1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * ip_vs_xmit.c: various packet transmitters for IPVS
4 *
5 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
6 * Julian Anastasov <ja@ssi.bg>
7 *
8 * Changes:
9 *
10 * Description of forwarding methods:
11 * - all transmitters are called from LOCAL_IN (remote clients) and
12 * LOCAL_OUT (local clients) but for ICMP can be called from FORWARD
13 * - not all connections have destination server, for example,
14 * connections in backup server when fwmark is used
15 * - bypass connections use daddr from packet
16 * - we can use dst without ref while sending in RCU section, we use
17 * ref when returning NF_ACCEPT for NAT-ed packet via loopback
18 * LOCAL_OUT rules:
19 * - skb->dev is NULL, skb->protocol is not set (both are set in POST_ROUTING)
20 * - skb->pkt_type is not set yet
21 * - the only place where we can see skb->sk != NULL
22 */
23
24 #define KMSG_COMPONENT "IPVS"
25 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
26
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
29 #include <linux/tcp.h> /* for tcphdr */
30 #include <net/ip.h>
31 #include <net/gue.h>
32 #include <net/gre.h>
33 #include <net/tcp.h> /* for csum_tcpudp_magic */
34 #include <net/udp.h>
35 #include <net/icmp.h> /* for icmp_send */
36 #include <net/route.h> /* for ip_route_output */
37 #include <net/ipv6.h>
38 #include <net/ip6_route.h>
39 #include <net/ip_tunnels.h>
40 #include <net/ip6_checksum.h>
41 #include <net/addrconf.h>
42 #include <linux/icmpv6.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv4.h>
45
46 #include <net/ip_vs.h>
47
48 enum {
49 IP_VS_RT_MODE_LOCAL = 1, /* Allow local dest */
50 IP_VS_RT_MODE_NON_LOCAL = 2, /* Allow non-local dest */
51 IP_VS_RT_MODE_RDR = 4, /* Allow redirect from remote daddr to
52 * local
53 */
54 IP_VS_RT_MODE_CONNECT = 8, /* Always bind route to saddr */
55 IP_VS_RT_MODE_KNOWN_NH = 16,/* Route via remote addr */
56 IP_VS_RT_MODE_TUNNEL = 32,/* Tunnel mode */
57 };
58
ip_vs_dest_dst_alloc(void)59 static inline struct ip_vs_dest_dst *ip_vs_dest_dst_alloc(void)
60 {
61 return kmalloc(sizeof(struct ip_vs_dest_dst), GFP_ATOMIC);
62 }
63
ip_vs_dest_dst_free(struct ip_vs_dest_dst * dest_dst)64 static inline void ip_vs_dest_dst_free(struct ip_vs_dest_dst *dest_dst)
65 {
66 kfree(dest_dst);
67 }
68
69 /*
70 * Destination cache to speed up outgoing route lookup
71 */
72 static inline void
__ip_vs_dst_set(struct ip_vs_dest * dest,struct ip_vs_dest_dst * dest_dst,struct dst_entry * dst,u32 dst_cookie)73 __ip_vs_dst_set(struct ip_vs_dest *dest, struct ip_vs_dest_dst *dest_dst,
74 struct dst_entry *dst, u32 dst_cookie)
75 {
76 struct ip_vs_dest_dst *old;
77
78 old = rcu_dereference_protected(dest->dest_dst,
79 lockdep_is_held(&dest->dst_lock));
80
81 if (dest_dst) {
82 dest_dst->dst_cache = dst;
83 dest_dst->dst_cookie = dst_cookie;
84 }
85 rcu_assign_pointer(dest->dest_dst, dest_dst);
86
87 if (old)
88 call_rcu(&old->rcu_head, ip_vs_dest_dst_rcu_free);
89 }
90
91 static inline struct ip_vs_dest_dst *
__ip_vs_dst_check(struct ip_vs_dest * dest)92 __ip_vs_dst_check(struct ip_vs_dest *dest)
93 {
94 struct ip_vs_dest_dst *dest_dst = rcu_dereference(dest->dest_dst);
95 struct dst_entry *dst;
96
97 if (!dest_dst)
98 return NULL;
99 dst = dest_dst->dst_cache;
100 if (dst->obsolete &&
101 dst->ops->check(dst, dest_dst->dst_cookie) == NULL)
102 return NULL;
103 return dest_dst;
104 }
105
106 static inline bool
__mtu_check_toobig_v6(const struct sk_buff * skb,u32 mtu)107 __mtu_check_toobig_v6(const struct sk_buff *skb, u32 mtu)
108 {
109 if (IP6CB(skb)->frag_max_size) {
110 /* frag_max_size tell us that, this packet have been
111 * defragmented by netfilter IPv6 conntrack module.
112 */
113 if (IP6CB(skb)->frag_max_size > mtu)
114 return true; /* largest fragment violate MTU */
115 }
116 else if (skb->len > mtu && !skb_is_gso(skb)) {
117 return true; /* Packet size violate MTU size */
118 }
119 return false;
120 }
121
122 /* Get route to daddr, update *saddr, optionally bind route to saddr */
do_output_route4(struct net * net,__be32 daddr,int rt_mode,__be32 * saddr)123 static struct rtable *do_output_route4(struct net *net, __be32 daddr,
124 int rt_mode, __be32 *saddr)
125 {
126 struct flowi4 fl4;
127 struct rtable *rt;
128 bool loop = false;
129
130 memset(&fl4, 0, sizeof(fl4));
131 fl4.daddr = daddr;
132 fl4.flowi4_flags = (rt_mode & IP_VS_RT_MODE_KNOWN_NH) ?
