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
8 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
9 * Peter Kese <peter.kese@ijs.si>
10 * Julian Anastasov <ja@ssi.bg>
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.
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
22 * Paul `Rusty' Russell properly handle non-linear skbs
23 * Harald Welte don't use nfcache
27 #define KMSG_COMPONENT "IPVS"
28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
30 #include <linux/module.h>
31 #include <linux/kernel.h>
33 #include <linux/tcp.h>
34 #include <linux/sctp.h>
35 #include <linux/icmp.h>
36 #include <linux/slab.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() */
46 #include <linux/netfilter.h>
47 #include <linux/netfilter_ipv4.h>
49 #ifdef CONFIG_IP_VS_IPV6
51 #include <linux/netfilter_ipv6.h>
52 #include <net/ip6_route.h>
55 #include <net/ip_vs.h>
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);
67 EXPORT_SYMBOL(ip_vs_conn_put);
68 #ifdef CONFIG_IP_VS_DEBUG
69 EXPORT_SYMBOL(ip_vs_get_debug_level);
71 EXPORT_SYMBOL(ip_vs_new_conn_out);
73 static unsigned int ip_vs_net_id __read_mostly;
74 /* netns cnt used for uniqueness */
75 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
77 /* ID used in ICMP lookups */
78 #define icmp_id(icmph) (((icmph)->un).echo.id)
79 #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier)
81 const char *ip_vs_proto_name(unsigned int proto)
96 #ifdef CONFIG_IP_VS_IPV6
101 sprintf(buf, "IP_%u", proto);
106 void ip_vs_init_hash_table(struct list_head *table, int rows)
109 INIT_LIST_HEAD(&table[rows]);
113 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
115 struct ip_vs_dest *dest = cp->dest;
116 struct netns_ipvs *ipvs = cp->ipvs;
118 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
119 struct ip_vs_cpu_stats *s;
120 struct ip_vs_service *svc;
122 s = this_cpu_ptr(dest->stats.cpustats);
123 u64_stats_update_begin(&s->syncp);
125 s->cnt.inbytes += skb->len;
126 u64_stats_update_end(&s->syncp);
129 svc = rcu_dereference(dest->svc);
130 s = this_cpu_ptr(svc->stats.cpustats);
131 u64_stats_update_begin(&s->syncp);
133 s->cnt.inbytes += skb->len;
134 u64_stats_update_end(&s->syncp);
137 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
138 u64_stats_update_begin(&s->syncp);
140 s->cnt.inbytes += skb->len;
141 u64_stats_update_end(&s->syncp);
147 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
149 struct ip_vs_dest *dest = cp->dest;
150 struct netns_ipvs *ipvs = cp->ipvs;
152 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
153 struct ip_vs_cpu_stats *s;
154 struct ip_vs_service *svc;
156 s = this_cpu_ptr(dest->stats.cpustats);
157 u64_stats_update_begin(&s->syncp);
159 s->cnt.outbytes += skb->len;
160 u64_stats_update_end(&s->syncp);
163 svc = rcu_dereference(dest->svc);
164 s = this_cpu_ptr(svc->stats.cpustats);
165 u64_stats_update_begin(&s->syncp);
167 s->cnt.outbytes += skb->len;
168 u64_stats_update_end(&s->syncp);
171 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
172 u64_stats_update_begin(&s->syncp);
174 s->cnt.outbytes += skb->len;
175 u64_stats_update_end(&s->syncp);
181 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
183 struct netns_ipvs *ipvs = svc->ipvs;
184 struct ip_vs_cpu_stats *s;
186 s = this_cpu_ptr(cp->dest->stats.cpustats);
187 u64_stats_update_begin(&s->syncp);
189 u64_stats_update_end(&s->syncp);
191 s = this_cpu_ptr(svc->stats.cpustats);
192 u64_stats_update_begin(&s->syncp);
194 u64_stats_update_end(&s->syncp);
196 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
197 u64_stats_update_begin(&s->syncp);
199 u64_stats_update_end(&s->syncp);
204 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
205 const struct sk_buff *skb,
206 struct ip_vs_proto_data *pd)
208 if (likely(pd->pp->state_transition))
209 pd->pp->state_transition(cp, direction, skb, pd);
213 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
214 struct sk_buff *skb, int protocol,
215 const union nf_inet_addr *caddr, __be16 cport,
216 const union nf_inet_addr *vaddr, __be16 vport,
217 struct ip_vs_conn_param *p)
219 ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
221 p->pe = rcu_dereference(svc->pe);
222 if (p->pe && p->pe->fill_param)
223 return p->pe->fill_param(p, skb);
229 * IPVS persistent scheduling function
230 * It creates a connection entry according to its template if exists,
231 * or selects a server and creates a connection entry plus a template.
232 * Locking: we are svc user (svc->refcnt), so we hold all dests too
233 * Protocols supported: TCP, UDP
235 static struct ip_vs_conn *
236 ip_vs_sched_persist(struct ip_vs_service *svc,
237 struct sk_buff *skb, __be16 src_port, __be16 dst_port,
238 int *ignored, struct ip_vs_iphdr *iph)
240 struct ip_vs_conn *cp = NULL;
241 struct ip_vs_dest *dest;
242 struct ip_vs_conn *ct;
243 __be16 dport = 0; /* destination port to forward */
245 struct ip_vs_conn_param param;
246 const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
247 union nf_inet_addr snet; /* source network of the client,
249 const union nf_inet_addr *src_addr, *dst_addr;
251 if (likely(!ip_vs_iph_inverse(iph))) {
252 src_addr = &iph->saddr;
253 dst_addr = &iph->daddr;
255 src_addr = &iph->daddr;
256 dst_addr = &iph->saddr;
260 /* Mask saddr with the netmask to adjust template granularity */
261 #ifdef CONFIG_IP_VS_IPV6
262 if (svc->af == AF_INET6)
263 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
264 (__force __u32) svc->netmask);
267 snet.ip = src_addr->ip & svc->netmask;
269 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
271 IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
272 IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
273 IP_VS_DBG_ADDR(svc->af, &snet));
276 * As far as we know, FTP is a very complicated network protocol, and
277 * it uses control connection and data connections. For active FTP,
278 * FTP server initialize data connection to the client, its source port
279 * is often 20. For passive FTP, FTP server tells the clients the port
280 * that it passively listens to, and the client issues the data
281 * connection. In the tunneling or direct routing mode, the load
282 * balancer is on the client-to-server half of connection, the port
283 * number is unknown to the load balancer. So, a conn template like
284 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
285 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
286 * is created for other persistent services.
