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 | ||
17a8f8e3 CG |
46 | enum { |
47 | IP_VS_RT_MODE_LOCAL = 1, /* Allow local dest */ | |
48 | IP_VS_RT_MODE_NON_LOCAL = 2, /* Allow non-local dest */ | |
49 | IP_VS_RT_MODE_RDR = 4, /* Allow redirect from remote daddr to | |
50 | * local | |
51 | */ | |
52 | }; | |
1da177e4 LT |
53 | |
54 | /* | |
55 | * Destination cache to speed up outgoing route lookup | |
56 | */ | |
57 | static inline void | |
714f095f HS |
58 | __ip_vs_dst_set(struct ip_vs_dest *dest, u32 rtos, struct dst_entry *dst, |
59 | u32 dst_cookie) | |
1da177e4 LT |
60 | { |
61 | struct dst_entry *old_dst; | |
62 | ||
63 | old_dst = dest->dst_cache; | |
64 | dest->dst_cache = dst; | |
65 | dest->dst_rtos = rtos; | |
714f095f | 66 | dest->dst_cookie = dst_cookie; |
1da177e4 LT |
67 | dst_release(old_dst); |
68 | } | |
69 | ||
70 | static inline struct dst_entry * | |
714f095f | 71 | __ip_vs_dst_check(struct ip_vs_dest *dest, u32 rtos) |
1da177e4 LT |
72 | { |
73 | struct dst_entry *dst = dest->dst_cache; | |
74 | ||
75 | if (!dst) | |
76 | return NULL; | |
714f095f HS |
77 | if ((dst->obsolete || rtos != dest->dst_rtos) && |
78 | dst->ops->check(dst, dest->dst_cookie) == NULL) { | |
1da177e4 LT |
79 | dest->dst_cache = NULL; |
80 | dst_release(dst); | |
81 | return NULL; | |
82 | } | |
83 | dst_hold(dst); | |
84 | return dst; | |
85 | } | |
86 | ||
17a8f8e3 | 87 | /* Get route to destination or remote server */ |
ad1b30b1 | 88 | static struct rtable * |
fc604767 JA |
89 | __ip_vs_get_out_rt(struct sk_buff *skb, struct ip_vs_dest *dest, |
90 | __be32 daddr, u32 rtos, int rt_mode) | |
1da177e4 | 91 | { |
fc604767 | 92 | struct net *net = dev_net(skb_dst(skb)->dev); |
1da177e4 | 93 | struct rtable *rt; /* Route to the other host */ |
fc604767 JA |
94 | struct rtable *ort; /* Original route */ |
95 | int local; | |
1da177e4 LT |
96 | |
97 | if (dest) { | |
98 | spin_lock(&dest->dst_lock); | |
99 | if (!(rt = (struct rtable *) | |
714f095f | 100 | __ip_vs_dst_check(dest, rtos))) { |
78fbfd8a | 101 | rt = ip_route_output(net, dest->addr.ip, 0, rtos, 0); |
b23dd4fe | 102 | if (IS_ERR(rt)) { |
1da177e4 | 103 | spin_unlock(&dest->dst_lock); |
14d5e834 HH |
104 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", |
105 | &dest->addr.ip); | |
1da177e4 LT |
106 | return NULL; |
107 | } | |
714f095f | 108 | __ip_vs_dst_set(dest, rtos, dst_clone(&rt->dst), 0); |
14d5e834 HH |
109 | IP_VS_DBG(10, "new dst %pI4, refcnt=%d, rtos=%X\n", |
110 | &dest->addr.ip, | |
d8d1f30b | 111 | atomic_read(&rt->dst.__refcnt), rtos); |
1da177e4 LT |
112 | } |
113 | spin_unlock(&dest->dst_lock); | |
114 | } else { | |
78fbfd8a | 115 | rt = ip_route_output(net, daddr, 0, rtos, 0); |
b23dd4fe | 116 | if (IS_ERR(rt)) { |
14d5e834 | 117 | IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", |
fc604767 | 118 | &daddr); |
1da177e4 LT |
119 | return NULL; |
120 | } | |
121 | } | |
122 | ||
fc604767 | 123 | local = rt->rt_flags & RTCF_LOCAL; |
17a8f8e3 CG |
124 | if (!((local ? IP_VS_RT_MODE_LOCAL : IP_VS_RT_MODE_NON_LOCAL) & |
125 | rt_mode)) { | |
fc604767 JA |
126 | IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI4\n", |
127 | (rt->rt_flags & RTCF_LOCAL) ? | |
128 | "local":"non-local", &rt->rt_dst); | |
129 | ip_rt_put(rt); | |
130 | return NULL; | |
131 | } | |
17a8f8e3 CG |
132 | if (local && !(rt_mode & IP_VS_RT_MODE_RDR) && |
133 | !((ort = skb_rtable(skb)) && ort->rt_flags & RTCF_LOCAL)) { | |
fc604767 JA |
134 | IP_VS_DBG_RL("Redirect from non-local address %pI4 to local " |
135 | "requires NAT method, dest: %pI4\n", | |
136 | &ip_hdr(skb)->daddr, &rt->rt_dst); | |
137 | ip_rt_put(rt); | |
138 | return NULL; | |
139 | } | |
140 | if (unlikely(!local && ipv4_is_loopback(ip_hdr(skb)->saddr))) { | |
141 | IP_VS_DBG_RL("Stopping traffic from loopback address %pI4 " | |
142 | "to non-local address, dest: %pI4\n", | |
143 | &ip_hdr(skb)->saddr, &rt->rt_dst); | |
144 | ip_rt_put(rt); | |
145 | return NULL; | |
146 | } | |
147 | ||
1da177e4 LT |
148 | return rt; |
149 | } | |
150 | ||
fc604767 JA |
151 | /* Reroute packet to local IPv4 stack after DNAT */ |
152 | static int | |
153 | __ip_vs_reroute_locally(struct sk_buff *skb) | |
154 | { | |
155 | struct rtable *rt = skb_rtable(skb); | |
156 | struct net_device *dev = rt->dst.dev; | |
157 | struct net *net = dev_net(dev); | |
158 | struct iphdr *iph = ip_hdr(skb); | |
159 | ||
c7537967 | 160 | if (rt_is_input_route(rt)) { |
fc604767 JA |
161 | unsigned long orefdst = skb->_skb_refdst; |
162 | ||
163 | if (ip_route_input(skb, iph->daddr, iph->saddr, | |
164 | iph->tos, skb->dev)) | |
165 | return 0; | |
166 | refdst_drop(orefdst); | |
167 | } else { | |
9d6ec938 DM |
168 | struct flowi4 fl4 = { |
169 | .daddr = iph->daddr, | |
170 | .saddr = iph->saddr, | |
171 | .flowi4_tos = RT_TOS(iph->tos), | |
172 | .flowi4_mark = skb->mark, | |
fc604767 | 173 | }; |
fc604767 | 174 | |
9d6ec938 | 175 | rt = ip_route_output_key(net, &fl4); |
b23dd4fe | 176 | if (IS_ERR(rt)) |
fc604767 JA |
177 | return 0; |
178 | if (!(rt->rt_flags & RTCF_LOCAL)) { | |
179 | ip_rt_put(rt); | |
180 | return 0; | |
181 | } | |
182 | /* Drop old route. */ | |
183 | skb_dst_drop(skb); | |
184 | skb_dst_set(skb, &rt->dst); | |
185 | } | |
186 | return 1; | |
187 | } | |
188 | ||
38cdcc9a | 189 | #ifdef CONFIG_IP_VS_IPV6 |
714f095f | 190 | |
fc604767 JA |
191 | static inline int __ip_vs_is_local_route6(struct rt6_info *rt) |
192 | { | |
193 | return rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK; | |
194 | } | |
195 | ||
714f095f HS |
196 | static struct dst_entry * |
197 | __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr, | |
198 | struct in6_addr *ret_saddr, int do_xfrm) | |
199 | { | |
200 | struct dst_entry *dst; | |
4c9483b2 DM |
201 | struct flowi6 fl6 = { |
202 | .daddr = *daddr, | |
714f095f HS |
203 | }; |
204 | ||
4c9483b2 | 205 | dst = ip6_route_output(net, NULL, &fl6); |
