2 * IPVS An implementation of the IP virtual server support for the
3 * LINUX operating system. IPVS is now implemented as a module
4 * over the Netfilter framework. IPVS can be used to build a
5 * high-performance and highly available server based on a
8 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
9 * Peter Kese <peter.kese@ijs.si>
10 * Julian Anastasov <ja@ssi.bg>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
17 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
18 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
22 * Paul `Rusty' Russell properly handle non-linear skbs
23 * Harald Welte don't use nfcache
27 #define KMSG_COMPONENT "IPVS"
28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
30 #include <linux/module.h>
31 #include <linux/kernel.h>
33 #include <linux/tcp.h>
34 #include <linux/sctp.h>
35 #include <linux/icmp.h>
36 #include <linux/slab.h>
41 #include <net/icmp.h> /* for icmp_send */
42 #include <net/route.h>
43 #include <net/ip6_checksum.h>
44 #include <net/netns/generic.h> /* net_generic() */
46 #include <linux/netfilter.h>
47 #include <linux/netfilter_ipv4.h>
49 #ifdef CONFIG_IP_VS_IPV6
51 #include <linux/netfilter_ipv6.h>
52 #include <net/ip6_route.h>
55 #include <net/ip_vs.h>
58 EXPORT_SYMBOL(register_ip_vs_scheduler);
59 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
60 EXPORT_SYMBOL(ip_vs_proto_name);
61 EXPORT_SYMBOL(ip_vs_conn_new);
62 EXPORT_SYMBOL(ip_vs_conn_in_get);
63 EXPORT_SYMBOL(ip_vs_conn_out_get);
64 #ifdef CONFIG_IP_VS_PROTO_TCP
65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
67 EXPORT_SYMBOL(ip_vs_conn_put);
68 #ifdef CONFIG_IP_VS_DEBUG
69 EXPORT_SYMBOL(ip_vs_get_debug_level);
72 static int ip_vs_net_id __read_mostly;
73 /* netns cnt used for uniqueness */
74 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
76 /* ID used in ICMP lookups */
77 #define icmp_id(icmph) (((icmph)->un).echo.id)
78 #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier)
80 const char *ip_vs_proto_name(unsigned int proto)
95 #ifdef CONFIG_IP_VS_IPV6
100 sprintf(buf, "IP_%u", proto);
105 void ip_vs_init_hash_table(struct list_head *table, int rows)
108 INIT_LIST_HEAD(&table[rows]);
112 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
114 struct ip_vs_dest *dest = cp->dest;
115 struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
117 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
118 struct ip_vs_cpu_stats *s;
119 struct ip_vs_service *svc;
121 s = this_cpu_ptr(dest->stats.cpustats);
123 u64_stats_update_begin(&s->syncp);
124 s->ustats.inbytes += skb->len;
125 u64_stats_update_end(&s->syncp);
128 svc = rcu_dereference(dest->svc);
129 s = this_cpu_ptr(svc->stats.cpustats);
131 u64_stats_update_begin(&s->syncp);
132 s->ustats.inbytes += skb->len;
133 u64_stats_update_end(&s->syncp);
136 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
138 u64_stats_update_begin(&s->syncp);
139 s->ustats.inbytes += skb->len;
140 u64_stats_update_end(&s->syncp);
146 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
148 struct ip_vs_dest *dest = cp->dest;
149 struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
151 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
152 struct ip_vs_cpu_stats *s;
153 struct ip_vs_service *svc;
155 s = this_cpu_ptr(dest->stats.cpustats);
157 u64_stats_update_begin(&s->syncp);
158 s->ustats.outbytes += skb->len;
159 u64_stats_update_end(&s->syncp);
162 svc = rcu_dereference(dest->svc);
163 s = this_cpu_ptr(svc->stats.cpustats);
165 u64_stats_update_begin(&s->syncp);
166 s->ustats.outbytes += skb->len;
167 u64_stats_update_end(&s->syncp);
170 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
172 u64_stats_update_begin(&s->syncp);
173 s->ustats.outbytes += skb->len;
174 u64_stats_update_end(&s->syncp);
180 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
182 struct netns_ipvs *ipvs = net_ipvs(svc->net);
183 struct ip_vs_cpu_stats *s;
185 s = this_cpu_ptr(cp->dest->stats.cpustats);
188 s = this_cpu_ptr(svc->stats.cpustats);
191 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
197 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
198 const struct sk_buff *skb,
199 struct ip_vs_proto_data *pd)
201 if (likely(pd->pp->state_transition))
202 pd->pp->state_transition(cp, direction, skb, pd);
206 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
207 struct sk_buff *skb, int protocol,
208 const union nf_inet_addr *caddr, __be16 cport,
209 const union nf_inet_addr *vaddr, __be16 vport,
210 struct ip_vs_conn_param *p)
212 ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
214 p->pe = rcu_dereference(svc->pe);
215 if (p->pe && p->pe->fill_param)
216 return p->pe->fill_param(p, skb);
222 * IPVS persistent scheduling function
223 * It creates a connection entry according to its template if exists,
224 * or selects a server and creates a connection entry plus a template.
