]>
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 | * | |
cb59155f JA |
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 | |
1da177e4 LT |
24 | */ |
25 | ||
9aada7ac HE |
26 | #define KMSG_COMPONENT "IPVS" |
27 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt | |
28 | ||
1da177e4 | 29 | #include <linux/kernel.h> |
5a0e3ad6 | 30 | #include <linux/slab.h> |
1da177e4 | 31 | #include <linux/tcp.h> /* for tcphdr */ |
c439cb2e | 32 | #include <net/ip.h> |
1da177e4 LT |
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 */ | |
38cdcc9a JV |
37 | #include <net/ipv6.h> |
38 | #include <net/ip6_route.h> | |
714f095f | 39 | #include <net/addrconf.h> |
38cdcc9a | 40 | #include <linux/icmpv6.h> |
1da177e4 LT |
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 | |
714f095f HS |
51 | __ip_vs_dst_set(struct ip_vs_dest *dest, u32 rtos, struct dst_entry *dst, |
52 | u32 dst_cookie) | |
1da177e4 LT |
53 | { |
54 | struct dst_entry *old_dst; | |
55 | ||
56 | old_dst = dest->dst_cache; | |
57 | dest->dst_cache = dst; | |
58 | dest->dst_rtos = rtos; | |
714f095f | 59 | dest->dst_cookie = dst_cookie; |
1da177e4 LT |
60 | dst_release(old_dst); |
61 | } | |
62 | ||
63 | static inline struct dst_entry * | |
714f095f | 64 | __ip_vs_dst_check(struct ip_vs_dest *dest, u32 rtos) |
1da177e4 LT |
65 | { |
66 | struct dst_entry *dst = dest->dst_cache; | |
67 | ||
68 | if (!dst) | |
69 | return NULL; | |
714f095f HS |
70 | if ((dst->obsolete || rtos != dest->dst_rtos) && |
71 | dst->ops->check(dst, dest->dst_cookie) == NULL) { | |
1da177e4 LT |
72 | dest->dst_cache = NULL; |
73 | dst_release(dst); | |
74 | return NULL; | |
75 | } | |
76 | dst_hold(dst); | |
77 | return dst; | |
78 | } | |
79 | ||
fc604767 JA |
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 | */ | |
ad1b30b1 | 85 | static struct rtable * |
fc604767 JA |
86 | __ip_vs_get_out_rt(struct sk_buff *skb, struct ip_vs_dest *dest, |
87 | __be32 daddr, u32 rtos, int rt_mode) | |
1da177e4 | 88 | { |
fc604767 | 89 | struct net *net = dev_net(skb_dst(skb)->dev); |
1da177e4 | 90 | struct rtable *rt; /* Route to the other host */ |
fc604767 JA |
91 | struct rtable *ort; /* Original route */ |
92 | int local; | |
1da177e4 LT |
93 | |
94 | if (dest) { | |
95 | spin_lock(&dest->dst_lock); | |
96 | if (!(rt = (struct rtable *) | |
714f095f | 97 | __ip_vs_dst_check(dest, rtos))) { |
1da177e4 LT |
98 | struct flowi fl = { |
99 | .oif = 0, | |
100 | .nl_u = { | |
101 | .ip4_u = { | |
e7ade46a | 102 | .daddr = dest->addr.ip, |
1da177e4 LT |
103 | .saddr = 0, |
104 | .tos = rtos, } }, | |
105 | }; | |
106 | ||
714f095f | 107 | if (ip_route_output_key(net, &rt, &fl)) { |
1da177e4 | 108 | spin_unlock(&dest->dst_lock); |
14d5e834 HH |
109 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", |
110 | &dest->addr.ip); | |
1da177e4 LT |
111 | return NULL; |
112 | } | |
714f095f | 113 | __ip_vs_dst_set(dest, rtos, dst_clone(&rt->dst), 0); |
14d5e834 HH |
114 | IP_VS_DBG(10, "new dst %pI4, refcnt=%d, rtos=%X\n", |
115 | &dest->addr.ip, | |
d8d1f30b | 116 | atomic_read(&rt->dst.__refcnt), rtos); |
1da177e4 LT |
117 | } |
118 | spin_unlock(&dest->dst_lock); | |
119 | } else { | |
120 | struct flowi fl = { | |
121 | .oif = 0, | |
122 | .nl_u = { | |
123 | .ip4_u = { | |
fc604767 | 124 | .daddr = daddr, |
1da177e4 LT |
125 | .saddr = 0, |
126 | .tos = rtos, } }, | |
127 | }; | |
128 | ||
714f095f | 129 | if (ip_route_output_key(net, &rt, &fl)) { |
14d5e834 | 130 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", |
fc604767 | 131 | &daddr); |
1da177e4 LT |
132 | return NULL; |
133 | } | |
134 | } | |
135 | ||
fc604767 JA |
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 | ||
1da177e4 LT |
160 | return rt; |
161 | } | |
162 | ||
fc604767 JA |
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 | ||
c7537967 | 172 | if (rt_is_input_route(rt)) { |
fc604767 JA |
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 | ||
38cdcc9a | 206 | #ifdef CONFIG_IP_VS_IPV6 |
714f095f | 207 | |
fc604767 JA |
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 | ||
714f095f HS |
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 | ||
fc604767 JA |
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 | */ | |
38cdcc9a | 252 | static struct rt6_info * |
fc604767 JA |
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) | |
38cdcc9a | 256 | { |
fc604767 | 257 | struct net *net = dev_net(skb_dst(skb)->dev); |
38cdcc9a | 258 | struct rt6_info *rt; /* Route to the other host */ |
fc604767 | 259 | struct rt6_info *ort; /* Original route */ |
714f095f | 260 | struct dst_entry *dst; |
fc604767 | 261 | int local; |
38cdcc9a JV |
262 | |
263 | if (dest) { | |
264 | spin_lock(&dest->dst_lock); | |
714f095f | 265 | rt = (struct rt6_info *)__ip_vs_dst_check(dest, 0); |
38cdcc9a | 266 | if (!rt) { |
714f095f | 267 | u32 cookie; |
38cdcc9a | 268 | |
714f095f HS |
269 | dst = __ip_vs_route_output_v6(net, &dest->addr.in6, |