289 int protocol = iph->protocol;
290 const union nf_inet_addr *vaddr = dst_addr;
293 if (dst_port == svc->port) {
295 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
297 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
299 if (svc->port != FTPPORT)
302 /* Note: persistent fwmark-based services and
303 * persistent port zero service are handled here.
305 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
306 * port zero template:
307 * <protocol,caddr,0,vaddr,0,daddr,0>
310 protocol = IPPROTO_IP;
314 /* return *ignored = -1 so NF_DROP can be used */
315 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
316 vaddr, vport, ¶m) < 0) {
322 /* Check if a template already exists */
323 ct = ip_vs_ct_in_get(¶m);
324 if (!ct || !ip_vs_check_template(ct, NULL)) {
325 struct ip_vs_scheduler *sched;
328 * No template found or the dest of the connection
329 * template is not available.
330 * return *ignored=0 i.e. ICMP and NF_DROP
332 sched = rcu_dereference(svc->scheduler);
334 /* read svc->sched_data after svc->scheduler */
336 dest = sched->schedule(svc, skb, iph);
341 IP_VS_DBG(1, "p-schedule: no dest found.\n");
342 kfree(param.pe_data);
347 if (dst_port == svc->port && svc->port != FTPPORT)
351 * This adds param.pe_data to the template,
352 * and thus param.pe_data will be destroyed
353 * when the template expires */
354 ct = ip_vs_conn_new(¶m, dest->af, &dest->addr, dport,
355 IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
357 kfree(param.pe_data);
362 ct->timeout = svc->timeout;
364 /* set destination with the found template */
366 kfree(param.pe_data);
370 if (dport == svc->port && dest->port)
373 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
374 && iph->protocol == IPPROTO_UDP) ?
375 IP_VS_CONN_F_ONE_PACKET : 0;
378 * Create a new connection according to the template
380 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
381 src_port, dst_addr, dst_port, ¶m);
383 cp = ip_vs_conn_new(¶m, dest->af, &dest->addr, dport, flags, dest,
394 ip_vs_control_add(cp, ct);
397 ip_vs_conn_stats(cp, svc);
403 * IPVS main scheduling function
404 * It selects a server according to the virtual service, and
405 * creates a connection entry.
406 * Protocols supported: TCP, UDP
410 * 1 : protocol tried to schedule (eg. on SYN), found svc but the
411 * svc/scheduler decides that this packet should be accepted with
412 * NF_ACCEPT because it must not be scheduled.
414 * 0 : scheduler can not find destination, so try bypass or
415 * return ICMP and then NF_DROP (ip_vs_leave).
417 * -1 : scheduler tried to schedule but fatal error occurred, eg.
418 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
419 * failure such as missing Call-ID, ENOMEM on skb_linearize
420 * or pe_data. In this case we should return NF_DROP without
421 * any attempts to send ICMP with ip_vs_leave.
424 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
425 struct ip_vs_proto_data *pd, int *ignored,
426 struct ip_vs_iphdr *iph)
428 struct ip_vs_protocol *pp = pd->pp;
429 struct ip_vs_conn *cp = NULL;
430 struct ip_vs_scheduler *sched;
431 struct ip_vs_dest *dest;
432 __be16 _ports[2], *pptr, cport, vport;
433 const void *caddr, *vaddr;
438 * IPv6 frags, only the first hit here.
440 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
444 if (likely(!ip_vs_iph_inverse(iph))) {
457 * FTPDATA needs this check when using local real server.
458 * Never schedule Active FTPDATA connections from real server.
459 * For LVS-NAT they must be already created. For other methods
460 * with persistence the connection is created on SYN+ACK.
462 if (cport == FTPDATA) {
463 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
464 "Not scheduling FTPDATA");
469 * Do not schedule replies from local real server.
471 if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
472 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
473 cp = pp->conn_in_get(svc->ipvs, svc->af, skb, iph);
474 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
477 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
478 "Not scheduling reply for existing"
480 __ip_vs_conn_put(cp);
488 if (svc->flags & IP_VS_SVC_F_PERSISTENT)
489 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
495 * Non-persistent service
497 if (!svc->fwmark && vport != svc->port) {
499 pr_err("Schedule: port zero only supported "
500 "in persistent services, "
501 "check your ipvs configuration\n");
505 sched = rcu_dereference(svc->scheduler);
507 /* read svc->sched_data after svc->scheduler */
509 dest = sched->schedule(svc, skb, iph);
514 IP_VS_DBG(1, "Schedule: no dest found.\n");
518 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
519 && iph->protocol == IPPROTO_UDP) ?
520 IP_VS_CONN_F_ONE_PACKET : 0;
523 * Create a connection entry.
526 struct ip_vs_conn_param p;
528 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
529 caddr, cport, vaddr, vport, &p);
530 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
531 dest->port ? dest->port : vport,
532 flags, dest, skb->mark);
539 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
540 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
542 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
543 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
544 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
545 cp->flags, refcount_read(&cp->refcnt));
547 ip_vs_conn_stats(cp, svc);
551 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
552 union nf_inet_addr *addr)
554 #ifdef CONFIG_IP_VS_IPV6
556 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
558 return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
562 * Pass or drop the packet.
563 * Called by ip_vs_in, when the virtual service is available but
564 * no destination is available for a new connection.
566 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
567 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
569 __be16 _ports[2], *pptr, dport;
570 struct netns_ipvs *ipvs = svc->ipvs;
571 struct net *net = ipvs->net;
573 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
576 dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
578 /* if it is fwmark-based service, the cache_bypass sysctl is up
579 and the destination is a non-local unicast, then create
580 a cache_bypass connection entry */
581 if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
582 !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
583 ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
585 struct ip_vs_conn *cp;
586 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
587 iph->protocol == IPPROTO_UDP) ?
588 IP_VS_CONN_F_ONE_PACKET : 0;
589 union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } };
591 /* create a new connection entry */
592 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
594 struct ip_vs_conn_param p;
595 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
596 &iph->saddr, pptr[0],
597 &iph->daddr, pptr[1], &p);
598 cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
599 IP_VS_CONN_F_BYPASS | flags,
606 ip_vs_in_stats(cp, skb);
609 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
611 /* transmit the first SYN packet */
612 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
613 /* do not touch skb anymore */
615 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
616 atomic_inc(&cp->control->in_pkts);
618 atomic_inc(&cp->in_pkts);
624 * When the virtual ftp service is presented, packets destined
625 * for other services on the VIP may get here (except services
626 * listed in the ipvs table), pass the packets, because it is
627 * not ipvs job to decide to drop the packets.