133 FLOWI_FLAG_KNOWN_NH : 0;
134
135 retry:
136 rt = ip_route_output_key(net, &fl4);
137 if (IS_ERR(rt)) {
138 /* Invalid saddr ? */
139 if (PTR_ERR(rt) == -EINVAL && *saddr &&
140 rt_mode & IP_VS_RT_MODE_CONNECT && !loop) {
141 *saddr = 0;
142 flowi4_update_output(&fl4, 0, daddr, 0);
143 goto retry;
144 }
145 IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", &daddr);
146 return NULL;
147 } else if (!*saddr && rt_mode & IP_VS_RT_MODE_CONNECT && fl4.saddr) {
148 ip_rt_put(rt);
149 *saddr = fl4.saddr;
150 flowi4_update_output(&fl4, 0, daddr, fl4.saddr);
151 loop = true;
152 goto retry;
153 }
154 *saddr = fl4.saddr;
155 return rt;
156 }
157
158 #ifdef CONFIG_IP_VS_IPV6
__ip_vs_is_local_route6(struct rt6_info * rt)159 static inline int __ip_vs_is_local_route6(struct rt6_info *rt)
160 {
161 return rt->dst.dev && rt->dst.dev->flags & IFF_LOOPBACK;
162 }
163 #endif
164
crosses_local_route_boundary(int skb_af,struct sk_buff * skb,int rt_mode,bool new_rt_is_local)165 static inline bool crosses_local_route_boundary(int skb_af, struct sk_buff *skb,
166 int rt_mode,
167 bool new_rt_is_local)
168 {
169 bool rt_mode_allow_local = !!(rt_mode & IP_VS_RT_MODE_LOCAL);
170 bool rt_mode_allow_non_local = !!(rt_mode & IP_VS_RT_MODE_NON_LOCAL);
171 bool rt_mode_allow_redirect = !!(rt_mode & IP_VS_RT_MODE_RDR);
172 bool source_is_loopback;
173 bool old_rt_is_local;
174
175 #ifdef CONFIG_IP_VS_IPV6
176 if (skb_af == AF_INET6) {
177 int addr_type = ipv6_addr_type(&ipv6_hdr(skb)->saddr);
178
179 source_is_loopback =
180 (!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
181 (addr_type & IPV6_ADDR_LOOPBACK);
182 old_rt_is_local = __ip_vs_is_local_route6(
183 dst_rt6_info(skb_dst(skb)));
184 } else
185 #endif
186 {
187 source_is_loopback = ipv4_is_loopback(ip_hdr(skb)->saddr);
188 old_rt_is_local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
189 }
190
191 if (unlikely(new_rt_is_local)) {
192 if (!rt_mode_allow_local)
193 return true;
194 if (!rt_mode_allow_redirect && !old_rt_is_local)
195 return true;
196 } else {
197 if (!rt_mode_allow_non_local)
198 return true;
199 if (source_is_loopback)
200 return true;
201 }
202 return false;
203 }
204
maybe_update_pmtu(int skb_af,struct sk_buff * skb,int mtu)205 static inline void maybe_update_pmtu(int skb_af, struct sk_buff *skb, int mtu)
206 {
207 struct sock *sk = skb->sk;
208 struct rtable *ort = skb_rtable(skb);
209
210 if (!skb->dev && sk && sk_fullsock(sk))
211 ort->dst.ops->update_pmtu(&ort->dst, sk, NULL, mtu, true);
212 }
213
ensure_mtu_is_adequate(struct netns_ipvs * ipvs,int skb_af,int rt_mode,struct ip_vs_iphdr * ipvsh,struct sk_buff * skb,int mtu)214 static inline bool ensure_mtu_is_adequate(struct netns_ipvs *ipvs, int skb_af,
215 int rt_mode,
216 struct ip_vs_iphdr *ipvsh,
217 struct sk_buff *skb, int mtu)
218 {
219 #ifdef CONFIG_IP_VS_IPV6
220 if (skb_af == AF_INET6) {
221 struct net *net = ipvs->net;
222
223 if (unlikely(__mtu_check_toobig_v6(skb, mtu))) {
224 if (!skb->dev)
225 skb->dev = net->loopback_dev;
226 /* only send ICMP too big on first fragment */
227 if (!ipvsh->fragoffs && !ip_vs_iph_icmp(ipvsh))
228 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
229 IP_VS_DBG(1, "frag needed for %pI6c\n",
230 &ipv6_hdr(skb)->saddr);
231 return false;
232 }
233 } else
234 #endif
235 {
236 /* If we're going to tunnel the packet and pmtu discovery
237 * is disabled, we'll just fragment it anyway
238 */
239 if ((rt_mode & IP_VS_RT_MODE_TUNNEL) && !sysctl_pmtu_disc(ipvs))
240 return true;
241
242 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_DF) &&
243 skb->len > mtu && !skb_is_gso(skb) &&
244 !ip_vs_iph_icmp(ipvsh))) {
245 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
246 htonl(mtu));
247 IP_VS_DBG(1, "frag needed for %pI4\n",
248 &ip_hdr(skb)->saddr);
249 return false;
250 }
251 }
252
253 return true;
254 }
255
decrement_ttl(struct netns_ipvs * ipvs,int skb_af,struct sk_buff * skb)256 static inline bool decrement_ttl(struct netns_ipvs *ipvs,
257 int skb_af,
258 struct sk_buff *skb)
259 {
260 struct net *net = ipvs->net;
261
262 #ifdef CONFIG_IP_VS_IPV6
263 if (skb_af == AF_INET6) {
264 struct dst_entry *dst = skb_dst(skb);
265
266 /* check and decrement ttl */
267 if (ipv6_hdr(skb)->hop_limit <= 1) {
268 struct inet6_dev *idev = __in6_dev_get_safely(skb->dev);
269
270 /* Force OUTPUT device used as source address */
271 skb->dev = dst->dev;
272 icmpv6_send(skb, ICMPV6_TIME_EXCEED,
273 ICMPV6_EXC_HOPLIMIT, 0);
274 IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
275
276 return false;
277 }
278
279 /* don't propagate ttl change to cloned packets */
280 if (skb_ensure_writable(skb, sizeof(struct ipv6hdr)))
281 return false;
282
283 ipv6_hdr(skb)->hop_limit--;
284 } else
285 #endif
286 {
287 if (ip_hdr(skb)->ttl <= 1) {
288 /* Tell the sender its packet died... */
289 IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
290 icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0);
291 return false;
292 }
293
294 /* don't propagate ttl change to cloned packets */
295 if (skb_ensure_writable(skb, sizeof(struct iphdr)))
296 return false;
297
298 /* Decrease ttl */
299 ip_decrease_ttl(ip_hdr(skb));
300 }
301
302 return true;
303 }
304
305 /* Get route to destination or remote server */
306 static int
__ip_vs_get_out_rt(struct netns_ipvs * ipvs,int skb_af,struct sk_buff * skb,struct ip_vs_dest * dest,__be32 daddr,int rt_mode,__be32 * ret_saddr,struct ip_vs_iphdr * ipvsh)307 __ip_vs_get_out_rt(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
308 struct ip_vs_dest *dest,
309 __be32 daddr, int rt_mode, __be32 *ret_saddr,
310 struct ip_vs_iphdr *ipvsh)
311 {
312 struct net *net = ipvs->net;
313 struct ip_vs_dest_dst *dest_dst;
314 struct rtable *rt; /* Route to the other host */
315 int mtu;
316 int local, noref = 1;
317
318 if (dest) {
319 dest_dst = __ip_vs_dst_check(dest);
320 if (likely(dest_dst))
321 rt = dst_rtable(dest_dst->dst_cache);
322 else {
323 dest_dst = ip_vs_dest_dst_alloc();
324 spin_lock_bh(&dest->dst_lock);
325 if (!dest_dst) {
326 __ip_vs_dst_set(dest, NULL, NULL, 0);
327 spin_unlock_bh(&dest->dst_lock);
328 goto err_unreach;
329 }
330 rt = do_output_route4(net, dest->addr.ip, rt_mode,
331 &dest_dst->dst_saddr.ip);
332 if (!rt) {
333 __ip_vs_dst_set(dest, NULL, NULL, 0);
334 spin_unlock_bh(&dest->dst_lock);
335 ip_vs_dest_dst_free(dest_dst);
336 goto err_unreach;
337 }
338 __ip_vs_dst_set(dest, dest_dst, &rt->dst, 0);
339 spin_unlock_bh(&dest->dst_lock);
340 IP_VS_DBG(10, "new dst %pI4, src %pI4, refcnt=%d\n",
341 &dest->addr.ip, &dest_dst->dst_saddr.ip,
342 rcuref_read(&rt->dst.__rcuref));
343 }
344 if (ret_saddr)
345 *ret_saddr = dest_dst->dst_saddr.ip;
346 } else {
347 __be32 saddr = htonl(INADDR_ANY);
348
349 noref = 0;
350
351 /* For such unconfigured boxes avoid many route lookups
