]>
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 | .fl4_dst = dest->addr.ip, | |
100 | .fl4_tos = rtos, | |
101 | }; | |
102 | ||
103 | if (ip_route_output_key(net, &rt, &fl)) { | |
104 | spin_unlock(&dest->dst_lock); | |
105 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", | |
106 | &dest->addr.ip); | |
107 | return NULL; | |
108 | } | |
109 | __ip_vs_dst_set(dest, rtos, dst_clone(&rt->dst), 0); | |
110 | IP_VS_DBG(10, "new dst %pI4, refcnt=%d, rtos=%X\n", | |
111 | &dest->addr.ip, | |
112 | atomic_read(&rt->dst.__refcnt), rtos); | |
113 | } | |
114 | spin_unlock(&dest->dst_lock); | |
115 | } else { | |
116 | struct flowi fl = { | |
117 | .fl4_dst = daddr, | |
118 | .fl4_tos = rtos, | |
119 | }; | |
120 | ||
121 | if (ip_route_output_key(net, &rt, &fl)) { | |
122 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", | |
123 | &daddr); | |
124 | return NULL; | |
125 | } | |
126 | } | |
127 | ||
128 | local = rt->rt_flags & RTCF_LOCAL; | |
129 | if (!((local ? 1 : 2) & rt_mode)) { | |
130 | IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI4\n", | |
131 | (rt->rt_flags & RTCF_LOCAL) ? | |
132 | "local":"non-local", &rt->rt_dst); | |
133 | ip_rt_put(rt); | |
134 | return NULL; | |
135 | } | |
136 | if (local && !(rt_mode & 4) && !((ort = skb_rtable(skb)) && | |
137 | ort->rt_flags & RTCF_LOCAL)) { | |
138 | IP_VS_DBG_RL("Redirect from non-local address %pI4 to local " | |
139 | "requires NAT method, dest: %pI4\n", | |
140 | &ip_hdr(skb)->daddr, &rt->rt_dst); | |
141 | ip_rt_put(rt); | |
142 | return NULL; | |
143 | } | |
144 | if (unlikely(!local && ipv4_is_loopback(ip_hdr(skb)->saddr))) { | |
145 | IP_VS_DBG_RL("Stopping traffic from loopback address %pI4 " | |
146 | "to non-local address, dest: %pI4\n", | |
147 | &ip_hdr(skb)->saddr, &rt->rt_dst); | |
148 | ip_rt_put(rt); | |
149 | return NULL; | |
150 | } | |
151 | ||
152 | return rt; | |
153 | } | |
154 | ||
155 | /* Reroute packet to local IPv4 stack after DNAT */ | |
156 | static int | |
157 | __ip_vs_reroute_locally(struct sk_buff *skb) | |
158 | { | |
159 | struct rtable *rt = skb_rtable(skb); | |
160 | struct net_device *dev = rt->dst.dev; | |
161 | struct net *net = dev_net(dev); | |
162 | struct iphdr *iph = ip_hdr(skb); | |
163 | ||
164 | if (rt_is_input_route(rt)) { | |
165 | unsigned long orefdst = skb->_skb_refdst; | |
166 | ||
167 | if (ip_route_input(skb, iph->daddr, iph->saddr, | |
168 | iph->tos, skb->dev)) | |
169 | return 0; | |
170 | refdst_drop(orefdst); | |
171 | } else { | |
172 | struct flowi fl = { | |
173 | .fl4_dst = iph->daddr, | |
174 | .fl4_src = iph->saddr, | |
175 | .fl4_tos = RT_TOS(iph->tos), | |
176 | .mark = skb->mark, | |
177 | }; | |
178 | struct rtable *rt; | |
179 | ||
180 | if (ip_route_output_key(net, &rt, &fl)) | |
181 | return 0; | |
182 | if (!(rt->rt_flags & RTCF_LOCAL)) { | |
183 | ip_rt_put(rt); | |
184 | return 0; | |
185 | } | |
186 | /* Drop old route. */ | |
187 | skb_dst_drop(skb); | |
188 | skb_dst_set(skb, &rt->dst); | |
189 | } | |
190 | return 1; | |
191 | } | |
192 | ||
193 | #ifdef CONFIG_IP_VS_IPV6 | |
194 | ||
195 | static inline int __ip_vs_is_local_route6(struct rt6_info *rt) | |
196 | { | |
197 | return rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK; | |
198 | } | |
199 | ||
200 | static struct dst_entry * | |
201 | __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr, | |
202 | struct in6_addr *ret_saddr, int do_xfrm) | |
203 | { | |
204 | struct dst_entry *dst; | |
205 | struct flowi fl = { | |
206 | .fl6_dst = *daddr, | |
207 | }; | |
208 | ||
209 | dst = ip6_route_output(net, NULL, &fl); | |
210 | if (dst->error) | |
211 | goto out_err; | |
212 | if (!ret_saddr) | |
213 | return dst; | |
214 | if (ipv6_addr_any(&fl.fl6_src) && | |
215 | ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev, | |
216 | &fl.fl6_dst, 0, &fl.fl6_src) < 0) | |
217 | goto out_err; | |
218 | if (do_xfrm && xfrm_lookup(net, &dst, &fl, NULL, 0) < 0) | |
219 | goto out_err; | |
220 | ipv6_addr_copy(ret_saddr, &fl.fl6_src); | |
221 | return dst; | |
222 | ||
223 | out_err: | |
224 | dst_release(dst); | |
225 | IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr); | |
226 | return NULL; | |
227 | } | |
228 | ||
229 | /* | |
230 | * Get route to destination or remote server | |
231 | * rt_mode: flags, &1=Allow local dest, &2=Allow non-local dest, | |
232 | * &4=Allow redirect from remote daddr to local | |
233 | */ | |
234 | static struct rt6_info * | |
235 | __ip_vs_get_out_rt_v6(struct sk_buff *skb, struct ip_vs_dest *dest, | |
236 | struct in6_addr *daddr, struct in6_addr *ret_saddr, | |
237 | int do_xfrm, int rt_mode) | |
238 | { | |
239 | struct net *net = dev_net(skb_dst(skb)->dev); | |
240 | struct rt6_info *rt; /* Route to the other host */ | |
241 | struct rt6_info *ort; /* Original route */ | |
242 | struct dst_entry *dst; | |
243 | int local; | |
244 | ||
245 | if (dest) { | |
246 | spin_lock(&dest->dst_lock); | |
247 | rt = (struct rt6_info *)__ip_vs_dst_check(dest, 0); | |
248 | if (!rt) { | |
249 | u32 cookie; | |
