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 int ip_vs_net_id __read_mostly;
73 #ifdef IP_VS_GENERIC_NETNS
74 EXPORT_SYMBOL(ip_vs_net_id);
76 /* netns cnt used for uniqueness */
77 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
79 /* ID used in ICMP lookups */
80 #define icmp_id(icmph) (((icmph)->un).echo.id)
81 #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier)
83 const char *ip_vs_proto_name(unsigned proto)
98 #ifdef CONFIG_IP_VS_IPV6
103 sprintf(buf, "IP_%d", proto);
108 void ip_vs_init_hash_table(struct list_head *table, int rows)
111 INIT_LIST_HEAD(&table[rows]);
115 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
117 struct ip_vs_dest *dest = cp->dest;
118 struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
120 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
121 struct ip_vs_cpu_stats *s;
123 s = this_cpu_ptr(dest->stats.cpustats);
125 u64_stats_update_begin(&s->syncp);
126 s->ustats.inbytes += skb->len;
127 u64_stats_update_end(&s->syncp);
129 s = this_cpu_ptr(dest->svc->stats.cpustats);
131 u64_stats_update_begin(&s->syncp);
132 s->ustats.inbytes += skb->len;
133 u64_stats_update_end(&s->syncp);
135 s = this_cpu_ptr(ipvs->cpustats);
137 u64_stats_update_begin(&s->syncp);
138 s->ustats.inbytes += skb->len;
139 u64_stats_update_end(&s->syncp);
145 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
147 struct ip_vs_dest *dest = cp->dest;
148 struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
150 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
151 struct ip_vs_cpu_stats *s;
153 s = this_cpu_ptr(dest->stats.cpustats);
155 u64_stats_update_begin(&s->syncp);
156 s->ustats.outbytes += skb->len;
157 u64_stats_update_end(&s->syncp);
159 s = this_cpu_ptr(dest->svc->stats.cpustats);
161 u64_stats_update_begin(&s->syncp);
162 s->ustats.outbytes += skb->len;
163 u64_stats_update_end(&s->syncp);
165 s = this_cpu_ptr(ipvs->cpustats);
167 u64_stats_update_begin(&s->syncp);
168 s->ustats.outbytes += skb->len;
169 u64_stats_update_end(&s->syncp);
175 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
177 struct netns_ipvs *ipvs = net_ipvs(svc->net);
178 struct ip_vs_cpu_stats *s;
180 s = this_cpu_ptr(cp->dest->stats.cpustats);
183 s = this_cpu_ptr(svc->stats.cpustats);
186 s = this_cpu_ptr(ipvs->cpustats);
192 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
193 const struct sk_buff *skb,
194 struct ip_vs_proto_data *pd)
196 if (unlikely(!pd->pp->state_transition))
198 return pd->pp->state_transition(cp, direction, skb, pd);
202 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
203 struct sk_buff *skb, int protocol,
204 const union nf_inet_addr *caddr, __be16 cport,
205 const union nf_inet_addr *vaddr, __be16 vport,
206 struct ip_vs_conn_param *p)
208 ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
211 if (p->pe && p->pe->fill_param)
212 return p->pe->fill_param(p, skb);
218 * IPVS persistent scheduling function
219 * It creates a connection entry according to its template if exists,
220 * or selects a server and creates a connection entry plus a template.
221 * Locking: we are svc user (svc->refcnt), so we hold all dests too
222 * Protocols supported: TCP, UDP
224 static struct ip_vs_conn *
225 ip_vs_sched_persist(struct ip_vs_service *svc,
227 __be16 src_port, __be16 dst_port, int *ignored)
229 struct ip_vs_conn *cp = NULL;
230 struct ip_vs_iphdr iph;
231 struct ip_vs_dest *dest;
232 struct ip_vs_conn *ct;
233 __be16 dport = 0; /* destination port to forward */
235 struct ip_vs_conn_param param;
236 union nf_inet_addr snet; /* source network of the client,
239 ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
241 /* Mask saddr with the netmask to adjust template granularity */
242 #ifdef CONFIG_IP_VS_IPV6
243 if (svc->af == AF_INET6)
244 ipv6_addr_prefix(&snet.in6, &iph.saddr.in6, svc->netmask);
247 snet.ip = iph.saddr.ip & svc->netmask;
249 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
251 IP_VS_DBG_ADDR(svc->af, &iph.saddr), ntohs(src_port),
252 IP_VS_DBG_ADDR(svc->af, &iph.daddr), ntohs(dst_port),
253 IP_VS_DBG_ADDR(svc->af, &snet));
256 * As far as we know, FTP is a very complicated network protocol, and
257 * it uses control connection and data connections. For active FTP,
258 * FTP server initialize data connection to the client, its source port
259 * is often 20. For passive FTP, FTP server tells the clients the port
260 * that it passively listens to, and the client issues the data
261 * connection. In the tunneling or direct routing mode, the load
262 * balancer is on the client-to-server half of connection, the port
263 * number is unknown to the load balancer. So, a conn template like
264 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
265 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
266 * is created for other persistent services.
269 int protocol = iph.protocol;
270 const union nf_inet_addr *vaddr = &iph.daddr;
271 const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
274 if (dst_port == svc->port) {
276 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
278 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
280 if (svc->port != FTPPORT)
283 /* Note: persistent fwmark-based services and
284 * persistent port zero service are handled here.
286 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
287 * port zero template:
288 * <protocol,caddr,0,vaddr,0,daddr,0>
291 protocol = IPPROTO_IP;
295 /* return *ignored = -1 so NF_DROP can be used */
296 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
297 vaddr, vport, ¶m) < 0) {
303 /* Check if a template already exists */
304 ct = ip_vs_ct_in_get(¶m);
305 if (!ct || !ip_vs_check_template(ct)) {
307 * No template found or the dest of the connection
308 * template is not available.