352 * for performance reasons because we do not remember saddr
353 */
354 rt_mode &= ~IP_VS_RT_MODE_CONNECT;
355 rt = do_output_route4(net, daddr, rt_mode, &saddr);
356 if (!rt)
357 goto err_unreach;
358 if (ret_saddr)
359 *ret_saddr = saddr;
360 }
361
362 local = (rt->rt_flags & RTCF_LOCAL) ? 1 : 0;
363 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
364 local))) {
365 IP_VS_DBG_RL("We are crossing local and non-local addresses"
366 " daddr=%pI4\n", &daddr);
367 goto err_put;
368 }
369
370 if (unlikely(local)) {
371 /* skb to local stack, preserve old route */
372 if (!noref)
373 ip_rt_put(rt);
374 return local;
375 }
376
377 if (!decrement_ttl(ipvs, skb_af, skb))
378 goto err_put;
379
380 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL))) {
381 mtu = dst_mtu(&rt->dst);
382 } else {
383 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
384 if (!dest)
385 goto err_put;
386 if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
387 mtu -= sizeof(struct udphdr) + sizeof(struct guehdr);
388 if ((dest->tun_flags &
389 IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
390 skb->ip_summed == CHECKSUM_PARTIAL)
391 mtu -= GUE_PLEN_REMCSUM + GUE_LEN_PRIV;
392 } else if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
393 IP_TUNNEL_DECLARE_FLAGS(tflags) = { };
394
395 if (dest->tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
396 __set_bit(IP_TUNNEL_CSUM_BIT, tflags);
397 mtu -= gre_calc_hlen(tflags);
398 }
399 if (mtu < 68) {
400 IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__);
401 goto err_put;
402 }
403 maybe_update_pmtu(skb_af, skb, mtu);
404 }
405
406 if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
407 goto err_put;
408
409 skb_dst_drop(skb);
410 if (noref)
411 skb_dst_set_noref(skb, &rt->dst);
412 else
413 skb_dst_set(skb, &rt->dst);
414
415 return local;
416
417 err_put:
418 if (!noref)
419 ip_rt_put(rt);
420 return -1;
421
422 err_unreach:
423 dst_link_failure(skb);
424 return -1;
425 }
426
427 #ifdef CONFIG_IP_VS_IPV6
428 static struct dst_entry *
__ip_vs_route_output_v6(struct net * net,struct in6_addr * daddr,struct in6_addr * ret_saddr,int do_xfrm,int rt_mode)429 __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr,
430 struct in6_addr *ret_saddr, int do_xfrm, int rt_mode)
431 {
432 struct dst_entry *dst;
433 struct flowi6 fl6 = {
434 .daddr = *daddr,
435 };
436
437 if (rt_mode & IP_VS_RT_MODE_KNOWN_NH)
438 fl6.flowi6_flags = FLOWI_FLAG_KNOWN_NH;
439
440 dst = ip6_route_output(net, NULL, &fl6);
441 if (dst->error)
442 goto out_err;
443 if (!ret_saddr)
444 return dst;
445 if (ipv6_addr_any(&fl6.saddr) &&
446 ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
447 &fl6.daddr, 0, &fl6.saddr) < 0)
448 goto out_err;
449 if (do_xfrm) {
450 dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
451 if (IS_ERR(dst)) {
452 dst = NULL;
453 goto out_err;
454 }
455 }
456 *ret_saddr = fl6.saddr;
457 return dst;
458
459 out_err:
460 dst_release(dst);
461 IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
462 return NULL;
463 }
464
465 /*
466 * Get route to destination or remote server
467 */
468 static int
__ip_vs_get_out_rt_v6(struct netns_ipvs * ipvs,int skb_af,struct sk_buff * skb,struct ip_vs_dest * dest,struct in6_addr * daddr,struct in6_addr * ret_saddr,struct ip_vs_iphdr * ipvsh,int do_xfrm,int rt_mode)469 __ip_vs_get_out_rt_v6(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
470 struct ip_vs_dest *dest,
471 struct in6_addr *daddr, struct in6_addr *ret_saddr,
472 struct ip_vs_iphdr *ipvsh, int do_xfrm, int rt_mode)
473 {
474 struct net *net = ipvs->net;
475 struct ip_vs_dest_dst *dest_dst;
476 struct rt6_info *rt; /* Route to the other host */
477 struct dst_entry *dst;
478 int mtu;
479 int local, noref = 1;
480
481 if (dest) {
482 dest_dst = __ip_vs_dst_check(dest);
483 if (likely(dest_dst))
484 rt = dst_rt6_info(dest_dst->dst_cache);
485 else {
486 u32 cookie;
487
488 dest_dst = ip_vs_dest_dst_alloc();
489 spin_lock_bh(&dest->dst_lock);
490 if (!dest_dst) {
491 __ip_vs_dst_set(dest, NULL, NULL, 0);
492 spin_unlock_bh(&dest->dst_lock);
493 goto err_unreach;
494 }
495 dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
496 &dest_dst->dst_saddr.in6,
497 do_xfrm, rt_mode);
498 if (!dst) {
499 __ip_vs_dst_set(dest, NULL, NULL, 0);
500 spin_unlock_bh(&dest->dst_lock);
501 ip_vs_dest_dst_free(dest_dst);
502 goto err_unreach;
503 }
504 rt = dst_rt6_info(dst);
505 cookie = rt6_get_cookie(rt);
506 __ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
507 spin_unlock_bh(&dest->dst_lock);
508 IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
509 &dest->addr.in6, &dest_dst->dst_saddr.in6,
510 rcuref_read(&rt->dst.__rcuref));
511 }
512 if (ret_saddr)
513 *ret_saddr = dest_dst->dst_saddr.in6;
514 } else {
515 noref = 0;
516 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm,
517 rt_mode);
518 if (!dst)
519 goto err_unreach;
520 rt = dst_rt6_info(dst);
521 }
522
523 local = __ip_vs_is_local_route6(rt);
524
525 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
526 local))) {
527 IP_VS_DBG_RL("We are crossing local and non-local addresses"
528 " daddr=%pI6\n", daddr);
529 goto err_put;
530 }
531
532 if (unlikely(local)) {
533 /* skb to local stack, preserve old route */
534 if (!noref)
535 dst_release(&rt->dst);
536 return local;
537 }
538
539 if (!decrement_ttl(ipvs, skb_af, skb))
540 goto err_put;
541
542 /* MTU checking */
543 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL)))
544 mtu = dst_mtu(&rt->dst);
545 else {
546 mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
547 if (!dest)
548 goto err_put;
549 if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
550 mtu -= sizeof(struct udphdr) + sizeof(struct guehdr);
551 if ((dest->tun_flags &
552 IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
553 skb->ip_summed == CHECKSUM_PARTIAL)
554 mtu -= GUE_PLEN_REMCSUM + GUE_LEN_PRIV;
555 } else if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
556 IP_TUNNEL_DECLARE_FLAGS(tflags) = { };
557
558 if (dest->tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
559 __set_bit(IP_TUNNEL_CSUM_BIT, tflags);
560 mtu -= gre_calc_hlen(tflags);
561 }
562 if (mtu < IPV6_MIN_MTU) {
563 IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
564 IPV6_MIN_MTU);
565 goto err_put;
566 }
567 maybe_update_pmtu(skb_af, skb, mtu);
568 }
569
570 if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
571 goto err_put;
572
573 skb_dst_drop(skb);
574 if (noref)
575 skb_dst_set_noref(skb, &rt->dst);
576 else
577 skb_dst_set(skb, &rt->dst);
578
579 return local;
580
581 err_put:
582 if (!noref)
583 dst_release(&rt->dst);
584 return -1;
585
586 err_unreach:
587 /* The ip6_link_failure function requires the dev field to be set
588 * in order to get the net (further for the sake of fwmark
589 * reflection).
