]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * ip_vs_xmit.c: various packet transmitters for IPVS | |
3 | * | |
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 | * Description of forwarding methods: | |
15 | * - all transmitters are called from LOCAL_IN (remote clients) and | |
16 | * LOCAL_OUT (local clients) but for ICMP can be called from FORWARD | |
17 | * - not all connections have destination server, for example, | |
18 | * connections in backup server when fwmark is used | |
19 | * - bypass connections use daddr from packet | |
20 | * LOCAL_OUT rules: | |
21 | * - skb->dev is NULL, skb->protocol is not set (both are set in POST_ROUTING) | |
22 | * - skb->pkt_type is not set yet | |
23 | * - the only place where we can see skb->sk != NULL | |
24 | */ | |
25 | ||
26 | #define KMSG_COMPONENT "IPVS" | |
27 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt | |
28 | ||
29 | #include <linux/kernel.h> | |
30 | #include <linux/slab.h> | |
31 | #include <linux/tcp.h> /* for tcphdr */ | |
32 | #include <net/ip.h> | |
33 | #include <net/tcp.h> /* for csum_tcpudp_magic */ | |
34 | #include <net/udp.h> | |
35 | #include <net/icmp.h> /* for icmp_send */ | |
36 | #include <net/route.h> /* for ip_route_output */ | |
37 | #include <net/ipv6.h> | |
38 | #include <net/ip6_route.h> | |
39 | #include <net/addrconf.h> | |
40 | #include <linux/icmpv6.h> | |
41 | #include <linux/netfilter.h> | |
42 | #include <linux/netfilter_ipv4.h> | |
43 | ||
44 | #include <net/ip_vs.h> | |
45 | ||
46 | ||
47 | /* | |
48 | * Destination cache to speed up outgoing route lookup | |
49 | */ | |
50 | static inline void | |
51 | __ip_vs_dst_set(struct ip_vs_dest *dest, u32 rtos, struct dst_entry *dst, | |
52 | u32 dst_cookie) | |
53 | { | |
54 | struct dst_entry *old_dst; | |
55 | ||
56 | old_dst = dest->dst_cache; | |
57 | dest->dst_cache = dst; | |
58 | dest->dst_rtos = rtos; | |
59 | dest->dst_cookie = dst_cookie; | |
60 | dst_release(old_dst); | |
61 | } | |
62 | ||
63 | static inline struct dst_entry * | |
64 | __ip_vs_dst_check(struct ip_vs_dest *dest, u32 rtos) | |
65 | { | |
66 | struct dst_entry *dst = dest->dst_cache; | |
67 | ||
68 | if (!dst) | |
69 | return NULL; | |
70 | if ((dst->obsolete || rtos != dest->dst_rtos) && | |
71 | dst->ops->check(dst, dest->dst_cookie) == NULL) { | |
72 | dest->dst_cache = NULL; | |
73 | dst_release(dst); | |
74 | return NULL; | |
75 | } | |
76 | dst_hold(dst); | |
77 | return dst; | |
78 | } | |
79 | ||
80 | /* | |
81 | * Get route to destination or remote server | |
82 | * rt_mode: flags, &1=Allow local dest, &2=Allow non-local dest, | |
83 | * &4=Allow redirect from remote daddr to local | |
84 | */ | |
85 | static struct rtable * | |
86 | __ip_vs_get_out_rt(struct sk_buff *skb, struct ip_vs_dest *dest, | |
87 | __be32 daddr, u32 rtos, int rt_mode) | |
88 | { | |
89 | struct net *net = dev_net(skb_dst(skb)->dev); | |
90 | struct rtable *rt; /* Route to the other host */ | |
91 | struct rtable *ort; /* Original route */ | |
92 | int local; | |
93 | ||
94 | if (dest) { | |
95 | spin_lock(&dest->dst_lock); | |
96 | if (!(rt = (struct rtable *) | |
97 | __ip_vs_dst_check(dest, rtos))) { | |
98 | struct flowi fl = { | |
99 | .oif = 0, | |
100 | .nl_u = { | |
101 | .ip4_u = { | |
102 | .daddr = dest->addr.ip, | |
103 | .saddr = 0, | |
104 | .tos = rtos, } }, | |
105 | }; | |
106 | ||
107 | if (ip_route_output_key(net, &rt, &fl)) { | |
108 | spin_unlock(&dest->dst_lock); | |
109 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", | |
110 | &dest->addr.ip); | |
111 | return NULL; | |
112 | } | |
113 | __ip_vs_dst_set(dest, rtos, dst_clone(&rt->dst), 0); | |
114 | IP_VS_DBG(10, "new dst %pI4, refcnt=%d, rtos=%X\n", | |
115 | &dest->addr.ip, | |
116 | atomic_read(&rt->dst.__refcnt), rtos); | |
117 | } | |
118 | spin_unlock(&dest->dst_lock); | |
119 | } else { | |
120 | struct flowi fl = { | |
121 | .oif = 0, | |
122 | .nl_u = { | |
123 | .ip4_u = { | |
124 | .daddr = daddr, | |
125 | .saddr = 0, | |
126 | .tos = rtos, } }, | |
127 | }; | |
128 | ||
129 | if (ip_route_output_key(net, &rt, &fl)) { | |
130 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", | |
131 | &daddr); | |
132 | return NULL; | |
133 | } | |
134 | } | |
135 | ||
136 | local = rt->rt_flags & RTCF_LOCAL; | |
137 | if (!((local ? 1 : 2) & rt_mode)) { | |
138 | IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI4\n", | |
139 | (rt->rt_flags & RTCF_LOCAL) ? | |
140 | "local":"non-local", &rt->rt_dst); | |
141 | ip_rt_put(rt); | |
142 | return NULL; | |
143 | } | |
144 | if (local && !(rt_mode & 4) && !((ort = skb_rtable(skb)) && | |
145 | ort->rt_flags & RTCF_LOCAL)) { | |
146 | IP_VS_DBG_RL("Redirect from non-local address %pI4 to local " | |
147 | "requires NAT method, dest: %pI4\n", | |
148 | &ip_hdr(skb)->daddr, &rt->rt_dst); | |
149 | ip_rt_put(rt); | |
150 | return NULL; | |
151 | } | |
152 | if (unlikely(!local && ipv4_is_loopback(ip_hdr(skb)->saddr))) { | |
153 | IP_VS_DBG_RL("Stopping traffic from loopback address %pI4 " | |
154 | "to non-local address, dest: %pI4\n", | |
155 | &ip_hdr(skb)->saddr, &rt->rt_dst); | |
156 | ip_rt_put(rt); | |
157 | return NULL; | |
158 | } | |
159 | ||
160 | return rt; | |
161 | } | |
162 | ||
163 | /* Reroute packet to local IPv4 stack after DNAT */ | |
164 | static int | |
165 | __ip_vs_reroute_locally(struct sk_buff *skb) | |
166 | { | |
167 | struct rtable *rt = skb_rtable(skb); | |
168 | struct net_device *dev = rt->dst.dev; | |
