xref: /linux/net/netfilter/ipvs/ip_vs_proto_udp.c (revision 8bf22c33e7a172fbc72464f4cc484d23a6b412ba)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * ip_vs_proto_udp.c:	UDP load balancing support for IPVS
4  *
5  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
6  *              Julian Anastasov <ja@ssi.bg>
7  *
8  * Changes:     Hans Schillstrom <hans.schillstrom@ericsson.com>
9  *              Network name space (netns) aware.
10  */
11 
12 #define pr_fmt(fmt) "IPVS: " fmt
13 
14 #include <linux/in.h>
15 #include <linux/ip.h>
16 #include <linux/kernel.h>
17 #include <linux/netfilter.h>
18 #include <linux/netfilter_ipv4.h>
19 #include <linux/udp.h>
20 #include <linux/indirect_call_wrapper.h>
21 
22 #include <net/ip_vs.h>
23 #include <net/ip.h>
24 #include <net/ip6_checksum.h>
25 
26 static int
27 udp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp,
28 	       unsigned int udphoff);
29 
30 static int
udp_conn_schedule(struct netns_ipvs * ipvs,int af,struct sk_buff * skb,struct ip_vs_proto_data * pd,int * verdict,struct ip_vs_conn ** cpp,struct ip_vs_iphdr * iph)31 udp_conn_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
32 		  struct ip_vs_proto_data *pd,
33 		  int *verdict, struct ip_vs_conn **cpp,
34 		  struct ip_vs_iphdr *iph)
35 {
36 	struct ip_vs_service *svc;
37 	struct udphdr _udph, *uh;
38 	__be16 _ports[2], *ports = NULL;
39 
40 	if (likely(!ip_vs_iph_icmp(iph))) {
41 		/* IPv6 fragments, only first fragment will hit this */
42 		uh = skb_header_pointer(skb, iph->len, sizeof(_udph), &_udph);
43 		if (uh)
44 			ports = &uh->source;
45 	} else {
46 		ports = skb_header_pointer(
47 			skb, iph->len, sizeof(_ports), &_ports);
48 	}
49 
50 	if (!ports) {
51 		*verdict = NF_DROP;
52 		return 0;
53 	}
54 
55 	if (likely(!ip_vs_iph_inverse(iph)))
56 		svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol,
57 					 &iph->daddr, ports[1]);
58 	else
59 		svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol,
60 					 &iph->saddr, ports[0]);
61 
62 	if (svc) {
63 		int ignored;
64 
65 		if (ip_vs_todrop(ipvs)) {
66 			/*
67 			 * It seems that we are very loaded.
68 			 * We have to drop this packet :(
69 			 */
70 			*verdict = NF_DROP;
71 			return 0;
72 		}
73 
74 		/*
75 		 * Let the virtual server select a real server for the
76 		 * incoming connection, and create a connection entry.
77 		 */
78 		*cpp = ip_vs_schedule(svc, skb, pd, &ignored, iph);
79 		if (!*cpp && ignored <= 0) {
80 			if (!ignored)
81 				*verdict = ip_vs_leave(svc, skb, pd, iph);
82 			else
83 				*verdict = NF_DROP;
84 			return 0;
85 		}
86 	}
87 	/* NF_ACCEPT */
88 	return 1;
89 }
90 
91 
92 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)93 udp_fast_csum_update(int af, struct udphdr *uhdr,
94 		     const union nf_inet_addr *oldip,
95 		     const union nf_inet_addr *newip,
96 		     __be16 oldport, __be16 newport)
97 {
98 #ifdef CONFIG_IP_VS_IPV6
99 	if (af == AF_INET6)
100 		uhdr->check =
101 			csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
102 					 ip_vs_check_diff2(oldport, newport,
103 						~csum_unfold(uhdr->check))));
104 	else
105 #endif
106 		uhdr->check =
107 			csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
108 					 ip_vs_check_diff2(oldport, newport,
109 						~csum_unfold(uhdr->check))));
110 	if (!uhdr->check)
111 		uhdr->check = CSUM_MANGLED_0;
112 }
113 
114 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)115 udp_partial_csum_update(int af, struct udphdr *uhdr,
116 		     const union nf_inet_addr *oldip,
117 		     const union nf_inet_addr *newip,
118 		     __be16 oldlen, __be16 newlen)
119 {
120 #ifdef CONFIG_IP_VS_IPV6
121 	if (af == AF_INET6)
122 		uhdr->check =
123 			~csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
124 					 ip_vs_check_diff2(oldlen, newlen,