590 */
591 if (!skb->dev)
592 skb->dev = skb_dst(skb)->dev;
593
594 dst_link_failure(skb);
595 return -1;
596 }
597 #endif
598
599
600 /* return NF_ACCEPT to allow forwarding or other NF_xxx on error */
ip_vs_tunnel_xmit_prepare(struct sk_buff * skb,struct ip_vs_conn * cp)601 static inline int ip_vs_tunnel_xmit_prepare(struct sk_buff *skb,
602 struct ip_vs_conn *cp)
603 {
604 int ret = NF_ACCEPT;
605
606 skb->ipvs_property = 1;
607 if (unlikely(cp->flags & IP_VS_CONN_F_NFCT))
608 ret = ip_vs_confirm_conntrack(skb);
609 if (ret == NF_ACCEPT) {
610 nf_reset_ct(skb);
611 skb_forward_csum(skb);
612 if (skb->dev)
613 skb_clear_tstamp(skb);
614 }
615 return ret;
616 }
617
618 /* In the event of a remote destination, it's possible that we would have
619 * matches against an old socket (particularly a TIME-WAIT socket). This
620 * causes havoc down the line (ip_local_out et. al. expect regular sockets
621 * and invalid memory accesses will happen) so simply drop the association
622 * in this case.
623 */
ip_vs_drop_early_demux_sk(struct sk_buff * skb)624 static inline void ip_vs_drop_early_demux_sk(struct sk_buff *skb)
625 {
626 /* If dev is set, the packet came from the LOCAL_IN callback and
627 * not from a local TCP socket.
628 */
629 if (skb->dev)
630 skb_orphan(skb);
631 }
632
633 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
ip_vs_nat_send_or_cont(int pf,struct sk_buff * skb,struct ip_vs_conn * cp,int local)634 static inline int ip_vs_nat_send_or_cont(int pf, struct sk_buff *skb,
635 struct ip_vs_conn *cp, int local)
636 {
637 int ret = NF_STOLEN;
638
639 skb->ipvs_property = 1;
640 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
641 ip_vs_notrack(skb);
642 else
643 ip_vs_update_conntrack(skb, cp, 1);
644
645 /* Remove the early_demux association unless it's bound for the
646 * exact same port and address on this host after translation.
647 */
648 if (!local || cp->vport != cp->dport ||
649 !ip_vs_addr_equal(cp->af, &cp->vaddr, &cp->daddr))
650 ip_vs_drop_early_demux_sk(skb);
651
652 if (!local) {
653 skb_forward_csum(skb);
654 if (skb->dev)
655 skb_clear_tstamp(skb);
656 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
657 NULL, skb_dst(skb)->dev, dst_output);
658 } else
659 ret = NF_ACCEPT;
660
661 return ret;
662 }
663
664 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
ip_vs_send_or_cont(int pf,struct sk_buff * skb,struct ip_vs_conn * cp,int local)665 static inline int ip_vs_send_or_cont(int pf, struct sk_buff *skb,
666 struct ip_vs_conn *cp, int local)
667 {
668 int ret = NF_STOLEN;
669
670 skb->ipvs_property = 1;
671 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
672 ip_vs_notrack(skb);
673 if (!local) {
674 ip_vs_drop_early_demux_sk(skb);
675 skb_forward_csum(skb);
676 if (skb->dev)
677 skb_clear_tstamp(skb);
678 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
679 NULL, skb_dst(skb)->dev, dst_output);
680 } else
681 ret = NF_ACCEPT;
682 return ret;
683 }
684
685
686 /*
687 * NULL transmitter (do nothing except return NF_ACCEPT)
688 */
689 int
ip_vs_null_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)690 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
691 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
692 {
693 /* we do not touch skb and do not need pskb ptr */
694 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
695 }
696
697
698 /*
699 * Bypass transmitter
700 * Let packets bypass the destination when the destination is not
701 * available, it may be only used in transparent cache cluster.
702 */
703 int
ip_vs_bypass_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)704 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
705 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
706 {
707 struct iphdr *iph = ip_hdr(skb);
708
709 if (__ip_vs_get_out_rt(cp->ipvs, cp->af, skb, NULL, iph->daddr,
710 IP_VS_RT_MODE_NON_LOCAL, NULL, ipvsh) < 0)
711 goto tx_error;
712
713 ip_send_check(iph);
714
715 /* Another hack: avoid icmp_send in ip_fragment */
716 skb->ignore_df = 1;
717
718 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
719
720 return NF_STOLEN;
721
722 tx_error:
723 kfree_skb(skb);
724 return NF_STOLEN;
725 }
726
727 #ifdef CONFIG_IP_VS_IPV6
728 int
ip_vs_bypass_xmit_v6(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)729 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
730 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
731 {
732 struct ipv6hdr *iph = ipv6_hdr(skb);
733
734 if (__ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, NULL,
735 &iph->daddr, NULL,
736 ipvsh, 0, IP_VS_RT_MODE_NON_LOCAL) < 0)
737 goto tx_error;
738
739 /* Another hack: avoid icmp_send in ip_fragment */
740 skb->ignore_df = 1;
741
742 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
743
744 return NF_STOLEN;
745
746 tx_error:
747 kfree_skb(skb);
748 return NF_STOLEN;
749 }
750 #endif
751
752 /*
753 * NAT transmitter (only for outside-to-inside nat forwarding)
754 * Not used for related ICMP
755 */
756 int
ip_vs_nat_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)757 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
758 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
759 {
760 struct rtable *rt; /* Route to the other host */
761 int local, rc, was_input;
762
763 /* check if it is a connection of no-client-port */
764 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
765 __be16 _pt, *p;
766
767 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
768 if (p == NULL)
769 goto tx_error;
770 ip_vs_conn_fill_cport(cp, *p);
771 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
772 }
773
774 was_input = rt_is_input_route(skb_rtable(skb));
775 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
776 IP_VS_RT_MODE_LOCAL |
777 IP_VS_RT_MODE_NON_LOCAL |
778 IP_VS_RT_MODE_RDR, NULL, ipvsh);
779 if (local < 0)
780 goto tx_error;
781 rt = skb_rtable(skb);
782 /*
783 * Avoid duplicate tuple in reply direction for NAT traffic
784 * to local address when connection is sync-ed
785 */
786 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
787 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
788 enum ip_conntrack_info ctinfo;
789 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
790
791 if (ct) {
792 IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, ipvsh->off,
793 "ip_vs_nat_xmit(): "
794 "stopping DNAT to local address");
795 goto tx_error;
796 }
797 }
798 #endif
799
800 /* From world but DNAT to loopback address? */
801 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
802 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, ipvsh->off,
803 "ip_vs_nat_xmit(): stopping DNAT to loopback "
804 "address");
805 goto tx_error;
806 }
