]>
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 | ||
9aada7ac HE |
27 | #define KMSG_COMPONENT "IPVS" |
28 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt | |
29 | ||
1da177e4 LT |
30 | #include <linux/module.h> |
31 | #include <linux/kernel.h> | |
32 | #include <linux/ip.h> | |
33 | #include <linux/tcp.h> | |
2906f66a | 34 | #include <linux/sctp.h> |
1da177e4 | 35 | #include <linux/icmp.h> |
5a0e3ad6 | 36 | #include <linux/slab.h> |
1da177e4 LT |
37 | |
38 | #include <net/ip.h> | |
39 | #include <net/tcp.h> | |
40 | #include <net/udp.h> | |
41 | #include <net/icmp.h> /* for icmp_send */ | |
42 | #include <net/route.h> | |
2c70b519 | 43 | #include <net/ip6_checksum.h> |
1da177e4 LT |
44 | |
45 | #include <linux/netfilter.h> | |
46 | #include <linux/netfilter_ipv4.h> | |
47 | ||
2a3b791e JV |
48 | #ifdef CONFIG_IP_VS_IPV6 |
49 | #include <net/ipv6.h> | |
50 | #include <linux/netfilter_ipv6.h> | |
489fdeda | 51 | #include <net/ip6_route.h> |
2a3b791e JV |
52 | #endif |
53 | ||
1da177e4 LT |
54 | #include <net/ip_vs.h> |
55 | ||
56 | ||
57 | EXPORT_SYMBOL(register_ip_vs_scheduler); | |
58 | EXPORT_SYMBOL(unregister_ip_vs_scheduler); | |
1da177e4 LT |
59 | EXPORT_SYMBOL(ip_vs_proto_name); |
60 | EXPORT_SYMBOL(ip_vs_conn_new); | |
61 | EXPORT_SYMBOL(ip_vs_conn_in_get); | |
62 | EXPORT_SYMBOL(ip_vs_conn_out_get); | |
63 | #ifdef CONFIG_IP_VS_PROTO_TCP | |
64 | EXPORT_SYMBOL(ip_vs_tcp_conn_listen); | |
65 | #endif | |
66 | EXPORT_SYMBOL(ip_vs_conn_put); | |
67 | #ifdef CONFIG_IP_VS_DEBUG | |
68 | EXPORT_SYMBOL(ip_vs_get_debug_level); | |
69 | #endif | |
1da177e4 LT |
70 | |
71 | ||
72 | /* ID used in ICMP lookups */ | |
73 | #define icmp_id(icmph) (((icmph)->un).echo.id) | |
2a3b791e | 74 | #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier) |
1da177e4 LT |
75 | |
76 | const char *ip_vs_proto_name(unsigned proto) | |
77 | { | |
78 | static char buf[20]; | |
79 | ||
80 | switch (proto) { | |
81 | case IPPROTO_IP: | |
82 | return "IP"; | |
83 | case IPPROTO_UDP: | |
84 | return "UDP"; | |
85 | case IPPROTO_TCP: | |
86 | return "TCP"; | |
2906f66a VMR |
87 | case IPPROTO_SCTP: |
88 | return "SCTP"; | |
1da177e4 LT |
89 | case IPPROTO_ICMP: |
90 | return "ICMP"; | |
2a3b791e JV |
91 | #ifdef CONFIG_IP_VS_IPV6 |
92 | case IPPROTO_ICMPV6: | |
93 | return "ICMPv6"; | |
94 | #endif | |
1da177e4 LT |
95 | default: |
96 | sprintf(buf, "IP_%d", proto); | |
97 | return buf; | |
98 | } | |
99 | } | |
100 | ||
101 | void ip_vs_init_hash_table(struct list_head *table, int rows) | |
102 | { | |
103 | while (--rows >= 0) | |
104 | INIT_LIST_HEAD(&table[rows]); | |
105 | } | |
106 | ||
107 | static inline void | |
108 | ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb) | |
109 | { | |
110 | struct ip_vs_dest *dest = cp->dest; | |
111 | if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) { | |
112 | spin_lock(&dest->stats.lock); | |
e9c0ce23 SW |
113 | dest->stats.ustats.inpkts++; |
114 | dest->stats.ustats.inbytes += skb->len; | |
1da177e4 LT |
115 | spin_unlock(&dest->stats.lock); |
116 | ||
117 | spin_lock(&dest->svc->stats.lock); | |
e9c0ce23 SW |
118 | dest->svc->stats.ustats.inpkts++; |
119 | dest->svc->stats.ustats.inbytes += skb->len; | |
1da177e4 LT |
120 | spin_unlock(&dest->svc->stats.lock); |
121 | ||
122 | spin_lock(&ip_vs_stats.lock); | |
e9c0ce23 SW |
123 | ip_vs_stats.ustats.inpkts++; |
124 | ip_vs_stats.ustats.inbytes += skb->len; | |
1da177e4 LT |
125 | spin_unlock(&ip_vs_stats.lock); |
126 | } | |
127 | } | |
128 | ||
129 | ||
130 | static inline void | |
131 | ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb) | |
132 | { | |
133 | struct ip_vs_dest *dest = cp->dest; | |
134 | if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) { | |
135 | spin_lock(&dest->stats.lock); | |
e9c0ce23 SW |
136 | dest->stats.ustats.outpkts++; |
137 | dest->stats.ustats.outbytes += skb->len; | |
1da177e4 LT |
138 | spin_unlock(&dest->stats.lock); |
139 | ||
140 | spin_lock(&dest->svc->stats.lock); | |
e9c0ce23 SW |
141 | dest->svc->stats.ustats.outpkts++; |
142 | dest->svc->stats.ustats.outbytes += skb->len; | |
1da177e4 LT |
143 | spin_unlock(&dest->svc->stats.lock); |
144 | ||
145 | spin_lock(&ip_vs_stats.lock); | |
e9c0ce23 SW |
146 | ip_vs_stats.ustats.outpkts++; |
147 | ip_vs_stats.ustats.outbytes += skb->len; | |
1da177e4 LT |
148 | spin_unlock(&ip_vs_stats.lock); |
149 | } | |
150 | } | |
151 | ||
152 | ||
153 | static inline void | |
154 | ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc) | |
155 | { | |
156 | spin_lock(&cp->dest->stats.lock); | |
e9c0ce23 | 157 | cp->dest->stats.ustats.conns++; |
1da177e4 LT |
158 | spin_unlock(&cp->dest->stats.lock); |
159 | ||
160 | spin_lock(&svc->stats.lock); | |
e9c0ce23 | 161 | svc->stats.ustats.conns++; |
1da177e4 LT |
162 | spin_unlock(&svc->stats.lock); |
163 | ||
164 | spin_lock(&ip_vs_stats.lock); | |
e9c0ce23 | 165 | ip_vs_stats.ustats.conns++; |
1da177e4 LT |
166 | spin_unlock(&ip_vs_stats.lock); |
167 | } | |
168 | ||
169 | ||
170 | static inline int | |
171 | ip_vs_set_state(struct ip_vs_conn *cp, int direction, | |
172 | const struct sk_buff *skb, | |
173 | struct ip_vs_protocol *pp) | |
174 | { | |
175 | if (unlikely(!pp->state_transition)) | |
176 | return 0; | |
177 | return pp->state_transition(cp, direction, skb, pp); | |
178 | } | |
179 | ||
f71499aa | 180 | static inline void |
85999283 SH |
181 | ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc, |
182 | struct sk_buff *skb, int protocol, | |
183 | const union nf_inet_addr *caddr, __be16 cport, | |
184 | const union nf_inet_addr *vaddr, __be16 vport, | |
185 | struct ip_vs_conn_param *p) | |
186 | { | |
187 | ip_vs_conn_fill_param(svc->af, protocol, caddr, cport, vaddr, vport, p); | |
188 | p->pe = svc->pe; | |
189 | if (p->pe && p->pe->fill_param) | |
f71499aa | 190 | p->pe->fill_param(p, skb); |
85999283 | 191 | } |
1da177e4 | 192 | |
1da177e4 LT |
193 | /* |
194 | * IPVS persistent scheduling function | |
195 | * It creates a connection entry according to its template if exists, | |
196 | * or selects a server and creates a connection entry plus a template. | |
197 | * Locking: we are svc user (svc->refcnt), so we hold all dests too | |
198 | * Protocols supported: TCP, UDP | |
199 | */ | |
200 | static struct ip_vs_conn * | |
201 | ip_vs_sched_persist(struct ip_vs_service *svc, | |
85999283 | 202 | struct sk_buff *skb, |
014d730d | 203 | __be16 ports[2]) |
1da177e4 LT |
204 | { |
205 | struct ip_vs_conn *cp = NULL; | |
28364a59 | 206 | struct ip_vs_iphdr iph; |
1da177e4 LT |
207 | struct ip_vs_dest *dest; |
208 | struct ip_vs_conn *ct; | |
5b57a98c | 209 | __be16 dport = 0; /* destination port to forward */ |
3575792e | 210 | unsigned int flags; |
f11017ec | 211 | struct ip_vs_conn_param param; |
28364a59 JV |
212 | union nf_inet_addr snet; /* source network of the client, |
213 | after masking */ | |
cd17f9ed JV |
214 | |
215 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); | |
1da177e4 LT |
216 | |
217 | /* Mask saddr with the netmask to adjust template granularity */ | |
cd17f9ed JV |
218 | #ifdef CONFIG_IP_VS_IPV6 |
219 | if (svc->af == AF_INET6) | |
220 | ipv6_addr_prefix(&snet.in6, &iph.saddr.in6, svc->netmask); | |
221 | else | |
222 | #endif | |
223 | snet.ip = iph.saddr.ip & svc->netmask; | |
1da177e4 | 224 | |
cd17f9ed JV |
225 | IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u " |
226 | "mnet %s\n", | |
227 | IP_VS_DBG_ADDR(svc->af, &iph.saddr), ntohs(ports[0]), | |
228 | IP_VS_DBG_ADDR(svc->af, &iph.daddr), ntohs(ports[1]), | |
229 | IP_VS_DBG_ADDR(svc->af, &snet)); | |
1da177e4 LT |
230 | |
231 | /* | |
232 | * As far as we know, FTP is a very complicated network protocol, and | |
233 | * it uses control connection and data connections. For active FTP, | |
234 | * FTP server initialize data connection to the client, its source port | |
235 | * is often 20. For passive FTP, FTP server tells the clients the port | |
236 | * that it passively listens to, and the client issues the data | |
237 | * connection. In the tunneling or direct routing mode, the load | |
238 | * balancer is on the client-to-server half of connection, the port | |