169 | struct net *net = dev_net(dev); | |
170 | struct iphdr *iph = ip_hdr(skb); | |
171 | ||
172 | if (rt->fl.iif) { | |
173 | unsigned long orefdst = skb->_skb_refdst; | |
174 | ||
175 | if (ip_route_input(skb, iph->daddr, iph->saddr, | |
176 | iph->tos, skb->dev)) | |
177 | return 0; | |
178 | refdst_drop(orefdst); | |
179 | } else { | |
180 | struct flowi fl = { | |
181 | .oif = 0, | |
182 | .nl_u = { | |
183 | .ip4_u = { | |
184 | .daddr = iph->daddr, | |
185 | .saddr = iph->saddr, | |
186 | .tos = RT_TOS(iph->tos), | |
187 | } | |
188 | }, | |
189 | .mark = skb->mark, | |
190 | }; | |
191 | struct rtable *rt; | |
192 | ||
193 | if (ip_route_output_key(net, &rt, &fl)) | |
194 | return 0; | |
195 | if (!(rt->rt_flags & RTCF_LOCAL)) { | |
196 | ip_rt_put(rt); | |
197 | return 0; | |
198 | } | |
199 | /* Drop old route. */ | |
200 | skb_dst_drop(skb); | |
201 | skb_dst_set(skb, &rt->dst); | |
202 | } | |
203 | return 1; | |
204 | } | |
205 | ||
206 | #ifdef CONFIG_IP_VS_IPV6 | |
207 | ||
208 | static inline int __ip_vs_is_local_route6(struct rt6_info *rt) | |
209 | { | |
210 | return rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK; | |
211 | } | |
212 | ||
213 | static struct dst_entry * | |
214 | __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr, | |
215 | struct in6_addr *ret_saddr, int do_xfrm) | |
216 | { | |
217 | struct dst_entry *dst; | |
218 | struct flowi fl = { | |
219 | .oif = 0, | |
220 | .nl_u = { | |
221 | .ip6_u = { | |
222 | .daddr = *daddr, | |
223 | }, | |
224 | }, | |
225 | }; | |
226 | ||
227 | dst = ip6_route_output(net, NULL, &fl); | |
228 | if (dst->error) | |
229 | goto out_err; | |
230 | if (!ret_saddr) | |
231 | return dst; | |
232 | if (ipv6_addr_any(&fl.fl6_src) && | |
233 | ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev, | |
234 | &fl.fl6_dst, 0, &fl.fl6_src) < 0) | |
235 | goto out_err; | |
236 | if (do_xfrm && xfrm_lookup(net, &dst, &fl, NULL, 0) < 0) | |
237 | goto out_err; | |
238 | ipv6_addr_copy(ret_saddr, &fl.fl6_src); | |
239 | return dst; | |
240 | ||
241 | out_err: | |
242 | dst_release(dst); | |
243 | IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr); | |
244 | return NULL; | |
245 | } | |
246 | ||
247 | /* | |
248 | * Get route to destination or remote server | |
249 | * rt_mode: flags, &1=Allow local dest, &2=Allow non-local dest, | |
250 | * &4=Allow redirect from remote daddr to local | |
251 | */ | |
252 | static struct rt6_info * | |
253 | __ip_vs_get_out_rt_v6(struct sk_buff *skb, struct ip_vs_dest *dest, | |
254 | struct in6_addr *daddr, struct in6_addr *ret_saddr, | |
255 | int do_xfrm, int rt_mode) | |
256 | { | |
257 | struct net *net = dev_net(skb_dst(skb)->dev); | |
258 | struct rt6_info *rt; /* Route to the other host */ | |
259 | struct rt6_info *ort; /* Original route */ | |
260 | struct dst_entry *dst; | |
261 | int local; | |
262 | ||
263 | if (dest) { | |
264 | spin_lock(&dest->dst_lock); | |
265 | rt = (struct rt6_info *)__ip_vs_dst_check(dest, 0); | |
266 | if (!rt) { | |
267 | u32 cookie; | |
268 | ||
269 | dst = __ip_vs_route_output_v6(net, &dest->addr.in6, | |
270 | &dest->dst_saddr, | |
271 | do_xfrm); | |
272 | if (!dst) { | |
273 | spin_unlock(&dest->dst_lock); | |
274 | return NULL; | |
275 | } | |
276 | rt = (struct rt6_info *) dst; | |
277 | cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0; | |
278 | __ip_vs_dst_set(dest, 0, dst_clone(&rt->dst), cookie); | |
279 | IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n", | |
280 | &dest->addr.in6, &dest->dst_saddr, | |
281 | atomic_read(&rt->dst.__refcnt)); | |
282 | } | |
283 | if (ret_saddr) | |
284 | ipv6_addr_copy(ret_saddr, &dest->dst_saddr); | |
285 | spin_unlock(&dest->dst_lock); | |
286 | } else { | |
287 | dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm); | |
288 | if (!dst) | |
289 | return NULL; | |
290 | rt = (struct rt6_info *) dst; | |
291 | } | |
292 | ||
293 | local = __ip_vs_is_local_route6(rt); | |
294 | if (!((local ? 1 : 2) & rt_mode)) { | |
295 | IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI6\n", | |
296 | local ? "local":"non-local", daddr); | |
297 | dst_release(&rt->dst); | |
298 | return NULL; | |
299 | } | |
300 | if (local && !(rt_mode & 4) && | |
301 | !((ort = (struct rt6_info *) skb_dst(skb)) && | |
302 | __ip_vs_is_local_route6(ort))) { | |
303 | IP_VS_DBG_RL("Redirect from non-local address %pI6 to local " | |
304 | "requires NAT method, dest: %pI6\n", | |
305 | &ipv6_hdr(skb)->daddr, daddr); | |
306 | dst_release(&rt->dst); | |
307 | return NULL; | |
308 | } | |
309 | if (unlikely(!local && (!skb->dev || skb->dev->flags & IFF_LOOPBACK) && | |
310 | ipv6_addr_type(&ipv6_hdr(skb)->saddr) & | |
311 | IPV6_ADDR_LOOPBACK)) { | |
312 | IP_VS_DBG_RL("Stopping traffic from loopback address %pI6 " | |
313 | "to non-local address, dest: %pI6\n", | |
314 | &ipv6_hdr(skb)->saddr, daddr); | |
315 | dst_release(&rt->dst); | |
316 | return NULL; | |
317 | } | |
318 | ||
319 | return rt; | |
320 | } | |
321 | #endif | |
322 | ||
323 | ||
324 | /* | |
325 | * Release dest->dst_cache before a dest is removed | |
326 | */ | |
327 | void | |
328 | ip_vs_dst_reset(struct ip_vs_dest *dest) | |
329 | { | |
330 | struct dst_entry *old_dst; | |
331 | ||
332 | old_dst = dest->dst_cache; | |
333 | dest->dst_cache = NULL; | |
334 | dst_release(old_dst); | |
335 | } | |
336 | ||
337 | #define IP_VS_XMIT_TUNNEL(skb, cp) \ | |
338 | ({ \ | |
339 | int __ret = NF_ACCEPT; \ | |
340 | \ | |