807
808 /* copy-on-write the packet before mangling it */
809 if (skb_ensure_writable(skb, sizeof(struct iphdr)))
810 goto tx_error;
811
812 if (skb_cow(skb, rt->dst.dev->hard_header_len))
813 goto tx_error;
814
815 /* mangle the packet */
816 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
817 goto tx_error;
818 ip_hdr(skb)->daddr = cp->daddr.ip;
819 ip_send_check(ip_hdr(skb));
820
821 IP_VS_DBG_PKT(10, AF_INET, pp, skb, ipvsh->off, "After DNAT");
822
823 /* FIXME: when application helper enlarges the packet and the length
824 is larger than the MTU of outgoing device, there will be still
825 MTU problem. */
826
827 /* Another hack: avoid icmp_send in ip_fragment */
828 skb->ignore_df = 1;
829
830 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
831
832 return rc;
833
834 tx_error:
835 kfree_skb(skb);
836 return NF_STOLEN;
837 }
838
839 #ifdef CONFIG_IP_VS_IPV6
840 int
ip_vs_nat_xmit_v6(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)841 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
842 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
843 {
844 struct rt6_info *rt; /* Route to the other host */
845 int local, rc;
846
847 /* check if it is a connection of no-client-port */
848 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT && !ipvsh->fragoffs)) {
849 __be16 _pt, *p;
850 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
851 if (p == NULL)
852 goto tx_error;
853 ip_vs_conn_fill_cport(cp, *p);
854 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
855 }
856
857 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
858 &cp->daddr.in6,
859 NULL, ipvsh, 0,
860 IP_VS_RT_MODE_LOCAL |
861 IP_VS_RT_MODE_NON_LOCAL |
862 IP_VS_RT_MODE_RDR);
863 if (local < 0)
864 goto tx_error;
865 rt = dst_rt6_info(skb_dst(skb));
866 /*
867 * Avoid duplicate tuple in reply direction for NAT traffic
868 * to local address when connection is sync-ed
869 */
870 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
871 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
872 enum ip_conntrack_info ctinfo;
873 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
874
875 if (ct) {
876 IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, ipvsh->off,
877 "ip_vs_nat_xmit_v6(): "
878 "stopping DNAT to local address");
879 goto tx_error;
880 }
881 }
882 #endif
883
884 /* From world but DNAT to loopback address? */
885 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
886 ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
887 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, ipvsh->off,
888 "ip_vs_nat_xmit_v6(): "
889 "stopping DNAT to loopback address");
890 goto tx_error;
891 }
892
893 /* copy-on-write the packet before mangling it */
894 if (skb_ensure_writable(skb, sizeof(struct ipv6hdr)))
895 goto tx_error;
896
897 if (skb_cow(skb, rt->dst.dev->hard_header_len))
898 goto tx_error;
899
900 /* mangle the packet */
901 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
902 goto tx_error;
903 ipv6_hdr(skb)->daddr = cp->daddr.in6;
904
905 IP_VS_DBG_PKT(10, AF_INET6, pp, skb, ipvsh->off, "After DNAT");
906
907 /* FIXME: when application helper enlarges the packet and the length
908 is larger than the MTU of outgoing device, there will be still
909 MTU problem. */
910
911 /* Another hack: avoid icmp_send in ip_fragment */
912 skb->ignore_df = 1;
913
914 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
915
916 return rc;
917
918 tx_error:
919 kfree_skb(skb);
920 return NF_STOLEN;
921 }
922 #endif
923
924 /* When forwarding a packet, we must ensure that we've got enough headroom
925 * for the encapsulation packet in the skb. This also gives us an
926 * opportunity to figure out what the payload_len, dsfield, ttl, and df
927 * values should be, so that we won't need to look at the old ip header
928 * again
929 */
930 static struct sk_buff *
ip_vs_prepare_tunneled_skb(struct sk_buff * skb,int skb_af,unsigned int max_headroom,__u8 * next_protocol,__u32 * payload_len,__u8 * dsfield,__u8 * ttl,__be16 * df)931 ip_vs_prepare_tunneled_skb(struct sk_buff *skb, int skb_af,
932 unsigned int max_headroom, __u8 *next_protocol,
933 __u32 *payload_len, __u8 *dsfield, __u8 *ttl,
934 __be16 *df)
935 {
936 struct sk_buff *new_skb = NULL;
937 struct iphdr *old_iph = NULL;
938 __u8 old_dsfield;
939 #ifdef CONFIG_IP_VS_IPV6
940 struct ipv6hdr *old_ipv6h = NULL;
941 #endif
942
943 ip_vs_drop_early_demux_sk(skb);
944
945 if (skb_headroom(skb) < max_headroom || skb_cloned(skb)) {
946 new_skb = skb_realloc_headroom(skb, max_headroom);
947 if (!new_skb)
948 goto error;
949 if (skb->sk)
950 skb_set_owner_w(new_skb, skb->sk);
951 consume_skb(skb);
952 skb = new_skb;
953 }
954
955 #ifdef CONFIG_IP_VS_IPV6
956 if (skb_af == AF_INET6) {
957 old_ipv6h = ipv6_hdr(skb);
958 *next_protocol = IPPROTO_IPV6;
959 if (payload_len)
960 *payload_len =
961 ntohs(old_ipv6h->payload_len) +
962 sizeof(*old_ipv6h);
963 old_dsfield = ipv6_get_dsfield(old_ipv6h);
964 *ttl = old_ipv6h->hop_limit;
965 if (df)
966 *df = 0;
967 } else
968 #endif
969 {
970 old_iph = ip_hdr(skb);
971 /* Copy DF, reset fragment offset and MF */
972 if (df)
973 *df = (old_iph->frag_off & htons(IP_DF));
974 *next_protocol = IPPROTO_IPIP;
975
976 /* fix old IP header checksum */
977 ip_send_check(old_iph);
978 old_dsfield = ipv4_get_dsfield(old_iph);
979 *ttl = old_iph->ttl;
980 if (payload_len)
981 *payload_len = skb_ip_totlen(skb);
982 }
983
984 /* Implement full-functionality option for ECN encapsulation */
985 *dsfield = INET_ECN_encapsulate(old_dsfield, old_dsfield);
986
987 return skb;
988 error:
989 kfree_skb(skb);
990 return ERR_PTR(-ENOMEM);
991 }
992
__tun_gso_type_mask(int encaps_af,int orig_af)993 static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
994 {
995 switch (encaps_af) {
996 case AF_INET:
997 return SKB_GSO_IPXIP4;
998 case AF_INET6:
999 return SKB_GSO_IPXIP6;
1000 default:
1001 return 0;
1002 }
1003 }
1004
1005 static int
ipvs_gue_encap(struct net * net,struct sk_buff * skb,struct ip_vs_conn * cp,__u8 * next_protocol)1006 ipvs_gue_encap(struct net *net, struct sk_buff *skb,
1007 struct ip_vs_conn *cp, __u8 *next_protocol)
1008 {
1009 __be16 dport;
1010 __be16 sport = udp_flow_src_port(net, skb, 0, 0, false);
1011 struct udphdr *udph; /* Our new UDP header */
1012 struct guehdr *gueh; /* Our new GUE header */
1013 size_t hdrlen, optlen = 0;
1014 void *data;
1015 bool need_priv = false;
1016
1017 if ((cp->dest->tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
1018 skb->ip_summed == CHECKSUM_PARTIAL) {
1019 optlen += GUE_PLEN_REMCSUM + GUE_LEN_PRIV;
1020 need_priv = true;
1021 }
1022
1023 hdrlen = sizeof(struct guehdr) + optlen;
1024
1025 skb_push(skb, hdrlen);
