1 /*
2 * IPVS An implementation of the IP virtual server support for the
3 * LINUX operating system. IPVS is now implemented as a module
4 * over the Netfilter framework. IPVS can be used to build a
5 * high-performance and highly available server based on a
6 * cluster of servers.
7 *
8 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
9 * Peter Kese <peter.kese@ijs.si>
10 * Julian Anastasov <ja@ssi.bg>
11 *
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
16 *
17 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
18 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
19 * and others.
20 *
21 * Changes:
22 * Paul `Rusty' Russell properly handle non-linear skbs
23 * Harald Welte don't use nfcache
24 *
25 */
26
27 #define KMSG_COMPONENT "IPVS"
28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
29
30 #include <linux/module.h>
31 #include <linux/kernel.h>
32 #include <linux/ip.h>
33 #include <linux/tcp.h>
34 #include <linux/sctp.h>
35 #include <linux/icmp.h>
36 #include <linux/slab.h>
37
38 #include <net/ip.h>
39 #include <net/tcp.h>
40 #include <net/udp.h>
41 #include <net/icmp.h> /* for icmp_send */
42 #include <net/route.h>
43 #include <net/ip6_checksum.h>
44 #include <net/netns/generic.h> /* net_generic() */
45
46 #include <linux/netfilter.h>
47 #include <linux/netfilter_ipv4.h>
48
49 #ifdef CONFIG_IP_VS_IPV6
50 #include <net/ipv6.h>
51 #include <linux/netfilter_ipv6.h>
52 #include <net/ip6_route.h>
53 #endif
54
55 #include <net/ip_vs.h>
56
57
58 EXPORT_SYMBOL(register_ip_vs_scheduler);
59 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
60 EXPORT_SYMBOL(ip_vs_proto_name);
61 EXPORT_SYMBOL(ip_vs_conn_new);
62 EXPORT_SYMBOL(ip_vs_conn_in_get);
63 EXPORT_SYMBOL(ip_vs_conn_out_get);
64 #ifdef CONFIG_IP_VS_PROTO_TCP
65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
66 #endif
67 EXPORT_SYMBOL(ip_vs_conn_put);
68 #ifdef CONFIG_IP_VS_DEBUG
69 EXPORT_SYMBOL(ip_vs_get_debug_level);
70 #endif
71
72 static int ip_vs_net_id __read_mostly;
73 /* netns cnt used for uniqueness */
74 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
75
76 /* ID used in ICMP lookups */
77 #define icmp_id(icmph) (((icmph)->un).echo.id)
78 #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier)
79
80 const char *ip_vs_proto_name(unsigned int proto)
81 {
82 static char buf[20];
83
84 switch (proto) {
85 case IPPROTO_IP:
86 return "IP";
87 case IPPROTO_UDP:
88 return "UDP";
89 case IPPROTO_TCP:
90 return "TCP";
91 case IPPROTO_SCTP:
92 return "SCTP";
93 case IPPROTO_ICMP:
94 return "ICMP";
95 #ifdef CONFIG_IP_VS_IPV6
96 case IPPROTO_ICMPV6:
97 return "ICMPv6";
98 #endif
99 default:
100 sprintf(buf, "IP_%u", proto);
101 return buf;
102 }
103 }
104
105 void ip_vs_init_hash_table(struct list_head *table, int rows)
106 {
107 while (--rows >= 0)
108 INIT_LIST_HEAD(&table[rows]);
109 }
110
111 static inline void
112 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
113 {
114 struct ip_vs_dest *dest = cp->dest;
115 struct netns_ipvs *ipvs = cp->ipvs;
116
117 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
118 struct ip_vs_cpu_stats *s;
119 struct ip_vs_service *svc;
120
121 s = this_cpu_ptr(dest->stats.cpustats);
122 u64_stats_update_begin(&s->syncp);
123 s->cnt.inpkts++;
124 s->cnt.inbytes += skb->len;
125 u64_stats_update_end(&s->syncp);
126
127 rcu_read_lock();
128 svc = rcu_dereference(dest->svc);
129 s = this_cpu_ptr(svc->stats.cpustats);
130 u64_stats_update_begin(&s->syncp);
131 s->cnt.inpkts++;
132 s->cnt.inbytes += skb->len;
133 u64_stats_update_end(&s->syncp);
134 rcu_read_unlock();
135
136 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
137 u64_stats_update_begin(&s->syncp);
138 s->cnt.inpkts++;
139 s->cnt.inbytes += skb->len;
140 u64_stats_update_end(&s->syncp);
141 }
142 }
143
144
145 static inline void
146 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
147 {
148 struct ip_vs_dest *dest = cp->dest;
149 struct netns_ipvs *ipvs = cp->ipvs;
150
151 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
152 struct ip_vs_cpu_stats *s;
153 struct ip_vs_service *svc;
154
155 s = this_cpu_ptr(dest->stats.cpustats);
156 u64_stats_update_begin(&s->syncp);
157 s->cnt.outpkts++;
158 s->cnt.outbytes += skb->len;
159 u64_stats_update_end(&s->syncp);
160
161 rcu_read_lock();
162 svc = rcu_dereference(dest->svc);
163 s = this_cpu_ptr(svc->stats.cpustats);
164 u64_stats_update_begin(&s->syncp);
165 s->cnt.outpkts++;
166 s->cnt.outbytes += skb->len;
167 u64_stats_update_end(&s->syncp);
168 rcu_read_unlock();
169
170 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
171 u64_stats_update_begin(&s->syncp);
172 s->cnt.outpkts++;
173 s->cnt.outbytes += skb->len;
174 u64_stats_update_end(&s->syncp);
175 }
176 }
177
178
179 static inline void
180 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
181 {
182 struct netns_ipvs *ipvs = svc->ipvs;
183 struct ip_vs_cpu_stats *s;
184
185 s = this_cpu_ptr(cp->dest->stats.cpustats);
186 u64_stats_update_begin(&s->syncp);
187 s->cnt.conns++;
188 u64_stats_update_end(&s->syncp);
189
190 s = this_cpu_ptr(svc->stats.cpustats);
191 u64_stats_update_begin(&s->syncp);
192 s->cnt.conns++;
193 u64_stats_update_end(&s->syncp);
194
195 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
196 u64_stats_update_begin(&s->syncp);
197 s->cnt.conns++;
198 u64_stats_update_end(&s->syncp);
199 }
200
201
202 static inline void
203 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
204 const struct sk_buff *skb,
205 struct ip_vs_proto_data *pd)
206 {
207 if (likely(pd->pp->state_transition))
208 pd->pp->state_transition(cp, direction, skb, pd);
209 }
210
211 static inline int
212 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
213 struct sk_buff *skb, int protocol,
214 const union nf_inet_addr *caddr, __be16 cport,
215 const union nf_inet_addr *vaddr, __be16 vport,
216 struct ip_vs_conn_param *p)
217 {
218 ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
219 vport, p);
220 p->pe = rcu_dereference(svc->pe);
221 if (p->pe && p->pe->fill_param)
222 return p->pe->fill_param(p, skb);
223
224 return 0;
225 }
226
227 /*
228 * IPVS persistent scheduling function
229 * It creates a connection entry according to its template if exists,
230 * or selects a server and creates a connection entry plus a template.
231 * Locking: we are svc user (svc->refcnt), so we hold all dests too
232 * Protocols supported: TCP, UDP
233 */
234 static struct ip_vs_conn *
235 ip_vs_sched_persist(struct ip_vs_service *svc,
236 struct sk_buff *skb, __be16 src_port, __be16 dst_port,
237 int *ignored, struct ip_vs_iphdr *iph)
238 {
239 struct ip_vs_conn *cp = NULL;
240 struct ip_vs_dest *dest;
241 struct ip_vs_conn *ct;
242 __be16 dport = 0; /* destination port to forward */
243 unsigned int flags;
244 struct ip_vs_conn_param param;
245 const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
246 union nf_inet_addr snet; /* source network of the client,
247 after masking */
248 const union nf_inet_addr *src_addr, *dst_addr;
249
250 if (likely(!ip_vs_iph_inverse(iph))) {
251 src_addr = &iph->saddr;
252 dst_addr = &iph->daddr;
253 } else {
254 src_addr = &iph->daddr;
255 dst_addr = &iph->saddr;
256 }
257
258
259 /* Mask saddr with the netmask to adjust template granularity */
260 #ifdef CONFIG_IP_VS_IPV6
261 if (svc->af == AF_INET6)
262 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
263 (__force __u32) svc->netmask);
264 else
265 #endif
266 snet.ip = src_addr->ip & svc->netmask;
267
268 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
269 "mnet %s\n",
270 IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
271 IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
272 IP_VS_DBG_ADDR(svc->af, &snet));
273
274 /*
275 * As far as we know, FTP is a very complicated network protocol, and
276 * it uses control connection and data connections. For active FTP,
277 * FTP server initialize data connection to the client, its source port
278 * is often 20. For passive FTP, FTP server tells the clients the port
279 * that it passively listens to, and the client issues the data
280 * connection. In the tunneling or direct routing mode, the load
281 * balancer is on the client-to-server half of connection, the port
282 * number is unknown to the load balancer. So, a conn template like
283 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
284 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
285 * is created for other persistent services.