714f095f HS |
206 | if (dst->error) |
207 | goto out_err; | |
208 | if (!ret_saddr) | |
209 | return dst; | |
4c9483b2 | 210 | if (ipv6_addr_any(&fl6.saddr) && |
714f095f | 211 | ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev, |
4c9483b2 | 212 | &fl6.daddr, 0, &fl6.saddr) < 0) |
714f095f | 213 | goto out_err; |
452edd59 | 214 | if (do_xfrm) { |
4c9483b2 | 215 | dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0); |
452edd59 DM |
216 | if (IS_ERR(dst)) { |
217 | dst = NULL; | |
218 | goto out_err; | |
219 | } | |
220 | } | |
4c9483b2 | 221 | ipv6_addr_copy(ret_saddr, &fl6.saddr); |
714f095f HS |
222 | return dst; |
223 | ||
224 | out_err: | |
225 | dst_release(dst); | |
226 | IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr); | |
227 | return NULL; | |
228 | } | |
229 | ||
fc604767 JA |
230 | /* |
231 | * Get route to destination or remote server | |
232 | * rt_mode: flags, &1=Allow local dest, &2=Allow non-local dest, | |
233 | * &4=Allow redirect from remote daddr to local | |
234 | */ | |
38cdcc9a | 235 | static struct rt6_info * |
fc604767 JA |
236 | __ip_vs_get_out_rt_v6(struct sk_buff *skb, struct ip_vs_dest *dest, |
237 | struct in6_addr *daddr, struct in6_addr *ret_saddr, | |
238 | int do_xfrm, int rt_mode) | |
38cdcc9a | 239 | { |
fc604767 | 240 | struct net *net = dev_net(skb_dst(skb)->dev); |
38cdcc9a | 241 | struct rt6_info *rt; /* Route to the other host */ |
fc604767 | 242 | struct rt6_info *ort; /* Original route */ |
714f095f | 243 | struct dst_entry *dst; |
fc604767 | 244 | int local; |
38cdcc9a JV |
245 | |
246 | if (dest) { | |
247 | spin_lock(&dest->dst_lock); | |
714f095f | 248 | rt = (struct rt6_info *)__ip_vs_dst_check(dest, 0); |
38cdcc9a | 249 | if (!rt) { |
714f095f | 250 | u32 cookie; |
38cdcc9a | 251 | |
714f095f HS |
252 | dst = __ip_vs_route_output_v6(net, &dest->addr.in6, |
253 | &dest->dst_saddr, | |
254 | do_xfrm); | |
255 | if (!dst) { | |
38cdcc9a | 256 | spin_unlock(&dest->dst_lock); |
38cdcc9a JV |
257 | return NULL; |
258 | } | |
714f095f HS |
259 | rt = (struct rt6_info *) dst; |
260 | cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0; | |
261 | __ip_vs_dst_set(dest, 0, dst_clone(&rt->dst), cookie); | |
262 | IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n", | |
263 | &dest->addr.in6, &dest->dst_saddr, | |
d8d1f30b | 264 | atomic_read(&rt->dst.__refcnt)); |
38cdcc9a | 265 | } |
714f095f HS |
266 | if (ret_saddr) |
267 | ipv6_addr_copy(ret_saddr, &dest->dst_saddr); | |
38cdcc9a JV |
268 | spin_unlock(&dest->dst_lock); |
269 | } else { | |
fc604767 | 270 | dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm); |
714f095f | 271 | if (!dst) |
38cdcc9a | 272 | return NULL; |
714f095f | 273 | rt = (struct rt6_info *) dst; |
38cdcc9a JV |
274 | } |
275 | ||
fc604767 JA |
276 | local = __ip_vs_is_local_route6(rt); |
277 | if (!((local ? 1 : 2) & rt_mode)) { | |
278 | IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI6\n", | |
279 | local ? "local":"non-local", daddr); | |
280 | dst_release(&rt->dst); | |
281 | return NULL; | |
282 | } | |
283 | if (local && !(rt_mode & 4) && | |
284 | !((ort = (struct rt6_info *) skb_dst(skb)) && | |
285 | __ip_vs_is_local_route6(ort))) { | |
286 | IP_VS_DBG_RL("Redirect from non-local address %pI6 to local " | |
287 | "requires NAT method, dest: %pI6\n", | |
288 | &ipv6_hdr(skb)->daddr, daddr); | |
289 | dst_release(&rt->dst); | |
290 | return NULL; | |
291 | } | |
292 | if (unlikely(!local && (!skb->dev || skb->dev->flags & IFF_LOOPBACK) && | |
293 | ipv6_addr_type(&ipv6_hdr(skb)->saddr) & | |
294 | IPV6_ADDR_LOOPBACK)) { | |
295 | IP_VS_DBG_RL("Stopping traffic from loopback address %pI6 " | |
296 | "to non-local address, dest: %pI6\n", | |
297 | &ipv6_hdr(skb)->saddr, daddr); | |
298 | dst_release(&rt->dst); | |
299 | return NULL; | |
300 | } | |
301 | ||
38cdcc9a JV |
302 | return rt; |
303 | } | |
304 | #endif | |
305 | ||
1da177e4 LT |
306 | |
307 | /* | |
308 | * Release dest->dst_cache before a dest is removed | |
309 | */ | |
310 | void | |
311 | ip_vs_dst_reset(struct ip_vs_dest *dest) | |
312 | { | |
313 | struct dst_entry *old_dst; | |
314 | ||
315 | old_dst = dest->dst_cache; | |
316 | dest->dst_cache = NULL; | |
317 | dst_release(old_dst); | |
318 | } | |
319 | ||
f4bc17cd JA |
320 | #define IP_VS_XMIT_TUNNEL(skb, cp) \ |
321 | ({ \ | |
322 | int __ret = NF_ACCEPT; \ | |
323 | \ | |
cf356d69 | 324 | (skb)->ipvs_property = 1; \ |
f4bc17cd JA |
325 | if (unlikely((cp)->flags & IP_VS_CONN_F_NFCT)) \ |
326 | __ret = ip_vs_confirm_conntrack(skb, cp); \ | |
327 | if (__ret == NF_ACCEPT) { \ | |
328 | nf_reset(skb); \ | |
4256f1aa | 329 | skb_forward_csum(skb); \ |
f4bc17cd JA |
330 | } \ |
331 | __ret; \ | |
332 | }) | |
333 | ||
fc604767 | 334 | #define IP_VS_XMIT_NAT(pf, skb, cp, local) \ |
1da177e4 | 335 | do { \ |
cf356d69 | 336 | (skb)->ipvs_property = 1; \ |
f4bc17cd | 337 | if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \ |
cf356d69 | 338 | ip_vs_notrack(skb); \ |
f4bc17cd JA |
339 | else \ |
340 | ip_vs_update_conntrack(skb, cp, 1); \ | |
fc604767 JA |
341 | if (local) \ |
342 | return NF_ACCEPT; \ | |
ccc7911f | 343 | skb_forward_csum(skb); \ |
38cdcc9a | 344 | NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \ |
f4bc17cd JA |
345 | skb_dst(skb)->dev, dst_output); \ |
346 | } while (0) | |
347 | ||
fc604767 | 348 | #define IP_VS_XMIT(pf, skb, cp, local) \ |
f4bc17cd | 349 | do { \ |
cf356d69 | 350 | (skb)->ipvs_property = 1; \ |
f4bc17cd | 351 | if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \ |
cf356d69 | 352 | ip_vs_notrack(skb); \ |
fc604767 JA |
353 | if (local) \ |
354 | return NF_ACCEPT; \ | |
f4bc17cd JA |
355 | skb_forward_csum(skb); \ |
356 | NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \ | |
357 | skb_dst(skb)->dev, dst_output); \ | |
1da177e4 LT |
358 | } while (0) |
359 | ||
360 | ||
361 | /* | |