1026
1027 gueh = (struct guehdr *)skb->data;
1028
1029 gueh->control = 0;
1030 gueh->version = 0;
1031 gueh->hlen = optlen >> 2;
1032 gueh->flags = 0;
1033 gueh->proto_ctype = *next_protocol;
1034
1035 data = &gueh[1];
1036
1037 if (need_priv) {
1038 __be32 *flags = data;
1039 u16 csum_start = skb_checksum_start_offset(skb);
1040 __be16 *pd;
1041
1042 gueh->flags |= GUE_FLAG_PRIV;
1043 *flags = 0;
1044 data += GUE_LEN_PRIV;
1045
1046 if (csum_start < hdrlen)
1047 return -EINVAL;
1048
1049 csum_start -= hdrlen;
1050 pd = data;
1051 pd[0] = htons(csum_start);
1052 pd[1] = htons(csum_start + skb->csum_offset);
1053
1054 if (!skb_is_gso(skb)) {
1055 skb->ip_summed = CHECKSUM_NONE;
1056 skb->encapsulation = 0;
1057 }
1058
1059 *flags |= GUE_PFLAG_REMCSUM;
1060 data += GUE_PLEN_REMCSUM;
1061 }
1062
1063 skb_push(skb, sizeof(struct udphdr));
1064 skb_reset_transport_header(skb);
1065
1066 udph = udp_hdr(skb);
1067
1068 dport = cp->dest->tun_port;
1069 udph->dest = dport;
1070 udph->source = sport;
1071 udph->len = htons(skb->len);
1072 udph->check = 0;
1073
1074 *next_protocol = IPPROTO_UDP;
1075
1076 return 0;
1077 }
1078
1079 static void
ipvs_gre_encap(struct net * net,struct sk_buff * skb,struct ip_vs_conn * cp,__u8 * next_protocol)1080 ipvs_gre_encap(struct net *net, struct sk_buff *skb,
1081 struct ip_vs_conn *cp, __u8 *next_protocol)
1082 {
1083 __be16 proto = *next_protocol == IPPROTO_IPIP ?
1084 htons(ETH_P_IP) : htons(ETH_P_IPV6);
1085 IP_TUNNEL_DECLARE_FLAGS(tflags) = { };
1086 size_t hdrlen;
1087
1088 if (cp->dest->tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
1089 __set_bit(IP_TUNNEL_CSUM_BIT, tflags);
1090
1091 hdrlen = gre_calc_hlen(tflags);
1092 gre_build_header(skb, hdrlen, tflags, proto, 0, 0);
1093
1094 *next_protocol = IPPROTO_GRE;
1095 }
1096
1097 /*
1098 * IP Tunneling transmitter
1099 *
1100 * This function encapsulates the packet in a new IP packet, its
1101 * destination will be set to cp->daddr. Most code of this function
1102 * is taken from ipip.c.
1103 *
1104 * It is used in VS/TUN cluster. The load balancer selects a real
1105 * server from a cluster based on a scheduling algorithm,
1106 * encapsulates the request packet and forwards it to the selected
1107 * server. For example, all real servers are configured with
1108 * "ifconfig tunl0 <Virtual IP Address> up". When the server receives
1109 * the encapsulated packet, it will decapsulate the packet, processe
1110 * the request and return the response packets directly to the client
1111 * without passing the load balancer. This can greatly increase the
1112 * scalability of virtual server.
1113 *
1114 * Used for ANY protocol
1115 */
1116 int
ip_vs_tunnel_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)1117 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1118 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1119 {
1120 struct netns_ipvs *ipvs = cp->ipvs;
1121 struct net *net = ipvs->net;
1122 struct rtable *rt; /* Route to the other host */
1123 __be32 saddr; /* Source for tunnel */
1124 struct net_device *tdev; /* Device to other host */
1125 __u8 next_protocol = 0;
1126 __u8 dsfield = 0;
1127 __u8 ttl = 0;
1128 __be16 df = 0;
1129 __be16 *dfp = NULL;
1130 struct iphdr *iph; /* Our new IP header */
1131 unsigned int max_headroom; /* The extra header space needed */
1132 int ret, local;
1133 int tun_type, gso_type;
1134 int tun_flags;
1135
1136 local = __ip_vs_get_out_rt(ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1137 IP_VS_RT_MODE_LOCAL |
1138 IP_VS_RT_MODE_NON_LOCAL |
1139 IP_VS_RT_MODE_CONNECT |
1140 IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
1141 if (local < 0)
1142 goto tx_error;
1143 if (local)
1144 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1145
1146 rt = skb_rtable(skb);
1147 tdev = rt->dst.dev;
1148
1149 /*
1150 * Okay, now see if we can stuff it in the buffer as-is.
1151 */
1152 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
1153
1154 tun_type = cp->dest->tun_type;
1155 tun_flags = cp->dest->tun_flags;
1156
1157 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1158 size_t gue_hdrlen, gue_optlen = 0;
1159
1160 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
1161 skb->ip_summed == CHECKSUM_PARTIAL) {
1162 gue_optlen += GUE_PLEN_REMCSUM + GUE_LEN_PRIV;
1163 }
1164 gue_hdrlen = sizeof(struct guehdr) + gue_optlen;
1165
1166 max_headroom += sizeof(struct udphdr) + gue_hdrlen;
1167 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
1168 IP_TUNNEL_DECLARE_FLAGS(tflags) = { };
1169 size_t gre_hdrlen;
1170
1171 if (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
1172 __set_bit(IP_TUNNEL_CSUM_BIT, tflags);
1173 gre_hdrlen = gre_calc_hlen(tflags);
1174
1175 max_headroom += gre_hdrlen;
1176 }
1177
1178 /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
1179 dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
1180 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1181 &next_protocol, NULL, &dsfield,
1182 &ttl, dfp);
1183 if (IS_ERR(skb))
1184 return NF_STOLEN;
1185
1186 gso_type = __tun_gso_type_mask(AF_INET, cp->af);
1187 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1188 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM) ||
1189 (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM))
1190 gso_type |= SKB_GSO_UDP_TUNNEL_CSUM;
1191 else
1192 gso_type |= SKB_GSO_UDP_TUNNEL;
1193 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
1194 skb->ip_summed == CHECKSUM_PARTIAL) {
1195 gso_type |= SKB_GSO_TUNNEL_REMCSUM;
1196 }
1197 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
1198 if (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
1199 gso_type |= SKB_GSO_GRE_CSUM;
1200 else
1201 gso_type |= SKB_GSO_GRE;
1202 }
1203
1204 if (iptunnel_handle_offloads(skb, gso_type))
1205 goto tx_error;
1206
1207 skb->transport_header = skb->network_header;
1208
1209 skb_set_inner_ipproto(skb, next_protocol);
1210 skb_set_inner_mac_header(skb, skb_inner_network_offset(skb));
1211
1212 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1213 bool check = false;
1214
1215 if (ipvs_gue_encap(net, skb, cp, &next_protocol))
1216 goto tx_error;
1217
1218 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM) ||
1219 (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM))
1220 check = true;
1221
1222 udp_set_csum(!check, skb, saddr, cp->daddr.ip, skb->len);
1223 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE)
1224 ipvs_gre_encap(net, skb, cp, &next_protocol);
1225
1226 skb_push(skb, sizeof(struct iphdr));
1227 skb_reset_network_header(skb);
1228 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1229
1230 /*
1231 * Push down and install the IPIP header.