270 | &dest->dst_saddr, | |
271 | do_xfrm); | |
272 | if (!dst) { | |
38cdcc9a | 273 | spin_unlock(&dest->dst_lock); |
38cdcc9a JV |
274 | return NULL; |
275 | } | |
714f095f HS |
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, | |
d8d1f30b | 281 | atomic_read(&rt->dst.__refcnt)); |
38cdcc9a | 282 | } |
714f095f HS |
283 | if (ret_saddr) |
284 | ipv6_addr_copy(ret_saddr, &dest->dst_saddr); | |
38cdcc9a JV |
285 | spin_unlock(&dest->dst_lock); |
286 | } else { | |
fc604767 | 287 | dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm); |
714f095f | 288 | if (!dst) |
38cdcc9a | 289 | return NULL; |
714f095f | 290 | rt = (struct rt6_info *) dst; |
38cdcc9a JV |
291 | } |
292 | ||
fc604767 JA |
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 | ||
38cdcc9a JV |
319 | return rt; |
320 | } | |
321 | #endif | |
322 | ||
1da177e4 LT |
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 | ||
f4bc17cd JA |
337 | #define IP_VS_XMIT_TUNNEL(skb, cp) \ |
338 | ({ \ | |
339 | int __ret = NF_ACCEPT; \ | |
340 | \ | |
cf356d69 | 341 | (skb)->ipvs_property = 1; \ |
f4bc17cd JA |
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); \ | |
4256f1aa | 346 | skb_forward_csum(skb); \ |
f4bc17cd JA |
347 | } \ |
348 | __ret; \ | |
349 | }) | |
350 | ||
fc604767 | 351 | #define IP_VS_XMIT_NAT(pf, skb, cp, local) \ |
1da177e4 | 352 | do { \ |
cf356d69 | 353 | (skb)->ipvs_property = 1; \ |
f4bc17cd | 354 | if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \ |
cf356d69 | 355 | ip_vs_notrack(skb); \ |
f4bc17cd JA |
356 | else \ |
357 | ip_vs_update_conntrack(skb, cp, 1); \ | |
fc604767 JA |
358 | if (local) \ |
359 | return NF_ACCEPT; \ | |
ccc7911f | 360 | skb_forward_csum(skb); \ |
38cdcc9a | 361 | NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \ |
f4bc17cd JA |
362 | skb_dst(skb)->dev, dst_output); \ |
363 | } while (0) | |
364 | ||
fc604767 | 365 | #define IP_VS_XMIT(pf, skb, cp, local) \ |
f4bc17cd | 366 | do { \ |
cf356d69 | 367 | (skb)->ipvs_property = 1; \ |
f4bc17cd | 368 | if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \ |
cf356d69 | 369 | ip_vs_notrack(skb); \ |
fc604767 JA |
370 | if (local) \ |
371 | return NF_ACCEPT; \ | |
f4bc17cd JA |
372 | skb_forward_csum(skb); \ |
373 | NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \ | |
374 | skb_dst(skb)->dev, dst_output); \ | |
1da177e4 LT |
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 */ | |
fc604767 | 386 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); |
1da177e4 LT |
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 */ | |
eddc9ec5 | 400 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 | 401 | int mtu; |
1da177e4 LT |
402 | |
403 | EnterFunction(10); | |
404 | ||
fc604767 JA |
405 | if (!(rt = __ip_vs_get_out_rt(skb, NULL, iph->daddr, |
406 | RT_TOS(iph->tos), 2))) | |
1da177e4 | 407 | goto tx_error_icmp; |
1da177e4 LT |
408 | |
409 | /* MTU checking */ | |
d8d1f30b | 410 | mtu = dst_mtu(&rt->dst); |
4412ec49 | 411 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) { |
1da177e4 LT |
412 | ip_rt_put(rt); |
413 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
1e3e238e | 414 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
1da177e4 LT |
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 | } | |
eddc9ec5 | 426 | ip_send_check(ip_hdr(skb)); |
1da177e4 LT |
427 | |
428 | /* drop old route */ | |
adf30907 | 429 | skb_dst_drop(skb); |
d8d1f30b | 430 | skb_dst_set(skb, &rt->dst); |
1da177e4 LT |
431 | |
432 | /* Another hack: avoid icmp_send in ip_fragment */ | |
433 | skb->local_df = 1; | |
434 | ||
fc604767 | 435 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0); |
1da177e4 LT |
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 | ||
b3cdd2a7 JV |
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; | |
b3cdd2a7 JV |
456 | |
457 | EnterFunction(10); | |
458 | ||
fc604767 | 459 | if (!(rt = __ip_vs_get_out_rt_v6(skb, NULL, &iph->daddr, NULL, 0, 2))) |
b3cdd2a7 | 460 | goto tx_error_icmp; |
b3cdd2a7 JV |
461 | |
462 | /* MTU checking */ | |
d8d1f30b | 463 | mtu = dst_mtu(&rt->dst); |
b3cdd2a7 | 464 | if (skb->len > mtu) { |
cb59155f JA |
465 | if (!skb->dev) { |
466 | struct net *net = dev_net(skb_dst(skb)->dev); | |
467 | ||
468 | skb->dev = net->loopback_dev; | |
469 | } | |
3ffe533c | 470 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
cb59155f | 471 | dst_release(&rt->dst); |
1e3e238e | 472 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
b3cdd2a7 JV |
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)) { | |
d8d1f30b | 482 | dst_release(&rt->dst); |
b3cdd2a7 JV |
483 | return NF_STOLEN; |
484 | } | |
485 | ||
486 | /* drop old route */ | |
adf30907 | 487 | skb_dst_drop(skb); |
d8d1f30b | 488 | skb_dst_set(skb, &rt->dst); |
b3cdd2a7 JV |
489 | |
490 | /* Another hack: avoid icmp_send in ip_fragment */ | |
491 | skb->local_df = 1; | |
492 | ||