250 | ||
251 | dst = __ip_vs_route_output_v6(net, &dest->addr.in6, | |
252 | &dest->dst_saddr, | |
253 | do_xfrm); | |
254 | if (!dst) { | |
255 | spin_unlock(&dest->dst_lock); | |
256 | return NULL; | |
257 | } | |
258 | rt = (struct rt6_info *) dst; | |
259 | cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0; | |
260 | __ip_vs_dst_set(dest, 0, dst_clone(&rt->dst), cookie); | |
261 | IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n", | |
262 | &dest->addr.in6, &dest->dst_saddr, | |
263 | atomic_read(&rt->dst.__refcnt)); | |
264 | } | |
265 | if (ret_saddr) | |
266 | ipv6_addr_copy(ret_saddr, &dest->dst_saddr); | |
267 | spin_unlock(&dest->dst_lock); | |
268 | } else { | |
269 | dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm); | |
270 | if (!dst) | |
271 | return NULL; | |
272 | rt = (struct rt6_info *) dst; | |
273 | } | |
274 | ||
275 | local = __ip_vs_is_local_route6(rt); | |
276 | if (!((local ? 1 : 2) & rt_mode)) { | |
277 | IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI6\n", | |
278 | local ? "local":"non-local", daddr); | |
279 | dst_release(&rt->dst); | |
280 | return NULL; | |
281 | } | |
282 | if (local && !(rt_mode & 4) && | |
283 | !((ort = (struct rt6_info *) skb_dst(skb)) && | |
284 | __ip_vs_is_local_route6(ort))) { | |
285 | IP_VS_DBG_RL("Redirect from non-local address %pI6 to local " | |
286 | "requires NAT method, dest: %pI6\n", | |
287 | &ipv6_hdr(skb)->daddr, daddr); | |
288 | dst_release(&rt->dst); | |
289 | return NULL; | |
290 | } | |
291 | if (unlikely(!local && (!skb->dev || skb->dev->flags & IFF_LOOPBACK) && | |
292 | ipv6_addr_type(&ipv6_hdr(skb)->saddr) & | |
293 | IPV6_ADDR_LOOPBACK)) { | |
294 | IP_VS_DBG_RL("Stopping traffic from loopback address %pI6 " | |
295 | "to non-local address, dest: %pI6\n", | |
296 | &ipv6_hdr(skb)->saddr, daddr); | |
297 | dst_release(&rt->dst); | |
298 | return NULL; | |
299 | } | |
300 | ||
301 | return rt; | |
302 | } | |
303 | #endif | |
304 | ||
305 | ||
306 | /* | |
307 | * Release dest->dst_cache before a dest is removed | |
308 | */ | |
309 | void | |
310 | ip_vs_dst_reset(struct ip_vs_dest *dest) | |
311 | { | |
312 | struct dst_entry *old_dst; | |
313 | ||
314 | old_dst = dest->dst_cache; | |
315 | dest->dst_cache = NULL; | |
316 | dst_release(old_dst); | |
317 | } | |
318 | ||
319 | #define IP_VS_XMIT_TUNNEL(skb, cp) \ | |
320 | ({ \ | |
321 | int __ret = NF_ACCEPT; \ | |
322 | \ | |
323 | (skb)->ipvs_property = 1; \ | |
324 | if (unlikely((cp)->flags & IP_VS_CONN_F_NFCT)) \ | |
325 | __ret = ip_vs_confirm_conntrack(skb, cp); \ | |
326 | if (__ret == NF_ACCEPT) { \ | |
327 | nf_reset(skb); \ | |
328 | skb_forward_csum(skb); \ | |
329 | } \ | |
330 | __ret; \ | |
331 | }) | |
332 | ||
333 | #define IP_VS_XMIT_NAT(pf, skb, cp, local) \ | |
334 | do { \ | |
335 | (skb)->ipvs_property = 1; \ | |
336 | if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \ | |
337 | ip_vs_notrack(skb); \ | |
338 | else \ | |
339 | ip_vs_update_conntrack(skb, cp, 1); \ | |
340 | if (local) \ | |
341 | return NF_ACCEPT; \ | |
342 | skb_forward_csum(skb); \ | |
343 | NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \ | |
344 | skb_dst(skb)->dev, dst_output); \ | |
345 | } while (0) | |
346 | ||
347 | #define IP_VS_XMIT(pf, skb, cp, local) \ | |
348 | do { \ | |
349 | (skb)->ipvs_property = 1; \ | |
350 | if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \ | |
351 | ip_vs_notrack(skb); \ | |
352 | if (local) \ | |
353 | return NF_ACCEPT; \ | |
354 | skb_forward_csum(skb); \ | |
355 | NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \ | |
356 | skb_dst(skb)->dev, dst_output); \ | |
357 | } while (0) | |
358 | ||
359 | ||
360 | /* | |
361 | * NULL transmitter (do nothing except return NF_ACCEPT) | |
362 | */ | |
363 | int | |
364 | ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
365 | struct ip_vs_protocol *pp) | |
366 | { | |
367 | /* we do not touch skb and do not need pskb ptr */ | |
368 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); | |
369 | } | |
370 | ||
371 | ||
372 | /* | |
373 | * Bypass transmitter | |
374 | * Let packets bypass the destination when the destination is not | |
375 | * available, it may be only used in transparent cache cluster. | |
376 | */ | |
377 | int | |
378 | ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
379 | struct ip_vs_protocol *pp) | |
380 | { | |
381 | struct rtable *rt; /* Route to the other host */ | |
382 | struct iphdr *iph = ip_hdr(skb); | |
383 | int mtu; | |
384 | ||
385 | EnterFunction(10); | |
386 | ||
387 | if (!(rt = __ip_vs_get_out_rt(skb, NULL, iph->daddr, | |
388 | RT_TOS(iph->tos), 2))) | |
389 | goto tx_error_icmp; | |
390 | ||
391 | /* MTU checking */ | |
392 | mtu = dst_mtu(&rt->dst); | |
393 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) { | |
394 | ip_rt_put(rt); | |
395 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
396 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
397 | goto tx_error; | |
398 | } | |
399 | ||
400 | /* | |
