]>
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 | ||
9aada7ac HE |
16 | #define KMSG_COMPONENT "IPVS" |
17 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt | |
18 | ||
1da177e4 | 19 | #include <linux/kernel.h> |
1da177e4 | 20 | #include <linux/tcp.h> /* for tcphdr */ |
c439cb2e | 21 | #include <net/ip.h> |
1da177e4 LT |
22 | #include <net/tcp.h> /* for csum_tcpudp_magic */ |
23 | #include <net/udp.h> | |
24 | #include <net/icmp.h> /* for icmp_send */ | |
25 | #include <net/route.h> /* for ip_route_output */ | |
38cdcc9a JV |
26 | #include <net/ipv6.h> |
27 | #include <net/ip6_route.h> | |
28 | #include <linux/icmpv6.h> | |
1da177e4 LT |
29 | #include <linux/netfilter.h> |
30 | #include <linux/netfilter_ipv4.h> | |
31 | ||
32 | #include <net/ip_vs.h> | |
33 | ||
34 | ||
35 | /* | |
36 | * Destination cache to speed up outgoing route lookup | |
37 | */ | |
38 | static inline void | |
39 | __ip_vs_dst_set(struct ip_vs_dest *dest, u32 rtos, struct dst_entry *dst) | |
40 | { | |
41 | struct dst_entry *old_dst; | |
42 | ||
43 | old_dst = dest->dst_cache; | |
44 | dest->dst_cache = dst; | |
45 | dest->dst_rtos = rtos; | |
46 | dst_release(old_dst); | |
47 | } | |
48 | ||
49 | static inline struct dst_entry * | |
50 | __ip_vs_dst_check(struct ip_vs_dest *dest, u32 rtos, u32 cookie) | |
51 | { | |
52 | struct dst_entry *dst = dest->dst_cache; | |
53 | ||
54 | if (!dst) | |
55 | return NULL; | |
38cdcc9a JV |
56 | if ((dst->obsolete |
57 | || (dest->af == AF_INET && rtos != dest->dst_rtos)) && | |
1da177e4 LT |
58 | dst->ops->check(dst, cookie) == NULL) { |
59 | dest->dst_cache = NULL; | |
60 | dst_release(dst); | |
61 | return NULL; | |
62 | } | |
63 | dst_hold(dst); | |
64 | return dst; | |
65 | } | |
66 | ||
ad1b30b1 | 67 | static struct rtable * |
1da177e4 LT |
68 | __ip_vs_get_out_rt(struct ip_vs_conn *cp, u32 rtos) |
69 | { | |
70 | struct rtable *rt; /* Route to the other host */ | |
71 | struct ip_vs_dest *dest = cp->dest; | |
72 | ||
73 | if (dest) { | |
74 | spin_lock(&dest->dst_lock); | |
75 | if (!(rt = (struct rtable *) | |
76 | __ip_vs_dst_check(dest, rtos, 0))) { | |
77 | struct flowi fl = { | |
78 | .oif = 0, | |
79 | .nl_u = { | |
80 | .ip4_u = { | |
e7ade46a | 81 | .daddr = dest->addr.ip, |
1da177e4 LT |
82 | .saddr = 0, |
83 | .tos = rtos, } }, | |
84 | }; | |
85 | ||
f206351a | 86 | if (ip_route_output_key(&init_net, &rt, &fl)) { |
1da177e4 | 87 | spin_unlock(&dest->dst_lock); |
14d5e834 HH |
88 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", |
89 | &dest->addr.ip); | |
1da177e4 LT |
90 | return NULL; |
91 | } | |
92 | __ip_vs_dst_set(dest, rtos, dst_clone(&rt->u.dst)); | |
14d5e834 HH |
93 | IP_VS_DBG(10, "new dst %pI4, refcnt=%d, rtos=%X\n", |
94 | &dest->addr.ip, | |
1da177e4 LT |
95 | atomic_read(&rt->u.dst.__refcnt), rtos); |
96 | } | |
97 | spin_unlock(&dest->dst_lock); | |
98 | } else { | |
99 | struct flowi fl = { | |
100 | .oif = 0, | |
101 | .nl_u = { | |
102 | .ip4_u = { | |
e7ade46a | 103 | .daddr = cp->daddr.ip, |
1da177e4 LT |
104 | .saddr = 0, |
105 | .tos = rtos, } }, | |
106 | }; | |
107 | ||
f206351a | 108 | if (ip_route_output_key(&init_net, &rt, &fl)) { |
14d5e834 HH |
109 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", |
110 | &cp->daddr.ip); | |
1da177e4 LT |
111 | return NULL; |
112 | } | |
113 | } | |
114 | ||
115 | return rt; | |
116 | } | |
117 | ||
38cdcc9a JV |
118 | #ifdef CONFIG_IP_VS_IPV6 |
119 | static struct rt6_info * | |
120 | __ip_vs_get_out_rt_v6(struct ip_vs_conn *cp) | |
121 | { | |
122 | struct rt6_info *rt; /* Route to the other host */ | |
123 | struct ip_vs_dest *dest = cp->dest; | |
124 | ||
125 | if (dest) { | |
126 | spin_lock(&dest->dst_lock); | |
127 | rt = (struct rt6_info *)__ip_vs_dst_check(dest, 0, 0); | |
128 | if (!rt) { | |
129 | struct flowi fl = { | |
130 | .oif = 0, | |
131 | .nl_u = { | |
132 | .ip6_u = { | |
133 | .daddr = dest->addr.in6, | |
134 | .saddr = { | |
135 | .s6_addr32 = | |
136 | { 0, 0, 0, 0 }, | |
137 | }, | |
138 | }, | |
139 | }, | |
140 | }; | |
141 | ||
142 | rt = (struct rt6_info *)ip6_route_output(&init_net, | |