239 | * number is unknown to the load balancer. So, a conn template like | |
240 | * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP | |
241 | * service, and a template like <caddr, 0, vaddr, vport, daddr, dport> | |
242 | * is created for other persistent services. | |
243 | */ | |
5b57a98c | 244 | { |
f11017ec SH |
245 | int protocol = iph.protocol; |
246 | const union nf_inet_addr *vaddr = &iph.daddr; | |
247 | const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) }; | |
248 | __be16 vport = 0; | |
249 | ||
5b57a98c SH |
250 | if (ports[1] == svc->port) { |
251 | /* non-FTP template: | |
252 | * <protocol, caddr, 0, vaddr, vport, daddr, dport> | |
253 | * FTP template: | |
254 | * <protocol, caddr, 0, vaddr, 0, daddr, 0> | |
1da177e4 LT |
255 | */ |
256 | if (svc->port != FTPPORT) | |
5b57a98c | 257 | vport = ports[1]; |
1da177e4 | 258 | } else { |
5b57a98c SH |
259 | /* Note: persistent fwmark-based services and |
260 | * persistent port zero service are handled here. | |
261 | * fwmark template: | |
262 | * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0> | |
263 | * port zero template: | |
264 | * <protocol,caddr,0,vaddr,0,daddr,0> | |
1da177e4 | 265 | */ |
28364a59 | 266 | if (svc->fwmark) { |
5b57a98c SH |
267 | protocol = IPPROTO_IP; |
268 | vaddr = &fwmark; | |
269 | } | |
1da177e4 | 270 | } |
f71499aa SH |
271 | ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0, |
272 | vaddr, vport, ¶m); | |
1da177e4 LT |
273 | } |
274 | ||
5b57a98c | 275 | /* Check if a template already exists */ |
f11017ec | 276 | ct = ip_vs_ct_in_get(¶m); |
5b57a98c SH |
277 | if (!ct || !ip_vs_check_template(ct)) { |
278 | /* No template found or the dest of the connection | |
279 | * template is not available. | |
280 | */ | |
281 | dest = svc->scheduler->schedule(svc, skb); | |
282 | if (!dest) { | |
283 | IP_VS_DBG(1, "p-schedule: no dest found.\n"); | |
85999283 | 284 | kfree(param.pe_data); |
5b57a98c SH |
285 | return NULL; |
286 | } | |
287 | ||
288 | if (ports[1] == svc->port && svc->port != FTPPORT) | |
289 | dport = dest->port; | |
290 | ||
85999283 SH |
291 | /* Create a template |
292 | * This adds param.pe_data to the template, | |
293 | * and thus param.pe_data will be destroyed | |
294 | * when the template expires */ | |
f11017ec | 295 | ct = ip_vs_conn_new(¶m, &dest->addr, dport, |
5b57a98c | 296 | IP_VS_CONN_F_TEMPLATE, dest); |
85999283 SH |
297 | if (ct == NULL) { |
298 | kfree(param.pe_data); | |
5b57a98c | 299 | return NULL; |
85999283 | 300 | } |
5b57a98c SH |
301 | |
302 | ct->timeout = svc->timeout; | |
85999283 | 303 | } else { |
5b57a98c SH |
304 | /* set destination with the found template */ |
305 | dest = ct->dest; | |
85999283 SH |
306 | kfree(param.pe_data); |
307 | } | |
5b57a98c SH |
308 | |
309 | dport = ports[1]; | |
310 | if (dport == svc->port && dest->port) | |
311 | dport = dest->port; | |
312 | ||
26ec037f NC |
313 | flags = (svc->flags & IP_VS_SVC_F_ONEPACKET |
314 | && iph.protocol == IPPROTO_UDP)? | |
315 | IP_VS_CONN_F_ONE_PACKET : 0; | |
316 | ||
1da177e4 LT |
317 | /* |
318 | * Create a new connection according to the template | |
319 | */ | |
f11017ec SH |
320 | ip_vs_conn_fill_param(svc->af, iph.protocol, &iph.saddr, ports[0], |
321 | &iph.daddr, ports[1], ¶m); | |
322 | cp = ip_vs_conn_new(¶m, &dest->addr, dport, flags, dest); | |
1da177e4 LT |
323 | if (cp == NULL) { |
324 | ip_vs_conn_put(ct); | |
325 | return NULL; | |
326 | } | |
327 | ||
328 | /* | |
329 | * Add its control | |
330 | */ | |
331 | ip_vs_control_add(cp, ct); | |
332 | ip_vs_conn_put(ct); | |
333 | ||
334 | ip_vs_conn_stats(cp, svc); | |
335 | return cp; | |
336 | } | |
337 | ||
338 | ||
339 | /* | |
340 | * IPVS main scheduling function | |
341 | * It selects a server according to the virtual service, and | |
342 | * creates a connection entry. | |
343 | * Protocols supported: TCP, UDP | |
344 | */ | |
345 | struct ip_vs_conn * | |
190ecd27 JA |
346 | ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb, |
347 | struct ip_vs_protocol *pp, int *ignored) | |
1da177e4 LT |
348 | { |
349 | struct ip_vs_conn *cp = NULL; | |
28364a59 | 350 | struct ip_vs_iphdr iph; |
1da177e4 | 351 | struct ip_vs_dest *dest; |
3575792e JA |
352 | __be16 _ports[2], *pptr; |
353 | unsigned int flags; | |
1da177e4 | 354 | |
190ecd27 | 355 | *ignored = 1; |
28364a59 JV |
356 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); |
357 | pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports); | |
1da177e4 LT |
358 | if (pptr == NULL) |
359 | return NULL; | |
360 | ||
190ecd27 JA |
361 | /* |
362 | * FTPDATA needs this check when using local real server. | |
363 | * Never schedule Active FTPDATA connections from real server. | |
364 | * For LVS-NAT they must be already created. For other methods | |
365 | * with persistence the connection is created on SYN+ACK. | |
366 | */ | |
367 | if (pptr[0] == FTPDATA) { | |
368 | IP_VS_DBG_PKT(12, pp, skb, 0, "Not scheduling FTPDATA"); | |
369 | return NULL; | |
370 | } | |
371 | ||
372 | /* | |
373 | * Do not schedule replies from local real server. It is risky | |
374 | * for fwmark services but mostly for persistent services. | |
375 | */ | |
376 | if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) && | |
377 | (svc->flags & IP_VS_SVC_F_PERSISTENT || svc->fwmark) && | |
378 | (cp = pp->conn_in_get(svc->af, skb, pp, &iph, iph.len, 1))) { | |
379 | IP_VS_DBG_PKT(12, pp, skb, 0, | |
380 | "Not scheduling reply for existing connection"); | |
381 | __ip_vs_conn_put(cp); | |
382 | return NULL; | |
383 | } | |
384 | ||
1da177e4 LT |
385 | /* |
386 | * Persistent service | |
387 | */ | |
190ecd27 JA |
388 | if (svc->flags & IP_VS_SVC_F_PERSISTENT) { |
389 | *ignored = 0; | |
1da177e4 | 390 | return ip_vs_sched_persist(svc, skb, pptr); |
190ecd27 | 391 | } |
1da177e4 LT |
392 | |
393 | /* | |
394 | * Non-persistent service | |
395 | */ | |
396 | if (!svc->fwmark && pptr[1] != svc->port) { | |
397 | if (!svc->port) | |
1e3e238e HE |
398 | pr_err("Schedule: port zero only supported " |
399 | "in persistent services, " | |
400 | "check your ipvs configuration\n"); | |
1da177e4 LT |
401 | return NULL; |
402 | } | |
403 | ||
190ecd27 JA |
404 | *ignored = 0; |
405 | ||
1da177e4 LT |
406 | dest = svc->scheduler->schedule(svc, skb); |
407 | if (dest == NULL) { | |
408 | IP_VS_DBG(1, "Schedule: no dest found.\n"); | |
409 | return NULL; | |
410 | } | |
411 | ||
26ec037f NC |
412 | flags = (svc->flags & IP_VS_SVC_F_ONEPACKET |
413 | && iph.protocol == IPPROTO_UDP)? | |
414 | IP_VS_CONN_F_ONE_PACKET : 0; | |
415 | ||
1da177e4 LT |
416 | /* |
417 | * Create a connection entry. | |
418 | */ | |
f11017ec SH |
419 | { |
420 | struct ip_vs_conn_param p; | |
421 | ip_vs_conn_fill_param(svc->af, iph.protocol, &iph.saddr, | |
422 | pptr[0], &iph.daddr, pptr[1], &p); | |
423 | cp = ip_vs_conn_new(&p, &dest->addr, | |
424 | dest->port ? dest->port : pptr[1], | |
425 | flags, dest); | |
426 | if (!cp) | |
427 | return NULL; | |
428 | } | |
1da177e4 | 429 | |
cd17f9ed JV |
430 | IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u " |
431 | "d:%s:%u conn->flags:%X conn->refcnt:%d\n", | |
432 | ip_vs_fwd_tag(cp), | |
433 | IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport), | |
434 | IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport), | |
435 | IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport), | |
436 | cp->flags, atomic_read(&cp->refcnt)); | |
1da177e4 LT |
437 | |
438 | ip_vs_conn_stats(cp, svc); | |
439 | return cp; | |
440 | } | |
441 | ||
442 | ||
443 | /* | |
444 | * Pass or drop the packet. | |
445 | * Called by ip_vs_in, when the virtual service is available but | |
446 | * no destination is available for a new connection. | |
447 | */ | |
448 | int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb, | |
449 | struct ip_vs_protocol *pp) | |
450 | { | |
014d730d | 451 | __be16 _ports[2], *pptr; |
28364a59 | 452 | struct ip_vs_iphdr iph; |
2a3b791e JV |
453 | int unicast; |
454 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); | |
1da177e4 | 455 | |