362 | * NULL transmitter (do nothing except return NF_ACCEPT) | |
363 | */ | |
364 | int | |
365 | ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
366 | struct ip_vs_protocol *pp) | |
367 | { | |
368 | /* we do not touch skb and do not need pskb ptr */ | |
fc604767 | 369 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); |
1da177e4 LT |
370 | } |
371 | ||
372 | ||
373 | /* | |
374 | * Bypass transmitter | |
375 | * Let packets bypass the destination when the destination is not | |
376 | * available, it may be only used in transparent cache cluster. | |
377 | */ | |
378 | int | |
379 | ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
380 | struct ip_vs_protocol *pp) | |
381 | { | |
382 | struct rtable *rt; /* Route to the other host */ | |
eddc9ec5 | 383 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 | 384 | int mtu; |
1da177e4 LT |
385 | |
386 | EnterFunction(10); | |
387 | ||
17a8f8e3 CG |
388 | if (!(rt = __ip_vs_get_out_rt(skb, NULL, iph->daddr, RT_TOS(iph->tos), |
389 | IP_VS_RT_MODE_NON_LOCAL))) | |
1da177e4 | 390 | goto tx_error_icmp; |
1da177e4 LT |
391 | |
392 | /* MTU checking */ | |
d8d1f30b | 393 | mtu = dst_mtu(&rt->dst); |
8f1b03a4 SH |
394 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF)) && |
395 | !skb_is_gso(skb)) { | |
1da177e4 LT |
396 | ip_rt_put(rt); |
397 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); | |
1e3e238e | 398 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
1da177e4 LT |
399 | goto tx_error; |
400 | } | |
401 | ||
402 | /* | |
403 | * Call ip_send_check because we are not sure it is called | |
404 | * after ip_defrag. Is copy-on-write needed? | |
405 | */ | |
406 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
407 | ip_rt_put(rt); | |
408 | return NF_STOLEN; | |
409 | } | |
eddc9ec5 | 410 | ip_send_check(ip_hdr(skb)); |
1da177e4 LT |
411 | |
412 | /* drop old route */ | |
adf30907 | 413 | skb_dst_drop(skb); |
d8d1f30b | 414 | skb_dst_set(skb, &rt->dst); |
1da177e4 LT |
415 | |
416 | /* Another hack: avoid icmp_send in ip_fragment */ | |
417 | skb->local_df = 1; | |
418 | ||
fc604767 | 419 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0); |
1da177e4 LT |
420 | |
421 | LeaveFunction(10); | |
422 | return NF_STOLEN; | |
423 | ||
424 | tx_error_icmp: | |
425 | dst_link_failure(skb); | |
426 | tx_error: | |
427 | kfree_skb(skb); | |
428 | LeaveFunction(10); | |
429 | return NF_STOLEN; | |
430 | } | |
431 | ||
b3cdd2a7 JV |
432 | #ifdef CONFIG_IP_VS_IPV6 |
433 | int | |
434 | ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
435 | struct ip_vs_protocol *pp) | |
436 | { | |
437 | struct rt6_info *rt; /* Route to the other host */ | |
438 | struct ipv6hdr *iph = ipv6_hdr(skb); | |
439 | int mtu; | |
b3cdd2a7 JV |
440 | |
441 | EnterFunction(10); | |
442 | ||
fc604767 | 443 | if (!(rt = __ip_vs_get_out_rt_v6(skb, NULL, &iph->daddr, NULL, 0, 2))) |
b3cdd2a7 | 444 | goto tx_error_icmp; |
b3cdd2a7 JV |
445 | |
446 | /* MTU checking */ | |
d8d1f30b | 447 | mtu = dst_mtu(&rt->dst); |
8f1b03a4 | 448 | if (skb->len > mtu && !skb_is_gso(skb)) { |
cb59155f JA |
449 | if (!skb->dev) { |
450 | struct net *net = dev_net(skb_dst(skb)->dev); | |
451 | ||
452 | skb->dev = net->loopback_dev; | |
453 | } | |
3ffe533c | 454 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
cb59155f | 455 | dst_release(&rt->dst); |
1e3e238e | 456 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
b3cdd2a7 JV |
457 | goto tx_error; |
458 | } | |
459 | ||
460 | /* | |
461 | * Call ip_send_check because we are not sure it is called | |
462 | * after ip_defrag. Is copy-on-write needed? | |
463 | */ | |
464 | skb = skb_share_check(skb, GFP_ATOMIC); | |
465 | if (unlikely(skb == NULL)) { | |
d8d1f30b | 466 | dst_release(&rt->dst); |
b3cdd2a7 JV |
467 | return NF_STOLEN; |
468 | } | |
469 | ||
470 | /* drop old route */ | |
adf30907 | 471 | skb_dst_drop(skb); |
d8d1f30b | 472 | skb_dst_set(skb, &rt->dst); |
b3cdd2a7 JV |
473 | |
474 | /* Another hack: avoid icmp_send in ip_fragment */ | |
475 | skb->local_df = 1; | |
476 | ||
fc604767 | 477 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0); |
b3cdd2a7 JV |
478 | |
479 | LeaveFunction(10); | |
480 | return NF_STOLEN; | |
481 | ||
482 | tx_error_icmp: | |
483 | dst_link_failure(skb); | |
484 | tx_error: | |
485 | kfree_skb(skb); | |
486 | LeaveFunction(10); | |
487 | return NF_STOLEN; | |
488 | } | |
489 | #endif | |
1da177e4 LT |
490 | |
491 | /* | |
492 | * NAT transmitter (only for outside-to-inside nat forwarding) | |
493 | * Not used for related ICMP | |
494 | */ | |
495 | int | |
496 | ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
497 | struct ip_vs_protocol *pp) | |
498 | { | |
499 | struct rtable *rt; /* Route to the other host */ | |
500 | int mtu; | |
eddc9ec5 | 501 | struct iphdr *iph = ip_hdr(skb); |
fc604767 | 502 | int local; |
1da177e4 LT |
503 | |
504 | EnterFunction(10); | |
505 | ||
506 | /* check if it is a connection of no-client-port */ | |
507 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
014d730d | 508 | __be16 _pt, *p; |
1da177e4 LT |
509 | p = skb_header_pointer(skb, iph->ihl*4, sizeof(_pt), &_pt); |
510 | if (p == NULL) | |
511 | goto tx_error; | |
512 | ip_vs_conn_fill_cport(cp, *p); | |
513 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
514 | } | |
515 | ||
fc604767 | 516 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, |
17a8f8e3 CG |
517 | RT_TOS(iph->tos), |
518 | IP_VS_RT_MODE_LOCAL | | |
519 | IP_VS_RT_MODE_NON_LOCAL | | |
520 | IP_VS_RT_MODE_RDR))) | |
1da177e4 | 521 | goto tx_error_icmp; |
fc604767 JA |
522 | local = rt->rt_flags & RTCF_LOCAL; |
523 | /* | |
524 | * Avoid duplicate tuple in reply direction for NAT traffic | |
525 | * to local address when connection is sync-ed | |
526 | */ | |
527 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
528 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