225 * Locking: we are svc user (svc->refcnt), so we hold all dests too
226 * Protocols supported: TCP, UDP
228 static struct ip_vs_conn *
229 ip_vs_sched_persist(struct ip_vs_service *svc,
230 struct sk_buff *skb, __be16 src_port, __be16 dst_port,
231 int *ignored, struct ip_vs_iphdr *iph)
233 struct ip_vs_conn *cp = NULL;
234 struct ip_vs_dest *dest;
235 struct ip_vs_conn *ct;
236 __be16 dport = 0; /* destination port to forward */
238 struct ip_vs_conn_param param;
239 const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
240 union nf_inet_addr snet; /* source network of the client,
243 /* Mask saddr with the netmask to adjust template granularity */
244 #ifdef CONFIG_IP_VS_IPV6
245 if (svc->af == AF_INET6)
246 ipv6_addr_prefix(&snet.in6, &iph->saddr.in6,
247 (__force __u32) svc->netmask);
250 snet.ip = iph->saddr.ip & svc->netmask;
252 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
254 IP_VS_DBG_ADDR(svc->af, &iph->saddr), ntohs(src_port),
255 IP_VS_DBG_ADDR(svc->af, &iph->daddr), ntohs(dst_port),
256 IP_VS_DBG_ADDR(svc->af, &snet));
259 * As far as we know, FTP is a very complicated network protocol, and
260 * it uses control connection and data connections. For active FTP,
261 * FTP server initialize data connection to the client, its source port
262 * is often 20. For passive FTP, FTP server tells the clients the port
263 * that it passively listens to, and the client issues the data
264 * connection. In the tunneling or direct routing mode, the load
265 * balancer is on the client-to-server half of connection, the port
266 * number is unknown to the load balancer. So, a conn template like
267 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
268 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
269 * is created for other persistent services.
272 int protocol = iph->protocol;
273 const union nf_inet_addr *vaddr = &iph->daddr;
276 if (dst_port == svc->port) {
278 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
280 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
282 if (svc->port != FTPPORT)
285 /* Note: persistent fwmark-based services and
286 * persistent port zero service are handled here.
288 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
289 * port zero template:
290 * <protocol,caddr,0,vaddr,0,daddr,0>
293 protocol = IPPROTO_IP;
297 /* return *ignored = -1 so NF_DROP can be used */
298 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
299 vaddr, vport, ¶m) < 0) {
305 /* Check if a template already exists */
306 ct = ip_vs_ct_in_get(¶m);
307 if (!ct || !ip_vs_check_template(ct)) {
308 struct ip_vs_scheduler *sched;
311 * No template found or the dest of the connection
312 * template is not available.
313 * return *ignored=0 i.e. ICMP and NF_DROP
315 sched = rcu_dereference(svc->scheduler);
316 dest = sched->schedule(svc, skb, iph);
318 IP_VS_DBG(1, "p-schedule: no dest found.\n");
319 kfree(param.pe_data);
324 if (dst_port == svc->port && svc->port != FTPPORT)
328 * This adds param.pe_data to the template,
329 * and thus param.pe_data will be destroyed
330 * when the template expires */
331 ct = ip_vs_conn_new(¶m, dest->af, &dest->addr, dport,
332 IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
334 kfree(param.pe_data);
339 ct->timeout = svc->timeout;
341 /* set destination with the found template */
343 kfree(param.pe_data);
347 if (dport == svc->port && dest->port)
350 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
351 && iph->protocol == IPPROTO_UDP) ?
352 IP_VS_CONN_F_ONE_PACKET : 0;
355 * Create a new connection according to the template
357 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol, &iph->saddr,
358 src_port, &iph->daddr, dst_port, ¶m);
360 cp = ip_vs_conn_new(¶m, dest->af, &dest->addr, dport, flags, dest,
371 ip_vs_control_add(cp, ct);
374 ip_vs_conn_stats(cp, svc);
380 * IPVS main scheduling function
381 * It selects a server according to the virtual service, and
382 * creates a connection entry.
383 * Protocols supported: TCP, UDP
387 * 1 : protocol tried to schedule (eg. on SYN), found svc but the
388 * svc/scheduler decides that this packet should be accepted with
389 * NF_ACCEPT because it must not be scheduled.
391 * 0 : scheduler can not find destination, so try bypass or
392 * return ICMP and then NF_DROP (ip_vs_leave).
394 * -1 : scheduler tried to schedule but fatal error occurred, eg.
395 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
396 * failure such as missing Call-ID, ENOMEM on skb_linearize
397 * or pe_data. In this case we should return NF_DROP without
398 * any attempts to send ICMP with ip_vs_leave.
401 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
402 struct ip_vs_proto_data *pd, int *ignored,
403 struct ip_vs_iphdr *iph)
405 struct ip_vs_protocol *pp = pd->pp;
406 struct ip_vs_conn *cp = NULL;
407 struct ip_vs_scheduler *sched;
408 struct ip_vs_dest *dest;
409 __be16 _ports[2], *pptr;
414 * IPv6 frags, only the first hit here.
416 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
421 * FTPDATA needs this check when using local real server.
422 * Never schedule Active FTPDATA connections from real server.
423 * For LVS-NAT they must be already created. For other methods
424 * with persistence the connection is created on SYN+ACK.
426 if (pptr[0] == FTPDATA) {
427 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
428 "Not scheduling FTPDATA");
433 * Do not schedule replies from local real server.
435 if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
436 (cp = pp->conn_in_get(svc->af, skb, iph, 1))) {
437 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
438 "Not scheduling reply for existing connection");
439 __ip_vs_conn_put(cp);
446 if (svc->flags & IP_VS_SVC_F_PERSISTENT)
447 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored,
453 * Non-persistent service
455 if (!svc->fwmark && pptr[1] != svc->port) {
457 pr_err("Schedule: port zero only supported "
458 "in persistent services, "
459 "check your ipvs configuration\n");
463 sched = rcu_dereference(svc->scheduler);
464 dest = sched->schedule(svc, skb, iph);
466 IP_VS_DBG(1, "Schedule: no dest found.\n");
470 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
471 && iph->protocol == IPPROTO_UDP) ?
472 IP_VS_CONN_F_ONE_PACKET : 0;
475 * Create a connection entry.