309 * return *ignored=0 i.e. ICMP and NF_DROP
311 dest = svc->scheduler->schedule(svc, skb);
313 IP_VS_DBG(1, "p-schedule: no dest found.\n");
314 kfree(param.pe_data);
319 if (dst_port == svc->port && svc->port != FTPPORT)
323 * This adds param.pe_data to the template,
324 * and thus param.pe_data will be destroyed
325 * when the template expires */
326 ct = ip_vs_conn_new(¶m, &dest->addr, dport,
327 IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
329 kfree(param.pe_data);
334 ct->timeout = svc->timeout;
336 /* set destination with the found template */
338 kfree(param.pe_data);
342 if (dport == svc->port && dest->port)
345 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
346 && iph.protocol == IPPROTO_UDP)?
347 IP_VS_CONN_F_ONE_PACKET : 0;
350 * Create a new connection according to the template
352 ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol, &iph.saddr,
353 src_port, &iph.daddr, dst_port, ¶m);
355 cp = ip_vs_conn_new(¶m, &dest->addr, dport, flags, dest, skb->mark);
365 ip_vs_control_add(cp, ct);
368 ip_vs_conn_stats(cp, svc);
374 * IPVS main scheduling function
375 * It selects a server according to the virtual service, and
376 * creates a connection entry.
377 * Protocols supported: TCP, UDP
381 * 1 : protocol tried to schedule (eg. on SYN), found svc but the
382 * svc/scheduler decides that this packet should be accepted with
383 * NF_ACCEPT because it must not be scheduled.
385 * 0 : scheduler can not find destination, so try bypass or
386 * return ICMP and then NF_DROP (ip_vs_leave).
388 * -1 : scheduler tried to schedule but fatal error occurred, eg.
389 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
390 * failure such as missing Call-ID, ENOMEM on skb_linearize
391 * or pe_data. In this case we should return NF_DROP without
392 * any attempts to send ICMP with ip_vs_leave.
395 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
396 struct ip_vs_proto_data *pd, int *ignored)
398 struct ip_vs_protocol *pp = pd->pp;
399 struct ip_vs_conn *cp = NULL;
400 struct ip_vs_iphdr iph;
401 struct ip_vs_dest *dest;
402 __be16 _ports[2], *pptr;
406 ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
407 pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
412 * FTPDATA needs this check when using local real server.
413 * Never schedule Active FTPDATA connections from real server.
414 * For LVS-NAT they must be already created. For other methods
415 * with persistence the connection is created on SYN+ACK.
417 if (pptr[0] == FTPDATA) {
418 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
419 "Not scheduling FTPDATA");
424 * Do not schedule replies from local real server.
426 if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
427 (cp = pp->conn_in_get(svc->af, skb, &iph, iph.len, 1))) {
428 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
429 "Not scheduling reply for existing connection");
430 __ip_vs_conn_put(cp);
437 if (svc->flags & IP_VS_SVC_F_PERSISTENT)
438 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored);
443 * Non-persistent service
445 if (!svc->fwmark && pptr[1] != svc->port) {
447 pr_err("Schedule: port zero only supported "
448 "in persistent services, "
449 "check your ipvs configuration\n");
453 dest = svc->scheduler->schedule(svc, skb);
455 IP_VS_DBG(1, "Schedule: no dest found.\n");
459 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
460 && iph.protocol == IPPROTO_UDP)?
461 IP_VS_CONN_F_ONE_PACKET : 0;
464 * Create a connection entry.
467 struct ip_vs_conn_param p;
469 ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol,
470 &iph.saddr, pptr[0], &iph.daddr, pptr[1],
472 cp = ip_vs_conn_new(&p, &dest->addr,
473 dest->port ? dest->port : pptr[1],
474 flags, dest, skb->mark);
481 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
482 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
484 IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
485 IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
486 IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
487 cp->flags, atomic_read(&cp->refcnt));
489 ip_vs_conn_stats(cp, svc);
495 * Pass or drop the packet.
496 * Called by ip_vs_in, when the virtual service is available but
497 * no destination is available for a new connection.
499 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
500 struct ip_vs_proto_data *pd)
503 struct netns_ipvs *ipvs;
504 __be16 _ports[2], *pptr;
505 struct ip_vs_iphdr iph;
508 ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
510 pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
512 ip_vs_service_put(svc);
517 #ifdef CONFIG_IP_VS_IPV6
518 if (svc->af == AF_INET6)
519 unicast = ipv6_addr_type(&iph.daddr.in6) & IPV6_ADDR_UNICAST;
522 unicast = (inet_addr_type(net, iph.daddr.ip) == RTN_UNICAST);
524 /* if it is fwmark-based service, the cache_bypass sysctl is up
525 and the destination is a non-local unicast, then create
526 a cache_bypass connection entry */
527 ipvs = net_ipvs(net);
528 if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
530 struct ip_vs_conn *cp;
531 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
532 iph.protocol == IPPROTO_UDP)?
533 IP_VS_CONN_F_ONE_PACKET : 0;
534 union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } };
536 ip_vs_service_put(svc);
538 /* create a new connection entry */
539 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
541 struct ip_vs_conn_param p;
542 ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol,
544 &iph.daddr, pptr[1], &p);
545 cp = ip_vs_conn_new(&p, &daddr, 0,
546 IP_VS_CONN_F_BYPASS | flags,
553 ip_vs_in_stats(cp, skb);
556 cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
558 /* transmit the first SYN packet */
559 ret = cp->packet_xmit(skb, cp, pd->pp);
560 /* do not touch skb anymore */
562 atomic_inc(&cp->in_pkts);
568 * When the virtual ftp service is presented, packets destined
569 * for other services on the VIP may get here (except services
570 * listed in the ipvs table), pass the packets, because it is
571 * not ipvs job to decide to drop the packets.