286 */
287 {
288 int protocol = iph->protocol;
289 const union nf_inet_addr *vaddr = dst_addr;
290 __be16 vport = 0;
291
292 if (dst_port == svc->port) {
293 /* non-FTP template:
294 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
295 * FTP template:
296 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
297 */
298 if (svc->port != FTPPORT)
299 vport = dst_port;
300 } else {
301 /* Note: persistent fwmark-based services and
302 * persistent port zero service are handled here.
303 * fwmark template:
304 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
305 * port zero template:
306 * <protocol,caddr,0,vaddr,0,daddr,0>
307 */
308 if (svc->fwmark) {
309 protocol = IPPROTO_IP;
310 vaddr = &fwmark;
311 }
312 }
313 /* return *ignored = -1 so NF_DROP can be used */
314 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
315 vaddr, vport, ¶m) < 0) {
316 *ignored = -1;
317 return NULL;
318 }
319 }
320
321 /* Check if a template already exists */
322 ct = ip_vs_ct_in_get(¶m);
323 if (!ct || !ip_vs_check_template(ct)) {
324 struct ip_vs_scheduler *sched;
325
326 /*
327 * No template found or the dest of the connection
328 * template is not available.
329 * return *ignored=0 i.e. ICMP and NF_DROP
330 */
331 sched = rcu_dereference(svc->scheduler);
332 if (sched) {
333 /* read svc->sched_data after svc->scheduler */
334 smp_rmb();
335 dest = sched->schedule(svc, skb, iph);
336 } else {
337 dest = NULL;
338 }
339 if (!dest) {
340 IP_VS_DBG(1, "p-schedule: no dest found.\n");
341 kfree(param.pe_data);
342 *ignored = 0;
343 return NULL;
344 }
345
346 if (dst_port == svc->port && svc->port != FTPPORT)
347 dport = dest->port;
348
349 /* Create a template
350 * This adds param.pe_data to the template,
351 * and thus param.pe_data will be destroyed
352 * when the template expires */
353 ct = ip_vs_conn_new(¶m, dest->af, &dest->addr, dport,
354 IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
355 if (ct == NULL) {
356 kfree(param.pe_data);
357 *ignored = -1;
358 return NULL;
359 }
360
361 ct->timeout = svc->timeout;
362 } else {
363 /* set destination with the found template */
364 dest = ct->dest;
365 kfree(param.pe_data);
366 }
367
368 dport = dst_port;
369 if (dport == svc->port && dest->port)
370 dport = dest->port;
371
372 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
373 && iph->protocol == IPPROTO_UDP) ?
374 IP_VS_CONN_F_ONE_PACKET : 0;
375
376 /*
377 * Create a new connection according to the template
378 */
379 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
380 src_port, dst_addr, dst_port, ¶m);
381
382 cp = ip_vs_conn_new(¶m, dest->af, &dest->addr, dport, flags, dest,
383 skb->mark);
384 if (cp == NULL) {
385 ip_vs_conn_put(ct);
386 *ignored = -1;
387 return NULL;
388 }
389
390 /*
391 * Add its control
392 */
393 ip_vs_control_add(cp, ct);
394 ip_vs_conn_put(ct);
395
396 ip_vs_conn_stats(cp, svc);
397 return cp;
398 }
399
400
401 /*
402 * IPVS main scheduling function
403 * It selects a server according to the virtual service, and
404 * creates a connection entry.
405 * Protocols supported: TCP, UDP
406 *
407 * Usage of *ignored
408 *
409 * 1 : protocol tried to schedule (eg. on SYN), found svc but the
410 * svc/scheduler decides that this packet should be accepted with
411 * NF_ACCEPT because it must not be scheduled.
412 *
413 * 0 : scheduler can not find destination, so try bypass or
414 * return ICMP and then NF_DROP (ip_vs_leave).
415 *
416 * -1 : scheduler tried to schedule but fatal error occurred, eg.
417 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
418 * failure such as missing Call-ID, ENOMEM on skb_linearize
419 * or pe_data. In this case we should return NF_DROP without
420 * any attempts to send ICMP with ip_vs_leave.
421 */
422 struct ip_vs_conn *
423 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
424 struct ip_vs_proto_data *pd, int *ignored,
425 struct ip_vs_iphdr *iph)
426 {
427 struct ip_vs_protocol *pp = pd->pp;
428 struct ip_vs_conn *cp = NULL;
429 struct ip_vs_scheduler *sched;
430 struct ip_vs_dest *dest;
431 __be16 _ports[2], *pptr, cport, vport;
432 const void *caddr, *vaddr;
433 unsigned int flags;
434
435 *ignored = 1;
436 /*
437 * IPv6 frags, only the first hit here.
438 */
439 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
440 if (pptr == NULL)
441 return NULL;
442
443 if (likely(!ip_vs_iph_inverse(iph))) {
444 cport = pptr[0];
445 caddr = &iph->saddr;
446 vport = pptr[1];
447 vaddr = &iph->daddr;
448 } else {
449 cport = pptr[1];
450 caddr = &iph->daddr;
451 vport = pptr[0];
452 vaddr = &iph->saddr;
453 }
454
455 /*
456 * FTPDATA needs this check when using local real server.
457 * Never schedule Active FTPDATA connections from real server.
458 * For LVS-NAT they must be already created. For other methods
459 * with persistence the connection is created on SYN+ACK.
460 */
461 if (cport == FTPDATA) {
462 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
463 "Not scheduling FTPDATA");
464 return NULL;
465 }
466
467 /*
468 * Do not schedule replies from local real server.
469 */
470 if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
471 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
472 cp = pp->conn_in_get(svc->ipvs, svc->af, skb, iph);
473 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
474
475 if (cp) {
476 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
477 "Not scheduling reply for existing"
478 " connection");
479 __ip_vs_conn_put(cp);
480 return NULL;
481 }
482 }
483
484 /*
485 * Persistent service
486 */
487 if (svc->flags & IP_VS_SVC_F_PERSISTENT)
488 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
489 iph);
490
491 *ignored = 0;
492
493 /*
494 * Non-persistent service
495 */
496 if (!svc->fwmark && vport != svc->port) {
497 if (!svc->port)
498 pr_err("Schedule: port zero only supported "
499 "in persistent services, "
500 "check your ipvs configuration\n");
501 return NULL;
502 }
503
504 sched = rcu_dereference(svc->scheduler);
505 if (sched) {
506 /* read svc->sched_data after svc->scheduler */
507 smp_rmb();
508 dest = sched->schedule(svc, skb, iph);
509 } else {
510 dest = NULL;
511 }
512 if (dest == NULL) {
513 IP_VS_DBG(1, "Schedule: no dest found.\n");
514 return NULL;
515 }
516
517 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
518 && iph->protocol == IPPROTO_UDP) ?
519 IP_VS_CONN_F_ONE_PACKET : 0;
520
521 /*
522 * Create a connection entry.
523 */
524 {
525 struct ip_vs_conn_param p;
526
527 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
528 caddr, cport, vaddr, vport, &p);
529 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
530 dest->port ? dest->port : vport,
531 flags, dest, skb->mark);
532 if (!cp) {
533 *ignored = -1;
534 return NULL;
535 }
536 }
537
538 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
539 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
540 ip_vs_fwd_tag(cp),
541 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
542 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
543 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
544 cp->flags, atomic_read(&cp->refcnt));
545
546 ip_vs_conn_stats(cp, svc);
547 return cp;
548 }
549
550 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
551 union nf_inet_addr *addr)
552 {
553 #ifdef CONFIG_IP_VS_IPV6
554 if (af == AF_INET6)
555 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
556 #endif
557 return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
558 }
559
560 /*
561 * Pass or drop the packet.