478 struct ip_vs_conn_param p;
480 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
481 &iph->saddr, pptr[0], &iph->daddr,
483 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
484 dest->port ? dest->port : pptr[1],
485 flags, dest, skb->mark);
492 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
493 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
495 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
496 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
497 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
498 cp->flags, atomic_read(&cp->refcnt));
500 ip_vs_conn_stats(cp, svc);
506 * Pass or drop the packet.
507 * Called by ip_vs_in, when the virtual service is available but
508 * no destination is available for a new connection.
510 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
511 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
513 __be16 _ports[2], *pptr;
516 struct netns_ipvs *ipvs;
520 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
528 #ifdef CONFIG_IP_VS_IPV6
529 if (svc->af == AF_INET6)
530 unicast = ipv6_addr_type(&iph->daddr.in6) & IPV6_ADDR_UNICAST;
533 unicast = (inet_addr_type(net, iph->daddr.ip) == RTN_UNICAST);
535 /* if it is fwmark-based service, the cache_bypass sysctl is up
536 and the destination is a non-local unicast, then create
537 a cache_bypass connection entry */
538 ipvs = net_ipvs(net);
539 if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
541 struct ip_vs_conn *cp;
542 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
543 iph->protocol == IPPROTO_UDP) ?
544 IP_VS_CONN_F_ONE_PACKET : 0;
545 union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } };
547 /* create a new connection entry */
548 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
550 struct ip_vs_conn_param p;
551 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
552 &iph->saddr, pptr[0],
553 &iph->daddr, pptr[1], &p);
554 cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
555 IP_VS_CONN_F_BYPASS | flags,
562 ip_vs_in_stats(cp, skb);
565 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
567 /* transmit the first SYN packet */
568 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
569 /* do not touch skb anymore */
571 atomic_inc(&cp->in_pkts);
578 * When the virtual ftp service is presented, packets destined
579 * for other services on the VIP may get here (except services
580 * listed in the ipvs table), pass the packets, because it is
581 * not ipvs job to decide to drop the packets.
583 if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT))
587 * Notify the client that the destination is unreachable, and
588 * release the socket buffer.
589 * Since it is in IP layer, the TCP socket is not actually
590 * created, the TCP RST packet cannot be sent, instead that
591 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
593 #ifdef CONFIG_IP_VS_IPV6
594 if (svc->af == AF_INET6) {
596 struct net *net_ = dev_net(skb_dst(skb)->dev);
598 skb->dev = net_->loopback_dev;
600 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
603 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
610 static int sysctl_snat_reroute(struct sk_buff *skb)
612 struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
613 return ipvs->sysctl_snat_reroute;
616 static int sysctl_nat_icmp_send(struct net *net)
618 struct netns_ipvs *ipvs = net_ipvs(net);
619 return ipvs->sysctl_nat_icmp_send;
622 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
624 return ipvs->sysctl_expire_nodest_conn;
629 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
630 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
631 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
635 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
637 return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
640 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
642 if (NF_INET_LOCAL_IN == hooknum)
643 return IP_DEFRAG_VS_IN;
644 if (NF_INET_FORWARD == hooknum)
645 return IP_DEFRAG_VS_FWD;
646 return IP_DEFRAG_VS_OUT;
649 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
654 err = ip_defrag(skb, user);
657 ip_send_check(ip_hdr(skb));
662 static int ip_vs_route_me_harder(int af, struct sk_buff *skb,
663 unsigned int hooknum)
665 if (!sysctl_snat_reroute(skb))
667 /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
668 if (NF_INET_LOCAL_IN == hooknum)
670 #ifdef CONFIG_IP_VS_IPV6
671 if (af == AF_INET6) {
672 struct dst_entry *dst = skb_dst(skb);
674 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
675 ip6_route_me_harder(skb) != 0)
679 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
680 ip_route_me_harder(skb, RTN_LOCAL) != 0)
687 * Packet has been made sufficiently writable in caller
688 * - inout: 1=in->out, 0=out->in
690 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
691 struct ip_vs_conn *cp, int inout)
693 struct iphdr *iph = ip_hdr(skb);
694 unsigned int icmp_offset = iph->ihl*4;
695 struct icmphdr *icmph = (struct icmphdr *)(skb_network_header(skb) +
697 struct iphdr *ciph = (struct iphdr *)(icmph + 1);
700 iph->saddr = cp->vaddr.ip;
702 ciph->daddr = cp->vaddr.ip;
705 iph->daddr = cp->daddr.ip;
707 ciph->saddr = cp->daddr.ip;
711 /* the TCP/UDP/SCTP port */
712 if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
713 IPPROTO_SCTP == ciph->protocol) {
714 __be16 *ports = (void *)ciph + ciph->ihl*4;
717 ports[1] = cp->vport;
719 ports[0] = cp->dport;
722 /* And finally the ICMP checksum */
724 icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
725 skb->ip_summed = CHECKSUM_UNNECESSARY;
728 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
729 "Forwarding altered outgoing ICMP");
731 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
732 "Forwarding altered incoming ICMP");
735 #ifdef CONFIG_IP_VS_IPV6
736 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