573 if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
574 ip_vs_service_put(svc);
578 ip_vs_service_put(svc);
581 * Notify the client that the destination is unreachable, and
582 * release the socket buffer.
583 * Since it is in IP layer, the TCP socket is not actually
584 * created, the TCP RST packet cannot be sent, instead that
585 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
587 #ifdef CONFIG_IP_VS_IPV6
588 if (svc->af == AF_INET6) {
590 struct net *net = dev_net(skb_dst(skb)->dev);
592 skb->dev = net->loopback_dev;
594 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
597 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
602 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
604 return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
607 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
609 if (NF_INET_LOCAL_IN == hooknum)
610 return IP_DEFRAG_VS_IN;
611 if (NF_INET_FORWARD == hooknum)
612 return IP_DEFRAG_VS_FWD;
613 return IP_DEFRAG_VS_OUT;
616 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
618 int err = ip_defrag(skb, user);
621 ip_send_check(ip_hdr(skb));
626 #ifdef CONFIG_IP_VS_IPV6
627 static inline int ip_vs_gather_frags_v6(struct sk_buff *skb, u_int32_t user)
629 /* TODO IPv6: Find out what to do here for IPv6 */
635 * Packet has been made sufficiently writable in caller
636 * - inout: 1=in->out, 0=out->in
638 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
639 struct ip_vs_conn *cp, int inout)
641 struct iphdr *iph = ip_hdr(skb);
642 unsigned int icmp_offset = iph->ihl*4;
643 struct icmphdr *icmph = (struct icmphdr *)(skb_network_header(skb) +
645 struct iphdr *ciph = (struct iphdr *)(icmph + 1);
648 iph->saddr = cp->vaddr.ip;
650 ciph->daddr = cp->vaddr.ip;
653 iph->daddr = cp->daddr.ip;
655 ciph->saddr = cp->daddr.ip;
659 /* the TCP/UDP/SCTP port */
660 if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
661 IPPROTO_SCTP == ciph->protocol) {
662 __be16 *ports = (void *)ciph + ciph->ihl*4;
665 ports[1] = cp->vport;
667 ports[0] = cp->dport;
670 /* And finally the ICMP checksum */
672 icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
673 skb->ip_summed = CHECKSUM_UNNECESSARY;
676 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
677 "Forwarding altered outgoing ICMP");
679 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
680 "Forwarding altered incoming ICMP");
683 #ifdef CONFIG_IP_VS_IPV6
684 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
685 struct ip_vs_conn *cp, int inout)
687 struct ipv6hdr *iph = ipv6_hdr(skb);
688 unsigned int icmp_offset = sizeof(struct ipv6hdr);
689 struct icmp6hdr *icmph = (struct icmp6hdr *)(skb_network_header(skb) +
691 struct ipv6hdr *ciph = (struct ipv6hdr *)(icmph + 1);
694 iph->saddr = cp->vaddr.in6;
695 ciph->daddr = cp->vaddr.in6;
697 iph->daddr = cp->daddr.in6;
698 ciph->saddr = cp->daddr.in6;
701 /* the TCP/UDP/SCTP port */
702 if (IPPROTO_TCP == ciph->nexthdr || IPPROTO_UDP == ciph->nexthdr ||
703 IPPROTO_SCTP == ciph->nexthdr) {
704 __be16 *ports = (void *)ciph + sizeof(struct ipv6hdr);
707 ports[1] = cp->vport;
709 ports[0] = cp->dport;
712 /* And finally the ICMP checksum */
713 icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
714 skb->len - icmp_offset,
716 skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
717 skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
718 skb->ip_summed = CHECKSUM_PARTIAL;
721 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
722 (void *)ciph - (void *)iph,
723 "Forwarding altered outgoing ICMPv6");
725 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
726 (void *)ciph - (void *)iph,
727 "Forwarding altered incoming ICMPv6");
731 /* Handle relevant response ICMP messages - forward to the right
732 * destination host. Used for NAT and local client.
734 static int handle_response_icmp(int af, struct sk_buff *skb,
735 union nf_inet_addr *snet,
736 __u8 protocol, struct ip_vs_conn *cp,
737 struct ip_vs_protocol *pp,
738 unsigned int offset, unsigned int ihl)
740 struct netns_ipvs *ipvs;
741 unsigned int verdict = NF_DROP;
743 if (IP_VS_FWD_METHOD(cp) != 0) {
744 pr_err("shouldn't reach here, because the box is on the "
745 "half connection in the tun/dr module.\n");
748 /* Ensure the checksum is correct */
749 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
750 /* Failed checksum! */
751 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
752 IP_VS_DBG_ADDR(af, snet));
756 if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
757 IPPROTO_SCTP == protocol)
758 offset += 2 * sizeof(__u16);
759 if (!skb_make_writable(skb, offset))
762 ipvs = net_ipvs(skb_net(skb));
764 #ifdef CONFIG_IP_VS_IPV6
766 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
769 ip_vs_nat_icmp(skb, pp, cp, 1);
771 #ifdef CONFIG_IP_VS_IPV6
772 if (af == AF_INET6) {
773 if (ipvs->sysctl_snat_reroute && ip6_route_me_harder(skb) != 0)
777 if ((ipvs->sysctl_snat_reroute ||
778 skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
779 ip_route_me_harder(skb, RTN_LOCAL) != 0)
782 /* do the statistics and put it back */
783 ip_vs_out_stats(cp, skb);
785 skb->ipvs_property = 1;
786 if (!(cp->flags & IP_VS_CONN_F_NFCT))
789 ip_vs_update_conntrack(skb, cp, 0);
793 __ip_vs_conn_put(cp);
799 * Handle ICMP messages in the inside-to-outside direction (outgoing).
