Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IPVS An implementation of the IP virtual server support for the | |
3 | * LINUX operating system. IPVS is now implemented as a module | |
4 | * over the Netfilter framework. IPVS can be used to build a | |
5 | * high-performance and highly available server based on a | |
6 | * cluster of servers. | |
7 | * | |
1da177e4 LT |
8 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> |
9 | * Peter Kese <peter.kese@ijs.si> | |
10 | * Julian Anastasov <ja@ssi.bg> | |
11 | * | |
12 | * This program is free software; you can redistribute it and/or | |
13 | * modify it under the terms of the GNU General Public License | |
14 | * as published by the Free Software Foundation; either version | |
15 | * 2 of the License, or (at your option) any later version. | |
16 | * | |
17 | * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese, | |
18 | * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms | |
19 | * and others. | |
20 | * | |
21 | * Changes: | |
22 | * Paul `Rusty' Russell properly handle non-linear skbs | |
6869c4d8 | 23 | * Harald Welte don't use nfcache |
1da177e4 LT |
24 | * |
25 | */ | |
26 | ||
27 | #include <linux/module.h> | |
28 | #include <linux/kernel.h> | |
29 | #include <linux/ip.h> | |
30 | #include <linux/tcp.h> | |
31 | #include <linux/icmp.h> | |
32 | ||
33 | #include <net/ip.h> | |
34 | #include <net/tcp.h> | |
35 | #include <net/udp.h> | |
36 | #include <net/icmp.h> /* for icmp_send */ | |
37 | #include <net/route.h> | |
38 | ||
39 | #include <linux/netfilter.h> | |
40 | #include <linux/netfilter_ipv4.h> | |
41 | ||
2a3b791e JV |
42 | #ifdef CONFIG_IP_VS_IPV6 |
43 | #include <net/ipv6.h> | |
44 | #include <linux/netfilter_ipv6.h> | |
45 | #endif | |
46 | ||
1da177e4 LT |
47 | #include <net/ip_vs.h> |
48 | ||
49 | ||
50 | EXPORT_SYMBOL(register_ip_vs_scheduler); | |
51 | EXPORT_SYMBOL(unregister_ip_vs_scheduler); | |
52 | EXPORT_SYMBOL(ip_vs_skb_replace); | |
53 | EXPORT_SYMBOL(ip_vs_proto_name); | |
54 | EXPORT_SYMBOL(ip_vs_conn_new); | |
55 | EXPORT_SYMBOL(ip_vs_conn_in_get); | |
56 | EXPORT_SYMBOL(ip_vs_conn_out_get); | |
57 | #ifdef CONFIG_IP_VS_PROTO_TCP | |
58 | EXPORT_SYMBOL(ip_vs_tcp_conn_listen); | |
59 | #endif | |
60 | EXPORT_SYMBOL(ip_vs_conn_put); | |
61 | #ifdef CONFIG_IP_VS_DEBUG | |
62 | EXPORT_SYMBOL(ip_vs_get_debug_level); | |
63 | #endif | |
1da177e4 LT |
64 | |
65 | ||
66 | /* ID used in ICMP lookups */ | |
67 | #define icmp_id(icmph) (((icmph)->un).echo.id) | |
2a3b791e | 68 | #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier) |
1da177e4 LT |
69 | |
70 | const char *ip_vs_proto_name(unsigned proto) | |
71 | { | |
72 | static char buf[20]; | |
73 | ||
74 | switch (proto) { | |
75 | case IPPROTO_IP: | |
76 | return "IP"; | |
77 | case IPPROTO_UDP: | |
78 | return "UDP"; | |
79 | case IPPROTO_TCP: | |
80 | return "TCP"; | |
81 | case IPPROTO_ICMP: | |
82 | return "ICMP"; | |
2a3b791e JV |
83 | #ifdef CONFIG_IP_VS_IPV6 |
84 | case IPPROTO_ICMPV6: | |
85 | return "ICMPv6"; | |
86 | #endif | |
1da177e4 LT |
87 | default: |
88 | sprintf(buf, "IP_%d", proto); | |
89 | return buf; | |
90 | } | |
91 | } | |
92 | ||
93 | void ip_vs_init_hash_table(struct list_head *table, int rows) | |
94 | { | |
95 | while (--rows >= 0) | |
96 | INIT_LIST_HEAD(&table[rows]); | |
97 | } | |
98 | ||
99 | static inline void | |
100 | ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb) | |
101 | { | |
102 | struct ip_vs_dest *dest = cp->dest; | |
103 | if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) { | |
104 | spin_lock(&dest->stats.lock); | |
105 | dest->stats.inpkts++; | |
106 | dest->stats.inbytes += skb->len; | |
107 | spin_unlock(&dest->stats.lock); | |
108 | ||
109 | spin_lock(&dest->svc->stats.lock); | |
110 | dest->svc->stats.inpkts++; | |
111 | dest->svc->stats.inbytes += skb->len; | |
112 | spin_unlock(&dest->svc->stats.lock); | |
113 | ||
114 | spin_lock(&ip_vs_stats.lock); | |
115 | ip_vs_stats.inpkts++; | |
116 | ip_vs_stats.inbytes += skb->len; | |
117 | spin_unlock(&ip_vs_stats.lock); | |
118 | } | |
119 | } | |
120 | ||
121 | ||
122 | static inline void | |
123 | ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb) | |
124 | { | |
125 | struct ip_vs_dest *dest = cp->dest; | |
126 | if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) { | |
127 | spin_lock(&dest->stats.lock); | |
128 | dest->stats.outpkts++; | |
129 | dest->stats.outbytes += skb->len; | |
130 | spin_unlock(&dest->stats.lock); | |
131 | ||
132 | spin_lock(&dest->svc->stats.lock); | |
133 | dest->svc->stats.outpkts++; | |
134 | dest->svc->stats.outbytes += skb->len; | |
135 | spin_unlock(&dest->svc->stats.lock); | |
136 | ||
137 | spin_lock(&ip_vs_stats.lock); | |
138 | ip_vs_stats.outpkts++; | |
139 | ip_vs_stats.outbytes += skb->len; | |
140 | spin_unlock(&ip_vs_stats.lock); | |
141 | } | |
142 | } | |
143 | ||
144 | ||
145 | static inline void | |
146 | ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc) | |
147 | { | |
148 | spin_lock(&cp->dest->stats.lock); | |
149 | cp->dest->stats.conns++; | |
150 | spin_unlock(&cp->dest->stats.lock); | |
151 | ||
152 | spin_lock(&svc->stats.lock); | |
153 | svc->stats.conns++; | |
154 | spin_unlock(&svc->stats.lock); | |
155 | ||
156 | spin_lock(&ip_vs_stats.lock); | |
157 | ip_vs_stats.conns++; | |
158 | spin_unlock(&ip_vs_stats.lock); | |
159 | } | |
160 | ||
161 | ||
162 | static inline int | |
163 | ip_vs_set_state(struct ip_vs_conn *cp, int direction, | |
164 | const struct sk_buff *skb, | |
165 | struct ip_vs_protocol *pp) | |
166 | { | |
167 | if (unlikely(!pp->state_transition)) | |
168 | return 0; | |
169 | return pp->state_transition(cp, direction, skb, pp); | |
170 | } | |
171 | ||
172 | ||
1da177e4 LT |
173 | /* |
174 | * IPVS persistent scheduling function | |
175 | * It creates a connection entry according to its template if exists, | |
176 | * or selects a server and creates a connection entry plus a template. | |
177 | * Locking: we are svc user (svc->refcnt), so we hold all dests too | |
178 | * Protocols supported: TCP, UDP | |
179 | */ | |
180 | static struct ip_vs_conn * | |
181 | ip_vs_sched_persist(struct ip_vs_service *svc, | |
182 | const struct sk_buff *skb, | |
014d730d | 183 | __be16 ports[2]) |
1da177e4 LT |
184 | { |
185 | struct ip_vs_conn *cp = NULL; | |
28364a59 | 186 | struct ip_vs_iphdr iph; |
1da177e4 LT |
187 | struct ip_vs_dest *dest; |
188 | struct ip_vs_conn *ct; | |
cd17f9ed | 189 | __be16 dport; /* destination port to forward */ |
28364a59 JV |
190 | union nf_inet_addr snet; /* source network of the client, |
191 | after masking */ | |
cd17f9ed JV |
192 | |
193 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); | |
1da177e4 LT |
194 | |
