?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
IPVS: Move IPVS to net/netfilter/ipvs
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / netfilter / ipvs / ip_vs_proto_udp.c
blob6eb6039d63434f6ecc4bd511edb390e0e91bbd6a
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 #include <net/ip6_checksum.h>
26
27 static struct ip_vs_conn *
28 udp_conn_in_get(int af, const struct sk_buff *skb, struct ip_vs_protocol *pp,
29 const struct ip_vs_iphdr *iph, unsigned int proto_off,
30 int inverse)
31 {
32 struct ip_vs_conn *cp;
33 __be16 _ports[2], *pptr;
34
35 pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
36 if (pptr == NULL)
37 return NULL;
38
39 if (likely(!inverse)) {
40 cp = ip_vs_conn_in_get(af, iph->protocol,
41 &iph->saddr, pptr[0],
42 &iph->daddr, pptr[1]);
43 } else {
44 cp = ip_vs_conn_in_get(af, iph->protocol,
45 &iph->daddr, pptr[1],
46 &iph->saddr, pptr[0]);
47 }
48
49 return cp;
50 }
51
52
53 static struct ip_vs_conn *
54 udp_conn_out_get(int af, const struct sk_buff *skb, struct ip_vs_protocol *pp,
55 const struct ip_vs_iphdr *iph, unsigned int proto_off,
56 int inverse)
57 {
58 struct ip_vs_conn *cp;
59 __be16 _ports[2], *pptr;
60
61 pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
62 if (pptr == NULL)
63 return NULL;
64
65 if (likely(!inverse)) {
66 cp = ip_vs_conn_out_get(af, iph->protocol,
67 &iph->saddr, pptr[0],
68 &iph->daddr, pptr[1]);
69 } else {
70 cp = ip_vs_conn_out_get(af, iph->protocol,
71 &iph->daddr, pptr[1],
72 &iph->saddr, pptr[0]);
73 }
74
75 return cp;
76 }
77
78
79 static int
80 udp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_protocol *pp,
81 int *verdict, struct ip_vs_conn **cpp)
82 {
83 struct ip_vs_service *svc;
84 struct udphdr _udph, *uh;
85 struct ip_vs_iphdr iph;
86
87 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
88
89 uh = skb_header_pointer(skb, iph.len, sizeof(_udph), &_udph);
90 if (uh == NULL) {
91 *verdict = NF_DROP;
92 return 0;
93 }
94
95 svc = ip_vs_service_get(af, skb->mark, iph.protocol,
96 &iph.daddr, uh->dest);
97 if (svc) {
98 if (ip_vs_todrop()) {
99 /*
100 * It seems that we are very loaded.
101 * We have to drop this packet :(
102 */
103 ip_vs_service_put(svc);
104 *verdict = NF_DROP;
105 return 0;
106 }
107
108 /*
109 * Let the virtual server select a real server for the
110 * incoming connection, and create a connection entry.
111 */
112 *cpp = ip_vs_schedule(svc, skb);
113 if (!*cpp) {
114 *verdict = ip_vs_leave(svc, skb, pp);
115 return 0;
116 }
117 ip_vs_service_put(svc);
118 }
119 return 1;
120 }
121
122
123 static inline void
124 udp_fast_csum_update(int af, struct udphdr *uhdr,
125 const union nf_inet_addr *oldip,
126 const union nf_inet_addr *newip,
127 __be16 oldport, __be16 newport)
128 {
129 #ifdef CONFIG_IP_VS_IPV6
130 if (af == AF_INET6)
131 uhdr->check =
132 csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
133 ip_vs_check_diff2(oldport, newport,
134 ~csum_unfold(uhdr->check))));
135 else
136 #endif
137 uhdr->check =
138 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
139 ip_vs_check_diff2(oldport, newport,
140 ~csum_unfold(uhdr->check))));
141 if (!uhdr->check)
142 uhdr->check = CSUM_MANGLED_0;
143 }
144
145 static inline void
146 udp_partial_csum_update(int af, struct udphdr *uhdr,