629 if (svc->port == FTPPORT && dport != FTPPORT)
632 if (unlikely(ip_vs_iph_icmp(iph)))
636 * Notify the client that the destination is unreachable, and
637 * release the socket buffer.
638 * Since it is in IP layer, the TCP socket is not actually
639 * created, the TCP RST packet cannot be sent, instead that
640 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
642 #ifdef CONFIG_IP_VS_IPV6
643 if (svc->af == AF_INET6) {
645 skb->dev = net->loopback_dev;
646 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
649 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
656 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
658 return ipvs->sysctl_snat_reroute;
661 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
663 return ipvs->sysctl_nat_icmp_send;
666 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
668 return ipvs->sysctl_expire_nodest_conn;
673 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
674 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
675 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
679 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
681 return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
684 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
686 if (NF_INET_LOCAL_IN == hooknum)
687 return IP_DEFRAG_VS_IN;
688 if (NF_INET_FORWARD == hooknum)
689 return IP_DEFRAG_VS_FWD;
690 return IP_DEFRAG_VS_OUT;
693 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
694 struct sk_buff *skb, u_int32_t user)
699 err = ip_defrag(ipvs->net, skb, user);
702 ip_send_check(ip_hdr(skb));
707 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
708 struct sk_buff *skb, unsigned int hooknum)
710 if (!sysctl_snat_reroute(ipvs))
712 /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
713 if (NF_INET_LOCAL_IN == hooknum)
715 #ifdef CONFIG_IP_VS_IPV6
716 if (af == AF_INET6) {
717 struct dst_entry *dst = skb_dst(skb);
719 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
720 ip6_route_me_harder(ipvs->net, skb) != 0)
724 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
725 ip_route_me_harder(ipvs->net, skb, RTN_LOCAL) != 0)
732 * Packet has been made sufficiently writable in caller
733 * - inout: 1=in->out, 0=out->in
735 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
736 struct ip_vs_conn *cp, int inout)
738 struct iphdr *iph = ip_hdr(skb);
739 unsigned int icmp_offset = iph->ihl*4;
740 struct icmphdr *icmph = (struct icmphdr *)(skb_network_header(skb) +
742 struct iphdr *ciph = (struct iphdr *)(icmph + 1);
745 iph->saddr = cp->vaddr.ip;
747 ciph->daddr = cp->vaddr.ip;
750 iph->daddr = cp->daddr.ip;
752 ciph->saddr = cp->daddr.ip;
756 /* the TCP/UDP/SCTP port */
757 if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
758 IPPROTO_SCTP == ciph->protocol) {
759 __be16 *ports = (void *)ciph + ciph->ihl*4;
762 ports[1] = cp->vport;
764 ports[0] = cp->dport;
767 /* And finally the ICMP checksum */
769 icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
770 skb->ip_summed = CHECKSUM_UNNECESSARY;
773 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
774 "Forwarding altered outgoing ICMP");
776 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
777 "Forwarding altered incoming ICMP");
780 #ifdef CONFIG_IP_VS_IPV6
781 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
782 struct ip_vs_conn *cp, int inout)
784 struct ipv6hdr *iph = ipv6_hdr(skb);
785 unsigned int icmp_offset = 0;
786 unsigned int offs = 0; /* header offset*/
788 struct icmp6hdr *icmph;
789 struct ipv6hdr *ciph;
790 unsigned short fragoffs;
792 ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
793 icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
794 offs = icmp_offset + sizeof(struct icmp6hdr);
795 ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
797 protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
800 iph->saddr = cp->vaddr.in6;
801 ciph->daddr = cp->vaddr.in6;
803 iph->daddr = cp->daddr.in6;
804 ciph->saddr = cp->daddr.in6;
807 /* the TCP/UDP/SCTP port */
808 if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
809 IPPROTO_SCTP == protocol)) {
810 __be16 *ports = (void *)(skb_network_header(skb) + offs);
812 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
813 ntohs(inout ? ports[1] : ports[0]),
814 ntohs(inout ? cp->vport : cp->dport));
816 ports[1] = cp->vport;
818 ports[0] = cp->dport;
821 /* And finally the ICMP checksum */
822 icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
823 skb->len - icmp_offset,
825 skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
826 skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
827 skb->ip_summed = CHECKSUM_PARTIAL;
830 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
831 (void *)ciph - (void *)iph,
832 "Forwarding altered outgoing ICMPv6");
834 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
835 (void *)ciph - (void *)iph,
836 "Forwarding altered incoming ICMPv6");
840 /* Handle relevant response ICMP messages - forward to the right
843 static int handle_response_icmp(int af, struct sk_buff *skb,
844 union nf_inet_addr *snet,
845 __u8 protocol, struct ip_vs_conn *cp,
846 struct ip_vs_protocol *pp,
847 unsigned int offset, unsigned int ihl,
848 unsigned int hooknum)
850 unsigned int verdict = NF_DROP;
852 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
855 /* Ensure the checksum is correct */
856 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
857 /* Failed checksum! */
858 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
859 IP_VS_DBG_ADDR(af, snet));
863 if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
864 IPPROTO_SCTP == protocol)
865 offset += 2 * sizeof(__u16);
866 if (!skb_make_writable(skb, offset))
869 #ifdef CONFIG_IP_VS_IPV6
871 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
874 ip_vs_nat_icmp(skb, pp, cp, 1);
876 if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
879 /* do the statistics and put it back */
880 ip_vs_out_stats(cp, skb);
882 skb->ipvs_property = 1;
883 if (!(cp->flags & IP_VS_CONN_F_NFCT))
886 ip_vs_update_conntrack(skb, cp, 0);
892 __ip_vs_conn_put(cp);
898 * Handle ICMP messages in the inside-to-outside direction (outgoing).
899 * Find any that might be relevant, check against existing connections.
900 * Currently handles error types - unreachable, quench, ttl exceeded.
902 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
903 int *related, unsigned int hooknum)
906 struct icmphdr _icmph, *ic;
907 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
908 struct ip_vs_iphdr ciph;
909 struct ip_vs_conn *cp;
910 struct ip_vs_protocol *pp;
911 unsigned int offset, ihl;
912 union nf_inet_addr snet;
916 /* reassemble IP fragments */
917 if (ip_is_fragment(ip_hdr(skb))) {
918 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
923 offset = ihl = iph->ihl * 4;
924 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
928 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
929 ic->type, ntohs(icmp_id(ic)),
930 &iph->saddr, &iph->daddr);
933 * Work through seeing if this is for us.
