IPVS: Extend functions for getting/creating connections
[safe/jmp/linux-2.6] / net / ipv4 / ipvs / ip_vs_proto_udp.c
1 /*
2  * ip_vs_proto_udp.c:   UDP load balancing support for IPVS
3  *
4  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
5  *              Julian Anastasov <ja@ssi.bg>
6  *
7  *              This program is free software; you can redistribute it and/or
8  *              modify it under the terms of the GNU General Public License
9  *              as published by the Free Software Foundation; either version
10  *              2 of the License, or (at your option) any later version.
11  *
12  * Changes:
13  *
14  */
15
16 #include <linux/in.h>
17 #include <linux/ip.h>
18 #include <linux/kernel.h>
19 #include <linux/netfilter.h>
20 #include <linux/netfilter_ipv4.h>
21 #include <linux/udp.h>
22
23 #include <net/ip_vs.h>
24 #include <net/ip.h>
25
26 static struct ip_vs_conn *
27 udp_conn_in_get(int af, const struct sk_buff *skb, struct ip_vs_protocol *pp,
28                 const struct ip_vs_iphdr *iph, unsigned int proto_off,
29                 int inverse)
30 {
31         struct ip_vs_conn *cp;
32         __be16 _ports[2], *pptr;
33
34         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
35         if (pptr == NULL)
36                 return NULL;
37
38         if (likely(!inverse)) {
39                 cp = ip_vs_conn_in_get(af, iph->protocol,
40                                        &iph->saddr, pptr[0],
41                                        &iph->daddr, pptr[1]);
42         } else {
43                 cp = ip_vs_conn_in_get(af, iph->protocol,
44                                        &iph->daddr, pptr[1],
45                                        &iph->saddr, pptr[0]);
46         }
47
48         return cp;
49 }
50
51
52 static struct ip_vs_conn *
53 udp_conn_out_get(int af, const struct sk_buff *skb, struct ip_vs_protocol *pp,
54                  const struct ip_vs_iphdr *iph, unsigned int proto_off,
55                  int inverse)
56 {
57         struct ip_vs_conn *cp;
58         __be16 _ports[2], *pptr;
59
60         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
61         if (pptr == NULL)
62                 return NULL;
63
64         if (likely(!inverse)) {
65                 cp = ip_vs_conn_out_get(af, iph->protocol,
66                                         &iph->saddr, pptr[0],
67                                         &iph->daddr, pptr[1]);
68         } else {
69                 cp = ip_vs_conn_out_get(af, iph->protocol,
70                                         &iph->daddr, pptr[1],
71                                         &iph->saddr, pptr[0]);
72         }
73
74         return cp;
75 }
76
77
78 static int
79 udp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_protocol *pp,
80                   int *verdict, struct ip_vs_conn **cpp)
81 {
82         struct ip_vs_service *svc;
83         struct udphdr _udph, *uh;
84         struct ip_vs_iphdr iph;
85
86         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
87
88         uh = skb_header_pointer(skb, iph.len, sizeof(_udph), &_udph);
89         if (uh == NULL) {
90                 *verdict = NF_DROP;
91                 return 0;
92         }
93
94         svc = ip_vs_service_get(af, skb->mark, iph.protocol,
95                                 &iph.daddr, uh->dest);
96         if (svc) {
97                 if (ip_vs_todrop()) {
98                         /*
99                          * It seems that we are very loaded.
100                          * We have to drop this packet :(
101                          */
102                         ip_vs_service_put(svc);
103                         *verdict = NF_DROP;
104                         return 0;
105                 }
106
107                 /*
108                  * Let the virtual server select a real server for the
109                  * incoming connection, and create a connection entry.
