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: Hans Schillstrom <hans.schillstrom@ericsson.com>
13 * Network name space (netns) aware.
14 *
15 */
16
17 #define KMSG_COMPONENT "IPVS"
18 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
19
20 #include <linux/in.h>
21 #include <linux/ip.h>
22 #include <linux/kernel.h>
23 #include <linux/netfilter.h>
24 #include <linux/netfilter_ipv4.h>
25 #include <linux/udp.h>
26
27 #include <net/ip_vs.h>
28 #include <net/ip.h>
29 #include <net/ip6_checksum.h>
30
31 static int
32 udp_conn_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
33 struct ip_vs_proto_data *pd,
34 int *verdict, struct ip_vs_conn **cpp,
35 struct ip_vs_iphdr *iph)
36 {
37 struct ip_vs_service *svc;
38 struct udphdr _udph, *uh;
39 __be16 _ports[2], *ports = NULL;
40
41 if (likely(!ip_vs_iph_icmp(iph))) {
42 /* IPv6 fragments, only first fragment will hit this */
43 uh = skb_header_pointer(skb, iph->len, sizeof(_udph), &_udph);
44 if (uh)
45 ports = &uh->source;
46 } else {
47 ports = skb_header_pointer(
48 skb, iph->len, sizeof(_ports), &_ports);
49 }
50
51 if (!ports) {
52 *verdict = NF_DROP;
53 return 0;
54 }
55
56 if (likely(!ip_vs_iph_inverse(iph)))
57 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol,
58 &iph->daddr, ports[1]);
59 else
60 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol,
61 &iph->saddr, ports[0]);
62
63 if (svc) {
64 int ignored;
65
66 if (ip_vs_todrop(ipvs)) {
67 /*
68 * It seems that we are very loaded.
69 * We have to drop this packet :(
70 */
71 *verdict = NF_DROP;
72 return 0;
73 }
74
75 /*
76 * Let the virtual server select a real server for the
77 * incoming connection, and create a connection entry.
78 */
79 *cpp = ip_vs_schedule(svc, skb, pd, &ignored, iph);
80 if (!*cpp && ignored <= 0) {
81 if (!ignored)
82 *verdict = ip_vs_leave(svc, skb, pd, iph);
83 else
84 *verdict = NF_DROP;
85 return 0;
86 }
87 }
88 /* NF_ACCEPT */
89 return 1;
90 }
91
92
93 static inline void
94 udp_fast_csum_update(int af, struct udphdr *uhdr,
95 const union nf_inet_addr *oldip,
96 const union nf_inet_addr *newip,
97 __be16 oldport, __be16 newport)
98 {
99 #ifdef CONFIG_IP_VS_IPV6
100 if (af == AF_INET6)
101 uhdr->check =
102 csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
103 ip_vs_check_diff2(oldport, newport,
104 ~csum_unfold(uhdr->check))));
105 else
106 #endif
107 uhdr->check =
108 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
109 ip_vs_check_diff2(oldport, newport,
110 ~csum_unfold(uhdr->check))));
111 if (!uhdr->check)
112 uhdr->check = CSUM_MANGLED_0;
113 }
114
115 static inline void
116 udp_partial_csum_update(int af, struct udphdr *uhdr,
117 const union nf_inet_addr *oldip,
118 const union nf_inet_addr *newip,
119 __be16 oldlen, __be16 newlen)
120 {
121 #ifdef CONFIG_IP_VS_IPV6
122 if (af == AF_INET6)
123 uhdr->check =
124 ~csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
125 ip_vs_check_diff2(oldlen, newlen,
126 csum_unfold(uhdr->check))));
127 else
128 #endif
129 uhdr->check =
130 ~csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
131 ip_vs_check_diff2(oldlen, newlen,
132 csum_unfold(uhdr->check))));
133 }
134
135
136 static int
137 udp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
138 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
139 {
140 struct udphdr *udph;
141 unsigned int udphoff = iph->len;
142 int oldlen;
143 int payload_csum = 0;
144
145 #ifdef CONFIG_IP_VS_IPV6
146 if (cp->af == AF_INET6 && iph->fragoffs)
147 return 1;
148 #endif
149 oldlen = skb->len - udphoff;
150
151 /* csum_check requires unshared skb */
152 if (!skb_make_writable(skb, udphoff+sizeof(*udph)))
153 return 0;
154
155 if (unlikely(cp->app != NULL)) {
156 int ret;
157
158 /* Some checks before mangling */
159 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
160 return 0;
161
162 /*
163 * Call application helper if needed
164 */
165 if (!(ret = ip_vs_app_pkt_out(cp, skb)))
166 return 0;
167 /* ret=2: csum update is needed after payload mangling */
168 if (ret == 1)
169 oldlen = skb->len - udphoff;
170 else
171 payload_csum = 1;
172 }
173
174 udph = (void *)skb_network_header(skb) + udphoff;
175 udph->source = cp->vport;
176
177 /*
178 * Adjust UDP checksums
179 */
180 if (skb->ip_summed == CHECKSUM_PARTIAL) {
181 udp_partial_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr,
182 htons(oldlen),
183 htons(skb->len - udphoff));
184 } else if (!payload_csum && (udph->check != 0)) {
185 /* Only port and addr are changed, do fast csum update */
186 udp_fast_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr,
187 cp->dport, cp->vport);
188 if (skb->ip_summed == CHECKSUM_COMPLETE)
189 skb->ip_summed = (cp->app && pp->csum_check) ?