341 | (skb)->ipvs_property = 1; \ | |
342 | if (unlikely((cp)->flags & IP_VS_CONN_F_NFCT)) \ | |
343 | __ret = ip_vs_confirm_conntrack(skb, cp); \ | |
344 | if (__ret == NF_ACCEPT) { \ | |
345 | nf_reset(skb); \ | |
346 | skb_forward_csum(skb); \ | |
347 | } \ | |
348 | __ret; \ | |
349 | }) | |
350 | ||
351 | #define IP_VS_XMIT_NAT(pf, skb, cp, local) \ | |
352 | do { \ | |
353 | (skb)->ipvs_property = 1; \ | |
354 | if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \ | |
355 | ip_vs_notrack(skb); \ | |
356 | else \ | |
357 | ip_vs_update_conntrack(skb, cp, 1); \ | |
358 | if (local) \ | |
359 | return NF_ACCEPT; \ | |
360 | skb_forward_csum(skb); \ | |
361 | NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \ | |
362 | skb_dst(skb)->dev, dst_output); \ | |
363 | } while (0) | |
364 | ||
365 | #define IP_VS_XMIT(pf, skb, cp, local) \ | |
366 | do { \ | |
367 | (skb)->ipvs_property = 1; \ | |
368 | if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \ | |
369 | ip_vs_notrack(skb); \ | |
370 | if (local) \ | |
371 | return NF_ACCEPT; \ | |
372 | skb_forward_csum(skb); \ | |
373 | NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \ | |
374 | skb_dst(skb)->dev, dst_output); \ | |
375 | } while (0) | |
376 | ||
377 | ||
378 | /* | |
379 | * NULL transmitter (do nothing except return NF_ACCEPT) | |
380 | */ | |
381 | int | |
382 | ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
383 | struct ip_vs_protocol *pp) | |
384 | { | |
385 | /* we do not touch skb and do not need pskb ptr */ | |
386 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); | |
387 | } | |
388 | ||
389 | ||
390 | /* | |
391 | * Bypass transmitter | |
392 | * Let packets bypass the destination when the destination is not | |
393 | * available, it may be only used in transparent cache cluster. | |
394 | */ | |
395 | int | |
396 | ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
397 | struct ip_vs_protocol *pp) | |
398 | { | |
399 | struct rtable *rt; /* Route to the other host */ | |
400 | struct iphdr *iph = ip_hdr(skb); | |
401 | int mtu; | |
402 | ||
403 | EnterFunction(10); | |
404 | ||
405 | if (!(rt = __ip_vs_get_out_rt(skb, NULL, iph->daddr, | |
406 | RT_TOS(iph->tos), 2))) | |
407 | goto tx_error_icmp; | |
408 | ||
409 | /* MTU checking */ | |
410 | mtu = dst_mtu(&rt->dst); | |
411 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) { | |
412 | ip_rt_put(rt); | |
413 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
414 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
415 | goto tx_error; | |
416 | } | |
417 | ||
418 | /* | |
419 | * Call ip_send_check because we are not sure it is called | |
420 | * after ip_defrag. Is copy-on-write needed? | |
421 | */ | |
422 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
423 | ip_rt_put(rt); | |
424 | return NF_STOLEN; | |
425 | } | |
426 | ip_send_check(ip_hdr(skb)); | |
427 | ||
428 | /* drop old route */ | |
429 | skb_dst_drop(skb); | |
430 | skb_dst_set(skb, &rt->dst); | |
431 | ||
432 | /* Another hack: avoid icmp_send in ip_fragment */ | |
433 | skb->local_df = 1; | |
434 | ||
435 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0); | |
436 | ||
437 | LeaveFunction(10); | |
438 | return NF_STOLEN; | |
439 | ||
440 | tx_error_icmp: | |
441 | dst_link_failure(skb); | |
442 | tx_error: | |
443 | kfree_skb(skb); | |
444 | LeaveFunction(10); | |
445 | return NF_STOLEN; | |
446 | } | |
447 | ||
448 | #ifdef CONFIG_IP_VS_IPV6 | |
449 | int | |
450 | ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
451 | struct ip_vs_protocol *pp) | |
452 | { | |
453 | struct rt6_info *rt; /* Route to the other host */ | |
454 | struct ipv6hdr *iph = ipv6_hdr(skb); | |
455 | int mtu; | |
456 | ||
457 | EnterFunction(10); | |
458 | ||
459 | if (!(rt = __ip_vs_get_out_rt_v6(skb, NULL, &iph->daddr, NULL, 0, 2))) | |
460 | goto tx_error_icmp; | |
461 | ||
462 | /* MTU checking */ | |
463 | mtu = dst_mtu(&rt->dst); | |
464 | if (skb->len > mtu) { | |
465 | if (!skb->dev) { | |
466 | struct net *net = dev_net(skb_dst(skb)->dev); | |
467 | ||
468 | skb->dev = net->loopback_dev; | |
469 | } | |
470 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); | |
471 | dst_release(&rt->dst); | |
472 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
473 | goto tx_error; | |
474 | } | |
475 | ||
476 | /* | |
477 | * Call ip_send_check because we are not sure it is called | |
478 | * after ip_defrag. Is copy-on-write needed? | |
479 | */ | |
480 | skb = skb_share_check(skb, GFP_ATOMIC); | |
481 | if (unlikely(skb == NULL)) { | |
482 | dst_release(&rt->dst); | |
483 | return NF_STOLEN; | |
484 | } | |
485 | ||
486 | /* drop old route */ | |
487 | skb_dst_drop(skb); | |
488 | skb_dst_set(skb, &rt->dst); | |
489 | ||
490 | /* Another hack: avoid icmp_send in ip_fragment */ | |
491 | skb->local_df = 1; | |
492 | ||
493 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0); | |
494 | ||
495 | LeaveFunction(10); | |
496 | return NF_STOLEN; | |
497 | ||
498 | tx_error_icmp: | |
499 | dst_link_failure(skb); | |
500 | tx_error: | |
501 | kfree_skb(skb); | |
502 | LeaveFunction(10); | |
503 | return NF_STOLEN; | |
504 | } | |
505 | #endif | |
506 | ||
507 | /* | |
508 | * NAT transmitter (only for outside-to-inside nat forwarding) | |
509 | * Not used for related ICMP | |
510 | */ | |
511 | int | |
512 | ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
513 | struct ip_vs_protocol *pp) | |
514 | { | |
515 | struct rtable *rt; /* Route to the other host */ | |