1232 */
1233 iph = ip_hdr(skb);
1234 iph->version = 4;
1235 iph->ihl = sizeof(struct iphdr)>>2;
1236 iph->frag_off = df;
1237 iph->protocol = next_protocol;
1238 iph->tos = dsfield;
1239 iph->daddr = cp->daddr.ip;
1240 iph->saddr = saddr;
1241 iph->ttl = ttl;
1242 ip_select_ident(net, skb, NULL);
1243
1244 /* Another hack: avoid icmp_send in ip_fragment */
1245 skb->ignore_df = 1;
1246
1247 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1248 if (ret == NF_ACCEPT)
1249 ip_local_out(net, skb->sk, skb);
1250 else if (ret == NF_DROP)
1251 kfree_skb(skb);
1252
1253 return NF_STOLEN;
1254
1255 tx_error:
1256 kfree_skb(skb);
1257 return NF_STOLEN;
1258 }
1259
1260 #ifdef CONFIG_IP_VS_IPV6
1261 int
ip_vs_tunnel_xmit_v6(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)1262 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1263 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1264 {
1265 struct netns_ipvs *ipvs = cp->ipvs;
1266 struct net *net = ipvs->net;
1267 struct rt6_info *rt; /* Route to the other host */
1268 struct in6_addr saddr; /* Source for tunnel */
1269 struct net_device *tdev; /* Device to other host */
1270 __u8 next_protocol = 0;
1271 __u32 payload_len = 0;
1272 __u8 dsfield = 0;
1273 __u8 ttl = 0;
1274 struct ipv6hdr *iph; /* Our new IP header */
1275 unsigned int max_headroom; /* The extra header space needed */
1276 int ret, local;
1277 int tun_type, gso_type;
1278 int tun_flags;
1279
1280 local = __ip_vs_get_out_rt_v6(ipvs, cp->af, skb, cp->dest,
1281 &cp->daddr.in6,
1282 &saddr, ipvsh, 1,
1283 IP_VS_RT_MODE_LOCAL |
1284 IP_VS_RT_MODE_NON_LOCAL |
1285 IP_VS_RT_MODE_TUNNEL);
1286 if (local < 0)
1287 goto tx_error;
1288 if (local)
1289 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1290
1291 rt = dst_rt6_info(skb_dst(skb));
1292 tdev = rt->dst.dev;
1293
1294 /*
1295 * Okay, now see if we can stuff it in the buffer as-is.
1296 */
1297 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
1298
1299 tun_type = cp->dest->tun_type;
1300 tun_flags = cp->dest->tun_flags;
1301
1302 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1303 size_t gue_hdrlen, gue_optlen = 0;
1304
1305 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
1306 skb->ip_summed == CHECKSUM_PARTIAL) {
1307 gue_optlen += GUE_PLEN_REMCSUM + GUE_LEN_PRIV;
1308 }
1309 gue_hdrlen = sizeof(struct guehdr) + gue_optlen;
1310
1311 max_headroom += sizeof(struct udphdr) + gue_hdrlen;
1312 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
1313 IP_TUNNEL_DECLARE_FLAGS(tflags) = { };
1314 size_t gre_hdrlen;
1315
1316 if (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
1317 __set_bit(IP_TUNNEL_CSUM_BIT, tflags);
1318 gre_hdrlen = gre_calc_hlen(tflags);
1319
1320 max_headroom += gre_hdrlen;
1321 }
1322
1323 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1324 &next_protocol, &payload_len,
1325 &dsfield, &ttl, NULL);
1326 if (IS_ERR(skb))
1327 return NF_STOLEN;
1328
1329 gso_type = __tun_gso_type_mask(AF_INET6, cp->af);
1330 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1331 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM) ||
1332 (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM))
1333 gso_type |= SKB_GSO_UDP_TUNNEL_CSUM;
1334 else
1335 gso_type |= SKB_GSO_UDP_TUNNEL;
1336 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
1337 skb->ip_summed == CHECKSUM_PARTIAL) {
1338 gso_type |= SKB_GSO_TUNNEL_REMCSUM;
1339 }
1340 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
1341 if (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
1342 gso_type |= SKB_GSO_GRE_CSUM;
1343 else
1344 gso_type |= SKB_GSO_GRE;
1345 }
1346
1347 if (iptunnel_handle_offloads(skb, gso_type))
1348 goto tx_error;
1349
1350 skb->transport_header = skb->network_header;
1351
1352 skb_set_inner_ipproto(skb, next_protocol);
1353 skb_set_inner_mac_header(skb, skb_inner_network_offset(skb));
1354
1355 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1356 bool check = false;
1357
1358 if (ipvs_gue_encap(net, skb, cp, &next_protocol))
1359 goto tx_error;
1360
1361 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM) ||
1362 (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM))
1363 check = true;
1364
1365 udp6_set_csum(!check, skb, &saddr, &cp->daddr.in6, skb->len);
1366 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE)
1367 ipvs_gre_encap(net, skb, cp, &next_protocol);
1368
1369 skb_push(skb, sizeof(struct ipv6hdr));
1370 skb_reset_network_header(skb);
1371 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1372
1373 /*
1374 * Push down and install the IPIP header.