562 * Called by ip_vs_in, when the virtual service is available but
563 * no destination is available for a new connection.
564 */
565 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
566 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
567 {
568 __be16 _ports[2], *pptr, dport;
569 struct netns_ipvs *ipvs = svc->ipvs;
570 struct net *net = ipvs->net;
571
572 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
573 if (!pptr)
574 return NF_DROP;
575 dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
576
577 /* if it is fwmark-based service, the cache_bypass sysctl is up
578 and the destination is a non-local unicast, then create
579 a cache_bypass connection entry */
580 if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
581 !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
582 ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
583 int ret;
584 struct ip_vs_conn *cp;
585 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
586 iph->protocol == IPPROTO_UDP) ?
587 IP_VS_CONN_F_ONE_PACKET : 0;
588 union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } };
589
590 /* create a new connection entry */
591 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
592 {
593 struct ip_vs_conn_param p;
594 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
595 &iph->saddr, pptr[0],
596 &iph->daddr, pptr[1], &p);
597 cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
598 IP_VS_CONN_F_BYPASS | flags,
599 NULL, skb->mark);
600 if (!cp)
601 return NF_DROP;
602 }
603
604 /* statistics */
605 ip_vs_in_stats(cp, skb);
606
607 /* set state */
608 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
609
610 /* transmit the first SYN packet */
611 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
612 /* do not touch skb anymore */
613
614 atomic_inc(&cp->in_pkts);
615 ip_vs_conn_put(cp);
616 return ret;
617 }
618
619 /*
620 * When the virtual ftp service is presented, packets destined
621 * for other services on the VIP may get here (except services
622 * listed in the ipvs table), pass the packets, because it is
623 * not ipvs job to decide to drop the packets.
624 */
625 if (svc->port == FTPPORT && dport != FTPPORT)
626 return NF_ACCEPT;
627
628 if (unlikely(ip_vs_iph_icmp(iph)))
629 return NF_DROP;
630
631 /*
632 * Notify the client that the destination is unreachable, and
633 * release the socket buffer.
634 * Since it is in IP layer, the TCP socket is not actually
635 * created, the TCP RST packet cannot be sent, instead that
636 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
637 */
638 #ifdef CONFIG_IP_VS_IPV6
639 if (svc->af == AF_INET6) {
640 if (!skb->dev)
641 skb->dev = net->loopback_dev;
642 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
643 } else
644 #endif
645 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
646
647 return NF_DROP;
648 }
649
650 #ifdef CONFIG_SYSCTL
651
652 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
653 {
654 return ipvs->sysctl_snat_reroute;
655 }
656
657 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
658 {
659 return ipvs->sysctl_nat_icmp_send;
660 }
661
662 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
663 {
664 return ipvs->sysctl_expire_nodest_conn;
665 }
666
667 #else
668
669 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
670 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
671 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
672
673 #endif
674
675 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
676 {
677 return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
678 }
679
680 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
681 {
682 if (NF_INET_LOCAL_IN == hooknum)
683 return IP_DEFRAG_VS_IN;
684 if (NF_INET_FORWARD == hooknum)
685 return IP_DEFRAG_VS_FWD;
686 return IP_DEFRAG_VS_OUT;
687 }
688
689 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
690 struct sk_buff *skb, u_int32_t user)
691 {
692 int err;
693
694 local_bh_disable();
695 err = ip_defrag(ipvs->net, skb, user);
696 local_bh_enable();
697 if (!err)
698 ip_send_check(ip_hdr(skb));
699
700 return err;
701 }
702
703 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
704 struct sk_buff *skb, unsigned int hooknum)
705 {
706 if (!sysctl_snat_reroute(ipvs))
707 return 0;
708 /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
709 if (NF_INET_LOCAL_IN == hooknum)
710 return 0;
711 #ifdef CONFIG_IP_VS_IPV6
712 if (af == AF_INET6) {
713 struct dst_entry *dst = skb_dst(skb);
714
715 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
716 ip6_route_me_harder(ipvs->net, skb) != 0)
717 return 1;
718 } else
719 #endif
720 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
721 ip_route_me_harder(ipvs->net, skb, RTN_LOCAL) != 0)
722 return 1;
723
724 return 0;
725 }
726
727 /*
728 * Packet has been made sufficiently writable in caller
729 * - inout: 1=in->out, 0=out->in
730 */
731 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
732 struct ip_vs_conn *cp, int inout)
733 {
734 struct iphdr *iph = ip_hdr(skb);
735 unsigned int icmp_offset = iph->ihl*4;
736 struct icmphdr *icmph = (struct icmphdr *)(skb_network_header(skb) +
737 icmp_offset);
738 struct iphdr *ciph = (struct iphdr *)(icmph + 1);
739
740 if (inout) {
741 iph->saddr = cp->vaddr.ip;
742 ip_send_check(iph);
743 ciph->daddr = cp->vaddr.ip;
744 ip_send_check(ciph);
745 } else {
746 iph->daddr = cp->daddr.ip;
747 ip_send_check(iph);
748 ciph->saddr = cp->daddr.ip;
749 ip_send_check(ciph);
750 }
751
752 /* the TCP/UDP/SCTP port */
753 if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
754 IPPROTO_SCTP == ciph->protocol) {
755 __be16 *ports = (void *)ciph + ciph->ihl*4;
756
757 if (inout)
758 ports[1] = cp->vport;
759 else
760 ports[0] = cp->dport;
761 }
762
763 /* And finally the ICMP checksum */
764 icmph->checksum = 0;
765 icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
766 skb->ip_summed = CHECKSUM_UNNECESSARY;
767
768 if (inout)
769 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
770 "Forwarding altered outgoing ICMP");
771 else
772 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
773 "Forwarding altered incoming ICMP");
774 }
775
776 #ifdef CONFIG_IP_VS_IPV6
777 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
778 struct ip_vs_conn *cp, int inout)
779 {
780 struct ipv6hdr *iph = ipv6_hdr(skb);
781 unsigned int icmp_offset = 0;
782 unsigned int offs = 0; /* header offset*/
783 int protocol;
784 struct icmp6hdr *icmph;
785 struct ipv6hdr *ciph;
786 unsigned short fragoffs;
787
788 ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
789 icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
790 offs = icmp_offset + sizeof(struct icmp6hdr);
791 ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
792
793 protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
794
795 if (inout) {
796 iph->saddr = cp->vaddr.in6;
797 ciph->daddr = cp->vaddr.in6;
798 } else {
799 iph->daddr = cp->daddr.in6;
800 ciph->saddr = cp->daddr.in6;
801 }
802
803 /* the TCP/UDP/SCTP port */
804 if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
805 IPPROTO_SCTP == protocol)) {
806 __be16 *ports = (void *)(skb_network_header(skb) + offs);
807
808 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
809 ntohs(inout ? ports[1] : ports[0]),
810 ntohs(inout ? cp->vport : cp->dport));
811 if (inout)
812 ports[1] = cp->vport;
813 else
814 ports[0] = cp->dport;
815 }
816
817 /* And finally the ICMP checksum */
818 icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
819 skb->len - icmp_offset,
820 IPPROTO_ICMPV6, 0);
821 skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
822 skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
823 skb->ip_summed = CHECKSUM_PARTIAL;
824
825 if (inout)
826 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
827 (void *)ciph - (void *)iph,
828 "Forwarding altered outgoing ICMPv6");
829 else
830 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
831 (void *)ciph - (void *)iph,
832 "Forwarding altered incoming ICMPv6");
833 }
834 #endif
835
836 /* Handle relevant response ICMP messages - forward to the right
837 * destination host.