529 | enum ip_conntrack_info ctinfo; | |
530 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
531 | ||
532 | if (ct && !nf_ct_is_untracked(ct)) { | |
0d79641a JA |
533 | IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, 0, |
534 | "ip_vs_nat_xmit(): " | |
fc604767 JA |
535 | "stopping DNAT to local address"); |
536 | goto tx_error_put; | |
537 | } | |
538 | } | |
539 | #endif | |
540 | ||
541 | /* From world but DNAT to loopback address? */ | |
c7537967 DM |
542 | if (local && ipv4_is_loopback(rt->rt_dst) && |
543 | rt_is_input_route(skb_rtable(skb))) { | |
0d79641a | 544 | IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, 0, "ip_vs_nat_xmit(): " |
fc604767 JA |
545 | "stopping DNAT to loopback address"); |
546 | goto tx_error_put; | |
547 | } | |
1da177e4 LT |
548 | |
549 | /* MTU checking */ | |
d8d1f30b | 550 | mtu = dst_mtu(&rt->dst); |
8f1b03a4 SH |
551 | if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF)) && |
552 | !skb_is_gso(skb)) { | |
1da177e4 | 553 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); |
0d79641a JA |
554 | IP_VS_DBG_RL_PKT(0, AF_INET, pp, skb, 0, |
555 | "ip_vs_nat_xmit(): frag needed for"); | |
fc604767 | 556 | goto tx_error_put; |
1da177e4 LT |
557 | } |
558 | ||
559 | /* copy-on-write the packet before mangling it */ | |
af1e1cf0 | 560 | if (!skb_make_writable(skb, sizeof(struct iphdr))) |
1da177e4 LT |
561 | goto tx_error_put; |
562 | ||
d8d1f30b | 563 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) |
1da177e4 LT |
564 | goto tx_error_put; |
565 | ||
1da177e4 | 566 | /* mangle the packet */ |
3db05fea | 567 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) |
fc604767 | 568 | goto tx_error_put; |
e7ade46a | 569 | ip_hdr(skb)->daddr = cp->daddr.ip; |
eddc9ec5 | 570 | ip_send_check(ip_hdr(skb)); |
1da177e4 | 571 | |
fc604767 JA |
572 | if (!local) { |
573 | /* drop old route */ | |
574 | skb_dst_drop(skb); | |
575 | skb_dst_set(skb, &rt->dst); | |
576 | } else { | |
577 | ip_rt_put(rt); | |
578 | /* | |
579 | * Some IPv4 replies get local address from routes, | |
580 | * not from iph, so while we DNAT after routing | |
581 | * we need this second input/output route. | |
582 | */ | |
583 | if (!__ip_vs_reroute_locally(skb)) | |
584 | goto tx_error; | |
585 | } | |
586 | ||
0d79641a | 587 | IP_VS_DBG_PKT(10, AF_INET, pp, skb, 0, "After DNAT"); |
1da177e4 LT |
588 | |
589 | /* FIXME: when application helper enlarges the packet and the length | |
590 | is larger than the MTU of outgoing device, there will be still | |
591 | MTU problem. */ | |
592 | ||
593 | /* Another hack: avoid icmp_send in ip_fragment */ | |
594 | skb->local_df = 1; | |
595 | ||
fc604767 | 596 | IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local); |
1da177e4 LT |
597 | |
598 | LeaveFunction(10); | |
599 | return NF_STOLEN; | |
600 | ||
601 | tx_error_icmp: | |
602 | dst_link_failure(skb); | |
603 | tx_error: | |
1da177e4 | 604 | kfree_skb(skb); |
f4bc17cd | 605 | LeaveFunction(10); |
1da177e4 LT |
606 | return NF_STOLEN; |
607 | tx_error_put: | |
608 | ip_rt_put(rt); | |
609 | goto tx_error; | |
610 | } | |
611 | ||
b3cdd2a7 JV |
612 | #ifdef CONFIG_IP_VS_IPV6 |
613 | int | |
614 | ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
615 | struct ip_vs_protocol *pp) | |
616 | { | |
617 | struct rt6_info *rt; /* Route to the other host */ | |
618 | int mtu; | |
fc604767 | 619 | int local; |
b3cdd2a7 JV |
620 | |
621 | EnterFunction(10); | |
622 | ||
623 | /* check if it is a connection of no-client-port */ | |
624 | if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) { | |
625 | __be16 _pt, *p; | |
626 | p = skb_header_pointer(skb, sizeof(struct ipv6hdr), | |
627 | sizeof(_pt), &_pt); | |
628 | if (p == NULL) | |
629 | goto tx_error; | |
630 | ip_vs_conn_fill_cport(cp, *p); | |
631 | IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p)); | |
632 | } | |
633 | ||
fc604767 JA |
634 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, |
635 | 0, 1|2|4))) | |
b3cdd2a7 | 636 | goto tx_error_icmp; |
fc604767 JA |
637 | local = __ip_vs_is_local_route6(rt); |
638 | /* | |
639 | * Avoid duplicate tuple in reply direction for NAT traffic | |
640 | * to local address when connection is sync-ed | |
641 | */ | |
642 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
643 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
644 | enum ip_conntrack_info ctinfo; | |
645 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
646 | ||
647 | if (ct && !nf_ct_is_untracked(ct)) { | |
0d79641a | 648 | IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, 0, |
fc604767 JA |
649 | "ip_vs_nat_xmit_v6(): " |
650 | "stopping DNAT to local address"); | |
651 | goto tx_error_put; | |
652 | } | |
653 | } | |
654 | #endif | |
655 | ||
656 | /* From world but DNAT to loopback address? */ | |
657 | if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) && | |
658 | ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) { | |
0d79641a | 659 | IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, 0, |
fc604767 JA |
660 | "ip_vs_nat_xmit_v6(): " |
661 | "stopping DNAT to loopback address"); | |
662 | goto tx_error_put; | |
663 | } | |
b3cdd2a7 JV |
664 | |
665 | /* MTU checking */ | |
d8d1f30b | 666 | mtu = dst_mtu(&rt->dst); |
8f1b03a4 | 667 | if (skb->len > mtu && !skb_is_gso(skb)) { |
cb59155f JA |
668 | if (!skb->dev) { |
669 | struct net *net = dev_net(skb_dst(skb)->dev); | |
670 | ||
671 | skb->dev = net->loopback_dev; | |
672 | } | |
3ffe533c | 673 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
0d79641a | 674 | IP_VS_DBG_RL_PKT(0, AF_INET6, pp, skb, 0, |
b3cdd2a7 | 675 | "ip_vs_nat_xmit_v6(): frag needed for"); |
fc604767 | 676 | goto tx_error_put; |
b3cdd2a7 JV |
677 | } |
678 | ||
679 | /* copy-on-write the packet before mangling it */ | |
680 | if (!skb_make_writable(skb, sizeof(struct ipv6hdr))) | |
681 | goto tx_error_put; | |
682 | ||