143 | NULL, &fl); | |
144 | if (!rt) { | |
145 | spin_unlock(&dest->dst_lock); | |
5b095d98 | 146 | IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", |
38ff4fa4 | 147 | &dest->addr.in6); |
38cdcc9a JV |
148 | return NULL; |
149 | } | |
150 | __ip_vs_dst_set(dest, 0, dst_clone(&rt->u.dst)); | |
5b095d98 | 151 | IP_VS_DBG(10, "new dst %pI6, refcnt=%d\n", |
38ff4fa4 | 152 | &dest->addr.in6, |
38cdcc9a JV |
153 | atomic_read(&rt->u.dst.__refcnt)); |
154 | } | |
155 | spin_unlock(&dest->dst_lock); | |
156 | } else { | |
157 | struct flowi fl = { | |
158 | .oif = 0, | |
159 | .nl_u = { | |
160 | .ip6_u = { | |
161 | .daddr = cp->daddr.in6, | |
162 | .saddr = { | |
163 | .s6_addr32 = { 0, 0, 0, 0 }, | |
164 | }, | |
165 | }, | |
166 | }, | |
167 | }; | |
168 | ||
169 | rt = (struct rt6_info *)ip6_route_output(&init_net, NULL, &fl); | |
170 | if (!rt) { | |
5b095d98 | 171 | IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", |
38ff4fa4 | 172 | &cp->daddr.in6); |
38cdcc9a JV |
173 | return NULL; |
174 | } | |
175 | } | |
176 | ||
177 | return rt; | |
178 | } | |
179 | #endif | |
180 | ||
1da177e4 LT |
181 | |
182 | /* | |
183 | * Release dest->dst_cache before a dest is removed | |
184 | */ | |
185 | void | |
186 | ip_vs_dst_reset(struct ip_vs_dest *dest) | |
187 | { | |
188 | struct dst_entry *old_dst; | |
189 | ||
190 | old_dst = dest->dst_cache; | |
191 | dest->dst_cache = NULL; | |
192 | dst_release(old_dst); | |
193 | } | |
194 | ||
38cdcc9a | 195 | #define IP_VS_XMIT(pf, skb, rt) \ |
1da177e4 | 196 | do { \ |
6869c4d8 | 197 | (skb)->ipvs_property = 1; \ |
ccc7911f | 198 | skb_forward_csum(skb); \ |
38cdcc9a | 199 | NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \ |
1da177e4 LT |
200 | (rt)->u.dst.dev, dst_output); \ |
201 | } while (0) | |
202 | ||
203 | ||
204 | /* | |
205 | * NULL transmitter (do nothing except return NF_ACCEPT) | |
206 | */ | |
207 | int | |
208 | ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
209 | struct ip_vs_protocol *pp) | |
210 | { | |
211 | /* we do not touch skb and do not need pskb ptr */ | |
212 | return NF_ACCEPT; | |
213 | } | |
214 | ||
215 | ||
216 | /* | |
217 | * Bypass transmitter | |
218 | * Let packets bypass the destination when the destination is not | |
219 | * available, it may be only used in transparent cache cluster. | |
220 | */ | |
221 | int | |
222 | ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
223 | struct ip_vs_protocol *pp) | |
224 | { | |
225 | struct rtable *rt; /* Route to the other host */ | |
eddc9ec5 | 226 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
227 | u8 tos = iph->tos; |
228 | int mtu; | |
229 | struct flowi fl = { | |
230 | .oif = 0, | |
231 | .nl_u = { | |
232 | .ip4_u = { | |
233 | .daddr = iph->daddr, | |
234 | .saddr = 0, | |
235 | .tos = RT_TOS(tos), } }, | |
236 | }; | |
237 | ||
238 | EnterFunction(10); | |
239 | ||
f206351a | 240 | if (ip_route_output_key(&init_net, &rt, &fl)) { |
1e3e238e HE |
241 | IP_VS_DBG_RL("%s(): ip_route_output error, dest: %pI4\n", |
242 | __func__, &iph->daddr); | |
1da177e4 LT |
243 | goto tx_error_icmp; |
244 | } | |
245 | ||
246 | /* MTU checking */ | |
247 | mtu = dst_mtu(&rt->u.dst); | |
4412ec49 | 248 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) { |
1da177e4 LT |
249 | ip_rt_put(rt); |
250 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
1e3e238e | 251 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
1da177e4 LT |
252 | goto tx_error; |
253 | } | |
254 | ||
255 | /* | |
256 | * Call ip_send_check because we are not sure it is called | |
257 | * after ip_defrag. Is copy-on-write needed? | |
258 | */ | |
259 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
260 | ip_rt_put(rt); | |
261 | return NF_STOLEN; | |
262 | } | |
eddc9ec5 | 263 | ip_send_check(ip_hdr(skb)); |
1da177e4 LT |
264 | |
265 | /* drop old route */ | |
adf30907 ED |
266 | skb_dst_drop(skb); |
267 | skb_dst_set(skb, &rt->u.dst); | |
1da177e4 LT |
268 | |
269 | /* Another hack: avoid icmp_send in ip_fragment */ | |
270 | skb->local_df = 1; | |