fc604767 | 493 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0); |
b3cdd2a7 JV |
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 | |
1da177e4 LT |
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; | |
eddc9ec5 | 517 | struct iphdr *iph = ip_hdr(skb); |
fc604767 | 518 | int local; |
1da177e4 LT |
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)) { | |
014d730d | 524 | __be16 _pt, *p; |
1da177e4 LT |
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 | ||
fc604767 JA |
532 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, |
533 | RT_TOS(iph->tos), 1|2|4))) | |
1da177e4 | 534 | goto tx_error_icmp; |
fc604767 JA |
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)) { | |
0d79641a JA |
546 | IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, 0, |
547 | "ip_vs_nat_xmit(): " | |
fc604767 JA |
548 | "stopping DNAT to local address"); |
549 | goto tx_error_put; | |
550 | } | |
551 | } | |
552 | #endif | |
553 | ||
554 | /* From world but DNAT to loopback address? */ | |
c7537967 DM |
555 | if (local && ipv4_is_loopback(rt->rt_dst) && |
556 | rt_is_input_route(skb_rtable(skb))) { | |
0d79641a | 557 | IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, 0, "ip_vs_nat_xmit(): " |
fc604767 JA |
558 | "stopping DNAT to loopback address"); |
559 | goto tx_error_put; | |
560 | } | |
1da177e4 LT |
561 | |
562 | /* MTU checking */ | |
d8d1f30b | 563 | mtu = dst_mtu(&rt->dst); |
4412ec49 | 564 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) { |
1da177e4 | 565 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); |
0d79641a JA |
566 | IP_VS_DBG_RL_PKT(0, AF_INET, pp, skb, 0, |
567 | "ip_vs_nat_xmit(): frag needed for"); | |
fc604767 | 568 | goto tx_error_put; |
1da177e4 LT |
569 | } |
570 | ||
571 | /* copy-on-write the packet before mangling it */ | |
af1e1cf0 | 572 | if (!skb_make_writable(skb, sizeof(struct iphdr))) |
1da177e4 LT |
573 | goto tx_error_put; |
574 | ||
d8d1f30b | 575 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) |
1da177e4 LT |
576 | goto tx_error_put; |
577 | ||
1da177e4 | 578 | /* mangle the packet */ |
3db05fea | 579 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) |
fc604767 | 580 | goto tx_error_put; |
e7ade46a | 581 | ip_hdr(skb)->daddr = cp->daddr.ip; |
eddc9ec5 | 582 | ip_send_check(ip_hdr(skb)); |
1da177e4 | 583 | |
fc604767 JA |
584 | if (!local) { |
585 | /* drop old route */ | |
586 | skb_dst_drop(skb); | |
587 | skb_dst_set(skb, &rt->dst); | |
588 | } else { | |
589 | ip_rt_put(rt); | |
590 | /* | |
591 | * Some IPv4 replies get local address from routes, | |
592 | * not from iph, so while we DNAT after routing | |
593 | * we need this second input/output route. | |
594 | */ | |
595 | if (!__ip_vs_reroute_locally(skb)) | |
596 | goto tx_error; | |
597 | } | |
598 | ||
0d79641a | 599 | IP_VS_DBG_PKT(10, AF_INET, pp, skb, 0, "After DNAT"); |
1da177e4 LT |
600 | |
601 | /* FIXME: when application helper enlarges the packet and the length | |
602 | is larger than the MTU of outgoing device, there will be still | |
603 | MTU problem. */ | |
604 | ||
605 | /* Another hack: avoid icmp_send in ip_fragment */ | |
606 | skb->local_df = 1; | |
607 | ||
fc604767 | 608 | IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local); |
1da177e4 LT |
609 | |
610 | LeaveFunction(10); | |
611 | return NF_STOLEN; | |
612 | ||
613 | tx_error_icmp: | |
614 | dst_link_failure(skb); | |
615 | tx_error: | |
1da177e4 | 616 | kfree_skb(skb); |
f4bc17cd | 617 | LeaveFunction(10); |
1da177e4 LT |
618 | return NF_STOLEN; |
619 | tx_error_put: | |
620 | ip_rt_put(rt); | |
621 | goto tx_error; | |
622 | } | |
623 | ||
b3cdd2a7 JV |
624 | #ifdef CONFIG_IP_VS_IPV6 |
625 | int | |
626 | ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
627 | struct ip_vs_protocol *pp) | |
628 | { | |
629 | struct rt6_info *rt; /* Route to the other host */ | |
630 | int mtu; | |
fc604767 | 631 | int local; |
b3cdd2a7 JV |
632 | |
633 | EnterFunction(10); | |
634 | ||
635 | /* check if it is a connection of no-client-port */ | |
636 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
637 | __be16 _pt, *p; | |
638 | p = skb_header_pointer(skb, sizeof(struct ipv6hdr), | |
639 | sizeof(_pt), &_pt); | |
640 | if (p == NULL) | |
641 | goto tx_error; | |
642 | ip_vs_conn_fill_cport(cp, *p); | |
643 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
644 | } | |
645 | ||
fc604767 JA |
646 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, |
647 | 0, 1|2|4))) | |
b3cdd2a7 | 648 | goto tx_error_icmp; |
fc604767 JA |
649 | local = __ip_vs_is_local_route6(rt); |
650 | /* | |
651 | * Avoid duplicate tuple in reply direction for NAT traffic | |
652 | * to local address when connection is sync-ed | |
653 | */ | |
654 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
655 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
656 | enum ip_conntrack_info ctinfo; | |
657 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