401 | * Call ip_send_check because we are not sure it is called | |
402 | * after ip_defrag. Is copy-on-write needed? | |
403 | */ | |
404 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
405 | ip_rt_put(rt); | |
406 | return NF_STOLEN; | |
407 | } | |
408 | ip_send_check(ip_hdr(skb)); | |
409 | ||
410 | /* drop old route */ | |
411 | skb_dst_drop(skb); | |
412 | skb_dst_set(skb, &rt->dst); | |
413 | ||
414 | /* Another hack: avoid icmp_send in ip_fragment */ | |
415 | skb->local_df = 1; | |
416 | ||
417 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0); | |
418 | ||
419 | LeaveFunction(10); | |
420 | return NF_STOLEN; | |
421 | ||
422 | tx_error_icmp: | |
423 | dst_link_failure(skb); | |
424 | tx_error: | |
425 | kfree_skb(skb); | |
426 | LeaveFunction(10); | |
427 | return NF_STOLEN; | |
428 | } | |
429 | ||
430 | #ifdef CONFIG_IP_VS_IPV6 | |
431 | int | |
432 | ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
433 | struct ip_vs_protocol *pp) | |
434 | { | |
435 | struct rt6_info *rt; /* Route to the other host */ | |
436 | struct ipv6hdr *iph = ipv6_hdr(skb); | |
437 | int mtu; | |
438 | ||
439 | EnterFunction(10); | |
440 | ||
441 | if (!(rt = __ip_vs_get_out_rt_v6(skb, NULL, &iph->daddr, NULL, 0, 2))) | |
442 | goto tx_error_icmp; | |
443 | ||
444 | /* MTU checking */ | |
445 | mtu = dst_mtu(&rt->dst); | |
446 | if (skb->len > mtu) { | |
447 | if (!skb->dev) { | |
448 | struct net *net = dev_net(skb_dst(skb)->dev); | |
449 | ||
450 | skb->dev = net->loopback_dev; | |
451 | } | |
452 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); | |
453 | dst_release(&rt->dst); | |
454 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
455 | goto tx_error; | |
456 | } | |
457 | ||
458 | /* | |
459 | * Call ip_send_check because we are not sure it is called | |
460 | * after ip_defrag. Is copy-on-write needed? | |
461 | */ | |
462 | skb = skb_share_check(skb, GFP_ATOMIC); | |
463 | if (unlikely(skb == NULL)) { | |
464 | dst_release(&rt->dst); | |
465 | return NF_STOLEN; | |
466 | } | |
467 | ||
468 | /* drop old route */ | |
469 | skb_dst_drop(skb); | |
470 | skb_dst_set(skb, &rt->dst); | |
471 | ||
472 | /* Another hack: avoid icmp_send in ip_fragment */ | |
473 | skb->local_df = 1; | |
474 | ||
475 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0); | |
476 | ||
477 | LeaveFunction(10); | |
478 | return NF_STOLEN; | |
479 | ||
480 | tx_error_icmp: | |
481 | dst_link_failure(skb); | |
482 | tx_error: | |
483 | kfree_skb(skb); | |
484 | LeaveFunction(10); | |
485 | return NF_STOLEN; | |
486 | } | |
487 | #endif | |
488 | ||
489 | /* | |
490 | * NAT transmitter (only for outside-to-inside nat forwarding) | |
491 | * Not used for related ICMP | |
492 | */ | |
493 | int | |
494 | ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
495 | struct ip_vs_protocol *pp) | |
496 | { | |
497 | struct rtable *rt; /* Route to the other host */ | |
498 | int mtu; | |
499 | struct iphdr *iph = ip_hdr(skb); | |
500 | int local; | |
501 | ||
502 | EnterFunction(10); | |
503 | ||
504 | /* check if it is a connection of no-client-port */ | |
505 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
506 | __be16 _pt, *p; | |
507 | p = skb_header_pointer(skb, iph->ihl*4, sizeof(_pt), &_pt); | |
508 | if (p == NULL) | |
509 | goto tx_error; | |
510 | ip_vs_conn_fill_cport(cp, *p); | |
511 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
512 | } | |
513 | ||
514 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, | |
515 | RT_TOS(iph->tos), 1|2|4))) | |
516 | goto tx_error_icmp; | |
517 | local = rt->rt_flags & RTCF_LOCAL; | |
518 | /* | |
519 | * Avoid duplicate tuple in reply direction for NAT traffic | |
520 | * to local address when connection is sync-ed | |
521 | */ | |
522 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
523 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
524 | enum ip_conntrack_info ctinfo; | |
525 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
526 | ||
527 | if (ct && !nf_ct_is_untracked(ct)) { | |
528 | IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, 0, | |
529 | "ip_vs_nat_xmit(): " | |
530 | "stopping DNAT to local address"); | |
531 | goto tx_error_put; | |
532 | } | |
533 | } | |
534 | #endif | |
535 | ||
536 | /* From world but DNAT to loopback address? */ | |
537 | if (local && ipv4_is_loopback(rt->rt_dst) && | |
538 | rt_is_input_route(skb_rtable(skb))) { | |
539 | IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, 0, "ip_vs_nat_xmit(): " | |
540 | "stopping DNAT to loopback address"); | |
541 | goto tx_error_put; | |
542 | } | |
543 | ||
544 | /* MTU checking */ | |
545 | mtu = dst_mtu(&rt->dst); | |
546 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) { | |
547 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
548 | IP_VS_DBG_RL_PKT(0, AF_INET, pp, skb, 0, | |
549 | "ip_vs_nat_xmit(): frag needed for"); | |
550 | goto tx_error_put; | |
551 | } | |
552 | ||
553 | /* copy-on-write the packet before mangling it */ | |
554 | if (!skb_make_writable(skb, sizeof(struct iphdr))) | |