110                  */
111                 *cpp = ip_vs_schedule(svc, skb);
112                 if (!*cpp) {
113                         *verdict = ip_vs_leave(svc, skb, pp);
114                         return 0;
115                 }
116                 ip_vs_service_put(svc);
117         }
118         return 1;
119 }
120
121
122 static inline void
123 udp_fast_csum_update(int af, struct udphdr *uhdr,
124                      const union nf_inet_addr *oldip,
125                      const union nf_inet_addr *newip,
126                      __be16 oldport, __be16 newport)
127 {
128 #ifdef CONFIG_IP_VS_IPV6
129         if (af == AF_INET6)
130                 uhdr->check =
131                         csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
132                                          ip_vs_check_diff2(oldport, newport,
133                                                 ~csum_unfold(uhdr->check))));
134         else
135 #endif
136                 uhdr->check =
137                         csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
138                                          ip_vs_check_diff2(oldport, newport,
139                                                 ~csum_unfold(uhdr->check))));
140         if (!uhdr->check)
141                 uhdr->check = CSUM_MANGLED_0;
142 }
143
144 static int
145 udp_snat_handler(struct sk_buff *skb,
146                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
147 {
148         struct udphdr *udph;
149         unsigned int udphoff;
150
151 #ifdef CONFIG_IP_VS_IPV6
152         if (cp->af == AF_INET6)
153                 udphoff = sizeof(struct ipv6hdr);
154         else
155 #endif
156                 udphoff = ip_hdrlen(skb);
157
158         /* csum_check requires unshared skb */
159         if (!skb_make_writable(skb, udphoff+sizeof(*udph)))
160                 return 0;
161
162         if (unlikely(cp->app != NULL)) {
163                 /* Some checks before mangling */
164                 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
165                         return 0;
166
167                 /*
168                  *      Call application helper if needed
169                  */
170                 if (!ip_vs_app_pkt_out(cp, skb))
171                         return 0;
172         }
173
174         udph = (void *)skb_network_header(skb) + udphoff;
175         udph->source = cp->vport;
176
177         /*
178          *      Adjust UDP checksums
179          */
180         if (!cp->app && (udph->check != 0)) {
181                 /* Only port and addr are changed, do fast csum update */
182                 udp_fast_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr,
183                                      cp->dport, cp->vport);
184                 if (skb->ip_summed == CHECKSUM_COMPLETE)
185                         skb->ip_summed = CHECKSUM_NONE;
186         } else {
187                 /* full checksum calculation */
188                 udph->check = 0;
189                 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
190 #ifdef CONFIG_IP_VS_IPV6
191                 if (cp->af == AF_INET6)
192                         udph->check = csum_ipv6_magic(&cp->vaddr.in6,
193                                                       &cp->caddr.in6,
194                                                       skb->len - udphoff,
195                                                       cp->protocol, skb->csum);
196                 else
197 #endif
198                         udph->check = csum_tcpudp_magic(cp->vaddr.ip,
199                                                         cp->caddr.ip,
200                                                         skb->len - udphoff,
201                                                         cp->protocol,
202                                                         skb->csum);
203                 if (udph->check == 0)
204                         udph->check = CSUM_MANGLED_0;
205                 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
206                           pp->name, udph->check,
207                           (char*)&(udph->check) - (char*)udph);
208         }
209         return 1;
210 }
211
212
213 static int
214 udp_dnat_handler(struct sk_buff *skb,
215                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
216 {
217         struct udphdr *udph;
218         unsigned int udphoff;
219
220 #ifdef CONFIG_IP_VS_IPV6
221         if (cp->af == AF_INET6)
222                 udphoff = sizeof(struct ipv6hdr);
223         else
224 #endif
225                 udphoff = ip_hdrlen(skb);
226
227         /* csum_check requires unshared skb */
228         if (!skb_make_writable(skb, udphoff+sizeof(*udph)))
229                 return 0;
230
231         if (unlikely(cp->app != NULL)) {
232                 /* Some checks before mangling */
233                 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
234                         return 0;
235
236                 /*
237                  *      Attempt ip_vs_app call.