28364a59 | 456 | pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports); |
1da177e4 LT |
457 | if (pptr == NULL) { |
458 | ip_vs_service_put(svc); | |
459 | return NF_DROP; | |
460 | } | |
461 | ||
2a3b791e JV |
462 | #ifdef CONFIG_IP_VS_IPV6 |
463 | if (svc->af == AF_INET6) | |
464 | unicast = ipv6_addr_type(&iph.daddr.in6) & IPV6_ADDR_UNICAST; | |
465 | else | |
466 | #endif | |
467 | unicast = (inet_addr_type(&init_net, iph.daddr.ip) == RTN_UNICAST); | |
468 | ||
1da177e4 | 469 | /* if it is fwmark-based service, the cache_bypass sysctl is up |
2a3b791e | 470 | and the destination is a non-local unicast, then create |
1da177e4 | 471 | a cache_bypass connection entry */ |
2a3b791e | 472 | if (sysctl_ip_vs_cache_bypass && svc->fwmark && unicast) { |
1da177e4 LT |
473 | int ret, cs; |
474 | struct ip_vs_conn *cp; | |
3575792e JA |
475 | unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET && |
476 | iph.protocol == IPPROTO_UDP)? | |
477 | IP_VS_CONN_F_ONE_PACKET : 0; | |
dff630dd | 478 | union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } }; |
1da177e4 LT |
479 | |
480 | ip_vs_service_put(svc); | |
481 | ||
482 | /* create a new connection entry */ | |
1e3e238e | 483 | IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__); |
f11017ec SH |
484 | { |
485 | struct ip_vs_conn_param p; | |
486 | ip_vs_conn_fill_param(svc->af, iph.protocol, | |
487 | &iph.saddr, pptr[0], | |
488 | &iph.daddr, pptr[1], &p); | |
489 | cp = ip_vs_conn_new(&p, &daddr, 0, | |
490 | IP_VS_CONN_F_BYPASS | flags, | |
491 | NULL); | |
492 | if (!cp) | |
493 | return NF_DROP; | |
494 | } | |
1da177e4 LT |
495 | |
496 | /* statistics */ | |
497 | ip_vs_in_stats(cp, skb); | |
498 | ||
499 | /* set state */ | |
500 | cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp); | |
501 | ||
502 | /* transmit the first SYN packet */ | |
503 | ret = cp->packet_xmit(skb, cp, pp); | |
504 | /* do not touch skb anymore */ | |
505 | ||
506 | atomic_inc(&cp->in_pkts); | |
507 | ip_vs_conn_put(cp); | |
508 | return ret; | |
509 | } | |
510 | ||
511 | /* | |
512 | * When the virtual ftp service is presented, packets destined | |
513 | * for other services on the VIP may get here (except services | |
514 | * listed in the ipvs table), pass the packets, because it is | |
515 | * not ipvs job to decide to drop the packets. | |
516 | */ | |
517 | if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) { | |
518 | ip_vs_service_put(svc); | |
519 | return NF_ACCEPT; | |
520 | } | |
521 | ||
522 | ip_vs_service_put(svc); | |
523 | ||
524 | /* | |
525 | * Notify the client that the destination is unreachable, and | |
526 | * release the socket buffer. | |
527 | * Since it is in IP layer, the TCP socket is not actually | |
528 | * created, the TCP RST packet cannot be sent, instead that | |
529 | * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ | |
530 | */ | |
2a3b791e JV |
531 | #ifdef CONFIG_IP_VS_IPV6 |
532 | if (svc->af == AF_INET6) | |
3ffe533c | 533 | icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); |
2a3b791e JV |
534 | else |
535 | #endif | |
536 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); | |
537 | ||
1da177e4 LT |
538 | return NF_DROP; |
539 | } | |
540 | ||
b1550f22 | 541 | __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset) |
1da177e4 | 542 | { |
d3bc23e7 | 543 | return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0)); |
1da177e4 LT |
544 | } |
545 | ||
1ca5bb54 JA |
546 | static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum) |
547 | { | |
548 | if (NF_INET_LOCAL_IN == hooknum) | |
549 | return IP_DEFRAG_VS_IN; | |
550 | if (NF_INET_FORWARD == hooknum) | |
551 | return IP_DEFRAG_VS_FWD; | |
552 | return IP_DEFRAG_VS_OUT; | |
553 | } | |
554 | ||
776c729e | 555 | static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user) |
1da177e4 | 556 | { |
776c729e HX |
557 | int err = ip_defrag(skb, user); |
558 | ||
559 | if (!err) | |
eddc9ec5 | 560 | ip_send_check(ip_hdr(skb)); |
776c729e HX |
561 | |
562 | return err; | |
1da177e4 LT |
563 | } |
564 | ||
2a3b791e JV |
565 | #ifdef CONFIG_IP_VS_IPV6 |
566 | static inline int ip_vs_gather_frags_v6(struct sk_buff *skb, u_int32_t user) | |
567 | { | |
568 | /* TODO IPv6: Find out what to do here for IPv6 */ | |
569 | return 0; | |
570 | } | |
571 | #endif | |
572 | ||
1da177e4 LT |
573 | /* |
574 | * Packet has been made sufficiently writable in caller | |
575 | * - inout: 1=in->out, 0=out->in | |
576 | */ | |
577 | void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp, | |
578 | struct ip_vs_conn *cp, int inout) | |
579 | { | |
eddc9ec5 | 580 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 | 581 | unsigned int icmp_offset = iph->ihl*4; |
d56f90a7 ACM |
582 | struct icmphdr *icmph = (struct icmphdr *)(skb_network_header(skb) + |
583 | icmp_offset); | |
1da177e4 LT |
584 | struct iphdr *ciph = (struct iphdr *)(icmph + 1); |
585 | ||
586 | if (inout) { | |
e7ade46a | 587 | iph->saddr = cp->vaddr.ip; |
1da177e4 | 588 | ip_send_check(iph); |
e7ade46a | 589 | ciph->daddr = cp->vaddr.ip; |
1da177e4 LT |
590 | ip_send_check(ciph); |
591 | } else { | |
e7ade46a | 592 | iph->daddr = cp->daddr.ip; |
1da177e4 | 593 | ip_send_check(iph); |
e7ade46a | 594 | ciph->saddr = cp->daddr.ip; |
1da177e4 LT |
595 | ip_send_check(ciph); |
596 | } | |
597 | ||
2906f66a VMR |
598 | /* the TCP/UDP/SCTP port */ |
599 | if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol || | |
600 | IPPROTO_SCTP == ciph->protocol) { | |
014d730d | 601 | __be16 *ports = (void *)ciph + ciph->ihl*4; |
1da177e4 LT |
602 | |
603 | if (inout) | |
604 | ports[1] = cp->vport; | |
605 | else | |
606 | ports[0] = cp->dport; | |
607 | } | |
608 | ||
609 | /* And finally the ICMP checksum */ | |
610 | icmph->checksum = 0; | |
611 | icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset); | |
612 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
613 | ||
614 | if (inout) | |
615 | IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph, | |
616 | "Forwarding altered outgoing ICMP"); | |
617 | else | |
618 | IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph, | |
619 | "Forwarding altered incoming ICMP"); | |
620 | } | |
621 | ||
b3cdd2a7 JV |
622 | #ifdef CONFIG_IP_VS_IPV6 |
623 | void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp, | |
624 | struct ip_vs_conn *cp, int inout) | |
625 | { | |
626 | struct ipv6hdr *iph = ipv6_hdr(skb); | |
627 | unsigned int icmp_offset = sizeof(struct ipv6hdr); | |
628 | struct icmp6hdr *icmph = (struct icmp6hdr *)(skb_network_header(skb) + | |
629 | icmp_offset); | |
630 | struct ipv6hdr *ciph = (struct ipv6hdr *)(icmph + 1); | |
631 | ||
632 | if (inout) { | |
633 | iph->saddr = cp->vaddr.in6; | |
634 | ciph->daddr = cp->vaddr.in6; | |
635 | } else { | |
636 | iph->daddr = cp->daddr.in6; | |
637 | ciph->saddr = cp->daddr.in6; | |
638 | } | |
639 | ||
2906f66a VMR |
640 | /* the TCP/UDP/SCTP port */ |
641 | if (IPPROTO_TCP == ciph->nexthdr || IPPROTO_UDP == ciph->nexthdr || | |
642 | IPPROTO_SCTP == ciph->nexthdr) { | |
b3cdd2a7 JV |
643 | __be16 *ports = (void *)ciph + sizeof(struct ipv6hdr); |
644 | ||
645 | if (inout) | |
646 | ports[1] = cp->vport; | |
647 | else | |
648 | ports[0] = cp->dport; | |
649 | } | |
650 | ||
651 | /* And finally the ICMP checksum */ | |
8870f842 SH |
652 | icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, |
653 | skb->len - icmp_offset, | |
654 | IPPROTO_ICMPV6, 0); | |
655 | skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset; | |
656 | skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum); | |
657 | skb->ip_summed = CHECKSUM_PARTIAL; | |
b3cdd2a7 JV |
658 | |
659 | if (inout) | |
660 | IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph, | |
661 | "Forwarding altered outgoing ICMPv6"); | |
662 | else | |
663 | IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph, | |
664 | "Forwarding altered incoming ICMPv6"); | |
665 | } | |
666 | #endif | |
667 | ||
4856c84c MT |
668 | /* Handle relevant response ICMP messages - forward to the right |
669 | * destination host. Used for NAT and local client. | |
670 | */ | |
f2428ed5 SH |