800 * Find any that might be relevant, check against existing connections.
801 * Currently handles error types - unreachable, quench, ttl exceeded.
803 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
804 unsigned int hooknum)
807 struct icmphdr _icmph, *ic;
808 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
809 struct ip_vs_iphdr ciph;
810 struct ip_vs_conn *cp;
811 struct ip_vs_protocol *pp;
812 unsigned int offset, ihl;
813 union nf_inet_addr snet;
817 /* reassemble IP fragments */
818 if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
819 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
824 offset = ihl = iph->ihl * 4;
825 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
829 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
830 ic->type, ntohs(icmp_id(ic)),
831 &iph->saddr, &iph->daddr);
834 * Work through seeing if this is for us.
835 * These checks are supposed to be in an order that means easy
836 * things are checked first to speed up processing.... however
837 * this means that some packets will manage to get a long way
838 * down this stack and then be rejected, but that's life.
840 if ((ic->type != ICMP_DEST_UNREACH) &&
841 (ic->type != ICMP_SOURCE_QUENCH) &&
842 (ic->type != ICMP_TIME_EXCEEDED)) {
847 /* Now find the contained IP header */
848 offset += sizeof(_icmph);
849 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
851 return NF_ACCEPT; /* The packet looks wrong, ignore */
853 pp = ip_vs_proto_get(cih->protocol);
857 /* Is the embedded protocol header present? */
858 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
862 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
863 "Checking outgoing ICMP for");
865 offset += cih->ihl * 4;
867 ip_vs_fill_iphdr(AF_INET, cih, &ciph);
868 /* The embedded headers contain source and dest in reverse order */
869 cp = pp->conn_out_get(AF_INET, skb, &ciph, offset, 1);
873 snet.ip = iph->saddr;
874 return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
878 #ifdef CONFIG_IP_VS_IPV6
879 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
880 unsigned int hooknum)
883 struct icmp6hdr _icmph, *ic;
884 struct ipv6hdr _ciph, *cih; /* The ip header contained
886 struct ip_vs_iphdr ciph;
887 struct ip_vs_conn *cp;
888 struct ip_vs_protocol *pp;
890 union nf_inet_addr snet;
894 /* reassemble IP fragments */
895 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
896 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
901 offset = sizeof(struct ipv6hdr);
902 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
906 IP_VS_DBG(12, "Outgoing ICMPv6 (%d,%d) %pI6->%pI6\n",
907 ic->icmp6_type, ntohs(icmpv6_id(ic)),
908 &iph->saddr, &iph->daddr);
911 * Work through seeing if this is for us.
912 * These checks are supposed to be in an order that means easy
913 * things are checked first to speed up processing.... however
914 * this means that some packets will manage to get a long way
915 * down this stack and then be rejected, but that's life.
917 if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
918 (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
919 (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
924 /* Now find the contained IP header */
925 offset += sizeof(_icmph);
926 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
928 return NF_ACCEPT; /* The packet looks wrong, ignore */
930 pp = ip_vs_proto_get(cih->nexthdr);
934 /* Is the embedded protocol header present? */
935 /* TODO: we don't support fragmentation at the moment anyways */
936 if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
939 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
940 "Checking outgoing ICMPv6 for");
942 offset += sizeof(struct ipv6hdr);
944 ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
945 /* The embedded headers contain source and dest in reverse order */
946 cp = pp->conn_out_get(AF_INET6, skb, &ciph, offset, 1);
950 ipv6_addr_copy(&snet.in6, &iph->saddr);
951 return handle_response_icmp(AF_INET6, skb, &snet, cih->nexthdr, cp,
952 pp, offset, sizeof(struct ipv6hdr));
957 * Check if sctp chunc is ABORT chunk
959 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
961 sctp_chunkhdr_t *sch, schunk;
962 sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
963 sizeof(schunk), &schunk);
966 if (sch->type == SCTP_CID_ABORT)
971 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
973 struct tcphdr _tcph, *th;
975 th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
981 /* Handle response packets: rewrite addresses and send away...
982 * Used for NAT and local client.
985 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
986 struct ip_vs_conn *cp, int ihl)
988 struct ip_vs_protocol *pp = pd->pp;
989 struct netns_ipvs *ipvs;
991 IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
993 if (!skb_make_writable(skb, ihl))
996 /* mangle the packet */
997 if (pp->snat_handler && !pp->snat_handler(skb, pp, cp))
1000 #ifdef CONFIG_IP_VS_IPV6
1002 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1006 ip_hdr(skb)->saddr = cp->vaddr.ip;
1007 ip_send_check(ip_hdr(skb));
1011 * nf_iterate does not expect change in the skb->dst->dev.
1012 * It looks like it is not fatal to enable this code for hooks
1013 * where our handlers are at the end of the chain list and
1014 * when all next handlers use skb->dst->dev and not outdev.
1015 * It will definitely route properly the inout NAT traffic
1016 * when multiple paths are used.
1019 /* For policy routing, packets originating from this
1020 * machine itself may be routed differently to packets
1021 * passing through. We want this packet to be routed as
1022 * if it came from this machine itself. So re-compute
1023 * the routing information.