934 * These checks are supposed to be in an order that means easy
935 * things are checked first to speed up processing.... however
936 * this means that some packets will manage to get a long way
937 * down this stack and then be rejected, but that's life.
939 if ((ic->type != ICMP_DEST_UNREACH) &&
940 (ic->type != ICMP_SOURCE_QUENCH) &&
941 (ic->type != ICMP_TIME_EXCEEDED)) {
946 /* Now find the contained IP header */
947 offset += sizeof(_icmph);
948 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
950 return NF_ACCEPT; /* The packet looks wrong, ignore */
952 pp = ip_vs_proto_get(cih->protocol);
956 /* Is the embedded protocol header present? */
957 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
961 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
962 "Checking outgoing ICMP for");
964 ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
966 /* The embedded headers contain source and dest in reverse order */
967 cp = pp->conn_out_get(ipvs, AF_INET, skb, &ciph);
971 snet.ip = iph->saddr;
972 return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
973 pp, ciph.len, ihl, hooknum);
976 #ifdef CONFIG_IP_VS_IPV6
977 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
978 int *related, unsigned int hooknum,
979 struct ip_vs_iphdr *ipvsh)
981 struct icmp6hdr _icmph, *ic;
982 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
983 struct ip_vs_conn *cp;
984 struct ip_vs_protocol *pp;
985 union nf_inet_addr snet;
989 ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
994 * Work through seeing if this is for us.
995 * These checks are supposed to be in an order that means easy
996 * things are checked first to speed up processing.... however
997 * this means that some packets will manage to get a long way
998 * down this stack and then be rejected, but that's life.
1000 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1004 /* Fragment header that is before ICMP header tells us that:
1005 * it's not an error message since they can't be fragmented.
1007 if (ipvsh->flags & IP6_FH_F_FRAG)
1010 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1011 ic->icmp6_type, ntohs(icmpv6_id(ic)),
1012 &ipvsh->saddr, &ipvsh->daddr);
1014 if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1016 return NF_ACCEPT; /* The packet looks wrong, ignore */
1018 pp = ip_vs_proto_get(ciph.protocol);
1022 /* The embedded headers contain source and dest in reverse order */
1023 cp = pp->conn_out_get(ipvs, AF_INET6, skb, &ciph);
1027 snet.in6 = ciph.saddr.in6;
1029 return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1030 pp, offset, sizeof(struct ipv6hdr),
1036 * Check if sctp chunc is ABORT chunk
1038 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1040 sctp_chunkhdr_t *sch, schunk;
1041 sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1042 sizeof(schunk), &schunk);
1045 if (sch->type == SCTP_CID_ABORT)
1050 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1052 struct tcphdr _tcph, *th;
1054 th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1060 static inline bool is_new_conn(const struct sk_buff *skb,
1061 struct ip_vs_iphdr *iph)
1063 switch (iph->protocol) {
1065 struct tcphdr _tcph, *th;
1067 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1072 case IPPROTO_SCTP: {
1073 sctp_chunkhdr_t *sch, schunk;
1075 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1076 sizeof(schunk), &schunk);
1079 return sch->type == SCTP_CID_INIT;
1086 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1087 int conn_reuse_mode)
1089 /* Controlled (FTP DATA or persistence)? */
1093 switch (cp->protocol) {
1095 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1096 (cp->state == IP_VS_TCP_S_CLOSE) ||
1097 ((conn_reuse_mode & 2) &&
1098 (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1099 (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1101 return cp->state == IP_VS_SCTP_S_CLOSED;
1107 /* Generic function to create new connections for outgoing RS packets
1109 * Pre-requisites for successful connection creation:
1110 * 1) Virtual Service is NOT fwmark based:
1111 * In fwmark-VS actual vaddr and vport are unknown to IPVS
1112 * 2) Real Server and Virtual Service were NOT configured without port:
1113 * This is to allow match of different VS to the same RS ip-addr
1115 struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc,
1116 struct ip_vs_dest *dest,
1117 struct sk_buff *skb,
1118 const struct ip_vs_iphdr *iph,
1122 struct ip_vs_conn_param param;
1123 struct ip_vs_conn *ct = NULL, *cp = NULL;
1124 const union nf_inet_addr *vaddr, *daddr, *caddr;
1125 union nf_inet_addr snet;
1132 daddr = &iph->saddr;
1133 caddr = &iph->daddr;
1135 /* check pre-requisites are satisfied */
1138 if (!vport || !dport)
1141 /* for persistent service first create connection template */
1142 if (svc->flags & IP_VS_SVC_F_PERSISTENT) {
1143 /* apply netmask the same way ingress-side does */
1144 #ifdef CONFIG_IP_VS_IPV6
1145 if (svc->af == AF_INET6)
1146 ipv6_addr_prefix(&snet.in6, &caddr->in6,
1147 (__force __u32)svc->netmask);
1150 snet.ip = caddr->ip & svc->netmask;
1151 /* fill params and create template if not existent */
1152 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol,
1156 ct = ip_vs_ct_in_get(¶m);
1157 /* check if template exists and points to the same dest */
1158 if (!ct || !ip_vs_check_template(ct, dest)) {
1159 ct = ip_vs_conn_new(¶m, dest->af, daddr, dport,
1160 IP_VS_CONN_F_TEMPLATE, dest, 0);
1162 kfree(param.pe_data);
1165 ct->timeout = svc->timeout;
1167 kfree(param.pe_data);
1171 /* connection flags */
1172 flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) &&
1173 iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0;
1174 /* create connection */
1175 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
1176 caddr, cport, vaddr, vport, ¶m);
1177 cp = ip_vs_conn_new(¶m, dest->af, daddr, dport, flags, dest, 0);
1184 ip_vs_control_add(cp, ct);
1187 ip_vs_conn_stats(cp, svc);
1189 /* return connection (will be used to handle outgoing packet) */
1190 IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u "
1191 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
1193 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
1194 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
1195 IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
1196 cp->flags, refcount_read(&cp->refcnt));
1201 /* Handle outgoing packets which are considered requests initiated by
1202 * real servers, so that subsequent responses from external client can be
1203 * routed to the right real server.
1204 * Used also for outgoing responses in OPS mode.