d8d1f30b | 683 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) |
b3cdd2a7 JV |
684 | goto tx_error_put; |
685 | ||
b3cdd2a7 JV |
686 | /* mangle the packet */ |
687 | if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp)) | |
688 | goto tx_error; | |
fc604767 JA |
689 | ipv6_addr_copy(&ipv6_hdr(skb)->daddr, &cp->daddr.in6); |
690 | ||
691 | if (!local || !skb->dev) { | |
692 | /* drop the old route when skb is not shared */ | |
693 | skb_dst_drop(skb); | |
694 | skb_dst_set(skb, &rt->dst); | |
695 | } else { | |
696 | /* destined to loopback, do we need to change route? */ | |
697 | dst_release(&rt->dst); | |
698 | } | |
b3cdd2a7 | 699 | |
0d79641a | 700 | IP_VS_DBG_PKT(10, AF_INET6, pp, skb, 0, "After DNAT"); |
b3cdd2a7 JV |
701 | |
702 | /* FIXME: when application helper enlarges the packet and the length | |
703 | is larger than the MTU of outgoing device, there will be still | |
704 | MTU problem. */ | |
705 | ||
706 | /* Another hack: avoid icmp_send in ip_fragment */ | |
707 | skb->local_df = 1; | |
708 | ||
fc604767 | 709 | IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local); |
b3cdd2a7 JV |
710 | |
711 | LeaveFunction(10); | |
712 | return NF_STOLEN; | |
713 | ||
714 | tx_error_icmp: | |
715 | dst_link_failure(skb); | |
716 | tx_error: | |
717 | LeaveFunction(10); | |
718 | kfree_skb(skb); | |
719 | return NF_STOLEN; | |
720 | tx_error_put: | |
d8d1f30b | 721 | dst_release(&rt->dst); |
b3cdd2a7 JV |
722 | goto tx_error; |
723 | } | |
724 | #endif | |
725 | ||
1da177e4 LT |
726 | |
727 | /* | |
728 | * IP Tunneling transmitter | |
729 | * | |
730 | * This function encapsulates the packet in a new IP packet, its | |
731 | * destination will be set to cp->daddr. Most code of this function | |
732 | * is taken from ipip.c. | |
733 | * | |
734 | * It is used in VS/TUN cluster. The load balancer selects a real | |
735 | * server from a cluster based on a scheduling algorithm, | |
736 | * encapsulates the request packet and forwards it to the selected | |
737 | * server. For example, all real servers are configured with | |
738 | * "ifconfig tunl0 <Virtual IP Address> up". When the server receives | |
739 | * the encapsulated packet, it will decapsulate the packet, processe | |
740 | * the request and return the response packets directly to the client | |
741 | * without passing the load balancer. This can greatly increase the | |
742 | * scalability of virtual server. | |
743 | * | |
744 | * Used for ANY protocol | |
745 | */ | |
746 | int | |
747 | ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
748 | struct ip_vs_protocol *pp) | |
749 | { | |
750 | struct rtable *rt; /* Route to the other host */ | |
751 | struct net_device *tdev; /* Device to other host */ | |
eddc9ec5 | 752 | struct iphdr *old_iph = ip_hdr(skb); |
1da177e4 | 753 | u8 tos = old_iph->tos; |
76ab608d | 754 | __be16 df = old_iph->frag_off; |
1da177e4 | 755 | struct iphdr *iph; /* Our new IP header */ |
c2636b4d | 756 | unsigned int max_headroom; /* The extra header space needed */ |
1da177e4 | 757 | int mtu; |
f4bc17cd | 758 | int ret; |
1da177e4 LT |
759 | |
760 | EnterFunction(10); | |
761 | ||
fc604767 | 762 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, |
17a8f8e3 CG |
763 | RT_TOS(tos), IP_VS_RT_MODE_LOCAL | |
764 | IP_VS_RT_MODE_NON_LOCAL))) | |
1da177e4 | 765 | goto tx_error_icmp; |
fc604767 JA |
766 | if (rt->rt_flags & RTCF_LOCAL) { |
767 | ip_rt_put(rt); | |
768 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); | |
769 | } | |
1da177e4 | 770 | |
d8d1f30b | 771 | tdev = rt->dst.dev; |
1da177e4 | 772 | |
d8d1f30b | 773 | mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr); |
1da177e4 | 774 | if (mtu < 68) { |
1e3e238e | 775 | IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__); |
fc604767 | 776 | goto tx_error_put; |
1da177e4 | 777 | } |
adf30907 ED |
778 | if (skb_dst(skb)) |
779 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
1da177e4 | 780 | |
4412ec49 | 781 | df |= (old_iph->frag_off & htons(IP_DF)); |
1da177e4 | 782 | |
8f1b03a4 SH |
783 | if ((old_iph->frag_off & htons(IP_DF) && |
784 | mtu < ntohs(old_iph->tot_len) && !skb_is_gso(skb))) { | |
1da177e4 | 785 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); |
1e3e238e | 786 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
fc604767 | 787 | goto tx_error_put; |
1da177e4 LT |
788 | } |
789 | ||
790 | /* | |
791 | * Okay, now see if we can stuff it in the buffer as-is. | |
792 | */ | |
793 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr); | |
794 | ||
795 | if (skb_headroom(skb) < max_headroom | |
796 | || skb_cloned(skb) || skb_shared(skb)) { | |
797 | struct sk_buff *new_skb = | |
798 | skb_realloc_headroom(skb, max_headroom); | |
799 | if (!new_skb) { | |
800 | ip_rt_put(rt); | |
801 | kfree_skb(skb); | |
1e3e238e | 802 | IP_VS_ERR_RL("%s(): no memory\n", __func__); |
1da177e4 LT |
803 | return NF_STOLEN; |
804 | } | |
805 | kfree_skb(skb); | |
806 | skb = new_skb; | |
eddc9ec5 | 807 | old_iph = ip_hdr(skb); |
1da177e4 LT |
808 | } |
809 | ||
714f095f | 810 | skb->transport_header = skb->network_header; |
1da177e4 LT |
811 | |
812 | /* fix old IP header checksum */ | |
813 | ip_send_check(old_iph); | |
814 | ||
e2d1bca7 ACM |
815 | skb_push(skb, sizeof(struct iphdr)); |
816 | skb_reset_network_header(skb); | |
1da177e4 LT |
817 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); |
818 | ||
819 | /* drop old route */ | |
adf30907 | 820 | skb_dst_drop(skb); |
d8d1f30b | 821 | skb_dst_set(skb, &rt->dst); |
1da177e4 LT |
822 | |
823 | /* | |
824 | * Push down and install the IPIP header. | |
825 | */ | |
eddc9ec5 | 826 | iph = ip_hdr(skb); |
1da177e4 LT |
827 | iph->version = 4; |
828 | iph->ihl = sizeof(struct iphdr)>>2; | |
829 | iph->frag_off = df; | |
830 | iph->protocol = IPPROTO_IPIP; | |
831 | iph->tos = tos; | |
832 | iph->daddr = rt->rt_dst; | |
833 | iph->saddr = rt->rt_src; | |