125 						csum_unfold(uhdr->check))));
126 	else
127 #endif
128 	uhdr->check =
129 		~csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
130 				ip_vs_check_diff2(oldlen, newlen,
131 						csum_unfold(uhdr->check))));
132 }
133 
134 
135 INDIRECT_CALLABLE_SCOPE int
udp_snat_handler(struct sk_buff * skb,struct ip_vs_protocol * pp,struct ip_vs_conn * cp,struct ip_vs_iphdr * iph)136 udp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
137 		 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
138 {
139 	struct udphdr *udph;
140 	unsigned int udphoff = iph->len;
141 	bool payload_csum = false;
142 	int oldlen;
143 
144 #ifdef CONFIG_IP_VS_IPV6
145 	if (cp->af == AF_INET6 && iph->fragoffs)
146 		return 1;
147 #endif
148 	oldlen = skb->len - udphoff;
149 
150 	/* csum_check requires unshared skb */
151 	if (skb_ensure_writable(skb, udphoff + sizeof(*udph)))
152 		return 0;
153 
154 	if (unlikely(cp->app != NULL)) {
155 		int ret;
156 
157 		/* Some checks before mangling */
158 		if (!udp_csum_check(cp->af, skb, pp, udphoff))
159 			return 0;
160 
161 		/*
162 		 *	Call application helper if needed
163 		 */
164 		if (!(ret = ip_vs_app_pkt_out(cp, skb, iph)))
165 			return 0;
166 		/* ret=2: csum update is needed after payload mangling */
167 		if (ret == 1)
168 			oldlen = skb->len - udphoff;
169 		else
170 			payload_csum = true;
171 	}
172 
173 	udph = (void *)skb_network_header(skb) + udphoff;
174 	udph->source = cp->vport;
175 
176 	/*
177 	 *	Adjust UDP checksums
178 	 */
179 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
180 		udp_partial_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr,
181 					htons(oldlen),
182 					htons(skb->len - udphoff));
183 	} else if (!payload_csum && (udph->check != 0)) {
184 		/* Only port and addr are changed, do fast csum update */
185 		udp_fast_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr,
186 				     cp->dport, cp->vport);
187 		if (skb->ip_summed == CHECKSUM_COMPLETE)
188 			skb->ip_summed = cp->app ?
189 					 CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
190 	} else {
191 		/* full checksum calculation */
192 		udph->check = 0;
193 		skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
194 #ifdef CONFIG_IP_VS_IPV6
195 		if (cp->af == AF_INET6)
196 			udph->check = csum_ipv6_magic(&cp->vaddr.in6,
197 						      &cp->caddr.in6,
198 						      skb->len - udphoff,
199 						      cp->protocol, skb->csum);
200 		else
201 #endif
202 			udph->check = csum_tcpudp_magic(cp->vaddr.ip,
203 							cp->caddr.ip,
204 							skb->len - udphoff,
205 							cp->protocol,
206 							skb->csum);
207 		if (udph->check == 0)
208 			udph->check = CSUM_MANGLED_0;
209 		skb->ip_summed = CHECKSUM_UNNECESSARY;
210 		IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
211 			  pp->name, udph->check,
212 			  (char*)&(udph->check) - (char*)udph);
213 	}
214 	return 1;
215 }
216 
217 
218 static int
udp_dnat_handler(struct sk_buff * skb,struct ip_vs_protocol * pp,struct ip_vs_conn * cp,struct ip_vs_iphdr * iph)219 udp_dnat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
220 		 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
221 {
222 	struct udphdr *udph;
223 	unsigned int udphoff = iph->len;
224 	bool payload_csum = false;
225 	int oldlen;
226 
227 #ifdef CONFIG_IP_VS_IPV6
228 	if (cp->af == AF_INET6 && iph->fragoffs)
229 		return 1;
230 #endif
231 	oldlen = skb->len - udphoff;
232 
233 	/* csum_check requires unshared skb */
234 	if (skb_ensure_writable(skb, udphoff + sizeof(*udph)))
235 		return 0;
236 
237 	if (unlikely(cp->app != NULL)) {
238 		int ret;
239 
240 		/* Some checks before mangling */
241 		if (!udp_csum_check(cp->af, skb, pp, udphoff))
242 			return 0;
243 
244 		/*
245 		 *	Attempt ip_vs_app call.