195 | /* Mask saddr with the netmask to adjust template granularity */ | |
cd17f9ed JV |
196 | #ifdef CONFIG_IP_VS_IPV6 |
197 | if (svc->af == AF_INET6) | |
198 | ipv6_addr_prefix(&snet.in6, &iph.saddr.in6, svc->netmask); | |
199 | else | |
200 | #endif | |
201 | snet.ip = iph.saddr.ip & svc->netmask; | |
1da177e4 | 202 | |
cd17f9ed JV |
203 | IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u " |
204 | "mnet %s\n", | |
205 | IP_VS_DBG_ADDR(svc->af, &iph.saddr), ntohs(ports[0]), | |
206 | IP_VS_DBG_ADDR(svc->af, &iph.daddr), ntohs(ports[1]), | |
207 | IP_VS_DBG_ADDR(svc->af, &snet)); | |
1da177e4 LT |
208 | |
209 | /* | |
210 | * As far as we know, FTP is a very complicated network protocol, and | |
211 | * it uses control connection and data connections. For active FTP, | |
212 | * FTP server initialize data connection to the client, its source port | |
213 | * is often 20. For passive FTP, FTP server tells the clients the port | |
214 | * that it passively listens to, and the client issues the data | |
215 | * connection. In the tunneling or direct routing mode, the load | |
216 | * balancer is on the client-to-server half of connection, the port | |
217 | * number is unknown to the load balancer. So, a conn template like | |
218 | * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP | |
219 | * service, and a template like <caddr, 0, vaddr, vport, daddr, dport> | |
220 | * is created for other persistent services. | |
221 | */ | |
222 | if (ports[1] == svc->port) { | |
223 | /* Check if a template already exists */ | |
224 | if (svc->port != FTPPORT) | |
cd17f9ed | 225 | ct = ip_vs_ct_in_get(svc->af, iph.protocol, &snet, 0, |
28364a59 | 226 | &iph.daddr, ports[1]); |
1da177e4 | 227 | else |
cd17f9ed | 228 | ct = ip_vs_ct_in_get(svc->af, iph.protocol, &snet, 0, |
28364a59 | 229 | &iph.daddr, 0); |
1da177e4 LT |
230 | |
231 | if (!ct || !ip_vs_check_template(ct)) { | |
232 | /* | |
233 | * No template found or the dest of the connection | |
234 | * template is not available. | |
235 | */ | |
236 | dest = svc->scheduler->schedule(svc, skb); | |
237 | if (dest == NULL) { | |
238 | IP_VS_DBG(1, "p-schedule: no dest found.\n"); | |
239 | return NULL; | |
240 | } | |
241 | ||
242 | /* | |
243 | * Create a template like <protocol,caddr,0, | |
244 | * vaddr,vport,daddr,dport> for non-ftp service, | |
245 | * and <protocol,caddr,0,vaddr,0,daddr,0> | |
246 | * for ftp service. | |
247 | */ | |
248 | if (svc->port != FTPPORT) | |
cd17f9ed | 249 | ct = ip_vs_conn_new(svc->af, iph.protocol, |
28364a59 JV |
250 | &snet, 0, |
251 | &iph.daddr, | |
1da177e4 | 252 | ports[1], |
28364a59 | 253 | &dest->addr, dest->port, |
87375ab4 | 254 | IP_VS_CONN_F_TEMPLATE, |
1da177e4 LT |
255 | dest); |
256 | else | |
cd17f9ed | 257 | ct = ip_vs_conn_new(svc->af, iph.protocol, |
28364a59 JV |
258 | &snet, 0, |
259 | &iph.daddr, 0, | |
260 | &dest->addr, 0, | |
87375ab4 | 261 | IP_VS_CONN_F_TEMPLATE, |
1da177e4 LT |
262 | dest); |
263 | if (ct == NULL) | |
264 | return NULL; | |
265 | ||
266 | ct->timeout = svc->timeout; | |
267 | } else { | |
268 | /* set destination with the found template */ | |
269 | dest = ct->dest; | |
270 | } | |
271 | dport = dest->port; | |
272 | } else { | |
273 | /* | |
274 | * Note: persistent fwmark-based services and persistent | |
275 | * port zero service are handled here. | |
276 | * fwmark template: <IPPROTO_IP,caddr,0,fwmark,0,daddr,0> | |
277 | * port zero template: <protocol,caddr,0,vaddr,0,daddr,0> | |
278 | */ | |
28364a59 JV |
279 | if (svc->fwmark) { |
280 | union nf_inet_addr fwmark = { | |
281 | .all = { 0, 0, 0, htonl(svc->fwmark) } | |
282 | }; | |
283 | ||
cd17f9ed | 284 | ct = ip_vs_ct_in_get(svc->af, IPPROTO_IP, &snet, 0, |
28364a59 JV |
285 | &fwmark, 0); |
286 | } else | |
cd17f9ed | 287 | ct = ip_vs_ct_in_get(svc->af, iph.protocol, &snet, 0, |
28364a59 | 288 | &iph.daddr, 0); |
1da177e4 LT |
289 | |
290 | if (!ct || !ip_vs_check_template(ct)) { | |
291 | /* | |
292 | * If it is not persistent port zero, return NULL, | |
293 | * otherwise create a connection template. | |
294 | */ | |
295 | if (svc->port) | |
296 | return NULL; | |
297 | ||
298 | dest = svc->scheduler->schedule(svc, skb); | |
299 | if (dest == NULL) { | |
300 | IP_VS_DBG(1, "p-schedule: no dest found.\n"); | |
301 | return NULL; | |
302 | } | |
303 | ||
304 | /* | |
305 | * Create a template according to the service | |
306 | */ | |
28364a59 JV |
307 | if (svc->fwmark) { |
308 | union nf_inet_addr fwmark = { | |
309 | .all = { 0, 0, 0, htonl(svc->fwmark) } | |
310 | }; | |
311 | ||
cd17f9ed | 312 | ct = ip_vs_conn_new(svc->af, IPPROTO_IP, |
28364a59 JV |
313 | &snet, 0, |
314 | &fwmark, 0, | |
315 | &dest->addr, 0, | |
87375ab4 | 316 | IP_VS_CONN_F_TEMPLATE, |
1da177e4 | 317 | dest); |
28364a59 | 318 | } else |
cd17f9ed | 319 | ct = ip_vs_conn_new(svc->af, iph.protocol, |
28364a59 JV |
320 | &snet, 0, |
321 | &iph.daddr, 0, | |
322 | &dest->addr, 0, | |
87375ab4 | 323 | IP_VS_CONN_F_TEMPLATE, |
1da177e4 LT |
324 | dest); |
325 | if (ct == NULL) | |
326 | return NULL; | |
327 | ||
328 | ct->timeout = svc->timeout; | |
329 | } else { | |
330 | /* set destination with the found template */ | |
331 | dest = ct->dest; | |
332 | } | |
333 | dport = ports[1]; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Create a new connection according to the template | |
338 | */ | |
cd17f9ed | 339 | cp = ip_vs_conn_new(svc->af, iph.protocol, |
28364a59 JV |
340 | &iph.saddr, ports[0], |
341 | &iph.daddr, ports[1], | |
342 | &dest->addr, dport, | |
1da177e4 LT |
343 | 0, |
344 | dest); | |
345 | if (cp == NULL) { | |
346 | ip_vs_conn_put(ct); | |
347 | return NULL; | |
348 | } | |
349 | ||
350 | /* | |
351 | * Add its control | |
352 | */ | |
353 | ip_vs_control_add(cp, ct); | |
354 | ip_vs_conn_put(ct); | |
355 | ||
356 | ip_vs_conn_stats(cp, svc); | |
357 | return cp; | |
358 | } | |
359 | ||
360 | ||
361 | /* | |
362 | * IPVS main scheduling function | |
363 | * It selects a server according to the virtual service, and | |
364 | * creates a connection entry. | |
365 | * Protocols supported: TCP, UDP | |
366 | */ | |
367 | struct ip_vs_conn * | |
368 | ip_vs_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) | |
369 | { | |
370 | struct ip_vs_conn *cp = NULL; | |
28364a59 | 371 | struct ip_vs_iphdr iph; |
1da177e4 | 372 | struct ip_vs_dest *dest; |
014d730d | 373 | __be16 _ports[2], *pptr; |
1da177e4 | 374 | |
28364a59 JV |
375 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); |
376 | pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports); | |
1da177e4 LT |
377 | if (pptr == NULL) |
378 | return NULL; | |
379 | ||
380 | /* | |
381 | * Persistent service | |
382 | */ | |
383 | if (svc->flags & IP_VS_SVC_F_PERSISTENT) | |
384 | return ip_vs_sched_persist(svc, skb, pptr); | |