834 | iph->ttl = old_iph->ttl; | |
d8d1f30b | 835 | ip_select_ident(iph, &rt->dst, NULL); |
1da177e4 LT |
836 | |
837 | /* Another hack: avoid icmp_send in ip_fragment */ | |
838 | skb->local_df = 1; | |
839 | ||
f4bc17cd JA |
840 | ret = IP_VS_XMIT_TUNNEL(skb, cp); |
841 | if (ret == NF_ACCEPT) | |
842 | ip_local_out(skb); | |
843 | else if (ret == NF_DROP) | |
844 | kfree_skb(skb); | |
1da177e4 LT |
845 | |
846 | LeaveFunction(10); | |
847 | ||
848 | return NF_STOLEN; | |
849 | ||
850 | tx_error_icmp: | |
851 | dst_link_failure(skb); | |
852 | tx_error: | |
853 | kfree_skb(skb); | |
854 | LeaveFunction(10); | |
855 | return NF_STOLEN; | |
fc604767 JA |
856 | tx_error_put: |
857 | ip_rt_put(rt); | |
858 | goto tx_error; | |
1da177e4 LT |
859 | } |
860 | ||
b3cdd2a7 JV |
861 | #ifdef CONFIG_IP_VS_IPV6 |
862 | int | |
863 | ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
864 | struct ip_vs_protocol *pp) | |
865 | { | |
866 | struct rt6_info *rt; /* Route to the other host */ | |
714f095f | 867 | struct in6_addr saddr; /* Source for tunnel */ |
b3cdd2a7 JV |
868 | struct net_device *tdev; /* Device to other host */ |
869 | struct ipv6hdr *old_iph = ipv6_hdr(skb); | |
b3cdd2a7 JV |
870 | struct ipv6hdr *iph; /* Our new IP header */ |
871 | unsigned int max_headroom; /* The extra header space needed */ | |
872 | int mtu; | |
f4bc17cd | 873 | int ret; |
b3cdd2a7 JV |
874 | |
875 | EnterFunction(10); | |
876 | ||
fc604767 JA |
877 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, |
878 | &saddr, 1, 1|2))) | |
b3cdd2a7 | 879 | goto tx_error_icmp; |
fc604767 JA |
880 | if (__ip_vs_is_local_route6(rt)) { |
881 | dst_release(&rt->dst); | |
882 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1); | |
883 | } | |
b3cdd2a7 | 884 | |
d8d1f30b | 885 | tdev = rt->dst.dev; |
b3cdd2a7 | 886 | |
d8d1f30b | 887 | mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr); |
714f095f | 888 | if (mtu < IPV6_MIN_MTU) { |
714f095f HS |
889 | IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__, |
890 | IPV6_MIN_MTU); | |
fc604767 | 891 | goto tx_error_put; |
b3cdd2a7 | 892 | } |
adf30907 ED |
893 | if (skb_dst(skb)) |
894 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); | |
b3cdd2a7 | 895 | |
8f1b03a4 SH |
896 | if (mtu < ntohs(old_iph->payload_len) + sizeof(struct ipv6hdr) && |
897 | !skb_is_gso(skb)) { | |
cb59155f JA |
898 | if (!skb->dev) { |
899 | struct net *net = dev_net(skb_dst(skb)->dev); | |
900 | ||
901 | skb->dev = net->loopback_dev; | |
902 | } | |
3ffe533c | 903 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
1e3e238e | 904 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
fc604767 | 905 | goto tx_error_put; |
b3cdd2a7 JV |
906 | } |
907 | ||
908 | /* | |
909 | * Okay, now see if we can stuff it in the buffer as-is. | |
910 | */ | |
911 | max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr); | |
912 | ||
913 | if (skb_headroom(skb) < max_headroom | |
914 | || skb_cloned(skb) || skb_shared(skb)) { | |
915 | struct sk_buff *new_skb = | |
916 | skb_realloc_headroom(skb, max_headroom); | |
917 | if (!new_skb) { | |
d8d1f30b | 918 | dst_release(&rt->dst); |
b3cdd2a7 | 919 | kfree_skb(skb); |
1e3e238e | 920 | IP_VS_ERR_RL("%s(): no memory\n", __func__); |
b3cdd2a7 JV |
921 | return NF_STOLEN; |
922 | } | |
923 | kfree_skb(skb); | |
924 | skb = new_skb; | |
925 | old_iph = ipv6_hdr(skb); | |
926 | } | |
927 | ||
714f095f | 928 | skb->transport_header = skb->network_header; |
b3cdd2a7 JV |
929 | |
930 | skb_push(skb, sizeof(struct ipv6hdr)); | |
931 | skb_reset_network_header(skb); | |
932 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); | |
933 | ||
934 | /* drop old route */ | |
adf30907 | 935 | skb_dst_drop(skb); |
d8d1f30b | 936 | skb_dst_set(skb, &rt->dst); |
b3cdd2a7 JV |
937 | |
938 | /* | |
939 | * Push down and install the IPIP header. | |
940 | */ | |
941 | iph = ipv6_hdr(skb); | |
942 | iph->version = 6; | |
943 | iph->nexthdr = IPPROTO_IPV6; | |
b7b45f47 HH |
944 | iph->payload_len = old_iph->payload_len; |
945 | be16_add_cpu(&iph->payload_len, sizeof(*old_iph)); | |
b3cdd2a7 JV |
946 | iph->priority = old_iph->priority; |
947 | memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl)); | |
714f095f HS |
948 | ipv6_addr_copy(&iph->daddr, &cp->daddr.in6); |
949 | ipv6_addr_copy(&iph->saddr, &saddr); | |
b3cdd2a7 JV |
950 | iph->hop_limit = old_iph->hop_limit; |
951 | ||
952 | /* Another hack: avoid icmp_send in ip_fragment */ | |
953 | skb->local_df = 1; | |
954 | ||
f4bc17cd JA |
955 | ret = IP_VS_XMIT_TUNNEL(skb, cp); |
956 | if (ret == NF_ACCEPT) | |
957 | ip6_local_out(skb); | |
958 | else if (ret == NF_DROP) | |
959 | kfree_skb(skb); | |
b3cdd2a7 JV |
960 | |
961 | LeaveFunction(10); | |
962 | ||
963 | return NF_STOLEN; | |
964 | ||
965 | tx_error_icmp: | |
966 | dst_link_failure(skb); | |
967 | tx_error: | |
968 | kfree_skb(skb); | |
969 | LeaveFunction(10); | |
970 | return NF_STOLEN; | |
fc604767 JA |
971 | tx_error_put: |
972 | dst_release(&rt->dst); | |
973 | goto tx_error; | |
b3cdd2a7 JV |
974 | } |
975 | #endif | |
976 | ||
1da177e4 LT |
977 | |
978 | /* | |
979 | * Direct Routing transmitter | |
980 | * Used for ANY protocol | |
981 | */ | |
982 | int | |
983 | ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
984 | struct ip_vs_protocol *pp) | |
985 | { | |
986 | struct rtable *rt; /* Route to the other host */ | |
eddc9ec5 | 987 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
988 | int mtu; |
989 | ||
990 | EnterFunction(10); | |
991 | ||
fc604767 | 992 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, |
17a8f8e3 CG |
993 | RT_TOS(iph->tos), |
994 | IP_VS_RT_MODE_LOCAL | | |
995 | IP_VS_RT_MODE_NON_LOCAL))) | |
1da177e4 | 996 | goto tx_error_icmp; |
fc604767 JA |
997 | if (rt->rt_flags & RTCF_LOCAL) { |
998 | ip_rt_put(rt); | |
999 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1); | |
1000 | } | |
1da177e4 LT |
1001 | |
1002 | /* MTU checking */ | |
d8d1f30b | 1003 | mtu = dst_mtu(&rt->dst); |
8f1b03a4 SH |
1004 | if ((iph->frag_off & htons(IP_DF)) && skb->len > mtu && |
1005 | !skb_is_gso(skb)) { | |
1da177e4 LT |
1006 | icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu)); |
1007 | ip_rt_put(rt); | |
1e3e238e | 1008 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
1da177e4 LT |
1009 | goto tx_error; |
1010 | } | |
1011 | ||
1012 | /* | |
1013 | * Call ip_send_check because we are not sure it is called | |
1014 | * after ip_defrag. Is copy-on-write needed? | |
1015 | */ | |
1016 | if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) { | |
1017 | ip_rt_put(rt); | |
1018 | return NF_STOLEN; | |
1019 | } | |
eddc9ec5 | 1020 | ip_send_check(ip_hdr(skb)); |
1da177e4 LT |
1021 | |
1022 | /* drop old route */ | |
adf30907 | 1023 | skb_dst_drop(skb); |
d8d1f30b | 1024 | skb_dst_set(skb, &rt->dst); |
1da177e4 LT |
1025 | |
1026 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1027 | skb->local_df = 1; | |
1028 | ||
fc604767 | 1029 | IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0); |
1da177e4 LT |
1030 | |
1031 | LeaveFunction(10); | |
1032 | return NF_STOLEN; | |
1033 | ||
1034 | tx_error_icmp: | |
1035 | dst_link_failure(skb); | |
1036 | tx_error: | |
1037 | kfree_skb(skb); | |
1038 | LeaveFunction(10); | |
1039 | return NF_STOLEN; | |
1040 | } | |
1041 | ||
b3cdd2a7 JV |
1042 | #ifdef CONFIG_IP_VS_IPV6 |
1043 | int | |
1044 | ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1045 | struct ip_vs_protocol *pp) | |
1046 | { | |
1047 | struct rt6_info *rt; /* Route to the other host */ | |
1048 | int mtu; | |
1049 | ||
1050 | EnterFunction(10); | |
1051 | ||
fc604767 JA |
1052 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, |
1053 | 0, 1|2))) | |
b3cdd2a7 | 1054 | goto tx_error_icmp; |
fc604767 JA |
1055 | if (__ip_vs_is_local_route6(rt)) { |
1056 | dst_release(&rt->dst); | |
1057 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1); | |
1058 | } | |
b3cdd2a7 JV |
1059 | |
1060 | /* MTU checking */ | |
d8d1f30b | 1061 | mtu = dst_mtu(&rt->dst); |
b3cdd2a7 | 1062 | if (skb->len > mtu) { |
cb59155f JA |
1063 | if (!skb->dev) { |
1064 | struct net *net = dev_net(skb_dst(skb)->dev); | |
1065 | ||
1066 | skb->dev = net->loopback_dev; | |
1067 | } | |
3ffe533c | 1068 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
d8d1f30b | 1069 | dst_release(&rt->dst); |
1e3e238e | 1070 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
b3cdd2a7 JV |
1071 | goto tx_error; |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * Call ip_send_check because we are not sure it is called | |
1076 | * after ip_defrag. Is copy-on-write needed? | |
1077 | */ | |
1078 | skb = skb_share_check(skb, GFP_ATOMIC); | |
1079 | if (unlikely(skb == NULL)) { | |
d8d1f30b | 1080 | dst_release(&rt->dst); |
b3cdd2a7 JV |
1081 | return NF_STOLEN; |
1082 | } | |
1083 | ||
1084 | /* drop old route */ | |
adf30907 | 1085 | skb_dst_drop(skb); |
d8d1f30b | 1086 | skb_dst_set(skb, &rt->dst); |
b3cdd2a7 JV |
1087 | |
1088 | /* Another hack: avoid icmp_send in ip_fragment */ | |
1089 | skb->local_df = 1; | |
1090 | ||
fc604767 | 1091 | IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0); |
b3cdd2a7 JV |
1092 | |
1093 | LeaveFunction(10); | |
1094 | return NF_STOLEN; | |
1095 | ||
1096 | tx_error_icmp: | |
1097 | dst_link_failure(skb); | |
1098 | tx_error: | |
1099 | kfree_skb(skb); | |
1100 | LeaveFunction(10); | |
1101 | return NF_STOLEN; | |
1102 | } | |
1103 | #endif | |
1104 | ||
1da177e4 LT |
1105 | |
1106 | /* | |
1107 | * ICMP packet transmitter | |
1108 | * called by the ip_vs_in_icmp | |
1109 | */ | |
1110 | int | |
1111 | ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1112 | struct ip_vs_protocol *pp, int offset) | |
1113 | { | |
1114 | struct rtable *rt; /* Route to the other host */ | |
1115 | int mtu; | |
1116 | int rc; | |
fc604767 | 1117 | int local; |
1da177e4 LT |
1118 | |
1119 | EnterFunction(10); | |
1120 | ||
1121 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
1122 | forwarded directly here, because there is no need to | |
1123 | translate address/port back */ | |
1124 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
1125 | if (cp->packet_xmit) | |
1126 | rc = cp->packet_xmit(skb, cp, pp); | |
1127 | else | |
1128 | rc = NF_ACCEPT; | |
1129 | /* do not touch skb anymore */ | |
1130 | atomic_inc(&cp->in_pkts); | |
1da177e4 LT |
1131 | goto out; |
1132 | } | |
1133 | ||
1134 | /* | |
1135 | * mangle and send the packet here (only for VS/NAT) | |
1136 | */ | |
1137 | ||
fc604767 | 1138 | if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip, |
17a8f8e3 CG |
1139 | RT_TOS(ip_hdr(skb)->tos), |
1140 | IP_VS_RT_MODE_LOCAL | | |
1141 | IP_VS_RT_MODE_NON_LOCAL | | |
1142 | IP_VS_RT_MODE_RDR))) | |
1da177e4 | 1143 | goto tx_error_icmp; |
fc604767 JA |
1144 | local = rt->rt_flags & RTCF_LOCAL; |
1145 | ||
1146 | /* | |
1147 | * Avoid duplicate tuple in reply direction for NAT traffic | |
1148 | * to local address when connection is sync-ed | |
1149 | */ | |
1150 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
1151 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
1152 | enum ip_conntrack_info ctinfo; | |
1153 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
1154 | ||
1155 | if (ct && !nf_ct_is_untracked(ct)) { | |
1156 | IP_VS_DBG(10, "%s(): " | |
1157 | "stopping DNAT to local address %pI4\n", | |
1158 | __func__, &cp->daddr.ip); | |
1159 | goto tx_error_put; | |
1160 | } | |
1161 | } | |
1162 | #endif | |
1163 | ||
1164 | /* From world but DNAT to loopback address? */ | |
c7537967 DM |
1165 | if (local && ipv4_is_loopback(rt->rt_dst) && |
1166 | rt_is_input_route(skb_rtable(skb))) { | |
fc604767 JA |
1167 | IP_VS_DBG(1, "%s(): " |
1168 | "stopping DNAT to loopback %pI4\n", | |
1169 | __func__, &cp->daddr.ip); | |
1170 | goto tx_error_put; | |
1171 | } | |
1da177e4 LT |
1172 | |
1173 | /* MTU checking */ | |
d8d1f30b | 1174 | mtu = dst_mtu(&rt->dst); |
8f1b03a4 SH |
1175 | if ((skb->len > mtu) && (ip_hdr(skb)->frag_off & htons(IP_DF)) && |
1176 | !skb_is_gso(skb)) { | |
1da177e4 | 1177 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); |
1e3e238e | 1178 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
fc604767 | 1179 | goto tx_error_put; |
1da177e4 LT |
1180 | } |
1181 | ||
1182 | /* copy-on-write the packet before mangling it */ | |
af1e1cf0 | 1183 | if (!skb_make_writable(skb, offset)) |
1da177e4 LT |
1184 | goto tx_error_put; |
1185 | ||
d8d1f30b | 1186 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) |
1da177e4 LT |
1187 | goto tx_error_put; |
1188 | ||
1da177e4 LT |
1189 | ip_vs_nat_icmp(skb, pp, cp, 0); |
1190 | ||
fc604767 JA |
1191 | if (!local) { |
1192 | /* drop the old route when skb is not shared */ | |
1193 | skb_dst_drop(skb); | |
1194 | skb_dst_set(skb, &rt->dst); | |
1195 | } else { | |
1196 | ip_rt_put(rt); | |
1197 | /* | |
1198 | * Some IPv4 replies get local address from routes, | |
1199 | * not from iph, so while we DNAT after routing | |
1200 | * we need this second input/output route. | |
1201 | */ | |
1202 | if (!__ip_vs_reroute_locally(skb)) | |
1203 | goto tx_error; | |
1204 | } | |
1205 | ||
1da177e4 LT |
1206 | /* Another hack: avoid icmp_send in ip_fragment */ |
1207 | skb->local_df = 1; | |
1208 | ||
fc604767 | 1209 | IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local); |
1da177e4 LT |
1210 | |
1211 | rc = NF_STOLEN; | |
1212 | goto out; | |
1213 | ||
1214 | tx_error_icmp: | |
1215 | dst_link_failure(skb); | |
1216 | tx_error: | |
1217 | dev_kfree_skb(skb); | |
1218 | rc = NF_STOLEN; | |
1219 | out: | |
1220 | LeaveFunction(10); | |
1221 | return rc; | |
1222 | tx_error_put: | |
1223 | ip_rt_put(rt); | |
1224 | goto tx_error; | |
1225 | } | |
b3cdd2a7 JV |
1226 | |
1227 | #ifdef CONFIG_IP_VS_IPV6 | |
1228 | int | |
1229 | ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1230 | struct ip_vs_protocol *pp, int offset) | |
1231 | { | |
1232 | struct rt6_info *rt; /* Route to the other host */ | |
1233 | int mtu; | |
1234 | int rc; | |
fc604767 | 1235 | int local; |
b3cdd2a7 JV |
1236 | |
1237 | EnterFunction(10); | |
1238 | ||
1239 | /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be | |
1240 | forwarded directly here, because there is no need to | |
1241 | translate address/port back */ | |
1242 | if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) { | |
1243 | if (cp->packet_xmit) | |
1244 | rc = cp->packet_xmit(skb, cp, pp); | |
1245 | else | |
1246 | rc = NF_ACCEPT; | |
1247 | /* do not touch skb anymore */ | |
1248 | atomic_inc(&cp->in_pkts); | |
1249 | goto out; | |
1250 | } | |
1251 | ||
1252 | /* | |
1253 | * mangle and send the packet here (only for VS/NAT) | |
1254 | */ | |
1255 | ||
fc604767 JA |
1256 | if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL, |
1257 | 0, 1|2|4))) | |
b3cdd2a7 JV |
1258 | goto tx_error_icmp; |
1259 | ||
fc604767 JA |
1260 | local = __ip_vs_is_local_route6(rt); |
1261 | /* | |
1262 | * Avoid duplicate tuple in reply direction for NAT traffic | |
1263 | * to local address when connection is sync-ed | |
1264 | */ | |
1265 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
1266 | if (cp->flags & IP_VS_CONN_F_SYNC && local) { | |
1267 | enum ip_conntrack_info ctinfo; | |
1268 | struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo); | |
1269 | ||
1270 | if (ct && !nf_ct_is_untracked(ct)) { | |
1271 | IP_VS_DBG(10, "%s(): " | |
1272 | "stopping DNAT to local address %pI6\n", | |
1273 | __func__, &cp->daddr.in6); | |
1274 | goto tx_error_put; | |
1275 | } | |
1276 | } | |
1277 | #endif | |
1278 | ||
1279 | /* From world but DNAT to loopback address? */ | |
1280 | if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) && | |
1281 | ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) { | |
1282 | IP_VS_DBG(1, "%s(): " | |
1283 | "stopping DNAT to loopback %pI6\n", | |
1284 | __func__, &cp->daddr.in6); | |
1285 | goto tx_error_put; | |
1286 | } | |
1287 | ||
b3cdd2a7 | 1288 | /* MTU checking */ |
d8d1f30b | 1289 | mtu = dst_mtu(&rt->dst); |
8f1b03a4 | 1290 | if (skb->len > mtu && !skb_is_gso(skb)) { |
cb59155f JA |
1291 | if (!skb->dev) { |
1292 | struct net *net = dev_net(skb_dst(skb)->dev); | |
1293 | ||
1294 | skb->dev = net->loopback_dev; | |
1295 | } | |
3ffe533c | 1296 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
1e3e238e | 1297 | IP_VS_DBG_RL("%s(): frag needed\n", __func__); |
fc604767 | 1298 | goto tx_error_put; |
b3cdd2a7 JV |
1299 | } |
1300 | ||
1301 | /* copy-on-write the packet before mangling it */ | |
1302 | if (!skb_make_writable(skb, offset)) | |
1303 | goto tx_error_put; | |
1304 | ||
d8d1f30b | 1305 | if (skb_cow(skb, rt->dst.dev->hard_header_len)) |
b3cdd2a7 JV |
1306 | goto tx_error_put; |
1307 | ||
b3cdd2a7 JV |
1308 | ip_vs_nat_icmp_v6(skb, pp, cp, 0); |
1309 | ||
fc604767 JA |
1310 | if (!local || !skb->dev) { |
1311 | /* drop the old route when skb is not shared */ | |
1312 | skb_dst_drop(skb); | |
1313 | skb_dst_set(skb, &rt->dst); | |
1314 | } else { | |
1315 | /* destined to loopback, do we need to change route? */ | |
1316 | dst_release(&rt->dst); | |
1317 | } | |
1318 | ||
b3cdd2a7 JV |
1319 | /* Another hack: avoid icmp_send in ip_fragment */ |
1320 | skb->local_df = 1; | |
1321 | ||
fc604767 | 1322 | IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local); |
b3cdd2a7 JV |
1323 | |
1324 | rc = NF_STOLEN; | |
1325 | goto out; | |
1326 | ||
1327 | tx_error_icmp: | |
1328 | dst_link_failure(skb); | |
1329 | tx_error: | |
1330 | dev_kfree_skb(skb); | |
1331 | rc = NF_STOLEN; | |
1332 | out: | |
1333 | LeaveFunction(10); | |
1334 | return rc; | |
1335 | tx_error_put: | |
d8d1f30b | 1336 | dst_release(&rt->dst); |
b3cdd2a7 JV |
1337 | goto tx_error; |
1338 | } | |
1339 | #endif |