737 struct ip_vs_conn *cp, int inout)
739 struct ipv6hdr *iph = ipv6_hdr(skb);
740 unsigned int icmp_offset = 0;
741 unsigned int offs = 0; /* header offset*/
743 struct icmp6hdr *icmph;
744 struct ipv6hdr *ciph;
745 unsigned short fragoffs;
747 ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
748 icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
749 offs = icmp_offset + sizeof(struct icmp6hdr);
750 ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
752 protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
755 iph->saddr = cp->vaddr.in6;
756 ciph->daddr = cp->vaddr.in6;
758 iph->daddr = cp->daddr.in6;
759 ciph->saddr = cp->daddr.in6;
762 /* the TCP/UDP/SCTP port */
763 if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
764 IPPROTO_SCTP == protocol)) {
765 __be16 *ports = (void *)(skb_network_header(skb) + offs);
767 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
768 ntohs(inout ? ports[1] : ports[0]),
769 ntohs(inout ? cp->vport : cp->dport));
771 ports[1] = cp->vport;
773 ports[0] = cp->dport;
776 /* And finally the ICMP checksum */
777 icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
778 skb->len - icmp_offset,
780 skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
781 skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
782 skb->ip_summed = CHECKSUM_PARTIAL;
785 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
786 (void *)ciph - (void *)iph,
787 "Forwarding altered outgoing ICMPv6");
789 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
790 (void *)ciph - (void *)iph,
791 "Forwarding altered incoming ICMPv6");
795 /* Handle relevant response ICMP messages - forward to the right
798 static int handle_response_icmp(int af, struct sk_buff *skb,
799 union nf_inet_addr *snet,
800 __u8 protocol, struct ip_vs_conn *cp,
801 struct ip_vs_protocol *pp,
802 unsigned int offset, unsigned int ihl,
803 unsigned int hooknum)
805 unsigned int verdict = NF_DROP;
807 if (IP_VS_FWD_METHOD(cp) != 0) {
808 pr_err("shouldn't reach here, because the box is on the "
809 "half connection in the tun/dr module.\n");
812 /* Ensure the checksum is correct */
813 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
814 /* Failed checksum! */
815 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
816 IP_VS_DBG_ADDR(af, snet));
820 if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
821 IPPROTO_SCTP == protocol)
822 offset += 2 * sizeof(__u16);
823 if (!skb_make_writable(skb, offset))
826 #ifdef CONFIG_IP_VS_IPV6
828 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
831 ip_vs_nat_icmp(skb, pp, cp, 1);
833 if (ip_vs_route_me_harder(af, skb, hooknum))
836 /* do the statistics and put it back */
837 ip_vs_out_stats(cp, skb);
839 skb->ipvs_property = 1;
840 if (!(cp->flags & IP_VS_CONN_F_NFCT))
843 ip_vs_update_conntrack(skb, cp, 0);
847 __ip_vs_conn_put(cp);
853 * Handle ICMP messages in the inside-to-outside direction (outgoing).
854 * Find any that might be relevant, check against existing connections.
855 * Currently handles error types - unreachable, quench, ttl exceeded.
857 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
858 unsigned int hooknum)
861 struct icmphdr _icmph, *ic;
862 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
863 struct ip_vs_iphdr ciph;
864 struct ip_vs_conn *cp;
865 struct ip_vs_protocol *pp;
866 unsigned int offset, ihl;
867 union nf_inet_addr snet;
871 /* reassemble IP fragments */
872 if (ip_is_fragment(ip_hdr(skb))) {
873 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
878 offset = ihl = iph->ihl * 4;
879 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
883 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
884 ic->type, ntohs(icmp_id(ic)),
885 &iph->saddr, &iph->daddr);
888 * Work through seeing if this is for us.
889 * These checks are supposed to be in an order that means easy
890 * things are checked first to speed up processing.... however
891 * this means that some packets will manage to get a long way
892 * down this stack and then be rejected, but that's life.
894 if ((ic->type != ICMP_DEST_UNREACH) &&
895 (ic->type != ICMP_SOURCE_QUENCH) &&
896 (ic->type != ICMP_TIME_EXCEEDED)) {
901 /* Now find the contained IP header */
902 offset += sizeof(_icmph);
903 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
905 return NF_ACCEPT; /* The packet looks wrong, ignore */
907 pp = ip_vs_proto_get(cih->protocol);
911 /* Is the embedded protocol header present? */
912 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
916 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
917 "Checking outgoing ICMP for");
919 ip_vs_fill_ip4hdr(cih, &ciph);
921 /* The embedded headers contain source and dest in reverse order */
922 cp = pp->conn_out_get(AF_INET, skb, &ciph, 1);
926 snet.ip = iph->saddr;
927 return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
928 pp, ciph.len, ihl, hooknum);
931 #ifdef CONFIG_IP_VS_IPV6
932 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
933 unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
935 struct icmp6hdr _icmph, *ic;
936 struct ipv6hdr _ip6h, *ip6h; /* The ip header contained within ICMP */
937 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
938 struct ip_vs_conn *cp;
939 struct ip_vs_protocol *pp;
940 union nf_inet_addr snet;
941 unsigned int writable;
944 ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
949 * Work through seeing if this is for us.
950 * These checks are supposed to be in an order that means easy
951 * things are checked first to speed up processing.... however
952 * this means that some packets will manage to get a long way
953 * down this stack and then be rejected, but that's life.
955 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
959 /* Fragment header that is before ICMP header tells us that:
960 * it's not an error message since they can't be fragmented.