671 | static int handle_response_icmp(int af, struct sk_buff *skb, |
672 | union nf_inet_addr *snet, | |
673 | __u8 protocol, struct ip_vs_conn *cp, | |
4856c84c MT |
674 | struct ip_vs_protocol *pp, |
675 | unsigned int offset, unsigned int ihl) | |
676 | { | |
677 | unsigned int verdict = NF_DROP; | |
678 | ||
679 | if (IP_VS_FWD_METHOD(cp) != 0) { | |
1e3e238e HE |
680 | pr_err("shouldn't reach here, because the box is on the " |
681 | "half connection in the tun/dr module.\n"); | |
4856c84c MT |
682 | } |
683 | ||
684 | /* Ensure the checksum is correct */ | |
685 | if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) { | |
686 | /* Failed checksum! */ | |
f2428ed5 SH |
687 | IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n", |
688 | IP_VS_DBG_ADDR(af, snet)); | |
4856c84c MT |
689 | goto out; |
690 | } | |
691 | ||
2906f66a VMR |
692 | if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol || |
693 | IPPROTO_SCTP == protocol) | |
4856c84c MT |
694 | offset += 2 * sizeof(__u16); |
695 | if (!skb_make_writable(skb, offset)) | |
696 | goto out; | |
697 | ||
f2428ed5 SH |
698 | #ifdef CONFIG_IP_VS_IPV6 |
699 | if (af == AF_INET6) | |
700 | ip_vs_nat_icmp_v6(skb, pp, cp, 1); | |
701 | else | |
702 | #endif | |
703 | ip_vs_nat_icmp(skb, pp, cp, 1); | |
4856c84c | 704 | |
f5a41847 JA |
705 | #ifdef CONFIG_IP_VS_IPV6 |
706 | if (af == AF_INET6) { | |
707 | if (sysctl_ip_vs_snat_reroute && ip6_route_me_harder(skb) != 0) | |
708 | goto out; | |
709 | } else | |
710 | #endif | |
711 | if ((sysctl_ip_vs_snat_reroute || | |
712 | skb_rtable(skb)->rt_flags & RTCF_LOCAL) && | |
713 | ip_route_me_harder(skb, RTN_LOCAL) != 0) | |
714 | goto out; | |
715 | ||
4856c84c MT |
716 | /* do the statistics and put it back */ |
717 | ip_vs_out_stats(cp, skb); | |
718 | ||
cf356d69 | 719 | skb->ipvs_property = 1; |
f4bc17cd | 720 | if (!(cp->flags & IP_VS_CONN_F_NFCT)) |
cf356d69 | 721 | ip_vs_notrack(skb); |
f4bc17cd JA |
722 | else |
723 | ip_vs_update_conntrack(skb, cp, 0); | |
4856c84c MT |
724 | verdict = NF_ACCEPT; |
725 | ||
726 | out: | |
727 | __ip_vs_conn_put(cp); | |
728 | ||
729 | return verdict; | |
730 | } | |
731 | ||
1da177e4 LT |
732 | /* |
733 | * Handle ICMP messages in the inside-to-outside direction (outgoing). | |
4856c84c | 734 | * Find any that might be relevant, check against existing connections. |
1da177e4 | 735 | * Currently handles error types - unreachable, quench, ttl exceeded. |
1da177e4 | 736 | */ |
1ca5bb54 JA |
737 | static int ip_vs_out_icmp(struct sk_buff *skb, int *related, |
738 | unsigned int hooknum) | |
1da177e4 | 739 | { |
1da177e4 LT |
740 | struct iphdr *iph; |
741 | struct icmphdr _icmph, *ic; | |
742 | struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */ | |
51ef348b | 743 | struct ip_vs_iphdr ciph; |
1da177e4 LT |
744 | struct ip_vs_conn *cp; |
745 | struct ip_vs_protocol *pp; | |
4856c84c | 746 | unsigned int offset, ihl; |
f2428ed5 | 747 | union nf_inet_addr snet; |
1da177e4 LT |
748 | |
749 | *related = 1; | |
750 | ||
751 | /* reassemble IP fragments */ | |
eddc9ec5 | 752 | if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) { |
1ca5bb54 | 753 | if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum))) |
1da177e4 | 754 | return NF_STOLEN; |
1da177e4 LT |
755 | } |
756 | ||
eddc9ec5 | 757 | iph = ip_hdr(skb); |
1da177e4 LT |
758 | offset = ihl = iph->ihl * 4; |
759 | ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); | |
760 | if (ic == NULL) | |
761 | return NF_DROP; | |
762 | ||
14d5e834 | 763 | IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n", |
1da177e4 | 764 | ic->type, ntohs(icmp_id(ic)), |
14d5e834 | 765 | &iph->saddr, &iph->daddr); |
1da177e4 LT |
766 | |
767 | /* | |
768 | * Work through seeing if this is for us. | |
769 | * These checks are supposed to be in an order that means easy | |
770 | * things are checked first to speed up processing.... however | |
771 | * this means that some packets will manage to get a long way | |
772 | * down this stack and then be rejected, but that's life. | |
773 | */ | |
774 | if ((ic->type != ICMP_DEST_UNREACH) && | |
775 | (ic->type != ICMP_SOURCE_QUENCH) && | |
776 | (ic->type != ICMP_TIME_EXCEEDED)) { | |
777 | *related = 0; | |
778 | return NF_ACCEPT; | |
779 | } | |
780 | ||
781 | /* Now find the contained IP header */ | |
782 | offset += sizeof(_icmph); | |
783 | cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); | |
784 | if (cih == NULL) | |
785 | return NF_ACCEPT; /* The packet looks wrong, ignore */ | |
786 | ||
787 | pp = ip_vs_proto_get(cih->protocol); | |
788 | if (!pp) | |
789 | return NF_ACCEPT; | |
790 | ||
791 | /* Is the embedded protocol header present? */ | |
4412ec49 | 792 | if (unlikely(cih->frag_off & htons(IP_OFFSET) && |
1da177e4 LT |
793 | pp->dont_defrag)) |
794 | return NF_ACCEPT; | |
795 | ||
796 | IP_VS_DBG_PKT(11, pp, skb, offset, "Checking outgoing ICMP for"); | |
797 | ||
798 | offset += cih->ihl * 4; | |
799 | ||
51ef348b | 800 | ip_vs_fill_iphdr(AF_INET, cih, &ciph); |
1da177e4 | 801 | /* The embedded headers contain source and dest in reverse order */ |
51ef348b | 802 | cp = pp->conn_out_get(AF_INET, skb, pp, &ciph, offset, 1); |
1da177e4 LT |
803 | if (!cp) |
804 | return NF_ACCEPT; | |
805 | ||
f2428ed5 SH |
806 | snet.ip = iph->saddr; |
807 | return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp, | |
808 | pp, offset, ihl); | |
1da177e4 LT |
809 | } |
810 | ||
2a3b791e | 811 | #ifdef CONFIG_IP_VS_IPV6 |
1ca5bb54 JA |
812 | static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related, |
813 | unsigned int hooknum) | |
2a3b791e JV |
814 | { |
815 | struct ipv6hdr *iph; | |
816 | struct icmp6hdr _icmph, *ic; | |
817 | struct ipv6hdr _ciph, *cih; /* The ip header contained | |
818 | within the ICMP */ | |
819 | struct ip_vs_iphdr ciph; | |
820 | struct ip_vs_conn *cp; | |
821 | struct ip_vs_protocol *pp; | |
f2428ed5 SH |
822 | unsigned int offset; |
823 | union nf_inet_addr snet; | |
2a3b791e JV |
824 | |
825 | *related = 1; | |
826 | ||
827 | /* reassemble IP fragments */ | |
828 | if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) { | |
1ca5bb54 | 829 | if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum))) |
2a3b791e JV |
830 | return NF_STOLEN; |
831 | } | |
832 | ||
833 | iph = ipv6_hdr(skb); | |
834 | offset = sizeof(struct ipv6hdr); | |
835 | ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); | |
836 | if (ic == NULL) | |
837 | return NF_DROP; | |
838 | ||
5b095d98 | 839 | IP_VS_DBG(12, "Outgoing ICMPv6 (%d,%d) %pI6->%pI6\n", |
2a3b791e | 840 | ic->icmp6_type, ntohs(icmpv6_id(ic)), |
38ff4fa4 | 841 | &iph->saddr, &iph->daddr); |
2a3b791e JV |
842 | |
843 | /* | |
844 | * Work through seeing if this is for us. | |
845 | * These checks are supposed to be in an order that means easy | |
846 | * things are checked first to speed up processing.... however | |
847 | * this means that some packets will manage to get a long way | |
848 | * down this stack and then be rejected, but that's life. | |
849 | */ | |
850 | if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) && | |
851 | (ic->icmp6_type != ICMPV6_PKT_TOOBIG) && | |
852 | (ic->icmp6_type != ICMPV6_TIME_EXCEED)) { | |
853 | *related = 0; | |
854 | return NF_ACCEPT; | |
855 | } | |
856 | ||
857 | /* Now find the contained IP header */ | |
858 | offset += sizeof(_icmph); | |
859 | cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); | |
860 | if (cih == NULL) | |
861 | return NF_ACCEPT; /* The packet looks wrong, ignore */ | |
862 | ||
863 | pp = ip_vs_proto_get(cih->nexthdr); | |
864 | if (!pp) | |
865 | return NF_ACCEPT; | |
866 | ||
867 | /* Is the embedded protocol header present? */ | |
868 | /* TODO: we don't support fragmentation at the moment anyways */ | |
869 | if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag)) | |
870 | return NF_ACCEPT; | |
871 | ||
872 | IP_VS_DBG_PKT(11, pp, skb, offset, "Checking outgoing ICMPv6 for"); | |
873 | ||
874 | offset += sizeof(struct ipv6hdr); | |
875 | ||
876 | ip_vs_fill_iphdr(AF_INET6, cih, &ciph); | |
877 | /* The embedded headers contain source and dest in reverse order */ | |