838 */
839 static int handle_response_icmp(int af, struct sk_buff *skb,
840 union nf_inet_addr *snet,
841 __u8 protocol, struct ip_vs_conn *cp,
842 struct ip_vs_protocol *pp,
843 unsigned int offset, unsigned int ihl,
844 unsigned int hooknum)
845 {
846 unsigned int verdict = NF_DROP;
847
848 if (IP_VS_FWD_METHOD(cp) != 0) {
849 pr_err("shouldn't reach here, because the box is on the "
850 "half connection in the tun/dr module.\n");
851 }
852
853 /* Ensure the checksum is correct */
854 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
855 /* Failed checksum! */
856 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
857 IP_VS_DBG_ADDR(af, snet));
858 goto out;
859 }
860
861 if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
862 IPPROTO_SCTP == protocol)
863 offset += 2 * sizeof(__u16);
864 if (!skb_make_writable(skb, offset))
865 goto out;
866
867 #ifdef CONFIG_IP_VS_IPV6
868 if (af == AF_INET6)
869 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
870 else
871 #endif
872 ip_vs_nat_icmp(skb, pp, cp, 1);
873
874 if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
875 goto out;
876
877 /* do the statistics and put it back */
878 ip_vs_out_stats(cp, skb);
879
880 skb->ipvs_property = 1;
881 if (!(cp->flags & IP_VS_CONN_F_NFCT))
882 ip_vs_notrack(skb);
883 else
884 ip_vs_update_conntrack(skb, cp, 0);
885 verdict = NF_ACCEPT;
886
887 out:
888 __ip_vs_conn_put(cp);
889
890 return verdict;
891 }
892
893 /*
894 * Handle ICMP messages in the inside-to-outside direction (outgoing).
895 * Find any that might be relevant, check against existing connections.
896 * Currently handles error types - unreachable, quench, ttl exceeded.
897 */
898 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
899 int *related, unsigned int hooknum)
900 {
901 struct iphdr *iph;
902 struct icmphdr _icmph, *ic;
903 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
904 struct ip_vs_iphdr ciph;
905 struct ip_vs_conn *cp;
906 struct ip_vs_protocol *pp;
907 unsigned int offset, ihl;
908 union nf_inet_addr snet;
909
910 *related = 1;
911
912 /* reassemble IP fragments */
913 if (ip_is_fragment(ip_hdr(skb))) {
914 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
915 return NF_STOLEN;
916 }
917
918 iph = ip_hdr(skb);
919 offset = ihl = iph->ihl * 4;
920 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
921 if (ic == NULL)
922 return NF_DROP;
923
924 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
925 ic->type, ntohs(icmp_id(ic)),
926 &iph->saddr, &iph->daddr);
927
928 /*
929 * Work through seeing if this is for us.
930 * These checks are supposed to be in an order that means easy
931 * things are checked first to speed up processing.... however
932 * this means that some packets will manage to get a long way
933 * down this stack and then be rejected, but that's life.
934 */
935 if ((ic->type != ICMP_DEST_UNREACH) &&
936 (ic->type != ICMP_SOURCE_QUENCH) &&
937 (ic->type != ICMP_TIME_EXCEEDED)) {
938 *related = 0;
939 return NF_ACCEPT;
940 }
941
942 /* Now find the contained IP header */
943 offset += sizeof(_icmph);
944 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
945 if (cih == NULL)
946 return NF_ACCEPT; /* The packet looks wrong, ignore */
947
948 pp = ip_vs_proto_get(cih->protocol);
949 if (!pp)
950 return NF_ACCEPT;
951
952 /* Is the embedded protocol header present? */
953 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
954 pp->dont_defrag))
955 return NF_ACCEPT;
956
957 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
958 "Checking outgoing ICMP for");
959
960 ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
961
962 /* The embedded headers contain source and dest in reverse order */
963 cp = pp->conn_out_get(ipvs, AF_INET, skb, &ciph);
964 if (!cp)
965 return NF_ACCEPT;
966
967 snet.ip = iph->saddr;
968 return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
969 pp, ciph.len, ihl, hooknum);
970 }
971
972 #ifdef CONFIG_IP_VS_IPV6
973 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
974 int *related, unsigned int hooknum,
975 struct ip_vs_iphdr *ipvsh)
976 {
977 struct icmp6hdr _icmph, *ic;
978 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
979 struct ip_vs_conn *cp;
980 struct ip_vs_protocol *pp;
981 union nf_inet_addr snet;
982 unsigned int offset;
983
984 *related = 1;
985 ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
986 if (ic == NULL)
987 return NF_DROP;
988
989 /*
990 * Work through seeing if this is for us.
991 * These checks are supposed to be in an order that means easy
992 * things are checked first to speed up processing.... however
993 * this means that some packets will manage to get a long way
994 * down this stack and then be rejected, but that's life.
995 */
996 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
997 *related = 0;
998 return NF_ACCEPT;
999 }
1000 /* Fragment header that is before ICMP header tells us that:
1001 * it's not an error message since they can't be fragmented.
1002 */
1003 if (ipvsh->flags & IP6_FH_F_FRAG)
1004 return NF_DROP;
1005
1006 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1007 ic->icmp6_type, ntohs(icmpv6_id(ic)),
1008 &ipvsh->saddr, &ipvsh->daddr);
1009
1010 if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1011 true, &ciph))
1012 return NF_ACCEPT; /* The packet looks wrong, ignore */
1013
1014 pp = ip_vs_proto_get(ciph.protocol);
1015 if (!pp)
1016 return NF_ACCEPT;
1017
1018 /* The embedded headers contain source and dest in reverse order */
1019 cp = pp->conn_out_get(ipvs, AF_INET6, skb, &ciph);
1020 if (!cp)
1021 return NF_ACCEPT;
1022
1023 snet.in6 = ciph.saddr.in6;
1024 offset = ciph.len;
1025 return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1026 pp, offset, sizeof(struct ipv6hdr),
1027 hooknum);
1028 }
1029 #endif
1030
1031 /*
1032 * Check if sctp chunc is ABORT chunk
1033 */
1034 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1035 {
1036 sctp_chunkhdr_t *sch, schunk;
1037 sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1038 sizeof(schunk), &schunk);
1039 if (sch == NULL)
1040 return 0;
1041 if (sch->type == SCTP_CID_ABORT)
1042 return 1;
1043 return 0;
1044 }
1045
1046 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1047 {
1048 struct tcphdr _tcph, *th;
1049
1050 th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1051 if (th == NULL)
1052 return 0;
1053 return th->rst;
1054 }
1055
1056 static inline bool is_new_conn(const struct sk_buff *skb,
1057 struct ip_vs_iphdr *iph)
1058 {
1059 switch (iph->protocol) {
1060 case IPPROTO_TCP: {
1061 struct tcphdr _tcph, *th;
1062
1063 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1064 if (th == NULL)
1065 return false;
1066 return th->syn;
1067 }
1068 case IPPROTO_SCTP: {
1069 sctp_chunkhdr_t *sch, schunk;
1070
1071 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1072 sizeof(schunk), &schunk);
1073 if (sch == NULL)
1074 return false;
1075 return sch->type == SCTP_CID_INIT;
1076 }
1077 default:
1078 return false;
1079 }
1080 }
1081
1082 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1083 int conn_reuse_mode)
1084 {
1085 /* Controlled (FTP DATA or persistence)? */
1086 if (cp->control)
1087 return false;
1088
1089 switch (cp->protocol) {
1090 case IPPROTO_TCP:
1091 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1092 ((conn_reuse_mode & 2) &&
1093 (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1094 (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1095 case IPPROTO_SCTP:
1096 return cp->state == IP_VS_SCTP_S_CLOSED;
1097 default:
1098 return false;
1099 }
1100 }
1101
1102 /* Handle response packets: rewrite addresses and send away...
1103 */
1104 static unsigned int
1105 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1106 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1107 unsigned int hooknum)
1108 {
1109 struct ip_vs_protocol *pp = pd->pp;
1110
1111 IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1112
1113 if (!skb_make_writable(skb, iph->len))
1114 goto drop;
1115
1116 /* mangle the packet */
1117 if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1118 goto drop;
1119
1120 #ifdef CONFIG_IP_VS_IPV6
1121 if (af == AF_INET6)
1122 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1123 else
1124 #endif
1125 {
1126 ip_hdr(skb)->saddr = cp->vaddr.ip;
1127 ip_send_check(ip_hdr(skb));
1128 }
1129
1130 /*
1131 * nf_iterate does not expect change in the skb->dst->dev.
1132 * It looks like it is not fatal to enable this code for hooks
1133 * where our handlers are at the end of the chain list and
1134 * when all next handlers use skb->dst->dev and not outdev.
1135 * It will definitely route properly the inout NAT traffic
1136 * when multiple paths are used.