1206 * Connection management is handled by persistent-engine specific callback.
1208 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum,
1209 struct netns_ipvs *ipvs,
1210 int af, struct sk_buff *skb,
1211 const struct ip_vs_iphdr *iph)
1213 struct ip_vs_dest *dest;
1214 struct ip_vs_conn *cp = NULL;
1215 __be16 _ports[2], *pptr;
1217 if (hooknum == NF_INET_LOCAL_IN)
1220 pptr = frag_safe_skb_hp(skb, iph->len,
1221 sizeof(_ports), _ports, iph);
1226 dest = ip_vs_find_real_service(ipvs, af, iph->protocol,
1227 &iph->saddr, pptr[0]);
1229 struct ip_vs_service *svc;
1230 struct ip_vs_pe *pe;
1232 svc = rcu_dereference(dest->svc);
1234 pe = rcu_dereference(svc->pe);
1235 if (pe && pe->conn_out)
1236 cp = pe->conn_out(svc, dest, skb, iph,
1245 /* Handle response packets: rewrite addresses and send away...
1248 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1249 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1250 unsigned int hooknum)
1252 struct ip_vs_protocol *pp = pd->pp;
1254 IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1256 if (!skb_make_writable(skb, iph->len))
1259 /* mangle the packet */
1260 if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1263 #ifdef CONFIG_IP_VS_IPV6
1265 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1269 ip_hdr(skb)->saddr = cp->vaddr.ip;
1270 ip_send_check(ip_hdr(skb));
1274 * nf_iterate does not expect change in the skb->dst->dev.
1275 * It looks like it is not fatal to enable this code for hooks
1276 * where our handlers are at the end of the chain list and
1277 * when all next handlers use skb->dst->dev and not outdev.
1278 * It will definitely route properly the inout NAT traffic
1279 * when multiple paths are used.
1282 /* For policy routing, packets originating from this
1283 * machine itself may be routed differently to packets
1284 * passing through. We want this packet to be routed as
1285 * if it came from this machine itself. So re-compute
1286 * the routing information.
1288 if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1291 IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1293 ip_vs_out_stats(cp, skb);
1294 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1295 skb->ipvs_property = 1;
1296 if (!(cp->flags & IP_VS_CONN_F_NFCT))
1299 ip_vs_update_conntrack(skb, cp, 0);
1313 * Check if outgoing packet belongs to the established ip_vs_conn.
1316 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1318 struct ip_vs_iphdr iph;
1319 struct ip_vs_protocol *pp;
1320 struct ip_vs_proto_data *pd;
1321 struct ip_vs_conn *cp;
1326 /* Already marked as IPVS request or reply? */
1327 if (skb->ipvs_property)
1330 sk = skb_to_full_sk(skb);
1331 /* Bad... Do not break raw sockets */
1332 if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1335 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1339 if (unlikely(!skb_dst(skb)))
1345 ip_vs_fill_iph_skb(af, skb, false, &iph);
1346 #ifdef CONFIG_IP_VS_IPV6
1347 if (af == AF_INET6) {
1348 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1350 int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1358 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1360 int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1366 pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1371 /* reassemble IP fragments */
1372 #ifdef CONFIG_IP_VS_IPV6
1375 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1376 if (ip_vs_gather_frags(ipvs, skb,
1377 ip_vs_defrag_user(hooknum)))
1380 ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1384 * Check if the packet belongs to an existing entry
1386 cp = pp->conn_out_get(ipvs, af, skb, &iph);
1389 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
1391 return handle_response(af, skb, pd, cp, &iph, hooknum);
1394 /* Check for real-server-started requests */
1395 if (atomic_read(&ipvs->conn_out_counter)) {
1396 /* Currently only for UDP:
1397 * connection oriented protocols typically use
1398 * ephemeral ports for outgoing connections, so
1399 * related incoming responses would not match any VS
1401 if (pp->protocol == IPPROTO_UDP) {
1402 cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph);
1404 return handle_response(af, skb, pd, cp, &iph,
1409 if (sysctl_nat_icmp_send(ipvs) &&
1410 (pp->protocol == IPPROTO_TCP ||
1411 pp->protocol == IPPROTO_UDP ||
1412 pp->protocol == IPPROTO_SCTP)) {
1413 __be16 _ports[2], *pptr;
1415 pptr = frag_safe_skb_hp(skb, iph.len,
1416 sizeof(_ports), _ports, &iph);
1418 return NF_ACCEPT; /* Not for me */
1419 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1422 * Notify the real server: there is no
1423 * existing entry if it is not RST
1424 * packet or not TCP packet.
1426 if ((iph.protocol != IPPROTO_TCP &&
1427 iph.protocol != IPPROTO_SCTP)
1428 || ((iph.protocol == IPPROTO_TCP
1429 && !is_tcp_reset(skb, iph.len))
1430 || (iph.protocol == IPPROTO_SCTP
1431 && !is_sctp_abort(skb,
1433 #ifdef CONFIG_IP_VS_IPV6
1434 if (af == AF_INET6) {
1436 skb->dev = ipvs->net->loopback_dev;
1438 ICMPV6_DEST_UNREACH,
1439 ICMPV6_PORT_UNREACH,
1445 ICMP_PORT_UNREACH, 0);
1452 IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1453 "ip_vs_out: packet continues traversal as normal");
1457 __ip_vs_conn_put(cp);
1462 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1463 * used only for VS/NAT.
1464 * Check if packet is reply for established ip_vs_conn.
1467 ip_vs_reply4(void *priv, struct sk_buff *skb,
1468 const struct nf_hook_state *state)
1470 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1474 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1475 * Check if packet is reply for established ip_vs_conn.
1478 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1479 const struct nf_hook_state *state)
1481 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1484 #ifdef CONFIG_IP_VS_IPV6
1487 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1488 * used only for VS/NAT.
1489 * Check if packet is reply for established ip_vs_conn.
1492 ip_vs_reply6(void *priv, struct sk_buff *skb,
1493 const struct nf_hook_state *state)
1495 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1499 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1500 * Check if packet is reply for established ip_vs_conn.