147 const union nf_inet_addr *oldip,
148 const union nf_inet_addr *newip,
149 __be16 oldlen, __be16 newlen)
150 {
151 #ifdef CONFIG_IP_VS_IPV6
152 if (af == AF_INET6)
153 uhdr->check =
154 csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
155 ip_vs_check_diff2(oldlen, newlen,
156 ~csum_unfold(uhdr->check))));
157 else
158 #endif
159 uhdr->check =
160 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
161 ip_vs_check_diff2(oldlen, newlen,
162 ~csum_unfold(uhdr->check))));
163 }
164
165
166 static int
167 udp_snat_handler(struct sk_buff *skb,
168 struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
169 {
170 struct udphdr *udph;
171 unsigned int udphoff;
172 int oldlen;
173
174 #ifdef CONFIG_IP_VS_IPV6
175 if (cp->af == AF_INET6)
176 udphoff = sizeof(struct ipv6hdr);
177 else
178 #endif
179 udphoff = ip_hdrlen(skb);
180 oldlen = skb->len - udphoff;
181
182 /* csum_check requires unshared skb */
183 if (!skb_make_writable(skb, udphoff+sizeof(*udph)))
184 return 0;
185
186 if (unlikely(cp->app != NULL)) {
187 /* Some checks before mangling */
188 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
189 return 0;
190
191 /*
192 * Call application helper if needed
193 */
194 if (!ip_vs_app_pkt_out(cp, skb))
195 return 0;
196 }
197
198 udph = (void *)skb_network_header(skb) + udphoff;
199 udph->source = cp->vport;
200
201 /*
202 * Adjust UDP checksums
203 */
204 if (skb->ip_summed == CHECKSUM_PARTIAL) {
205 udp_partial_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr,
206 htonl(oldlen),
207 htonl(skb->len - udphoff));
208 } else if (!cp->app && (udph->check != 0)) {
209 /* Only port and addr are changed, do fast csum update */
210 udp_fast_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr,
211 cp->dport, cp->vport);
212 if (skb->ip_summed == CHECKSUM_COMPLETE)
213 skb->ip_summed = CHECKSUM_NONE;
214 } else {
215 /* full checksum calculation */
216 udph->check = 0;
217 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
218 #ifdef CONFIG_IP_VS_IPV6
219 if (cp->af == AF_INET6)
220 udph->check = csum_ipv6_magic(&cp->vaddr.in6,
221 &cp->caddr.in6,
222 skb->len - udphoff,
223 cp->protocol, skb->csum);
224 else
225 #endif
226 udph->check = csum_tcpudp_magic(cp->vaddr.ip,
227 cp->caddr.ip,
228 skb->len - udphoff,
229 cp->protocol,
230 skb->csum);
231 if (udph->check == 0)
232 udph->check = CSUM_MANGLED_0;
233 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
234 pp->name, udph->check,
235 (char*)&(udph->check) - (char*)udph);
236 }
237 return 1;
238 }
239
240
241 static int
242 udp_dnat_handler(struct sk_buff *skb,
243 struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
244 {
245 struct udphdr *udph;
246 unsigned int udphoff;
247 int oldlen;
248
249 #ifdef CONFIG_IP_VS_IPV6
250 if (cp->af == AF_INET6)
251 udphoff = sizeof(struct ipv6hdr);
252 else
253 #endif
254 udphoff = ip_hdrlen(skb);
255 oldlen = skb->len - udphoff;
256
257 /* csum_check requires unshared skb */
258 if (!skb_make_writable(skb, udphoff+sizeof(*udph)))
259 return 0;
260
261 if (unlikely(cp->app != NULL)) {
262 /* Some checks before mangling */
263 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
264 return 0;
265
266 /*
267 * Attempt ip_vs_app call.
