Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[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  * Version:     $Id: ip_vs_proto_udp.c,v 1.3 2002/11/30 01:50:35 wensong Exp $
5  *
6  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
7  *              Julian Anastasov <ja@ssi.bg>
8  *
9  *              This program is free software; you can redistribute it and/or
10  *              modify it under the terms of the GNU General Public License
11  *              as published by the Free Software Foundation; either version
12  *              2 of the License, or (at your option) any later version.
13  *
14  * Changes:
15  *
16  */
17
18 #include <linux/in.h>
19 #include <linux/ip.h>
20 #include <linux/kernel.h>
21 #include <linux/netfilter_ipv4.h>
22 #include <linux/udp.h>
23
24 #include <net/ip_vs.h>
25 #include <net/ip.h>
26
27 static struct ip_vs_conn *
28 udp_conn_in_get(const struct sk_buff *skb, struct ip_vs_protocol *pp,
29                 const struct iphdr *iph, unsigned int proto_off, 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(iph->protocol,
40                                        iph->saddr, pptr[0],
41                                        iph->daddr, pptr[1]);
42         } else {
43                 cp = ip_vs_conn_in_get(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(const struct sk_buff *skb, struct ip_vs_protocol *pp,
54                  const struct iphdr *iph, unsigned int proto_off, int inverse)
55 {
56         struct ip_vs_conn *cp;
57         __be16 _ports[2], *pptr;
58
59         pptr = skb_header_pointer(skb, ip_hdrlen(skb),
60                                   sizeof(_ports), _ports);
61         if (pptr == NULL)
62                 return NULL;
63
64         if (likely(!inverse)) {
65                 cp = ip_vs_conn_out_get(iph->protocol,
66                                         iph->saddr, pptr[0],
67                                         iph->daddr, pptr[1]);
68         } else {
69                 cp = ip_vs_conn_out_get(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(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
85         uh = skb_header_pointer(skb, ip_hdrlen(skb),
86                                 sizeof(_udph), &_udph);
87         if (uh == NULL) {
88                 *verdict = NF_DROP;
89                 return 0;
90         }
91
92         if ((svc = ip_vs_service_get(skb->mark, ip_hdr(skb)->protocol,
93                                      ip_hdr(skb)->daddr, uh->dest))) {
94                 if (ip_vs_todrop()) {
95                         /*
96                          * It seems that we are very loaded.
97                          * We have to drop this packet :(
98                          */
99                         ip_vs_service_put(svc);
100                         *verdict = NF_DROP;
101                         return 0;
102                 }
103
104                 /*
105                  * Let the virtual server select a real server for the
106                  * incoming connection, and create a connection entry.
107                  */
108                 *cpp = ip_vs_schedule(svc, skb);
109                 if (!*cpp) {
110                         *verdict = ip_vs_leave(svc, skb, pp);
111                         return 0;
112                 }
113                 ip_vs_service_put(svc);
114         }
115         return 1;
116 }
117
118
119 static inline void
120 udp_fast_csum_update(struct udphdr *uhdr, __be32 oldip, __be32 newip,
121                      __be16 oldport, __be16 newport)
122 {
123         uhdr->check =
124                 csum_fold(ip_vs_check_diff4(oldip, newip,
125                                  ip_vs_check_diff2(oldport, newport,
126                                         ~csum_unfold(uhdr->check))));
127         if (!uhdr->check)
128                 uhdr->check = CSUM_MANGLED_0;
129 }
130
131 static int
132 udp_snat_handler(struct sk_buff **pskb,
133                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
134 {
135         struct udphdr *udph;
136         const unsigned int udphoff = ip_hdrlen(*pskb);
137
138         /* csum_check requires unshared skb */
139         if (!ip_vs_make_skb_writable(pskb, udphoff+sizeof(*udph)))
140                 return 0;
141
142         if (unlikely(cp->app != NULL)) {
143                 /* Some checks before mangling */
144                 if (pp->csum_check && !pp->csum_check(*pskb, pp))
145                         return 0;
146
147                 /*
148                  *      Call application helper if needed
149                  */
150                 if (!ip_vs_app_pkt_out(cp, pskb))
151                         return 0;
152         }
153
154         udph = (void *)ip_hdr(*pskb) + udphoff;
155         udph->source = cp->vport;
156
157         /*
158          *      Adjust UDP checksums
159          */
160         if (!cp->app && (udph->check != 0)) {
161                 /* Only port and addr are changed, do fast csum update */
162                 udp_fast_csum_update(udph, cp->daddr, cp->vaddr,
163                                      cp->dport, cp->vport);
164                 if ((*pskb)->ip_summed == CHECKSUM_COMPLETE)
165                         (*pskb)->ip_summed = CHECKSUM_NONE;
166         } else {
167                 /* full checksum calculation */
168                 udph->check = 0;
169                 (*pskb)->csum = skb_checksum(*pskb, udphoff,
170                                              (*pskb)->len - udphoff, 0);
171                 udph->check = csum_tcpudp_magic(cp->vaddr, cp->caddr,
172                                                 (*pskb)->len - udphoff,
173                                                 cp->protocol,
174                                                 (*pskb)->csum);
175                 if (udph->check == 0)
176                         udph->check = CSUM_MANGLED_0;
177                 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
178                           pp->name, udph->check,
179                           (char*)&(udph->check) - (char*)udph);
180         }
181         return 1;
182 }
183
184
185 static int
186 udp_dnat_handler(struct sk_buff **pskb,
187                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
188 {
189         struct udphdr *udph;
190         unsigned int udphoff = ip_hdrlen(*pskb);
191
192         /* csum_check requires unshared skb */
193         if (!ip_vs_make_skb_writable(pskb, udphoff+sizeof(*udph)))
194                 return 0;
195
196         if (unlikely(cp->app != NULL)) {
197                 /* Some checks before mangling */
198                 if (pp->csum_check && !pp->csum_check(*pskb, pp))
199                         return 0;
200
201                 /*
202                  *      Attempt ip_vs_app call.