516 | int mtu; | |
517 | struct iphdr *iph = ip_hdr(skb); | |
518 | int local; | |
519 | ||
520 | EnterFunction(10); | |
521 | ||
522 | /* check if it is a connection of no-client-port */ | |
523 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
524 | __be16 _pt, *p; | |
525 | p = skb_header_pointer(skb, iph->ihl*4, sizeof(_pt), &_pt); | |
526 | if (p == NULL) | |
527 | goto tx_error; | |
528 | ip_vs_conn_fill_cport(cp, *p); | |
529 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
530 | } | |
531 | ||
532 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, | |
533 | RT_TOS(iph->tos), 1|2|4))) | |
534 | goto tx_error_icmp; | |
535 | local = rt->rt_flags & RTCF_LOCAL; | |
536 | /* | |
537 | * Avoid duplicate tuple in reply direction for NAT traffic | |
538 | * to local address when connection is sync-ed | |
539 | */ | |
540 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
541 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
542 | enum ip_conntrack_info ctinfo; | |
543 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
544 | ||
545 | if (ct && !nf_ct_is_untracked(ct)) { | |
546 | IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, 0, | |
547 | "ip_vs_nat_xmit(): " | |
548 | "stopping DNAT to local address"); | |
549 | goto tx_error_put; | |
550 | } | |
551 | } | |
552 | #endif | |
553 | ||
554 | /* From world but DNAT to loopback address? */ | |
555 | if (local && ipv4_is_loopback(rt->rt_dst) && skb_rtable(skb)->fl.iif) { | |
556 | IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, 0, "ip_vs_nat_xmit(): " | |
557 | "stopping DNAT to loopback address"); | |
558 | goto tx_error_put; | |
559 | } | |
560 | ||
561 | /* MTU checking */ | |
562 | mtu = dst_mtu(&rt->dst); | |
563 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) { | |
564 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
565 | IP_VS_DBG_RL_PKT(0, AF_INET, pp, skb, 0, | |
566 | "ip_vs_nat_xmit(): frag needed for"); | |
567 | goto tx_error_put; | |
568 | } | |
569 | ||
570 | /* copy-on-write the packet before mangling it */ | |
571 | if (!skb_make_writable(skb, sizeof(struct iphdr))) | |
572 | goto tx_error_put; | |
573 | ||
574 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) | |
575 | goto tx_error_put; | |
576 | ||
577 | /* mangle the packet */ | |
578 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) | |
579 | goto tx_error_put; | |
580 | ip_hdr(skb)->daddr = cp->daddr.ip; | |
581 | ip_send_check(ip_hdr(skb)); | |
582 | ||
583 | if (!local) { | |
584 | /* drop old route */ | |
585 | skb_dst_drop(skb); | |
586 | skb_dst_set(skb, &rt->dst); | |
587 | } else { | |
588 | ip_rt_put(rt); | |
589 | /* | |
590 | * Some IPv4 replies get local address from routes, | |
591 | * not from iph, so while we DNAT after routing | |
592 | * we need this second input/output route. | |
593 | */ | |
594 | if (!__ip_vs_reroute_locally(skb)) | |
595 | goto tx_error; | |
596 | } | |
597 | ||
598 | IP_VS_DBG_PKT(10, AF_INET, pp, skb, 0, "After DNAT"); | |
599 | ||
600 | /* FIXME: when application helper enlarges the packet and the length | |
601 | is larger than the MTU of outgoing device, there will be still | |
602 | MTU problem. */ | |
603 | ||
604 | /* Another hack: avoid icmp_send in ip_fragment */ | |
605 | skb->local_df = 1; | |
606 | ||
607 | IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local); | |
608 | ||
609 | LeaveFunction(10); | |
610 | return NF_STOLEN; | |
611 | ||
612 | tx_error_icmp: | |
613 | dst_link_failure(skb); | |
614 | tx_error: | |
615 | kfree_skb(skb); | |
616 | LeaveFunction(10); | |
617 | return NF_STOLEN; | |
618 | tx_error_put: | |
619 | ip_rt_put(rt); | |
620 | goto tx_error; | |
621 | } | |
622 | ||
623 | #ifdef CONFIG_IP_VS_IPV6 | |
624 | int | |
625 | ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
626 | struct ip_vs_protocol *pp) | |
627 | { | |
628 | struct rt6_info *rt; /* Route to the other host */ | |
629 | int mtu; | |
630 | int local; | |
631 | ||
632 | EnterFunction(10); | |
633 | ||
634 | /* check if it is a connection of no-client-port */ | |
635 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
636 | __be16 _pt, *p; | |
637 | p = skb_header_pointer(skb, sizeof(struct ipv6hdr), | |
638 | sizeof(_pt), &_pt); | |
639 | if (p == NULL) | |
640 | goto tx_error; | |
641 | ip_vs_conn_fill_cport(cp, *p); | |
642 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
643 | } | |
644 | ||
645 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, | |
646 | 0, 1|2|4))) | |
647 | goto tx_error_icmp; | |
648 | local = __ip_vs_is_local_route6(rt); | |
649 | /* | |
650 | * Avoid duplicate tuple in reply direction for NAT traffic | |
651 | * to local address when connection is sync-ed | |
652 | */ | |
653 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
654 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
655 | enum ip_conntrack_info ctinfo; | |
656 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
657 | ||
658 | if (ct && !nf_ct_is_untracked(ct)) { | |
659 | IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, 0, | |
660 | "ip_vs_nat_xmit_v6(): " | |
661 | "stopping DNAT to local address"); | |
662 | goto tx_error_put; | |
663 | } | |
664 | } | |
665 | #endif | |
666 | ||
667 | /* From world but DNAT to loopback address? */ | |
668 | if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) && | |
669 | ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) { | |
670 | IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, 0, | |
