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_%d", 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 = net_ipvs(skb_net(skb));
117 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
118 struct ip_vs_cpu_stats *s;
120 s = this_cpu_ptr(dest->stats.cpustats);
122 u64_stats_update_begin(&s->syncp);
123 s->ustats.inbytes += skb->len;
124 u64_stats_update_end(&s->syncp);
126 s = this_cpu_ptr(dest->svc->stats.cpustats);
128 u64_stats_update_begin(&s->syncp);
129 s->ustats.inbytes += skb->len;
130 u64_stats_update_end(&s->syncp);
132 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
134 u64_stats_update_begin(&s->syncp);
135 s->ustats.inbytes += skb->len;
136 u64_stats_update_end(&s->syncp);
142 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
144 struct ip_vs_dest *dest = cp->dest;
145 struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
147 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
148 struct ip_vs_cpu_stats *s;
150 s = this_cpu_ptr(dest->stats.cpustats);
152 u64_stats_update_begin(&s->syncp);
153 s->ustats.outbytes += skb->len;
154 u64_stats_update_end(&s->syncp);
156 s = this_cpu_ptr(dest->svc->stats.cpustats);
158 u64_stats_update_begin(&s->syncp);
159 s->ustats.outbytes += skb->len;
160 u64_stats_update_end(&s->syncp);
162 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
164 u64_stats_update_begin(&s->syncp);
165 s->ustats.outbytes += skb->len;
166 u64_stats_update_end(&s->syncp);
172 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
174 struct netns_ipvs *ipvs = net_ipvs(svc->net);
175 struct ip_vs_cpu_stats *s;
177 s = this_cpu_ptr(cp->dest->stats.cpustats);
180 s = this_cpu_ptr(svc->stats.cpustats);
183 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
189 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
190 const struct sk_buff *skb,
191 struct ip_vs_proto_data *pd)
193 if (likely(pd->pp->state_transition))
194 pd->pp->state_transition(cp, direction, skb, pd);
198 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
199 struct sk_buff *skb, int protocol,
200 const union nf_inet_addr *caddr, __be16 cport,
201 const union nf_inet_addr *vaddr, __be16 vport,
202 struct ip_vs_conn_param *p)
204 ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
207 if (p->pe && p->pe->fill_param)
208 return p->pe->fill_param(p, skb);
214 * IPVS persistent scheduling function
215 * It creates a connection entry according to its template if exists,
216 * or selects a server and creates a connection entry plus a template.
217 * Locking: we are svc user (svc->refcnt), so we hold all dests too
218 * Protocols supported: TCP, UDP
220 static struct ip_vs_conn *
221 ip_vs_sched_persist(struct ip_vs_service *svc,
222 struct sk_buff *skb, __be16 src_port, __be16 dst_port,
223 int *ignored, struct ip_vs_iphdr *iph)
225 struct ip_vs_conn *cp = NULL;
226 struct ip_vs_dest *dest;
227 struct ip_vs_conn *ct;
228 __be16 dport = 0; /* destination port to forward */
230 struct ip_vs_conn_param param;
231 const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
232 union nf_inet_addr snet; /* source network of the client,
235 /* Mask saddr with the netmask to adjust template granularity */
236 #ifdef CONFIG_IP_VS_IPV6
237 if (svc->af == AF_INET6)
238 ipv6_addr_prefix(&snet.in6, &iph->saddr.in6, svc->netmask);
241 snet.ip = iph->saddr.ip & svc->netmask;
243 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
245 IP_VS_DBG_ADDR(svc->af, &iph->saddr), ntohs(src_port),
246 IP_VS_DBG_ADDR(svc->af, &iph->daddr), ntohs(dst_port),
247 IP_VS_DBG_ADDR(svc->af, &snet));
250 * As far as we know, FTP is a very complicated network protocol, and
251 * it uses control connection and data connections. For active FTP,
252 * FTP server initialize data connection to the client, its source port
253 * is often 20. For passive FTP, FTP server tells the clients the port
254 * that it passively listens to, and the client issues the data
255 * connection. In the tunneling or direct routing mode, the load
256 * balancer is on the client-to-server half of connection, the port
257 * number is unknown to the load balancer. So, a conn template like
258 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
259 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
260 * is created for other persistent services.
263 int protocol = iph->protocol;
264 const union nf_inet_addr *vaddr = &iph->daddr;
267 if (dst_port == svc->port) {
269 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
271 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
273 if (svc->port != FTPPORT)
276 /* Note: persistent fwmark-based services and
277 * persistent port zero service are handled here.
279 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
280 * port zero template:
281 * <protocol,caddr,0,vaddr,0,daddr,0>
284 protocol = IPPROTO_IP;
288 /* return *ignored = -1 so NF_DROP can be used */
289 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
290 vaddr, vport, ¶m) < 0) {
296 /* Check if a template already exists */
297 ct = ip_vs_ct_in_get(¶m);
298 if (!ct || !ip_vs_check_template(ct)) {
300 * No template found or the dest of the connection
301 * template is not available.
302 * return *ignored=0 i.e. ICMP and NF_DROP
305 dest = svc->scheduler->schedule(svc, skb);
308 IP_VS_DBG(1, "p-schedule: no dest found.\n");
309 kfree(param.pe_data);
314 if (dst_port == svc->port && svc->port != FTPPORT)
318 * This adds param.pe_data to the template,
319 * and thus param.pe_data will be destroyed
320 * when the template expires */
321 ct = ip_vs_conn_new(¶m, &dest->addr, dport,
322 IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
325 kfree(param.pe_data);
330 ct->timeout = svc->timeout;
332 /* set destination with the found template */
334 kfree(param.pe_data);
338 if (dport == svc->port && dest->port)
341 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
342 && iph->protocol == IPPROTO_UDP) ?
343 IP_VS_CONN_F_ONE_PACKET : 0;
346 * Create a new connection according to the template
348 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol, &iph->saddr,
349 src_port, &iph->daddr, dst_port, ¶m);
351 cp = ip_vs_conn_new(¶m, &dest->addr, dport, flags, dest, skb->mark);
361 ip_vs_control_add(cp, ct);
364 ip_vs_conn_stats(cp, svc);
370 * IPVS main scheduling function
371 * It selects a server according to the virtual service, and
372 * creates a connection entry.
373 * Protocols supported: TCP, UDP
377 * 1 : protocol tried to schedule (eg. on SYN), found svc but the
378 * svc/scheduler decides that this packet should be accepted with
379 * NF_ACCEPT because it must not be scheduled.
381 * 0 : scheduler can not find destination, so try bypass or
382 * return ICMP and then NF_DROP (ip_vs_leave).
384 * -1 : scheduler tried to schedule but fatal error occurred, eg.
385 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
386 * failure such as missing Call-ID, ENOMEM on skb_linearize
387 * or pe_data. In this case we should return NF_DROP without
388 * any attempts to send ICMP with ip_vs_leave.
391 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
392 struct ip_vs_proto_data *pd, int *ignored,
393 struct ip_vs_iphdr *iph)
395 struct ip_vs_protocol *pp = pd->pp;
396 struct ip_vs_conn *cp = NULL;
397 struct ip_vs_dest *dest;
398 __be16 _ports[2], *pptr;
403 * IPv6 frags, only the first hit here.
405 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
410 * FTPDATA needs this check when using local real server.
411 * Never schedule Active FTPDATA connections from real server.
412 * For LVS-NAT they must be already created. For other methods
413 * with persistence the connection is created on SYN+ACK.
415 if (pptr[0] == FTPDATA) {
416 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
417 "Not scheduling FTPDATA");
422 * Do not schedule replies from local real server.
424 if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
425 (cp = pp->conn_in_get(svc->af, skb, iph, 1))) {
426 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
427 "Not scheduling reply for existing connection");
428 __ip_vs_conn_put(cp);
435 if (svc->flags & IP_VS_SVC_F_PERSISTENT)
436 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored,
442 * Non-persistent service
444 if (!svc->fwmark && pptr[1] != svc->port) {
446 pr_err("Schedule: port zero only supported "
447 "in persistent services, "
448 "check your ipvs configuration\n");
453 dest = svc->scheduler->schedule(svc, skb);
456 IP_VS_DBG(1, "Schedule: no dest found.\n");
460 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
461 && iph->protocol == IPPROTO_UDP) ?
462 IP_VS_CONN_F_ONE_PACKET : 0;
465 * Create a connection entry.
468 struct ip_vs_conn_param p;
470 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
471 &iph->saddr, pptr[0], &iph->daddr,
473 cp = ip_vs_conn_new(&p, &dest->addr,
474 dest->port ? dest->port : pptr[1],
475 flags, dest, skb->mark);
483 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
484 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
486 IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
487 IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
488 IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
489 cp->flags, atomic_read(&cp->refcnt));
491 ip_vs_conn_stats(cp, svc);
497 * Pass or drop the packet.
498 * Called by ip_vs_in, when the virtual service is available but
499 * no destination is available for a new connection.
501 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
502 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
504 __be16 _ports[2], *pptr;
507 struct netns_ipvs *ipvs;
511 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
513 ip_vs_service_put(svc);
520 #ifdef CONFIG_IP_VS_IPV6
521 if (svc->af == AF_INET6)
522 unicast = ipv6_addr_type(&iph->daddr.in6) & IPV6_ADDR_UNICAST;
525 unicast = (inet_addr_type(net, iph->daddr.ip) == RTN_UNICAST);
527 /* if it is fwmark-based service, the cache_bypass sysctl is up
528 and the destination is a non-local unicast, then create
529 a cache_bypass connection entry */
530 ipvs = net_ipvs(net);
531 if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
533 struct ip_vs_conn *cp;
534 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
535 iph->protocol == IPPROTO_UDP) ?
536 IP_VS_CONN_F_ONE_PACKET : 0;
537 union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } };
539 ip_vs_service_put(svc);
541 /* create a new connection entry */
542 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
544 struct ip_vs_conn_param p;
545 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
546 &iph->saddr, pptr[0],
547 &iph->daddr, pptr[1], &p);
548 cp = ip_vs_conn_new(&p, &daddr, 0,
549 IP_VS_CONN_F_BYPASS | flags,
556 ip_vs_in_stats(cp, skb);
559 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
561 /* transmit the first SYN packet */
562 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
563 /* do not touch skb anymore */
565 atomic_inc(&cp->in_pkts);
572 * When the virtual ftp service is presented, packets destined
573 * for other services on the VIP may get here (except services
574 * listed in the ipvs table), pass the packets, because it is
575 * not ipvs job to decide to drop the packets.
577 if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
578 ip_vs_service_put(svc);
582 ip_vs_service_put(svc);
585 * Notify the client that the destination is unreachable, and
586 * release the socket buffer.
587 * Since it is in IP layer, the TCP socket is not actually
588 * created, the TCP RST packet cannot be sent, instead that
589 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
591 #ifdef CONFIG_IP_VS_IPV6
592 if (svc->af == AF_INET6) {
594 struct net *net = dev_net(skb_dst(skb)->dev);
596 skb->dev = net->loopback_dev;
598 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
601 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
608 static int sysctl_snat_reroute(struct sk_buff *skb)
610 struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
611 return ipvs->sysctl_snat_reroute;
614 static int sysctl_nat_icmp_send(struct net *net)
616 struct netns_ipvs *ipvs = net_ipvs(net);
617 return ipvs->sysctl_nat_icmp_send;
620 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
622 return ipvs->sysctl_expire_nodest_conn;
627 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
628 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
629 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
633 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
635 return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
638 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
640 if (NF_INET_LOCAL_IN == hooknum)
641 return IP_DEFRAG_VS_IN;
642 if (NF_INET_FORWARD == hooknum)
643 return IP_DEFRAG_VS_FWD;
644 return IP_DEFRAG_VS_OUT;
647 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
649 int err = ip_defrag(skb, user);
652 ip_send_check(ip_hdr(skb));
657 static int ip_vs_route_me_harder(int af, struct sk_buff *skb)
659 #ifdef CONFIG_IP_VS_IPV6
660 if (af == AF_INET6) {
661 if (sysctl_snat_reroute(skb) && ip6_route_me_harder(skb) != 0)
665 if ((sysctl_snat_reroute(skb) ||
666 skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
667 ip_route_me_harder(skb, RTN_LOCAL) != 0)
674 * Packet has been made sufficiently writable in caller
675 * - inout: 1=in->out, 0=out->in
677 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
678 struct ip_vs_conn *cp, int inout)
680 struct iphdr *iph = ip_hdr(skb);
681 unsigned int icmp_offset = iph->ihl*4;
682 struct icmphdr *icmph = (struct icmphdr *)(skb_network_header(skb) +
684 struct iphdr *ciph = (struct iphdr *)(icmph + 1);
687 iph->saddr = cp->vaddr.ip;
689 ciph->daddr = cp->vaddr.ip;
692 iph->daddr = cp->daddr.ip;
694 ciph->saddr = cp->daddr.ip;
698 /* the TCP/UDP/SCTP port */
699 if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
700 IPPROTO_SCTP == ciph->protocol) {
701 __be16 *ports = (void *)ciph + ciph->ihl*4;
704 ports[1] = cp->vport;
706 ports[0] = cp->dport;
709 /* And finally the ICMP checksum */
711 icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
712 skb->ip_summed = CHECKSUM_UNNECESSARY;
715 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
716 "Forwarding altered outgoing ICMP");
718 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
719 "Forwarding altered incoming ICMP");
722 #ifdef CONFIG_IP_VS_IPV6
723 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
724 struct ip_vs_conn *cp, int inout)
726 struct ipv6hdr *iph = ipv6_hdr(skb);
727 unsigned int icmp_offset = 0;
728 unsigned int offs = 0; /* header offset*/
730 struct icmp6hdr *icmph;
731 struct ipv6hdr *ciph;
732 unsigned short fragoffs;
734 ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
735 icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
736 offs = icmp_offset + sizeof(struct icmp6hdr);
737 ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
739 protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
742 iph->saddr = cp->vaddr.in6;
743 ciph->daddr = cp->vaddr.in6;
745 iph->daddr = cp->daddr.in6;
746 ciph->saddr = cp->daddr.in6;
749 /* the TCP/UDP/SCTP port */
750 if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
751 IPPROTO_SCTP == protocol)) {
752 __be16 *ports = (void *)(skb_network_header(skb) + offs);
754 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
755 ntohs(inout ? ports[1] : ports[0]),
756 ntohs(inout ? cp->vport : cp->dport));
758 ports[1] = cp->vport;
760 ports[0] = cp->dport;
763 /* And finally the ICMP checksum */
764 icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
765 skb->len - icmp_offset,
767 skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
768 skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
769 skb->ip_summed = CHECKSUM_PARTIAL;
772 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
773 (void *)ciph - (void *)iph,
774 "Forwarding altered outgoing ICMPv6");
776 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
777 (void *)ciph - (void *)iph,
778 "Forwarding altered incoming ICMPv6");
782 /* Handle relevant response ICMP messages - forward to the right
785 static int handle_response_icmp(int af, struct sk_buff *skb,
786 union nf_inet_addr *snet,
787 __u8 protocol, struct ip_vs_conn *cp,
788 struct ip_vs_protocol *pp,
789 unsigned int offset, unsigned int ihl)
791 unsigned int verdict = NF_DROP;
793 if (IP_VS_FWD_METHOD(cp) != 0) {
794 pr_err("shouldn't reach here, because the box is on the "
795 "half connection in the tun/dr module.\n");
798 /* Ensure the checksum is correct */
799 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
800 /* Failed checksum! */
801 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
802 IP_VS_DBG_ADDR(af, snet));
806 if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
807 IPPROTO_SCTP == protocol)
808 offset += 2 * sizeof(__u16);
809 if (!skb_make_writable(skb, offset))
812 #ifdef CONFIG_IP_VS_IPV6
814 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
817 ip_vs_nat_icmp(skb, pp, cp, 1);
819 if (ip_vs_route_me_harder(af, skb))
822 /* do the statistics and put it back */
823 ip_vs_out_stats(cp, skb);
825 skb->ipvs_property = 1;
826 if (!(cp->flags & IP_VS_CONN_F_NFCT))
829 ip_vs_update_conntrack(skb, cp, 0);
833 __ip_vs_conn_put(cp);
839 * Handle ICMP messages in the inside-to-outside direction (outgoing).
840 * Find any that might be relevant, check against existing connections.
841 * Currently handles error types - unreachable, quench, ttl exceeded.
843 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
844 unsigned int hooknum)
847 struct icmphdr _icmph, *ic;
848 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
849 struct ip_vs_iphdr ciph;
850 struct ip_vs_conn *cp;
851 struct ip_vs_protocol *pp;
852 unsigned int offset, ihl;
853 union nf_inet_addr snet;
857 /* reassemble IP fragments */
858 if (ip_is_fragment(ip_hdr(skb))) {
859 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
864 offset = ihl = iph->ihl * 4;
865 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
869 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
870 ic->type, ntohs(icmp_id(ic)),
871 &iph->saddr, &iph->daddr);
874 * Work through seeing if this is for us.
875 * These checks are supposed to be in an order that means easy
876 * things are checked first to speed up processing.... however
877 * this means that some packets will manage to get a long way
878 * down this stack and then be rejected, but that's life.
880 if ((ic->type != ICMP_DEST_UNREACH) &&
881 (ic->type != ICMP_SOURCE_QUENCH) &&
882 (ic->type != ICMP_TIME_EXCEEDED)) {
887 /* Now find the contained IP header */
888 offset += sizeof(_icmph);
889 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
891 return NF_ACCEPT; /* The packet looks wrong, ignore */
893 pp = ip_vs_proto_get(cih->protocol);
897 /* Is the embedded protocol header present? */
898 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
902 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
903 "Checking outgoing ICMP for");
905 ip_vs_fill_ip4hdr(cih, &ciph);
907 /* The embedded headers contain source and dest in reverse order */
908 cp = pp->conn_out_get(AF_INET, skb, &ciph, 1);
912 snet.ip = iph->saddr;
913 return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
917 #ifdef CONFIG_IP_VS_IPV6
918 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
919 unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
921 struct icmp6hdr _icmph, *ic;
922 struct ipv6hdr _ip6h, *ip6h; /* The ip header contained within ICMP */
923 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
924 struct ip_vs_conn *cp;
925 struct ip_vs_protocol *pp;
926 union nf_inet_addr snet;
927 unsigned int writable;
930 ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
935 * Work through seeing if this is for us.
936 * These checks are supposed to be in an order that means easy
937 * things are checked first to speed up processing.... however
938 * this means that some packets will manage to get a long way
939 * down this stack and then be rejected, but that's life.
941 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
945 /* Fragment header that is before ICMP header tells us that:
946 * it's not an error message since they can't be fragmented.
948 if (ipvsh->flags & IP6_FH_F_FRAG)
951 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
952 ic->icmp6_type, ntohs(icmpv6_id(ic)),
953 &ipvsh->saddr, &ipvsh->daddr);
955 /* Now find the contained IP header */
956 ciph.len = ipvsh->len + sizeof(_icmph);
957 ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
959 return NF_ACCEPT; /* The packet looks wrong, ignore */
960 ciph.saddr.in6 = ip6h->saddr; /* conn_out_get() handles reverse order */
961 ciph.daddr.in6 = ip6h->daddr;
962 /* skip possible IPv6 exthdrs of contained IPv6 packet */
963 ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
964 if (ciph.protocol < 0)
965 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
967 pp = ip_vs_proto_get(ciph.protocol);
971 /* The embedded headers contain source and dest in reverse order */
972 cp = pp->conn_out_get(AF_INET6, skb, &ciph, 1);
976 snet.in6 = ciph.saddr.in6;
978 return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
979 pp, writable, sizeof(struct ipv6hdr));
984 * Check if sctp chunc is ABORT chunk
986 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
988 sctp_chunkhdr_t *sch, schunk;
989 sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
990 sizeof(schunk), &schunk);
993 if (sch->type == SCTP_CID_ABORT)
998 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1000 struct tcphdr _tcph, *th;
1002 th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1008 /* Handle response packets: rewrite addresses and send away...
1011 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1012 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
1014 struct ip_vs_protocol *pp = pd->pp;
1016 IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1018 if (!skb_make_writable(skb, iph->len))
1021 /* mangle the packet */
1022 if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1025 #ifdef CONFIG_IP_VS_IPV6
1027 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1031 ip_hdr(skb)->saddr = cp->vaddr.ip;
1032 ip_send_check(ip_hdr(skb));
1036 * nf_iterate does not expect change in the skb->dst->dev.
1037 * It looks like it is not fatal to enable this code for hooks
1038 * where our handlers are at the end of the chain list and
1039 * when all next handlers use skb->dst->dev and not outdev.
1040 * It will definitely route properly the inout NAT traffic
1041 * when multiple paths are used.
1044 /* For policy routing, packets originating from this
1045 * machine itself may be routed differently to packets
1046 * passing through. We want this packet to be routed as
1047 * if it came from this machine itself. So re-compute
1048 * the routing information.
1050 if (ip_vs_route_me_harder(af, skb))
1053 IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1055 ip_vs_out_stats(cp, skb);
1056 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1057 skb->ipvs_property = 1;
1058 if (!(cp->flags & IP_VS_CONN_F_NFCT))
1061 ip_vs_update_conntrack(skb, cp, 0);
1075 * Check if outgoing packet belongs to the established ip_vs_conn.
1078 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1080 struct net *net = NULL;
1081 struct ip_vs_iphdr iph;
1082 struct ip_vs_protocol *pp;
1083 struct ip_vs_proto_data *pd;
1084 struct ip_vs_conn *cp;
1088 /* Already marked as IPVS request or reply? */
1089 if (skb->ipvs_property)
1092 /* Bad... Do not break raw sockets */
1093 if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1095 struct sock *sk = skb->sk;
1096 struct inet_sock *inet = inet_sk(skb->sk);
1098 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1102 if (unlikely(!skb_dst(skb)))
1106 if (!net_ipvs(net)->enable)
1109 ip_vs_fill_iph_skb(af, skb, &iph);
1110 #ifdef CONFIG_IP_VS_IPV6
1111 if (af == AF_INET6) {
1112 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1113 struct sk_buff *reasm = skb_nfct_reasm(skb);
1114 /* Save fw mark for coming frags */
1115 reasm->ipvs_property = 1;
1116 reasm->mark = skb->mark;
1118 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1120 int verdict = ip_vs_out_icmp_v6(skb, &related,
1128 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1130 int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1136 pd = ip_vs_proto_data_get(net, iph.protocol);
1141 /* reassemble IP fragments */
1142 #ifdef CONFIG_IP_VS_IPV6
1145 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1146 if (ip_vs_gather_frags(skb,
1147 ip_vs_defrag_user(hooknum)))
1150 ip_vs_fill_ip4hdr(skb_network_header(skb), &iph);
1154 * Check if the packet belongs to an existing entry
1156 cp = pp->conn_out_get(af, skb, &iph, 0);
1159 return handle_response(af, skb, pd, cp, &iph);
1160 if (sysctl_nat_icmp_send(net) &&
1161 (pp->protocol == IPPROTO_TCP ||
1162 pp->protocol == IPPROTO_UDP ||
1163 pp->protocol == IPPROTO_SCTP)) {
1164 __be16 _ports[2], *pptr;
1166 pptr = frag_safe_skb_hp(skb, iph.len,
1167 sizeof(_ports), _ports, &iph);
1169 return NF_ACCEPT; /* Not for me */
1170 if (ip_vs_has_real_service(net, af, iph.protocol, &iph.saddr,
1173 * Notify the real server: there is no
1174 * existing entry if it is not RST
1175 * packet or not TCP packet.
1177 if ((iph.protocol != IPPROTO_TCP &&
1178 iph.protocol != IPPROTO_SCTP)
1179 || ((iph.protocol == IPPROTO_TCP
1180 && !is_tcp_reset(skb, iph.len))
1181 || (iph.protocol == IPPROTO_SCTP
1182 && !is_sctp_abort(skb,
1184 #ifdef CONFIG_IP_VS_IPV6
1185 if (af == AF_INET6) {
1187 skb->dev = net->loopback_dev;
1189 ICMPV6_DEST_UNREACH,
1190 ICMPV6_PORT_UNREACH,
1196 ICMP_PORT_UNREACH, 0);
1201 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1202 "ip_vs_out: packet continues traversal as normal");
1207 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1208 * used only for VS/NAT.
1209 * Check if packet is reply for established ip_vs_conn.
1212 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1213 const struct net_device *in, const struct net_device *out,
1214 int (*okfn)(struct sk_buff *))
1216 return ip_vs_out(hooknum, skb, AF_INET);
1220 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1221 * Check if packet is reply for established ip_vs_conn.
1224 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1225 const struct net_device *in, const struct net_device *out,
1226 int (*okfn)(struct sk_buff *))
1228 unsigned int verdict;
1230 /* Disable BH in LOCAL_OUT until all places are fixed */
1232 verdict = ip_vs_out(hooknum, skb, AF_INET);
1237 #ifdef CONFIG_IP_VS_IPV6
1240 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1241 * used only for VS/NAT.
1242 * Check if packet is reply for established ip_vs_conn.
1245 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1246 const struct net_device *in, const struct net_device *out,
1247 int (*okfn)(struct sk_buff *))
1249 return ip_vs_out(hooknum, skb, AF_INET6);
1253 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1254 * Check if packet is reply for established ip_vs_conn.
1257 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1258 const struct net_device *in, const struct net_device *out,
1259 int (*okfn)(struct sk_buff *))
1261 unsigned int verdict;
1263 /* Disable BH in LOCAL_OUT until all places are fixed */
1265 verdict = ip_vs_out(hooknum, skb, AF_INET6);
1273 * Handle ICMP messages in the outside-to-inside direction (incoming).
1274 * Find any that might be relevant, check against existing connections,
1275 * forward to the right destination host if relevant.
1276 * Currently handles error types - unreachable, quench, ttl exceeded.
1279 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1281 struct net *net = NULL;
1283 struct icmphdr _icmph, *ic;
1284 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
1285 struct ip_vs_iphdr ciph;
1286 struct ip_vs_conn *cp;
1287 struct ip_vs_protocol *pp;
1288 struct ip_vs_proto_data *pd;
1289 unsigned int offset, offset2, ihl, verdict;
1294 /* reassemble IP fragments */
1295 if (ip_is_fragment(ip_hdr(skb))) {
1296 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1301 offset = ihl = iph->ihl * 4;
1302 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1306 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1307 ic->type, ntohs(icmp_id(ic)),
1308 &iph->saddr, &iph->daddr);
1311 * Work through seeing if this is for us.
1312 * These checks are supposed to be in an order that means easy
1313 * things are checked first to speed up processing.... however
1314 * this means that some packets will manage to get a long way
1315 * down this stack and then be rejected, but that's life.
1317 if ((ic->type != ICMP_DEST_UNREACH) &&
1318 (ic->type != ICMP_SOURCE_QUENCH) &&
1319 (ic->type != ICMP_TIME_EXCEEDED)) {
1324 /* Now find the contained IP header */
1325 offset += sizeof(_icmph);
1326 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1328 return NF_ACCEPT; /* The packet looks wrong, ignore */
1332 /* Special case for errors for IPIP packets */
1334 if (cih->protocol == IPPROTO_IPIP) {
1335 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1337 /* Error for our IPIP must arrive at LOCAL_IN */
1338 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1340 offset += cih->ihl * 4;
1341 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1343 return NF_ACCEPT; /* The packet looks wrong, ignore */
1347 pd = ip_vs_proto_data_get(net, cih->protocol);
1352 /* Is the embedded protocol header present? */
1353 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1357 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1358 "Checking incoming ICMP for");
1361 ip_vs_fill_ip4hdr(cih, &ciph);
1364 /* The embedded headers contain source and dest in reverse order.
1365 * For IPIP this is error for request, not for reply.
1367 cp = pp->conn_in_get(AF_INET, skb, &ciph, ipip ? 0 : 1);
1373 /* Ensure the checksum is correct */
1374 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1375 /* Failed checksum! */
1376 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1382 __be32 info = ic->un.gateway;
1384 /* Update the MTU */
1385 if (ic->type == ICMP_DEST_UNREACH &&
1386 ic->code == ICMP_FRAG_NEEDED) {
1387 struct ip_vs_dest *dest = cp->dest;
1388 u32 mtu = ntohs(ic->un.frag.mtu);
1390 /* Strip outer IP and ICMP, go to IPIP header */
1391 __skb_pull(skb, ihl + sizeof(_icmph));
1392 offset2 -= ihl + sizeof(_icmph);
1393 skb_reset_network_header(skb);
1394 IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1395 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1396 ipv4_update_pmtu(skb, dev_net(skb->dev),
1398 /* Client uses PMTUD? */
1399 if (!(cih->frag_off & htons(IP_DF)))
1401 /* Prefer the resulting PMTU */
1403 struct ip_vs_dest_dst *dest_dst;
1406 dest_dst = rcu_dereference(dest->dest_dst);
1408 mtu = dst_mtu(dest_dst->dst_cache);
1411 if (mtu > 68 + sizeof(struct iphdr))
1412 mtu -= sizeof(struct iphdr);
1415 /* Strip outer IP, ICMP and IPIP, go to IP header of
1418 __skb_pull(skb, offset2);
1419 skb_reset_network_header(skb);
1420 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1421 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1422 ic->type, ic->code, ntohl(info));
1423 icmp_send(skb, ic->type, ic->code, info);
1424 /* ICMP can be shorter but anyways, account it */
1425 ip_vs_out_stats(cp, skb);
1429 verdict = NF_STOLEN;
1433 /* do the statistics and put it back */
1434 ip_vs_in_stats(cp, skb);
1435 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1436 offset += 2 * sizeof(__u16);
1437 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1440 __ip_vs_conn_put(cp);
1445 #ifdef CONFIG_IP_VS_IPV6
1446 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1447 unsigned int hooknum, struct ip_vs_iphdr *iph)
1449 struct net *net = NULL;
1450 struct ipv6hdr _ip6h, *ip6h;
1451 struct icmp6hdr _icmph, *ic;
1452 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1453 struct ip_vs_conn *cp;
1454 struct ip_vs_protocol *pp;
1455 struct ip_vs_proto_data *pd;
1456 unsigned int offs_ciph, writable, verdict;
1460 ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1465 * Work through seeing if this is for us.
1466 * These checks are supposed to be in an order that means easy
1467 * things are checked first to speed up processing.... however
1468 * this means that some packets will manage to get a long way
1469 * down this stack and then be rejected, but that's life.
1471 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1475 /* Fragment header that is before ICMP header tells us that:
1476 * it's not an error message since they can't be fragmented.
1478 if (iph->flags & IP6_FH_F_FRAG)
1481 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1482 ic->icmp6_type, ntohs(icmpv6_id(ic)),
1483 &iph->saddr, &iph->daddr);
1485 /* Now find the contained IP header */
1486 ciph.len = iph->len + sizeof(_icmph);
1487 offs_ciph = ciph.len; /* Save ip header offset */
1488 ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
1490 return NF_ACCEPT; /* The packet looks wrong, ignore */
1491 ciph.saddr.in6 = ip6h->saddr; /* conn_in_get() handles reverse order */
1492 ciph.daddr.in6 = ip6h->daddr;
1493 /* skip possible IPv6 exthdrs of contained IPv6 packet */
1494 ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
1495 if (ciph.protocol < 0)
1496 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
1499 pd = ip_vs_proto_data_get(net, ciph.protocol);
1504 /* Cannot handle fragmented embedded protocol */
1508 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offs_ciph,
1509 "Checking incoming ICMPv6 for");
1511 /* The embedded headers contain source and dest in reverse order
1512 * if not from localhost
1514 cp = pp->conn_in_get(AF_INET6, skb, &ciph,
1515 (hooknum == NF_INET_LOCAL_OUT) ? 0 : 1);
1519 /* VS/TUN, VS/DR and LOCALNODE just let it go */
1520 if ((hooknum == NF_INET_LOCAL_OUT) &&
1521 (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1522 __ip_vs_conn_put(cp);
1526 /* do the statistics and put it back */
1527 ip_vs_in_stats(cp, skb);
1529 /* Need to mangle contained IPv6 header in ICMPv6 packet */
1530 writable = ciph.len;
1531 if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1532 IPPROTO_SCTP == ciph.protocol)
1533 writable += 2 * sizeof(__u16); /* Also mangle ports */
1535 verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, writable, hooknum, &ciph);
1537 __ip_vs_conn_put(cp);
1545 * Check if it's for virtual services, look it up,
1546 * and send it on its way...
1549 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1552 struct ip_vs_iphdr iph;
1553 struct ip_vs_protocol *pp;
1554 struct ip_vs_proto_data *pd;
1555 struct ip_vs_conn *cp;
1557 struct netns_ipvs *ipvs;
1559 /* Already marked as IPVS request or reply? */
1560 if (skb->ipvs_property)
1565 * - remote client: only PACKET_HOST
1566 * - route: used for struct net when skb->dev is unset
1568 if (unlikely((skb->pkt_type != PACKET_HOST &&
1569 hooknum != NF_INET_LOCAL_OUT) ||
1571 ip_vs_fill_iph_skb(af, skb, &iph);
1572 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1573 " ignored in hook %u\n",
1574 skb->pkt_type, iph.protocol,
1575 IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1578 /* ipvs enabled in this netns ? */
1580 ipvs = net_ipvs(net);
1581 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1584 ip_vs_fill_iph_skb(af, skb, &iph);
1586 /* Bad... Do not break raw sockets */
1587 if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1589 struct sock *sk = skb->sk;
1590 struct inet_sock *inet = inet_sk(skb->sk);
1592 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1596 #ifdef CONFIG_IP_VS_IPV6
1597 if (af == AF_INET6) {
1598 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1599 struct sk_buff *reasm = skb_nfct_reasm(skb);
1600 /* Save fw mark for coming frags. */
1601 reasm->ipvs_property = 1;
1602 reasm->mark = skb->mark;
1604 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1606 int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1614 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1616 int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1622 /* Protocol supported? */
1623 pd = ip_vs_proto_data_get(net, iph.protocol);
1628 * Check if the packet belongs to an existing connection entry
1630 cp = pp->conn_in_get(af, skb, &iph, 0);
1631 if (unlikely(!cp) && !iph.fragoffs) {
1632 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1633 * replayed fragment zero will already have created the cp
1637 /* Schedule and create new connection entry into &cp */
1638 if (!pp->conn_schedule(af, skb, pd, &v, &cp, &iph))
1642 if (unlikely(!cp)) {
1643 /* sorry, all this trouble for a no-hit :) */
1644 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1645 "ip_vs_in: packet continues traversal as normal");
1646 if (iph.fragoffs && !skb_nfct_reasm(skb)) {
1647 /* Fragment that couldn't be mapped to a conn entry
1648 * and don't have any pointer to a reasm skb
1649 * is missing module nf_defrag_ipv6
1651 IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1652 IP_VS_DBG_PKT(7, af, pp, skb, 0, "unhandled fragment");
1657 IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1658 /* Check the server status */
1659 if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1660 /* the destination server is not available */
1662 if (sysctl_expire_nodest_conn(ipvs)) {
1663 /* try to expire the connection immediately */
1664 ip_vs_conn_expire_now(cp);
1666 /* don't restart its timer, and silently
1668 __ip_vs_conn_put(cp);
1672 ip_vs_in_stats(cp, skb);
1673 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1674 if (cp->packet_xmit)
1675 ret = cp->packet_xmit(skb, cp, pp, &iph);
1676 /* do not touch skb anymore */
1678 IP_VS_DBG_RL("warning: packet_xmit is null");
1682 /* Increase its packet counter and check if it is needed
1683 * to be synchronized
1685 * Sync connection if it is about to close to
1686 * encorage the standby servers to update the connections timeout
1688 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1691 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1692 pkts = sysctl_sync_threshold(ipvs);
1694 pkts = atomic_add_return(1, &cp->in_pkts);
1696 if (ipvs->sync_state & IP_VS_STATE_MASTER)
1697 ip_vs_sync_conn(net, cp, pkts);
1704 * AF_INET handler in NF_INET_LOCAL_IN chain
1705 * Schedule and forward packets from remote clients
1708 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1709 const struct net_device *in,
1710 const struct net_device *out,
1711 int (*okfn)(struct sk_buff *))
1713 return ip_vs_in(hooknum, skb, AF_INET);
1717 * AF_INET handler in NF_INET_LOCAL_OUT chain
1718 * Schedule and forward packets from local clients
1721 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1722 const struct net_device *in, const struct net_device *out,
1723 int (*okfn)(struct sk_buff *))
1725 unsigned int verdict;
1727 /* Disable BH in LOCAL_OUT until all places are fixed */
1729 verdict = ip_vs_in(hooknum, skb, AF_INET);
1734 #ifdef CONFIG_IP_VS_IPV6
1737 * AF_INET6 fragment handling
1738 * Copy info from first fragment, to the rest of them.
1741 ip_vs_preroute_frag6(unsigned int hooknum, struct sk_buff *skb,
1742 const struct net_device *in,
1743 const struct net_device *out,
1744 int (*okfn)(struct sk_buff *))
1746 struct sk_buff *reasm = skb_nfct_reasm(skb);
1749 /* Skip if not a "replay" from nf_ct_frag6_output or first fragment.
1750 * ipvs_property is set when checking first fragment
1751 * in ip_vs_in() and ip_vs_out().
1754 IP_VS_DBG(2, "Fragment recv prop:%d\n", reasm->ipvs_property);
1755 if (!reasm || !reasm->ipvs_property)
1759 if (!net_ipvs(net)->enable)
1762 /* Copy stored fw mark, saved in ip_vs_{in,out} */
1763 skb->mark = reasm->mark;
1769 * AF_INET6 handler in NF_INET_LOCAL_IN chain
1770 * Schedule and forward packets from remote clients
1773 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1774 const struct net_device *in,
1775 const struct net_device *out,
1776 int (*okfn)(struct sk_buff *))
1778 return ip_vs_in(hooknum, skb, AF_INET6);
1782 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
1783 * Schedule and forward packets from local clients
1786 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1787 const struct net_device *in, const struct net_device *out,
1788 int (*okfn)(struct sk_buff *))
1790 unsigned int verdict;
1792 /* Disable BH in LOCAL_OUT until all places are fixed */
1794 verdict = ip_vs_in(hooknum, skb, AF_INET6);
1803 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1804 * related packets destined for 0.0.0.0/0.
1805 * When fwmark-based virtual service is used, such as transparent
1806 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1807 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1808 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1809 * and send them to ip_vs_in_icmp.
1812 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1813 const struct net_device *in, const struct net_device *out,
1814 int (*okfn)(struct sk_buff *))
1818 struct netns_ipvs *ipvs;
1820 if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1823 /* ipvs enabled in this netns ? */
1825 ipvs = net_ipvs(net);
1826 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1829 return ip_vs_in_icmp(skb, &r, hooknum);
1832 #ifdef CONFIG_IP_VS_IPV6
1834 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1835 const struct net_device *in, const struct net_device *out,
1836 int (*okfn)(struct sk_buff *))
1840 struct netns_ipvs *ipvs;
1841 struct ip_vs_iphdr iphdr;
1843 ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
1844 if (iphdr.protocol != IPPROTO_ICMPV6)
1847 /* ipvs enabled in this netns ? */
1849 ipvs = net_ipvs(net);
1850 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1853 return ip_vs_in_icmp_v6(skb, &r, hooknum, &iphdr);
1858 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1859 /* After packet filtering, change source only for VS/NAT */
1861 .hook = ip_vs_reply4,
1862 .owner = THIS_MODULE,
1864 .hooknum = NF_INET_LOCAL_IN,
1865 .priority = NF_IP_PRI_NAT_SRC - 2,
1867 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1868 * or VS/NAT(change destination), so that filtering rules can be
1869 * applied to IPVS. */
1871 .hook = ip_vs_remote_request4,
1872 .owner = THIS_MODULE,
1874 .hooknum = NF_INET_LOCAL_IN,
1875 .priority = NF_IP_PRI_NAT_SRC - 1,
1877 /* Before ip_vs_in, change source only for VS/NAT */
1879 .hook = ip_vs_local_reply4,
1880 .owner = THIS_MODULE,
1882 .hooknum = NF_INET_LOCAL_OUT,