658 | ||
659 | if (ct && !nf_ct_is_untracked(ct)) { | |
0d79641a | 660 | IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, 0, |
fc604767 JA |
661 | "ip_vs_nat_xmit_v6(): " |
662 | "stopping DNAT to local address"); | |
663 | goto tx_error_put; | |
664 | } | |
665 | } | |
666 | #endif | |
667 | ||
668 | /* From world but DNAT to loopback address? */ | |
669 | if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) && | |
670 | ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) { | |
0d79641a | 671 | IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, 0, |
fc604767 JA |
672 | "ip_vs_nat_xmit_v6(): " |
673 | "stopping DNAT to loopback address"); | |
674 | goto tx_error_put; | |
675 | } | |
b3cdd2a7 JV |
676 | |
677 | /* MTU checking */ | |
d8d1f30b | 678 | mtu = dst_mtu(&rt->dst); |
b3cdd2a7 | 679 | if (skb->len > mtu) { |
cb59155f JA |
680 | if (!skb->dev) { |
681 | struct net *net = dev_net(skb_dst(skb)->dev); | |
682 | ||
683 | skb->dev = net->loopback_dev; | |
684 | } | |
3ffe533c | 685 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
0d79641a | 686 | IP_VS_DBG_RL_PKT(0, AF_INET6, pp, skb, 0, |
b3cdd2a7 | 687 | "ip_vs_nat_xmit_v6(): frag needed for"); |
fc604767 | 688 | goto tx_error_put; |
b3cdd2a7 JV |
689 | } |
690 | ||
691 | /* copy-on-write the packet before mangling it */ | |
692 | if (!skb_make_writable(skb, sizeof(struct ipv6hdr))) | |
693 | goto tx_error_put; | |
694 | ||
d8d1f30b | 695 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) |
b3cdd2a7 JV |
696 | goto tx_error_put; |
697 | ||
b3cdd2a7 JV |
698 | /* mangle the packet */ |
699 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) | |
700 | goto tx_error; | |
fc604767 JA |
701 | ipv6_addr_copy(&ipv6_hdr(skb)->daddr, &cp->daddr.in6); |
702 | ||
703 | if (!local || !skb->dev) { | |
704 | /* drop the old route when skb is not shared */ | |
705 | skb_dst_drop(skb); | |
706 | skb_dst_set(skb, &rt->dst); | |
707 | } else { | |
708 | /* destined to loopback, do we need to change route? */ | |
709 | dst_release(&rt->dst); | |
710 | } | |
b3cdd2a7 | 711 | |
0d79641a | 712 | IP_VS_DBG_PKT(10, AF_INET6, pp, skb, 0, "After DNAT"); |
b3cdd2a7 JV |
713 | |
714 | /* FIXME: when application helper enlarges the packet and the length | |
715 | is larger than the MTU of outgoing device, there will be still | |
716 | MTU problem. */ | |
717 | ||
718 | /* Another hack: avoid icmp_send in ip_fragment */ | |
719 | skb->local_df = 1; | |
720 | ||
fc604767 | 721 | IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local); |
b3cdd2a7 JV |
722 | |
723 | LeaveFunction(10); | |
724 | return NF_STOLEN; | |
725 | ||
726 | tx_error_icmp: | |
727 | dst_link_failure(skb); | |
728 | tx_error: | |
729 | LeaveFunction(10); | |
730 | kfree_skb(skb); | |
731 | return NF_STOLEN; | |
732 | tx_error_put: | |
d8d1f30b | 733 | dst_release(&rt->dst); |
b3cdd2a7 JV |
734 | goto tx_error; |
735 | } | |
736 | #endif | |
737 | ||
1da177e4 LT |
738 | |
739 | /* | |
740 | * IP Tunneling transmitter | |
741 | * | |
742 | * This function encapsulates the packet in a new IP packet, its | |
743 | * destination will be set to cp->daddr. Most code of this function | |
744 | * is taken from ipip.c. | |
745 | * | |
746 | * It is used in VS/TUN cluster. The load balancer selects a real | |
747 | * server from a cluster based on a scheduling algorithm, | |
748 | * encapsulates the request packet and forwards it to the selected | |
749 | * server. For example, all real servers are configured with | |
750 | * "ifconfig tunl0 <Virtual IP Address> up". When the server receives | |
751 | * the encapsulated packet, it will decapsulate the packet, processe | |
752 | * the request and return the response packets directly to the client | |
753 | * without passing the load balancer. This can greatly increase the | |
754 | * scalability of virtual server. | |
755 | * | |
756 | * Used for ANY protocol | |
757 | */ | |
758 | int | |
759 | ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
760 | struct ip_vs_protocol *pp) | |
761 | { | |
762 | struct rtable *rt; /* Route to the other host */ | |
763 | struct net_device *tdev; /* Device to other host */ | |
eddc9ec5 | 764 | struct iphdr *old_iph = ip_hdr(skb); |
1da177e4 | 765 | u8 tos = old_iph->tos; |
76ab608d | 766 | __be16 df = old_iph->frag_off; |
1da177e4 | 767 | struct iphdr *iph; /* Our new IP header */ |
c2636b4d | 768 | unsigned int max_headroom; /* The extra header space needed */ |
1da177e4 | 769 | int mtu; |
f4bc17cd | 770 | int ret; |
1da177e4 LT |
771 | |
772 | EnterFunction(10); | |
773 | ||
fc604767 JA |
774 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, |
775 | RT_TOS(tos), 1|2))) | |
1da177e4 | 776 | goto tx_error_icmp; |
fc604767 JA |
777 | if (rt->rt_flags & RTCF_LOCAL) { |
778 | ip_rt_put(rt); | |
779 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); | |
780 | } | |
1da177e4 | 781 | |