1503 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1504 const struct nf_hook_state *state)
1506 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1512 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1513 struct ip_vs_proto_data *pd,
1514 int *verdict, struct ip_vs_conn **cpp,
1515 struct ip_vs_iphdr *iph)
1517 struct ip_vs_protocol *pp = pd->pp;
1519 if (!iph->fragoffs) {
1520 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1521 * replayed fragment zero will already have created the cp
1524 /* Schedule and create new connection entry into cpp */
1525 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1529 if (unlikely(!*cpp)) {
1530 /* sorry, all this trouble for a no-hit :) */
1531 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1532 "ip_vs_in: packet continues traversal as normal");
1533 if (iph->fragoffs) {
1534 /* Fragment that couldn't be mapped to a conn entry
1535 * is missing module nf_defrag_ipv6
1537 IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1538 IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1539 "unhandled fragment");
1541 *verdict = NF_ACCEPT;
1549 * Handle ICMP messages in the outside-to-inside direction (incoming).
1550 * Find any that might be relevant, check against existing connections,
1551 * forward to the right destination host if relevant.
1552 * Currently handles error types - unreachable, quench, ttl exceeded.
1555 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1556 unsigned int hooknum)
1559 struct icmphdr _icmph, *ic;
1560 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
1561 struct ip_vs_iphdr ciph;
1562 struct ip_vs_conn *cp;
1563 struct ip_vs_protocol *pp;
1564 struct ip_vs_proto_data *pd;
1565 unsigned int offset, offset2, ihl, verdict;
1566 bool ipip, new_cp = false;
1570 /* reassemble IP fragments */
1571 if (ip_is_fragment(ip_hdr(skb))) {
1572 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1577 offset = ihl = iph->ihl * 4;
1578 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1582 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1583 ic->type, ntohs(icmp_id(ic)),
1584 &iph->saddr, &iph->daddr);
1587 * Work through seeing if this is for us.
1588 * These checks are supposed to be in an order that means easy
1589 * things are checked first to speed up processing.... however
1590 * this means that some packets will manage to get a long way
1591 * down this stack and then be rejected, but that's life.
1593 if ((ic->type != ICMP_DEST_UNREACH) &&
1594 (ic->type != ICMP_SOURCE_QUENCH) &&
1595 (ic->type != ICMP_TIME_EXCEEDED)) {
1600 /* Now find the contained IP header */
1601 offset += sizeof(_icmph);
1602 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1604 return NF_ACCEPT; /* The packet looks wrong, ignore */
1606 /* Special case for errors for IPIP packets */
1608 if (cih->protocol == IPPROTO_IPIP) {
1609 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1611 /* Error for our IPIP must arrive at LOCAL_IN */
1612 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1614 offset += cih->ihl * 4;
1615 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1617 return NF_ACCEPT; /* The packet looks wrong, ignore */
1621 pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1626 /* Is the embedded protocol header present? */
1627 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1631 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1632 "Checking incoming ICMP for");
1635 ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1638 /* The embedded headers contain source and dest in reverse order.
1639 * For IPIP this is error for request, not for reply.
1641 cp = pp->conn_in_get(ipvs, AF_INET, skb, &ciph);
1646 if (!sysctl_schedule_icmp(ipvs))
1649 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1656 /* Ensure the checksum is correct */
1657 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1658 /* Failed checksum! */
1659 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1665 __be32 info = ic->un.gateway;
1666 __u8 type = ic->type;
1667 __u8 code = ic->code;
1669 /* Update the MTU */
1670 if (ic->type == ICMP_DEST_UNREACH &&
1671 ic->code == ICMP_FRAG_NEEDED) {
1672 struct ip_vs_dest *dest = cp->dest;
1673 u32 mtu = ntohs(ic->un.frag.mtu);
1674 __be16 frag_off = cih->frag_off;
1676 /* Strip outer IP and ICMP, go to IPIP header */
1677 if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1679 offset2 -= ihl + sizeof(_icmph);
1680 skb_reset_network_header(skb);
1681 IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1682 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1683 ipv4_update_pmtu(skb, ipvs->net,
1685 /* Client uses PMTUD? */
1686 if (!(frag_off & htons(IP_DF)))
1688 /* Prefer the resulting PMTU */
1690 struct ip_vs_dest_dst *dest_dst;
1693 dest_dst = rcu_dereference(dest->dest_dst);
1695 mtu = dst_mtu(dest_dst->dst_cache);
1698 if (mtu > 68 + sizeof(struct iphdr))
1699 mtu -= sizeof(struct iphdr);
1702 /* Strip outer IP, ICMP and IPIP, go to IP header of
1705 if (pskb_pull(skb, offset2) == NULL)
1707 skb_reset_network_header(skb);
1708 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1709 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1710 type, code, ntohl(info));
1711 icmp_send(skb, type, code, info);
1712 /* ICMP can be shorter but anyways, account it */
1713 ip_vs_out_stats(cp, skb);
1717 verdict = NF_STOLEN;
1721 /* do the statistics and put it back */
1722 ip_vs_in_stats(cp, skb);
1723 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1724 IPPROTO_SCTP == cih->protocol)
1725 offset += 2 * sizeof(__u16);
1726 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1729 if (likely(!new_cp))
1730 __ip_vs_conn_put(cp);
1737 #ifdef CONFIG_IP_VS_IPV6
1738 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1739 int *related, unsigned int hooknum,
1740 struct ip_vs_iphdr *iph)
1742 struct icmp6hdr _icmph, *ic;
1743 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1744 struct ip_vs_conn *cp;
1745 struct ip_vs_protocol *pp;
1746 struct ip_vs_proto_data *pd;
1747 unsigned int offset, verdict;
1748 bool new_cp = false;
1752 ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1757 * Work through seeing if this is for us.
1758 * These checks are supposed to be in an order that means easy
1759 * things are checked first to speed up processing.... however
1760 * this means that some packets will manage to get a long way
1761 * down this stack and then be rejected, but that's life.
1763 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1767 /* Fragment header that is before ICMP header tells us that:
1768 * it's not an error message since they can't be fragmented.
1770 if (iph->flags & IP6_FH_F_FRAG)
1773 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1774 ic->icmp6_type, ntohs(icmpv6_id(ic)),
1775 &iph->saddr, &iph->daddr);
1777 offset = iph->len + sizeof(_icmph);
1778 if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1781 pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1786 /* Cannot handle fragmented embedded protocol */
1790 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1791 "Checking incoming ICMPv6 for");
1793 /* The embedded headers contain source and dest in reverse order
1794 * if not from localhost
1796 cp = pp->conn_in_get(ipvs, AF_INET6, skb, &ciph);
1801 if (!sysctl_schedule_icmp(ipvs))
1804 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1810 /* VS/TUN, VS/DR and LOCALNODE just let it go */
1811 if ((hooknum == NF_INET_LOCAL_OUT) &&
1812 (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1813 verdict = NF_ACCEPT;
1817 /* do the statistics and put it back */
1818 ip_vs_in_stats(cp, skb);
1820 /* Need to mangle contained IPv6 header in ICMPv6 packet */
1822 if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1823 IPPROTO_SCTP == ciph.protocol)
1824 offset += 2 * sizeof(__u16); /* Also mangle ports */
1826 verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1829 if (likely(!new_cp))
1830 __ip_vs_conn_put(cp);
1840 * Check if it's for virtual services, look it up,
1841 * and send it on its way...