385 | ||
386 | /* | |
387 | * Non-persistent service | |
388 | */ | |
389 | if (!svc->fwmark && pptr[1] != svc->port) { | |
390 | if (!svc->port) | |
391 | IP_VS_ERR("Schedule: port zero only supported " | |
392 | "in persistent services, " | |
393 | "check your ipvs configuration\n"); | |
394 | return NULL; | |
395 | } | |
396 | ||
397 | dest = svc->scheduler->schedule(svc, skb); | |
398 | if (dest == NULL) { | |
399 | IP_VS_DBG(1, "Schedule: no dest found.\n"); | |
400 | return NULL; | |
401 | } | |
402 | ||
403 | /* | |
404 | * Create a connection entry. | |
405 | */ | |
cd17f9ed | 406 | cp = ip_vs_conn_new(svc->af, iph.protocol, |
28364a59 JV |
407 | &iph.saddr, pptr[0], |
408 | &iph.daddr, pptr[1], | |
409 | &dest->addr, dest->port ? dest->port : pptr[1], | |
1da177e4 LT |
410 | 0, |
411 | dest); | |
412 | if (cp == NULL) | |
413 | return NULL; | |
414 | ||
cd17f9ed JV |
415 | IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u " |
416 | "d:%s:%u conn->flags:%X conn->refcnt:%d\n", | |
417 | ip_vs_fwd_tag(cp), | |
418 | IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport), | |
419 | IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport), | |
420 | IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport), | |
421 | cp->flags, atomic_read(&cp->refcnt)); | |
1da177e4 LT |
422 | |
423 | ip_vs_conn_stats(cp, svc); | |
424 | return cp; | |
425 | } | |
426 | ||
427 | ||
428 | /* | |
429 | * Pass or drop the packet. | |
430 | * Called by ip_vs_in, when the virtual service is available but | |
431 | * no destination is available for a new connection. | |
432 | */ | |
433 | int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb, | |
434 | struct ip_vs_protocol *pp) | |
435 | { | |
014d730d | 436 | __be16 _ports[2], *pptr; |
28364a59 | 437 | struct ip_vs_iphdr iph; |
2a3b791e JV |
438 | int unicast; |
439 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); | |
1da177e4 | 440 | |
28364a59 | 441 | pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports); |
1da177e4 LT |
442 | if (pptr == NULL) { |
443 | ip_vs_service_put(svc); | |
444 | return NF_DROP; | |
445 | } | |
446 | ||
2a3b791e JV |
447 | #ifdef CONFIG_IP_VS_IPV6 |
448 | if (svc->af == AF_INET6) | |
449 | unicast = ipv6_addr_type(&iph.daddr.in6) & IPV6_ADDR_UNICAST; | |
450 | else | |
451 | #endif | |
452 | unicast = (inet_addr_type(&init_net, iph.daddr.ip) == RTN_UNICAST); | |
453 | ||
1da177e4 | 454 | /* if it is fwmark-based service, the cache_bypass sysctl is up |
2a3b791e | 455 | and the destination is a non-local unicast, then create |
1da177e4 | 456 | a cache_bypass connection entry */ |
2a3b791e | 457 | if (sysctl_ip_vs_cache_bypass && svc->fwmark && unicast) { |
1da177e4 LT |
458 | int ret, cs; |
459 | struct ip_vs_conn *cp; | |
460 | ||
461 | ip_vs_service_put(svc); | |
462 | ||
463 | /* create a new connection entry */ | |
464 | IP_VS_DBG(6, "ip_vs_leave: create a cache_bypass entry\n"); | |
2a3b791e | 465 | cp = ip_vs_conn_new(svc->af, iph.protocol, |
28364a59 JV |
466 | &iph.saddr, pptr[0], |
467 | &iph.daddr, pptr[1], | |
1da177e4 LT |
468 | 0, 0, |
469 | IP_VS_CONN_F_BYPASS, | |
470 | NULL); | |
471 | if (cp == NULL) | |
472 | return NF_DROP; | |
473 | ||
474 | /* statistics */ | |
475 | ip_vs_in_stats(cp, skb); | |
476 | ||
477 | /* set state */ | |
478 | cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp); | |
479 | ||
480 | /* transmit the first SYN packet */ | |
481 | ret = cp->packet_xmit(skb, cp, pp); | |
482 | /* do not touch skb anymore */ | |
483 | ||
484 | atomic_inc(&cp->in_pkts); | |
485 | ip_vs_conn_put(cp); | |
486 | return ret; | |
487 | } | |
488 | ||
489 | /* | |
490 | * When the virtual ftp service is presented, packets destined | |
491 | * for other services on the VIP may get here (except services | |
492 | * listed in the ipvs table), pass the packets, because it is | |
493 | * not ipvs job to decide to drop the packets. | |
494 | */ | |
495 | if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) { | |
496 | ip_vs_service_put(svc); | |
497 | return NF_ACCEPT; | |
498 | } | |
499 | ||
500 | ip_vs_service_put(svc); | |
501 | ||
502 | /* | |
503 | * Notify the client that the destination is unreachable, and | |
504 | * release the socket buffer. | |
505 | * Since it is in IP layer, the TCP socket is not actually | |
506 | * created, the TCP RST packet cannot be sent, instead that | |
507 | * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ | |
508 | */ | |
2a3b791e JV |
509 | #ifdef CONFIG_IP_VS_IPV6 |
510 | if (svc->af == AF_INET6) | |
511 | icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, | |
512 | skb->dev); | |
513 | else | |
514 | #endif | |
515 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); | |
516 | ||
1da177e4 LT |
517 | return NF_DROP; |
518 | } | |
519 | ||
520 | ||
521 | /* | |
6e23ae2a | 522 | * It is hooked before NF_IP_PRI_NAT_SRC at the NF_INET_POST_ROUTING |
1da177e4 LT |
523 | * chain, and is used for VS/NAT. |
524 | * It detects packets for VS/NAT connections and sends the packets | |
525 | * immediately. This can avoid that iptable_nat mangles the packets | |
526 | * for VS/NAT. | |
527 | */ | |
528 | static unsigned int ip_vs_post_routing(unsigned int hooknum, | |
3db05fea | 529 | struct sk_buff *skb, |
1da177e4 LT |
530 | const struct net_device *in, |
531 | const struct net_device *out, | |
532 | int (*okfn)(struct sk_buff *)) | |
533 | { | |
3db05fea | 534 | if (!skb->ipvs_property) |
1da177e4 | 535 | return NF_ACCEPT; |
1da177e4 | 536 | /* The packet was sent from IPVS, exit this chain */ |
abbcc739 | 537 | return NF_STOP; |
1da177e4 LT |
538 | } |
539 | ||
b1550f22 | 540 | __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset) |
1da177e4 | 541 | { |
d3bc23e7 | 542 | return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0)); |
1da177e4 LT |
543 | } |
544 | ||
776c729e | 545 | static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user) |
1da177e4 | 546 | { |
776c729e HX |
547 | int err = ip_defrag(skb, user); |
548 | ||
549 | if (!err) | |
eddc9ec5 | 550 | ip_send_check(ip_hdr(skb)); |
776c729e HX |
551 | |
552 | return err; | |
1da177e4 LT |
553 | } |
554 | ||
2a3b791e JV |
555 | #ifdef CONFIG_IP_VS_IPV6 |
556 | static inline int ip_vs_gather_frags_v6(struct sk_buff *skb, u_int32_t user) | |
557 | { | |
558 | /* TODO IPv6: Find out what to do here for IPv6 */ | |
559 | return 0; | |
560 | } | |
561 | #endif | |
562 | ||
1da177e4 LT |
563 | /* |
564 | * Packet has been made sufficiently writable in caller | |
565 | * - inout: 1=in->out, 0=out->in | |
566 | */ | |
567 | void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp, | |
568 | struct ip_vs_conn *cp, int inout) | |
569 | { | |