203                  *      It will fix ip_vs_conn
204                  */
205                 if (!ip_vs_app_pkt_in(cp, pskb))
206                         return 0;
207         }
208
209         udph = (void *)ip_hdr(*pskb) + udphoff;
210         udph->dest = cp->dport;
211
212         /*
213          *      Adjust UDP checksums
214          */
215         if (!cp->app && (udph->check != 0)) {
216                 /* Only port and addr are changed, do fast csum update */
217                 udp_fast_csum_update(udph, cp->vaddr, cp->daddr,
218                                      cp->vport, cp->dport);
219                 if ((*pskb)->ip_summed == CHECKSUM_COMPLETE)
220                         (*pskb)->ip_summed = CHECKSUM_NONE;
221         } else {
222                 /* full checksum calculation */
223                 udph->check = 0;
224                 (*pskb)->csum = skb_checksum(*pskb, udphoff,
225                                              (*pskb)->len - udphoff, 0);
226                 udph->check = csum_tcpudp_magic(cp->caddr, cp->daddr,
227                                                 (*pskb)->len - udphoff,
228                                                 cp->protocol,
229                                                 (*pskb)->csum);
230                 if (udph->check == 0)
231                         udph->check = CSUM_MANGLED_0;
232                 (*pskb)->ip_summed = CHECKSUM_UNNECESSARY;
233         }
234         return 1;
235 }
236
237
238 static int
239 udp_csum_check(struct sk_buff *skb, struct ip_vs_protocol *pp)
240 {
241         struct udphdr _udph, *uh;
242         const unsigned int udphoff = ip_hdrlen(skb);
243
244         uh = skb_header_pointer(skb, udphoff, sizeof(_udph), &_udph);
245         if (uh == NULL)
246                 return 0;
247
248         if (uh->check != 0) {
249                 switch (skb->ip_summed) {
250                 case CHECKSUM_NONE:
251                         skb->csum = skb_checksum(skb, udphoff,
252                                                  skb->len - udphoff, 0);
253                 case CHECKSUM_COMPLETE:
254                         if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
255                                               ip_hdr(skb)->daddr,
256                                               skb->len - udphoff,
257                                               ip_hdr(skb)->protocol,
258                                               skb->csum)) {
259                                 IP_VS_DBG_RL_PKT(0, pp, skb, 0,
260                                                  "Failed checksum for");
261                                 return 0;
262                         }
263                         break;
264                 default:
265                         /* No need to checksum. */
266                         break;
267                 }
268         }
269         return 1;
270 }
271
272
273 /*
274  *      Note: the caller guarantees that only one of register_app,
275  *      unregister_app or app_conn_bind is called each time.