962 if (ipvsh->flags & IP6_FH_F_FRAG)
965 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
966 ic->icmp6_type, ntohs(icmpv6_id(ic)),
967 &ipvsh->saddr, &ipvsh->daddr);
969 /* Now find the contained IP header */
970 ciph.len = ipvsh->len + sizeof(_icmph);
971 ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
973 return NF_ACCEPT; /* The packet looks wrong, ignore */
974 ciph.saddr.in6 = ip6h->saddr; /* conn_out_get() handles reverse order */
975 ciph.daddr.in6 = ip6h->daddr;
976 /* skip possible IPv6 exthdrs of contained IPv6 packet */
977 ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
978 if (ciph.protocol < 0)
979 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
981 pp = ip_vs_proto_get(ciph.protocol);
985 /* The embedded headers contain source and dest in reverse order */
986 cp = pp->conn_out_get(AF_INET6, skb, &ciph, 1);
990 snet.in6 = ciph.saddr.in6;
992 return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
993 pp, writable, sizeof(struct ipv6hdr),
999 * Check if sctp chunc is ABORT chunk
1001 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1003 sctp_chunkhdr_t *sch, schunk;
1004 sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1005 sizeof(schunk), &schunk);
1008 if (sch->type == SCTP_CID_ABORT)
1013 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1015 struct tcphdr _tcph, *th;
1017 th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1023 static inline bool is_new_conn(const struct sk_buff *skb,
1024 struct ip_vs_iphdr *iph)
1026 switch (iph->protocol) {
1028 struct tcphdr _tcph, *th;
1030 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1035 case IPPROTO_SCTP: {
1036 sctp_chunkhdr_t *sch, schunk;
1038 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1039 sizeof(schunk), &schunk);
1042 return sch->type == SCTP_CID_INIT;
1049 /* Handle response packets: rewrite addresses and send away...
1052 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1053 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1054 unsigned int hooknum)
1056 struct ip_vs_protocol *pp = pd->pp;
1058 IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1060 if (!skb_make_writable(skb, iph->len))
1063 /* mangle the packet */
1064 if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1067 #ifdef CONFIG_IP_VS_IPV6
1069 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1073 ip_hdr(skb)->saddr = cp->vaddr.ip;
1074 ip_send_check(ip_hdr(skb));
1078 * nf_iterate does not expect change in the skb->dst->dev.
1079 * It looks like it is not fatal to enable this code for hooks
1080 * where our handlers are at the end of the chain list and
1081 * when all next handlers use skb->dst->dev and not outdev.
1082 * It will definitely route properly the inout NAT traffic
1083 * when multiple paths are used.
1086 /* For policy routing, packets originating from this
1087 * machine itself may be routed differently to packets
1088 * passing through. We want this packet to be routed as
1089 * if it came from this machine itself. So re-compute
1090 * the routing information.
1092 if (ip_vs_route_me_harder(af, skb, hooknum))
1095 IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1097 ip_vs_out_stats(cp, skb);
1098 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1099 skb->ipvs_property = 1;
1100 if (!(cp->flags & IP_VS_CONN_F_NFCT))
1103 ip_vs_update_conntrack(skb, cp, 0);
1117 * Check if outgoing packet belongs to the established ip_vs_conn.
1120 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1122 struct net *net = NULL;
1123 struct ip_vs_iphdr iph;
1124 struct ip_vs_protocol *pp;
1125 struct ip_vs_proto_data *pd;
1126 struct ip_vs_conn *cp;
1130 /* Already marked as IPVS request or reply? */
1131 if (skb->ipvs_property)
1134 /* Bad... Do not break raw sockets */
1135 if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1137 struct sock *sk = skb->sk;
1138 struct inet_sock *inet = inet_sk(skb->sk);
1140 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1144 if (unlikely(!skb_dst(skb)))
1148 if (!net_ipvs(net)->enable)
1151 ip_vs_fill_iph_skb(af, skb, &iph);
1152 #ifdef CONFIG_IP_VS_IPV6
1153 if (af == AF_INET6) {
1154 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1156 int verdict = ip_vs_out_icmp_v6(skb, &related,
1164 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1166 int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1172 pd = ip_vs_proto_data_get(net, iph.protocol);
1177 /* reassemble IP fragments */
1178 #ifdef CONFIG_IP_VS_IPV6
1181 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1182 if (ip_vs_gather_frags(skb,
1183 ip_vs_defrag_user(hooknum)))
1186 ip_vs_fill_ip4hdr(skb_network_header(skb), &iph);
1190 * Check if the packet belongs to an existing entry
1192 cp = pp->conn_out_get(af, skb, &iph, 0);
1195 return handle_response(af, skb, pd, cp, &iph, hooknum);
1196 if (sysctl_nat_icmp_send(net) &&
1197 (pp->protocol == IPPROTO_TCP ||
1198 pp->protocol == IPPROTO_UDP ||
1199 pp->protocol == IPPROTO_SCTP)) {
1200 __be16 _ports[2], *pptr;
1202 pptr = frag_safe_skb_hp(skb, iph.len,
1203 sizeof(_ports), _ports, &iph);
1205 return NF_ACCEPT; /* Not for me */
1206 if (ip_vs_has_real_service(net, af, iph.protocol, &iph.saddr,
1209 * Notify the real server: there is no
1210 * existing entry if it is not RST
1211 * packet or not TCP packet.
1213 if ((iph.protocol != IPPROTO_TCP &&
1214 iph.protocol != IPPROTO_SCTP)
1215 || ((iph.protocol == IPPROTO_TCP
1216 && !is_tcp_reset(skb, iph.len))
1217 || (iph.protocol == IPPROTO_SCTP
1218 && !is_sctp_abort(skb,
1220 #ifdef CONFIG_IP_VS_IPV6
1221 if (af == AF_INET6) {
1223 skb->dev = net->loopback_dev;
1225 ICMPV6_DEST_UNREACH,
1226 ICMPV6_PORT_UNREACH,
1232 ICMP_PORT_UNREACH, 0);
1237 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1238 "ip_vs_out: packet continues traversal as normal");
1243 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1244 * used only for VS/NAT.
1245 * Check if packet is reply for established ip_vs_conn.
1248 ip_vs_reply4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1249 const struct net_device *in, const struct net_device *out,
1250 int (*okfn)(struct sk_buff *))
1252 return ip_vs_out(ops->hooknum, skb, AF_INET);
1256 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1257 * Check if packet is reply for established ip_vs_conn.