1844 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1846 struct ip_vs_iphdr iph;
1847 struct ip_vs_protocol *pp;
1848 struct ip_vs_proto_data *pd;
1849 struct ip_vs_conn *cp;
1851 int conn_reuse_mode;
1854 /* Already marked as IPVS request or reply? */
1855 if (skb->ipvs_property)
1860 * - remote client: only PACKET_HOST
1861 * - route: used for struct net when skb->dev is unset
1863 if (unlikely((skb->pkt_type != PACKET_HOST &&
1864 hooknum != NF_INET_LOCAL_OUT) ||
1866 ip_vs_fill_iph_skb(af, skb, false, &iph);
1867 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1868 " ignored in hook %u\n",
1869 skb->pkt_type, iph.protocol,
1870 IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1873 /* ipvs enabled in this netns ? */
1874 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1877 ip_vs_fill_iph_skb(af, skb, false, &iph);
1879 /* Bad... Do not break raw sockets */
1880 sk = skb_to_full_sk(skb);
1881 if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1884 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1888 #ifdef CONFIG_IP_VS_IPV6
1889 if (af == AF_INET6) {
1890 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1892 int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
1900 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1902 int verdict = ip_vs_in_icmp(ipvs, skb, &related,
1909 /* Protocol supported? */
1910 pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1911 if (unlikely(!pd)) {
1912 /* The only way we'll see this packet again is if it's
1913 * encapsulated, so mark it with ipvs_property=1 so we
1914 * skip it if we're ignoring tunneled packets
1916 if (sysctl_ignore_tunneled(ipvs))
1917 skb->ipvs_property = 1;
1923 * Check if the packet belongs to an existing connection entry
1925 cp = pp->conn_in_get(ipvs, af, skb, &iph);
1927 conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1928 if (conn_reuse_mode && !iph.fragoffs && is_new_conn(skb, &iph) && cp) {
1929 bool uses_ct = false, resched = false;
1931 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1932 unlikely(!atomic_read(&cp->dest->weight))) {
1934 uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1935 } else if (is_new_conn_expected(cp, conn_reuse_mode)) {
1936 uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1937 if (!atomic_read(&cp->n_control)) {
1940 /* Do not reschedule controlling connection
1941 * that uses conntrack while it is still
1942 * referenced by controlled connection(s).
1949 if (!atomic_read(&cp->n_control))
1950 ip_vs_conn_expire_now(cp);
1951 __ip_vs_conn_put(cp);
1958 if (unlikely(!cp)) {
1961 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
1965 IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
1967 /* Check the server status */
1968 if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1969 /* the destination server is not available */
1971 if (sysctl_expire_nodest_conn(ipvs)) {
1972 /* try to expire the connection immediately */
1973 ip_vs_conn_expire_now(cp);
1975 /* don't restart its timer, and silently
1977 __ip_vs_conn_put(cp);
1981 ip_vs_in_stats(cp, skb);
1982 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1983 if (cp->packet_xmit)
1984 ret = cp->packet_xmit(skb, cp, pp, &iph);
1985 /* do not touch skb anymore */
1987 IP_VS_DBG_RL("warning: packet_xmit is null");
1991 /* Increase its packet counter and check if it is needed
1992 * to be synchronized
1994 * Sync connection if it is about to close to
1995 * encorage the standby servers to update the connections timeout
1997 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
2000 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
2001 pkts = sysctl_sync_threshold(ipvs);
2003 pkts = atomic_add_return(1, &cp->in_pkts);
2005 if (ipvs->sync_state & IP_VS_STATE_MASTER)
2006 ip_vs_sync_conn(ipvs, cp, pkts);
2007 else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
2008 /* increment is done inside ip_vs_sync_conn too */
2009 atomic_inc(&cp->control->in_pkts);
2016 * AF_INET handler in NF_INET_LOCAL_IN chain
2017 * Schedule and forward packets from remote clients
2020 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
2021 const struct nf_hook_state *state)
2023 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2027 * AF_INET handler in NF_INET_LOCAL_OUT chain
2028 * Schedule and forward packets from local clients
2031 ip_vs_local_request4(void *priv, struct sk_buff *skb,
2032 const struct nf_hook_state *state)
2034 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2037 #ifdef CONFIG_IP_VS_IPV6
2040 * AF_INET6 handler in NF_INET_LOCAL_IN chain
2041 * Schedule and forward packets from remote clients
2044 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
2045 const struct nf_hook_state *state)
2047 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2051 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
2052 * Schedule and forward packets from local clients
2055 ip_vs_local_request6(void *priv, struct sk_buff *skb,
2056 const struct nf_hook_state *state)
2058 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2065 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
2066 * related packets destined for 0.0.0.0/0.
2067 * When fwmark-based virtual service is used, such as transparent
2068 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
2069 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
2070 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
2071 * and send them to ip_vs_in_icmp.