271 | ||
38cdcc9a | 272 | IP_VS_XMIT(PF_INET, skb, rt); |
1da177e4 LT |
273 | |
274 | LeaveFunction(10); | |
275 | return NF_STOLEN; | |
276 | ||
277 | tx_error_icmp: | |
278 | dst_link_failure(skb); | |
279 | tx_error: | |
280 | kfree_skb(skb); | |
281 | LeaveFunction(10); | |
282 | return NF_STOLEN; | |
283 | } | |
284 | ||
b3cdd2a7 JV |
285 | #ifdef CONFIG_IP_VS_IPV6 |
286 | int | |
287 | ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
288 | struct ip_vs_protocol *pp) | |
289 | { | |
290 | struct rt6_info *rt; /* Route to the other host */ | |
291 | struct ipv6hdr *iph = ipv6_hdr(skb); | |
292 | int mtu; | |
293 | struct flowi fl = { | |
294 | .oif = 0, | |
295 | .nl_u = { | |
296 | .ip6_u = { | |
297 | .daddr = iph->daddr, | |
298 | .saddr = { .s6_addr32 = {0, 0, 0, 0} }, } }, | |
299 | }; | |
300 | ||
301 | EnterFunction(10); | |
302 | ||
303 | rt = (struct rt6_info *)ip6_route_output(&init_net, NULL, &fl); | |
304 | if (!rt) { | |
1e3e238e HE |
305 | IP_VS_DBG_RL("%s(): ip6_route_output error, dest: %pI6\n", |
306 | __func__, &iph->daddr); | |
b3cdd2a7 JV |
307 | goto tx_error_icmp; |
308 | } | |
309 | ||
310 | /* MTU checking */ | |
311 | mtu = dst_mtu(&rt->u.dst); | |
312 | if (skb->len > mtu) { | |
313 | dst_release(&rt->u.dst); | |
314 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); | |
1e3e238e | 315 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
b3cdd2a7 JV |
316 | goto tx_error; |
317 | } | |
318 | ||
319 | /* | |
320 | * Call ip_send_check because we are not sure it is called | |
321 | * after ip_defrag. Is copy-on-write needed? | |
322 | */ | |
323 | skb = skb_share_check(skb, GFP_ATOMIC); | |
324 | if (unlikely(skb == NULL)) { | |
325 | dst_release(&rt->u.dst); | |
326 | return NF_STOLEN; | |
327 | } | |
328 | ||
329 | /* drop old route */ | |
adf30907 ED |
330 | skb_dst_drop(skb); |
331 | skb_dst_set(skb, &rt->u.dst); | |
b3cdd2a7 JV |
332 | |
333 | /* Another hack: avoid icmp_send in ip_fragment */ | |
334 | skb->local_df = 1; | |
335 | ||
336 | IP_VS_XMIT(PF_INET6, skb, rt); | |
337 | ||
338 | LeaveFunction(10); | |
339 | return NF_STOLEN; | |
340 | ||
341 | tx_error_icmp: | |
342 | dst_link_failure(skb); | |
343 | tx_error: | |
344 | kfree_skb(skb); | |
345 | LeaveFunction(10); | |
346 | return NF_STOLEN; | |
347 | } | |
348 | #endif | |
1da177e4 LT |
349 | |
350 | /* | |
351 | * NAT transmitter (only for outside-to-inside nat forwarding) | |
352 | * Not used for related ICMP | |
353 | */ | |
354 | int | |
355 | ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
356 | struct ip_vs_protocol *pp) | |
357 | { | |
358 | struct rtable *rt; /* Route to the other host */ | |
359 | int mtu; | |
eddc9ec5 | 360 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
361 | |
362 | EnterFunction(10); | |
363 | ||
364 | /* check if it is a connection of no-client-port */ | |
365 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
014d730d | 366 | __be16 _pt, *p; |
1da177e4 LT |
367 | p = skb_header_pointer(skb, iph->ihl*4, sizeof(_pt), &_pt); |
368 | if (p == NULL) | |
369 | goto tx_error; | |
370 | ip_vs_conn_fill_cport(cp, *p); | |
371 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
372 | } | |
373 | ||
374 | if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(iph->tos)))) | |
375 | goto tx_error_icmp; | |
376 | ||
377 | /* MTU checking */ | |
378 | mtu = dst_mtu(&rt->u.dst); | |
4412ec49 | 379 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) { |
1da177e4 LT |
380 | ip_rt_put(rt); |
381 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
382 | IP_VS_DBG_RL_PKT(0, pp, skb, 0, "ip_vs_nat_xmit(): frag needed for"); | |
383 | goto tx_error; | |
384 | } | |
385 | ||
386 | /* copy-on-write the packet before mangling it */ | |
af1e1cf0 | 387 | if (!skb_make_writable(skb, sizeof(struct iphdr))) |
1da177e4 LT |
388 | goto tx_error_put; |
389 | ||
390 | if (skb_cow(skb, rt->u.dst.dev->hard_header_len)) | |
391 | goto tx_error_put; | |
392 | ||
393 | /* drop old route */ | |
adf30907 ED |
394 | skb_dst_drop(skb); |
395 | skb_dst_set(skb, &rt->u.dst); | |
1da177e4 LT |