eddc9ec5 | 570 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 | 571 | unsigned int icmp_offset = iph->ihl*4; |
d56f90a7 ACM |
572 | struct icmphdr *icmph = (struct icmphdr *)(skb_network_header(skb) + |
573 | icmp_offset); | |
1da177e4 LT |
574 | struct iphdr *ciph = (struct iphdr *)(icmph + 1); |
575 | ||
576 | if (inout) { | |
e7ade46a | 577 | iph->saddr = cp->vaddr.ip; |
1da177e4 | 578 | ip_send_check(iph); |
e7ade46a | 579 | ciph->daddr = cp->vaddr.ip; |
1da177e4 LT |
580 | ip_send_check(ciph); |
581 | } else { | |
e7ade46a | 582 | iph->daddr = cp->daddr.ip; |
1da177e4 | 583 | ip_send_check(iph); |
e7ade46a | 584 | ciph->saddr = cp->daddr.ip; |
1da177e4 LT |
585 | ip_send_check(ciph); |
586 | } | |
587 | ||
588 | /* the TCP/UDP port */ | |
589 | if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol) { | |
014d730d | 590 | __be16 *ports = (void *)ciph + ciph->ihl*4; |
1da177e4 LT |
591 | |
592 | if (inout) | |
593 | ports[1] = cp->vport; | |
594 | else | |
595 | ports[0] = cp->dport; | |
596 | } | |
597 | ||
598 | /* And finally the ICMP checksum */ | |
599 | icmph->checksum = 0; | |
600 | icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset); | |
601 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
602 | ||
603 | if (inout) | |
604 | IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph, | |
605 | "Forwarding altered outgoing ICMP"); | |
606 | else | |
607 | IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph, | |
608 | "Forwarding altered incoming ICMP"); | |
609 | } | |
610 | ||
b3cdd2a7 JV |
611 | #ifdef CONFIG_IP_VS_IPV6 |
612 | void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp, | |
613 | struct ip_vs_conn *cp, int inout) | |
614 | { | |
615 | struct ipv6hdr *iph = ipv6_hdr(skb); | |
616 | unsigned int icmp_offset = sizeof(struct ipv6hdr); | |
617 | struct icmp6hdr *icmph = (struct icmp6hdr *)(skb_network_header(skb) + | |
618 | icmp_offset); | |
619 | struct ipv6hdr *ciph = (struct ipv6hdr *)(icmph + 1); | |
620 | ||
621 | if (inout) { | |
622 | iph->saddr = cp->vaddr.in6; | |
623 | ciph->daddr = cp->vaddr.in6; | |
624 | } else { | |
625 | iph->daddr = cp->daddr.in6; | |
626 | ciph->saddr = cp->daddr.in6; | |
627 | } | |
628 | ||
629 | /* the TCP/UDP port */ | |
630 | if (IPPROTO_TCP == ciph->nexthdr || IPPROTO_UDP == ciph->nexthdr) { | |
631 | __be16 *ports = (void *)ciph + sizeof(struct ipv6hdr); | |
632 | ||
633 | if (inout) | |
634 | ports[1] = cp->vport; | |
635 | else | |
636 | ports[0] = cp->dport; | |
637 | } | |
638 | ||
639 | /* And finally the ICMP checksum */ | |
640 | icmph->icmp6_cksum = 0; | |
641 | /* TODO IPv6: is this correct for ICMPv6? */ | |
642 | ip_vs_checksum_complete(skb, icmp_offset); | |
643 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
644 | ||
645 | if (inout) | |
646 | IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph, | |
647 | "Forwarding altered outgoing ICMPv6"); | |
648 | else | |
649 | IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph, | |
650 | "Forwarding altered incoming ICMPv6"); | |
651 | } | |
652 | #endif | |
653 | ||
1da177e4 LT |
654 | /* |
655 | * Handle ICMP messages in the inside-to-outside direction (outgoing). | |
656 | * Find any that might be relevant, check against existing connections, | |
657 | * forward to the right destination host if relevant. | |
658 | * Currently handles error types - unreachable, quench, ttl exceeded. | |
659 | * (Only used in VS/NAT) | |
660 | */ | |
3db05fea | 661 | static int ip_vs_out_icmp(struct sk_buff *skb, int *related) |
1da177e4 | 662 | { |
1da177e4 LT |
663 | struct iphdr *iph; |
664 | struct icmphdr _icmph, *ic; | |
665 | struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */ | |
51ef348b | 666 | struct ip_vs_iphdr ciph; |
1da177e4 LT |
667 | struct ip_vs_conn *cp; |
668 | struct ip_vs_protocol *pp; | |
669 | unsigned int offset, ihl, verdict; | |
670 | ||
671 | *related = 1; | |
672 | ||
673 | /* reassemble IP fragments */ | |
eddc9ec5 | 674 | if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) { |
776c729e | 675 | if (ip_vs_gather_frags(skb, IP_DEFRAG_VS_OUT)) |
1da177e4 | 676 | return NF_STOLEN; |
1da177e4 LT |
677 | } |
678 | ||
eddc9ec5 | 679 | iph = ip_hdr(skb); |
1da177e4 LT |
680 | offset = ihl = iph->ihl * 4; |
681 | ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); | |
682 | if (ic == NULL) | |
683 | return NF_DROP; | |
684 | ||
685 | IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n", | |
686 | ic->type, ntohs(icmp_id(ic)), | |
687 | NIPQUAD(iph->saddr), NIPQUAD(iph->daddr)); | |
688 | ||
689 | /* | |
690 | * Work through seeing if this is for us. | |
691 | * These checks are supposed to be in an order that means easy | |
692 | * things are checked first to speed up processing.... however | |
693 | * this means that some packets will manage to get a long way | |
694 | * down this stack and then be rejected, but that's life. | |
695 | */ | |
696 | if ((ic->type != ICMP_DEST_UNREACH) && | |
697 | (ic->type != ICMP_SOURCE_QUENCH) && | |
698 | (ic->type != ICMP_TIME_EXCEEDED)) { | |
699 | *related = 0; | |
700 | return NF_ACCEPT; | |
701 | } | |
702 | ||
703 | /* Now find the contained IP header */ | |
704 | offset += sizeof(_icmph); | |
705 | cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); | |
706 | if (cih == NULL) | |
707 | return NF_ACCEPT; /* The packet looks wrong, ignore */ | |
708 | ||
709 | pp = ip_vs_proto_get(cih->protocol); | |
710 | if (!pp) | |
711 | return NF_ACCEPT; | |
712 | ||
713 | /* Is the embedded protocol header present? */ | |
4412ec49 | 714 | if (unlikely(cih->frag_off & htons(IP_OFFSET) && |
1da177e4 LT |
715 | pp->dont_defrag)) |
716 | return NF_ACCEPT; | |
717 | ||
718 | IP_VS_DBG_PKT(11, pp, skb, offset, "Checking outgoing ICMP for"); | |
719 | ||
720 | offset += cih->ihl * 4; | |
721 | ||
51ef348b | 722 | ip_vs_fill_iphdr(AF_INET, cih, &ciph); |
1da177e4 | 723 | /* The embedded headers contain source and dest in reverse order */ |
51ef348b | 724 | cp = pp->conn_out_get(AF_INET, skb, pp, &ciph, offset, 1); |
1da177e4 LT |
725 | if (!cp) |
726 | return NF_ACCEPT; | |
727 | ||
728 | verdict = NF_DROP; | |
729 | ||
730 | if (IP_VS_FWD_METHOD(cp) != 0) { | |
464c4f18 | 731 | IP_VS_ERR("shouldn't reach here, because the box is on the " |
1da177e4 LT |
732 | "half connection in the tun/dr module.\n"); |
733 | } | |
734 | ||
735 | /* Ensure the checksum is correct */ | |
60476372 | 736 | if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) { |
1da177e4 LT |
737 | /* Failed checksum! */ |
738 | IP_VS_DBG(1, "Forward ICMP: failed checksum from %d.%d.%d.%d!\n", | |
739 | NIPQUAD(iph->saddr)); | |
740 | goto out; | |
741 | } | |
742 | ||
743 | if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol) | |
744 | offset += 2 * sizeof(__u16); | |