d8d1f30b | 782 | tdev = rt->dst.dev; |
1da177e4 | 783 | |
d8d1f30b | 784 | mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr); |
1da177e4 | 785 | if (mtu < 68) { |
1e3e238e | 786 | IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__); |
fc604767 | 787 | goto tx_error_put; |
1da177e4 | 788 | } |
adf30907 ED |
789 | if (skb_dst(skb)) |
790 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
1da177e4 | 791 | |
4412ec49 | 792 | df |= (old_iph->frag_off & htons(IP_DF)); |
1da177e4 | 793 | |
4412ec49 | 794 | if ((old_iph->frag_off & htons(IP_DF)) |
1da177e4 LT |
795 | && mtu < ntohs(old_iph->tot_len)) { |
796 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
1e3e238e | 797 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
fc604767 | 798 | goto tx_error_put; |
1da177e4 LT |
799 | } |
800 | ||
801 | /* | |
802 | * Okay, now see if we can stuff it in the buffer as-is. | |
803 | */ | |
804 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr); | |
805 | ||
806 | if (skb_headroom(skb) < max_headroom | |
807 | || skb_cloned(skb) || skb_shared(skb)) { | |
808 | struct sk_buff *new_skb = | |
809 | skb_realloc_headroom(skb, max_headroom); | |
810 | if (!new_skb) { | |
811 | ip_rt_put(rt); | |
812 | kfree_skb(skb); | |
1e3e238e | 813 | IP_VS_ERR_RL("%s(): no memory\n", __func__); |
1da177e4 LT |
814 | return NF_STOLEN; |
815 | } | |
816 | kfree_skb(skb); | |
817 | skb = new_skb; | |
eddc9ec5 | 818 | old_iph = ip_hdr(skb); |
1da177e4 LT |
819 | } |
820 | ||
714f095f | 821 | skb->transport_header = skb->network_header; |
1da177e4 LT |
822 | |
823 | /* fix old IP header checksum */ | |
824 | ip_send_check(old_iph); | |
825 | ||
e2d1bca7 ACM |
826 | skb_push(skb, sizeof(struct iphdr)); |
827 | skb_reset_network_header(skb); | |
1da177e4 LT |
828 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); |
829 | ||
830 | /* drop old route */ | |
adf30907 | 831 | skb_dst_drop(skb); |
d8d1f30b | 832 | skb_dst_set(skb, &rt->dst); |
1da177e4 LT |
833 | |
834 | /* | |
835 | * Push down and install the IPIP header. | |
836 | */ | |
eddc9ec5 | 837 | iph = ip_hdr(skb); |
1da177e4 LT |
838 | iph->version = 4; |
839 | iph->ihl = sizeof(struct iphdr)>>2; | |
840 | iph->frag_off = df; | |
841 | iph->protocol = IPPROTO_IPIP; | |
842 | iph->tos = tos; | |
843 | iph->daddr = rt->rt_dst; | |
844 | iph->saddr = rt->rt_src; | |
845 | iph->ttl = old_iph->ttl; | |
d8d1f30b | 846 | ip_select_ident(iph, &rt->dst, NULL); |
1da177e4 LT |
847 | |
848 | /* Another hack: avoid icmp_send in ip_fragment */ | |
849 | skb->local_df = 1; | |
850 | ||
f4bc17cd JA |
851 | ret = IP_VS_XMIT_TUNNEL(skb, cp); |
852 | if (ret == NF_ACCEPT) | |
853 | ip_local_out(skb); | |
854 | else if (ret == NF_DROP) | |
855 | kfree_skb(skb); | |
1da177e4 LT |
856 | |
857 | LeaveFunction(10); | |
858 | ||
859 | return NF_STOLEN; | |
860 | ||
861 | tx_error_icmp: | |
862 | dst_link_failure(skb); | |
863 | tx_error: | |
864 | kfree_skb(skb); | |
865 | LeaveFunction(10); | |
866 | return NF_STOLEN; | |
fc604767 JA |
867 | tx_error_put: |
868 | ip_rt_put(rt); | |
869 | goto tx_error; | |
1da177e4 LT |
870 | } |
871 | ||
b3cdd2a7 JV |
872 | #ifdef CONFIG_IP_VS_IPV6 |
873 | int | |
874 | ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
875 | struct ip_vs_protocol *pp) | |
876 | { | |
877 | struct rt6_info *rt; /* Route to the other host */ | |
714f095f | 878 | struct in6_addr saddr; /* Source for tunnel */ |
b3cdd2a7 JV |
879 | struct net_device *tdev; /* Device to other host */ |
880 | struct ipv6hdr *old_iph = ipv6_hdr(skb); | |
b3cdd2a7 JV |
881 | struct ipv6hdr *iph; /* Our new IP header */ |
882 | unsigned int max_headroom; /* The extra header space needed */ | |
883 | int mtu; | |
f4bc17cd | 884 | int ret; |
b3cdd2a7 JV |
885 | |
886 | EnterFunction(10); | |
887 | ||
fc604767 JA |
888 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, |
889 | &saddr, 1, 1|2))) | |
b3cdd2a7 | 890 | goto tx_error_icmp; |
fc604767 JA |
891 | if (__ip_vs_is_local_route6(rt)) { |
892 | dst_release(&rt->dst); | |
893 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1); | |
894 | } | |
b3cdd2a7 | 895 | |
d8d1f30b | 896 | tdev = rt->dst.dev; |
b3cdd2a7 | 897 | |
d8d1f30b | 898 | mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr); |
714f095f | 899 | if (mtu < IPV6_MIN_MTU) { |
714f095f HS |
900 | IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__, |
901 | IPV6_MIN_MTU); | |
fc604767 | 902 | goto tx_error_put; |
b3cdd2a7 | 903 | } |
adf30907 ED |
904 | if (skb_dst(skb)) |
905 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
b3cdd2a7 JV |
906 | |
907 | if (mtu < ntohs(old_iph->payload_len) + sizeof(struct ipv6hdr)) { | |
cb59155f JA |
908 | if (!skb->dev) { |
909 | struct net *net = dev_net(skb_dst(skb)->dev); | |
910 | ||
911 | skb->dev = net->loopback_dev; | |
912 | } | |
3ffe533c | 913 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
1e3e238e | 914 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
