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