246 		 *	It will fix ip_vs_conn
247 		 */
248 		if (!(ret = ip_vs_app_pkt_in(cp, skb, iph)))
249 			return 0;
250 		/* ret=2: csum update is needed after payload mangling */
251 		if (ret == 1)
252 			oldlen = skb->len - udphoff;
253 		else
254 			payload_csum = true;
255 	}
256 
257 	udph = (void *)skb_network_header(skb) + udphoff;
258 	udph->dest = cp->dport;
259 
260 	/*
261 	 *	Adjust UDP checksums
262 	 */
263 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
264 		udp_partial_csum_update(cp->af, udph, &cp->vaddr, &cp->daddr,
265 					htons(oldlen),
266 					htons(skb->len - udphoff));
267 	} else if (!payload_csum && (udph->check != 0)) {
268 		/* Only port and addr are changed, do fast csum update */
269 		udp_fast_csum_update(cp->af, udph, &cp->vaddr, &cp->daddr,
270 				     cp->vport, cp->dport);
271 		if (skb->ip_summed == CHECKSUM_COMPLETE)
272 			skb->ip_summed = cp->app ?
273 					 CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
274 	} else {
275 		/* full checksum calculation */
276 		udph->check = 0;
277 		skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0);
278 #ifdef CONFIG_IP_VS_IPV6
279 		if (cp->af == AF_INET6)
280 			udph->check = csum_ipv6_magic(&cp->caddr.in6,
281 						      &cp->daddr.in6,
282 						      skb->len - udphoff,
283 						      cp->protocol, skb->csum);
284 		else
285 #endif
286 			udph->check = csum_tcpudp_magic(cp->caddr.ip,
287 							cp->daddr.ip,
288 							skb->len - udphoff,
289 							cp->protocol,
290 							skb->csum);
291 		if (udph->check == 0)
292 			udph->check = CSUM_MANGLED_0;
293 		skb->ip_summed = CHECKSUM_UNNECESSARY;
294 	}
295 	return 1;
296 }
297 
298 
299 static int
udp_csum_check(int af,struct sk_buff * skb,struct ip_vs_protocol * pp,unsigned int udphoff)300 udp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp,
301 	       unsigned int udphoff)
302 {
303 	struct udphdr _udph, *uh;
304 
305 	uh = skb_header_pointer(skb, udphoff, sizeof(_udph), &_udph);
306 	if (uh == NULL)
307 		return 0;
308 
309 	if (uh->check != 0) {
310 		switch (skb->ip_summed) {
311 		case CHECKSUM_NONE:
312 			skb->csum = skb_checksum(skb, udphoff,
313 						 skb->len - udphoff, 0);
314 			fallthrough;
315 		case CHECKSUM_COMPLETE:
316 #ifdef CONFIG_IP_VS_IPV6
317 			if (af == AF_INET6) {
318 				if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
319 						    &ipv6_hdr(skb)->daddr,
320 						    skb->len - udphoff,
321 						    IPPROTO_UDP,
322 						    skb->csum)) {
323 					IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
324 							 "Failed checksum for");
325 					return 0;
326 				}
327 			} else
328 #endif
329 				if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
330 						      ip_hdr(skb)->daddr,
331 						      skb->len - udphoff,
332 						      ip_hdr(skb)->protocol,
333 						      skb->csum)) {
334 					IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
335 							 "Failed checksum for");
336 					return 0;
337 				}
338 			break;
339 		default:
340 			/* No need to checksum. */
341 			break;
342 		}
343 	}
344 	return 1;
345 }
346 
udp_app_hashkey(__be16 port)347 static inline __u16 udp_app_hashkey(__be16 port)
348 {
349 	return (((__force u16)port >> UDP_APP_TAB_BITS) ^ (__force u16)port)
350 		& UDP_APP_TAB_MASK;
351 }
352 
353 
udp_register_app(struct netns_ipvs * ipvs,struct ip_vs_app * inc)354 static int udp_register_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
355 {
356 	struct ip_vs_app *i;
357 	__u16 hash;
358 	__be16 port = inc->port;
359 	int ret = 0;
360 	struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_UDP);
361 
362 	hash = udp_app_hashkey(port);
363 
364 	list_for_each_entry(i, &ipvs->udp_apps[hash], p_list) {
365 		if (i->port == port) {
366 			ret = -EEXIST;
367 			goto out;
368 		}
369 	}
370 	list_add_rcu(&inc->p_list, &ipvs->udp_apps[hash]);