fc604767 | 915 | goto tx_error_put; |
b3cdd2a7 JV |
916 | } |
917 | ||
918 | /* | |
919 | * Okay, now see if we can stuff it in the buffer as-is. | |
920 | */ | |
921 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr); | |
922 | ||
923 | if (skb_headroom(skb) < max_headroom | |
924 | || skb_cloned(skb) || skb_shared(skb)) { | |
925 | struct sk_buff *new_skb = | |
926 | skb_realloc_headroom(skb, max_headroom); | |
927 | if (!new_skb) { | |
d8d1f30b | 928 | dst_release(&rt->dst); |
b3cdd2a7 | 929 | kfree_skb(skb); |
1e3e238e | 930 | IP_VS_ERR_RL("%s(): no memory\n", __func__); |
b3cdd2a7 JV |
931 | return NF_STOLEN; |
932 | } | |
933 | kfree_skb(skb); | |
934 | skb = new_skb; | |
935 | old_iph = ipv6_hdr(skb); | |
936 | } | |
937 | ||
714f095f | 938 | skb->transport_header = skb->network_header; |
b3cdd2a7 JV |
939 | |
940 | skb_push(skb, sizeof(struct ipv6hdr)); | |
941 | skb_reset_network_header(skb); | |
942 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); | |
943 | ||
944 | /* drop old route */ | |
adf30907 | 945 | skb_dst_drop(skb); |
d8d1f30b | 946 | skb_dst_set(skb, &rt->dst); |
b3cdd2a7 JV |
947 | |
948 | /* | |
949 | * Push down and install the IPIP header. | |
950 | */ | |
951 | iph = ipv6_hdr(skb); | |
952 | iph->version = 6; | |
953 | iph->nexthdr = IPPROTO_IPV6; | |
b7b45f47 HH |
954 | iph->payload_len = old_iph->payload_len; |
955 | be16_add_cpu(&iph->payload_len, sizeof(*old_iph)); | |
b3cdd2a7 JV |
956 | iph->priority = old_iph->priority; |
957 | memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl)); | |
714f095f HS |
958 | ipv6_addr_copy(&iph->daddr, &cp->daddr.in6); |
959 | ipv6_addr_copy(&iph->saddr, &saddr); | |
b3cdd2a7 JV |
960 | iph->hop_limit = old_iph->hop_limit; |
961 | ||
962 | /* Another hack: avoid icmp_send in ip_fragment */ | |
963 | skb->local_df = 1; | |
964 | ||
f4bc17cd JA |
965 | ret = IP_VS_XMIT_TUNNEL(skb, cp); |
966 | if (ret == NF_ACCEPT) | |
967 | ip6_local_out(skb); | |
968 | else if (ret == NF_DROP) | |
969 | kfree_skb(skb); | |
b3cdd2a7 JV |
970 | |
971 | LeaveFunction(10); | |
972 | ||
973 | return NF_STOLEN; | |
974 | ||
975 | tx_error_icmp: | |
976 | dst_link_failure(skb); | |
977 | tx_error: | |
978 | kfree_skb(skb); | |
979 | LeaveFunction(10); | |
980 | return NF_STOLEN; | |
fc604767 JA |
981 | tx_error_put: |
982 | dst_release(&rt->dst); | |
983 | goto tx_error; | |
b3cdd2a7 JV |
984 | } |
985 | #endif | |
986 | ||
1da177e4 LT |
987 | |
988 | /* | |
989 | * Direct Routing transmitter | |
990 | * Used for ANY protocol | |
991 | */ | |
992 | int | |
993 | ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
994 | struct ip_vs_protocol *pp) | |
995 | { | |
996 | struct rtable *rt; /* Route to the other host */ | |
eddc9ec5 | 997 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
998 | int mtu; |
999 | ||
1000 | EnterFunction(10); | |
1001 | ||
fc604767 JA |
1002 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, |
1003 | RT_TOS(iph->tos), 1|2))) | |
1da177e4 | 1004 | goto tx_error_icmp; |
fc604767 JA |
1005 | if (rt->rt_flags & RTCF_LOCAL) { |
1006 | ip_rt_put(rt); | |
1007 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); | |
1008 | } | |
1da177e4 LT |
1009 | |
1010 | /* MTU checking */ | |
d8d1f30b | 1011 | mtu = dst_mtu(&rt->dst); |
4412ec49 | 1012 | if ((iph->frag_off & htons(IP_DF)) && skb->len > mtu) { |
1da177e4 LT |
1013 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); |
1014 | ip_rt_put(rt); | |
1e3e238e | 1015 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
1da177e4 LT |
1016 | goto tx_error; |
1017 | } | |
1018 | ||
1019 | /* | |
1020 | * Call ip_send_check because we are not sure it is called | |
1021 | * after ip_defrag. Is copy-on-write needed? | |
1022 | */ | |
1023 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
1024 | ip_rt_put(rt); | |
1025 | return NF_STOLEN; | |
1026 | } | |
eddc9ec5 | 1027 | ip_send_check(ip_hdr(skb)); |
1da177e4 LT |
1028 | |
1029 | /* drop old route */ | |
adf30907 | 1030 | skb_dst_drop(skb); |
d8d1f30b | 1031 | skb_dst_set(skb, &rt->dst); |
1da177e4 LT |
1032 | |
1033 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1034 | skb->local_df = 1; | |
1035 | ||
fc604767 | 1036 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0); |
1da177e4 LT |
1037 | |
1038 | LeaveFunction(10); | |
1039 | return NF_STOLEN; | |
1040 | ||
1041 | tx_error_icmp: | |
1042 | dst_link_failure(skb); | |
1043 | tx_error: | |
1044 | kfree_skb(skb); | |
1045 | LeaveFunction(10); | |
1046 | return NF_STOLEN; | |
1047 | } | |
1048 | ||
b3cdd2a7 JV |
1049 | #ifdef CONFIG_IP_VS_IPV6 |
1050 | int | |
1051 | ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1052 | struct ip_vs_protocol *pp) | |
1053 | { | |
1054 | struct rt6_info *rt; /* Route to the other host */ | |
1055 | int mtu; | |
1056 | ||
1057 | EnterFunction(10); | |
1058 | ||
fc604767 JA |
1059 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, |