396 | |
397 | /* mangle the packet */ | |
3db05fea | 398 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) |
1da177e4 | 399 | goto tx_error; |
e7ade46a | 400 | ip_hdr(skb)->daddr = cp->daddr.ip; |
eddc9ec5 | 401 | ip_send_check(ip_hdr(skb)); |
1da177e4 LT |
402 | |
403 | IP_VS_DBG_PKT(10, pp, skb, 0, "After DNAT"); | |
404 | ||
405 | /* FIXME: when application helper enlarges the packet and the length | |
406 | is larger than the MTU of outgoing device, there will be still | |
407 | MTU problem. */ | |
408 | ||
409 | /* Another hack: avoid icmp_send in ip_fragment */ | |
410 | skb->local_df = 1; | |
411 | ||
38cdcc9a | 412 | IP_VS_XMIT(PF_INET, skb, rt); |
1da177e4 LT |
413 | |
414 | LeaveFunction(10); | |
415 | return NF_STOLEN; | |
416 | ||
417 | tx_error_icmp: | |
418 | dst_link_failure(skb); | |
419 | tx_error: | |
420 | LeaveFunction(10); | |
421 | kfree_skb(skb); | |
422 | return NF_STOLEN; | |
423 | tx_error_put: | |
424 | ip_rt_put(rt); | |
425 | goto tx_error; | |
426 | } | |
427 | ||
b3cdd2a7 JV |
428 | #ifdef CONFIG_IP_VS_IPV6 |
429 | int | |
430 | ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
431 | struct ip_vs_protocol *pp) | |
432 | { | |
433 | struct rt6_info *rt; /* Route to the other host */ | |
434 | int mtu; | |
435 | ||
436 | EnterFunction(10); | |
437 | ||
438 | /* check if it is a connection of no-client-port */ | |
439 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
440 | __be16 _pt, *p; | |
441 | p = skb_header_pointer(skb, sizeof(struct ipv6hdr), | |
442 | sizeof(_pt), &_pt); | |
443 | if (p == NULL) | |
444 | goto tx_error; | |
445 | ip_vs_conn_fill_cport(cp, *p); | |
446 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
447 | } | |
448 | ||
449 | rt = __ip_vs_get_out_rt_v6(cp); | |
450 | if (!rt) | |
451 | goto tx_error_icmp; | |
452 | ||
453 | /* MTU checking */ | |
454 | mtu = dst_mtu(&rt->u.dst); | |
455 | if (skb->len > mtu) { | |
456 | dst_release(&rt->u.dst); | |
457 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); | |
458 | IP_VS_DBG_RL_PKT(0, pp, skb, 0, | |
459 | "ip_vs_nat_xmit_v6(): frag needed for"); | |
460 | goto tx_error; | |
461 | } | |
462 | ||
463 | /* copy-on-write the packet before mangling it */ | |
464 | if (!skb_make_writable(skb, sizeof(struct ipv6hdr))) | |
465 | goto tx_error_put; | |
466 | ||
467 | if (skb_cow(skb, rt->u.dst.dev->hard_header_len)) | |
468 | goto tx_error_put; | |
469 | ||
470 | /* drop old route */ | |
adf30907 ED |
471 | skb_dst_drop(skb); |
472 | skb_dst_set(skb, &rt->u.dst); | |
b3cdd2a7 JV |
473 | |
474 | /* mangle the packet */ | |
475 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) | |
476 | goto tx_error; | |
477 | ipv6_hdr(skb)->daddr = cp->daddr.in6; | |
478 | ||
479 | IP_VS_DBG_PKT(10, pp, skb, 0, "After DNAT"); | |
480 | ||
481 | /* FIXME: when application helper enlarges the packet and the length | |
482 | is larger than the MTU of outgoing device, there will be still | |
483 | MTU problem. */ | |
484 | ||
485 | /* Another hack: avoid icmp_send in ip_fragment */ | |
486 | skb->local_df = 1; | |
487 | ||
488 | IP_VS_XMIT(PF_INET6, skb, rt); | |
489 | ||
490 | LeaveFunction(10); | |
491 | return NF_STOLEN; | |
492 | ||
493 | tx_error_icmp: | |
494 | dst_link_failure(skb); | |
495 | tx_error: | |
496 | LeaveFunction(10); | |
497 | kfree_skb(skb); | |
498 | return NF_STOLEN; | |
499 | tx_error_put: | |
500 | dst_release(&rt->u.dst); | |
501 | goto tx_error; | |
502 | } | |
503 | #endif | |
504 | ||
1da177e4 LT |
505 | |
506 | /* | |
507 | * IP Tunneling transmitter | |
508 | * | |
509 | * This function encapsulates the packet in a new IP packet, its | |
510 | * destination will be set to cp->daddr. Most code of this function | |
511 | * is taken from ipip.c. | |
512 | * | |
513 | * It is used in VS/TUN cluster. The load balancer selects a real | |
514 | * server from a cluster based on a scheduling algorithm, | |
515 | * encapsulates the request packet and forwards it to the selected | |
516 | * server. For example, all real servers are configured with | |