671 | "ip_vs_nat_xmit_v6(): " | |
672 | "stopping DNAT to loopback address"); | |
673 | goto tx_error_put; | |
674 | } | |
675 | ||
676 | /* MTU checking */ | |
677 | mtu = dst_mtu(&rt->dst); | |
678 | if (skb->len > mtu) { | |
679 | if (!skb->dev) { | |
680 | struct net *net = dev_net(skb_dst(skb)->dev); | |
681 | ||
682 | skb->dev = net->loopback_dev; | |
683 | } | |
684 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); | |
685 | IP_VS_DBG_RL_PKT(0, AF_INET6, pp, skb, 0, | |
686 | "ip_vs_nat_xmit_v6(): frag needed for"); | |
687 | goto tx_error_put; | |
688 | } | |
689 | ||
690 | /* copy-on-write the packet before mangling it */ | |
691 | if (!skb_make_writable(skb, sizeof(struct ipv6hdr))) | |
692 | goto tx_error_put; | |
693 | ||
694 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) | |
695 | goto tx_error_put; | |
696 | ||
697 | /* mangle the packet */ | |
698 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) | |
699 | goto tx_error; | |
700 | ipv6_addr_copy(&ipv6_hdr(skb)->daddr, &cp->daddr.in6); | |
701 | ||
702 | if (!local || !skb->dev) { | |
703 | /* drop the old route when skb is not shared */ | |
704 | skb_dst_drop(skb); | |
705 | skb_dst_set(skb, &rt->dst); | |
706 | } else { | |
707 | /* destined to loopback, do we need to change route? */ | |
708 | dst_release(&rt->dst); | |
709 | } | |
710 | ||
711 | IP_VS_DBG_PKT(10, AF_INET6, pp, skb, 0, "After DNAT"); | |
712 | ||
713 | /* FIXME: when application helper enlarges the packet and the length | |
714 | is larger than the MTU of outgoing device, there will be still | |
715 | MTU problem. */ | |
716 | ||
717 | /* Another hack: avoid icmp_send in ip_fragment */ | |
718 | skb->local_df = 1; | |
719 | ||
720 | IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local); | |
721 | ||
722 | LeaveFunction(10); | |
723 | return NF_STOLEN; | |
724 | ||
725 | tx_error_icmp: | |
726 | dst_link_failure(skb); | |
727 | tx_error: | |
728 | LeaveFunction(10); | |
729 | kfree_skb(skb); | |
730 | return NF_STOLEN; | |
731 | tx_error_put: | |
732 | dst_release(&rt->dst); | |
733 | goto tx_error; | |
734 | } | |
735 | #endif | |
736 | ||
737 | ||
738 | /* | |
739 | * IP Tunneling transmitter | |
740 | * | |
741 | * This function encapsulates the packet in a new IP packet, its | |
742 | * destination will be set to cp->daddr. Most code of this function | |
743 | * is taken from ipip.c. | |
744 | * | |
745 | * It is used in VS/TUN cluster. The load balancer selects a real | |
746 | * server from a cluster based on a scheduling algorithm, | |
747 | * encapsulates the request packet and forwards it to the selected | |
748 | * server. For example, all real servers are configured with | |
749 | * "ifconfig tunl0 <Virtual IP Address> up". When the server receives | |
750 | * the encapsulated packet, it will decapsulate the packet, processe | |
751 | * the request and return the response packets directly to the client | |
752 | * without passing the load balancer. This can greatly increase the | |
753 | * scalability of virtual server. | |
754 | * | |
755 | * Used for ANY protocol | |
756 | */ | |
757 | int | |
758 | ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
759 | struct ip_vs_protocol *pp) | |
760 | { | |
761 | struct rtable *rt; /* Route to the other host */ | |
762 | struct net_device *tdev; /* Device to other host */ | |
763 | struct iphdr *old_iph = ip_hdr(skb); | |
764 | u8 tos = old_iph->tos; | |
765 | __be16 df = old_iph->frag_off; | |
766 | struct iphdr *iph; /* Our new IP header */ | |
767 | unsigned int max_headroom; /* The extra header space needed */ | |
768 | int mtu; | |
769 | int ret; | |
770 | ||
771 | EnterFunction(10); | |
772 | ||
773 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, | |
774 | RT_TOS(tos), 1|2))) | |
775 | goto tx_error_icmp; | |
776 | if (rt->rt_flags & RTCF_LOCAL) { | |
777 | ip_rt_put(rt); | |
778 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); | |
779 | } | |
780 | ||
781 | tdev = rt->dst.dev; | |
782 | ||
783 | mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr); | |
784 | if (mtu < 68) { | |
785 | IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__); | |
786 | goto tx_error_put; | |
787 | } | |
788 | if (skb_dst(skb)) | |
789 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
790 | ||
791 | df |= (old_iph->frag_off & htons(IP_DF)); | |
792 | ||
793 | if ((old_iph->frag_off & htons(IP_DF)) | |
794 | && mtu < ntohs(old_iph->tot_len)) { | |
795 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
796 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
797 | goto tx_error_put; | |
798 | } | |
799 | ||
800 | /* | |
801 | * Okay, now see if we can stuff it in the buffer as-is. | |
802 | */ | |
803 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr); | |
804 | ||
805 | if (skb_headroom(skb) < max_headroom | |
806 | || skb_cloned(skb) || skb_shared(skb)) { | |
807 | struct sk_buff *new_skb = | |
808 | skb_realloc_headroom(skb, max_headroom); | |
809 | if (!new_skb) { | |
810 | ip_rt_put(rt); | |
811 | kfree_skb(skb); | |
812 | IP_VS_ERR_RL("%s(): no memory\n", __func__); | |
813 | return NF_STOLEN; | |
814 | } | |
815 | kfree_skb(skb); | |
816 | skb = new_skb; | |
817 | old_iph = ip_hdr(skb); | |
818 | } | |
819 | ||
820 | skb->transport_header = skb->network_header; | |
821 | ||
822 | /* fix old IP header checksum */ | |
823 | ip_send_check(old_iph); | |
824 | ||
825 | skb_push(skb, sizeof(struct iphdr)); | |