1025 ipvs = net_ipvs(skb_net(skb));
1027 #ifdef CONFIG_IP_VS_IPV6
1028 if (af == AF_INET6) {
1029 if (ipvs->sysctl_snat_reroute && ip6_route_me_harder(skb) != 0)
1033 if ((ipvs->sysctl_snat_reroute ||
1034 skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
1035 ip_route_me_harder(skb, RTN_LOCAL) != 0)
1038 IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1040 ip_vs_out_stats(cp, skb);
1041 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1042 skb->ipvs_property = 1;
1043 if (!(cp->flags & IP_VS_CONN_F_NFCT))
1046 ip_vs_update_conntrack(skb, cp, 0);
1060 * Check if outgoing packet belongs to the established ip_vs_conn.
1063 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1065 struct net *net = NULL;
1066 struct ip_vs_iphdr iph;
1067 struct ip_vs_protocol *pp;
1068 struct ip_vs_proto_data *pd;
1069 struct ip_vs_conn *cp;
1070 struct netns_ipvs *ipvs;
1074 /* Already marked as IPVS request or reply? */
1075 if (skb->ipvs_property)
1078 /* Bad... Do not break raw sockets */
1079 if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1081 struct sock *sk = skb->sk;
1082 struct inet_sock *inet = inet_sk(skb->sk);
1084 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1088 if (unlikely(!skb_dst(skb)))
1092 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1093 #ifdef CONFIG_IP_VS_IPV6
1094 if (af == AF_INET6) {
1095 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1097 int verdict = ip_vs_out_icmp_v6(skb, &related,
1102 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1106 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1108 int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1112 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1115 pd = ip_vs_proto_data_get(net, iph.protocol);
1120 /* reassemble IP fragments */
1121 #ifdef CONFIG_IP_VS_IPV6
1122 if (af == AF_INET6) {
1123 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1124 if (ip_vs_gather_frags_v6(skb,
1125 ip_vs_defrag_user(hooknum)))
1129 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1132 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_MF|IP_OFFSET) &&
1133 !pp->dont_defrag)) {
1134 if (ip_vs_gather_frags(skb,
1135 ip_vs_defrag_user(hooknum)))
1138 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1142 * Check if the packet belongs to an existing entry
1144 cp = pp->conn_out_get(af, skb, &iph, iph.len, 0);
1145 ipvs = net_ipvs(net);
1148 return handle_response(af, skb, pd, cp, iph.len);
1149 if (ipvs->sysctl_nat_icmp_send &&
1150 (pp->protocol == IPPROTO_TCP ||
1151 pp->protocol == IPPROTO_UDP ||
1152 pp->protocol == IPPROTO_SCTP)) {
1153 __be16 _ports[2], *pptr;
1155 pptr = skb_header_pointer(skb, iph.len,
1156 sizeof(_ports), _ports);
1158 return NF_ACCEPT; /* Not for me */
1159 if (ip_vs_lookup_real_service(net, af, iph.protocol,
1163 * Notify the real server: there is no
1164 * existing entry if it is not RST
1165 * packet or not TCP packet.
1167 if ((iph.protocol != IPPROTO_TCP &&
1168 iph.protocol != IPPROTO_SCTP)
1169 || ((iph.protocol == IPPROTO_TCP
1170 && !is_tcp_reset(skb, iph.len))
1171 || (iph.protocol == IPPROTO_SCTP
1172 && !is_sctp_abort(skb,
1174 #ifdef CONFIG_IP_VS_IPV6
1175 if (af == AF_INET6) {
1177 dev_net(skb_dst(skb)->dev);
1180 skb->dev = net->loopback_dev;
1182 ICMPV6_DEST_UNREACH,
1183 ICMPV6_PORT_UNREACH,
1189 ICMP_PORT_UNREACH, 0);
1194 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1195 "ip_vs_out: packet continues traversal as normal");
1200 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1201 * used only for VS/NAT.
1202 * Check if packet is reply for established ip_vs_conn.
1205 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1206 const struct net_device *in, const struct net_device *out,
1207 int (*okfn)(struct sk_buff *))
1209 return ip_vs_out(hooknum, skb, AF_INET);
1213 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1214 * Check if packet is reply for established ip_vs_conn.
1217 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1218 const struct net_device *in, const struct net_device *out,
1219 int (*okfn)(struct sk_buff *))
1221 unsigned int verdict;
1223 /* Disable BH in LOCAL_OUT until all places are fixed */
1225 verdict = ip_vs_out(hooknum, skb, AF_INET);
1230 #ifdef CONFIG_IP_VS_IPV6
1233 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1234 * used only for VS/NAT.
1235 * Check if packet is reply for established ip_vs_conn.
1238 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1239 const struct net_device *in, const struct net_device *out,
1240 int (*okfn)(struct sk_buff *))
1242 return ip_vs_out(hooknum, skb, AF_INET6);
1246 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1247 * Check if packet is reply for established ip_vs_conn.
1250 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1251 const struct net_device *in, const struct net_device *out,
1252 int (*okfn)(struct sk_buff *))
1254 unsigned int verdict;
1256 /* Disable BH in LOCAL_OUT until all places are fixed */
1258 verdict = ip_vs_out(hooknum, skb, AF_INET6);
1266 * Handle ICMP messages in the outside-to-inside direction (incoming).
1267 * Find any that might be relevant, check against existing connections,
1268 * forward to the right destination host if relevant.
1269 * Currently handles error types - unreachable, quench, ttl exceeded.
1272 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1274 struct net *net = NULL;
1276 struct icmphdr _icmph, *ic;
1277 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
1278 struct ip_vs_iphdr ciph;
1279 struct ip_vs_conn *cp;
1280 struct ip_vs_protocol *pp;
1281 struct ip_vs_proto_data *pd;
1282 unsigned int offset, ihl, verdict;
1283 union nf_inet_addr snet;
1287 /* reassemble IP fragments */
1288 if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
1289 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1294 offset = ihl = iph->ihl * 4;
1295 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1299 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1300 ic->type, ntohs(icmp_id(ic)),
1301 &iph->saddr, &iph->daddr);
1304 * Work through seeing if this is for us.