190 CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
191 } else {
192 /* full checksum calculation */
193 udph->check = 0;
194 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
195 #ifdef CONFIG_IP_VS_IPV6
196 if (cp->af == AF_INET6)
197 udph->check = csum_ipv6_magic(&cp->vaddr.in6,
198 &cp->caddr.in6,
199 skb->len - udphoff,
200 cp->protocol, skb->csum);
201 else
202 #endif
203 udph->check = csum_tcpudp_magic(cp->vaddr.ip,
204 cp->caddr.ip,
205 skb->len - udphoff,
206 cp->protocol,
207 skb->csum);
208 if (udph->check == 0)
209 udph->check = CSUM_MANGLED_0;
210 skb->ip_summed = CHECKSUM_UNNECESSARY;
211 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
212 pp->name, udph->check,
213 (char*)&(udph->check) - (char*)udph);
214 }
215 return 1;
216 }
217
218
219 static int
220 udp_dnat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
221 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
222 {
223 struct udphdr *udph;
224 unsigned int udphoff = iph->len;
225 int oldlen;
226 int payload_csum = 0;
227
228 #ifdef CONFIG_IP_VS_IPV6
229 if (cp->af == AF_INET6 && iph->fragoffs)
230 return 1;
231 #endif
232 oldlen = skb->len - udphoff;
233
234 /* csum_check requires unshared skb */
235 if (!skb_make_writable(skb, udphoff+sizeof(*udph)))
236 return 0;
237
238 if (unlikely(cp->app != NULL)) {
239 int ret;
240
241 /* Some checks before mangling */
242 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
243 return 0;
244
245 /*
246 * Attempt ip_vs_app call.
247 * It will fix ip_vs_conn
248 */
249 if (!(ret = ip_vs_app_pkt_in(cp, skb)))
250 return 0;
251 /* ret=2: csum update is needed after payload mangling */
252 if (ret == 1)
253 oldlen = skb->len - udphoff;
254 else
255 payload_csum = 1;
256 }
257
258 udph = (void *)skb_network_header(skb) + udphoff;
259 udph->dest = cp->dport;
260
261 /*
262 * Adjust UDP checksums
263 */
264 if (skb->ip_summed == CHECKSUM_PARTIAL) {
265 udp_partial_csum_update(cp->af, udph, &cp->vaddr, &cp->daddr,
266 htons(oldlen),
267 htons(skb->len - udphoff));
268 } else if (!payload_csum && (udph->check != 0)) {
269 /* Only port and addr are changed, do fast csum update */
270 udp_fast_csum_update(cp->af, udph, &cp->vaddr, &cp->daddr,
271 cp->vport, cp->dport);
272 if (skb->ip_summed == CHECKSUM_COMPLETE)
273 skb->ip_summed = (cp->app && pp->csum_check) ?
274 CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
275 } else {
276 /* full checksum calculation */
277 udph->check = 0;
278 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
279 #ifdef CONFIG_IP_VS_IPV6
280 if (cp->af == AF_INET6)
281 udph->check = csum_ipv6_magic(&cp->caddr.in6,
282 &cp->daddr.in6,
283 skb->len - udphoff,
284 cp->protocol, skb->csum);
285 else
286 #endif
287 udph->check = csum_tcpudp_magic(cp->caddr.ip,
288 cp->daddr.ip,
289 skb->len - udphoff,
290 cp->protocol,
291 skb->csum);
292 if (udph->check == 0)
293 udph->check = CSUM_MANGLED_0;
294 skb->ip_summed = CHECKSUM_UNNECESSARY;
295 }
296 return 1;
297 }
298
299
300 static int
301 udp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp)
302 {
303 struct udphdr _udph, *uh;
304 unsigned int udphoff;
305
306 #ifdef CONFIG_IP_VS_IPV6
307 if (af == AF_INET6)
308 udphoff = sizeof(struct ipv6hdr);
309 else
310 #endif
311 udphoff = ip_hdrlen(skb);
312
313 uh = skb_header_pointer(skb, udphoff, sizeof(_udph), &_udph);
314 if (uh == NULL)
315 return 0;
316
317 if (uh->check != 0) {
318 switch (skb->ip_summed) {
319 case CHECKSUM_NONE:
320 skb->csum = skb_checksum(skb, udphoff,
321 skb->len - udphoff, 0);
322 /* fall through */
323 case CHECKSUM_COMPLETE:
324 #ifdef CONFIG_IP_VS_IPV6
325 if (af == AF_INET6) {
326 if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
327 &ipv6_hdr(skb)->daddr,
328 skb->len - udphoff,
329 ipv6_hdr(skb)->nexthdr,
330 skb->csum)) {
331 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
332 "Failed checksum for");
333 return 0;
334 }
335 } else
336 #endif
337 if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
338 ip_hdr(skb)->daddr,
339 skb->len - udphoff,
340 ip_hdr(skb)->protocol,
341 skb->csum)) {
342 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
343 "Failed checksum for");