268 * It will fix ip_vs_conn
269 */
270 if (!ip_vs_app_pkt_in(cp, skb))
271 return 0;
272 }
273
274 udph = (void *)skb_network_header(skb) + udphoff;
275 udph->dest = cp->dport;
276
277 /*
278 * Adjust UDP checksums
279 */
280 if (skb->ip_summed == CHECKSUM_PARTIAL) {
281 udp_partial_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr,
282 htonl(oldlen),
283 htonl(skb->len - udphoff));
284 } else if (!cp->app && (udph->check != 0)) {
285 /* Only port and addr are changed, do fast csum update */
286 udp_fast_csum_update(cp->af, udph, &cp->vaddr, &cp->daddr,
287 cp->vport, cp->dport);
288 if (skb->ip_summed == CHECKSUM_COMPLETE)
289 skb->ip_summed = CHECKSUM_NONE;
290 } else {
291 /* full checksum calculation */
292 udph->check = 0;
293 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
294 #ifdef CONFIG_IP_VS_IPV6
295 if (cp->af == AF_INET6)
296 udph->check = csum_ipv6_magic(&cp->caddr.in6,
297 &cp->daddr.in6,
298 skb->len - udphoff,
299 cp->protocol, skb->csum);
300 else
301 #endif
302 udph->check = csum_tcpudp_magic(cp->caddr.ip,
303 cp->daddr.ip,
304 skb->len - udphoff,
305 cp->protocol,
306 skb->csum);
307 if (udph->check == 0)
308 udph->check = CSUM_MANGLED_0;
309 skb->ip_summed = CHECKSUM_UNNECESSARY;
310 }
311 return 1;
312 }
313
314
315 static int
316 udp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp)
317 {
318 struct udphdr _udph, *uh;
319 unsigned int udphoff;
320
321 #ifdef CONFIG_IP_VS_IPV6
322 if (af == AF_INET6)
323 udphoff = sizeof(struct ipv6hdr);
324 else
325 #endif
326 udphoff = ip_hdrlen(skb);
327
328 uh = skb_header_pointer(skb, udphoff, sizeof(_udph), &_udph);
329 if (uh == NULL)
330 return 0;
331
332 if (uh->check != 0) {
333 switch (skb->ip_summed) {
334 case CHECKSUM_NONE:
335 skb->csum = skb_checksum(skb, udphoff,
336 skb->len - udphoff, 0);
337 case CHECKSUM_COMPLETE:
338 #ifdef CONFIG_IP_VS_IPV6
339 if (af == AF_INET6) {
340 if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
341 &ipv6_hdr(skb)->daddr,
342 skb->len - udphoff,
343 ipv6_hdr(skb)->nexthdr,
344 skb->csum)) {
345 IP_VS_DBG_RL_PKT(0, pp, skb, 0,
346 "Failed checksum for");
347 return 0;
348 }
349 } else
350 #endif
351 if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
352 ip_hdr(skb)->daddr,
353 skb->len - udphoff,
354 ip_hdr(skb)->protocol,
355 skb->csum)) {
356 IP_VS_DBG_RL_PKT(0, pp, skb, 0,
357 "Failed checksum for");
358 return 0;
359 }
360 break;
361 default:
362 /* No need to checksum. */
363 break;
364 }
365 }
366 return 1;
367 }
368
369
370 /*
371 * Note: the caller guarantees that only one of register_app,
372 * unregister_app or app_conn_bind is called each time.
373 */
374
375 #define UDP_APP_TAB_BITS 4
376 #define UDP_APP_TAB_SIZE (1 << UDP_APP_TAB_BITS)
377 #define UDP_APP_TAB_MASK (UDP_APP_TAB_SIZE - 1)
378
379 static struct list_head udp_apps[UDP_APP_TAB_SIZE];
380 static DEFINE_SPINLOCK(udp_app_lock);
381
382 static inline __u16 udp_app_hashkey(__be16 port)
383 {
384 return (((__force u16)port >> UDP_APP_TAB_BITS) ^ (__force u16)port)
385 & UDP_APP_TAB_MASK;
386 }
387
388
389 static int udp_register_app(struct ip_vs_app *inc)
390 {
391 struct ip_vs_app *i;
392 __u16 hash;
393 __be16 port = inc->port;
394 int ret = 0;
395
396 hash = udp_app_hashkey(port);
397
398
399 spin_lock_bh(&udp_app_lock);
400 list_for_each_entry(i, &udp_apps[hash], p_list) {
401 if (i->port == port) {
402 ret = -EEXIST;
403 goto out;