517 | * "ifconfig tunl0 <Virtual IP Address> up". When the server receives | |
518 | * the encapsulated packet, it will decapsulate the packet, processe | |
519 | * the request and return the response packets directly to the client | |
520 | * without passing the load balancer. This can greatly increase the | |
521 | * scalability of virtual server. | |
522 | * | |
523 | * Used for ANY protocol | |
524 | */ | |
525 | int | |
526 | ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
527 | struct ip_vs_protocol *pp) | |
528 | { | |
529 | struct rtable *rt; /* Route to the other host */ | |
530 | struct net_device *tdev; /* Device to other host */ | |
eddc9ec5 | 531 | struct iphdr *old_iph = ip_hdr(skb); |
1da177e4 | 532 | u8 tos = old_iph->tos; |
76ab608d | 533 | __be16 df = old_iph->frag_off; |
2e07fa9c | 534 | sk_buff_data_t old_transport_header = skb->transport_header; |
1da177e4 | 535 | struct iphdr *iph; /* Our new IP header */ |
c2636b4d | 536 | unsigned int max_headroom; /* The extra header space needed */ |
1da177e4 LT |
537 | int mtu; |
538 | ||
539 | EnterFunction(10); | |
540 | ||
4412ec49 | 541 | if (skb->protocol != htons(ETH_P_IP)) { |
1e3e238e | 542 | IP_VS_DBG_RL("%s(): protocol error, " |
1da177e4 | 543 | "ETH_P_IP: %d, skb protocol: %d\n", |
1e3e238e | 544 | __func__, htons(ETH_P_IP), skb->protocol); |
1da177e4 LT |
545 | goto tx_error; |
546 | } | |
547 | ||
548 | if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(tos)))) | |
549 | goto tx_error_icmp; | |
550 | ||
551 | tdev = rt->u.dst.dev; | |
552 | ||
553 | mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr); | |
554 | if (mtu < 68) { | |
555 | ip_rt_put(rt); | |
1e3e238e | 556 | IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__); |
1da177e4 LT |
557 | goto tx_error; |
558 | } | |
adf30907 ED |
559 | if (skb_dst(skb)) |
560 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
1da177e4 | 561 | |
4412ec49 | 562 | df |= (old_iph->frag_off & htons(IP_DF)); |
1da177e4 | 563 | |
4412ec49 | 564 | if ((old_iph->frag_off & htons(IP_DF)) |
1da177e4 LT |
565 | && mtu < ntohs(old_iph->tot_len)) { |
566 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
567 | ip_rt_put(rt); | |
1e3e238e | 568 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
1da177e4 LT |
569 | goto tx_error; |
570 | } | |
571 | ||
572 | /* | |
573 | * Okay, now see if we can stuff it in the buffer as-is. | |
574 | */ | |
575 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr); | |
576 | ||
577 | if (skb_headroom(skb) < max_headroom | |
578 | || skb_cloned(skb) || skb_shared(skb)) { | |
579 | struct sk_buff *new_skb = | |
580 | skb_realloc_headroom(skb, max_headroom); | |
581 | if (!new_skb) { | |
582 | ip_rt_put(rt); | |
583 | kfree_skb(skb); | |
1e3e238e | 584 | IP_VS_ERR_RL("%s(): no memory\n", __func__); |
1da177e4 LT |
585 | return NF_STOLEN; |
586 | } | |
587 | kfree_skb(skb); | |
588 | skb = new_skb; | |
eddc9ec5 | 589 | old_iph = ip_hdr(skb); |
1da177e4 LT |
590 | } |
591 | ||
b0e380b1 | 592 | skb->transport_header = old_transport_header; |
1da177e4 LT |
593 | |
594 | /* fix old IP header checksum */ | |
595 | ip_send_check(old_iph); | |
596 | ||
e2d1bca7 ACM |
597 | skb_push(skb, sizeof(struct iphdr)); |
598 | skb_reset_network_header(skb); | |
1da177e4 LT |
599 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); |
600 | ||
601 | /* drop old route */ | |
adf30907 ED |
602 | skb_dst_drop(skb); |
603 | skb_dst_set(skb, &rt->u.dst); | |
1da177e4 LT |
604 | |
605 | /* | |
606 | * Push down and install the IPIP header. | |
607 | */ | |
eddc9ec5 | 608 | iph = ip_hdr(skb); |
1da177e4 LT |
609 | iph->version = 4; |
610 | iph->ihl = sizeof(struct iphdr)>>2; | |
611 | iph->frag_off = df; | |
612 | iph->protocol = IPPROTO_IPIP; | |
613 | iph->tos = tos; | |
614 | iph->daddr = rt->rt_dst; | |
615 | iph->saddr = rt->rt_src; | |
616 | iph->ttl = old_iph->ttl; | |
1da177e4 | 617 | ip_select_ident(iph, &rt->u.dst, NULL); |
1da177e4 LT |
618 | |
619 | /* Another hack: avoid icmp_send in ip_fragment */ | |
620 | skb->local_df = 1; | |
621 | ||
c439cb2e | 622 | ip_local_out(skb); |