1375 */
1376 iph = ipv6_hdr(skb);
1377 iph->version = 6;
1378 iph->nexthdr = next_protocol;
1379 iph->payload_len = htons(payload_len);
1380 memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
1381 ipv6_change_dsfield(iph, 0, dsfield);
1382 iph->daddr = cp->daddr.in6;
1383 iph->saddr = saddr;
1384 iph->hop_limit = ttl;
1385
1386 /* Another hack: avoid icmp_send in ip_fragment */
1387 skb->ignore_df = 1;
1388
1389 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1390 if (ret == NF_ACCEPT)
1391 ip6_local_out(net, skb->sk, skb);
1392 else if (ret == NF_DROP)
1393 kfree_skb(skb);
1394
1395 return NF_STOLEN;
1396
1397 tx_error:
1398 kfree_skb(skb);
1399 return NF_STOLEN;
1400 }
1401 #endif
1402
1403
1404 /*
1405 * Direct Routing transmitter
1406 * Used for ANY protocol
1407 */
1408 int
ip_vs_dr_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)1409 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1410 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1411 {
1412 int local;
1413
1414 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1415 IP_VS_RT_MODE_LOCAL |
1416 IP_VS_RT_MODE_NON_LOCAL |
1417 IP_VS_RT_MODE_KNOWN_NH, NULL, ipvsh);
1418 if (local < 0)
1419 goto tx_error;
1420 if (local)
1421 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1422
1423 ip_send_check(ip_hdr(skb));
1424
1425 /* Another hack: avoid icmp_send in ip_fragment */
1426 skb->ignore_df = 1;
1427
1428 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
1429
1430 return NF_STOLEN;
1431
1432 tx_error:
1433 kfree_skb(skb);
1434 return NF_STOLEN;
1435 }
1436
1437 #ifdef CONFIG_IP_VS_IPV6
1438 int
ip_vs_dr_xmit_v6(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)1439 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1440 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1441 {
1442 int local;
1443
1444 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1445 &cp->daddr.in6,
1446 NULL, ipvsh, 0,
1447 IP_VS_RT_MODE_LOCAL |
1448 IP_VS_RT_MODE_NON_LOCAL |
1449 IP_VS_RT_MODE_KNOWN_NH);
1450 if (local < 0)
1451 goto tx_error;
1452 if (local)
1453 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1454
1455 /* Another hack: avoid icmp_send in ip_fragment */
1456 skb->ignore_df = 1;
1457
1458 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
1459
1460 return NF_STOLEN;
1461
1462 tx_error:
1463 kfree_skb(skb);
1464 return NF_STOLEN;
1465 }
1466 #endif
1467
1468
1469 /*
1470 * ICMP packet transmitter
1471 * called by the ip_vs_in_icmp
1472 */
1473 int
ip_vs_icmp_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,int offset,unsigned int hooknum,struct ip_vs_iphdr * iph)1474 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1475 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1476 struct ip_vs_iphdr *iph)
1477 {
1478 struct rtable *rt; /* Route to the other host */
1479 int rc;
1480 int local;
1481 int rt_mode, was_input;
1482
1483 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1484 forwarded directly here, because there is no need to
1485 translate address/port back */
1486 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1487 if (cp->packet_xmit)
1488 rc = cp->packet_xmit(skb, cp, pp, iph);
1489 else
1490 rc = NF_ACCEPT;
1491 /* do not touch skb anymore */
1492 atomic_inc(&cp->in_pkts);
1493 return rc;
1494 }
1495
1496 /*
1497 * mangle and send the packet here (only for VS/NAT)
1498 */
1499 was_input = rt_is_input_route(skb_rtable(skb));
1500
1501 /* LOCALNODE from FORWARD hook is not supported */
1502 rt_mode = (hooknum != NF_INET_FORWARD) ?
1503 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1504 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1505 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip, rt_mode,
1506 NULL, iph);
1507 if (local < 0)
1508 goto tx_error;
1509 rt = skb_rtable(skb);
1510
1511 /*
1512 * Avoid duplicate tuple in reply direction for NAT traffic
1513 * to local address when connection is sync-ed
1514 */
1515 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1516 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1517 enum ip_conntrack_info ctinfo;
1518 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1519
1520 if (ct) {
1521 IP_VS_DBG(10, "%s(): "
1522 "stopping DNAT to local address %pI4\n",
1523 __func__, &cp->daddr.ip);
1524 goto tx_error;
1525 }
1526 }
1527 #endif
1528
1529 /* From world but DNAT to loopback address? */
1530 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
1531 IP_VS_DBG(1, "%s(): "
1532 "stopping DNAT to loopback %pI4\n",
1533 __func__, &cp->daddr.ip);
1534 goto tx_error;
1535 }
1536
1537 /* copy-on-write the packet before mangling it */
1538 if (skb_ensure_writable(skb, offset))
1539 goto tx_error;
1540
1541 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1542 goto tx_error;
1543
1544 ip_vs_nat_icmp(skb, pp, cp, 0);
1545
1546 /* Another hack: avoid icmp_send in ip_fragment */
1547 skb->ignore_df = 1;
1548
1549 return ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
1550
1551 tx_error:
1552 kfree_skb(skb);
1553 rc = NF_STOLEN;
1554 return rc;
1555 }
1556
1557 #ifdef CONFIG_IP_VS_IPV6
1558 int
ip_vs_icmp_xmit_v6(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,int offset,unsigned int hooknum,struct ip_vs_iphdr * ipvsh)1559 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1560 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1561 struct ip_vs_iphdr *ipvsh)
1562 {
1563 struct rt6_info *rt; /* Route to the other host */
1564 int rc;
1565 int local;
1566 int rt_mode;
1567
1568 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1569 forwarded directly here, because there is no need to
1570 translate address/port back */
1571 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1572 if (cp->packet_xmit)
1573 rc = cp->packet_xmit(skb, cp, pp, ipvsh);
1574 else
1575 rc = NF_ACCEPT;
1576 /* do not touch skb anymore */
1577 atomic_inc(&cp->in_pkts);
1578 return rc;
1579 }
1580
1581 /*
1582 * mangle and send the packet here (only for VS/NAT)
1583 */
1584
1585 /* LOCALNODE from FORWARD hook is not supported */
1586 rt_mode = (hooknum != NF_INET_FORWARD) ?
1587 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1588 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1589 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1590 &cp->daddr.in6, NULL, ipvsh, 0, rt_mode);
1591 if (local < 0)
1592 goto tx_error;
1593 rt = dst_rt6_info(skb_dst(skb));
1594 /*
1595 * Avoid duplicate tuple in reply direction for NAT traffic
1596 * to local address when connection is sync-ed
1597 */
1598 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1599 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1600 enum ip_conntrack_info ctinfo;
1601 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1602
1603 if (ct) {
1604 IP_VS_DBG(10, "%s(): "
1605 "stopping DNAT to local address %pI6\n",
1606 __func__, &cp->daddr.in6);
1607 goto tx_error;
1608 }
1609 }
1610 #endif
1611
1612 /* From world but DNAT to loopback address? */
1613 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1614 ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
1615 IP_VS_DBG(1, "%s(): "
1616 "stopping DNAT to loopback %pI6\n",
1617 __func__, &cp->daddr.in6);
1618 goto tx_error;
1619 }
1620
1621 /* copy-on-write the packet before mangling it */
1622 if (skb_ensure_writable(skb, offset))
1623 goto tx_error;
1624
1625 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1626 goto tx_error;
1627
1628 ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1629
1630 /* Another hack: avoid icmp_send in ip_fragment */
1631 skb->ignore_df = 1;
1632
1633 return ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
1634
1635 tx_error:
1636 kfree_skb(skb);
1637 rc = NF_STOLEN;
1638 return rc;
1639 }
1640 #endif
1641