238                  *      It will fix ip_vs_conn
239                  */
240                 if (!ip_vs_app_pkt_in(cp, skb))
241                         return 0;
242         }
243
244         udph = (void *)skb_network_header(skb) + udphoff;
245         udph->dest = cp->dport;
246
247         /*
248          *      Adjust UDP checksums
249          */
250         if (!cp->app && (udph->check != 0)) {
251                 /* Only port and addr are changed, do fast csum update */
252                 udp_fast_csum_update(cp->af, udph, &cp->vaddr, &cp->daddr,
253                                      cp->vport, cp->dport);
254                 if (skb->ip_summed == CHECKSUM_COMPLETE)
255                         skb->ip_summed = CHECKSUM_NONE;
256         } else {
257                 /* full checksum calculation */
258                 udph->check = 0;
259                 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
260 #ifdef CONFIG_IP_VS_IPV6
261                 if (cp->af == AF_INET6)
262                         udph->check = csum_ipv6_magic(&cp->caddr.in6,
263                                                       &cp->daddr.in6,
264                                                       skb->len - udphoff,
265                                                       cp->protocol, skb->csum);
266                 else
267 #endif
268                         udph->check = csum_tcpudp_magic(cp->caddr.ip,
269                                                         cp->daddr.ip,
270                                                         skb->len - udphoff,
271                                                         cp->protocol,
272                                                         skb->csum);
273                 if (udph->check == 0)
274                         udph->check = CSUM_MANGLED_0;
275                 skb->ip_summed = CHECKSUM_UNNECESSARY;
276         }
277         return 1;
278 }
279
280
281 static int
282 udp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp)
283 {
284         struct udphdr _udph, *uh;
285         unsigned int udphoff;
286
287 #ifdef CONFIG_IP_VS_IPV6
288         if (af == AF_INET6)
289                 udphoff = sizeof(struct ipv6hdr);
290         else
291 #endif
292                 udphoff = ip_hdrlen(skb);
293
294         uh = skb_header_pointer(skb, udphoff, sizeof(_udph), &_udph);
295         if (uh == NULL)
296                 return 0;
297
298         if (uh->check != 0) {
299                 switch (skb->ip_summed) {
300                 case CHECKSUM_NONE:
301                         skb->csum = skb_checksum(skb, udphoff,
302                                                  skb->len - udphoff, 0);
303                 case CHECKSUM_COMPLETE:
304 #ifdef CONFIG_IP_VS_IPV6
305                         if (af == AF_INET6) {
306                                 if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
307                                                     &ipv6_hdr(skb)->daddr,
308                                                     skb->len - udphoff,
309                                                     ipv6_hdr(skb)->nexthdr,
310                                                     skb->csum)) {
311                                         IP_VS_DBG_RL_PKT(0, pp, skb, 0,
312                                                          "Failed checksum for");
313                                         return 0;
314                                 }
315                         } else
316 #endif
317                                 if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
318                                                       ip_hdr(skb)->daddr,
319                                                       skb->len - udphoff,
320                                                       ip_hdr(skb)->protocol,
321                                                       skb->csum)) {
322                                         IP_VS_DBG_RL_PKT(0, pp, skb, 0,
323                                                          "Failed checksum for");
324                                         return 0;
325                                 }
326                         break;
327                 default:
328                         /* No need to checksum. */
329                         break;
330                 }
331         }
332         return 1;
333 }
334
335
336 /*
337  *      Note: the caller guarantees that only one of register_app,
338  *      unregister_app or app_conn_bind is called each time.
339  */
340
341 #define UDP_APP_TAB_BITS        4
342 #define UDP_APP_TAB_SIZE        (1 << UDP_APP_TAB_BITS)
343 #define UDP_APP_TAB_MASK        (UDP_APP_TAB_SIZE - 1)
344
345 static struct list_head udp_apps[UDP_APP_TAB_SIZE];
346 static DEFINE_SPINLOCK(udp_app_lock);
347
348 static inline __u16 udp_app_hashkey(__be16 port)
349 {
350         return (((__force u16)port >> UDP_APP_TAB_BITS) ^ (__force u16)port)
351                 & UDP_APP_TAB_MASK;