1260 ip_vs_local_reply4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1261 const struct net_device *in, const struct net_device *out,
1262 int (*okfn)(struct sk_buff *))
1264 return ip_vs_out(ops->hooknum, skb, AF_INET);
1267 #ifdef CONFIG_IP_VS_IPV6
1270 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1271 * used only for VS/NAT.
1272 * Check if packet is reply for established ip_vs_conn.
1275 ip_vs_reply6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1276 const struct net_device *in, const struct net_device *out,
1277 int (*okfn)(struct sk_buff *))
1279 return ip_vs_out(ops->hooknum, skb, AF_INET6);
1283 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1284 * Check if packet is reply for established ip_vs_conn.
1287 ip_vs_local_reply6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1288 const struct net_device *in, const struct net_device *out,
1289 int (*okfn)(struct sk_buff *))
1291 return ip_vs_out(ops->hooknum, skb, AF_INET6);
1297 * Handle ICMP messages in the outside-to-inside direction (incoming).
1298 * Find any that might be relevant, check against existing connections,
1299 * forward to the right destination host if relevant.
1300 * Currently handles error types - unreachable, quench, ttl exceeded.
1303 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1305 struct net *net = NULL;
1307 struct icmphdr _icmph, *ic;
1308 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
1309 struct ip_vs_iphdr ciph;
1310 struct ip_vs_conn *cp;
1311 struct ip_vs_protocol *pp;
1312 struct ip_vs_proto_data *pd;
1313 unsigned int offset, offset2, ihl, verdict;
1318 /* reassemble IP fragments */
1319 if (ip_is_fragment(ip_hdr(skb))) {
1320 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1325 offset = ihl = iph->ihl * 4;
1326 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1330 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1331 ic->type, ntohs(icmp_id(ic)),
1332 &iph->saddr, &iph->daddr);
1335 * Work through seeing if this is for us.
1336 * These checks are supposed to be in an order that means easy
1337 * things are checked first to speed up processing.... however
1338 * this means that some packets will manage to get a long way
1339 * down this stack and then be rejected, but that's life.
1341 if ((ic->type != ICMP_DEST_UNREACH) &&
1342 (ic->type != ICMP_SOURCE_QUENCH) &&
1343 (ic->type != ICMP_TIME_EXCEEDED)) {
1348 /* Now find the contained IP header */
1349 offset += sizeof(_icmph);
1350 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1352 return NF_ACCEPT; /* The packet looks wrong, ignore */
1356 /* Special case for errors for IPIP packets */
1358 if (cih->protocol == IPPROTO_IPIP) {
1359 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1361 /* Error for our IPIP must arrive at LOCAL_IN */
1362 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1364 offset += cih->ihl * 4;
1365 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1367 return NF_ACCEPT; /* The packet looks wrong, ignore */
1371 pd = ip_vs_proto_data_get(net, cih->protocol);
1376 /* Is the embedded protocol header present? */
1377 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1381 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1382 "Checking incoming ICMP for");
1385 ip_vs_fill_ip4hdr(cih, &ciph);
1388 /* The embedded headers contain source and dest in reverse order.
1389 * For IPIP this is error for request, not for reply.
1391 cp = pp->conn_in_get(AF_INET, skb, &ciph, ipip ? 0 : 1);
1397 /* Ensure the checksum is correct */
1398 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1399 /* Failed checksum! */
1400 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1406 __be32 info = ic->un.gateway;
1407 __u8 type = ic->type;
1408 __u8 code = ic->code;
1410 /* Update the MTU */
1411 if (ic->type == ICMP_DEST_UNREACH &&
1412 ic->code == ICMP_FRAG_NEEDED) {
1413 struct ip_vs_dest *dest = cp->dest;
1414 u32 mtu = ntohs(ic->un.frag.mtu);
1415 __be16 frag_off = cih->frag_off;
1417 /* Strip outer IP and ICMP, go to IPIP header */
1418 if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1420 offset2 -= ihl + sizeof(_icmph);
1421 skb_reset_network_header(skb);
1422 IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1423 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1424 ipv4_update_pmtu(skb, dev_net(skb->dev),
1426 /* Client uses PMTUD? */
1427 if (!(frag_off & htons(IP_DF)))
1429 /* Prefer the resulting PMTU */
1431 struct ip_vs_dest_dst *dest_dst;
1434 dest_dst = rcu_dereference(dest->dest_dst);
1436 mtu = dst_mtu(dest_dst->dst_cache);
1439 if (mtu > 68 + sizeof(struct iphdr))
1440 mtu -= sizeof(struct iphdr);
1443 /* Strip outer IP, ICMP and IPIP, go to IP header of
1446 if (pskb_pull(skb, offset2) == NULL)
1448 skb_reset_network_header(skb);
1449 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1450 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1451 type, code, ntohl(info));
1452 icmp_send(skb, type, code, info);
1453 /* ICMP can be shorter but anyways, account it */
1454 ip_vs_out_stats(cp, skb);
1458 verdict = NF_STOLEN;
1462 /* do the statistics and put it back */
1463 ip_vs_in_stats(cp, skb);
1464 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1465 IPPROTO_SCTP == cih->protocol)
1466 offset += 2 * sizeof(__u16);
1467 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1470 __ip_vs_conn_put(cp);
1475 #ifdef CONFIG_IP_VS_IPV6
1476 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1477 unsigned int hooknum, struct ip_vs_iphdr *iph)
1479 struct net *net = NULL;
1480 struct ipv6hdr _ip6h, *ip6h;
1481 struct icmp6hdr _icmph, *ic;
1482 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1483 struct ip_vs_conn *cp;
1484 struct ip_vs_protocol *pp;
1485 struct ip_vs_proto_data *pd;
1486 unsigned int offs_ciph, writable, verdict;
1490 ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1495 * Work through seeing if this is for us.
1496 * These checks are supposed to be in an order that means easy
1497 * things are checked first to speed up processing.... however
1498 * this means that some packets will manage to get a long way
1499 * down this stack and then be rejected, but that's life.
1501 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1505 /* Fragment header that is before ICMP header tells us that:
1506 * it's not an error message since they can't be fragmented.
1508 if (iph->flags & IP6_FH_F_FRAG)
1511 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1512 ic->icmp6_type, ntohs(icmpv6_id(ic)),
1513 &iph->saddr, &iph->daddr);
1515 /* Now find the contained IP header */
1516 ciph.len = iph->len + sizeof(_icmph);
1517 offs_ciph = ciph.len; /* Save ip header offset */
1518 ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
1520 return NF_ACCEPT; /* The packet looks wrong, ignore */
1521 ciph.saddr.in6 = ip6h->saddr; /* conn_in_get() handles reverse order */
1522 ciph.daddr.in6 = ip6h->daddr;
1523 /* skip possible IPv6 exthdrs of contained IPv6 packet */
1524 ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
1525 if (ciph.protocol < 0)
1526 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
1529 pd = ip_vs_proto_data_get(net, ciph.protocol);
1534 /* Cannot handle fragmented embedded protocol */
1538 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offs_ciph,
1539 "Checking incoming ICMPv6 for");
1541 /* The embedded headers contain source and dest in reverse order
1542 * if not from localhost
1544 cp = pp->conn_in_get(AF_INET6, skb, &ciph,
1545 (hooknum == NF_INET_LOCAL_OUT) ? 0 : 1);
1549 /* VS/TUN, VS/DR and LOCALNODE just let it go */
1550 if ((hooknum == NF_INET_LOCAL_OUT) &&
1551 (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1552 __ip_vs_conn_put(cp);
1556 /* do the statistics and put it back */
1557 ip_vs_in_stats(cp, skb);
1559 /* Need to mangle contained IPv6 header in ICMPv6 packet */
1560 writable = ciph.len;
1561 if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1562 IPPROTO_SCTP == ciph.protocol)
1563 writable += 2 * sizeof(__u16); /* Also mangle ports */
1565 verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, writable, hooknum, &ciph);
1567 __ip_vs_conn_put(cp);
1575 * Check if it's for virtual services, look it up,
1576 * and send it on its way...
1579 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1582 struct ip_vs_iphdr iph;
1583 struct ip_vs_protocol *pp;
1584 struct ip_vs_proto_data *pd;
1585 struct ip_vs_conn *cp;
1587 struct netns_ipvs *ipvs;
1589 /* Already marked as IPVS request or reply? */
1590 if (skb->ipvs_property)
1595 * - remote client: only PACKET_HOST
1596 * - route: used for struct net when skb->dev is unset
1598 if (unlikely((skb->pkt_type != PACKET_HOST &&
1599 hooknum != NF_INET_LOCAL_OUT) ||
1601 ip_vs_fill_iph_skb(af, skb, &iph);
1602 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1603 " ignored in hook %u\n",
1604 skb->pkt_type, iph.protocol,
1605 IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1608 /* ipvs enabled in this netns ? */
1610 ipvs = net_ipvs(net);
1611 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1614 ip_vs_fill_iph_skb(af, skb, &iph);
1616 /* Bad... Do not break raw sockets */
1617 if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1619 struct sock *sk = skb->sk;
1620 struct inet_sock *inet = inet_sk(skb->sk);
1622 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1626 #ifdef CONFIG_IP_VS_IPV6
1627 if (af == AF_INET6) {
1628 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1630 int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1638 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1640 int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1646 /* Protocol supported? */
1647 pd = ip_vs_proto_data_get(net, iph.protocol);
1652 * Check if the packet belongs to an existing connection entry
1654 cp = pp->conn_in_get(af, skb, &iph, 0);
1656 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp && cp->dest &&
1657 unlikely(!atomic_read(&cp->dest->weight)) && !iph.fragoffs &&
1658 is_new_conn(skb, &iph)) {
1659 ip_vs_conn_expire_now(cp);
1660 __ip_vs_conn_put(cp);
1664 if (unlikely(!cp) && !iph.fragoffs) {
1665 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1666 * replayed fragment zero will already have created the cp
1670 /* Schedule and create new connection entry into &cp */
1671 if (!pp->conn_schedule(af, skb, pd, &v, &cp, &iph))
1675 if (unlikely(!cp)) {
1676 /* sorry, all this trouble for a no-hit :) */
1677 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1678 "ip_vs_in: packet continues traversal as normal");
1680 /* Fragment that couldn't be mapped to a conn entry
1681 * is missing module nf_defrag_ipv6
1683 IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1684 IP_VS_DBG_PKT(7, af, pp, skb, 0, "unhandled fragment");
1689 IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1690 /* Check the server status */
1691 if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1692 /* the destination server is not available */
1694 if (sysctl_expire_nodest_conn(ipvs)) {
1695 /* try to expire the connection immediately */
1696 ip_vs_conn_expire_now(cp);
1698 /* don't restart its timer, and silently
1700 __ip_vs_conn_put(cp);
1704 ip_vs_in_stats(cp, skb);
1705 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1706 if (cp->packet_xmit)