2074 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
2075 const struct nf_hook_state *state)
2078 struct netns_ipvs *ipvs = net_ipvs(state->net);
2080 if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
2083 /* ipvs enabled in this netns ? */
2084 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2087 return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
2090 #ifdef CONFIG_IP_VS_IPV6
2092 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
2093 const struct nf_hook_state *state)
2096 struct netns_ipvs *ipvs = net_ipvs(state->net);
2097 struct ip_vs_iphdr iphdr;
2099 ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
2100 if (iphdr.protocol != IPPROTO_ICMPV6)
2103 /* ipvs enabled in this netns ? */
2104 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2107 return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
2112 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
2113 /* After packet filtering, change source only for VS/NAT */
2115 .hook = ip_vs_reply4,
2117 .hooknum = NF_INET_LOCAL_IN,
2118 .priority = NF_IP_PRI_NAT_SRC - 2,
2120 /* After packet filtering, forward packet through VS/DR, VS/TUN,
2121 * or VS/NAT(change destination), so that filtering rules can be
2122 * applied to IPVS. */
2124 .hook = ip_vs_remote_request4,
2126 .hooknum = NF_INET_LOCAL_IN,
2127 .priority = NF_IP_PRI_NAT_SRC - 1,
2129 /* Before ip_vs_in, change source only for VS/NAT */
2131 .hook = ip_vs_local_reply4,
2133 .hooknum = NF_INET_LOCAL_OUT,
2134 .priority = NF_IP_PRI_NAT_DST + 1,
2136 /* After mangle, schedule and forward local requests */
2138 .hook = ip_vs_local_request4,
2140 .hooknum = NF_INET_LOCAL_OUT,
2141 .priority = NF_IP_PRI_NAT_DST + 2,
2143 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2144 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2146 .hook = ip_vs_forward_icmp,
2148 .hooknum = NF_INET_FORWARD,
2151 /* After packet filtering, change source only for VS/NAT */
2153 .hook = ip_vs_reply4,
2155 .hooknum = NF_INET_FORWARD,
2158 #ifdef CONFIG_IP_VS_IPV6
2159 /* After packet filtering, change source only for VS/NAT */
2161 .hook = ip_vs_reply6,
2163 .hooknum = NF_INET_LOCAL_IN,
2164 .priority = NF_IP6_PRI_NAT_SRC - 2,
2166 /* After packet filtering, forward packet through VS/DR, VS/TUN,
2167 * or VS/NAT(change destination), so that filtering rules can be
2168 * applied to IPVS. */
2170 .hook = ip_vs_remote_request6,
2172 .hooknum = NF_INET_LOCAL_IN,
2173 .priority = NF_IP6_PRI_NAT_SRC - 1,
2175 /* Before ip_vs_in, change source only for VS/NAT */
2177 .hook = ip_vs_local_reply6,
2179 .hooknum = NF_INET_LOCAL_OUT,
2180 .priority = NF_IP6_PRI_NAT_DST + 1,
2182 /* After mangle, schedule and forward local requests */
2184 .hook = ip_vs_local_request6,
2186 .hooknum = NF_INET_LOCAL_OUT,
2187 .priority = NF_IP6_PRI_NAT_DST + 2,
2189 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2190 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2192 .hook = ip_vs_forward_icmp_v6,
2194 .hooknum = NF_INET_FORWARD,
2197 /* After packet filtering, change source only for VS/NAT */
2199 .hook = ip_vs_reply6,
2201 .hooknum = NF_INET_FORWARD,
2207 * Initialize IP Virtual Server netns mem.
2209 static int __net_init __ip_vs_init(struct net *net)
2211 struct netns_ipvs *ipvs;
2214 ipvs = net_generic(net, ip_vs_net_id);
2218 /* Hold the beast until a service is registerd */
2221 /* Counters used for creating unique names */
2222 ipvs->gen = atomic_read(&ipvs_netns_cnt);
2223 atomic_inc(&ipvs_netns_cnt);
2226 if (ip_vs_estimator_net_init(ipvs) < 0)
2227 goto estimator_fail;
2229 if (ip_vs_control_net_init(ipvs) < 0)
2232 if (ip_vs_protocol_net_init(ipvs) < 0)
2235 if (ip_vs_app_net_init(ipvs) < 0)
2238 if (ip_vs_conn_net_init(ipvs) < 0)
2241 if (ip_vs_sync_net_init(ipvs) < 0)
2244 ret = nf_register_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2254 ip_vs_sync_net_cleanup(ipvs);
2256 ip_vs_conn_net_cleanup(ipvs);
2258 ip_vs_app_net_cleanup(ipvs);
2260 ip_vs_protocol_net_cleanup(ipvs);
2262 ip_vs_control_net_cleanup(ipvs);
2264 ip_vs_estimator_net_cleanup(ipvs);
2270 static void __net_exit __ip_vs_cleanup(struct net *net)
2272 struct netns_ipvs *ipvs = net_ipvs(net);
2274 nf_unregister_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2275 ip_vs_service_net_cleanup(ipvs); /* ip_vs_flush() with locks */
2276 ip_vs_conn_net_cleanup(ipvs);
2277 ip_vs_app_net_cleanup(ipvs);
2278 ip_vs_protocol_net_cleanup(ipvs);
2279 ip_vs_control_net_cleanup(ipvs);
2280 ip_vs_estimator_net_cleanup(ipvs);
2281 IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2285 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2287 struct netns_ipvs *ipvs = net_ipvs(net);
2289 ipvs->enable = 0; /* Disable packet reception */
2291 ip_vs_sync_net_cleanup(ipvs);
2295 static struct pernet_operations ipvs_core_ops = {
2296 .init = __ip_vs_init,
2297 .exit = __ip_vs_cleanup,
2298 .id = &ip_vs_net_id,
2299 .size = sizeof(struct netns_ipvs),
2302 static struct pernet_operations ipvs_core_dev_ops = {
2303 .exit = __ip_vs_dev_cleanup,
2307 * Initialize IP Virtual Server
2309 static int __init ip_vs_init(void)
2313 ret = ip_vs_control_init();
2315 pr_err("can't setup control.\n");
2319 ip_vs_protocol_init();
2321 ret = ip_vs_conn_init();
2323 pr_err("can't setup connection table.\n");
2324 goto cleanup_protocol;
2327 ret = register_pernet_subsys(&ipvs_core_ops); /* Alloc ip_vs struct */
2331 ret = register_pernet_device(&ipvs_core_dev_ops);
2335 ret = ip_vs_register_nl_ioctl();
2337 pr_err("can't register netlink/ioctl.\n");
2341 pr_info("ipvs loaded.\n");
2346 unregister_pernet_device(&ipvs_core_dev_ops);
2348 unregister_pernet_subsys(&ipvs_core_ops);
2350 ip_vs_conn_cleanup();
2352 ip_vs_protocol_cleanup();
2353 ip_vs_control_cleanup();
2358 static void __exit ip_vs_cleanup(void)
2360 ip_vs_unregister_nl_ioctl();
2361 unregister_pernet_device(&ipvs_core_dev_ops);
2362 unregister_pernet_subsys(&ipvs_core_ops); /* free ip_vs struct */
2363 ip_vs_conn_cleanup();
2364 ip_vs_protocol_cleanup();
2365 ip_vs_control_cleanup();
2366 pr_info("ipvs unloaded.\n");
2369 module_init(ip_vs_init);
2370 module_exit(ip_vs_cleanup);
2371 MODULE_LICENSE("GPL");