344 return 0;
345 }
346 break;
347 default:
348 /* No need to checksum. */
349 break;
350 }
351 }
352 return 1;
353 }
354
355 static inline __u16 udp_app_hashkey(__be16 port)
356 {
357 return (((__force u16)port >> UDP_APP_TAB_BITS) ^ (__force u16)port)
358 & UDP_APP_TAB_MASK;
359 }
360
361
362 static int udp_register_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
363 {
364 struct ip_vs_app *i;
365 __u16 hash;
366 __be16 port = inc->port;
367 int ret = 0;
368 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_UDP);
369
370 hash = udp_app_hashkey(port);
371
372 list_for_each_entry(i, &ipvs->udp_apps[hash], p_list) {
373 if (i->port == port) {
374 ret = -EEXIST;
375 goto out;
376 }
377 }
378 list_add_rcu(&inc->p_list, &ipvs->udp_apps[hash]);
379 atomic_inc(&pd->appcnt);
380
381 out:
382 return ret;
383 }
384
385
386 static void
387 udp_unregister_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
388 {
389 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_UDP);
390
391 atomic_dec(&pd->appcnt);
392 list_del_rcu(&inc->p_list);
393 }
394
395
396 static int udp_app_conn_bind(struct ip_vs_conn *cp)
397 {
398 struct netns_ipvs *ipvs = cp->ipvs;
399 int hash;
400 struct ip_vs_app *inc;
401 int result = 0;
402
403 /* Default binding: bind app only for NAT */
404 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
405 return 0;
406
407 /* Lookup application incarnations and bind the right one */
408 hash = udp_app_hashkey(cp->vport);
409
410 list_for_each_entry_rcu(inc, &ipvs->udp_apps[hash], p_list) {
411 if (inc->port == cp->vport) {
412 if (unlikely(!ip_vs_app_inc_get(inc)))
413 break;
414
415 IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->"
416 "%s:%u to app %s on port %u\n",
417 __func__,
418 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
419 ntohs(cp->cport),
420 IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
421 ntohs(cp->vport),
422 inc->name, ntohs(inc->port));
423
424 cp->app = inc;
425 if (inc->init_conn)
426 result = inc->init_conn(inc, cp);
427 break;
428 }
429 }
430
431 return result;
432 }
433
434
435 static const int udp_timeouts[IP_VS_UDP_S_LAST+1] = {
436 [IP_VS_UDP_S_NORMAL] = 5*60*HZ,
437 [IP_VS_UDP_S_LAST] = 2*HZ,
438 };
439
440 static const char *const udp_state_name_table[IP_VS_UDP_S_LAST+1] = {
441 [IP_VS_UDP_S_NORMAL] = "UDP",
442 [IP_VS_UDP_S_LAST] = "BUG!",
443 };
444
445 static const char * udp_state_name(int state)
446 {
447 if (state >= IP_VS_UDP_S_LAST)
448 return "ERR!";
449 return udp_state_name_table[state] ? udp_state_name_table[state] : "?";
450 }
451
452 static void
453 udp_state_transition(struct ip_vs_conn *cp, int direction,
454 const struct sk_buff *skb,
455 struct ip_vs_proto_data *pd)
456 {
457 if (unlikely(!pd)) {
458 pr_err("UDP no ns data\n");
459 return;
460 }
461
462 cp->timeout = pd->timeout_table[IP_VS_UDP_S_NORMAL];
463 }
464
465 static int __udp_init(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd)
466 {
467 ip_vs_init_hash_table(ipvs->udp_apps, UDP_APP_TAB_SIZE);
468 pd->timeout_table = ip_vs_create_timeout_table((int *)udp_timeouts,
469 sizeof(udp_timeouts));
470 if (!pd->timeout_table)
471 return -ENOMEM;
472 return 0;
473 }
474
475 static void __udp_exit(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd)
476 {
477 kfree(pd->timeout_table);
478 }
479
480
481 struct ip_vs_protocol ip_vs_protocol_udp = {
482 .name = "UDP",
483 .protocol = IPPROTO_UDP,
484 .num_states = IP_VS_UDP_S_LAST,
485 .dont_defrag = 0,
486 .init = NULL,
487 .exit = NULL,
488 .init_netns = __udp_init,
489 .exit_netns = __udp_exit,
490 .conn_schedule = udp_conn_schedule,
491 .conn_in_get = ip_vs_conn_in_get_proto,
492 .conn_out_get = ip_vs_conn_out_get_proto,
493 .snat_handler = udp_snat_handler,
494 .dnat_handler = udp_dnat_handler,
495 .csum_check = udp_csum_check,
496 .state_transition = udp_state_transition,
497 .state_name = udp_state_name,
498 .register_app = udp_register_app,
499 .unregister_app = udp_unregister_app,
500 .app_conn_bind = udp_app_conn_bind,
501 .debug_packet = ip_vs_tcpudp_debug_packet,
502 .timeout_change = NULL,
503 };
504
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.