af1e1cf0 | 745 | if (!skb_make_writable(skb, offset)) |
1da177e4 | 746 | goto out; |
1da177e4 LT |
747 | |
748 | ip_vs_nat_icmp(skb, pp, cp, 1); | |
749 | ||
750 | /* do the statistics and put it back */ | |
751 | ip_vs_out_stats(cp, skb); | |
752 | ||
6869c4d8 | 753 | skb->ipvs_property = 1; |
1da177e4 LT |
754 | verdict = NF_ACCEPT; |
755 | ||
756 | out: | |
757 | __ip_vs_conn_put(cp); | |
758 | ||
759 | return verdict; | |
760 | } | |
761 | ||
2a3b791e JV |
762 | #ifdef CONFIG_IP_VS_IPV6 |
763 | static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related) | |
764 | { | |
765 | struct ipv6hdr *iph; | |
766 | struct icmp6hdr _icmph, *ic; | |
767 | struct ipv6hdr _ciph, *cih; /* The ip header contained | |
768 | within the ICMP */ | |
769 | struct ip_vs_iphdr ciph; | |
770 | struct ip_vs_conn *cp; | |
771 | struct ip_vs_protocol *pp; | |
772 | unsigned int offset, verdict; | |
773 | ||
774 | *related = 1; | |
775 | ||
776 | /* reassemble IP fragments */ | |
777 | if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) { | |
778 | if (ip_vs_gather_frags_v6(skb, IP_DEFRAG_VS_OUT)) | |
779 | return NF_STOLEN; | |
780 | } | |
781 | ||
782 | iph = ipv6_hdr(skb); | |
783 | offset = sizeof(struct ipv6hdr); | |
784 | ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); | |
785 | if (ic == NULL) | |
786 | return NF_DROP; | |
787 | ||
788 | IP_VS_DBG(12, "Outgoing ICMPv6 (%d,%d) " NIP6_FMT "->" NIP6_FMT "\n", | |
789 | ic->icmp6_type, ntohs(icmpv6_id(ic)), | |
790 | NIP6(iph->saddr), NIP6(iph->daddr)); | |
791 | ||
792 | /* | |
793 | * Work through seeing if this is for us. | |
794 | * These checks are supposed to be in an order that means easy | |
795 | * things are checked first to speed up processing.... however | |
796 | * this means that some packets will manage to get a long way | |
797 | * down this stack and then be rejected, but that's life. | |
798 | */ | |
799 | if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) && | |
800 | (ic->icmp6_type != ICMPV6_PKT_TOOBIG) && | |
801 | (ic->icmp6_type != ICMPV6_TIME_EXCEED)) { | |
802 | *related = 0; | |
803 | return NF_ACCEPT; | |
804 | } | |
805 | ||
806 | /* Now find the contained IP header */ | |
807 | offset += sizeof(_icmph); | |
808 | cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); | |
809 | if (cih == NULL) | |
810 | return NF_ACCEPT; /* The packet looks wrong, ignore */ | |
811 | ||
812 | pp = ip_vs_proto_get(cih->nexthdr); | |
813 | if (!pp) | |
814 | return NF_ACCEPT; | |
815 | ||
816 | /* Is the embedded protocol header present? */ | |
817 | /* TODO: we don't support fragmentation at the moment anyways */ | |
818 | if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag)) | |
819 | return NF_ACCEPT; | |
820 | ||
821 | IP_VS_DBG_PKT(11, pp, skb, offset, "Checking outgoing ICMPv6 for"); | |
822 | ||
823 | offset += sizeof(struct ipv6hdr); | |
824 | ||
825 | ip_vs_fill_iphdr(AF_INET6, cih, &ciph); | |
826 | /* The embedded headers contain source and dest in reverse order */ | |
827 | cp = pp->conn_out_get(AF_INET6, skb, pp, &ciph, offset, 1); | |
828 | if (!cp) | |
829 | return NF_ACCEPT; | |
830 | ||
831 | verdict = NF_DROP; | |
832 | ||
833 | if (IP_VS_FWD_METHOD(cp) != 0) { | |
834 | IP_VS_ERR("shouldn't reach here, because the box is on the " | |
835 | "half connection in the tun/dr module.\n"); | |
836 | } | |
837 | ||
838 | /* Ensure the checksum is correct */ | |
839 | if (!skb_csum_unnecessary(skb) | |
840 | && ip_vs_checksum_complete(skb, sizeof(struct ipv6hdr))) { | |
841 | /* Failed checksum! */ | |
842 | IP_VS_DBG(1, "Forward ICMPv6: failed checksum from " | |
843 | NIP6_FMT "!\n", | |
844 | NIP6(iph->saddr)); | |
845 | goto out; | |
846 | } | |
847 | ||
848 | if (IPPROTO_TCP == cih->nexthdr || IPPROTO_UDP == cih->nexthdr) | |
849 | offset += 2 * sizeof(__u16); | |
850 | if (!skb_make_writable(skb, offset)) | |
851 | goto out; | |
852 | ||
853 | ip_vs_nat_icmp_v6(skb, pp, cp, 1); | |
854 | ||
855 | /* do the statistics and put it back */ | |
856 | ip_vs_out_stats(cp, skb); | |
857 | ||
858 | skb->ipvs_property = 1; | |
859 | verdict = NF_ACCEPT; | |
860 | ||
861 | out: | |
862 | __ip_vs_conn_put(cp); | |
863 | ||
864 | return verdict; | |
865 | } | |
866 | #endif | |
867 | ||
868 | static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len) | |
1da177e4 LT |
869 | { |
870 | struct tcphdr _tcph, *th; | |
871 | ||
2a3b791e | 872 | th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph); |
1da177e4 LT |
873 | if (th == NULL) |
874 | return 0; | |
875 | return th->rst; | |
876 | } | |
877 | ||
878 | /* | |
6e23ae2a | 879 | * It is hooked at the NF_INET_FORWARD chain, used only for VS/NAT. |
1da177e4 LT |
880 | * Check if outgoing packet belongs to the established ip_vs_conn, |
881 | * rewrite addresses of the packet and send it on its way... | |
882 | */ | |
883 | static unsigned int | |
3db05fea | 884 | ip_vs_out(unsigned int hooknum, struct sk_buff *skb, |
1da177e4 LT |
885 | const struct net_device *in, const struct net_device *out, |
886 | int (*okfn)(struct sk_buff *)) | |
887 | { | |
51ef348b | 888 | struct ip_vs_iphdr iph; |
1da177e4 LT |
889 | struct ip_vs_protocol *pp; |
890 | struct ip_vs_conn *cp; | |
2a3b791e | 891 | int af; |
1da177e4 LT |
892 | |
893 | EnterFunction(11); | |
894 | ||
2a3b791e JV |
895 | af = (skb->protocol == __constant_htons(ETH_P_IP)) ? AF_INET : AF_INET6; |
896 | ||
6869c4d8 | 897 | if (skb->ipvs_property) |
1da177e4 LT |
898 | return NF_ACCEPT; |
899 | ||
2a3b791e JV |
900 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); |
901 | #ifdef CONFIG_IP_VS_IPV6 | |
902 | if (af == AF_INET6) { | |
903 | if (unlikely(iph.protocol == IPPROTO_ICMPV6)) { | |
904 | int related, verdict = ip_vs_out_icmp_v6(skb, &related); | |
1da177e4 | 905 | |
2a3b791e JV |
906 | if (related) |
907 | return verdict; | |
908 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); | |
909 | } | |
910 | } else | |
911 | #endif | |
912 | if (unlikely(iph.protocol == IPPROTO_ICMP)) { | |
913 | int related, verdict = ip_vs_out_icmp(skb, &related); | |
914 | ||
915 | if (related) | |
916 | return verdict; | |
917 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); | |
918 | } | |
1da177e4 | 919 | |
51ef348b | 920 | pp = ip_vs_proto_get(iph.protocol); |
1da177e4 LT |
921 | if (unlikely(!pp)) |
922 | return NF_ACCEPT; | |
923 | ||
924 | /* reassemble IP fragments */ | |
2a3b791e JV |
925 | #ifdef CONFIG_IP_VS_IPV6 |
926 | if (af == AF_INET6) { | |
927 | if (unlikely(iph.protocol == IPPROTO_ICMPV6)) { | |
928 | int related, verdict = ip_vs_out_icmp_v6(skb, &related); | |
1da177e4 | 929 | |
2a3b791e JV |
930 | if (related) |
931 | return verdict; | |
932 | ||
933 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); | |
934 | } | |
935 | } else | |