1da177e4 LT |
623 | |
624 | LeaveFunction(10); | |
625 | ||
626 | return NF_STOLEN; | |
627 | ||
628 | tx_error_icmp: | |
629 | dst_link_failure(skb); | |
630 | tx_error: | |
631 | kfree_skb(skb); | |
632 | LeaveFunction(10); | |
633 | return NF_STOLEN; | |
634 | } | |
635 | ||
b3cdd2a7 JV |
636 | #ifdef CONFIG_IP_VS_IPV6 |
637 | int | |
638 | ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
639 | struct ip_vs_protocol *pp) | |
640 | { | |
641 | struct rt6_info *rt; /* Route to the other host */ | |
642 | struct net_device *tdev; /* Device to other host */ | |
643 | struct ipv6hdr *old_iph = ipv6_hdr(skb); | |
644 | sk_buff_data_t old_transport_header = skb->transport_header; | |
645 | struct ipv6hdr *iph; /* Our new IP header */ | |
646 | unsigned int max_headroom; /* The extra header space needed */ | |
647 | int mtu; | |
648 | ||
649 | EnterFunction(10); | |
650 | ||
651 | if (skb->protocol != htons(ETH_P_IPV6)) { | |
1e3e238e | 652 | IP_VS_DBG_RL("%s(): protocol error, " |
b3cdd2a7 | 653 | "ETH_P_IPV6: %d, skb protocol: %d\n", |
1e3e238e | 654 | __func__, htons(ETH_P_IPV6), skb->protocol); |
b3cdd2a7 JV |
655 | goto tx_error; |
656 | } | |
657 | ||
658 | rt = __ip_vs_get_out_rt_v6(cp); | |
659 | if (!rt) | |
660 | goto tx_error_icmp; | |
661 | ||
662 | tdev = rt->u.dst.dev; | |
663 | ||
664 | mtu = dst_mtu(&rt->u.dst) - sizeof(struct ipv6hdr); | |
665 | /* TODO IPv6: do we need this check in IPv6? */ | |
666 | if (mtu < 1280) { | |
667 | dst_release(&rt->u.dst); | |
1e3e238e | 668 | IP_VS_DBG_RL("%s(): mtu less than 1280\n", __func__); |
b3cdd2a7 JV |
669 | goto tx_error; |
670 | } | |
adf30907 ED |
671 | if (skb_dst(skb)) |
672 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
b3cdd2a7 JV |
673 | |
674 | if (mtu < ntohs(old_iph->payload_len) + sizeof(struct ipv6hdr)) { | |
675 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); | |
676 | dst_release(&rt->u.dst); | |
1e3e238e | 677 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
b3cdd2a7 JV |
678 | goto tx_error; |
679 | } | |
680 | ||
681 | /* | |
682 | * Okay, now see if we can stuff it in the buffer as-is. | |
683 | */ | |
684 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr); | |
685 | ||
686 | if (skb_headroom(skb) < max_headroom | |
687 | || skb_cloned(skb) || skb_shared(skb)) { | |
688 | struct sk_buff *new_skb = | |
689 | skb_realloc_headroom(skb, max_headroom); | |
690 | if (!new_skb) { | |
691 | dst_release(&rt->u.dst); | |
692 | kfree_skb(skb); | |
1e3e238e | 693 | IP_VS_ERR_RL("%s(): no memory\n", __func__); |
b3cdd2a7 JV |
694 | return NF_STOLEN; |
695 | } | |
696 | kfree_skb(skb); | |
697 | skb = new_skb; | |
698 | old_iph = ipv6_hdr(skb); | |
699 | } | |
700 | ||
701 | skb->transport_header = old_transport_header; | |
702 | ||
703 | skb_push(skb, sizeof(struct ipv6hdr)); | |
704 | skb_reset_network_header(skb); | |
705 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); | |
706 | ||
707 | /* drop old route */ | |
adf30907 ED |
708 | skb_dst_drop(skb); |
709 | skb_dst_set(skb, &rt->u.dst); | |
b3cdd2a7 JV |
710 | |
711 | /* | |
712 | * Push down and install the IPIP header. | |
713 | */ | |
714 | iph = ipv6_hdr(skb); | |
715 | iph->version = 6; | |
716 | iph->nexthdr = IPPROTO_IPV6; | |
b7b45f47 HH |
717 | iph->payload_len = old_iph->payload_len; |
718 | be16_add_cpu(&iph->payload_len, sizeof(*old_iph)); | |
b3cdd2a7 JV |
719 | iph->priority = old_iph->priority; |
720 | memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl)); | |
721 | iph->daddr = rt->rt6i_dst.addr; | |
722 | iph->saddr = cp->vaddr.in6; /* rt->rt6i_src.addr; */ | |
723 | iph->hop_limit = old_iph->hop_limit; | |
724 | ||
725 | /* Another hack: avoid icmp_send in ip_fragment */ | |
726 | skb->local_df = 1; | |
727 | ||
728 | ip6_local_out(skb); | |
729 | ||
730 | LeaveFunction(10); | |
731 | ||
732 | return NF_STOLEN; | |
733 | ||
734 | tx_error_icmp: | |
735 | dst_link_failure(skb); | |
736 | tx_error: | |
737 | kfree_skb(skb); | |
738 | LeaveFunction(10); | |
739 | return NF_STOLEN; | |
740 | } | |
741 | #endif | |
742 | ||
1da177e4 LT |