555 | goto tx_error_put; | |
556 | ||
557 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) | |
558 | goto tx_error_put; | |
559 | ||
560 | /* mangle the packet */ | |
561 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) | |
562 | goto tx_error_put; | |
563 | ip_hdr(skb)->daddr = cp->daddr.ip; | |
564 | ip_send_check(ip_hdr(skb)); | |
565 | ||
566 | if (!local) { | |
567 | /* drop old route */ | |
568 | skb_dst_drop(skb); | |
569 | skb_dst_set(skb, &rt->dst); | |
570 | } else { | |
571 | ip_rt_put(rt); | |
572 | /* | |
573 | * Some IPv4 replies get local address from routes, | |
574 | * not from iph, so while we DNAT after routing | |
575 | * we need this second input/output route. | |
576 | */ | |
577 | if (!__ip_vs_reroute_locally(skb)) | |
578 | goto tx_error; | |
579 | } | |
580 | ||
581 | IP_VS_DBG_PKT(10, AF_INET, pp, skb, 0, "After DNAT"); | |
582 | ||
583 | /* FIXME: when application helper enlarges the packet and the length | |
584 | is larger than the MTU of outgoing device, there will be still | |
585 | MTU problem. */ | |
586 | ||
587 | /* Another hack: avoid icmp_send in ip_fragment */ | |
588 | skb->local_df = 1; | |
589 | ||
590 | IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local); | |
591 | ||
592 | LeaveFunction(10); | |
593 | return NF_STOLEN; | |
594 | ||
595 | tx_error_icmp: | |
596 | dst_link_failure(skb); | |
597 | tx_error: | |
598 | kfree_skb(skb); | |
599 | LeaveFunction(10); | |
600 | return NF_STOLEN; | |
601 | tx_error_put: | |
602 | ip_rt_put(rt); | |
603 | goto tx_error; | |
604 | } | |
605 | ||
606 | #ifdef CONFIG_IP_VS_IPV6 | |
607 | int | |
608 | ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
609 | struct ip_vs_protocol *pp) | |
610 | { | |
611 | struct rt6_info *rt; /* Route to the other host */ | |
612 | int mtu; | |
613 | int local; | |
614 | ||
615 | EnterFunction(10); | |
616 | ||
617 | /* check if it is a connection of no-client-port */ | |
618 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
619 | __be16 _pt, *p; | |
620 | p = skb_header_pointer(skb, sizeof(struct ipv6hdr), | |
621 | sizeof(_pt), &_pt); | |
622 | if (p == NULL) | |
623 | goto tx_error; | |
624 | ip_vs_conn_fill_cport(cp, *p); | |
625 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
626 | } | |
627 | ||
628 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, | |
629 | 0, 1|2|4))) | |
630 | goto tx_error_icmp; | |
631 | local = __ip_vs_is_local_route6(rt); | |
632 | /* | |
633 | * Avoid duplicate tuple in reply direction for NAT traffic | |
634 | * to local address when connection is sync-ed | |
635 | */ | |
636 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
637 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
638 | enum ip_conntrack_info ctinfo; | |
639 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
640 | ||
641 | if (ct && !nf_ct_is_untracked(ct)) { | |
642 | IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, 0, | |
643 | "ip_vs_nat_xmit_v6(): " | |
644 | "stopping DNAT to local address"); | |
645 | goto tx_error_put; | |
646 | } | |
647 | } | |
648 | #endif | |
649 | ||
650 | /* From world but DNAT to loopback address? */ | |
651 | if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) && | |
652 | ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) { | |
653 | IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, 0, | |
654 | "ip_vs_nat_xmit_v6(): " | |
655 | "stopping DNAT to loopback address"); | |
656 | goto tx_error_put; | |
657 | } | |
658 | ||
659 | /* MTU checking */ | |
660 | mtu = dst_mtu(&rt->dst); | |
661 | if (skb->len > mtu) { | |
662 | if (!skb->dev) { | |
663 | struct net *net = dev_net(skb_dst(skb)->dev); | |
664 | ||
665 | skb->dev = net->loopback_dev; | |
666 | } | |
667 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); | |
668 | IP_VS_DBG_RL_PKT(0, AF_INET6, pp, skb, 0, | |
669 | "ip_vs_nat_xmit_v6(): frag needed for"); | |
670 | goto tx_error_put; | |
671 | } | |
672 | ||
673 | /* copy-on-write the packet before mangling it */ | |
674 | if (!skb_make_writable(skb, sizeof(struct ipv6hdr))) | |
675 | goto tx_error_put; | |
676 | ||
677 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) | |
678 | goto tx_error_put; | |
679 | ||
680 | /* mangle the packet */ | |
681 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) | |
682 | goto tx_error; | |
683 | ipv6_addr_copy(&ipv6_hdr(skb)->daddr, &cp->daddr.in6); | |
684 | ||
685 | if (!local || !skb->dev) { | |
686 | /* drop the old route when skb is not shared */ | |
687 | skb_dst_drop(skb); | |
688 | skb_dst_set(skb, &rt->dst); | |
689 | } else { | |
690 | /* destined to loopback, do we need to change route? */ | |
691 | dst_release(&rt->dst); | |
692 | } | |
693 | ||
694 | IP_VS_DBG_PKT(10, AF_INET6, pp, skb, 0, "After DNAT"); | |
695 | ||
696 | /* FIXME: when application helper enlarges the packet and the length | |
697 | is larger than the MTU of outgoing device, there will be still | |
698 | MTU problem. */ | |
699 | ||
700 | /* Another hack: avoid icmp_send in ip_fragment */ | |
701 | skb->local_df = 1; | |