878 | cp = pp->conn_out_get(AF_INET6, skb, pp, &ciph, offset, 1); | |
879 | if (!cp) | |
880 | return NF_ACCEPT; | |
881 | ||
178f5e49 | 882 | ipv6_addr_copy(&snet.in6, &iph->saddr); |
f2428ed5 SH |
883 | return handle_response_icmp(AF_INET6, skb, &snet, cih->nexthdr, cp, |
884 | pp, offset, sizeof(struct ipv6hdr)); | |
2a3b791e JV |
885 | } |
886 | #endif | |
887 | ||
2906f66a VMR |
888 | /* |
889 | * Check if sctp chunc is ABORT chunk | |
890 | */ | |
891 | static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len) | |
892 | { | |
893 | sctp_chunkhdr_t *sch, schunk; | |
894 | sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t), | |
895 | sizeof(schunk), &schunk); | |
896 | if (sch == NULL) | |
897 | return 0; | |
898 | if (sch->type == SCTP_CID_ABORT) | |
899 | return 1; | |
900 | return 0; | |
901 | } | |
902 | ||
2a3b791e | 903 | static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len) |
1da177e4 LT |
904 | { |
905 | struct tcphdr _tcph, *th; | |
906 | ||
2a3b791e | 907 | th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph); |
1da177e4 LT |
908 | if (th == NULL) |
909 | return 0; | |
910 | return th->rst; | |
911 | } | |
912 | ||
4856c84c MT |
913 | /* Handle response packets: rewrite addresses and send away... |
914 | * Used for NAT and local client. | |
915 | */ | |
916 | static unsigned int | |
917 | handle_response(int af, struct sk_buff *skb, struct ip_vs_protocol *pp, | |
918 | struct ip_vs_conn *cp, int ihl) | |
919 | { | |
920 | IP_VS_DBG_PKT(11, pp, skb, 0, "Outgoing packet"); | |
921 | ||
922 | if (!skb_make_writable(skb, ihl)) | |
923 | goto drop; | |
924 | ||
925 | /* mangle the packet */ | |
926 | if (pp->snat_handler && !pp->snat_handler(skb, pp, cp)) | |
927 | goto drop; | |
928 | ||
929 | #ifdef CONFIG_IP_VS_IPV6 | |
930 | if (af == AF_INET6) | |
931 | ipv6_hdr(skb)->saddr = cp->vaddr.in6; | |
932 | else | |
933 | #endif | |
934 | { | |
935 | ip_hdr(skb)->saddr = cp->vaddr.ip; | |
936 | ip_send_check(ip_hdr(skb)); | |
937 | } | |
938 | ||
8a803040 JA |
939 | /* |
940 | * nf_iterate does not expect change in the skb->dst->dev. | |
941 | * It looks like it is not fatal to enable this code for hooks | |
942 | * where our handlers are at the end of the chain list and | |
943 | * when all next handlers use skb->dst->dev and not outdev. | |
944 | * It will definitely route properly the inout NAT traffic | |
945 | * when multiple paths are used. | |
946 | */ | |
947 | ||
4856c84c MT |
948 | /* For policy routing, packets originating from this |
949 | * machine itself may be routed differently to packets | |
950 | * passing through. We want this packet to be routed as | |
951 | * if it came from this machine itself. So re-compute | |
952 | * the routing information. | |
953 | */ | |
954 | #ifdef CONFIG_IP_VS_IPV6 | |
f5a41847 JA |
955 | if (af == AF_INET6) { |
956 | if (sysctl_ip_vs_snat_reroute && ip6_route_me_harder(skb) != 0) | |
957 | goto drop; | |
958 | } else | |
4856c84c | 959 | #endif |
f5a41847 JA |
960 | if ((sysctl_ip_vs_snat_reroute || |
961 | skb_rtable(skb)->rt_flags & RTCF_LOCAL) && | |
962 | ip_route_me_harder(skb, RTN_LOCAL) != 0) | |
963 | goto drop; | |
4856c84c | 964 | |
4856c84c MT |
965 | IP_VS_DBG_PKT(10, pp, skb, 0, "After SNAT"); |
966 | ||
967 | ip_vs_out_stats(cp, skb); | |
968 | ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pp); | |
cf356d69 | 969 | skb->ipvs_property = 1; |
f4bc17cd | 970 | if (!(cp->flags & IP_VS_CONN_F_NFCT)) |
cf356d69 | 971 | ip_vs_notrack(skb); |
f4bc17cd JA |
972 | else |
973 | ip_vs_update_conntrack(skb, cp, 0); | |
4856c84c MT |
974 | ip_vs_conn_put(cp); |
975 | ||
4856c84c MT |
976 | LeaveFunction(11); |
977 | return NF_ACCEPT; | |
978 | ||
979 | drop: | |
980 | ip_vs_conn_put(cp); | |
981 | kfree_skb(skb); | |
f4bc17cd | 982 | LeaveFunction(11); |
4856c84c MT |
983 | return NF_STOLEN; |
984 | } | |
985 | ||
1da177e4 | 986 | /* |
4856c84c | 987 | * Check if outgoing packet belongs to the established ip_vs_conn. |
1da177e4 LT |
988 | */ |
989 | static unsigned int | |
fc604767 | 990 | ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af) |
1da177e4 | 991 | { |
51ef348b | 992 | struct ip_vs_iphdr iph; |
1da177e4 LT |
993 | struct ip_vs_protocol *pp; |
994 | struct ip_vs_conn *cp; | |
1da177e4 LT |
995 | |
996 | EnterFunction(11); | |
997 | ||
fc604767 | 998 | /* Already marked as IPVS request or reply? */ |
6869c4d8 | 999 | if (skb->ipvs_property) |
1da177e4 LT |
1000 | return NF_ACCEPT; |
1001 | ||
fc604767 JA |
1002 | /* Bad... Do not break raw sockets */ |
1003 | if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT && | |
1004 | af == AF_INET)) { | |
1005 | struct sock *sk = skb->sk; | |
1006 | struct inet_sock *inet = inet_sk(skb->sk); | |
1007 | ||
1008 | if (inet && sk->sk_family == PF_INET && inet->nodefrag) | |
1009 | return NF_ACCEPT; | |
1010 | } | |
1011 | ||
1012 | if (unlikely(!skb_dst(skb))) | |
1013 | return NF_ACCEPT; | |
1014 | ||
2a3b791e JV |
1015 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); |
1016 | #ifdef CONFIG_IP_VS_IPV6 | |
1017 | if (af == AF_INET6) { | |
1018 | if (unlikely(iph.protocol == IPPROTO_ICMPV6)) { | |
1ca5bb54 JA |
1019 | int related; |
1020 | int verdict = ip_vs_out_icmp_v6(skb, &related, | |
1021 | hooknum); | |
1da177e4 | 1022 | |
f5a41847 | 1023 | if (related) |
2a3b791e JV |
1024 | return verdict; |
1025 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); | |
1026 | } | |
1027 | } else | |
1028 | #endif | |
1029 | if (unlikely(iph.protocol == IPPROTO_ICMP)) { | |
1ca5bb54 JA |
1030 | int related; |
1031 | int verdict = ip_vs_out_icmp(skb, &related, hooknum); | |
2a3b791e | 1032 | |
f5a41847 | 1033 | if (related) |
2a3b791e JV |
1034 | return verdict; |
1035 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); | |
1036 | } | |
1da177e4 | 1037 | |
51ef348b | 1038 | pp = ip_vs_proto_get(iph.protocol); |
1da177e4 LT |
1039 | if (unlikely(!pp)) |
1040 | return NF_ACCEPT; | |
1041 | ||
1042 | /* reassemble IP fragments */ | |
2a3b791e JV |
1043 | #ifdef CONFIG_IP_VS_IPV6 |
1044 | if (af == AF_INET6) { | |
1ca5bb54 JA |
1045 | if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) { |
1046 | if (ip_vs_gather_frags_v6(skb, | |
1047 | ip_vs_defrag_user(hooknum))) | |
1048 | return NF_STOLEN; | |
2a3b791e | 1049 | } |
1ca5bb54 JA |
1050 | |
1051 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); | |
2a3b791e JV |
1052 | } else |
1053 | #endif | |
1054 | if (unlikely(ip_hdr(skb)->frag_off & htons(IP_MF|IP_OFFSET) && | |
1055 | !pp->dont_defrag)) { | |
1ca5bb54 JA |
1056 | if (ip_vs_gather_frags(skb, |
1057 | ip_vs_defrag_user(hooknum))) | |
2a3b791e JV |
1058 | return NF_STOLEN; |
1059 | ||
1060 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); | |
1061 | } | |
1da177e4 LT |
1062 | |
1063 | /* | |
1064 | * Check if the packet belongs to an existing entry | |
1065 | */ | |
2a3b791e | 1066 | cp = pp->conn_out_get(af, skb, pp, &iph, iph.len, 0); |
1da177e4 LT |
1067 | |
1068 | if (unlikely(!cp)) { | |
1069 | if (sysctl_ip_vs_nat_icmp_send && | |
1070 | (pp->protocol == IPPROTO_TCP || | |
2906f66a VMR |
1071 | pp->protocol == IPPROTO_UDP || |
1072 | pp->protocol == IPPROTO_SCTP)) { | |
014d730d | 1073 | __be16 _ports[2], *pptr; |
1da177e4 | 1074 | |
51ef348b | 1075 | pptr = skb_header_pointer(skb, iph.len, |
1da177e4 LT |
1076 | sizeof(_ports), _ports); |
1077 | if (pptr == NULL) | |
1078 | return NF_ACCEPT; /* Not for me */ | |
7937df15 JV |
1079 | if (ip_vs_lookup_real_service(af, iph.protocol, |
1080 | &iph.saddr, | |
2a3b791e | 1081 | pptr[0])) { |
1da177e4 LT |
1082 | /* |
1083 | * Notify the real server: there is no | |
1084 | * existing entry if it is not RST | |
1085 | * packet or not TCP packet. | |
1086 | */ | |
2906f66a VMR |
1087 | if ((iph.protocol != IPPROTO_TCP && |
1088 | iph.protocol != IPPROTO_SCTP) | |
1089 | || ((iph.protocol == IPPROTO_TCP | |
1090 | && !is_tcp_reset(skb, iph.len)) | |
1091 | || (iph.protocol == IPPROTO_SCTP | |
1092 | && !is_sctp_abort(skb, | |
1093 | iph.len)))) { | |
2a3b791e JV |
1094 | #ifdef CONFIG_IP_VS_IPV6 |
1095 | if (af == AF_INET6) | |
1096 | icmpv6_send(skb, | |
1097 | ICMPV6_DEST_UNREACH, | |