743 | |
744 | /* | |
745 | * Direct Routing transmitter | |
746 | * Used for ANY protocol | |
747 | */ | |
748 | int | |
749 | ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
750 | struct ip_vs_protocol *pp) | |
751 | { | |
752 | struct rtable *rt; /* Route to the other host */ | |
eddc9ec5 | 753 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
754 | int mtu; |
755 | ||
756 | EnterFunction(10); | |
757 | ||
758 | if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(iph->tos)))) | |
759 | goto tx_error_icmp; | |
760 | ||
761 | /* MTU checking */ | |
762 | mtu = dst_mtu(&rt->u.dst); | |
4412ec49 | 763 | if ((iph->frag_off & htons(IP_DF)) && skb->len > mtu) { |
1da177e4 LT |
764 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); |
765 | ip_rt_put(rt); | |
1e3e238e | 766 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
1da177e4 LT |
767 | goto tx_error; |
768 | } | |
769 | ||
770 | /* | |
771 | * Call ip_send_check because we are not sure it is called | |
772 | * after ip_defrag. Is copy-on-write needed? | |
773 | */ | |
774 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
775 | ip_rt_put(rt); | |
776 | return NF_STOLEN; | |
777 | } | |
eddc9ec5 | 778 | ip_send_check(ip_hdr(skb)); |
1da177e4 LT |
779 | |
780 | /* drop old route */ | |
adf30907 ED |
781 | skb_dst_drop(skb); |
782 | skb_dst_set(skb, &rt->u.dst); | |
1da177e4 LT |
783 | |
784 | /* Another hack: avoid icmp_send in ip_fragment */ | |
785 | skb->local_df = 1; | |
786 | ||
38cdcc9a | 787 | IP_VS_XMIT(PF_INET, skb, rt); |
1da177e4 LT |
788 | |
789 | LeaveFunction(10); | |
790 | return NF_STOLEN; | |
791 | ||
792 | tx_error_icmp: | |
793 | dst_link_failure(skb); | |
794 | tx_error: | |
795 | kfree_skb(skb); | |
796 | LeaveFunction(10); | |
797 | return NF_STOLEN; | |
798 | } | |
799 | ||
b3cdd2a7 JV |
800 | #ifdef CONFIG_IP_VS_IPV6 |
801 | int | |
802 | ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
803 | struct ip_vs_protocol *pp) | |
804 | { | |
805 | struct rt6_info *rt; /* Route to the other host */ | |
806 | int mtu; | |
807 | ||
808 | EnterFunction(10); | |
809 | ||
810 | rt = __ip_vs_get_out_rt_v6(cp); | |
811 | if (!rt) | |
812 | goto tx_error_icmp; | |
813 | ||
814 | /* MTU checking */ | |
815 | mtu = dst_mtu(&rt->u.dst); | |
816 | if (skb->len > mtu) { | |
817 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); | |
818 | dst_release(&rt->u.dst); | |
1e3e238e | 819 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
b3cdd2a7 JV |
820 | goto tx_error; |
821 | } | |
822 | ||
823 | /* | |
824 | * Call ip_send_check because we are not sure it is called | |
825 | * after ip_defrag. Is copy-on-write needed? | |
826 | */ | |
827 | skb = skb_share_check(skb, GFP_ATOMIC); | |
828 | if (unlikely(skb == NULL)) { | |
829 | dst_release(&rt->u.dst); | |
830 | return NF_STOLEN; | |
831 | } | |
832 | ||
833 | /* drop old route */ | |
adf30907 ED |
834 | skb_dst_drop(skb); |
835 | skb_dst_set(skb, &rt->u.dst); | |
b3cdd2a7 JV |
836 | |
837 | /* Another hack: avoid icmp_send in ip_fragment */ | |
838 | skb->local_df = 1; | |
839 | ||
840 | IP_VS_XMIT(PF_INET6, skb, rt); | |
841 | ||
842 | LeaveFunction(10); | |
843 | return NF_STOLEN; | |
844 | ||
845 | tx_error_icmp: | |
846 | dst_link_failure(skb); | |
847 | tx_error: | |
848 | kfree_skb(skb); | |
849 | LeaveFunction(10); | |
850 | return NF_STOLEN; | |
851 | } | |
852 | #endif | |
853 | ||
1da177e4 LT |
854 | |
855 | /* | |
856 | * ICMP packet transmitter | |
857 | * called by the ip_vs_in_icmp | |
858 | */ | |
859 | int | |
860 | ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
861 | struct ip_vs_protocol *pp, int offset) | |
862 | { | |
863 | struct rtable *rt; /* Route to the other host */ | |
864 | int mtu; | |
865 | int rc; | |
866 | ||
867 | EnterFunction(10); | |
868 | ||
869 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
870 | forwarded directly here, because there is no need to | |
871 | translate address/port back */ | |
872 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
873 | if (cp->packet_xmit) | |
874 | rc = cp->packet_xmit(skb, cp, pp); | |