1305 * These checks are supposed to be in an order that means easy
1306 * things are checked first to speed up processing.... however
1307 * this means that some packets will manage to get a long way
1308 * down this stack and then be rejected, but that's life.
1310 if ((ic->type != ICMP_DEST_UNREACH) &&
1311 (ic->type != ICMP_SOURCE_QUENCH) &&
1312 (ic->type != ICMP_TIME_EXCEEDED)) {
1317 /* Now find the contained IP header */
1318 offset += sizeof(_icmph);
1319 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1321 return NF_ACCEPT; /* The packet looks wrong, ignore */
1324 pd = ip_vs_proto_data_get(net, cih->protocol);
1329 /* Is the embedded protocol header present? */
1330 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1334 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1335 "Checking incoming ICMP for");
1337 offset += cih->ihl * 4;
1339 ip_vs_fill_iphdr(AF_INET, cih, &ciph);
1340 /* The embedded headers contain source and dest in reverse order */
1341 cp = pp->conn_in_get(AF_INET, skb, &ciph, offset, 1);
1343 /* The packet could also belong to a local client */
1344 cp = pp->conn_out_get(AF_INET, skb, &ciph, offset, 1);
1346 snet.ip = iph->saddr;
1347 return handle_response_icmp(AF_INET, skb, &snet,
1348 cih->protocol, cp, pp,
1356 /* Ensure the checksum is correct */
1357 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1358 /* Failed checksum! */
1359 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1364 /* do the statistics and put it back */
1365 ip_vs_in_stats(cp, skb);
1366 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1367 offset += 2 * sizeof(__u16);
1368 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
1369 /* LOCALNODE from FORWARD hook is not supported */
1370 if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD &&
1371 skb_rtable(skb)->rt_flags & RTCF_LOCAL) {
1372 IP_VS_DBG(1, "%s(): "
1373 "local delivery to %pI4 but in FORWARD\n",
1374 __func__, &skb_rtable(skb)->rt_dst);
1379 __ip_vs_conn_put(cp);
1384 #ifdef CONFIG_IP_VS_IPV6
1386 ip_vs_in_icmp_v6(struct sk_buff *skb, int *related, unsigned int hooknum)
1388 struct net *net = NULL;
1389 struct ipv6hdr *iph;
1390 struct icmp6hdr _icmph, *ic;
1391 struct ipv6hdr _ciph, *cih; /* The ip header contained
1393 struct ip_vs_iphdr ciph;
1394 struct ip_vs_conn *cp;
1395 struct ip_vs_protocol *pp;
1396 struct ip_vs_proto_data *pd;
1397 unsigned int offset, verdict;
1398 union nf_inet_addr snet;
1399 struct rt6_info *rt;
1403 /* reassemble IP fragments */
1404 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1405 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
1409 iph = ipv6_hdr(skb);
1410 offset = sizeof(struct ipv6hdr);
1411 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1415 IP_VS_DBG(12, "Incoming ICMPv6 (%d,%d) %pI6->%pI6\n",
1416 ic->icmp6_type, ntohs(icmpv6_id(ic)),
1417 &iph->saddr, &iph->daddr);
1420 * Work through seeing if this is for us.
1421 * These checks are supposed to be in an order that means easy
1422 * things are checked first to speed up processing.... however
1423 * this means that some packets will manage to get a long way
1424 * down this stack and then be rejected, but that's life.
1426 if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
1427 (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
1428 (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
1433 /* Now find the contained IP header */
1434 offset += sizeof(_icmph);
1435 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1437 return NF_ACCEPT; /* The packet looks wrong, ignore */
1440 pd = ip_vs_proto_data_get(net, cih->nexthdr);
1445 /* Is the embedded protocol header present? */
1446 /* TODO: we don't support fragmentation at the moment anyways */
1447 if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
1450 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1451 "Checking incoming ICMPv6 for");
1453 offset += sizeof(struct ipv6hdr);
1455 ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
1456 /* The embedded headers contain source and dest in reverse order */
1457 cp = pp->conn_in_get(AF_INET6, skb, &ciph, offset, 1);
1459 /* The packet could also belong to a local client */
1460 cp = pp->conn_out_get(AF_INET6, skb, &ciph, offset, 1);
1462 ipv6_addr_copy(&snet.in6, &iph->saddr);
1463 return handle_response_icmp(AF_INET6, skb, &snet,
1466 sizeof(struct ipv6hdr));
1473 /* do the statistics and put it back */
1474 ip_vs_in_stats(cp, skb);
1475 if (IPPROTO_TCP == cih->nexthdr || IPPROTO_UDP == cih->nexthdr ||
1476 IPPROTO_SCTP == cih->nexthdr)
1477 offset += 2 * sizeof(__u16);
1478 verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset);
1479 /* LOCALNODE from FORWARD hook is not supported */
1480 if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD &&
1481 (rt = (struct rt6_info *) skb_dst(skb)) &&
1482 rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK) {
1483 IP_VS_DBG(1, "%s(): "
1484 "local delivery to %pI6 but in FORWARD\n",
1485 __func__, &rt->rt6i_dst);
1489 __ip_vs_conn_put(cp);
1497 * Check if it's for virtual services, look it up,
1498 * and send it on its way...