1060 | 0, 1|2))) | |
b3cdd2a7 | 1061 | goto tx_error_icmp; |
fc604767 JA |
1062 | if (__ip_vs_is_local_route6(rt)) { |
1063 | dst_release(&rt->dst); | |
1064 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1); | |
1065 | } | |
b3cdd2a7 JV |
1066 | |
1067 | /* MTU checking */ | |
d8d1f30b | 1068 | mtu = dst_mtu(&rt->dst); |
b3cdd2a7 | 1069 | if (skb->len > mtu) { |
cb59155f JA |
1070 | if (!skb->dev) { |
1071 | struct net *net = dev_net(skb_dst(skb)->dev); | |
1072 | ||
1073 | skb->dev = net->loopback_dev; | |
1074 | } | |
3ffe533c | 1075 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
d8d1f30b | 1076 | dst_release(&rt->dst); |
1e3e238e | 1077 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
b3cdd2a7 JV |
1078 | goto tx_error; |
1079 | } | |
1080 | ||
1081 | /* | |
1082 | * Call ip_send_check because we are not sure it is called | |
1083 | * after ip_defrag. Is copy-on-write needed? | |
1084 | */ | |
1085 | skb = skb_share_check(skb, GFP_ATOMIC); | |
1086 | if (unlikely(skb == NULL)) { | |
d8d1f30b | 1087 | dst_release(&rt->dst); |
b3cdd2a7 JV |
1088 | return NF_STOLEN; |
1089 | } | |
1090 | ||
1091 | /* drop old route */ | |
adf30907 | 1092 | skb_dst_drop(skb); |
d8d1f30b | 1093 | skb_dst_set(skb, &rt->dst); |
b3cdd2a7 JV |
1094 | |
1095 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1096 | skb->local_df = 1; | |
1097 | ||
fc604767 | 1098 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0); |
b3cdd2a7 JV |
1099 | |
1100 | LeaveFunction(10); | |
1101 | return NF_STOLEN; | |
1102 | ||
1103 | tx_error_icmp: | |
1104 | dst_link_failure(skb); | |
1105 | tx_error: | |
1106 | kfree_skb(skb); | |
1107 | LeaveFunction(10); | |
1108 | return NF_STOLEN; | |
1109 | } | |
1110 | #endif | |
1111 | ||
1da177e4 LT |
1112 | |
1113 | /* | |
1114 | * ICMP packet transmitter | |
1115 | * called by the ip_vs_in_icmp | |
1116 | */ | |
1117 | int | |
1118 | ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1119 | struct ip_vs_protocol *pp, int offset) | |
1120 | { | |
1121 | struct rtable *rt; /* Route to the other host */ | |
1122 | int mtu; | |
1123 | int rc; | |
fc604767 | 1124 | int local; |
1da177e4 LT |
1125 | |
1126 | EnterFunction(10); | |
1127 | ||
1128 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
1129 | forwarded directly here, because there is no need to | |
1130 | translate address/port back */ | |
1131 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
1132 | if (cp->packet_xmit) | |
1133 | rc = cp->packet_xmit(skb, cp, pp); | |
1134 | else | |
1135 | rc = NF_ACCEPT; | |
1136 | /* do not touch skb anymore */ | |
1137 | atomic_inc(&cp->in_pkts); | |
1da177e4 LT |
1138 | goto out; |
1139 | } | |
1140 | ||
1141 | /* | |
1142 | * mangle and send the packet here (only for VS/NAT) | |
1143 | */ | |
1144 | ||
fc604767 JA |
1145 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, |
1146 | RT_TOS(ip_hdr(skb)->tos), 1|2|4))) | |
1da177e4 | 1147 | goto tx_error_icmp; |
fc604767 JA |
1148 | local = rt->rt_flags & RTCF_LOCAL; |
1149 | ||
1150 | /* | |
1151 | * Avoid duplicate tuple in reply direction for NAT traffic | |
1152 | * to local address when connection is sync-ed | |
1153 | */ | |
1154 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
1155 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
1156 | enum ip_conntrack_info ctinfo; | |
1157 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
1158 | ||
1159 | if (ct && !nf_ct_is_untracked(ct)) { | |
1160 | IP_VS_DBG(10, "%s(): " | |
1161 | "stopping DNAT to local address %pI4\n", | |
1162 | __func__, &cp->daddr.ip); | |
1163 | goto tx_error_put; | |
1164 | } | |
1165 | } | |
1166 | #endif | |
1167 | ||
1168 | /* From world but DNAT to loopback address? */ | |
c7537967 DM |
1169 | if (local && ipv4_is_loopback(rt->rt_dst) && |
1170 | rt_is_input_route(skb_rtable(skb))) { | |
fc604767 JA |
1171 | IP_VS_DBG(1, "%s(): " |
1172 | "stopping DNAT to loopback %pI4\n", | |
1173 | __func__, &cp->daddr.ip); | |
1174 | goto tx_error_put; | |
1175 | } | |
1da177e4 LT |
1176 | |
1177 | /* MTU checking */ | |
d8d1f30b | 1178 | mtu = dst_mtu(&rt->dst); |
eddc9ec5 | 1179 | if ((skb->len > mtu) && (ip_hdr(skb)->frag_off & htons(IP_DF))) { |
1da177e4 | 1180 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); |
1e3e238e | 1181 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
fc604767 | 1182 | goto tx_error_put; |
1da177e4 LT |
1183 | } |
1184 | ||
1185 | /* copy-on-write the packet before mangling it */ | |
af1e1cf0 | 1186 | if (!skb_make_writable(skb, offset)) |
1da177e4 LT |
1187 | goto tx_error_put; |
1188 | ||
d8d1f30b | 1189 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) |
1da177e4 LT |
1190 | goto tx_error_put; |
1191 | ||
1da177e4 LT |
1192 | ip_vs_nat_icmp(skb, pp, cp, 0); |
1193 | ||
fc604767 JA |
1194 | if (!local) { |
1195 | /* drop the old route when skb is not shared */ | |
1196 | skb_dst_drop(skb); | |