1098 | ICMPV6_PORT_UNREACH, | |
3ffe533c | 1099 | 0); |
2a3b791e JV |
1100 | else |
1101 | #endif | |
1102 | icmp_send(skb, | |
1103 | ICMP_DEST_UNREACH, | |
1104 | ICMP_PORT_UNREACH, 0); | |
1da177e4 | 1105 | return NF_DROP; |
5af149cc | 1106 | } |
1da177e4 LT |
1107 | } |
1108 | } | |
1109 | IP_VS_DBG_PKT(12, pp, skb, 0, | |
1110 | "packet continues traversal as normal"); | |
1111 | return NF_ACCEPT; | |
1112 | } | |
1113 | ||
4856c84c | 1114 | return handle_response(af, skb, pp, cp, iph.len); |
1da177e4 LT |
1115 | } |
1116 | ||
fc604767 JA |
1117 | /* |
1118 | * It is hooked at the NF_INET_FORWARD chain, used only for VS/NAT. | |
1119 | * Check if packet is reply for established ip_vs_conn. | |
1120 | */ | |
1121 | static unsigned int | |
1122 | ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb, | |
1123 | const struct net_device *in, const struct net_device *out, | |
1124 | int (*okfn)(struct sk_buff *)) | |
1125 | { | |
1126 | return ip_vs_out(hooknum, skb, AF_INET); | |
1127 | } | |
1128 | ||
1129 | /* | |
1130 | * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT. | |
1131 | * Check if packet is reply for established ip_vs_conn. | |
1132 | */ | |
1133 | static unsigned int | |
1134 | ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb, | |
1135 | const struct net_device *in, const struct net_device *out, | |
1136 | int (*okfn)(struct sk_buff *)) | |
1137 | { | |
1138 | unsigned int verdict; | |
1139 | ||
1140 | /* Disable BH in LOCAL_OUT until all places are fixed */ | |
1141 | local_bh_disable(); | |
1142 | verdict = ip_vs_out(hooknum, skb, AF_INET); | |
1143 | local_bh_enable(); | |
1144 | return verdict; | |
1145 | } | |
1146 | ||
1147 | #ifdef CONFIG_IP_VS_IPV6 | |
1148 | ||
1149 | /* | |
1150 | * It is hooked at the NF_INET_FORWARD chain, used only for VS/NAT. | |
1151 | * Check if packet is reply for established ip_vs_conn. | |
1152 | */ | |
1153 | static unsigned int | |
1154 | ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb, | |
1155 | const struct net_device *in, const struct net_device *out, | |
1156 | int (*okfn)(struct sk_buff *)) | |
1157 | { | |
1158 | return ip_vs_out(hooknum, skb, AF_INET6); | |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT. | |
1163 | * Check if packet is reply for established ip_vs_conn. | |
1164 | */ | |
1165 | static unsigned int | |
1166 | ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb, | |
1167 | const struct net_device *in, const struct net_device *out, | |
1168 | int (*okfn)(struct sk_buff *)) | |
1169 | { | |
1170 | unsigned int verdict; | |
1171 | ||
1172 | /* Disable BH in LOCAL_OUT until all places are fixed */ | |
1173 | local_bh_disable(); | |
1174 | verdict = ip_vs_out(hooknum, skb, AF_INET6); | |
1175 | local_bh_enable(); | |
1176 | return verdict; | |
1177 | } | |
1178 | ||
1179 | #endif | |
1da177e4 LT |
1180 | |
1181 | /* | |
1182 | * Handle ICMP messages in the outside-to-inside direction (incoming). | |
1183 | * Find any that might be relevant, check against existing connections, | |
1184 | * forward to the right destination host if relevant. | |
1185 | * Currently handles error types - unreachable, quench, ttl exceeded. | |
1186 | */ | |
e905a9ed | 1187 | static int |
3db05fea | 1188 | ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum) |
1da177e4 | 1189 | { |
1da177e4 LT |
1190 | struct iphdr *iph; |
1191 | struct icmphdr _icmph, *ic; | |
1192 | struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */ | |
51ef348b | 1193 | struct ip_vs_iphdr ciph; |
1da177e4 LT |
1194 | struct ip_vs_conn *cp; |
1195 | struct ip_vs_protocol *pp; | |
1196 | unsigned int offset, ihl, verdict; | |
f2428ed5 | 1197 | union nf_inet_addr snet; |
1da177e4 LT |
1198 | |
1199 | *related = 1; | |
1200 | ||
1201 | /* reassemble IP fragments */ | |
eddc9ec5 | 1202 | if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) { |
1ca5bb54 | 1203 | if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum))) |
1da177e4 | 1204 | return NF_STOLEN; |
1da177e4 LT |
1205 | } |
1206 | ||
eddc9ec5 | 1207 | iph = ip_hdr(skb); |
1da177e4 LT |
1208 | offset = ihl = iph->ihl * 4; |
1209 | ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); | |
1210 | if (ic == NULL) | |
1211 | return NF_DROP; | |
1212 | ||
14d5e834 | 1213 | IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n", |
1da177e4 | 1214 | ic->type, ntohs(icmp_id(ic)), |
14d5e834 | 1215 | &iph->saddr, &iph->daddr); |
1da177e4 LT |
1216 | |
1217 | /* | |
1218 | * Work through seeing if this is for us. | |
1219 | * These checks are supposed to be in an order that means easy | |
1220 | * things are checked first to speed up processing.... however | |
1221 | * this means that some packets will manage to get a long way | |
1222 | * down this stack and then be rejected, but that's life. | |
1223 | */ | |
1224 | if ((ic->type != ICMP_DEST_UNREACH) && | |
1225 | (ic->type != ICMP_SOURCE_QUENCH) && | |
1226 | (ic->type != ICMP_TIME_EXCEEDED)) { | |
1227 | *related = 0; | |
1228 | return NF_ACCEPT; | |
1229 | } | |
1230 | ||
1231 | /* Now find the contained IP header */ | |
1232 | offset += sizeof(_icmph); | |
1233 | cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); | |
1234 | if (cih == NULL) | |
1235 | return NF_ACCEPT; /* The packet looks wrong, ignore */ | |
1236 | ||
1237 | pp = ip_vs_proto_get(cih->protocol); | |
1238 | if (!pp) | |
1239 | return NF_ACCEPT; | |
1240 | ||
1241 | /* Is the embedded protocol header present? */ | |
4412ec49 | 1242 | if (unlikely(cih->frag_off & htons(IP_OFFSET) && |
1da177e4 LT |
1243 | pp->dont_defrag)) |
1244 | return NF_ACCEPT; | |
1245 | ||
1246 | IP_VS_DBG_PKT(11, pp, skb, offset, "Checking incoming ICMP for"); | |
1247 | ||
1248 | offset += cih->ihl * 4; | |
1249 | ||
51ef348b | 1250 | ip_vs_fill_iphdr(AF_INET, cih, &ciph); |
1da177e4 | 1251 | /* The embedded headers contain source and dest in reverse order */ |
51ef348b | 1252 | cp = pp->conn_in_get(AF_INET, skb, pp, &ciph, offset, 1); |
4856c84c MT |
1253 | if (!cp) { |
1254 | /* The packet could also belong to a local client */ | |
1255 | cp = pp->conn_out_get(AF_INET, skb, pp, &ciph, offset, 1); | |
f2428ed5 SH |
1256 | if (cp) { |
1257 | snet.ip = iph->saddr; | |
1258 | return handle_response_icmp(AF_INET, skb, &snet, | |
1259 | cih->protocol, cp, pp, | |
4856c84c | 1260 | offset, ihl); |
f2428ed5 | 1261 | } |
1da177e4 | 1262 | return NF_ACCEPT; |
4856c84c | 1263 | } |
1da177e4 LT |
1264 | |
1265 | verdict = NF_DROP; | |
1266 | ||
1267 | /* Ensure the checksum is correct */ | |
60476372 | 1268 | if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) { |
1da177e4 | 1269 | /* Failed checksum! */ |
14d5e834 HH |
1270 | IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n", |
1271 | &iph->saddr); | |
1da177e4 LT |
1272 | goto out; |
1273 | } | |
1274 | ||
1275 | /* do the statistics and put it back */ | |
1276 | ip_vs_in_stats(cp, skb); | |
1277 | if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol) | |
1278 | offset += 2 * sizeof(__u16); | |
1279 | verdict = ip_vs_icmp_xmit(skb, cp, pp, offset); | |
489fdeda JA |
1280 | /* LOCALNODE from FORWARD hook is not supported */ |
1281 | if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD && | |
1282 | skb_rtable(skb)->rt_flags & RTCF_LOCAL) { | |
1283 | IP_VS_DBG(1, "%s(): " | |
1284 | "local delivery to %pI4 but in FORWARD\n", | |
1285 | __func__, &skb_rtable(skb)->rt_dst); | |
1286 | verdict = NF_DROP; | |
1287 | } | |
1da177e4 LT |
1288 | |
1289 | out: | |
1290 | __ip_vs_conn_put(cp); | |
1291 | ||
1292 | return verdict; | |
1293 | } | |
1294 | ||
2a3b791e JV |
1295 | #ifdef CONFIG_IP_VS_IPV6 |
1296 | static int | |
1297 | ip_vs_in_icmp_v6(struct sk_buff *skb, int *related, unsigned int hooknum) | |
1298 | { | |
1299 | struct ipv6hdr *iph; | |
1300 | struct icmp6hdr _icmph, *ic; | |
1301 | struct ipv6hdr _ciph, *cih; /* The ip header contained | |
1302 | within the ICMP */ | |
1303 | struct ip_vs_iphdr ciph; | |
1304 | struct ip_vs_conn *cp; | |
1305 | struct ip_vs_protocol *pp; | |
1306 | unsigned int offset, verdict; | |
f2428ed5 | 1307 | union nf_inet_addr snet; |
489fdeda | 1308 | struct rt6_info *rt; |