1501 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1504 struct ip_vs_iphdr iph;
1505 struct ip_vs_protocol *pp;
1506 struct ip_vs_proto_data *pd;
1507 struct ip_vs_conn *cp;
1508 int ret, restart, pkts;
1509 struct netns_ipvs *ipvs;
1511 /* Already marked as IPVS request or reply? */
1512 if (skb->ipvs_property)
1517 * - remote client: only PACKET_HOST
1518 * - route: used for struct net when skb->dev is unset
1520 if (unlikely((skb->pkt_type != PACKET_HOST &&
1521 hooknum != NF_INET_LOCAL_OUT) ||
1523 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1524 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1525 " ignored in hook %u\n",
1526 skb->pkt_type, iph.protocol,
1527 IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1530 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1532 /* Bad... Do not break raw sockets */
1533 if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1535 struct sock *sk = skb->sk;
1536 struct inet_sock *inet = inet_sk(skb->sk);
1538 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1542 #ifdef CONFIG_IP_VS_IPV6
1543 if (af == AF_INET6) {
1544 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1546 int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
1550 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1554 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1556 int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1560 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1564 /* Protocol supported? */
1565 pd = ip_vs_proto_data_get(net, iph.protocol);
1570 * Check if the packet belongs to an existing connection entry
1572 cp = pp->conn_in_get(af, skb, &iph, iph.len, 0);
1574 if (unlikely(!cp)) {
1577 if (!pp->conn_schedule(af, skb, pd, &v, &cp))
1581 if (unlikely(!cp)) {
1582 /* sorry, all this trouble for a no-hit :) */
1583 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1584 "ip_vs_in: packet continues traversal as normal");
1588 IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1590 ipvs = net_ipvs(net);
1591 /* Check the server status */
1592 if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1593 /* the destination server is not available */
1595 if (ipvs->sysctl_expire_nodest_conn) {
1596 /* try to expire the connection immediately */
1597 ip_vs_conn_expire_now(cp);
1599 /* don't restart its timer, and silently
1601 __ip_vs_conn_put(cp);
1605 ip_vs_in_stats(cp, skb);
1606 restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1607 if (cp->packet_xmit)
1608 ret = cp->packet_xmit(skb, cp, pp);
1609 /* do not touch skb anymore */
1611 IP_VS_DBG_RL("warning: packet_xmit is null");
1615 /* Increase its packet counter and check if it is needed
1616 * to be synchronized
1618 * Sync connection if it is about to close to
1619 * encorage the standby servers to update the connections timeout
1621 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1624 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1625 pkts = ipvs->sysctl_sync_threshold[0];
1627 pkts = atomic_add_return(1, &cp->in_pkts);
1629 if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1630 cp->protocol == IPPROTO_SCTP) {
1631 if ((cp->state == IP_VS_SCTP_S_ESTABLISHED &&
1632 (pkts % ipvs->sysctl_sync_threshold[1]
1633 == ipvs->sysctl_sync_threshold[0])) ||
1634 (cp->old_state != cp->state &&
1635 ((cp->state == IP_VS_SCTP_S_CLOSED) ||
1636 (cp->state == IP_VS_SCTP_S_SHUT_ACK_CLI) ||
1637 (cp->state == IP_VS_SCTP_S_SHUT_ACK_SER)))) {
1638 ip_vs_sync_conn(net, cp);
1643 /* Keep this block last: TCP and others with pp->num_states <= 1 */
1644 else if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1645 (((cp->protocol != IPPROTO_TCP ||
1646 cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1647 (pkts % ipvs->sysctl_sync_threshold[1]
1648 == ipvs->sysctl_sync_threshold[0])) ||
1649 ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) &&
1650 ((cp->state == IP_VS_TCP_S_FIN_WAIT) ||
1651 (cp->state == IP_VS_TCP_S_CLOSE) ||
1652 (cp->state == IP_VS_TCP_S_CLOSE_WAIT) ||
1653 (cp->state == IP_VS_TCP_S_TIME_WAIT)))))
1654 ip_vs_sync_conn(net, cp);
1656 cp->old_state = cp->state;
1663 * AF_INET handler in NF_INET_LOCAL_IN chain
1664 * Schedule and forward packets from remote clients
1667 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1668 const struct net_device *in,
1669 const struct net_device *out,
1670 int (*okfn)(struct sk_buff *))
1672 return ip_vs_in(hooknum, skb, AF_INET);
1676 * AF_INET handler in NF_INET_LOCAL_OUT chain
1677 * Schedule and forward packets from local clients
1680 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1681 const struct net_device *in, const struct net_device *out,
1682 int (*okfn)(struct sk_buff *))
1684 unsigned int verdict;
1686 /* Disable BH in LOCAL_OUT until all places are fixed */
1688 verdict = ip_vs_in(hooknum, skb, AF_INET);
1693 #ifdef CONFIG_IP_VS_IPV6
1696 * AF_INET6 handler in NF_INET_LOCAL_IN chain
1697 * Schedule and forward packets from remote clients
1700 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1701 const struct net_device *in,
1702 const struct net_device *out,
1703 int (*okfn)(struct sk_buff *))
1705 return ip_vs_in(hooknum, skb, AF_INET6);
1709 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
1710 * Schedule and forward packets from local clients
1713 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1714 const struct net_device *in, const struct net_device *out,
1715 int (*okfn)(struct sk_buff *))
1717 unsigned int verdict;
1719 /* Disable BH in LOCAL_OUT until all places are fixed */
1721 verdict = ip_vs_in(hooknum, skb, AF_INET6);
1730 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1731 * related packets destined for 0.0.0.0/0.
1732 * When fwmark-based virtual service is used, such as transparent
1733 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1734 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1735 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1736 * and send them to ip_vs_in_icmp.
1739 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1740 const struct net_device *in, const struct net_device *out,
1741 int (*okfn)(struct sk_buff *))
1745 if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1748 return ip_vs_in_icmp(skb, &r, hooknum);
1751 #ifdef CONFIG_IP_VS_IPV6
1753 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1754 const struct net_device *in, const struct net_device *out,
1755 int (*okfn)(struct sk_buff *))
1759 if (ipv6_hdr(skb)->nexthdr != IPPROTO_ICMPV6)
1762 return ip_vs_in_icmp_v6(skb, &r, hooknum);
1767 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1768 /* After packet filtering, change source only for VS/NAT */
1770 .hook = ip_vs_reply4,
1771 .owner = THIS_MODULE,
1773 .hooknum = NF_INET_LOCAL_IN,
1776 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1777 * or VS/NAT(change destination), so that filtering rules can be
1778 * applied to IPVS. */
1780 .hook = ip_vs_remote_request4,
1781 .owner = THIS_MODULE,
1783 .hooknum = NF_INET_LOCAL_IN,
1786 /* Before ip_vs_in, change source only for VS/NAT */
1788 .hook = ip_vs_local_reply4,
1789 .owner = THIS_MODULE,
1791 .hooknum = NF_INET_LOCAL_OUT,
1794 /* After mangle, schedule and forward local requests */
1796 .hook = ip_vs_local_request4,
1797 .owner = THIS_MODULE,
1799 .hooknum = NF_INET_LOCAL_OUT,
1802 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1803 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1805 .hook = ip_vs_forward_icmp,
1806 .owner = THIS_MODULE,
1808 .hooknum = NF_INET_FORWARD,
1811 /* After packet filtering, change source only for VS/NAT */
1813 .hook = ip_vs_reply4,
1814 .owner = THIS_MODULE,
1816 .hooknum = NF_INET_FORWARD,
1819 #ifdef CONFIG_IP_VS_IPV6
1820 /* After packet filtering, change source only for VS/NAT */
1822 .hook = ip_vs_reply6,
1823 .owner = THIS_MODULE,
1825 .hooknum = NF_INET_LOCAL_IN,
1828 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1829 * or VS/NAT(change destination), so that filtering rules can be
1830 * applied to IPVS. */
1832 .hook = ip_vs_remote_request6,
1833 .owner = THIS_MODULE,
1835 .hooknum = NF_INET_LOCAL_IN,
1838 /* Before ip_vs_in, change source only for VS/NAT */
1840 .hook = ip_vs_local_reply6,
1841 .owner = THIS_MODULE,
1843 .hooknum = NF_INET_LOCAL_OUT,
1846 /* After mangle, schedule and forward local requests */
1848 .hook = ip_vs_local_request6,
1849 .owner = THIS_MODULE,
1851 .hooknum = NF_INET_LOCAL_OUT,
1854 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1855 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1857 .hook = ip_vs_forward_icmp_v6,
1858 .owner = THIS_MODULE,
1860 .hooknum = NF_INET_FORWARD,
1863 /* After packet filtering, change source only for VS/NAT */
1865 .hook = ip_vs_reply6,
1866 .owner = THIS_MODULE,
1868 .hooknum = NF_INET_FORWARD,
1874 * Initialize IP Virtual Server netns mem.
1876 static int __net_init __ip_vs_init(struct net *net)
1878 struct netns_ipvs *ipvs;
1880 ipvs = net_generic(net, ip_vs_net_id);
1882 pr_err("%s(): no memory.\n", __func__);
1886 /* Counters used for creating unique names */
1887 ipvs->gen = atomic_read(&ipvs_netns_cnt);
1888 atomic_inc(&ipvs_netns_cnt);
1890 printk(KERN_INFO "IPVS: Creating netns size=%lu id=%d\n",
1891 sizeof(struct netns_ipvs), ipvs->gen);
1895 static void __net_exit __ip_vs_cleanup(struct net *net)
1897 IP_VS_DBG(10, "ipvs netns %d released\n", net_ipvs(net)->gen);
1900 static struct pernet_operations ipvs_core_ops = {
1901 .init = __ip_vs_init,
1902 .exit = __ip_vs_cleanup,
1903 .id = &ip_vs_net_id,
1904 .size = sizeof(struct netns_ipvs),
1908 * Initialize IP Virtual Server
1910 static int __init ip_vs_init(void)
1914 ret = register_pernet_subsys(&ipvs_core_ops); /* Alloc ip_vs struct */
1918 ip_vs_estimator_init();
1919 ret = ip_vs_control_init();
1921 pr_err("can't setup control.\n");
1922 goto cleanup_estimator;
1925 ip_vs_protocol_init();
1927 ret = ip_vs_app_init();
1929 pr_err("can't setup application helper.\n");
1930 goto cleanup_protocol;
1933 ret = ip_vs_conn_init();
1935 pr_err("can't setup connection table.\n");
1939 ret = ip_vs_sync_init();
1941 pr_err("can't setup sync data.\n");
1945 ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1947 pr_err("can't register hooks.\n");
1951 pr_info("ipvs loaded.\n");
1955 ip_vs_sync_cleanup();
1957 ip_vs_conn_cleanup();
1959 ip_vs_app_cleanup();
1961 ip_vs_protocol_cleanup();
1962 ip_vs_control_cleanup();
1964 ip_vs_estimator_cleanup();
1965 unregister_pernet_subsys(&ipvs_core_ops); /* free ip_vs struct */
1969 static void __exit ip_vs_cleanup(void)
1971 nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1972 ip_vs_sync_cleanup();
1973 ip_vs_conn_cleanup();
1974 ip_vs_app_cleanup();
1975 ip_vs_protocol_cleanup();
1976 ip_vs_control_cleanup();
1977 ip_vs_estimator_cleanup();
1978 unregister_pernet_subsys(&ipvs_core_ops); /* free ip_vs struct */
1979 pr_info("ipvs unloaded.\n");
1982 module_init(ip_vs_init);
1983 module_exit(ip_vs_cleanup);
1984 MODULE_LICENSE("GPL");