702 | ||
703 | IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local); | |
704 | ||
705 | LeaveFunction(10); | |
706 | return NF_STOLEN; | |
707 | ||
708 | tx_error_icmp: | |
709 | dst_link_failure(skb); | |
710 | tx_error: | |
711 | LeaveFunction(10); | |
712 | kfree_skb(skb); | |
713 | return NF_STOLEN; | |
714 | tx_error_put: | |
715 | dst_release(&rt->dst); | |
716 | goto tx_error; | |
717 | } | |
718 | #endif | |
719 | ||
720 | ||
721 | /* | |
722 | * IP Tunneling transmitter | |
723 | * | |
724 | * This function encapsulates the packet in a new IP packet, its | |
725 | * destination will be set to cp->daddr. Most code of this function | |
726 | * is taken from ipip.c. | |
727 | * | |
728 | * It is used in VS/TUN cluster. The load balancer selects a real | |
729 | * server from a cluster based on a scheduling algorithm, | |
730 | * encapsulates the request packet and forwards it to the selected | |
731 | * server. For example, all real servers are configured with | |
732 | * "ifconfig tunl0 <Virtual IP Address> up". When the server receives | |
733 | * the encapsulated packet, it will decapsulate the packet, processe | |
734 | * the request and return the response packets directly to the client | |
735 | * without passing the load balancer. This can greatly increase the | |
736 | * scalability of virtual server. | |
737 | * | |
738 | * Used for ANY protocol | |
739 | */ | |
740 | int | |
741 | ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
742 | struct ip_vs_protocol *pp) | |
743 | { | |
744 | struct rtable *rt; /* Route to the other host */ | |
745 | struct net_device *tdev; /* Device to other host */ | |
746 | struct iphdr *old_iph = ip_hdr(skb); | |
747 | u8 tos = old_iph->tos; | |
748 | __be16 df = old_iph->frag_off; | |
749 | struct iphdr *iph; /* Our new IP header */ | |
750 | unsigned int max_headroom; /* The extra header space needed */ | |
751 | int mtu; | |
752 | int ret; | |
753 | ||
754 | EnterFunction(10); | |
755 | ||
756 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, | |
757 | RT_TOS(tos), 1|2))) | |
758 | goto tx_error_icmp; | |
759 | if (rt->rt_flags & RTCF_LOCAL) { | |
760 | ip_rt_put(rt); | |
761 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); | |
762 | } | |
763 | ||
764 | tdev = rt->dst.dev; | |
765 | ||
766 | mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr); | |
767 | if (mtu < 68) { | |
768 | IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__); | |
769 | goto tx_error_put; | |
770 | } | |
771 | if (skb_dst(skb)) | |
772 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
773 | ||
774 | df |= (old_iph->frag_off & htons(IP_DF)); | |
775 | ||
776 | if ((old_iph->frag_off & htons(IP_DF)) | |
777 | && mtu < ntohs(old_iph->tot_len)) { | |
778 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
779 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
780 | goto tx_error_put; | |
781 | } | |
782 | ||
783 | /* | |
784 | * Okay, now see if we can stuff it in the buffer as-is. | |
785 | */ | |
786 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr); | |
787 | ||
788 | if (skb_headroom(skb) < max_headroom | |
789 | || skb_cloned(skb) || skb_shared(skb)) { | |
790 | struct sk_buff *new_skb = | |
791 | skb_realloc_headroom(skb, max_headroom); | |
792 | if (!new_skb) { | |
793 | ip_rt_put(rt); | |
794 | kfree_skb(skb); | |
795 | IP_VS_ERR_RL("%s(): no memory\n", __func__); | |
796 | return NF_STOLEN; | |
797 | } | |
798 | kfree_skb(skb); | |
799 | skb = new_skb; | |
800 | old_iph = ip_hdr(skb); | |
801 | } | |
802 | ||
803 | skb->transport_header = skb->network_header; | |
804 | ||
805 | /* fix old IP header checksum */ | |
806 | ip_send_check(old_iph); | |
807 | ||
808 | skb_push(skb, sizeof(struct iphdr)); | |
809 | skb_reset_network_header(skb); | |
810 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); | |
811 | ||
812 | /* drop old route */ | |
813 | skb_dst_drop(skb); | |
814 | skb_dst_set(skb, &rt->dst); | |
815 | ||
816 | /* | |
817 | * Push down and install the IPIP header. | |
818 | */ | |
819 | iph = ip_hdr(skb); | |
820 | iph->version = 4; | |
821 | iph->ihl = sizeof(struct iphdr)>>2; | |
822 | iph->frag_off = df; | |
823 | iph->protocol = IPPROTO_IPIP; | |
824 | iph->tos = tos; | |
825 | iph->daddr = rt->rt_dst; | |
826 | iph->saddr = rt->rt_src; | |
827 | iph->ttl = old_iph->ttl; | |
828 | ip_select_ident(iph, &rt->dst, NULL); | |
829 | ||
830 | /* Another hack: avoid icmp_send in ip_fragment */ | |
831 | skb->local_df = 1; | |
832 | ||
833 | ret = IP_VS_XMIT_TUNNEL(skb, cp); | |
834 | if (ret == NF_ACCEPT) | |
835 | ip_local_out(skb); | |
836 | else if (ret == NF_DROP) | |
837 | kfree_skb(skb); | |
838 | ||
839 | LeaveFunction(10); | |
840 | ||
841 | return NF_STOLEN; | |
842 | ||
843 | tx_error_icmp: | |
844 | dst_link_failure(skb); | |
845 | tx_error: | |
846 | kfree_skb(skb); | |
847 | LeaveFunction(10); | |
848 | return NF_STOLEN; | |
849 | tx_error_put: | |
850 | ip_rt_put(rt); | |
851 | goto tx_error; | |
852 | } | |
853 | ||
854 | #ifdef CONFIG_IP_VS_IPV6 | |