936 | #endif | |
937 | if (unlikely(ip_hdr(skb)->frag_off & htons(IP_MF|IP_OFFSET) && | |
938 | !pp->dont_defrag)) { | |
939 | if (ip_vs_gather_frags(skb, IP_DEFRAG_VS_OUT)) | |
940 | return NF_STOLEN; | |
941 | ||
942 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); | |
943 | } | |
1da177e4 LT |
944 | |
945 | /* | |
946 | * Check if the packet belongs to an existing entry | |
947 | */ | |
2a3b791e | 948 | cp = pp->conn_out_get(af, skb, pp, &iph, iph.len, 0); |
1da177e4 LT |
949 | |
950 | if (unlikely(!cp)) { | |
951 | if (sysctl_ip_vs_nat_icmp_send && | |
952 | (pp->protocol == IPPROTO_TCP || | |
953 | pp->protocol == IPPROTO_UDP)) { | |
014d730d | 954 | __be16 _ports[2], *pptr; |
1da177e4 | 955 | |
51ef348b | 956 | pptr = skb_header_pointer(skb, iph.len, |
1da177e4 LT |
957 | sizeof(_ports), _ports); |
958 | if (pptr == NULL) | |
959 | return NF_ACCEPT; /* Not for me */ | |
7937df15 JV |
960 | if (ip_vs_lookup_real_service(af, iph.protocol, |
961 | &iph.saddr, | |
2a3b791e | 962 | pptr[0])) { |
1da177e4 LT |
963 | /* |
964 | * Notify the real server: there is no | |
965 | * existing entry if it is not RST | |
966 | * packet or not TCP packet. | |
967 | */ | |
51ef348b | 968 | if (iph.protocol != IPPROTO_TCP |
2a3b791e JV |
969 | || !is_tcp_reset(skb, iph.len)) { |
970 | #ifdef CONFIG_IP_VS_IPV6 | |
971 | if (af == AF_INET6) | |
972 | icmpv6_send(skb, | |
973 | ICMPV6_DEST_UNREACH, | |
974 | ICMPV6_PORT_UNREACH, | |
975 | 0, skb->dev); | |
976 | else | |
977 | #endif | |
978 | icmp_send(skb, | |
979 | ICMP_DEST_UNREACH, | |
980 | ICMP_PORT_UNREACH, 0); | |
1da177e4 LT |
981 | return NF_DROP; |
982 | } | |
983 | } | |
984 | } | |
985 | IP_VS_DBG_PKT(12, pp, skb, 0, | |
986 | "packet continues traversal as normal"); | |
987 | return NF_ACCEPT; | |
988 | } | |
989 | ||
990 | IP_VS_DBG_PKT(11, pp, skb, 0, "Outgoing packet"); | |
991 | ||
51ef348b | 992 | if (!skb_make_writable(skb, iph.len)) |
1da177e4 LT |
993 | goto drop; |
994 | ||
995 | /* mangle the packet */ | |
3db05fea | 996 | if (pp->snat_handler && !pp->snat_handler(skb, pp, cp)) |
1da177e4 | 997 | goto drop; |
2a3b791e JV |
998 | |
999 | #ifdef CONFIG_IP_VS_IPV6 | |
1000 | if (af == AF_INET6) | |
1001 | ipv6_hdr(skb)->saddr = cp->vaddr.in6; | |
1002 | else | |
1003 | #endif | |
1004 | { | |
1005 | ip_hdr(skb)->saddr = cp->vaddr.ip; | |
1006 | ip_send_check(ip_hdr(skb)); | |
1007 | } | |
1da177e4 | 1008 | |
e905a9ed YH |
1009 | /* For policy routing, packets originating from this |
1010 | * machine itself may be routed differently to packets | |
1011 | * passing through. We want this packet to be routed as | |
1012 | * if it came from this machine itself. So re-compute | |
1013 | * the routing information. | |
1014 | */ | |
2a3b791e JV |
1015 | #ifdef CONFIG_IP_VS_IPV6 |
1016 | if (af == AF_INET6) { | |
1017 | if (ip6_route_me_harder(skb) != 0) | |
1018 | goto drop; | |
1019 | } else | |
1020 | #endif | |
1021 | if (ip_route_me_harder(skb, RTN_LOCAL) != 0) | |
1022 | goto drop; | |
901eaf6c | 1023 | |
1da177e4 LT |
1024 | IP_VS_DBG_PKT(10, pp, skb, 0, "After SNAT"); |
1025 | ||
1026 | ip_vs_out_stats(cp, skb); | |
1027 | ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pp); | |
1028 | ip_vs_conn_put(cp); | |
1029 | ||
6869c4d8 | 1030 | skb->ipvs_property = 1; |
1da177e4 LT |
1031 | |
1032 | LeaveFunction(11); | |
1033 | return NF_ACCEPT; | |
1034 | ||
1035 | drop: | |
1036 | ip_vs_conn_put(cp); | |
3db05fea | 1037 | kfree_skb(skb); |
1da177e4 LT |
1038 | return NF_STOLEN; |
1039 | } | |
1040 | ||
1041 | ||
1042 | /* | |
1043 | * Handle ICMP messages in the outside-to-inside direction (incoming). | |
1044 | * Find any that might be relevant, check against existing connections, | |
1045 | * forward to the right destination host if relevant. | |
1046 | * Currently handles error types - unreachable, quench, ttl exceeded. | |
1047 | */ | |
e905a9ed | 1048 | static int |
3db05fea | 1049 | ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum) |
1da177e4 | 1050 | { |
1da177e4 LT |
1051 | struct iphdr *iph; |
1052 | struct icmphdr _icmph, *ic; | |
1053 | struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */ | |
51ef348b | 1054 | struct ip_vs_iphdr ciph; |
1da177e4 LT |
1055 | struct ip_vs_conn *cp; |
1056 | struct ip_vs_protocol *pp; | |
1057 | unsigned int offset, ihl, verdict; | |
1058 | ||
1059 | *related = 1; | |
1060 | ||
1061 | /* reassemble IP fragments */ | |
eddc9ec5 | 1062 | if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) { |
6e23ae2a | 1063 | if (ip_vs_gather_frags(skb, hooknum == NF_INET_LOCAL_IN ? |
776c729e | 1064 | IP_DEFRAG_VS_IN : IP_DEFRAG_VS_FWD)) |
1da177e4 | 1065 | return NF_STOLEN; |
1da177e4 LT |
1066 | } |
1067 | ||
eddc9ec5 | 1068 | iph = ip_hdr(skb); |
1da177e4 LT |
1069 | offset = ihl = iph->ihl * 4; |
1070 | ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); | |
1071 | if (ic == NULL) | |
1072 | return NF_DROP; | |
1073 | ||
1074 | IP_VS_DBG(12, "Incoming ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n", | |
1075 | ic->type, ntohs(icmp_id(ic)), | |
1076 | NIPQUAD(iph->saddr), NIPQUAD(iph->daddr)); | |
1077 | ||
1078 | /* | |
1079 | * Work through seeing if this is for us. | |
1080 | * These checks are supposed to be in an order that means easy | |
1081 | * things are checked first to speed up processing.... however | |
1082 | * this means that some packets will manage to get a long way | |
1083 | * down this stack and then be rejected, but that's life. | |
1084 | */ | |
1085 | if ((ic->type != ICMP_DEST_UNREACH) && | |
1086 | (ic->type != ICMP_SOURCE_QUENCH) && | |
1087 | (ic->type != ICMP_TIME_EXCEEDED)) { | |
1088 | *related = 0; | |
1089 | return NF_ACCEPT; | |
1090 | } | |
1091 | ||
1092 | /* Now find the contained IP header */ | |
1093 | offset += sizeof(_icmph); | |
1094 | cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); | |
1095 | if (cih == NULL) | |
1096 | return NF_ACCEPT; /* The packet looks wrong, ignore */ | |
1097 | ||
1098 | pp = ip_vs_proto_get(cih->protocol); | |
1099 | if (!pp) | |
1100 | return NF_ACCEPT; | |
1101 | ||
1102 | /* Is the embedded protocol header present? */ | |
4412ec49 | 1103 | if (unlikely(cih->frag_off & htons(IP_OFFSET) && |
1da177e4 LT |
1104 | pp->dont_defrag)) |
1105 | return NF_ACCEPT; | |
1106 | ||
1107 | IP_VS_DBG_PKT(11, pp, skb, offset, "Checking incoming ICMP for"); | |
1108 | ||
1109 | offset += cih->ihl * 4; | |
1110 | ||
51ef348b | 1111 | ip_vs_fill_iphdr(AF_INET, cih, &ciph); |
1da177e4 | 1112 | /* The embedded headers contain source and dest in reverse order */ |
51ef348b | 1113 | cp = pp->conn_in_get(AF_INET, skb, pp, &ciph, offset, 1); |
1da177e4 LT |
1114 | if (!cp) |
1115 | return NF_ACCEPT; | |
1116 | ||
1117 | verdict = NF_DROP; | |
1118 | ||
