | if (cp->packet_xmit) | |
945 | rc = cp->packet_xmit(skb, cp, pp); | |
946 | else | |
947 | rc = NF_ACCEPT; | |
948 | /* do not touch skb anymore */ | |
949 | atomic_inc(&cp->in_pkts); | |
950 | goto out; | |
951 | } | |
952 | ||
953 | /* | |
954 | * mangle and send the packet here (only for VS/NAT) | |
955 | */ | |
956 | ||
957 | rt = __ip_vs_get_out_rt_v6(cp); | |
958 | if (!rt) | |
959 | goto tx_error_icmp; | |
960 | ||
961 | /* MTU checking */ | |
962 | mtu = dst_mtu(&rt->u.dst); | |
963 | if (skb->len > mtu) { | |
964 | dst_release(&rt->u.dst); | |
965 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); | |
966 | IP_VS_DBG_RL("ip_vs_in_icmp(): frag needed\n"); | |
967 | goto tx_error; | |
968 | } | |
969 | ||
970 | /* copy-on-write the packet before mangling it */ | |
971 | if (!skb_make_writable(skb, offset)) | |
972 | goto tx_error_put; | |
973 | ||
974 | if (skb_cow(skb, rt->u.dst.dev->hard_header_len)) | |
975 | goto tx_error_put; | |
976 | ||
977 | /* drop the old route when skb is not shared */ | |
adf30907 ED |
978 | skb_dst_drop(skb); |
979 | skb_dst_set(skb, &rt->u.dst); | |
b3cdd2a7 JV |
980 | |
981 | ip_vs_nat_icmp_v6(skb, pp, cp, 0); | |
982 | ||
983 | /* Another hack: avoid icmp_send in ip_fragment */ | |
984 | skb->local_df = 1; | |
985 | ||
986 | IP_VS_XMIT(PF_INET6, skb, rt); | |
987 | ||
988 | rc = NF_STOLEN; | |
989 | goto out; | |
990 | ||
991 | tx_error_icmp: | |
992 | dst_link_failure(skb); | |
993 | tx_error: | |
994 | dev_kfree_skb(skb); | |
995 | rc = NF_STOLEN; | |
996 | out: | |
997 | LeaveFunction(10); | |
998 | return rc; | |
999 | tx_error_put: | |
1000 | dst_release(&rt->u.dst); | |
1001 | goto tx_error; | |
1002 | } | |
1003 | #endif |