404 }
405 }
406 list_add(&inc->p_list, &udp_apps[hash]);
407 atomic_inc(&ip_vs_protocol_udp.appcnt);
408
409 out:
410 spin_unlock_bh(&udp_app_lock);
411 return ret;
412 }
413
414
415 static void
416 udp_unregister_app(struct ip_vs_app *inc)
417 {
418 spin_lock_bh(&udp_app_lock);
419 atomic_dec(&ip_vs_protocol_udp.appcnt);
420 list_del(&inc->p_list);
421 spin_unlock_bh(&udp_app_lock);
422 }
423
424
425 static int udp_app_conn_bind(struct ip_vs_conn *cp)
426 {
427 int hash;
428 struct ip_vs_app *inc;
429 int result = 0;
430
431 /* Default binding: bind app only for NAT */
432 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
433 return 0;
434
435 /* Lookup application incarnations and bind the right one */
436 hash = udp_app_hashkey(cp->vport);
437
438 spin_lock(&udp_app_lock);
439 list_for_each_entry(inc, &udp_apps[hash], p_list) {
440 if (inc->port == cp->vport) {
441 if (unlikely(!ip_vs_app_inc_get(inc)))
442 break;
443 spin_unlock(&udp_app_lock);
444
445 IP_VS_DBG_BUF(9, "%s: Binding conn %s:%u->"
446 "%s:%u to app %s on port %u\n",
447 __func__,
448 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
449 ntohs(cp->cport),
450 IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
451 ntohs(cp->vport),
452 inc->name, ntohs(inc->port));
453
454 cp->app = inc;
455 if (inc->init_conn)
456 result = inc->init_conn(inc, cp);
457 goto out;
458 }
459 }
460 spin_unlock(&udp_app_lock);
461
462 out:
463 return result;
464 }
465
466
467 static int udp_timeouts[IP_VS_UDP_S_LAST+1] = {
468 [IP_VS_UDP_S_NORMAL] = 5*60*HZ,
469 [IP_VS_UDP_S_LAST] = 2*HZ,
470 };
471
472 static char * udp_state_name_table[IP_VS_UDP_S_LAST+1] = {
473 [IP_VS_UDP_S_NORMAL] = "UDP",
474 [IP_VS_UDP_S_LAST] = "BUG!",
475 };
476
477
478 static int
479 udp_set_state_timeout(struct ip_vs_protocol *pp, char *sname, int to)
480 {
481 return ip_vs_set_state_timeout(pp->timeout_table, IP_VS_UDP_S_LAST,
482 udp_state_name_table, sname, to);
483 }
484
485 static const char * udp_state_name(int state)
486 {
487 if (state >= IP_VS_UDP_S_LAST)
488 return "ERR!";
489 return udp_state_name_table[state] ? udp_state_name_table[state] : "?";
490 }
491
492 static int
493 udp_state_transition(struct ip_vs_conn *cp, int direction,
494 const struct sk_buff *skb,
495 struct ip_vs_protocol *pp)
496 {
497 cp->timeout = pp->timeout_table[IP_VS_UDP_S_NORMAL];
498 return 1;
499 }
500
501 static void udp_init(struct ip_vs_protocol *pp)
502 {
503 IP_VS_INIT_HASH_TABLE(udp_apps);
504 pp->timeout_table = udp_timeouts;
505 }
506
507 static void udp_exit(struct ip_vs_protocol *pp)
508 {
509 }
510
511
512 struct ip_vs_protocol ip_vs_protocol_udp = {
513 .name = "UDP",
514 .protocol = IPPROTO_UDP,
515 .num_states = IP_VS_UDP_S_LAST,
516 .dont_defrag = 0,
517 .init = udp_init,
518 .exit = udp_exit,
519 .conn_schedule = udp_conn_schedule,
520 .conn_in_get = udp_conn_in_get,
521 .conn_out_get = udp_conn_out_get,
522 .snat_handler = udp_snat_handler,
523 .dnat_handler = udp_dnat_handler,
524 .csum_check = udp_csum_check,
525 .state_transition = udp_state_transition,
526 .state_name = udp_state_name,
527 .register_app = udp_register_app,
528 .unregister_app = udp_unregister_app,
529 .app_conn_bind = udp_app_conn_bind,
530 .debug_packet = ip_vs_tcpudp_debug_packet,
531 .timeout_change = NULL,
532 .set_state_timeout = udp_set_state_timeout,
533 };
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree