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);
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);
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)
80 const char *ip_vs_proto_name(unsigned int proto)
95 #ifdef CONFIG_IP_VS_IPV6
100 sprintf(buf, "IP_%u", proto);
105 void ip_vs_init_hash_table(struct list_head *table, int rows)
108 INIT_LIST_HEAD(&table[rows]);
112 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
114 struct ip_vs_dest *dest = cp->dest;
115 struct netns_ipvs *ipvs = cp->ipvs;
117 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
118 struct ip_vs_cpu_stats *s;
119 struct ip_vs_service *svc;
121 s = this_cpu_ptr(dest->stats.cpustats);
122 u64_stats_update_begin(&s->syncp);
124 s->cnt.inbytes += skb->len;
125 u64_stats_update_end(&s->syncp);
128 svc = rcu_dereference(dest->svc);
129 s = this_cpu_ptr(svc->stats.cpustats);
130 u64_stats_update_begin(&s->syncp);
132 s->cnt.inbytes += skb->len;
133 u64_stats_update_end(&s->syncp);
136 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
137 u64_stats_update_begin(&s->syncp);
139 s->cnt.inbytes += skb->len;
140 u64_stats_update_end(&s->syncp);
146 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
148 struct ip_vs_dest *dest = cp->dest;
149 struct netns_ipvs *ipvs = cp->ipvs;
151 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
152 struct ip_vs_cpu_stats *s;
153 struct ip_vs_service *svc;
155 s = this_cpu_ptr(dest->stats.cpustats);
156 u64_stats_update_begin(&s->syncp);
158 s->cnt.outbytes += skb->len;
159 u64_stats_update_end(&s->syncp);
162 svc = rcu_dereference(dest->svc);
163 s = this_cpu_ptr(svc->stats.cpustats);
164 u64_stats_update_begin(&s->syncp);
166 s->cnt.outbytes += skb->len;
167 u64_stats_update_end(&s->syncp);
170 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
171 u64_stats_update_begin(&s->syncp);
173 s->cnt.outbytes += skb->len;
174 u64_stats_update_end(&s->syncp);
180 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
182 struct netns_ipvs *ipvs = svc->ipvs;
183 struct ip_vs_cpu_stats *s;
185 s = this_cpu_ptr(cp->dest->stats.cpustats);
186 u64_stats_update_begin(&s->syncp);
188 u64_stats_update_end(&s->syncp);
190 s = this_cpu_ptr(svc->stats.cpustats);
191 u64_stats_update_begin(&s->syncp);
193 u64_stats_update_end(&s->syncp);
195 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
196 u64_stats_update_begin(&s->syncp);
198 u64_stats_update_end(&s->syncp);
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)
207 if (likely(pd->pp->state_transition))
208 pd->pp->state_transition(cp, direction, skb, pd);
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)
218 ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
220 p->pe = rcu_dereference(svc->pe);
221 if (p->pe && p->pe->fill_param)
222 return p->pe->fill_param(p, skb);
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
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)
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 */
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,
248 const union nf_inet_addr *src_addr, *dst_addr;
250 if (likely(!ip_vs_iph_inverse(iph))) {
251 src_addr = &iph->saddr;
252 dst_addr = &iph->daddr;
254 src_addr = &iph->daddr;
255 dst_addr = &iph->saddr;
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);
266 snet.ip = src_addr->ip & svc->netmask;
268 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
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));
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.
288 int protocol = iph->protocol;
289 const union nf_inet_addr *vaddr = dst_addr;
292 if (dst_port == svc->port) {
294 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
296 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
298 if (svc->port != FTPPORT)
301 /* Note: persistent fwmark-based services and
302 * persistent port zero service are handled here.
304 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
305 * port zero template:
306 * <protocol,caddr,0,vaddr,0,daddr,0>
309 protocol = IPPROTO_IP;
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) {
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;
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
331 sched = rcu_dereference(svc->scheduler);
333 /* read svc->sched_data after svc->scheduler */
335 dest = sched->schedule(svc, skb, iph);
340 IP_VS_DBG(1, "p-schedule: no dest found.\n");
341 kfree(param.pe_data);
346 if (dst_port == svc->port && svc->port != FTPPORT)
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);
356 kfree(param.pe_data);
361 ct->timeout = svc->timeout;
363 /* set destination with the found template */
365 kfree(param.pe_data);
369 if (dport == svc->port && dest->port)
372 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
373 && iph->protocol == IPPROTO_UDP) ?
374 IP_VS_CONN_F_ONE_PACKET : 0;
377 * Create a new connection according to the template
379 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
380 src_port, dst_addr, dst_port, ¶m);
382 cp = ip_vs_conn_new(¶m, dest->af, &dest->addr, dport, flags, dest,
393 ip_vs_control_add(cp, ct);
396 ip_vs_conn_stats(cp, svc);
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
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.
413 * 0 : scheduler can not find destination, so try bypass or
414 * return ICMP and then NF_DROP (ip_vs_leave).
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.
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)
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;
437 * IPv6 frags, only the first hit here.
439 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
443 if (likely(!ip_vs_iph_inverse(iph))) {
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.
461 if (cport == FTPDATA) {
462 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
463 "Not scheduling FTPDATA");
468 * Do not schedule replies from local real server.
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;
476 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
477 "Not scheduling reply for existing"
479 __ip_vs_conn_put(cp);
487 if (svc->flags & IP_VS_SVC_F_PERSISTENT)
488 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
494 * Non-persistent service
496 if (!svc->fwmark && vport != svc->port) {
498 pr_err("Schedule: port zero only supported "
499 "in persistent services, "
500 "check your ipvs configuration\n");
504 sched = rcu_dereference(svc->scheduler);
506 /* read svc->sched_data after svc->scheduler */
508 dest = sched->schedule(svc, skb, iph);
513 IP_VS_DBG(1, "Schedule: no dest found.\n");
517 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
518 && iph->protocol == IPPROTO_UDP) ?
519 IP_VS_CONN_F_ONE_PACKET : 0;
522 * Create a connection entry.
525 struct ip_vs_conn_param p;
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);
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",
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));
546 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);
563 * Pass or drop the packet.
564 * Called by ip_vs_in, when the virtual service is available but
565 * no destination is available for a new connection.
567 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
568 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
570 __be16 _ports[2], *pptr, dport;
571 struct netns_ipvs *ipvs = svc->ipvs;
572 struct net *net = ipvs->net;
574 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
577 dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
579 /* if it is fwmark-based service, the cache_bypass sysctl is up
580 and the destination is a non-local unicast, then create
581 a cache_bypass connection entry */
582 if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
583 !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
584 ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
586 struct ip_vs_conn *cp;
587 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
588 iph->protocol == IPPROTO_UDP) ?
589 IP_VS_CONN_F_ONE_PACKET : 0;
590 union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } };
592 /* create a new connection entry */
593 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
595 struct ip_vs_conn_param p;
596 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
597 &iph->saddr, pptr[0],
598 &iph->daddr, pptr[1], &p);
599 cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
600 IP_VS_CONN_F_BYPASS | flags,
607 ip_vs_in_stats(cp, skb);
610 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
612 /* transmit the first SYN packet */
613 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
614 /* do not touch skb anymore */
616 atomic_inc(&cp->in_pkts);
622 * When the virtual ftp service is presented, packets destined
623 * for other services on the VIP may get here (except services
624 * listed in the ipvs table), pass the packets, because it is
625 * not ipvs job to decide to drop the packets.
627 if (svc->port == FTPPORT && dport != FTPPORT)
630 if (unlikely(ip_vs_iph_icmp(iph)))
634 * Notify the client that the destination is unreachable, and
635 * release the socket buffer.
636 * Since it is in IP layer, the TCP socket is not actually
637 * created, the TCP RST packet cannot be sent, instead that
638 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
640 #ifdef CONFIG_IP_VS_IPV6
641 if (svc->af == AF_INET6) {
643 skb->dev = net->loopback_dev;
644 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
647 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
654 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
656 return ipvs->sysctl_snat_reroute;
659 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
661 return ipvs->sysctl_nat_icmp_send;
664 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
666 return ipvs->sysctl_expire_nodest_conn;
671 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
672 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
673 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
677 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
679 return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
682 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
684 if (NF_INET_LOCAL_IN == hooknum)
685 return IP_DEFRAG_VS_IN;
686 if (NF_INET_FORWARD == hooknum)
687 return IP_DEFRAG_VS_FWD;
688 return IP_DEFRAG_VS_OUT;
691 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
692 struct sk_buff *skb, u_int32_t user)
697 err = ip_defrag(skb, user);
700 ip_send_check(ip_hdr(skb));
705 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
706 struct sk_buff *skb, unsigned int hooknum)
708 if (!sysctl_snat_reroute(ipvs))
710 /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
711 if (NF_INET_LOCAL_IN == hooknum)
713 #ifdef CONFIG_IP_VS_IPV6
714 if (af == AF_INET6) {
715 struct dst_entry *dst = skb_dst(skb);
717 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
718 ip6_route_me_harder(ipvs->net, skb) != 0)
722 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
723 ip_route_me_harder(ipvs->net, skb, RTN_LOCAL) != 0)
730 * Packet has been made sufficiently writable in caller
731 * - inout: 1=in->out, 0=out->in
733 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
734 struct ip_vs_conn *cp, int inout)
736 struct iphdr *iph = ip_hdr(skb);
737 unsigned int icmp_offset = iph->ihl*4;
738 struct icmphdr *icmph = (struct icmphdr *)(skb_network_header(skb) +
740 struct iphdr *ciph = (struct iphdr *)(icmph + 1);
743 iph->saddr = cp->vaddr.ip;
745 ciph->daddr = cp->vaddr.ip;
748 iph->daddr = cp->daddr.ip;
750 ciph->saddr = cp->daddr.ip;
754 /* the TCP/UDP/SCTP port */
755 if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
756 IPPROTO_SCTP == ciph->protocol) {
757 __be16 *ports = (void *)ciph + ciph->ihl*4;
760 ports[1] = cp->vport;
762 ports[0] = cp->dport;
765 /* And finally the ICMP checksum */
767 icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
768 skb->ip_summed = CHECKSUM_UNNECESSARY;
771 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
772 "Forwarding altered outgoing ICMP");
774 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
775 "Forwarding altered incoming ICMP");
778 #ifdef CONFIG_IP_VS_IPV6
779 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
780 struct ip_vs_conn *cp, int inout)
782 struct ipv6hdr *iph = ipv6_hdr(skb);
783 unsigned int icmp_offset = 0;
784 unsigned int offs = 0; /* header offset*/
786 struct icmp6hdr *icmph;
787 struct ipv6hdr *ciph;
788 unsigned short fragoffs;
790 ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
791 icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
792 offs = icmp_offset + sizeof(struct icmp6hdr);
793 ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
795 protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
798 iph->saddr = cp->vaddr.in6;
799 ciph->daddr = cp->vaddr.in6;
801 iph->daddr = cp->daddr.in6;
802 ciph->saddr = cp->daddr.in6;
805 /* the TCP/UDP/SCTP port */
806 if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
807 IPPROTO_SCTP == protocol)) {
808 __be16 *ports = (void *)(skb_network_header(skb) + offs);
810 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
811 ntohs(inout ? ports[1] : ports[0]),
812 ntohs(inout ? cp->vport : cp->dport));
814 ports[1] = cp->vport;
816 ports[0] = cp->dport;
819 /* And finally the ICMP checksum */
820 icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
821 skb->len - icmp_offset,
823 skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
824 skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
825 skb->ip_summed = CHECKSUM_PARTIAL;
828 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
829 (void *)ciph - (void *)iph,
830 "Forwarding altered outgoing ICMPv6");
832 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
833 (void *)ciph - (void *)iph,
834 "Forwarding altered incoming ICMPv6");
838 /* Handle relevant response ICMP messages - forward to the right
841 static int handle_response_icmp(int af, struct sk_buff *skb,
842 union nf_inet_addr *snet,
843 __u8 protocol, struct ip_vs_conn *cp,
844 struct ip_vs_protocol *pp,
845 unsigned int offset, unsigned int ihl,
846 unsigned int hooknum)
848 unsigned int verdict = NF_DROP;
850 if (IP_VS_FWD_METHOD(cp) != 0) {
851 pr_err("shouldn't reach here, because the box is on the "
852 "half connection in the tun/dr module.\n");
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);
890 __ip_vs_conn_put(cp);
896 * Handle ICMP messages in the inside-to-outside direction (outgoing).
897 * Find any that might be relevant, check against existing connections.
898 * Currently handles error types - unreachable, quench, ttl exceeded.
900 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
901 int *related, unsigned int hooknum)
904 struct icmphdr _icmph, *ic;
905 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
906 struct ip_vs_iphdr ciph;
907 struct ip_vs_conn *cp;
908 struct ip_vs_protocol *pp;
909 unsigned int offset, ihl;
910 union nf_inet_addr snet;
914 /* reassemble IP fragments */
915 if (ip_is_fragment(ip_hdr(skb))) {
916 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
921 offset = ihl = iph->ihl * 4;
922 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
926 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
927 ic->type, ntohs(icmp_id(ic)),
928 &iph->saddr, &iph->daddr);
931 * Work through seeing if this is for us.
932 * These checks are supposed to be in an order that means easy
933 * things are checked first to speed up processing.... however
934 * this means that some packets will manage to get a long way
935 * down this stack and then be rejected, but that's life.
937 if ((ic->type != ICMP_DEST_UNREACH) &&
938 (ic->type != ICMP_SOURCE_QUENCH) &&
939 (ic->type != ICMP_TIME_EXCEEDED)) {
944 /* Now find the contained IP header */
945 offset += sizeof(_icmph);
946 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
948 return NF_ACCEPT; /* The packet looks wrong, ignore */
950 pp = ip_vs_proto_get(cih->protocol);
954 /* Is the embedded protocol header present? */
955 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
959 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
960 "Checking outgoing ICMP for");
962 ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
964 /* The embedded headers contain source and dest in reverse order */
965 cp = pp->conn_out_get(ipvs, AF_INET, skb, &ciph);
969 snet.ip = iph->saddr;
970 return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
971 pp, ciph.len, ihl, hooknum);
974 #ifdef CONFIG_IP_VS_IPV6
975 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
976 int *related, unsigned int hooknum,
977 struct ip_vs_iphdr *ipvsh)
979 struct icmp6hdr _icmph, *ic;
980 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
981 struct ip_vs_conn *cp;
982 struct ip_vs_protocol *pp;
983 union nf_inet_addr snet;
987 ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
992 * Work through seeing if this is for us.
993 * These checks are supposed to be in an order that means easy
994 * things are checked first to speed up processing.... however
995 * this means that some packets will manage to get a long way
996 * down this stack and then be rejected, but that's life.
998 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1002 /* Fragment header that is before ICMP header tells us that:
1003 * it's not an error message since they can't be fragmented.
1005 if (ipvsh->flags & IP6_FH_F_FRAG)
1008 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1009 ic->icmp6_type, ntohs(icmpv6_id(ic)),
1010 &ipvsh->saddr, &ipvsh->daddr);
1012 if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1014 return NF_ACCEPT; /* The packet looks wrong, ignore */
1016 pp = ip_vs_proto_get(ciph.protocol);
1020 /* The embedded headers contain source and dest in reverse order */
1021 cp = pp->conn_out_get(ipvs, AF_INET6, skb, &ciph);
1025 snet.in6 = ciph.saddr.in6;
1027 return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1028 pp, offset, sizeof(struct ipv6hdr),
1034 * Check if sctp chunc is ABORT chunk
1036 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1038 sctp_chunkhdr_t *sch, schunk;
1039 sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1040 sizeof(schunk), &schunk);
1043 if (sch->type == SCTP_CID_ABORT)
1048 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1050 struct tcphdr _tcph, *th;
1052 th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1058 static inline bool is_new_conn(const struct sk_buff *skb,
1059 struct ip_vs_iphdr *iph)
1061 switch (iph->protocol) {
1063 struct tcphdr _tcph, *th;
1065 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1070 case IPPROTO_SCTP: {
1071 sctp_chunkhdr_t *sch, schunk;
1073 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1074 sizeof(schunk), &schunk);
1077 return sch->type == SCTP_CID_INIT;
1084 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1085 int conn_reuse_mode)
1087 /* Controlled (FTP DATA or persistence)? */
1091 switch (cp->protocol) {
1093 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1094 ((conn_reuse_mode & 2) &&
1095 (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1096 (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1098 return cp->state == IP_VS_SCTP_S_CLOSED;
1104 /* Handle response packets: rewrite addresses and send away...
1107 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1108 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1109 unsigned int hooknum)
1111 struct ip_vs_protocol *pp = pd->pp;
1113 IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1115 if (!skb_make_writable(skb, iph->len))
1118 /* mangle the packet */
1119 if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1122 #ifdef CONFIG_IP_VS_IPV6
1124 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1128 ip_hdr(skb)->saddr = cp->vaddr.ip;
1129 ip_send_check(ip_hdr(skb));
1133 * nf_iterate does not expect change in the skb->dst->dev.
1134 * It looks like it is not fatal to enable this code for hooks
1135 * where our handlers are at the end of the chain list and
1136 * when all next handlers use skb->dst->dev and not outdev.
1137 * It will definitely route properly the inout NAT traffic
1138 * when multiple paths are used.
1141 /* For policy routing, packets originating from this
1142 * machine itself may be routed differently to packets
1143 * passing through. We want this packet to be routed as
1144 * if it came from this machine itself. So re-compute
1145 * the routing information.
1147 if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1150 IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1152 ip_vs_out_stats(cp, skb);
1153 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1154 skb->ipvs_property = 1;
1155 if (!(cp->flags & IP_VS_CONN_F_NFCT))
1158 ip_vs_update_conntrack(skb, cp, 0);
1172 * Check if outgoing packet belongs to the established ip_vs_conn.
1175 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1177 struct net *net = ipvs->net;
1178 struct ip_vs_iphdr iph;
1179 struct ip_vs_protocol *pp;
1180 struct ip_vs_proto_data *pd;
1181 struct ip_vs_conn *cp;
1185 /* Already marked as IPVS request or reply? */
1186 if (skb->ipvs_property)
1189 /* Bad... Do not break raw sockets */
1190 if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1192 struct sock *sk = skb->sk;
1193 struct inet_sock *inet = inet_sk(skb->sk);
1195 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1199 if (unlikely(!skb_dst(skb)))
1205 ip_vs_fill_iph_skb(af, skb, false, &iph);
1206 #ifdef CONFIG_IP_VS_IPV6
1207 if (af == AF_INET6) {
1208 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1210 int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1218 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1220 int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1226 pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1231 /* reassemble IP fragments */
1232 #ifdef CONFIG_IP_VS_IPV6
1235 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1236 if (ip_vs_gather_frags(ipvs, skb,
1237 ip_vs_defrag_user(hooknum)))
1240 ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1244 * Check if the packet belongs to an existing entry
1246 cp = pp->conn_out_get(ipvs, af, skb, &iph);
1249 return handle_response(af, skb, pd, cp, &iph, hooknum);
1250 if (sysctl_nat_icmp_send(ipvs) &&
1251 (pp->protocol == IPPROTO_TCP ||
1252 pp->protocol == IPPROTO_UDP ||
1253 pp->protocol == IPPROTO_SCTP)) {
1254 __be16 _ports[2], *pptr;
1256 pptr = frag_safe_skb_hp(skb, iph.len,
1257 sizeof(_ports), _ports, &iph);
1259 return NF_ACCEPT; /* Not for me */
1260 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1263 * Notify the real server: there is no
1264 * existing entry if it is not RST
1265 * packet or not TCP packet.
1267 if ((iph.protocol != IPPROTO_TCP &&
1268 iph.protocol != IPPROTO_SCTP)
1269 || ((iph.protocol == IPPROTO_TCP
1270 && !is_tcp_reset(skb, iph.len))
1271 || (iph.protocol == IPPROTO_SCTP
1272 && !is_sctp_abort(skb,
1274 #ifdef CONFIG_IP_VS_IPV6
1275 if (af == AF_INET6) {
1277 skb->dev = net->loopback_dev;
1279 ICMPV6_DEST_UNREACH,
1280 ICMPV6_PORT_UNREACH,
1286 ICMP_PORT_UNREACH, 0);
1291 IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1292 "ip_vs_out: packet continues traversal as normal");
1297 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1298 * used only for VS/NAT.
1299 * Check if packet is reply for established ip_vs_conn.
1302 ip_vs_reply4(void *priv, struct sk_buff *skb,
1303 const struct nf_hook_state *state)
1305 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1309 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1310 * Check if packet is reply for established ip_vs_conn.
1313 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1314 const struct nf_hook_state *state)
1316 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1319 #ifdef CONFIG_IP_VS_IPV6
1322 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1323 * used only for VS/NAT.
1324 * Check if packet is reply for established ip_vs_conn.
1327 ip_vs_reply6(void *priv, struct sk_buff *skb,
1328 const struct nf_hook_state *state)
1330 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1334 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1335 * Check if packet is reply for established ip_vs_conn.
1338 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1339 const struct nf_hook_state *state)
1341 return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1347 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1348 struct ip_vs_proto_data *pd,
1349 int *verdict, struct ip_vs_conn **cpp,
1350 struct ip_vs_iphdr *iph)
1352 struct ip_vs_protocol *pp = pd->pp;
1354 if (!iph->fragoffs) {
1355 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1356 * replayed fragment zero will already have created the cp
1359 /* Schedule and create new connection entry into cpp */
1360 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1364 if (unlikely(!*cpp)) {
1365 /* sorry, all this trouble for a no-hit :) */
1366 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1367 "ip_vs_in: packet continues traversal as normal");
1368 if (iph->fragoffs) {
1369 /* Fragment that couldn't be mapped to a conn entry
1370 * is missing module nf_defrag_ipv6
1372 IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1373 IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1374 "unhandled fragment");
1376 *verdict = NF_ACCEPT;
1384 * Handle ICMP messages in the outside-to-inside direction (incoming).
1385 * Find any that might be relevant, check against existing connections,
1386 * forward to the right destination host if relevant.
1387 * Currently handles error types - unreachable, quench, ttl exceeded.
1390 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1391 unsigned int hooknum)
1394 struct icmphdr _icmph, *ic;
1395 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
1396 struct ip_vs_iphdr ciph;
1397 struct ip_vs_conn *cp;
1398 struct ip_vs_protocol *pp;
1399 struct ip_vs_proto_data *pd;
1400 unsigned int offset, offset2, ihl, verdict;
1401 bool ipip, new_cp = false;
1405 /* reassemble IP fragments */
1406 if (ip_is_fragment(ip_hdr(skb))) {
1407 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1412 offset = ihl = iph->ihl * 4;
1413 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1417 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1418 ic->type, ntohs(icmp_id(ic)),
1419 &iph->saddr, &iph->daddr);
1422 * Work through seeing if this is for us.
1423 * These checks are supposed to be in an order that means easy
1424 * things are checked first to speed up processing.... however
1425 * this means that some packets will manage to get a long way
1426 * down this stack and then be rejected, but that's life.
1428 if ((ic->type != ICMP_DEST_UNREACH) &&
1429 (ic->type != ICMP_SOURCE_QUENCH) &&
1430 (ic->type != ICMP_TIME_EXCEEDED)) {
1435 /* Now find the contained IP header */
1436 offset += sizeof(_icmph);
1437 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1439 return NF_ACCEPT; /* The packet looks wrong, ignore */
1441 /* Special case for errors for IPIP packets */
1443 if (cih->protocol == IPPROTO_IPIP) {
1444 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1446 /* Error for our IPIP must arrive at LOCAL_IN */
1447 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1449 offset += cih->ihl * 4;
1450 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1452 return NF_ACCEPT; /* The packet looks wrong, ignore */
1456 pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1461 /* Is the embedded protocol header present? */
1462 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1466 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1467 "Checking incoming ICMP for");
1470 ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1473 /* The embedded headers contain source and dest in reverse order.
1474 * For IPIP this is error for request, not for reply.
1476 cp = pp->conn_in_get(ipvs, AF_INET, skb, &ciph);
1481 if (!sysctl_schedule_icmp(ipvs))
1484 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1491 /* Ensure the checksum is correct */
1492 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1493 /* Failed checksum! */
1494 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1500 __be32 info = ic->un.gateway;
1501 __u8 type = ic->type;
1502 __u8 code = ic->code;
1504 /* Update the MTU */
1505 if (ic->type == ICMP_DEST_UNREACH &&
1506 ic->code == ICMP_FRAG_NEEDED) {
1507 struct ip_vs_dest *dest = cp->dest;
1508 u32 mtu = ntohs(ic->un.frag.mtu);
1509 __be16 frag_off = cih->frag_off;
1511 /* Strip outer IP and ICMP, go to IPIP header */
1512 if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1514 offset2 -= ihl + sizeof(_icmph);
1515 skb_reset_network_header(skb);
1516 IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1517 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1518 ipv4_update_pmtu(skb, ipvs->net,
1520 /* Client uses PMTUD? */
1521 if (!(frag_off & htons(IP_DF)))
1523 /* Prefer the resulting PMTU */
1525 struct ip_vs_dest_dst *dest_dst;
1528 dest_dst = rcu_dereference(dest->dest_dst);
1530 mtu = dst_mtu(dest_dst->dst_cache);
1533 if (mtu > 68 + sizeof(struct iphdr))
1534 mtu -= sizeof(struct iphdr);
1537 /* Strip outer IP, ICMP and IPIP, go to IP header of
1540 if (pskb_pull(skb, offset2) == NULL)
1542 skb_reset_network_header(skb);
1543 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1544 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1545 type, code, ntohl(info));
1546 icmp_send(skb, type, code, info);
1547 /* ICMP can be shorter but anyways, account it */
1548 ip_vs_out_stats(cp, skb);
1552 verdict = NF_STOLEN;
1556 /* do the statistics and put it back */
1557 ip_vs_in_stats(cp, skb);
1558 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1559 IPPROTO_SCTP == cih->protocol)
1560 offset += 2 * sizeof(__u16);
1561 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1564 if (likely(!new_cp))
1565 __ip_vs_conn_put(cp);
1572 #ifdef CONFIG_IP_VS_IPV6
1573 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1574 int *related, unsigned int hooknum,
1575 struct ip_vs_iphdr *iph)
1577 struct icmp6hdr _icmph, *ic;
1578 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1579 struct ip_vs_conn *cp;
1580 struct ip_vs_protocol *pp;
1581 struct ip_vs_proto_data *pd;
1582 unsigned int offset, verdict;
1583 bool new_cp = false;
1587 ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1592 * Work through seeing if this is for us.
1593 * These checks are supposed to be in an order that means easy
1594 * things are checked first to speed up processing.... however
1595 * this means that some packets will manage to get a long way
1596 * down this stack and then be rejected, but that's life.
1598 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1602 /* Fragment header that is before ICMP header tells us that:
1603 * it's not an error message since they can't be fragmented.
1605 if (iph->flags & IP6_FH_F_FRAG)
1608 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1609 ic->icmp6_type, ntohs(icmpv6_id(ic)),
1610 &iph->saddr, &iph->daddr);
1612 offset = iph->len + sizeof(_icmph);
1613 if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1616 pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1621 /* Cannot handle fragmented embedded protocol */
1625 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1626 "Checking incoming ICMPv6 for");
1628 /* The embedded headers contain source and dest in reverse order
1629 * if not from localhost
1631 cp = pp->conn_in_get(ipvs, AF_INET6, skb, &ciph);
1636 if (!sysctl_schedule_icmp(ipvs))
1639 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1645 /* VS/TUN, VS/DR and LOCALNODE just let it go */
1646 if ((hooknum == NF_INET_LOCAL_OUT) &&
1647 (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1648 verdict = NF_ACCEPT;
1652 /* do the statistics and put it back */
1653 ip_vs_in_stats(cp, skb);
1655 /* Need to mangle contained IPv6 header in ICMPv6 packet */
1657 if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1658 IPPROTO_SCTP == ciph.protocol)
1659 offset += 2 * sizeof(__u16); /* Also mangle ports */
1661 verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1664 if (likely(!new_cp))
1665 __ip_vs_conn_put(cp);
1675 * Check if it's for virtual services, look it up,
1676 * and send it on its way...
1679 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1681 struct ip_vs_iphdr iph;
1682 struct ip_vs_protocol *pp;
1683 struct ip_vs_proto_data *pd;
1684 struct ip_vs_conn *cp;
1686 int conn_reuse_mode;
1688 /* Already marked as IPVS request or reply? */
1689 if (skb->ipvs_property)
1694 * - remote client: only PACKET_HOST
1695 * - route: used for struct net when skb->dev is unset
1697 if (unlikely((skb->pkt_type != PACKET_HOST &&
1698 hooknum != NF_INET_LOCAL_OUT) ||
1700 ip_vs_fill_iph_skb(af, skb, false, &iph);
1701 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1702 " ignored in hook %u\n",
1703 skb->pkt_type, iph.protocol,
1704 IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1707 /* ipvs enabled in this netns ? */
1708 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1711 ip_vs_fill_iph_skb(af, skb, false, &iph);
1713 /* Bad... Do not break raw sockets */
1714 if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1716 struct sock *sk = skb->sk;
1717 struct inet_sock *inet = inet_sk(skb->sk);
1719 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1723 #ifdef CONFIG_IP_VS_IPV6
1724 if (af == AF_INET6) {
1725 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1727 int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
1735 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1737 int verdict = ip_vs_in_icmp(ipvs, skb, &related,
1744 /* Protocol supported? */
1745 pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1746 if (unlikely(!pd)) {
1747 /* The only way we'll see this packet again is if it's
1748 * encapsulated, so mark it with ipvs_property=1 so we
1749 * skip it if we're ignoring tunneled packets
1751 if (sysctl_ignore_tunneled(ipvs))
1752 skb->ipvs_property = 1;
1758 * Check if the packet belongs to an existing connection entry
1760 cp = pp->conn_in_get(ipvs, af, skb, &iph);
1762 conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1763 if (conn_reuse_mode && !iph.fragoffs &&
1764 is_new_conn(skb, &iph) && cp &&
1765 ((unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1766 unlikely(!atomic_read(&cp->dest->weight))) ||
1767 unlikely(is_new_conn_expected(cp, conn_reuse_mode)))) {
1768 if (!atomic_read(&cp->n_control))
1769 ip_vs_conn_expire_now(cp);
1770 __ip_vs_conn_put(cp);
1774 if (unlikely(!cp)) {
1777 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
1781 IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
1783 /* Check the server status */
1784 if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1785 /* the destination server is not available */
1787 if (sysctl_expire_nodest_conn(ipvs)) {
1788 /* try to expire the connection immediately */
1789 ip_vs_conn_expire_now(cp);
1791 /* don't restart its timer, and silently
1793 __ip_vs_conn_put(cp);
1797 ip_vs_in_stats(cp, skb);
1798 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1799 if (cp->packet_xmit)
1800 ret = cp->packet_xmit(skb, cp, pp, &iph);
1801 /* do not touch skb anymore */
1803 IP_VS_DBG_RL("warning: packet_xmit is null");
1807 /* Increase its packet counter and check if it is needed
1808 * to be synchronized
1810 * Sync connection if it is about to close to
1811 * encorage the standby servers to update the connections timeout
1813 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1816 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1817 pkts = sysctl_sync_threshold(ipvs);
1819 pkts = atomic_add_return(1, &cp->in_pkts);
1821 if (ipvs->sync_state & IP_VS_STATE_MASTER)
1822 ip_vs_sync_conn(ipvs, cp, pkts);
1829 * AF_INET handler in NF_INET_LOCAL_IN chain
1830 * Schedule and forward packets from remote clients
1833 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
1834 const struct nf_hook_state *state)
1836 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
1840 * AF_INET handler in NF_INET_LOCAL_OUT chain
1841 * Schedule and forward packets from local clients
1844 ip_vs_local_request4(void *priv, struct sk_buff *skb,
1845 const struct nf_hook_state *state)
1847 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
1850 #ifdef CONFIG_IP_VS_IPV6
1853 * AF_INET6 handler in NF_INET_LOCAL_IN chain
1854 * Schedule and forward packets from remote clients
1857 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
1858 const struct nf_hook_state *state)
1860 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
1864 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
1865 * Schedule and forward packets from local clients
1868 ip_vs_local_request6(void *priv, struct sk_buff *skb,
1869 const struct nf_hook_state *state)
1871 return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
1878 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1879 * related packets destined for 0.0.0.0/0.
1880 * When fwmark-based virtual service is used, such as transparent
1881 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1882 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1883 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1884 * and send them to ip_vs_in_icmp.
1887 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
1888 const struct nf_hook_state *state)
1891 struct netns_ipvs *ipvs = net_ipvs(state->net);
1893 if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1896 /* ipvs enabled in this netns ? */
1897 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1900 return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
1903 #ifdef CONFIG_IP_VS_IPV6
1905 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
1906 const struct nf_hook_state *state)
1909 struct netns_ipvs *ipvs = net_ipvs(state->net);
1910 struct ip_vs_iphdr iphdr;
1912 ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
1913 if (iphdr.protocol != IPPROTO_ICMPV6)
1916 /* ipvs enabled in this netns ? */
1917 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1920 return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
1925 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1926 /* After packet filtering, change source only for VS/NAT */
1928 .hook = ip_vs_reply4,
1929 .owner = THIS_MODULE,
1931 .hooknum = NF_INET_LOCAL_IN,
1932 .priority = NF_IP_PRI_NAT_SRC - 2,
1934 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1935 * or VS/NAT(change destination), so that filtering rules can be
1936 * applied to IPVS. */
1938 .hook = ip_vs_remote_request4,
1939 .owner = THIS_MODULE,
1941 .hooknum = NF_INET_LOCAL_IN,
1942 .priority = NF_IP_PRI_NAT_SRC - 1,
1944 /* Before ip_vs_in, change source only for VS/NAT */
1946 .hook = ip_vs_local_reply4,
1947 .owner = THIS_MODULE,
1949 .hooknum = NF_INET_LOCAL_OUT,
1950 .priority = NF_IP_PRI_NAT_DST + 1,
1952 /* After mangle, schedule and forward local requests */
1954 .hook = ip_vs_local_request4,
1955 .owner = THIS_MODULE,
1957 .hooknum = NF_INET_LOCAL_OUT,
1958 .priority = NF_IP_PRI_NAT_DST + 2,
1960 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1961 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1963 .hook = ip_vs_forward_icmp,
1964 .owner = THIS_MODULE,
1966 .hooknum = NF_INET_FORWARD,
1969 /* After packet filtering, change source only for VS/NAT */
1971 .hook = ip_vs_reply4,
1972 .owner = THIS_MODULE,
1974 .hooknum = NF_INET_FORWARD,
1977 #ifdef CONFIG_IP_VS_IPV6
1978 /* After packet filtering, change source only for VS/NAT */
1980 .hook = ip_vs_reply6,
1981 .owner = THIS_MODULE,
1983 .hooknum = NF_INET_LOCAL_IN,
1984 .priority = NF_IP6_PRI_NAT_SRC - 2,
1986 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1987 * or VS/NAT(change destination), so that filtering rules can be
1988 * applied to IPVS. */
1990 .hook = ip_vs_remote_request6,
1991 .owner = THIS_MODULE,
1993 .hooknum = NF_INET_LOCAL_IN,
1994 .priority = NF_IP6_PRI_NAT_SRC - 1,
1996 /* Before ip_vs_in, change source only for VS/NAT */
1998 .hook = ip_vs_local_reply6,
1999 .owner = THIS_MODULE,
2001 .hooknum = NF_INET_LOCAL_OUT,
2002 .priority = NF_IP6_PRI_NAT_DST + 1,
2004 /* After mangle, schedule and forward local requests */
2006 .hook = ip_vs_local_request6,
2007 .owner = THIS_MODULE,
2009 .hooknum = NF_INET_LOCAL_OUT,
2010 .priority = NF_IP6_PRI_NAT_DST + 2,
2012 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2013 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2015 .hook = ip_vs_forward_icmp_v6,
2016 .owner = THIS_MODULE,
2018 .hooknum = NF_INET_FORWARD,
2021 /* After packet filtering, change source only for VS/NAT */
2023 .hook = ip_vs_reply6,
2024 .owner = THIS_MODULE,
2026 .hooknum = NF_INET_FORWARD,
2032 * Initialize IP Virtual Server netns mem.
2034 static int __net_init __ip_vs_init(struct net *net)
2036 struct netns_ipvs *ipvs;
2038 ipvs = net_generic(net, ip_vs_net_id);
2042 /* Hold the beast until a service is registerd */
2045 /* Counters used for creating unique names */
2046 ipvs->gen = atomic_read(&ipvs_netns_cnt);
2047 atomic_inc(&ipvs_netns_cnt);
2050 if (ip_vs_estimator_net_init(ipvs) < 0)
2051 goto estimator_fail;
2053 if (ip_vs_control_net_init(ipvs) < 0)
2056 if (ip_vs_protocol_net_init(ipvs) < 0)
2059 if (ip_vs_app_net_init(ipvs) < 0)
2062 if (ip_vs_conn_net_init(ipvs) < 0)
2065 if (ip_vs_sync_net_init(ipvs) < 0)
2068 printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2069 sizeof(struct netns_ipvs), ipvs->gen);
2076 ip_vs_conn_net_cleanup(ipvs);
2078 ip_vs_app_net_cleanup(ipvs);
2080 ip_vs_protocol_net_cleanup(ipvs);
2082 ip_vs_control_net_cleanup(ipvs);
2084 ip_vs_estimator_net_cleanup(ipvs);
2090 static void __net_exit __ip_vs_cleanup(struct net *net)
2092 struct netns_ipvs *ipvs = net_ipvs(net);
2094 ip_vs_service_net_cleanup(ipvs); /* ip_vs_flush() with locks */
2095 ip_vs_conn_net_cleanup(ipvs);
2096 ip_vs_app_net_cleanup(ipvs);
2097 ip_vs_protocol_net_cleanup(ipvs);
2098 ip_vs_control_net_cleanup(ipvs);
2099 ip_vs_estimator_net_cleanup(ipvs);
2100 IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2104 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2106 struct netns_ipvs *ipvs = net_ipvs(net);
2108 ipvs->enable = 0; /* Disable packet reception */
2110 ip_vs_sync_net_cleanup(ipvs);
2114 static struct pernet_operations ipvs_core_ops = {
2115 .init = __ip_vs_init,
2116 .exit = __ip_vs_cleanup,
2117 .id = &ip_vs_net_id,
2118 .size = sizeof(struct netns_ipvs),
2121 static struct pernet_operations ipvs_core_dev_ops = {
2122 .exit = __ip_vs_dev_cleanup,
2126 * Initialize IP Virtual Server
2128 static int __init ip_vs_init(void)
2132 ret = ip_vs_control_init();
2134 pr_err("can't setup control.\n");
2138 ip_vs_protocol_init();
2140 ret = ip_vs_conn_init();
2142 pr_err("can't setup connection table.\n");
2143 goto cleanup_protocol;
2146 ret = register_pernet_subsys(&ipvs_core_ops); /* Alloc ip_vs struct */
2150 ret = register_pernet_device(&ipvs_core_dev_ops);
2154 ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2156 pr_err("can't register hooks.\n");
2160 ret = ip_vs_register_nl_ioctl();
2162 pr_err("can't register netlink/ioctl.\n");
2166 pr_info("ipvs loaded.\n");
2171 nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2173 unregister_pernet_device(&ipvs_core_dev_ops);
2175 unregister_pernet_subsys(&ipvs_core_ops);
2177 ip_vs_conn_cleanup();
2179 ip_vs_protocol_cleanup();
2180 ip_vs_control_cleanup();
2185 static void __exit ip_vs_cleanup(void)
2187 ip_vs_unregister_nl_ioctl();
2188 nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2189 unregister_pernet_device(&ipvs_core_dev_ops);
2190 unregister_pernet_subsys(&ipvs_core_ops); /* free ip_vs struct */
2191 ip_vs_conn_cleanup();
2192 ip_vs_protocol_cleanup();
2193 ip_vs_control_cleanup();
2194 pr_info("ipvs unloaded.\n");
2197 module_init(ip_vs_init);
2198 module_exit(ip_vs_cleanup);
2199 MODULE_LICENSE("GPL");