1137 */
1138
1139 /* For policy routing, packets originating from this
1140 * machine itself may be routed differently to packets
1141 * passing through. We want this packet to be routed as
1142 * if it came from this machine itself. So re-compute
1143 * the routing information.
1144 */
1145 if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1146 goto drop;
1147
1148 IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1149
1150 ip_vs_out_stats(cp, skb);
1151 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1152 skb->ipvs_property = 1;
1153 if (!(cp->flags & IP_VS_CONN_F_NFCT))
1154 ip_vs_notrack(skb);
1155 else
1156 ip_vs_update_conntrack(skb, cp, 0);
1157 ip_vs_conn_put(cp);
1158
1159 LeaveFunction(11);
1160 return NF_ACCEPT;
1161
1162 drop:
1163 ip_vs_conn_put(cp);
1164 kfree_skb(skb);
1165 LeaveFunction(11);
1166 return NF_STOLEN;
1167 }
1168
1169 /*
1170 * Check if outgoing packet belongs to the established ip_vs_conn.
1171 */
1172 static unsigned int
1173 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1174 {
1175 struct ip_vs_iphdr iph;
1176 struct ip_vs_protocol *pp;
1177 struct ip_vs_proto_data *pd;
1178 struct ip_vs_conn *cp;
1179 struct sock *sk;
1180
1181 EnterFunction(11);
1182
1183 /* Already marked as IPVS request or reply? */
1184 if (skb->ipvs_property)
1185 return NF_ACCEPT;
1186
1187 sk = skb_to_full_sk(skb);
1188 /* Bad... Do not break raw sockets */
1189 if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1190 af == AF_INET)) {
1191
1192 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1193 return NF_ACCEPT;
1194 }
1195
1196 if (unlikely(!skb_dst(skb)))
1197 return NF_ACCEPT;
1198
1199 if (!ipvs->enable)
1200 return NF_ACCEPT;
1201
1202 ip_vs_fill_iph_skb(af, skb, false, &iph);
1203 #ifdef CONFIG_IP_VS_IPV6
1204 if (af == AF_INET6) {
1205 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1206 int related;
1207 int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1208 hooknum, &iph);
1209
1210 if (related)
1211 return verdict;
1212 }
1213 } else
1214 #endif
1215 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1216 int related;
1217 int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1218
1219 if (related)
1220 return verdict;
1221 }
1222
1223 pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1224 if (unlikely(!pd))
1225 return NF_ACCEPT;
1226 pp = pd->pp;
1227
1228 /* reassemble IP fragments */
1229 #ifdef CONFIG_IP_VS_IPV6
1230 if (af == AF_INET)
1231 #endif
1232 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1233 if (ip_vs_gather_frags(ipvs, skb,
1234 ip_vs_defrag_user(hooknum)))
1235 return NF_STOLEN;
1236
1237 ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1238 }
1239
1240 /*
1241 * Check if the packet belongs to an existing entry
1242 */
1243 cp = pp->conn_out_get(ipvs, af, skb, &iph);
1244
1245 if (likely(cp))
1246 return handle_response(af, skb, pd, cp, &iph, hooknum);
1247 if (sysctl_nat_icmp_send(ipvs) &&
1248 (pp->protocol == IPPROTO_TCP ||
1249 pp->protocol == IPPROTO_UDP ||
1250 pp->protocol == IPPROTO_SCTP)) {
1251 __be16 _ports[2], *pptr;
1252
1253 pptr = frag_safe_skb_hp(skb, iph.len,
1254 sizeof(_ports), _ports, &iph);
1255 if (pptr == NULL)
1256 return NF_ACCEPT; /* Not for me */
1257 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1258 pptr[0])) {
1259 /*
1260 * Notify the real server: there is no
1261 * existing entry if it is not RST
1262 * packet or not TCP packet.
1263 */
1264 if ((iph.protocol != IPPROTO_TCP &&
1265 iph.protocol != IPPROTO_SCTP)
1266 || ((iph.protocol == IPPROTO_TCP
1267 && !is_tcp_reset(skb, iph.len))
1268 || (iph.protocol == IPPROTO_SCTP
1269 && !is_sctp_abort(skb,
1270 iph.len)))) {
1271 #ifdef CONFIG_IP_VS_IPV6
1272 if (af == AF_INET6) {
1273 if (!skb->dev)
1274 skb->dev = ipvs->net->loopback_dev;
1275 icmpv6_send(skb,
1276 ICMPV6_DEST_UNREACH,
1277 ICMPV6_PORT_UNREACH,
1278 0);
1279 } else
1280 #endif
1281 icmp_send(skb,
1282 ICMP_DEST_UNREACH,
1283 ICMP_PORT_UNREACH, 0);
1284 return NF_DROP;
1285 }
1286 }
1287 }
1288 IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1289 "ip_vs_out: packet continues traversal as normal");
1290 return NF_ACCEPT;
1291 }
1292
1293 /*
1294 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1295 * used only for VS/NAT.
1296 * Check if packet is reply for established ip_vs_conn.
1297 */
1298 static unsigned int
1299 ip_vs_reply4(void *priv, struct sk_buff *skb,
1300 const struct nf_hook_state *state)
1301 {
1302 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1303 }
1304
1305 /*
1306 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1307 * Check if packet is reply for established ip_vs_conn.
1308 */
1309 static unsigned int
1310 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1311 const struct nf_hook_state *state)
1312 {
1313 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1314 }
1315
1316 #ifdef CONFIG_IP_VS_IPV6
1317
1318 /*
1319 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1320 * used only for VS/NAT.
1321 * Check if packet is reply for established ip_vs_conn.
1322 */
1323 static unsigned int
1324 ip_vs_reply6(void *priv, struct sk_buff *skb,
1325 const struct nf_hook_state *state)
1326 {
1327 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1328 }
1329
1330 /*
1331 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1332 * Check if packet is reply for established ip_vs_conn.
1333 */
1334 static unsigned int
1335 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1336 const struct nf_hook_state *state)
1337 {
1338 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1339 }
1340
1341 #endif
1342
1343 static unsigned int
1344 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1345 struct ip_vs_proto_data *pd,
1346 int *verdict, struct ip_vs_conn **cpp,
1347 struct ip_vs_iphdr *iph)
1348 {
1349 struct ip_vs_protocol *pp = pd->pp;
1350
1351 if (!iph->fragoffs) {
1352 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1353 * replayed fragment zero will already have created the cp
1354 */
1355
1356 /* Schedule and create new connection entry into cpp */
1357 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1358 return 0;
1359 }
1360
1361 if (unlikely(!*cpp)) {
1362 /* sorry, all this trouble for a no-hit :) */
1363 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1364 "ip_vs_in: packet continues traversal as normal");
1365 if (iph->fragoffs) {
1366 /* Fragment that couldn't be mapped to a conn entry
1367 * is missing module nf_defrag_ipv6
1368 */
1369 IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1370 IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1371 "unhandled fragment");
1372 }
1373 *verdict = NF_ACCEPT;
1374 return 0;
1375 }
1376
1377 return 1;
1378 }
1379
1380 /*
1381 * Handle ICMP messages in the outside-to-inside direction (incoming).
1382 * Find any that might be relevant, check against existing connections,
1383 * forward to the right destination host if relevant.
1384 * Currently handles error types - unreachable, quench, ttl exceeded.
1385 */
1386 static int
1387 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1388 unsigned int hooknum)
1389 {
1390 struct iphdr *iph;
1391 struct icmphdr _icmph, *ic;
1392 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
1393 struct ip_vs_iphdr ciph;
1394 struct ip_vs_conn *cp;
1395 struct ip_vs_protocol *pp;
1396 struct ip_vs_proto_data *pd;
1397 unsigned int offset, offset2, ihl, verdict;
1398 bool ipip, new_cp = false;
1399
1400 *related = 1;
1401
1402 /* reassemble IP fragments */
1403 if (ip_is_fragment(ip_hdr(skb))) {
1404 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1405 return NF_STOLEN;
1406 }
1407
1408 iph = ip_hdr(skb);
1409 offset = ihl = iph->ihl * 4;
1410 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1411 if (ic == NULL)
1412 return NF_DROP;
1413
1414 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1415 ic->type, ntohs(icmp_id(ic)),
1416 &iph->saddr, &iph->daddr);
1417
1418 /*
1419 * Work through seeing if this is for us.
1420 * These checks are supposed to be in an order that means easy
1421 * things are checked first to speed up processing.... however
1422 * this means that some packets will manage to get a long way
1423 * down this stack and then be rejected, but that's life.
1424 */
1425 if ((ic->type != ICMP_DEST_UNREACH) &&
1426 (ic->type != ICMP_SOURCE_QUENCH) &&
1427 (ic->type != ICMP_TIME_EXCEEDED)) {
1428 *related = 0;
1429 return NF_ACCEPT;
1430 }
1431
1432 /* Now find the contained IP header */
1433 offset += sizeof(_icmph);
1434 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1435 if (cih == NULL)
1436 return NF_ACCEPT; /* The packet looks wrong, ignore */
1437
1438 /* Special case for errors for IPIP packets */
1439 ipip = false;
1440 if (cih->protocol == IPPROTO_IPIP) {
1441 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1442 return NF_ACCEPT;
1443 /* Error for our IPIP must arrive at LOCAL_IN */
1444 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1445 return NF_ACCEPT;
1446 offset += cih->ihl * 4;
1447 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1448 if (cih == NULL)
1449 return NF_ACCEPT; /* The packet looks wrong, ignore */
1450 ipip = true;
1451 }
1452
1453 pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1454 if (!pd)
1455 return NF_ACCEPT;
1456 pp = pd->pp;
1457
1458 /* Is the embedded protocol header present? */
1459 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1460 pp->dont_defrag))
1461 return NF_ACCEPT;
1462
1463 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1464 "Checking incoming ICMP for");
1465
1466 offset2 = offset;
1467 ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1468 offset = ciph.len;
1469
1470 /* The embedded headers contain source and dest in reverse order.
1471 * For IPIP this is error for request, not for reply.
1472 */
1473 cp = pp->conn_in_get(ipvs, AF_INET, skb, &ciph);
1474
1475 if (!cp) {
1476 int v;
1477
1478 if (!sysctl_schedule_icmp(ipvs))
1479 return NF_ACCEPT;
1480
1481 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1482 return v;
1483 new_cp = true;
1484 }
1485
1486 verdict = NF_DROP;
1487
1488 /* Ensure the checksum is correct */
1489 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1490 /* Failed checksum! */
1491 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1492 &iph->saddr);
1493 goto out;
1494 }
1495
1496 if (ipip) {
1497 __be32 info = ic->un.gateway;
1498 __u8 type = ic->type;
1499 __u8 code = ic->code;
1500
1501 /* Update the MTU */
1502 if (ic->type == ICMP_DEST_UNREACH &&
1503 ic->code == ICMP_FRAG_NEEDED) {
1504 struct ip_vs_dest *dest = cp->dest;
1505 u32 mtu = ntohs(ic->un.frag.mtu);
1506 __be16 frag_off = cih->frag_off;
1507
1508 /* Strip outer IP and ICMP, go to IPIP header */
1509 if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1510 goto ignore_ipip;
1511 offset2 -= ihl + sizeof(_icmph);
1512 skb_reset_network_header(skb);
1513 IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1514 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1515 ipv4_update_pmtu(skb, ipvs->net,
1516 mtu, 0, 0, 0, 0);
1517 /* Client uses PMTUD? */
1518 if (!(frag_off & htons(IP_DF)))
1519 goto ignore_ipip;
1520 /* Prefer the resulting PMTU */
1521 if (dest) {
1522 struct ip_vs_dest_dst *dest_dst;
1523
1524 rcu_read_lock();
1525 dest_dst = rcu_dereference(dest->dest_dst);
1526 if (dest_dst)
1527 mtu = dst_mtu(dest_dst->dst_cache);
1528 rcu_read_unlock();
1529 }
1530 if (mtu > 68 + sizeof(struct iphdr))
1531 mtu -= sizeof(struct iphdr);
1532 info = htonl(mtu);
1533 }
1534 /* Strip outer IP, ICMP and IPIP, go to IP header of
1535 * original request.
1536 */
1537 if (pskb_pull(skb, offset2) == NULL)
1538 goto ignore_ipip;
1539 skb_reset_network_header(skb);
1540 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1541 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1542 type, code, ntohl(info));
1543 icmp_send(skb, type, code, info);
1544 /* ICMP can be shorter but anyways, account it */
1545 ip_vs_out_stats(cp, skb);
1546
1547 ignore_ipip:
1548 consume_skb(skb);
1549 verdict = NF_STOLEN;
1550 goto out;
1551 }
1552
1553 /* do the statistics and put it back */
1554 ip_vs_in_stats(cp, skb);
1555 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1556 IPPROTO_SCTP == cih->protocol)
1557 offset += 2 * sizeof(__u16);
1558 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1559
1560 out:
1561 if (likely(!new_cp))
1562 __ip_vs_conn_put(cp);
1563 else
1564 ip_vs_conn_put(cp);
1565
1566 return verdict;
1567 }
1568
1569 #ifdef CONFIG_IP_VS_IPV6
1570 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1571 int *related, unsigned int hooknum,
1572 struct ip_vs_iphdr *iph)
1573 {
1574 struct icmp6hdr _icmph, *ic;
1575 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1576 struct ip_vs_conn *cp;
1577 struct ip_vs_protocol *pp;
1578 struct ip_vs_proto_data *pd;
1579 unsigned int offset, verdict;
1580 bool new_cp = false;
1581
1582 *related = 1;
1583
1584 ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1585 if (ic == NULL)
1586 return NF_DROP;
1587
1588 /*
1589 * Work through seeing if this is for us.
1590 * These checks are supposed to be in an order that means easy
1591 * things are checked first to speed up processing.... however
1592 * this means that some packets will manage to get a long way
1593 * down this stack and then be rejected, but that's life.
1594 */
1595 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1596 *related = 0;
1597 return NF_ACCEPT;
1598 }
1599 /* Fragment header that is before ICMP header tells us that:
1600 * it's not an error message since they can't be fragmented.
1601 */
1602 if (iph->flags & IP6_FH_F_FRAG)
1603 return NF_DROP;
1604
1605 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1606 ic->icmp6_type, ntohs(icmpv6_id(ic)),
1607 &iph->saddr, &iph->daddr);
1608
1609 offset = iph->len + sizeof(_icmph);
1610 if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1611 return NF_ACCEPT;
1612
1613 pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1614 if (!pd)
1615 return NF_ACCEPT;
1616 pp = pd->pp;
1617
1618 /* Cannot handle fragmented embedded protocol */
1619 if (ciph.fragoffs)
1620 return NF_ACCEPT;
1621
1622 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1623 "Checking incoming ICMPv6 for");
1624
1625 /* The embedded headers contain source and dest in reverse order
1626 * if not from localhost
1627 */
1628 cp = pp->conn_in_get(ipvs, AF_INET6, skb, &ciph);
1629
1630 if (!cp) {
1631 int v;
1632
1633 if (!sysctl_schedule_icmp(ipvs))
1634 return NF_ACCEPT;
1635
1636 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1637 return v;
1638
1639 new_cp = true;
1640 }
1641
1642 /* VS/TUN, VS/DR and LOCALNODE just let it go */
1643 if ((hooknum == NF_INET_LOCAL_OUT) &&
1644 (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1645 verdict = NF_ACCEPT;
1646 goto out;
1647 }
1648
1649 /* do the statistics and put it back */
1650 ip_vs_in_stats(cp, skb);
1651
1652 /* Need to mangle contained IPv6 header in ICMPv6 packet */
1653 offset = ciph.len;
1654 if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1655 IPPROTO_SCTP == ciph.protocol)
1656 offset += 2 * sizeof(__u16); /* Also mangle ports */
1657
1658 verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1659
1660 out:
1661 if (likely(!new_cp))
1662 __ip_vs_conn_put(cp);
1663 else
1664 ip_vs_conn_put(cp);
1665
1666 return verdict;
1667 }
1668 #endif
1669
1670
1671 /*
1672 * Check if it's for virtual services, look it up,
1673 * and send it on its way...
1674 */
1675 static unsigned int
1676 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1677 {
1678 struct ip_vs_iphdr iph;
1679 struct ip_vs_protocol *pp;
1680 struct ip_vs_proto_data *pd;
1681 struct ip_vs_conn *cp;
1682 int ret, pkts;
1683 int conn_reuse_mode;
1684 struct sock *sk;
1685
1686 /* Already marked as IPVS request or reply? */
1687 if (skb->ipvs_property)
1688 return NF_ACCEPT;
1689
1690 /*
1691 * Big tappo:
1692 * - remote client: only PACKET_HOST
1693 * - route: used for struct net when skb->dev is unset
1694 */
1695 if (unlikely((skb->pkt_type != PACKET_HOST &&
1696 hooknum != NF_INET_LOCAL_OUT) ||
1697 !skb_dst(skb))) {
1698 ip_vs_fill_iph_skb(af, skb, false, &iph);
1699 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1700 " ignored in hook %u\n",
1701 skb->pkt_type, iph.protocol,
1702 IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1703 return NF_ACCEPT;
1704 }
1705 /* ipvs enabled in this netns ? */
1706 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1707 return NF_ACCEPT;
1708
1709 ip_vs_fill_iph_skb(af, skb, false, &iph);
1710
1711 /* Bad... Do not break raw sockets */
1712 sk = skb_to_full_sk(skb);
1713 if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1714 af == AF_INET)) {
1715
1716 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1717 return NF_ACCEPT;
1718 }
1719
1720 #ifdef CONFIG_IP_VS_IPV6
1721 if (af == AF_INET6) {
1722 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1723 int related;
1724 int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
1725 hooknum, &iph);
1726
1727 if (related)
1728 return verdict;
1729 }
1730 } else
1731 #endif
1732 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1733 int related;
1734 int verdict = ip_vs_in_icmp(ipvs, skb, &related,
1735 hooknum);
1736
1737 if (related)
1738 return verdict;
1739 }
1740
1741 /* Protocol supported? */
1742 pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1743 if (unlikely(!pd)) {
1744 /* The only way we'll see this packet again is if it's
1745 * encapsulated, so mark it with ipvs_property=1 so we
1746 * skip it if we're ignoring tunneled packets
1747 */
1748 if (sysctl_ignore_tunneled(ipvs))
1749 skb->ipvs_property = 1;
1750
1751 return NF_ACCEPT;
1752 }
1753 pp = pd->pp;
1754 /*
1755 * Check if the packet belongs to an existing connection entry
1756 */
1757 cp = pp->conn_in_get(ipvs, af, skb, &iph);
1758
1759 conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1760 if (conn_reuse_mode && !iph.fragoffs && is_new_conn(skb, &iph) && cp) {
1761 bool uses_ct = false, resched = false;
1762
1763 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1764 unlikely(!atomic_read(&cp->dest->weight))) {
1765 resched = true;
1766 uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1767 } else if (is_new_conn_expected(cp, conn_reuse_mode)) {
1768 uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1769 if (!atomic_read(&cp->n_control)) {
1770 resched = true;
1771 } else {
1772 /* Do not reschedule controlling connection
1773 * that uses conntrack while it is still
1774 * referenced by controlled connection(s).
1775 */
1776 resched = !uses_ct;
1777 }
1778 }
1779
1780 if (resched) {
1781 if (!atomic_read(&cp->n_control))
1782 ip_vs_conn_expire_now(cp);
1783 __ip_vs_conn_put(cp);
1784 if (uses_ct)
1785 return NF_DROP;
1786 cp = NULL;
1787 }
1788 }
1789
1790 if (unlikely(!cp)) {
1791 int v;
1792
1793 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
1794 return v;
1795 }
1796
1797 IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
1798
1799 /* Check the server status */
1800 if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1801 /* the destination server is not available */
1802
1803 if (sysctl_expire_nodest_conn(ipvs)) {
1804 /* try to expire the connection immediately */
1805 ip_vs_conn_expire_now(cp);
1806 }
1807 /* don't restart its timer, and silently
1808 drop the packet. */
1809 __ip_vs_conn_put(cp);
1810 return NF_DROP;
1811 }
1812
1813 ip_vs_in_stats(cp, skb);
1814 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1815 if (cp->packet_xmit)
1816 ret = cp->packet_xmit(skb, cp, pp, &iph);
1817 /* do not touch skb anymore */
1818 else {
1819 IP_VS_DBG_RL("warning: packet_xmit is null");
1820 ret = NF_ACCEPT;
1821 }
1822
1823 /* Increase its packet counter and check if it is needed
1824 * to be synchronized
1825 *
1826 * Sync connection if it is about to close to
1827 * encorage the standby servers to update the connections timeout
1828 *
1829 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1830 */
1831
1832 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1833 pkts = sysctl_sync_threshold(ipvs);
1834 else
1835 pkts = atomic_add_return(1, &cp->in_pkts);
1836
1837 if (ipvs->sync_state & IP_VS_STATE_MASTER)
1838 ip_vs_sync_conn(ipvs, cp, pkts);
1839
1840 ip_vs_conn_put(cp);
1841 return ret;
1842 }
1843
1844 /*
1845 * AF_INET handler in NF_INET_LOCAL_IN chain
1846 * Schedule and forward packets from remote clients
1847 */
1848 static unsigned int
1849 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
1850 const struct nf_hook_state *state)
1851 {
1852 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
1853 }
1854
1855 /*
1856 * AF_INET handler in NF_INET_LOCAL_OUT chain
1857 * Schedule and forward packets from local clients
1858 */
1859 static unsigned int
1860 ip_vs_local_request4(void *priv, struct sk_buff *skb,
1861 const struct nf_hook_state *state)
1862 {
1863 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
1864 }
1865
1866 #ifdef CONFIG_IP_VS_IPV6
1867
1868 /*
1869 * AF_INET6 handler in NF_INET_LOCAL_IN chain
1870 * Schedule and forward packets from remote clients
1871 */
1872 static unsigned int
1873 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
1874 const struct nf_hook_state *state)
1875 {
1876 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
1877 }
1878
1879 /*
1880 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
1881 * Schedule and forward packets from local clients
1882 */
1883 static unsigned int
1884 ip_vs_local_request6(void *priv, struct sk_buff *skb,
1885 const struct nf_hook_state *state)
1886 {
1887 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
1888 }
1889
1890 #endif
1891
1892
1893 /*
1894 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1895 * related packets destined for 0.0.0.0/0.
1896 * When fwmark-based virtual service is used, such as transparent
1897 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1898 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1899 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1900 * and send them to ip_vs_in_icmp.