352 }
353
354
355 static int udp_register_app(struct ip_vs_app *inc)
356 {
357         struct ip_vs_app *i;
358         __u16 hash;
359         __be16 port = inc->port;
360         int ret = 0;
361
362         hash = udp_app_hashkey(port);
363
364
365         spin_lock_bh(&udp_app_lock);
366         list_for_each_entry(i, &udp_apps[hash], p_list) {
367                 if (i->port == port) {
368                         ret = -EEXIST;
369                         goto out;
370                 }
371         }
372         list_add(&inc->p_list, &udp_apps[hash]);
373         atomic_inc(&ip_vs_protocol_udp.appcnt);
374
375   out:
376         spin_unlock_bh(&udp_app_lock);
377         return ret;
378 }
379
380
381 static void
382 udp_unregister_app(struct ip_vs_app *inc)
383 {
384         spin_lock_bh(&udp_app_lock);
385         atomic_dec(&ip_vs_protocol_udp.appcnt);
386         list_del(&inc->p_list);
387         spin_unlock_bh(&udp_app_lock);
388 }
389
390
391 static int udp_app_conn_bind(struct ip_vs_conn *cp)
392 {
393         int hash;
394         struct ip_vs_app *inc;
395         int result = 0;
396
397         /* Default binding: bind app only for NAT */
398         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
399                 return 0;
400
401         /* Lookup application incarnations and bind the right one */
402         hash = udp_app_hashkey(cp->vport);
403
404         spin_lock(&udp_app_lock);
405         list_for_each_entry(inc, &udp_apps[hash], p_list) {
406                 if (inc->port == cp->vport) {
407                         if (unlikely(!ip_vs_app_inc_get(inc)))
408                                 break;
409                         spin_unlock(&udp_app_lock);
410
411                         IP_VS_DBG(9, "%s: Binding conn %u.%u.%u.%u:%u->"
412                                   "%u.%u.%u.%u:%u to app %s on port %u\n",
413                                   __func__,
414                                   NIPQUAD(cp->caddr.ip), ntohs(cp->cport),
415                                   NIPQUAD(cp->vaddr.ip), ntohs(cp->vport),
416                                   inc->name, ntohs(inc->port));
417                         cp->app = inc;
418                         if (inc->init_conn)
419                                 result = inc->init_conn(inc, cp);
420                         goto out;
421                 }
422         }
423         spin_unlock(&udp_app_lock);
424
425   out:
426         return result;
427 }
428
429
430 static int udp_timeouts[IP_VS_UDP_S_LAST+1] = {
431         [IP_VS_UDP_S_NORMAL]            =       5*60*HZ,
432         [IP_VS_UDP_S_LAST]              =       2*HZ,
433 };
434
435 static char * udp_state_name_table[IP_VS_UDP_S_LAST+1] = {
436         [IP_VS_UDP_S_NORMAL]            =       "UDP",
437         [IP_VS_UDP_S_LAST]              =       "BUG!",
438 };
439
440
441 static int
442 udp_set_state_timeout(struct ip_vs_protocol *pp, char *sname, int to)
443 {
444         return ip_vs_set_state_timeout(pp->timeout_table, IP_VS_UDP_S_LAST,
445                                        udp_state_name_table, sname, to);
446 }
447
448 static const char * udp_state_name(int state)
449 {
450         if (state >= IP_VS_UDP_S_LAST)
451                 return "ERR!";
452         return udp_state_name_table[state] ? udp_state_name_table[state] : "?";
453 }
454
455 static int
456 udp_state_transition(struct ip_vs_conn *cp, int direction,
457                      const struct sk_buff *skb,
458                      struct ip_vs_protocol *pp)
459 {
460         cp->timeout = pp->timeout_table[IP_VS_UDP_S_NORMAL];
461         return 1;
462 }
463
464 static void udp_init(struct ip_vs_protocol *pp)
465 {
466         IP_VS_INIT_HASH_TABLE(udp_apps);
467         pp->timeout_table = udp_timeouts;
468 }
469
470 static void udp_exit(struct ip_vs_protocol *pp)
471 {
472 }
473
474
475 struct ip_vs_protocol ip_vs_protocol_udp = {
476         .name =                 "UDP",
477         .protocol =             IPPROTO_UDP,
478         .num_states =           IP_VS_UDP_S_LAST,
479         .dont_defrag =          0,
480         .init =                 udp_init,
481         .exit =                 udp_exit,
482         .conn_schedule =        udp_conn_schedule,
483         .conn_in_get =          udp_conn_in_get,
484         .conn_out_get =         udp_conn_out_get,
485         .snat_handler =         udp_snat_handler,
486         .dnat_handler =         udp_dnat_handler,
487         .csum_check =           udp_csum_check,
488         .state_transition =     udp_state_transition,
489         .state_name =           udp_state_name,
490         .register_app =         udp_register_app,
491         .unregister_app =       udp_unregister_app,
492         .app_conn_bind =        udp_app_conn_bind,
493         .debug_packet =         ip_vs_tcpudp_debug_packet,
494         .timeout_change =       NULL,
495         .set_state_timeout =    udp_set_state_timeout,
496 };