1197 | skb_dst_set(skb, &rt->dst); | |
1198 | } else { | |
1199 | ip_rt_put(rt); | |
1200 | /* | |
1201 | * Some IPv4 replies get local address from routes, | |
1202 | * not from iph, so while we DNAT after routing | |
1203 | * we need this second input/output route. | |
1204 | */ | |
1205 | if (!__ip_vs_reroute_locally(skb)) | |
1206 | goto tx_error; | |
1207 | } | |
1208 | ||
1da177e4 LT |
1209 | /* Another hack: avoid icmp_send in ip_fragment */ |
1210 | skb->local_df = 1; | |
1211 | ||
fc604767 | 1212 | IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local); |
1da177e4 LT |
1213 | |
1214 | rc = NF_STOLEN; | |
1215 | goto out; | |
1216 | ||
1217 | tx_error_icmp: | |
1218 | dst_link_failure(skb); | |
1219 | tx_error: | |
1220 | dev_kfree_skb(skb); | |
1221 | rc = NF_STOLEN; | |
1222 | out: | |
1223 | LeaveFunction(10); | |
1224 | return rc; | |
1225 | tx_error_put: | |
1226 | ip_rt_put(rt); | |
1227 | goto tx_error; | |
1228 | } | |
b3cdd2a7 JV |
1229 | |
1230 | #ifdef CONFIG_IP_VS_IPV6 | |
1231 | int | |
1232 | ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1233 | struct ip_vs_protocol *pp, int offset) | |
1234 | { | |
1235 | struct rt6_info *rt; /* Route to the other host */ | |
1236 | int mtu; | |
1237 | int rc; | |
fc604767 | 1238 | int local; |
b3cdd2a7 JV |
1239 | |
1240 | EnterFunction(10); | |
1241 | ||
1242 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
1243 | forwarded directly here, because there is no need to | |
1244 | translate address/port back */ | |
1245 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
1246 | if (cp->packet_xmit) | |
1247 | rc = cp->packet_xmit(skb, cp, pp); | |
1248 | else | |
1249 | rc = NF_ACCEPT; | |
1250 | /* do not touch skb anymore */ | |
1251 | atomic_inc(&cp->in_pkts); | |
1252 | goto out; | |
1253 | } | |
1254 | ||
1255 | /* | |
1256 | * mangle and send the packet here (only for VS/NAT) | |
1257 | */ | |
1258 | ||
fc604767 JA |
1259 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, |
1260 | 0, 1|2|4))) | |
b3cdd2a7 JV |
1261 | goto tx_error_icmp; |
1262 | ||
fc604767 JA |
1263 | local = __ip_vs_is_local_route6(rt); |
1264 | /* | |
1265 | * Avoid duplicate tuple in reply direction for NAT traffic | |
1266 | * to local address when connection is sync-ed | |
1267 | */ | |
1268 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
1269 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
1270 | enum ip_conntrack_info ctinfo; | |
1271 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
1272 | ||
1273 | if (ct && !nf_ct_is_untracked(ct)) { | |
1274 | IP_VS_DBG(10, "%s(): " | |
1275 | "stopping DNAT to local address %pI6\n", | |
1276 | __func__, &cp->daddr.in6); | |
1277 | goto tx_error_put; | |
1278 | } | |
1279 | } | |
1280 | #endif | |
1281 | ||
1282 | /* From world but DNAT to loopback address? */ | |
1283 | if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) && | |
1284 | ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) { | |
1285 | IP_VS_DBG(1, "%s(): " | |
1286 | "stopping DNAT to loopback %pI6\n", | |
1287 | __func__, &cp->daddr.in6); | |
1288 | goto tx_error_put; | |
1289 | } | |
1290 | ||
b3cdd2a7 | 1291 | /* MTU checking */ |
d8d1f30b | 1292 | mtu = dst_mtu(&rt->dst); |
b3cdd2a7 | 1293 | if (skb->len > mtu) { |
cb59155f JA |
1294 | if (!skb->dev) { |
1295 | struct net *net = dev_net(skb_dst(skb)->dev); | |
1296 | ||
1297 | skb->dev = net->loopback_dev; | |
1298 | } | |
3ffe533c | 1299 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
1e3e238e | 1300 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
fc604767 | 1301 | goto tx_error_put; |
b3cdd2a7 JV |
1302 | } |
1303 | ||
1304 | /* copy-on-write the packet before mangling it */ | |
1305 | if (!skb_make_writable(skb, offset)) | |
1306 | goto tx_error_put; | |
1307 | ||
d8d1f30b | 1308 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) |
b3cdd2a7 JV |
1309 | goto tx_error_put; |
1310 | ||
b3cdd2a7 JV |
1311 | ip_vs_nat_icmp_v6(skb, pp, cp, 0); |
1312 | ||
fc604767 JA |
1313 | if (!local || !skb->dev) { |
1314 | /* drop the old route when skb is not shared */ | |
1315 | skb_dst_drop(skb); | |
1316 | skb_dst_set(skb, &rt->dst); | |
1317 | } else { | |
1318 | /* destined to loopback, do we need to change route? */ | |
1319 | dst_release(&rt->dst); | |
1320 | } | |
1321 | ||
b3cdd2a7 JV |
1322 | /* Another hack: avoid icmp_send in ip_fragment */ |
1323 | skb->local_df = 1; | |
1324 | ||
fc604767 | 1325 | IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local); |
b3cdd2a7 JV |
1326 | |
1327 | rc = NF_STOLEN; | |
1328 | goto out; | |
1329 | ||
1330 | tx_error_icmp: | |
1331 | dst_link_failure(skb); | |
1332 | tx_error: | |
1333 | dev_kfree_skb(skb); | |
1334 | rc = NF_STOLEN; | |
1335 | out: | |
1336 | LeaveFunction(10); | |
1337 | return rc; | |
1338 | tx_error_put: | |
d8d1f30b | 1339 | dst_release(&rt->dst); |
b3cdd2a7 JV |
1340 | goto tx_error; |
1341 | } | |
1342 | #endif |