2a3b791e JV |
1309 | |
1310 | *related = 1; | |
1311 | ||
1312 | /* reassemble IP fragments */ | |
1313 | if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) { | |
1ca5bb54 | 1314 | if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum))) |
2a3b791e JV |
1315 | return NF_STOLEN; |
1316 | } | |
1317 | ||
1318 | iph = ipv6_hdr(skb); | |
1319 | offset = sizeof(struct ipv6hdr); | |
1320 | ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); | |
1321 | if (ic == NULL) | |
1322 | return NF_DROP; | |
1323 | ||
5b095d98 | 1324 | IP_VS_DBG(12, "Incoming ICMPv6 (%d,%d) %pI6->%pI6\n", |
2a3b791e | 1325 | ic->icmp6_type, ntohs(icmpv6_id(ic)), |
38ff4fa4 | 1326 | &iph->saddr, &iph->daddr); |
2a3b791e JV |
1327 | |
1328 | /* | |
1329 | * Work through seeing if this is for us. | |
1330 | * These checks are supposed to be in an order that means easy | |
1331 | * things are checked first to speed up processing.... however | |
1332 | * this means that some packets will manage to get a long way | |
1333 | * down this stack and then be rejected, but that's life. | |
1334 | */ | |
1335 | if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) && | |
1336 | (ic->icmp6_type != ICMPV6_PKT_TOOBIG) && | |
1337 | (ic->icmp6_type != ICMPV6_TIME_EXCEED)) { | |
1338 | *related = 0; | |
1339 | return NF_ACCEPT; | |
1340 | } | |
1341 | ||
1342 | /* Now find the contained IP header */ | |
1343 | offset += sizeof(_icmph); | |
1344 | cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); | |
1345 | if (cih == NULL) | |
1346 | return NF_ACCEPT; /* The packet looks wrong, ignore */ | |
1347 | ||
1348 | pp = ip_vs_proto_get(cih->nexthdr); | |
1349 | if (!pp) | |
1350 | return NF_ACCEPT; | |
1351 | ||
1352 | /* Is the embedded protocol header present? */ | |
1353 | /* TODO: we don't support fragmentation at the moment anyways */ | |
1354 | if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag)) | |
1355 | return NF_ACCEPT; | |
1356 | ||
1357 | IP_VS_DBG_PKT(11, pp, skb, offset, "Checking incoming ICMPv6 for"); | |
1358 | ||
1359 | offset += sizeof(struct ipv6hdr); | |
1360 | ||
1361 | ip_vs_fill_iphdr(AF_INET6, cih, &ciph); | |
1362 | /* The embedded headers contain source and dest in reverse order */ | |
1363 | cp = pp->conn_in_get(AF_INET6, skb, pp, &ciph, offset, 1); | |
f2428ed5 SH |
1364 | if (!cp) { |
1365 | /* The packet could also belong to a local client */ | |
1366 | cp = pp->conn_out_get(AF_INET6, skb, pp, &ciph, offset, 1); | |
1367 | if (cp) { | |
178f5e49 | 1368 | ipv6_addr_copy(&snet.in6, &iph->saddr); |
f2428ed5 SH |
1369 | return handle_response_icmp(AF_INET6, skb, &snet, |
1370 | cih->nexthdr, | |
1371 | cp, pp, offset, | |
1372 | sizeof(struct ipv6hdr)); | |
1373 | } | |
2a3b791e | 1374 | return NF_ACCEPT; |
f2428ed5 | 1375 | } |
2a3b791e JV |
1376 | |
1377 | verdict = NF_DROP; | |
1378 | ||
1379 | /* do the statistics and put it back */ | |
1380 | ip_vs_in_stats(cp, skb); | |
2906f66a VMR |
1381 | if (IPPROTO_TCP == cih->nexthdr || IPPROTO_UDP == cih->nexthdr || |
1382 | IPPROTO_SCTP == cih->nexthdr) | |
2a3b791e JV |
1383 | offset += 2 * sizeof(__u16); |
1384 | verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset); | |
489fdeda JA |
1385 | /* LOCALNODE from FORWARD hook is not supported */ |
1386 | if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD && | |
1387 | (rt = (struct rt6_info *) skb_dst(skb)) && | |
1388 | rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK) { | |
1389 | IP_VS_DBG(1, "%s(): " | |
1390 | "local delivery to %pI6 but in FORWARD\n", | |
1391 | __func__, &rt->rt6i_dst); | |
1392 | verdict = NF_DROP; | |
1393 | } | |
2a3b791e JV |
1394 | |
1395 | __ip_vs_conn_put(cp); | |
1396 | ||
1397 | return verdict; | |
1398 | } | |
1399 | #endif | |
1400 | ||
1401 | ||
1da177e4 LT |
1402 | /* |
1403 | * Check if it's for virtual services, look it up, | |
1404 | * and send it on its way... | |
1405 | */ | |
1406 | static unsigned int | |
3db05fea | 1407 | ip_vs_in(unsigned int hooknum, struct sk_buff *skb, |
1da177e4 LT |
1408 | const struct net_device *in, const struct net_device *out, |
1409 | int (*okfn)(struct sk_buff *)) | |
1410 | { | |
51ef348b | 1411 | struct ip_vs_iphdr iph; |
1da177e4 LT |
1412 | struct ip_vs_protocol *pp; |
1413 | struct ip_vs_conn *cp; | |
1e66dafc | 1414 | int ret, restart, af, pkts; |
2a3b791e | 1415 | |
fc604767 JA |
1416 | /* Already marked as IPVS request or reply? */ |
1417 | if (skb->ipvs_property) | |
1418 | return NF_ACCEPT; | |
1419 | ||
60678040 | 1420 | af = (skb->protocol == htons(ETH_P_IP)) ? AF_INET : AF_INET6; |
51ef348b | 1421 | |
2a3b791e | 1422 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); |
1da177e4 LT |
1423 | |
1424 | /* | |
4856c84c MT |
1425 | * Big tappo: only PACKET_HOST, including loopback for local client |
1426 | * Don't handle local packets on IPv6 for now | |
1da177e4 | 1427 | */ |
f2428ed5 | 1428 | if (unlikely(skb->pkt_type != PACKET_HOST)) { |
51ef348b JV |
1429 | IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s ignored\n", |
1430 | skb->pkt_type, | |
1431 | iph.protocol, | |
2a3b791e | 1432 | IP_VS_DBG_ADDR(af, &iph.daddr)); |
1da177e4 LT |
1433 | return NF_ACCEPT; |
1434 | } | |
1435 | ||
94b26551 JV |
1436 | #ifdef CONFIG_IP_VS_IPV6 |
1437 | if (af == AF_INET6) { | |
1438 | if (unlikely(iph.protocol == IPPROTO_ICMPV6)) { | |
1ca5bb54 JA |
1439 | int related; |
1440 | int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum); | |
1da177e4 | 1441 | |
94b26551 JV |
1442 | if (related) |
1443 | return verdict; | |
1444 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); | |
1445 | } | |
1446 | } else | |
1447 | #endif | |
1448 | if (unlikely(iph.protocol == IPPROTO_ICMP)) { | |
1ca5bb54 JA |
1449 | int related; |
1450 | int verdict = ip_vs_in_icmp(skb, &related, hooknum); | |
94b26551 JV |
1451 | |
1452 | if (related) | |
1453 | return verdict; | |
1454 | ip_vs_fill_iphdr(af, skb_network_header(skb), &iph); | |
1455 | } | |
1da177e4 LT |
1456 | |
1457 | /* Protocol supported? */ | |
51ef348b | 1458 | pp = ip_vs_proto_get(iph.protocol); |
1da177e4 LT |
1459 | if (unlikely(!pp)) |
1460 | return NF_ACCEPT; | |
1461 | ||
1da177e4 LT |
1462 | /* |
1463 | * Check if the packet belongs to an existing connection entry | |
1464 | */ | |
2a3b791e | 1465 | cp = pp->conn_in_get(af, skb, pp, &iph, iph.len, 0); |
1da177e4 LT |
1466 | |
1467 | if (unlikely(!cp)) { | |
1468 | int v; | |
1469 | ||
4856c84c MT |
1470 | /* For local client packets, it could be a response */ |
1471 | cp = pp->conn_out_get(af, skb, pp, &iph, iph.len, 0); | |
1472 | if (cp) | |
1473 | return handle_response(af, skb, pp, cp, iph.len); | |
1474 | ||
2a3b791e | 1475 | if (!pp->conn_schedule(af, skb, pp, &v, &cp)) |
1da177e4 LT |
1476 | return v; |
1477 | } | |
1478 | ||
1479 | if (unlikely(!cp)) { | |
1480 | /* sorry, all this trouble for a no-hit :) */ | |
1481 | IP_VS_DBG_PKT(12, pp, skb, 0, | |
1482 | "packet continues traversal as normal"); | |
1483 | return NF_ACCEPT; | |
1484 | } | |
1485 | ||
1486 | IP_VS_DBG_PKT(11, pp, skb, 0, "Incoming packet"); | |
1487 | ||
1488 | /* Check the server status */ | |
1489 | if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) { | |
1490 | /* the destination server is not available */ | |
1491 | ||
1492 | if (sysctl_ip_vs_expire_nodest_conn) { | |
1493 | /* try to expire the connection immediately */ | |
1494 | ip_vs_conn_expire_now(cp); | |
1da177e4 | 1495 | } |
dc8103f2 JA |
1496 | /* don't restart its timer, and silently |
1497 | drop the packet. */ | |
1498 | __ip_vs_conn_put(cp); | |
1da177e4 LT |
1499 | return NF_DROP; |
1500 | } | |
1501 | ||
1502 | ip_vs_in_stats(cp, skb); | |
1503 | restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp); | |
1504 | if (cp->packet_xmit) | |
1505 | ret = cp->packet_xmit(skb, cp, pp); | |
1506 | /* do not touch skb anymore */ | |
1507 | else { | |
1508 | IP_VS_DBG_RL("warning: packet_xmit is null"); | |
1509 | ret = NF_ACCEPT; | |
1510 | } | |
1511 | ||
efac5276 RB |
1512 | /* Increase its packet counter and check if it is needed |
1513 | * to be synchronized | |
1514 | * | |
1515 | * Sync connection if it is about to close to | |