875 | else | |
876 | rc = NF_ACCEPT; | |
877 | /* do not touch skb anymore */ | |
878 | atomic_inc(&cp->in_pkts); | |
1da177e4 LT |
879 | goto out; |
880 | } | |
881 | ||
882 | /* | |
883 | * mangle and send the packet here (only for VS/NAT) | |
884 | */ | |
885 | ||
eddc9ec5 | 886 | if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(ip_hdr(skb)->tos)))) |
1da177e4 LT |
887 | goto tx_error_icmp; |
888 | ||
889 | /* MTU checking */ | |
890 | mtu = dst_mtu(&rt->u.dst); | |
eddc9ec5 | 891 | if ((skb->len > mtu) && (ip_hdr(skb)->frag_off & htons(IP_DF))) { |
1da177e4 LT |
892 | ip_rt_put(rt); |
893 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); | |
1e3e238e | 894 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
1da177e4 LT |
895 | goto tx_error; |
896 | } | |
897 | ||
898 | /* copy-on-write the packet before mangling it */ | |
af1e1cf0 | 899 | if (!skb_make_writable(skb, offset)) |
1da177e4 LT |
900 | goto tx_error_put; |
901 | ||
902 | if (skb_cow(skb, rt->u.dst.dev->hard_header_len)) | |
903 | goto tx_error_put; | |
904 | ||
905 | /* drop the old route when skb is not shared */ | |
adf30907 ED |
906 | skb_dst_drop(skb); |
907 | skb_dst_set(skb, &rt->u.dst); | |
1da177e4 LT |
908 | |
909 | ip_vs_nat_icmp(skb, pp, cp, 0); | |
910 | ||
911 | /* Another hack: avoid icmp_send in ip_fragment */ | |
912 | skb->local_df = 1; | |
913 | ||
38cdcc9a | 914 | IP_VS_XMIT(PF_INET, skb, rt); |
1da177e4 LT |
915 | |
916 | rc = NF_STOLEN; | |
917 | goto out; | |
918 | ||
919 | tx_error_icmp: | |
920 | dst_link_failure(skb); | |
921 | tx_error: | |
922 | dev_kfree_skb(skb); | |
923 | rc = NF_STOLEN; | |
924 | out: | |
925 | LeaveFunction(10); | |
926 | return rc; | |
927 | tx_error_put: | |
928 | ip_rt_put(rt); | |
929 | goto tx_error; | |
930 | } | |
b3cdd2a7 JV |
931 | |
932 | #ifdef CONFIG_IP_VS_IPV6 | |
933 | int | |
934 | ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
935 | struct ip_vs_protocol *pp, int offset) | |
936 | { | |
937 | struct rt6_info *rt; /* Route to the other host */ | |
938 | int mtu; | |
939 | int rc; | |
940 | ||
941 | EnterFunction(10); | |
942 | ||
943 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
944 | forwarded directly here, because there is no need to | |
945 | translate address/port back */ | |
946 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
947 | if (cp->packet_xmit) | |
948 | rc = cp->packet_xmit(skb, cp, pp); | |
949 | else | |
950 | rc = NF_ACCEPT; | |
951 | /* do not touch skb anymore */ | |
952 | atomic_inc(&cp->in_pkts); | |
953 | goto out; | |
954 | } | |
955 | ||
956 | /* | |
957 | * mangle and send the packet here (only for VS/NAT) | |
958 | */ | |
959 | ||
960 | rt = __ip_vs_get_out_rt_v6(cp); | |
961 | if (!rt) | |
962 | goto tx_error_icmp; | |
963 | ||
964 | /* MTU checking */ | |
965 | mtu = dst_mtu(&rt->u.dst); | |
966 | if (skb->len > mtu) { | |
967 | dst_release(&rt->u.dst); | |
968 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); | |
1e3e238e | 969 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
b3cdd2a7 JV |
970 | goto tx_error; |
971 | } | |
972 | ||
973 | /* copy-on-write the packet before mangling it */ | |
974 | if (!skb_make_writable(skb, offset)) | |
975 | goto tx_error_put; | |
976 | ||
977 | if (skb_cow(skb, rt->u.dst.dev->hard_header_len)) | |
978 | goto tx_error_put; | |
979 | ||
980 | /* drop the old route when skb is not shared */ | |
adf30907 ED |
981 | skb_dst_drop(skb); |
982 | skb_dst_set(skb, &rt->u.dst); | |
b3cdd2a7 JV |
983 | |
984 | ip_vs_nat_icmp_v6(skb, pp, cp, 0); | |
985 | ||
986 | /* Another hack: avoid icmp_send in ip_fragment */ | |
987 | skb->local_df = 1; | |
988 | ||
989 | IP_VS_XMIT(PF_INET6, skb, rt); | |
990 | ||
991 | rc = NF_STOLEN; | |
992 | goto out; | |
993 | ||
994 | tx_error_icmp: | |
995 | dst_link_failure(skb); | |
996 | tx_error: | |
997 | dev_kfree_skb(skb); | |
998 | rc = NF_STOLEN; | |
999 | out: | |
1000 | LeaveFunction(10); | |
1001 | return rc; | |
1002 | tx_error_put: | |
1003 | dst_release(&rt->u.dst); | |
1004 | goto tx_error; | |
1005 | } | |
1006 | #endif |