826 | skb_reset_network_header(skb); | |
827 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); | |
828 | ||
829 | /* drop old route */ | |
830 | skb_dst_drop(skb); | |
831 | skb_dst_set(skb, &rt->dst); | |
832 | ||
833 | /* | |
834 | * Push down and install the IPIP header. | |
835 | */ | |
836 | iph = ip_hdr(skb); | |
837 | iph->version = 4; | |
838 | iph->ihl = sizeof(struct iphdr)>>2; | |
839 | iph->frag_off = df; | |
840 | iph->protocol = IPPROTO_IPIP; | |
841 | iph->tos = tos; | |
842 | iph->daddr = rt->rt_dst; | |
843 | iph->saddr = rt->rt_src; | |
844 | iph->ttl = old_iph->ttl; | |
845 | ip_select_ident(iph, &rt->dst, NULL); | |
846 | ||
847 | /* Another hack: avoid icmp_send in ip_fragment */ | |
848 | skb->local_df = 1; | |
849 | ||
850 | ret = IP_VS_XMIT_TUNNEL(skb, cp); | |
851 | if (ret == NF_ACCEPT) | |
852 | ip_local_out(skb); | |
853 | else if (ret == NF_DROP) | |
854 | kfree_skb(skb); | |
855 | ||
856 | LeaveFunction(10); | |
857 | ||
858 | return NF_STOLEN; | |
859 | ||
860 | tx_error_icmp: | |
861 | dst_link_failure(skb); | |
862 | tx_error: | |
863 | kfree_skb(skb); | |
864 | LeaveFunction(10); | |
865 | return NF_STOLEN; | |
866 | tx_error_put: | |
867 | ip_rt_put(rt); | |
868 | goto tx_error; | |
869 | } | |
870 | ||
871 | #ifdef CONFIG_IP_VS_IPV6 | |
872 | int | |
873 | ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
874 | struct ip_vs_protocol *pp) | |
875 | { | |
876 | struct rt6_info *rt; /* Route to the other host */ | |
877 | struct in6_addr saddr; /* Source for tunnel */ | |
878 | struct net_device *tdev; /* Device to other host */ | |
879 | struct ipv6hdr *old_iph = ipv6_hdr(skb); | |
880 | struct ipv6hdr *iph; /* Our new IP header */ | |
881 | unsigned int max_headroom; /* The extra header space needed */ | |
882 | int mtu; | |
883 | int ret; | |
884 | ||
885 | EnterFunction(10); | |
886 | ||
887 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, | |
888 | &saddr, 1, 1|2))) | |
889 | goto tx_error_icmp; | |
890 | if (__ip_vs_is_local_route6(rt)) { | |
891 | dst_release(&rt->dst); | |
892 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1); | |
893 | } | |
894 | ||
895 | tdev = rt->dst.dev; | |
896 | ||
897 | mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr); | |
898 | if (mtu < IPV6_MIN_MTU) { | |
899 | IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__, | |
900 | IPV6_MIN_MTU); | |
901 | goto tx_error_put; | |
902 | } | |
903 | if (skb_dst(skb)) | |
904 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
905 | ||
906 | if (mtu < ntohs(old_iph->payload_len) + sizeof(struct ipv6hdr)) { | |
907 | if (!skb->dev) { | |
908 | struct net *net = dev_net(skb_dst(skb)->dev); | |
909 | ||
910 | skb->dev = net->loopback_dev; | |
911 | } | |
912 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); | |
913 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
914 | goto tx_error_put; | |
915 | } | |
916 | ||
917 | /* | |
918 | * Okay, now see if we can stuff it in the buffer as-is. | |
919 | */ | |
920 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr); | |
921 | ||
922 | if (skb_headroom(skb) < max_headroom | |
923 | || skb_cloned(skb) || skb_shared(skb)) { | |
924 | struct sk_buff *new_skb = | |
925 | skb_realloc_headroom(skb, max_headroom); | |
926 | if (!new_skb) { | |
927 | dst_release(&rt->dst); | |
928 | kfree_skb(skb); | |
929 | IP_VS_ERR_RL("%s(): no memory\n", __func__); | |
930 | return NF_STOLEN; | |
931 | } | |
932 | kfree_skb(skb); | |
933 | skb = new_skb; | |
934 | old_iph = ipv6_hdr(skb); | |
935 | } | |
936 | ||
937 | skb->transport_header = skb->network_header; | |
938 | ||
939 | skb_push(skb, sizeof(struct ipv6hdr)); | |
940 | skb_reset_network_header(skb); | |
941 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); | |
942 | ||
943 | /* drop old route */ | |
944 | skb_dst_drop(skb); | |
945 | skb_dst_set(skb, &rt->dst); | |
946 | ||
947 | /* | |
948 | * Push down and install the IPIP header. | |
949 | */ | |
950 | iph = ipv6_hdr(skb); | |
951 | iph->version = 6; | |
952 | iph->nexthdr = IPPROTO_IPV6; | |
953 | iph->payload_len = old_iph->payload_len; | |
954 | be16_add_cpu(&iph->payload_len, sizeof(*old_iph)); | |
955 | iph->priority = old_iph->priority; | |
956 | memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl)); | |
957 | ipv6_addr_copy(&iph->daddr, &cp->daddr.in6); | |
958 | ipv6_addr_copy(&iph->saddr, &saddr); | |
959 | iph->hop_limit = old_iph->hop_limit; | |
960 | ||
961 | /* Another hack: avoid icmp_send in ip_fragment */ | |
962 | skb->local_df = 1; | |
963 | ||
964 | ret = IP_VS_XMIT_TUNNEL(skb, cp); | |
965 | if (ret == NF_ACCEPT) | |
966 | ip6_local_out(skb); | |
967 | else if (ret == NF_DROP) | |
968 | kfree_skb(skb); | |
969 | ||
970 | LeaveFunction(10); | |
971 | ||
972 | return NF_STOLEN; | |
973 | ||
974 | tx_error_icmp: | |
975 | dst_link_failure(skb); | |
976 | tx_error: | |
977 | kfree_skb(skb); | |
978 | LeaveFunction(10); | |
979 | return NF_STOLEN; | |
980 | tx_error_put: | |
981 | dst_release(&rt->dst); | |
982 | goto tx_error; | |
983 | } | |
984 | #endif | |
985 | ||
986 | ||
987 | /* | |
988 | * Direct Routing transmitter | |
989 | * Used for ANY protocol | |
990 | */ | |
991 | int | |
992 | ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
993 | struct ip_vs_protocol *pp) | |
994 | { | |
995 | struct rtable *rt; /* Route to the other host */ | |
996 | struct iphdr *iph = ip_hdr(skb); | |
997 | int mtu; | |
998 | ||
999 | EnterFunction(10); | |