1516 | * encorage the standby servers to update the connections timeout | |
1517 | */ | |
1e66dafc | 1518 | pkts = atomic_add_return(1, &cp->in_pkts); |
2906f66a VMR |
1519 | if (af == AF_INET && (ip_vs_sync_state & IP_VS_STATE_MASTER) && |
1520 | cp->protocol == IPPROTO_SCTP) { | |
1521 | if ((cp->state == IP_VS_SCTP_S_ESTABLISHED && | |
8ed2163f | 1522 | (pkts % sysctl_ip_vs_sync_threshold[1] |
2906f66a VMR |
1523 | == sysctl_ip_vs_sync_threshold[0])) || |
1524 | (cp->old_state != cp->state && | |
1525 | ((cp->state == IP_VS_SCTP_S_CLOSED) || | |
1526 | (cp->state == IP_VS_SCTP_S_SHUT_ACK_CLI) || | |
1527 | (cp->state == IP_VS_SCTP_S_SHUT_ACK_SER)))) { | |
1528 | ip_vs_sync_conn(cp); | |
1529 | goto out; | |
1530 | } | |
1531 | } | |
1532 | ||
8ed2163f JA |
1533 | /* Keep this block last: TCP and others with pp->num_states <= 1 */ |
1534 | else if (af == AF_INET && | |
c6883f58 | 1535 | (ip_vs_sync_state & IP_VS_STATE_MASTER) && |
efac5276 RB |
1536 | (((cp->protocol != IPPROTO_TCP || |
1537 | cp->state == IP_VS_TCP_S_ESTABLISHED) && | |
1e66dafc | 1538 | (pkts % sysctl_ip_vs_sync_threshold[1] |
efac5276 RB |
1539 | == sysctl_ip_vs_sync_threshold[0])) || |
1540 | ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) && | |
1541 | ((cp->state == IP_VS_TCP_S_FIN_WAIT) || | |
9abfe315 | 1542 | (cp->state == IP_VS_TCP_S_CLOSE) || |
9d3a0de7 RB |
1543 | (cp->state == IP_VS_TCP_S_CLOSE_WAIT) || |
1544 | (cp->state == IP_VS_TCP_S_TIME_WAIT))))) | |
1da177e4 | 1545 | ip_vs_sync_conn(cp); |
2906f66a | 1546 | out: |
efac5276 | 1547 | cp->old_state = cp->state; |
1da177e4 LT |
1548 | |
1549 | ip_vs_conn_put(cp); | |
1550 | return ret; | |
1551 | } | |
1552 | ||
1553 | ||
1554 | /* | |
6e23ae2a | 1555 | * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP |
1da177e4 LT |
1556 | * related packets destined for 0.0.0.0/0. |
1557 | * When fwmark-based virtual service is used, such as transparent | |
1558 | * cache cluster, TCP packets can be marked and routed to ip_vs_in, | |
1559 | * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and | |
6e23ae2a | 1560 | * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain |
1da177e4 LT |
1561 | * and send them to ip_vs_in_icmp. |
1562 | */ | |
1563 | static unsigned int | |
3db05fea | 1564 | ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb, |
1da177e4 LT |
1565 | const struct net_device *in, const struct net_device *out, |
1566 | int (*okfn)(struct sk_buff *)) | |
1567 | { | |
1568 | int r; | |
1569 | ||
3db05fea | 1570 | if (ip_hdr(skb)->protocol != IPPROTO_ICMP) |
1da177e4 LT |
1571 | return NF_ACCEPT; |
1572 | ||
3db05fea | 1573 | return ip_vs_in_icmp(skb, &r, hooknum); |
1da177e4 LT |
1574 | } |
1575 | ||
2a3b791e JV |
1576 | #ifdef CONFIG_IP_VS_IPV6 |
1577 | static unsigned int | |
1578 | ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb, | |
1579 | const struct net_device *in, const struct net_device *out, | |
1580 | int (*okfn)(struct sk_buff *)) | |
1581 | { | |
1582 | int r; | |
1583 | ||
1584 | if (ipv6_hdr(skb)->nexthdr != IPPROTO_ICMPV6) | |
1585 | return NF_ACCEPT; | |
1586 | ||
1587 | return ip_vs_in_icmp_v6(skb, &r, hooknum); | |
1588 | } | |
1589 | #endif | |
1590 | ||
1da177e4 | 1591 | |
1999414a | 1592 | static struct nf_hook_ops ip_vs_ops[] __read_mostly = { |
41c5b317 PM |
1593 | /* After packet filtering, forward packet through VS/DR, VS/TUN, |
1594 | * or VS/NAT(change destination), so that filtering rules can be | |
1595 | * applied to IPVS. */ | |
1596 | { | |
1597 | .hook = ip_vs_in, | |
1598 | .owner = THIS_MODULE, | |
1599 | .pf = PF_INET, | |
1600 | .hooknum = NF_INET_LOCAL_IN, | |
1601 | .priority = 100, | |
1602 | }, | |
fc604767 | 1603 | /* Before ip_vs_in, change source only for VS/NAT */ |
41c5b317 | 1604 | { |
fc604767 | 1605 | .hook = ip_vs_local_reply4, |
41c5b317 PM |
1606 | .owner = THIS_MODULE, |
1607 | .pf = PF_INET, | |
fc604767 JA |
1608 | .hooknum = NF_INET_LOCAL_OUT, |
1609 | .priority = -99, | |
41c5b317 PM |
1610 | }, |
1611 | /* After packet filtering (but before ip_vs_out_icmp), catch icmp | |
1612 | * destined for 0.0.0.0/0, which is for incoming IPVS connections */ | |
1613 | { | |
1614 | .hook = ip_vs_forward_icmp, | |
1615 | .owner = THIS_MODULE, | |
1616 | .pf = PF_INET, | |
1617 | .hooknum = NF_INET_FORWARD, | |
1618 | .priority = 99, | |
1619 | }, | |
fc604767 JA |
1620 | /* After packet filtering, change source only for VS/NAT */ |
1621 | { | |
1622 | .hook = ip_vs_reply4, | |
1623 | .owner = THIS_MODULE, | |
1624 | .pf = PF_INET, | |
1625 | .hooknum = NF_INET_FORWARD, | |
1626 | .priority = 100, | |
1627 | }, | |
473b23d3 JV |
1628 | #ifdef CONFIG_IP_VS_IPV6 |
1629 | /* After packet filtering, forward packet through VS/DR, VS/TUN, | |
1630 | * or VS/NAT(change destination), so that filtering rules can be | |
1631 | * applied to IPVS. */ | |
1632 | { | |
1633 | .hook = ip_vs_in, | |
1634 | .owner = THIS_MODULE, | |
1635 | .pf = PF_INET6, | |
1636 | .hooknum = NF_INET_LOCAL_IN, | |
1637 | .priority = 100, | |
1638 | }, | |
fc604767 | 1639 | /* Before ip_vs_in, change source only for VS/NAT */ |
473b23d3 | 1640 | { |
fc604767 | 1641 | .hook = ip_vs_local_reply6, |
473b23d3 | 1642 | .owner = THIS_MODULE, |
fc604767 JA |
1643 | .pf = PF_INET, |
1644 | .hooknum = NF_INET_LOCAL_OUT, | |
1645 | .priority = -99, | |
473b23d3 JV |
1646 | }, |
1647 | /* After packet filtering (but before ip_vs_out_icmp), catch icmp | |
1648 | * destined for 0.0.0.0/0, which is for incoming IPVS connections */ | |
1649 | { | |
1650 | .hook = ip_vs_forward_icmp_v6, | |
1651 | .owner = THIS_MODULE, | |
1652 | .pf = PF_INET6, | |
1653 | .hooknum = NF_INET_FORWARD, | |
1654 | .priority = 99, | |
1655 | }, | |
fc604767 JA |
1656 | /* After packet filtering, change source only for VS/NAT */ |
1657 | { | |
1658 | .hook = ip_vs_reply6, | |
1659 | .owner = THIS_MODULE, | |
1660 | .pf = PF_INET6, | |
1661 | .hooknum = NF_INET_FORWARD, | |
1662 | .priority = 100, | |
1663 | }, | |
473b23d3 | 1664 | #endif |
1da177e4 LT |
1665 | }; |
1666 | ||
1667 | ||
1668 | /* | |
1669 | * Initialize IP Virtual Server | |
1670 | */ | |
1671 | static int __init ip_vs_init(void) | |
1672 | { | |
1673 | int ret; | |
1674 | ||
a919cf4b SW |
1675 | ip_vs_estimator_init(); |
1676 | ||
1da177e4 LT |
1677 | ret = ip_vs_control_init(); |
1678 | if (ret < 0) { | |
1e3e238e | 1679 | pr_err("can't setup control.\n"); |
a919cf4b | 1680 | goto cleanup_estimator; |
1da177e4 LT |
1681 | } |
1682 | ||
1683 | ip_vs_protocol_init(); | |
1684 | ||
1685 | ret = ip_vs_app_init(); | |
1686 | if (ret < 0) { | |
1e3e238e | 1687 | pr_err("can't setup application helper.\n"); |
1da177e4 LT |
1688 | goto cleanup_protocol; |
1689 | } | |
1690 | ||
1691 | ret = ip_vs_conn_init(); | |
1692 | if (ret < 0) { | |
1e3e238e | 1693 | pr_err("can't setup connection table.\n"); |
1da177e4 LT |
1694 | goto cleanup_app; |
1695 | } | |
1696 | ||
41c5b317 | 1697 | ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops)); |
1da177e4 | 1698 | if (ret < 0) { |
1e3e238e | 1699 | pr_err("can't register hooks.\n"); |
1da177e4 LT |
1700 | goto cleanup_conn; |
1701 | } | |
1702 | ||
1e3e238e | 1703 | pr_info("ipvs loaded.\n"); |
1da177e4 LT |
1704 | return ret; |
1705 | ||
1da177e4 LT |
1706 | cleanup_conn: |
1707 | ip_vs_conn_cleanup(); | |
1708 | cleanup_app: | |
1709 | ip_vs_app_cleanup(); | |
1710 | cleanup_protocol: | |
1711 | ip_vs_protocol_cleanup(); | |
1712 | ip_vs_control_cleanup(); | |
a919cf4b SW |
1713 | cleanup_estimator: |
1714 | ip_vs_estimator_cleanup(); | |
1da177e4 LT |
1715 | return ret; |
1716 | } | |
1717 | ||
1718 | static void __exit ip_vs_cleanup(void) | |
1719 | { | |
41c5b317 | 1720 | nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops)); |
1da177e4 LT |
1721 | ip_vs_conn_cleanup(); |
1722 | ip_vs_app_cleanup(); | |
1723 | ip_vs_protocol_cleanup(); | |
1724 | ip_vs_control_cleanup(); | |
a919cf4b | 1725 | ip_vs_estimator_cleanup(); |
1e3e238e | 1726 | pr_info("ipvs unloaded.\n"); |
1da177e4 LT |
1727 | } |
1728 | ||
1729 | module_init(ip_vs_init); | |
1730 | module_exit(ip_vs_cleanup); | |
1731 | MODULE_LICENSE("GPL"); |