1901 */
1902 static unsigned int
1903 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
1904 const struct nf_hook_state *state)
1905 {
1906 int r;
1907 struct netns_ipvs *ipvs = net_ipvs(state->net);
1908
1909 if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1910 return NF_ACCEPT;
1911
1912 /* ipvs enabled in this netns ? */
1913 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1914 return NF_ACCEPT;
1915
1916 return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
1917 }
1918
1919 #ifdef CONFIG_IP_VS_IPV6
1920 static unsigned int
1921 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
1922 const struct nf_hook_state *state)
1923 {
1924 int r;
1925 struct netns_ipvs *ipvs = net_ipvs(state->net);
1926 struct ip_vs_iphdr iphdr;
1927
1928 ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
1929 if (iphdr.protocol != IPPROTO_ICMPV6)
1930 return NF_ACCEPT;
1931
1932 /* ipvs enabled in this netns ? */
1933 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1934 return NF_ACCEPT;
1935
1936 return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
1937 }
1938 #endif
1939
1940
1941 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1942 /* After packet filtering, change source only for VS/NAT */
1943 {
1944 .hook = ip_vs_reply4,
1945 .pf = NFPROTO_IPV4,
1946 .hooknum = NF_INET_LOCAL_IN,
1947 .priority = NF_IP_PRI_NAT_SRC - 2,
1948 },
1949 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1950 * or VS/NAT(change destination), so that filtering rules can be
1951 * applied to IPVS. */
1952 {
1953 .hook = ip_vs_remote_request4,
1954 .pf = NFPROTO_IPV4,
1955 .hooknum = NF_INET_LOCAL_IN,
1956 .priority = NF_IP_PRI_NAT_SRC - 1,
1957 },
1958 /* Before ip_vs_in, change source only for VS/NAT */
1959 {
1960 .hook = ip_vs_local_reply4,
1961 .pf = NFPROTO_IPV4,
1962 .hooknum = NF_INET_LOCAL_OUT,
1963 .priority = NF_IP_PRI_NAT_DST + 1,
1964 },
1965 /* After mangle, schedule and forward local requests */
1966 {
1967 .hook = ip_vs_local_request4,
1968 .pf = NFPROTO_IPV4,
1969 .hooknum = NF_INET_LOCAL_OUT,
1970 .priority = NF_IP_PRI_NAT_DST + 2,
1971 },
1972 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1973 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1974 {
1975 .hook = ip_vs_forward_icmp,
1976 .pf = NFPROTO_IPV4,
1977 .hooknum = NF_INET_FORWARD,
1978 .priority = 99,
1979 },
1980 /* After packet filtering, change source only for VS/NAT */
1981 {
1982 .hook = ip_vs_reply4,
1983 .pf = NFPROTO_IPV4,
1984 .hooknum = NF_INET_FORWARD,
1985 .priority = 100,
1986 },
1987 #ifdef CONFIG_IP_VS_IPV6
1988 /* After packet filtering, change source only for VS/NAT */
1989 {
1990 .hook = ip_vs_reply6,
1991 .pf = NFPROTO_IPV6,
1992 .hooknum = NF_INET_LOCAL_IN,
1993 .priority = NF_IP6_PRI_NAT_SRC - 2,
1994 },
1995 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1996 * or VS/NAT(change destination), so that filtering rules can be
1997 * applied to IPVS. */
1998 {
1999 .hook = ip_vs_remote_request6,
2000 .pf = NFPROTO_IPV6,
2001 .hooknum = NF_INET_LOCAL_IN,
2002 .priority = NF_IP6_PRI_NAT_SRC - 1,
2003 },
2004 /* Before ip_vs_in, change source only for VS/NAT */
2005 {
2006 .hook = ip_vs_local_reply6,
2007 .pf = NFPROTO_IPV6,
2008 .hooknum = NF_INET_LOCAL_OUT,
2009 .priority = NF_IP6_PRI_NAT_DST + 1,
2010 },
2011 /* After mangle, schedule and forward local requests */
2012 {
2013 .hook = ip_vs_local_request6,
2014 .pf = NFPROTO_IPV6,
2015 .hooknum = NF_INET_LOCAL_OUT,
2016 .priority = NF_IP6_PRI_NAT_DST + 2,
2017 },
2018 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2019 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2020 {
2021 .hook = ip_vs_forward_icmp_v6,
2022 .pf = NFPROTO_IPV6,
2023 .hooknum = NF_INET_FORWARD,
2024 .priority = 99,
2025 },
2026 /* After packet filtering, change source only for VS/NAT */
2027 {
2028 .hook = ip_vs_reply6,
2029 .pf = NFPROTO_IPV6,
2030 .hooknum = NF_INET_FORWARD,
2031 .priority = 100,
2032 },
2033 #endif
2034 };
2035 /*
2036 * Initialize IP Virtual Server netns mem.
2037 */
2038 static int __net_init __ip_vs_init(struct net *net)
2039 {
2040 struct netns_ipvs *ipvs;
2041
2042 ipvs = net_generic(net, ip_vs_net_id);
2043 if (ipvs == NULL)
2044 return -ENOMEM;
2045
2046 /* Hold the beast until a service is registerd */
2047 ipvs->enable = 0;
2048 ipvs->net = net;
2049 /* Counters used for creating unique names */
2050 ipvs->gen = atomic_read(&ipvs_netns_cnt);
2051 atomic_inc(&ipvs_netns_cnt);
2052 net->ipvs = ipvs;
2053
2054 if (ip_vs_estimator_net_init(ipvs) < 0)
2055 goto estimator_fail;
2056
2057 if (ip_vs_control_net_init(ipvs) < 0)
2058 goto control_fail;
2059
2060 if (ip_vs_protocol_net_init(ipvs) < 0)
2061 goto protocol_fail;
2062
2063 if (ip_vs_app_net_init(ipvs) < 0)
2064 goto app_fail;
2065
2066 if (ip_vs_conn_net_init(ipvs) < 0)
2067 goto conn_fail;
2068
2069 if (ip_vs_sync_net_init(ipvs) < 0)
2070 goto sync_fail;
2071
2072 printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2073 sizeof(struct netns_ipvs), ipvs->gen);
2074 return 0;
2075 /*
2076 * Error handling
2077 */
2078
2079 sync_fail:
2080 ip_vs_conn_net_cleanup(ipvs);
2081 conn_fail:
2082 ip_vs_app_net_cleanup(ipvs);
2083 app_fail:
2084 ip_vs_protocol_net_cleanup(ipvs);
2085 protocol_fail:
2086 ip_vs_control_net_cleanup(ipvs);
2087 control_fail:
2088 ip_vs_estimator_net_cleanup(ipvs);
2089 estimator_fail:
2090 net->ipvs = NULL;
2091 return -ENOMEM;
2092 }
2093
2094 static void __net_exit __ip_vs_cleanup(struct net *net)
2095 {
2096 struct netns_ipvs *ipvs = net_ipvs(net);
2097
2098 ip_vs_service_net_cleanup(ipvs); /* ip_vs_flush() with locks */
2099 ip_vs_conn_net_cleanup(ipvs);
2100 ip_vs_app_net_cleanup(ipvs);
2101 ip_vs_protocol_net_cleanup(ipvs);
2102 ip_vs_control_net_cleanup(ipvs);
2103 ip_vs_estimator_net_cleanup(ipvs);
2104 IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2105 net->ipvs = NULL;
2106 }
2107
2108 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2109 {
2110 struct netns_ipvs *ipvs = net_ipvs(net);
2111 EnterFunction(2);
2112 ipvs->enable = 0; /* Disable packet reception */
2113 smp_wmb();
2114 ip_vs_sync_net_cleanup(ipvs);
2115 LeaveFunction(2);
2116 }
2117
2118 static struct pernet_operations ipvs_core_ops = {
2119 .init = __ip_vs_init,
2120 .exit = __ip_vs_cleanup,
2121 .id = &ip_vs_net_id,
2122 .size = sizeof(struct netns_ipvs),
2123 };
2124
2125 static struct pernet_operations ipvs_core_dev_ops = {
2126 .exit = __ip_vs_dev_cleanup,
2127 };
2128
2129 /*
2130 * Initialize IP Virtual Server
2131 */
2132 static int __init ip_vs_init(void)
2133 {
2134 int ret;
2135
2136 ret = ip_vs_control_init();
2137 if (ret < 0) {
2138 pr_err("can't setup control.\n");
2139 goto exit;
2140 }
2141
2142 ip_vs_protocol_init();
2143
2144 ret = ip_vs_conn_init();
2145 if (ret < 0) {
2146 pr_err("can't setup connection table.\n");
2147 goto cleanup_protocol;
2148 }
2149
2150 ret = register_pernet_subsys(&ipvs_core_ops); /* Alloc ip_vs struct */
2151 if (ret < 0)
2152 goto cleanup_conn;
2153
2154 ret = register_pernet_device(&ipvs_core_dev_ops);
2155 if (ret < 0)
2156 goto cleanup_sub;
2157
2158 ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2159 if (ret < 0) {
2160 pr_err("can't register hooks.\n");
2161 goto cleanup_dev;
2162 }
2163
2164 ret = ip_vs_register_nl_ioctl();
2165 if (ret < 0) {
2166 pr_err("can't register netlink/ioctl.\n");
2167 goto cleanup_hooks;
2168 }
2169
2170 pr_info("ipvs loaded.\n");
2171
2172 return ret;
2173
2174 cleanup_hooks:
2175 nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2176 cleanup_dev:
2177 unregister_pernet_device(&ipvs_core_dev_ops);
2178 cleanup_sub:
2179 unregister_pernet_subsys(&ipvs_core_ops);
2180 cleanup_conn:
2181 ip_vs_conn_cleanup();
2182 cleanup_protocol:
2183 ip_vs_protocol_cleanup();
2184 ip_vs_control_cleanup();
2185 exit:
2186 return ret;
2187 }
2188
2189 static void __exit ip_vs_cleanup(void)
2190 {
2191 ip_vs_unregister_nl_ioctl();
2192 nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2193 unregister_pernet_device(&ipvs_core_dev_ops);
2194 unregister_pernet_subsys(&ipvs_core_ops); /* free ip_vs struct */
2195 ip_vs_conn_cleanup();
2196 ip_vs_protocol_cleanup();
2197 ip_vs_control_cleanup();
2198 pr_info("ipvs unloaded.\n");
2199 }
2200
2201 module_init(ip_vs_init);
2202 module_exit(ip_vs_cleanup);
2203 MODULE_LICENSE("GPL");
2204
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.