1119 | /* Ensure the checksum is correct */ | |
60476372 | 1120 | if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) { |
1da177e4 LT |
1121 | /* Failed checksum! */ |
1122 | IP_VS_DBG(1, "Incoming ICMP: failed checksum from %d.%d.%d.%d!\n", | |
1123 | NIPQUAD(iph->saddr)); | |
1124 | goto out; | |
1125 | } | |
1126 | ||
1127 | /* do the statistics and put it back */ | |
1128 | ip_vs_in_stats(cp, skb); | |
1129 | if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol) | |
1130 | offset += 2 * sizeof(__u16); | |
1131 | verdict = ip_vs_icmp_xmit(skb, cp, pp, offset); | |
1132 | /* do not touch skb anymore */ | |
1133 | ||
1134 | out: | |
1135 | __ip_vs_conn_put(cp); | |
1136 | ||
1137 | return verdict; | |
1138 | } | |
1139 | ||
2a3b791e JV |
1140 | #ifdef CONFIG_IP_VS_IPV6 |
1141 | static int | |
1142 | ip_vs_in_icmp_v6(struct sk_buff *skb, int *related, unsigned int hooknum) | |
1143 | { | |
1144 | struct ipv6hdr *iph; | |
1145 | struct icmp6hdr _icmph, *ic; | |
1146 | struct ipv6hdr _ciph, *cih; /* The ip header contained | |
1147 | within the ICMP */ | |
1148 | struct ip_vs_iphdr ciph; | |
1149 | struct ip_vs_conn *cp; | |
1150 | struct ip_vs_protocol *pp; | |
1151 | unsigned int offset, verdict; | |
1152 | ||
1153 | *related = 1; | |
1154 | ||
1155 | /* reassemble IP fragments */ | |
1156 | if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) { | |
1157 | if (ip_vs_gather_frags_v6(skb, hooknum == NF_INET_LOCAL_IN ? | |
1158 | IP_DEFRAG_VS_IN : | |
1159 | IP_DEFRAG_VS_FWD)) | |
1160 | return NF_STOLEN; | |
1161 | } | |
1162 | ||
1163 | iph = ipv6_hdr(skb); | |
1164 | offset = sizeof(struct ipv6hdr); | |
1165 | ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); | |
1166 | if (ic == NULL) | |
1167 | return NF_DROP; | |
1168 | ||
1169 | IP_VS_DBG(12, "Incoming ICMPv6 (%d,%d) " NIP6_FMT "->" NIP6_FMT "\n", | |
1170 | ic->icmp6_type, ntohs(icmpv6_id(ic)), | |
1171 | NIP6(iph->saddr), NIP6(iph->daddr)); | |
1172 | ||
1173 | /* | |
1174 | * Work through seeing if this is for us. | |
1175 | * These checks are supposed to be in an order that means easy | |
1176 | * things are checked first to speed up processing.... however | |
1177 | * this means that some packets will manage to get a long way | |
1178 | * down this stack and then be rejected, but that's life. | |
1179 | */ | |
1180 | if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) && | |
1181 | (ic->icmp6_type != ICMPV6_PKT_TOOBIG) && | |
1182 | (ic->icmp6_type != ICMPV6_TIME_EXCEED)) { | |
1183 | *related = 0; | |
1184 | return NF_ACCEPT; | |
1185 | } | |
1186 | ||
1187 | /* Now find the contained IP header */ | |
1188 | offset += sizeof(_icmph); | |
1189 | cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); | |
1190 | if (cih == NULL) | |
1191 | return NF_ACCEPT; /* The packet looks wrong, ignore */ | |
1192 | ||
1193 | pp = ip_vs_proto_get(cih->nexthdr); | |
1194 | if (!pp) | |
1195 | return NF_ACCEPT; | |
1196 | ||
1197 | /* Is the embedded protocol header present? */ | |
1198 | /* TODO: we don't support fragmentation at the moment anyways */ | |
1199 | if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag)) | |
1200 | return NF_ACCEPT; | |
1201 | ||
1202 | IP_VS_DBG_PKT(11, pp, skb, offset, "Checking incoming ICMPv6 for"); | |
1203 | ||
1204 | offset += sizeof(struct ipv6hdr); | |
1205 | ||
1206 | ip_vs_fill_iphdr(AF_INET6, cih, &ciph); | |
1207 | /* The embedded headers contain source and dest in reverse order */ | |
1208 | cp = pp->conn_in_get(AF_INET6, skb, pp, &ciph, offset, 1); | |
1209 | if (!cp) | |
1210 | return NF_ACCEPT; | |
1211 | ||
1212 | verdict = NF_DROP; | |
1213 | ||
1214 | /* do the statistics and put it back */ | |
1215 | ip_vs_in_stats(cp, skb); | |
1216 | if (IPPROTO_TCP == cih->nexthdr || IPPROTO_UDP == cih->nexthdr) | |
1217 | offset += 2 * sizeof(__u16); | |
1218 | verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset); | |
1219 | /* do not touch skb anymore */ | |
1220 | ||
1221 | __ip_vs_conn_put(cp); | |
1222 | ||
1223 | return verdict; | |
1224 | } | |
1225 | #endif | |
1226 | ||
1227 | ||
1da177e4 LT |
1228 | /* |
1229 | * Check if it's for virtual services, look it up, | |
1230 | * and send it on its way... | |
1231 | */ | |
1232 | static unsigned int | |
3db05fea | 1233 | ip_vs_in(unsigned int hooknum, struct sk_buff *skb, |
1da177e4 LT |
1234 | const struct net_device *in, const struct net_device *out, |
1235 | int (*okfn)(struct sk_buff *)) | |
1236 | { | |
51ef348b | 1237 | struct ip_vs_iphdr iph; |
1da177e4 LT |
1238 | struct ip_vs_protocol *pp; |
1239 | struct ip_vs_conn *cp; | |
2a3b791e JV |
1240 | int ret, restart, af; |
1241 | ||
1242 | af = (skb->protocol == __constant_htons(ETH_P_IP)) ? AF_INET : AF_INET6; | |
51ef348b | 1243 | |
2a3b791e | 1244 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); |
1da177e4 LT |
1245 | |
1246 | /* | |
1247 | * Big tappo: only PACKET_HOST (neither loopback nor mcasts) | |
1248 | * ... don't know why 1st test DOES NOT include 2nd (?) | |
1249 | */ | |
1250 | if (unlikely(skb->pkt_type != PACKET_HOST | |
0cc217e1 | 1251 | || skb->dev->flags & IFF_LOOPBACK || skb->sk)) { |
51ef348b JV |
1252 | IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s ignored\n", |
1253 | skb->pkt_type, | |
1254 | iph.protocol, | |
2a3b791e | 1255 | IP_VS_DBG_ADDR(af, &iph.daddr)); |
1da177e4 LT |
1256 | return NF_ACCEPT; |
1257 | } | |
1258 | ||
51ef348b | 1259 | if (unlikely(iph.protocol == IPPROTO_ICMP)) { |
3db05fea | 1260 | int related, verdict = ip_vs_in_icmp(skb, &related, hooknum); |
1da177e4 LT |
1261 | |
1262 | if (related) | |
1263 | return verdict; | |
2a3b791e | 1264 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); |
1da177e4 LT |
1265 | } |
1266 | ||
1267 | /* Protocol supported? */ | |
51ef348b | 1268 | pp = ip_vs_proto_get(iph.protocol); |
1da177e4 LT |
1269 | if (unlikely(!pp)) |
1270 | return NF_ACCEPT; | |
1271 | ||
1da177e4 LT |
1272 | /* |
1273 | * Check if the packet belongs to an existing connection entry | |
1274 | */ | |
2a3b791e | 1275 | cp = pp->conn_in_get(af, skb, pp, &iph, iph.len, 0); |
1da177e4 LT |
1276 | |
1277 | if (unlikely(!cp)) { | |
1278 | int v; | |
1279 | ||
2a3b791e | 1280 | if (!pp->conn_schedule(af, skb, pp, &v, &cp)) |
1da177e4 LT |
1281 | return v; |
1282 | } | |
1283 | ||
1284 | if (unlikely(!cp)) { | |
1285 | /* sorry, all this trouble for a no-hit :) */ | |
1286 | IP_VS_DBG_PKT(12, pp, skb, 0, | |
1287 | "packet continues traversal as normal"); | |
1288 | return NF_ACCEPT; | |
1289 | } | |
1290 | ||
1291 | IP_VS_DBG_PKT(11, pp, skb, 0, "Incoming packet"); | |
1292 | ||
1293 | /* Check the server status */ | |
1294 | if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) { | |
1295 | /* the destination server is not available */ | |