276  */
277
278 #define UDP_APP_TAB_BITS        4
279 #define UDP_APP_TAB_SIZE        (1 << UDP_APP_TAB_BITS)
280 #define UDP_APP_TAB_MASK        (UDP_APP_TAB_SIZE - 1)
281
282 static struct list_head udp_apps[UDP_APP_TAB_SIZE];
283 static DEFINE_SPINLOCK(udp_app_lock);
284
285 static inline __u16 udp_app_hashkey(__be16 port)
286 {
287         return (((__force u16)port >> UDP_APP_TAB_BITS) ^ (__force u16)port)
288                 & UDP_APP_TAB_MASK;
289 }
290
291
292 static int udp_register_app(struct ip_vs_app *inc)
293 {
294         struct ip_vs_app *i;
295         __u16 hash;
296         __be16 port = inc->port;
297         int ret = 0;
298
299         hash = udp_app_hashkey(port);
300
301
302         spin_lock_bh(&udp_app_lock);
303         list_for_each_entry(i, &udp_apps[hash], p_list) {
304                 if (i->port == port) {
305                         ret = -EEXIST;
306                         goto out;
307                 }
308         }
309         list_add(&inc->p_list, &udp_apps[hash]);
310         atomic_inc(&ip_vs_protocol_udp.appcnt);
311
312   out:
313         spin_unlock_bh(&udp_app_lock);
314         return ret;
315 }
316
317
318 static void
319 udp_unregister_app(struct ip_vs_app *inc)
320 {
321         spin_lock_bh(&udp_app_lock);
322         atomic_dec(&ip_vs_protocol_udp.appcnt);
323         list_del(&inc->p_list);
324         spin_unlock_bh(&udp_app_lock);
325 }
326
327
328 static int udp_app_conn_bind(struct ip_vs_conn *cp)
329 {
330         int hash;
331         struct ip_vs_app *inc;
332         int result = 0;
333
334         /* Default binding: bind app only for NAT */
335         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
336                 return 0;
337
338         /* Lookup application incarnations and bind the right one */
339         hash = udp_app_hashkey(cp->vport);
340
341         spin_lock(&udp_app_lock);
342         list_for_each_entry(inc, &udp_apps[hash], p_list) {
343                 if (inc->port == cp->vport) {
344                         if (unlikely(!ip_vs_app_inc_get(inc)))
345                                 break;
346                         spin_unlock(&udp_app_lock);
347
348                         IP_VS_DBG(9, "%s: Binding conn %u.%u.%u.%u:%u->"
349                                   "%u.%u.%u.%u:%u to app %s on port %u\n",
350                                   __FUNCTION__,
351                                   NIPQUAD(cp->caddr), ntohs(cp->cport),
352                                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
353                                   inc->name, ntohs(inc->port));
354                         cp->app = inc;
355                         if (inc->init_conn)
356                                 result = inc->init_conn(inc, cp);
357                         goto out;
358                 }
359         }
360         spin_unlock(&udp_app_lock);
361
362   out:
363         return result;
364 }
365
366
367 static int udp_timeouts[IP_VS_UDP_S_LAST+1] = {
368         [IP_VS_UDP_S_NORMAL]            =       5*60*HZ,
369         [IP_VS_UDP_S_LAST]              =       2*HZ,
370 };
371
372 static char * udp_state_name_table[IP_VS_UDP_S_LAST+1] = {
373         [IP_VS_UDP_S_NORMAL]            =       "UDP",
374         [IP_VS_UDP_S_LAST]              =       "BUG!",
375 };
376
377
378 static int
379 udp_set_state_timeout(struct ip_vs_protocol *pp, char *sname, int to)
380 {
381         return ip_vs_set_state_timeout(pp->timeout_table, IP_VS_UDP_S_LAST,
382                                        udp_state_name_table, sname, to);
383 }
384
385 static const char * udp_state_name(int state)
386 {
387         if (state >= IP_VS_UDP_S_LAST)
388                 return "ERR!";
389         return udp_state_name_table[state] ? udp_state_name_table[state] : "?";
390 }
391
392 static int
393 udp_state_transition(struct ip_vs_conn *cp, int direction,
394                      const struct sk_buff *skb,
395                      struct ip_vs_protocol *pp)
396 {
397         cp->timeout = pp->timeout_table[IP_VS_UDP_S_NORMAL];
398         return 1;
399 }
400
401 static void udp_init(struct ip_vs_protocol *pp)
402 {
403         IP_VS_INIT_HASH_TABLE(udp_apps);
404         pp->timeout_table = udp_timeouts;
405 }
406
407 static void udp_exit(struct ip_vs_protocol *pp)
408 {
409 }
410
411
412 struct ip_vs_protocol ip_vs_protocol_udp = {
413         .name =                 "UDP",
414         .protocol =             IPPROTO_UDP,
415         .dont_defrag =          0,
416         .init =                 udp_init,
417         .exit =                 udp_exit,
418         .conn_schedule =        udp_conn_schedule,
419         .conn_in_get =          udp_conn_in_get,
420         .conn_out_get =         udp_conn_out_get,
421         .snat_handler =         udp_snat_handler,
422         .dnat_handler =         udp_dnat_handler,
423         .csum_check =           udp_csum_check,
424         .state_transition =     udp_state_transition,
425         .state_name =           udp_state_name,
426         .register_app =         udp_register_app,
427         .unregister_app =       udp_unregister_app,
428         .app_conn_bind =        udp_app_conn_bind,
429         .debug_packet =         ip_vs_tcpudp_debug_packet,
430         .timeout_change =       NULL,
431         .set_state_timeout =    udp_set_state_timeout,
432 };