1000 | ||
1001 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, | |
1002 | RT_TOS(iph->tos), 1|2))) | |
1003 | goto tx_error_icmp; | |
1004 | if (rt->rt_flags & RTCF_LOCAL) { | |
1005 | ip_rt_put(rt); | |
1006 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); | |
1007 | } | |
1008 | ||
1009 | /* MTU checking */ | |
1010 | mtu = dst_mtu(&rt->dst); | |
1011 | if ((iph->frag_off & htons(IP_DF)) && skb->len > mtu) { | |
1012 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
1013 | ip_rt_put(rt); | |
1014 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
1015 | goto tx_error; | |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * Call ip_send_check because we are not sure it is called | |
1020 | * after ip_defrag. Is copy-on-write needed? | |
1021 | */ | |
1022 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
1023 | ip_rt_put(rt); | |
1024 | return NF_STOLEN; | |
1025 | } | |
1026 | ip_send_check(ip_hdr(skb)); | |
1027 | ||
1028 | /* drop old route */ | |
1029 | skb_dst_drop(skb); | |
1030 | skb_dst_set(skb, &rt->dst); | |
1031 | ||
1032 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1033 | skb->local_df = 1; | |
1034 | ||
1035 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0); | |
1036 | ||
1037 | LeaveFunction(10); | |
1038 | return NF_STOLEN; | |
1039 | ||
1040 | tx_error_icmp: | |
1041 | dst_link_failure(skb); | |
1042 | tx_error: | |
1043 | kfree_skb(skb); | |
1044 | LeaveFunction(10); | |
1045 | return NF_STOLEN; | |
1046 | } | |
1047 | ||
1048 | #ifdef CONFIG_IP_VS_IPV6 | |
1049 | int | |
1050 | ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1051 | struct ip_vs_protocol *pp) | |
1052 | { | |
1053 | struct rt6_info *rt; /* Route to the other host */ | |
1054 | int mtu; | |
1055 | ||
1056 | EnterFunction(10); | |
1057 | ||
1058 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, | |
1059 | 0, 1|2))) | |
1060 | goto tx_error_icmp; | |
1061 | if (__ip_vs_is_local_route6(rt)) { | |
1062 | dst_release(&rt->dst); | |
1063 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1); | |
1064 | } | |
1065 | ||
1066 | /* MTU checking */ | |
1067 | mtu = dst_mtu(&rt->dst); | |
1068 | if (skb->len > mtu) { | |
1069 | if (!skb->dev) { | |
1070 | struct net *net = dev_net(skb_dst(skb)->dev); | |
1071 | ||
1072 | skb->dev = net->loopback_dev; | |
1073 | } | |
1074 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); | |
1075 | dst_release(&rt->dst); | |
1076 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
1077 | goto tx_error; | |
1078 | } | |
1079 | ||
1080 | /* | |
1081 | * Call ip_send_check because we are not sure it is called | |
1082 | * after ip_defrag. Is copy-on-write needed? | |
1083 | */ | |
1084 | skb = skb_share_check(skb, GFP_ATOMIC); | |
1085 | if (unlikely(skb == NULL)) { | |
1086 | dst_release(&rt->dst); | |
1087 | return NF_STOLEN; | |
1088 | } | |
1089 | ||
1090 | /* drop old route */ | |
1091 | skb_dst_drop(skb); | |
1092 | skb_dst_set(skb, &rt->dst); | |
1093 | ||
1094 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1095 | skb->local_df = 1; | |
1096 | ||
1097 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0); | |
1098 | ||
1099 | LeaveFunction(10); | |
1100 | return NF_STOLEN; | |
1101 | ||
1102 | tx_error_icmp: | |
1103 | dst_link_failure(skb); | |
1104 | tx_error: | |
1105 | kfree_skb(skb); | |
1106 | LeaveFunction(10); | |
1107 | return NF_STOLEN; | |
1108 | } | |
1109 | #endif | |
1110 | ||
1111 | ||
1112 | /* | |
1113 | * ICMP packet transmitter | |
1114 | * called by the ip_vs_in_icmp | |
1115 | */ | |
1116 | int | |
1117 | ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1118 | struct ip_vs_protocol *pp, int offset) | |
1119 | { | |
1120 | struct rtable *rt; /* Route to the other host */ | |
1121 | int mtu; | |
1122 | int rc; | |
1123 | int local; | |
1124 | ||
1125 | EnterFunction(10); | |
1126 | ||
1127 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
1128 | forwarded directly here, because there is no need to | |
1129 | translate address/port back */ | |
1130 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
1131 | if (cp->packet_xmit) | |
1132 | rc = cp->packet_xmit(skb, cp, pp); | |
1133 | else | |
1134 | rc = NF_ACCEPT; | |
1135 | /* do not touch skb anymore */ | |
1136 | atomic_inc(&cp->in_pkts); | |
1137 | goto out; | |
1138 | } | |
1139 | ||
1140 | /* | |
1141 | * mangle and send the packet here (only for VS/NAT) | |
1142 | */ | |
1143 | ||
1144 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, | |
1145 | RT_TOS(ip_hdr(skb)->tos), 1|2|4))) | |
1146 | goto tx_error_icmp; | |
1147 | local = rt->rt_flags & RTCF_LOCAL; | |
1148 | ||
1149 | /* | |
1150 | * Avoid duplicate tuple in reply direction for NAT traffic | |
1151 | * to local address when connection is sync-ed | |
1152 | */ | |
1153 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
1154 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
1155 | enum ip_conntrack_info ctinfo; | |
1156 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
1157 | ||
1158 | if (ct && !nf_ct_is_untracked(ct)) { | |
1159 | IP_VS_DBG(10, "%s(): " | |
1160 | "stopping DNAT to local address %pI4\n", | |
1161 | __func__, &cp->daddr.ip); | |
1162 | goto tx_error_put; | |
1163 | } | |
1164 | } | |
1165 | #endif | |
1166 | ||
1167 | /* From world but DNAT to loopback address? */ | |
1168 | if (local && ipv4_is_loopback(rt->rt_dst) && skb_rtable(skb)->fl.iif) { | |
1169 | IP_VS_DBG(1, "%s(): " | |
1170 | "stopping DNAT to loopback %pI4\n", | |