855 | int | |
856 | ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
857 | struct ip_vs_protocol *pp) | |
858 | { | |
859 | struct rt6_info *rt; /* Route to the other host */ | |
860 | struct in6_addr saddr; /* Source for tunnel */ | |
861 | struct net_device *tdev; /* Device to other host */ | |
862 | struct ipv6hdr *old_iph = ipv6_hdr(skb); | |
863 | struct ipv6hdr *iph; /* Our new IP header */ | |
864 | unsigned int max_headroom; /* The extra header space needed */ | |
865 | int mtu; | |
866 | int ret; | |
867 | ||
868 | EnterFunction(10); | |
869 | ||
870 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, | |
871 | &saddr, 1, 1|2))) | |
872 | goto tx_error_icmp; | |
873 | if (__ip_vs_is_local_route6(rt)) { | |
874 | dst_release(&rt->dst); | |
875 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1); | |
876 | } | |
877 | ||
878 | tdev = rt->dst.dev; | |
879 | ||
880 | mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr); | |
881 | if (mtu < IPV6_MIN_MTU) { | |
882 | IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__, | |
883 | IPV6_MIN_MTU); | |
884 | goto tx_error_put; | |
885 | } | |
886 | if (skb_dst(skb)) | |
887 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
888 | ||
889 | if (mtu < ntohs(old_iph->payload_len) + sizeof(struct ipv6hdr)) { | |
890 | if (!skb->dev) { | |
891 | struct net *net = dev_net(skb_dst(skb)->dev); | |
892 | ||
893 | skb->dev = net->loopback_dev; | |
894 | } | |
895 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); | |
896 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
897 | goto tx_error_put; | |
898 | } | |
899 | ||
900 | /* | |
901 | * Okay, now see if we can stuff it in the buffer as-is. | |
902 | */ | |
903 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr); | |
904 | ||
905 | if (skb_headroom(skb) < max_headroom | |
906 | || skb_cloned(skb) || skb_shared(skb)) { | |
907 | struct sk_buff *new_skb = | |
908 | skb_realloc_headroom(skb, max_headroom); | |
909 | if (!new_skb) { | |
910 | dst_release(&rt->dst); | |
911 | kfree_skb(skb); | |
912 | IP_VS_ERR_RL("%s(): no memory\n", __func__); | |
913 | return NF_STOLEN; | |
914 | } | |
915 | kfree_skb(skb); | |
916 | skb = new_skb; | |
917 | old_iph = ipv6_hdr(skb); | |
918 | } | |
919 | ||
920 | skb->transport_header = skb->network_header; | |
921 | ||
922 | skb_push(skb, sizeof(struct ipv6hdr)); | |
923 | skb_reset_network_header(skb); | |
924 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); | |
925 | ||
926 | /* drop old route */ | |
927 | skb_dst_drop(skb); | |
928 | skb_dst_set(skb, &rt->dst); | |
929 | ||
930 | /* | |
931 | * Push down and install the IPIP header. | |
932 | */ | |
933 | iph = ipv6_hdr(skb); | |
934 | iph->version = 6; | |
935 | iph->nexthdr = IPPROTO_IPV6; | |
936 | iph->payload_len = old_iph->payload_len; | |
937 | be16_add_cpu(&iph->payload_len, sizeof(*old_iph)); | |
938 | iph->priority = old_iph->priority; | |
939 | memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl)); | |
940 | ipv6_addr_copy(&iph->daddr, &cp->daddr.in6); | |
941 | ipv6_addr_copy(&iph->saddr, &saddr); | |
942 | iph->hop_limit = old_iph->hop_limit; | |
943 | ||
944 | /* Another hack: avoid icmp_send in ip_fragment */ | |
945 | skb->local_df = 1; | |
946 | ||
947 | ret = IP_VS_XMIT_TUNNEL(skb, cp); | |
948 | if (ret == NF_ACCEPT) | |
949 | ip6_local_out(skb); | |
950 | else if (ret == NF_DROP) | |
951 | kfree_skb(skb); | |
952 | ||
953 | LeaveFunction(10); | |
954 | ||
955 | return NF_STOLEN; | |
956 | ||
957 | tx_error_icmp: | |
958 | dst_link_failure(skb); | |
959 | tx_error: | |
960 | kfree_skb(skb); | |
961 | LeaveFunction(10); | |
962 | return NF_STOLEN; | |
963 | tx_error_put: | |
964 | dst_release(&rt->dst); | |
965 | goto tx_error; | |
966 | } | |
967 | #endif | |
968 | ||
969 | ||
970 | /* | |
971 | * Direct Routing transmitter | |
972 | * Used for ANY protocol | |
973 | */ | |
974 | int | |
975 | ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
976 | struct ip_vs_protocol *pp) | |
977 | { | |
978 | struct rtable *rt; /* Route to the other host */ | |
979 | struct iphdr *iph = ip_hdr(skb); | |
980 | int mtu; | |
981 | ||
982 | EnterFunction(10); | |
983 | ||
984 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, | |
985 | RT_TOS(iph->tos), 1|2))) | |
986 | goto tx_error_icmp; | |
987 | if (rt->rt_flags & RTCF_LOCAL) { | |
988 | ip_rt_put(rt); | |
989 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); | |
990 | } | |
991 | ||
992 | /* MTU checking */ | |
993 | mtu = dst_mtu(&rt->dst); | |
994 | if ((iph->frag_off & htons(IP_DF)) && skb->len > mtu) { | |
995 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
996 | ip_rt_put(rt); | |
997 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
998 | goto tx_error; | |
999 | } | |
1000 | ||
1001 | /* | |
1002 | * Call ip_send_check because we are not sure it is called | |
1003 | * after ip_defrag. Is copy-on-write needed? | |
1004 | */ | |
1005 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
1006 | ip_rt_put(rt); | |
1007 | return NF_STOLEN; | |