1883 .priority = NF_IP_PRI_NAT_DST + 1,
1885 /* After mangle, schedule and forward local requests */
1887 .hook = ip_vs_local_request4,
1888 .owner = THIS_MODULE,
1890 .hooknum = NF_INET_LOCAL_OUT,
1891 .priority = NF_IP_PRI_NAT_DST + 2,
1893 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1894 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1896 .hook = ip_vs_forward_icmp,
1897 .owner = THIS_MODULE,
1899 .hooknum = NF_INET_FORWARD,
1902 /* After packet filtering, change source only for VS/NAT */
1904 .hook = ip_vs_reply4,
1905 .owner = THIS_MODULE,
1907 .hooknum = NF_INET_FORWARD,
1910 #ifdef CONFIG_IP_VS_IPV6
1911 /* After mangle & nat fetch 2:nd fragment and following */
1913 .hook = ip_vs_preroute_frag6,
1914 .owner = THIS_MODULE,
1916 .hooknum = NF_INET_PRE_ROUTING,
1917 .priority = NF_IP6_PRI_NAT_DST + 1,
1919 /* After packet filtering, change source only for VS/NAT */
1921 .hook = ip_vs_reply6,
1922 .owner = THIS_MODULE,
1924 .hooknum = NF_INET_LOCAL_IN,
1925 .priority = NF_IP6_PRI_NAT_SRC - 2,
1927 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1928 * or VS/NAT(change destination), so that filtering rules can be
1929 * applied to IPVS. */
1931 .hook = ip_vs_remote_request6,
1932 .owner = THIS_MODULE,
1934 .hooknum = NF_INET_LOCAL_IN,
1935 .priority = NF_IP6_PRI_NAT_SRC - 1,
1937 /* Before ip_vs_in, change source only for VS/NAT */
1939 .hook = ip_vs_local_reply6,
1940 .owner = THIS_MODULE,
1942 .hooknum = NF_INET_LOCAL_OUT,
1943 .priority = NF_IP6_PRI_NAT_DST + 1,
1945 /* After mangle, schedule and forward local requests */
1947 .hook = ip_vs_local_request6,
1948 .owner = THIS_MODULE,
1950 .hooknum = NF_INET_LOCAL_OUT,
1951 .priority = NF_IP6_PRI_NAT_DST + 2,
1953 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1954 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1956 .hook = ip_vs_forward_icmp_v6,
1957 .owner = THIS_MODULE,
1959 .hooknum = NF_INET_FORWARD,
1962 /* After packet filtering, change source only for VS/NAT */
1964 .hook = ip_vs_reply6,
1965 .owner = THIS_MODULE,
1967 .hooknum = NF_INET_FORWARD,
1973 * Initialize IP Virtual Server netns mem.
1975 static int __net_init __ip_vs_init(struct net *net)
1977 struct netns_ipvs *ipvs;
1979 ipvs = net_generic(net, ip_vs_net_id);
1983 /* Hold the beast until a service is registerd */
1986 /* Counters used for creating unique names */
1987 ipvs->gen = atomic_read(&ipvs_netns_cnt);
1988 atomic_inc(&ipvs_netns_cnt);
1991 if (ip_vs_estimator_net_init(net) < 0)
1992 goto estimator_fail;
1994 if (ip_vs_control_net_init(net) < 0)
1997 if (ip_vs_protocol_net_init(net) < 0)
2000 if (ip_vs_app_net_init(net) < 0)
2003 if (ip_vs_conn_net_init(net) < 0)
2006 if (ip_vs_sync_net_init(net) < 0)
2009 printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2010 sizeof(struct netns_ipvs), ipvs->gen);
2017 ip_vs_conn_net_cleanup(net);
2019 ip_vs_app_net_cleanup(net);
2021 ip_vs_protocol_net_cleanup(net);
2023 ip_vs_control_net_cleanup(net);
2025 ip_vs_estimator_net_cleanup(net);
2031 static void __net_exit __ip_vs_cleanup(struct net *net)
2033 ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2034 ip_vs_conn_net_cleanup(net);
2035 ip_vs_app_net_cleanup(net);
2036 ip_vs_protocol_net_cleanup(net);
2037 ip_vs_control_net_cleanup(net);
2038 ip_vs_estimator_net_cleanup(net);
2039 IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2043 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2046 net_ipvs(net)->enable = 0; /* Disable packet reception */
2048 ip_vs_sync_net_cleanup(net);
2052 static struct pernet_operations ipvs_core_ops = {
2053 .init = __ip_vs_init,
2054 .exit = __ip_vs_cleanup,
2055 .id = &ip_vs_net_id,
2056 .size = sizeof(struct netns_ipvs),
2059 static struct pernet_operations ipvs_core_dev_ops = {
2060 .exit = __ip_vs_dev_cleanup,
2064 * Initialize IP Virtual Server
2066 static int __init ip_vs_init(void)
2070 ret = ip_vs_control_init();
2072 pr_err("can't setup control.\n");
2076 ip_vs_protocol_init();
2078 ret = ip_vs_conn_init();
2080 pr_err("can't setup connection table.\n");
2081 goto cleanup_protocol;
2084 ret = register_pernet_subsys(&ipvs_core_ops); /* Alloc ip_vs struct */
2088 ret = register_pernet_device(&ipvs_core_dev_ops);
2092 ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2094 pr_err("can't register hooks.\n");
2098 ret = ip_vs_register_nl_ioctl();
2100 pr_err("can't register netlink/ioctl.\n");
2104 pr_info("ipvs loaded.\n");
2109 nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2111 unregister_pernet_device(&ipvs_core_dev_ops);
2113 unregister_pernet_subsys(&ipvs_core_ops);
2115 ip_vs_conn_cleanup();
2117 ip_vs_protocol_cleanup();
2118 ip_vs_control_cleanup();
2123 static void __exit ip_vs_cleanup(void)
2125 ip_vs_unregister_nl_ioctl();
2126 nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2127 unregister_pernet_device(&ipvs_core_dev_ops);
2128 unregister_pernet_subsys(&ipvs_core_ops); /* free ip_vs struct */
2129 ip_vs_conn_cleanup();
2130 ip_vs_protocol_cleanup();
2131 ip_vs_control_cleanup();
2132 pr_info("ipvs unloaded.\n");
2135 module_init(ip_vs_init);
2136 module_exit(ip_vs_cleanup);
2137 MODULE_LICENSE("GPL");