1171 | __func__, &cp->daddr.ip); | |
1172 | goto tx_error_put; | |
1173 | } | |
1174 | ||
1175 | /* MTU checking */ | |
1176 | mtu = dst_mtu(&rt->dst); | |
1177 | if ((skb->len > mtu) && (ip_hdr(skb)->frag_off & htons(IP_DF))) { | |
1178 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); | |
1179 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
1180 | goto tx_error_put; | |
1181 | } | |
1182 | ||
1183 | /* copy-on-write the packet before mangling it */ | |
1184 | if (!skb_make_writable(skb, offset)) | |
1185 | goto tx_error_put; | |
1186 | ||
1187 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) | |
1188 | goto tx_error_put; | |
1189 | ||
1190 | ip_vs_nat_icmp(skb, pp, cp, 0); | |
1191 | ||
1192 | if (!local) { | |
1193 | /* drop the old route when skb is not shared */ | |
1194 | skb_dst_drop(skb); | |
1195 | skb_dst_set(skb, &rt->dst); | |
1196 | } else { | |
1197 | ip_rt_put(rt); | |
1198 | /* | |
1199 | * Some IPv4 replies get local address from routes, | |
1200 | * not from iph, so while we DNAT after routing | |
1201 | * we need this second input/output route. | |
1202 | */ | |
1203 | if (!__ip_vs_reroute_locally(skb)) | |
1204 | goto tx_error; | |
1205 | } | |
1206 | ||
1207 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1208 | skb->local_df = 1; | |
1209 | ||
1210 | IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local); | |
1211 | ||
1212 | rc = NF_STOLEN; | |
1213 | goto out; | |
1214 | ||
1215 | tx_error_icmp: | |
1216 | dst_link_failure(skb); | |
1217 | tx_error: | |
1218 | dev_kfree_skb(skb); | |
1219 | rc = NF_STOLEN; | |
1220 | out: | |
1221 | LeaveFunction(10); | |
1222 | return rc; | |
1223 | tx_error_put: | |
1224 | ip_rt_put(rt); | |
1225 | goto tx_error; | |
1226 | } | |
1227 | ||
1228 | #ifdef CONFIG_IP_VS_IPV6 | |
1229 | int | |
1230 | ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1231 | struct ip_vs_protocol *pp, int offset) | |
1232 | { | |
1233 | struct rt6_info *rt; /* Route to the other host */ | |
1234 | int mtu; | |
1235 | int rc; | |
1236 | int local; | |
1237 | ||
1238 | EnterFunction(10); | |
1239 | ||
1240 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
1241 | forwarded directly here, because there is no need to | |
1242 | translate address/port back */ | |
1243 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
1244 | if (cp->packet_xmit) | |
1245 | rc = cp->packet_xmit(skb, cp, pp); | |
1246 | else | |
1247 | rc = NF_ACCEPT; | |
1248 | /* do not touch skb anymore */ | |
1249 | atomic_inc(&cp->in_pkts); | |
1250 | goto out; | |
1251 | } | |
1252 | ||
1253 | /* | |
1254 | * mangle and send the packet here (only for VS/NAT) | |
1255 | */ | |
1256 | ||
1257 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, | |
1258 | 0, 1|2|4))) | |
1259 | goto tx_error_icmp; | |
1260 | ||
1261 | local = __ip_vs_is_local_route6(rt); | |
1262 | /* | |
1263 | * Avoid duplicate tuple in reply direction for NAT traffic | |
1264 | * to local address when connection is sync-ed | |
1265 | */ | |
1266 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
1267 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
1268 | enum ip_conntrack_info ctinfo; | |
1269 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
1270 | ||
1271 | if (ct && !nf_ct_is_untracked(ct)) { | |
1272 | IP_VS_DBG(10, "%s(): " | |
1273 | "stopping DNAT to local address %pI6\n", | |
1274 | __func__, &cp->daddr.in6); | |
1275 | goto tx_error_put; | |
1276 | } | |
1277 | } | |
1278 | #endif | |
1279 | ||
1280 | /* From world but DNAT to loopback address? */ | |
1281 | if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) && | |
1282 | ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) { | |
1283 | IP_VS_DBG(1, "%s(): " | |
1284 | "stopping DNAT to loopback %pI6\n", | |
1285 | __func__, &cp->daddr.in6); | |
1286 | goto tx_error_put; | |
1287 | } | |
1288 | ||
1289 | /* MTU checking */ | |
1290 | mtu = dst_mtu(&rt->dst); | |
1291 | if (skb->len > mtu) { | |
1292 | if (!skb->dev) { | |
1293 | struct net *net = dev_net(skb_dst(skb)->dev); | |
1294 | ||
1295 | skb->dev = net->loopback_dev; | |
1296 | } | |
1297 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); | |
1298 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
1299 | goto tx_error_put; | |
1300 | } | |
1301 | ||
1302 | /* copy-on-write the packet before mangling it */ | |
1303 | if (!skb_make_writable(skb, offset)) | |
1304 | goto tx_error_put; | |
1305 | ||
1306 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) | |
1307 | goto tx_error_put; | |
1308 | ||
1309 | ip_vs_nat_icmp_v6(skb, pp, cp, 0); | |
1310 | ||
1311 | if (!local || !skb->dev) { | |
1312 | /* drop the old route when skb is not shared */ | |
1313 | skb_dst_drop(skb); | |
1314 | skb_dst_set(skb, &rt->dst); | |
1315 | } else { | |
1316 | /* destined to loopback, do we need to change route? */ | |
1317 | dst_release(&rt->dst); | |
1318 | } | |
1319 | ||
1320 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1321 | skb->local_df = 1; | |
1322 | ||
1323 | IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local); | |
1324 | ||
1325 | rc = NF_STOLEN; | |
1326 | goto out; | |
1327 | ||
1328 | tx_error_icmp: | |
1329 | dst_link_failure(skb); | |
1330 | tx_error: | |
1331 | dev_kfree_skb(skb); | |
1332 | rc = NF_STOLEN; | |
1333 | out: | |
1334 | LeaveFunction(10); | |
1335 | return rc; | |
1336 | tx_error_put: | |
1337 | dst_release(&rt->dst); | |
1338 | goto tx_error; | |
1339 | } | |
1340 | #endif |