1008 | } | |
1009 | ip_send_check(ip_hdr(skb)); | |
1010 | ||
1011 | /* drop old route */ | |
1012 | skb_dst_drop(skb); | |
1013 | skb_dst_set(skb, &rt->dst); | |
1014 | ||
1015 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1016 | skb->local_df = 1; | |
1017 | ||
1018 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0); | |
1019 | ||
1020 | LeaveFunction(10); | |
1021 | return NF_STOLEN; | |
1022 | ||
1023 | tx_error_icmp: | |
1024 | dst_link_failure(skb); | |
1025 | tx_error: | |
1026 | kfree_skb(skb); | |
1027 | LeaveFunction(10); | |
1028 | return NF_STOLEN; | |
1029 | } | |
1030 | ||
1031 | #ifdef CONFIG_IP_VS_IPV6 | |
1032 | int | |
1033 | ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1034 | struct ip_vs_protocol *pp) | |
1035 | { | |
1036 | struct rt6_info *rt; /* Route to the other host */ | |
1037 | int mtu; | |
1038 | ||
1039 | EnterFunction(10); | |
1040 | ||
1041 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, | |
1042 | 0, 1|2))) | |
1043 | goto tx_error_icmp; | |
1044 | if (__ip_vs_is_local_route6(rt)) { | |
1045 | dst_release(&rt->dst); | |
1046 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1); | |
1047 | } | |
1048 | ||
1049 | /* MTU checking */ | |
1050 | mtu = dst_mtu(&rt->dst); | |
1051 | if (skb->len > mtu) { | |
1052 | if (!skb->dev) { | |
1053 | struct net *net = dev_net(skb_dst(skb)->dev); | |
1054 | ||
1055 | skb->dev = net->loopback_dev; | |
1056 | } | |
1057 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); | |
1058 | dst_release(&rt->dst); | |
1059 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
1060 | goto tx_error; | |
1061 | } | |
1062 | ||
1063 | /* | |
1064 | * Call ip_send_check because we are not sure it is called | |
1065 | * after ip_defrag. Is copy-on-write needed? | |
1066 | */ | |
1067 | skb = skb_share_check(skb, GFP_ATOMIC); | |
1068 | if (unlikely(skb == NULL)) { | |
1069 | dst_release(&rt->dst); | |
1070 | return NF_STOLEN; | |
1071 | } | |
1072 | ||
1073 | /* drop old route */ | |
1074 | skb_dst_drop(skb); | |
1075 | skb_dst_set(skb, &rt->dst); | |
1076 | ||
1077 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1078 | skb->local_df = 1; | |
1079 | ||
1080 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0); | |
1081 | ||
1082 | LeaveFunction(10); | |
1083 | return NF_STOLEN; | |
1084 | ||
1085 | tx_error_icmp: | |
1086 | dst_link_failure(skb); | |
1087 | tx_error: | |
1088 | kfree_skb(skb); | |
1089 | LeaveFunction(10); | |
1090 | return NF_STOLEN; | |
1091 | } | |
1092 | #endif | |
1093 | ||
1094 | ||
1095 | /* | |
1096 | * ICMP packet transmitter | |
1097 | * called by the ip_vs_in_icmp | |
1098 | */ | |
1099 | int | |
1100 | ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1101 | struct ip_vs_protocol *pp, int offset) | |
1102 | { | |
1103 | struct rtable *rt; /* Route to the other host */ | |
1104 | int mtu; | |
1105 | int rc; | |
1106 | int local; | |
1107 | ||
1108 | EnterFunction(10); | |
1109 | ||
1110 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
1111 | forwarded directly here, because there is no need to | |
1112 | translate address/port back */ | |
1113 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
1114 | if (cp->packet_xmit) | |
1115 | rc = cp->packet_xmit(skb, cp, pp); | |
1116 | else | |
1117 | rc = NF_ACCEPT; | |
1118 | /* do not touch skb anymore */ | |
1119 | atomic_inc(&cp->in_pkts); | |
1120 | goto out; | |
1121 | } | |
1122 | ||
1123 | /* | |
1124 | * mangle and send the packet here (only for VS/NAT) | |
1125 | */ | |
1126 | ||
1127 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, | |
1128 | RT_TOS(ip_hdr(skb)->tos), 1|2|4))) | |
1129 | goto tx_error_icmp; | |
1130 | local = rt->rt_flags & RTCF_LOCAL; | |
1131 | ||
1132 | /* | |
1133 | * Avoid duplicate tuple in reply direction for NAT traffic | |
1134 | * to local address when connection is sync-ed | |
1135 | */ | |
1136 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
1137 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
1138 | enum ip_conntrack_info ctinfo; | |
1139 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
1140 | ||
1141 | if (ct && !nf_ct_is_untracked(ct)) { | |
1142 | IP_VS_DBG(10, "%s(): " | |
1143 | "stopping DNAT to local address %pI4\n", | |
1144 | __func__, &cp->daddr.ip); | |
1145 | goto tx_error_put; | |
1146 | } | |
1147 | } | |
1148 | #endif | |
1149 | ||
1150 | /* From world but DNAT to loopback address? */ | |
1151 | if (local && ipv4_is_loopback(rt->rt_dst) && | |
1152 | rt_is_input_route(skb_rtable(skb))) { | |
1153 | IP_VS_DBG(1, "%s(): " | |
1154 | "stopping DNAT to loopback %pI4\n", | |
1155 | __func__, &cp->daddr.ip); | |
1156 | goto tx_error_put; | |
1157 | } | |
1158 | ||
1159 | /* MTU checking */ | |
1160 | mtu = dst_mtu(&rt->dst); | |
1161 | if ((skb->len > mtu) && (ip_hdr(skb)->frag_off & htons(IP_DF))) { | |
1162 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); | |
1163 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
1164 | goto tx_error_put; | |
1165 | } | |
1166 | ||