371 	atomic_inc(&pd->appcnt);
372 
373   out:
374 	return ret;
375 }
376 
377 
378 static void
udp_unregister_app(struct netns_ipvs * ipvs,struct ip_vs_app * inc)379 udp_unregister_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
380 {
381 	struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_UDP);
382 
383 	atomic_dec(&pd->appcnt);
384 	list_del_rcu(&inc->p_list);
385 }
386 
387 
udp_app_conn_bind(struct ip_vs_conn * cp)388 static int udp_app_conn_bind(struct ip_vs_conn *cp)
389 {
390 	struct netns_ipvs *ipvs = cp->ipvs;
391 	int hash;
392 	struct ip_vs_app *inc;
393 	int result = 0;
394 
395 	/* Default binding: bind app only for NAT */
396 	if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
397 		return 0;
398 
399 	/* Lookup application incarnations and bind the right one */
400 	hash = udp_app_hashkey(cp->vport);
401 
402 	list_for_each_entry_rcu(inc, &ipvs->udp_apps[hash], p_list) {
403 		if (inc->port == cp->vport) {
404 			if (unlikely(!ip_vs_app_inc_get(inc)))
405 				break;
406 
407 			IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->"
408 				      "%s:%u to app %s on port %u\n",
409 				      __func__,
410 				      IP_VS_DBG_ADDR(cp->af, &cp->caddr),
411 				      ntohs(cp->cport),
412 				      IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
413 				      ntohs(cp->vport),
414 				      inc->name, ntohs(inc->port));
415 
416 			cp->app = inc;
417 			if (inc->init_conn)
418 				result = inc->init_conn(inc, cp);
419 			break;
420 		}
421 	}
422 
423 	return result;
424 }
425 
426 
427 static const int udp_timeouts[IP_VS_UDP_S_LAST+1] = {
428 	[IP_VS_UDP_S_NORMAL]		=	5*60*HZ,
429 	[IP_VS_UDP_S_LAST]		=	2*HZ,
430 };
431 
432 static const char *const udp_state_name_table[IP_VS_UDP_S_LAST+1] = {
433 	[IP_VS_UDP_S_NORMAL]		=	"UDP",
434 	[IP_VS_UDP_S_LAST]		=	"BUG!",
435 };
436 
udp_state_name(int state)437 static const char * udp_state_name(int state)
438 {
439 	if (state >= IP_VS_UDP_S_LAST)
440 		return "ERR!";
441 	return udp_state_name_table[state] ? udp_state_name_table[state] : "?";
442 }
443 
444 static void
udp_state_transition(struct ip_vs_conn * cp,int direction,const struct sk_buff * skb,struct ip_vs_proto_data * pd)445 udp_state_transition(struct ip_vs_conn *cp, int direction,
446 		     const struct sk_buff *skb,
447 		     struct ip_vs_proto_data *pd)
448 {
449 	if (unlikely(!pd)) {
450 		pr_err("UDP no ns data\n");
451 		return;
452 	}
453 
454 	cp->timeout = pd->timeout_table[IP_VS_UDP_S_NORMAL];
455 	if (direction == IP_VS_DIR_OUTPUT)
456 		ip_vs_control_assure_ct(cp);
457 }
458 
__udp_init(struct netns_ipvs * ipvs,struct ip_vs_proto_data * pd)459 static int __udp_init(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd)
460 {
461 	ip_vs_init_hash_table(ipvs->udp_apps, UDP_APP_TAB_SIZE);
462 	pd->timeout_table = ip_vs_create_timeout_table((int *)udp_timeouts,
463 							sizeof(udp_timeouts));
464 	if (!pd->timeout_table)
465 		return -ENOMEM;
466 	return 0;
467 }
468 
__udp_exit(struct netns_ipvs * ipvs,struct ip_vs_proto_data * pd)469 static void __udp_exit(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd)
470 {
471 	kfree(pd->timeout_table);
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 =			NULL,
481 	.exit =			NULL,
482 	.init_netns =		__udp_init,
483 	.exit_netns =		__udp_exit,
484 	.conn_schedule =	udp_conn_schedule,
485 	.conn_in_get =		ip_vs_conn_in_get_proto,
486 	.conn_out_get =		ip_vs_conn_out_get_proto,
487 	.snat_handler =		udp_snat_handler,
488 	.dnat_handler =		udp_dnat_handler,
489 	.state_transition =	udp_state_transition,
490 	.state_name =		udp_state_name,
491 	.register_app =		udp_register_app,
492 	.unregister_app =	udp_unregister_app,
493 	.app_conn_bind =	udp_app_conn_bind,
494 	.debug_packet =		ip_vs_tcpudp_debug_packet,
495 	.timeout_change =	NULL,
496 };
497