1296 | ||
1297 | if (sysctl_ip_vs_expire_nodest_conn) { | |
1298 | /* try to expire the connection immediately */ | |
1299 | ip_vs_conn_expire_now(cp); | |
1da177e4 | 1300 | } |
dc8103f2 JA |
1301 | /* don't restart its timer, and silently |
1302 | drop the packet. */ | |
1303 | __ip_vs_conn_put(cp); | |
1da177e4 LT |
1304 | return NF_DROP; |
1305 | } | |
1306 | ||
1307 | ip_vs_in_stats(cp, skb); | |
1308 | restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp); | |
1309 | if (cp->packet_xmit) | |
1310 | ret = cp->packet_xmit(skb, cp, pp); | |
1311 | /* do not touch skb anymore */ | |
1312 | else { | |
1313 | IP_VS_DBG_RL("warning: packet_xmit is null"); | |
1314 | ret = NF_ACCEPT; | |
1315 | } | |
1316 | ||
efac5276 RB |
1317 | /* Increase its packet counter and check if it is needed |
1318 | * to be synchronized | |
1319 | * | |
1320 | * Sync connection if it is about to close to | |
1321 | * encorage the standby servers to update the connections timeout | |
1322 | */ | |
1da177e4 | 1323 | atomic_inc(&cp->in_pkts); |
c6883f58 JV |
1324 | if (af == AF_INET && |
1325 | (ip_vs_sync_state & IP_VS_STATE_MASTER) && | |
efac5276 RB |
1326 | (((cp->protocol != IPPROTO_TCP || |
1327 | cp->state == IP_VS_TCP_S_ESTABLISHED) && | |
1328 | (atomic_read(&cp->in_pkts) % sysctl_ip_vs_sync_threshold[1] | |
1329 | == sysctl_ip_vs_sync_threshold[0])) || | |
1330 | ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) && | |
1331 | ((cp->state == IP_VS_TCP_S_FIN_WAIT) || | |
9d3a0de7 RB |
1332 | (cp->state == IP_VS_TCP_S_CLOSE_WAIT) || |
1333 | (cp->state == IP_VS_TCP_S_TIME_WAIT))))) | |
1da177e4 | 1334 | ip_vs_sync_conn(cp); |
efac5276 | 1335 | cp->old_state = cp->state; |
1da177e4 LT |
1336 | |
1337 | ip_vs_conn_put(cp); | |
1338 | return ret; | |
1339 | } | |
1340 | ||
1341 | ||
1342 | /* | |
6e23ae2a | 1343 | * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP |
1da177e4 LT |
1344 | * related packets destined for 0.0.0.0/0. |
1345 | * When fwmark-based virtual service is used, such as transparent | |
1346 | * cache cluster, TCP packets can be marked and routed to ip_vs_in, | |
1347 | * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and | |
6e23ae2a | 1348 | * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain |
1da177e4 LT |
1349 | * and send them to ip_vs_in_icmp. |
1350 | */ | |
1351 | static unsigned int | |
3db05fea | 1352 | ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb, |
1da177e4 LT |
1353 | const struct net_device *in, const struct net_device *out, |
1354 | int (*okfn)(struct sk_buff *)) | |
1355 | { | |
1356 | int r; | |
1357 | ||
3db05fea | 1358 | if (ip_hdr(skb)->protocol != IPPROTO_ICMP) |
1da177e4 LT |
1359 | return NF_ACCEPT; |
1360 | ||
3db05fea | 1361 | return ip_vs_in_icmp(skb, &r, hooknum); |
1da177e4 LT |
1362 | } |
1363 | ||
2a3b791e JV |
1364 | #ifdef CONFIG_IP_VS_IPV6 |
1365 | static unsigned int | |
1366 | ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb, | |
1367 | const struct net_device *in, const struct net_device *out, | |
1368 | int (*okfn)(struct sk_buff *)) | |
1369 | { | |
1370 | int r; | |
1371 | ||
1372 | if (ipv6_hdr(skb)->nexthdr != IPPROTO_ICMPV6) | |
1373 | return NF_ACCEPT; | |
1374 | ||
1375 | return ip_vs_in_icmp_v6(skb, &r, hooknum); | |
1376 | } | |
1377 | #endif | |
1378 | ||
1da177e4 | 1379 | |
1999414a | 1380 | static struct nf_hook_ops ip_vs_ops[] __read_mostly = { |
41c5b317 PM |
1381 | /* After packet filtering, forward packet through VS/DR, VS/TUN, |
1382 | * or VS/NAT(change destination), so that filtering rules can be | |
1383 | * applied to IPVS. */ | |
1384 | { | |
1385 | .hook = ip_vs_in, | |
1386 | .owner = THIS_MODULE, | |
1387 | .pf = PF_INET, | |
1388 | .hooknum = NF_INET_LOCAL_IN, | |
1389 | .priority = 100, | |
1390 | }, | |
1391 | /* After packet filtering, change source only for VS/NAT */ | |
1392 | { | |
1393 | .hook = ip_vs_out, | |
1394 | .owner = THIS_MODULE, | |
1395 | .pf = PF_INET, | |
1396 | .hooknum = NF_INET_FORWARD, | |
1397 | .priority = 100, | |
1398 | }, | |
1399 | /* After packet filtering (but before ip_vs_out_icmp), catch icmp | |
1400 | * destined for 0.0.0.0/0, which is for incoming IPVS connections */ | |
1401 | { | |
1402 | .hook = ip_vs_forward_icmp, | |
1403 | .owner = THIS_MODULE, | |
1404 | .pf = PF_INET, | |
1405 | .hooknum = NF_INET_FORWARD, | |
1406 | .priority = 99, | |
1407 | }, | |
1408 | /* Before the netfilter connection tracking, exit from POST_ROUTING */ | |
1409 | { | |
1410 | .hook = ip_vs_post_routing, | |
1411 | .owner = THIS_MODULE, | |
1412 | .pf = PF_INET, | |
1413 | .hooknum = NF_INET_POST_ROUTING, | |
1414 | .priority = NF_IP_PRI_NAT_SRC-1, | |
1415 | }, | |
1da177e4 LT |
1416 | }; |
1417 | ||
1418 | ||
1419 | /* | |
1420 | * Initialize IP Virtual Server | |
1421 | */ | |
1422 | static int __init ip_vs_init(void) | |
1423 | { | |
1424 | int ret; | |
1425 | ||
a919cf4b SW |
1426 | ip_vs_estimator_init(); |
1427 | ||
1da177e4 LT |
1428 | ret = ip_vs_control_init(); |
1429 | if (ret < 0) { | |
1430 | IP_VS_ERR("can't setup control.\n"); | |
a919cf4b | 1431 | goto cleanup_estimator; |
1da177e4 LT |
1432 | } |
1433 | ||
1434 | ip_vs_protocol_init(); | |
1435 | ||
1436 | ret = ip_vs_app_init(); | |
1437 | if (ret < 0) { | |
1438 | IP_VS_ERR("can't setup application helper.\n"); | |
1439 | goto cleanup_protocol; | |
1440 | } | |
1441 | ||
1442 | ret = ip_vs_conn_init(); | |
1443 | if (ret < 0) { | |
1444 | IP_VS_ERR("can't setup connection table.\n"); | |
1445 | goto cleanup_app; | |
1446 | } | |
1447 | ||
41c5b317 | 1448 | ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops)); |
1da177e4 | 1449 | if (ret < 0) { |
41c5b317 | 1450 | IP_VS_ERR("can't register hooks.\n"); |
1da177e4 LT |
1451 | goto cleanup_conn; |
1452 | } | |
1453 | ||
1da177e4 LT |
1454 | IP_VS_INFO("ipvs loaded.\n"); |
1455 | return ret; | |
1456 | ||
1da177e4 LT |
1457 | cleanup_conn: |
1458 | ip_vs_conn_cleanup(); | |
1459 | cleanup_app: | |
1460 | ip_vs_app_cleanup(); | |
1461 | cleanup_protocol: | |
1462 | ip_vs_protocol_cleanup(); | |
1463 | ip_vs_control_cleanup(); | |
a919cf4b SW |
1464 | cleanup_estimator: |
1465 | ip_vs_estimator_cleanup(); | |
1da177e4 LT |
1466 | return ret; |
1467 | } | |
1468 | ||
1469 | static void __exit ip_vs_cleanup(void) | |
1470 | { | |
41c5b317 | 1471 | nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops)); |
1da177e4 LT |
1472 | ip_vs_conn_cleanup(); |
1473 | ip_vs_app_cleanup(); | |
1474 | ip_vs_protocol_cleanup(); | |
1475 | ip_vs_control_cleanup(); | |
a919cf4b | 1476 | ip_vs_estimator_cleanup(); |
1da177e4 LT |
1477 | IP_VS_INFO("ipvs unloaded.\n"); |
1478 | } | |
1479 | ||
1480 | module_init(ip_vs_init); | |
1481 | module_exit(ip_vs_cleanup); | |
1482 | MODULE_LICENSE("GPL"); |