1167 | /* copy-on-write the packet before mangling it */ | |
1168 | if (!skb_make_writable(skb, offset)) | |
1169 | goto tx_error_put; | |
1170 | ||
1171 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) | |
1172 | goto tx_error_put; | |
1173 | ||
1174 | ip_vs_nat_icmp(skb, pp, cp, 0); | |
1175 | ||
1176 | if (!local) { | |
1177 | /* drop the old route when skb is not shared */ | |
1178 | skb_dst_drop(skb); | |
1179 | skb_dst_set(skb, &rt->dst); | |
1180 | } else { | |
1181 | ip_rt_put(rt); | |
1182 | /* | |
1183 | * Some IPv4 replies get local address from routes, | |
1184 | * not from iph, so while we DNAT after routing | |
1185 | * we need this second input/output route. | |
1186 | */ | |
1187 | if (!__ip_vs_reroute_locally(skb)) | |
1188 | goto tx_error; | |
1189 | } | |
1190 | ||
1191 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1192 | skb->local_df = 1; | |
1193 | ||
1194 | IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local); | |
1195 | ||
1196 | rc = NF_STOLEN; | |
1197 | goto out; | |
1198 | ||
1199 | tx_error_icmp: | |
1200 | dst_link_failure(skb); | |
1201 | tx_error: | |
1202 | dev_kfree_skb(skb); | |
1203 | rc = NF_STOLEN; | |
1204 | out: | |
1205 | LeaveFunction(10); | |
1206 | return rc; | |
1207 | tx_error_put: | |
1208 | ip_rt_put(rt); | |
1209 | goto tx_error; | |
1210 | } | |
1211 | ||
1212 | #ifdef CONFIG_IP_VS_IPV6 | |
1213 | int | |
1214 | ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1215 | struct ip_vs_protocol *pp, int offset) | |
1216 | { | |
1217 | struct rt6_info *rt; /* Route to the other host */ | |
1218 | int mtu; | |
1219 | int rc; | |
1220 | int local; | |
1221 | ||
1222 | EnterFunction(10); | |
1223 | ||
1224 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
1225 | forwarded directly here, because there is no need to | |
1226 | translate address/port back */ | |
1227 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
1228 | if (cp->packet_xmit) | |
1229 | rc = cp->packet_xmit(skb, cp, pp); | |
1230 | else | |
1231 | rc = NF_ACCEPT; | |
1232 | /* do not touch skb anymore */ | |
1233 | atomic_inc(&cp->in_pkts); | |
1234 | goto out; | |
1235 | } | |
1236 | ||
1237 | /* | |
1238 | * mangle and send the packet here (only for VS/NAT) | |
1239 | */ | |
1240 | ||
1241 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, | |
1242 | 0, 1|2|4))) | |
1243 | goto tx_error_icmp; | |
1244 | ||
1245 | local = __ip_vs_is_local_route6(rt); | |
1246 | /* | |
1247 | * Avoid duplicate tuple in reply direction for NAT traffic | |
1248 | * to local address when connection is sync-ed | |
1249 | */ | |
1250 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
1251 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
1252 | enum ip_conntrack_info ctinfo; | |
1253 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
1254 | ||
1255 | if (ct && !nf_ct_is_untracked(ct)) { | |
1256 | IP_VS_DBG(10, "%s(): " | |
1257 | "stopping DNAT to local address %pI6\n", | |
1258 | __func__, &cp->daddr.in6); | |
1259 | goto tx_error_put; | |
1260 | } | |
1261 | } | |
1262 | #endif | |
1263 | ||
1264 | /* From world but DNAT to loopback address? */ | |
1265 | if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) && | |
1266 | ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) { | |
1267 | IP_VS_DBG(1, "%s(): " | |
1268 | "stopping DNAT to loopback %pI6\n", | |
1269 | __func__, &cp->daddr.in6); | |
1270 | goto tx_error_put; | |
1271 | } | |
1272 | ||
1273 | /* MTU checking */ | |
1274 | mtu = dst_mtu(&rt->dst); | |
1275 | if (skb->len > mtu) { | |
1276 | if (!skb->dev) { | |
1277 | struct net *net = dev_net(skb_dst(skb)->dev); | |
1278 | ||
1279 | skb->dev = net->loopback_dev; | |
1280 | } | |
1281 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); | |
1282 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); | |
1283 | goto tx_error_put; | |
1284 | } | |
1285 | ||
1286 | /* copy-on-write the packet before mangling it */ | |
1287 | if (!skb_make_writable(skb, offset)) | |
1288 | goto tx_error_put; | |
1289 | ||
1290 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) | |
1291 | goto tx_error_put; | |
1292 | ||
1293 | ip_vs_nat_icmp_v6(skb, pp, cp, 0); | |
1294 | ||
1295 | if (!local || !skb->dev) { | |
1296 | /* drop the old route when skb is not shared */ | |
1297 | skb_dst_drop(skb); | |
1298 | skb_dst_set(skb, &rt->dst); | |
1299 | } else { | |
1300 | /* destined to loopback, do we need to change route? */ | |
1301 | dst_release(&rt->dst); | |
1302 | } | |
1303 | ||
1304 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1305 | skb->local_df = 1; | |
1306 | ||
1307 | IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local); | |
1308 | ||
1309 | rc = NF_STOLEN; | |
1310 | goto out; | |
1311 | ||
1312 | tx_error_icmp: | |
1313 | dst_link_failure(skb); | |
1314 | tx_error: | |
1315 | dev_kfree_skb(skb); | |
1316 | rc = NF_STOLEN; | |
1317 | out: | |
1318 | LeaveFunction(10); | |
1319 | return rc; | |
1320 | tx_error_put: | |
1321 | dst_release(&rt->dst); | |
1322 | goto tx_error; | |
1323 | } | |
1324 | #endif |