1707 ret = cp->packet_xmit(skb, cp, pp, &iph);
1708 /* do not touch skb anymore */
1710 IP_VS_DBG_RL("warning: packet_xmit is null");
1714 /* Increase its packet counter and check if it is needed
1715 * to be synchronized
1717 * Sync connection if it is about to close to
1718 * encorage the standby servers to update the connections timeout
1720 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1723 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1724 pkts = sysctl_sync_threshold(ipvs);
1726 pkts = atomic_add_return(1, &cp->in_pkts);
1728 if (ipvs->sync_state & IP_VS_STATE_MASTER)
1729 ip_vs_sync_conn(net, cp, pkts);
1736 * AF_INET handler in NF_INET_LOCAL_IN chain
1737 * Schedule and forward packets from remote clients
1740 ip_vs_remote_request4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1741 const struct net_device *in,
1742 const struct net_device *out,
1743 int (*okfn)(struct sk_buff *))
1745 return ip_vs_in(ops->hooknum, skb, AF_INET);
1749 * AF_INET handler in NF_INET_LOCAL_OUT chain
1750 * Schedule and forward packets from local clients
1753 ip_vs_local_request4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1754 const struct net_device *in, const struct net_device *out,
1755 int (*okfn)(struct sk_buff *))
1757 return ip_vs_in(ops->hooknum, skb, AF_INET);
1760 #ifdef CONFIG_IP_VS_IPV6
1763 * AF_INET6 handler in NF_INET_LOCAL_IN chain
1764 * Schedule and forward packets from remote clients
1767 ip_vs_remote_request6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1768 const struct net_device *in,
1769 const struct net_device *out,
1770 int (*okfn)(struct sk_buff *))
1772 return ip_vs_in(ops->hooknum, skb, AF_INET6);
1776 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
1777 * Schedule and forward packets from local clients
1780 ip_vs_local_request6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1781 const struct net_device *in, const struct net_device *out,
1782 int (*okfn)(struct sk_buff *))
1784 return ip_vs_in(ops->hooknum, skb, AF_INET6);
1791 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1792 * related packets destined for 0.0.0.0/0.
1793 * When fwmark-based virtual service is used, such as transparent
1794 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1795 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1796 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1797 * and send them to ip_vs_in_icmp.
1800 ip_vs_forward_icmp(const struct nf_hook_ops *ops, struct sk_buff *skb,
1801 const struct net_device *in, const struct net_device *out,
1802 int (*okfn)(struct sk_buff *))
1806 struct netns_ipvs *ipvs;
1808 if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1811 /* ipvs enabled in this netns ? */
1813 ipvs = net_ipvs(net);
1814 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1817 return ip_vs_in_icmp(skb, &r, ops->hooknum);
1820 #ifdef CONFIG_IP_VS_IPV6
1822 ip_vs_forward_icmp_v6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1823 const struct net_device *in, const struct net_device *out,
1824 int (*okfn)(struct sk_buff *))
1828 struct netns_ipvs *ipvs;
1829 struct ip_vs_iphdr iphdr;
1831 ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
1832 if (iphdr.protocol != IPPROTO_ICMPV6)
1835 /* ipvs enabled in this netns ? */
1837 ipvs = net_ipvs(net);
1838 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1841 return ip_vs_in_icmp_v6(skb, &r, ops->hooknum, &iphdr);
1846 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1847 /* After packet filtering, change source only for VS/NAT */
1849 .hook = ip_vs_reply4,
1850 .owner = THIS_MODULE,
1852 .hooknum = NF_INET_LOCAL_IN,
1853 .priority = NF_IP_PRI_NAT_SRC - 2,
1855 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1856 * or VS/NAT(change destination), so that filtering rules can be
1857 * applied to IPVS. */
1859 .hook = ip_vs_remote_request4,
1860 .owner = THIS_MODULE,
1862 .hooknum = NF_INET_LOCAL_IN,
1863 .priority = NF_IP_PRI_NAT_SRC - 1,
1865 /* Before ip_vs_in, change source only for VS/NAT */
1867 .hook = ip_vs_local_reply4,
1868 .owner = THIS_MODULE,
1870 .hooknum = NF_INET_LOCAL_OUT,
1871 .priority = NF_IP_PRI_NAT_DST + 1,
1873 /* After mangle, schedule and forward local requests */
1875 .hook = ip_vs_local_request4,
1876 .owner = THIS_MODULE,
1878 .hooknum = NF_INET_LOCAL_OUT,
1879 .priority = NF_IP_PRI_NAT_DST + 2,
1881 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1882 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1884 .hook = ip_vs_forward_icmp,
1885 .owner = THIS_MODULE,
1887 .hooknum = NF_INET_FORWARD,
1890 /* After packet filtering, change source only for VS/NAT */
1892 .hook = ip_vs_reply4,
1893 .owner = THIS_MODULE,
1895 .hooknum = NF_INET_FORWARD,
1898 #ifdef CONFIG_IP_VS_IPV6
1899 /* After packet filtering, change source only for VS/NAT */
1901 .hook = ip_vs_reply6,
1902 .owner = THIS_MODULE,
1904 .hooknum = NF_INET_LOCAL_IN,
1905 .priority = NF_IP6_PRI_NAT_SRC - 2,
1907 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1908 * or VS/NAT(change destination), so that filtering rules can be
1909 * applied to IPVS. */
1911 .hook = ip_vs_remote_request6,
1912 .owner = THIS_MODULE,
1914 .hooknum = NF_INET_LOCAL_IN,
1915 .priority = NF_IP6_PRI_NAT_SRC - 1,
1917 /* Before ip_vs_in, change source only for VS/NAT */
1919 .hook = ip_vs_local_reply6,
1920 .owner = THIS_MODULE,
1922 .hooknum = NF_INET_LOCAL_OUT,
1923 .priority = NF_IP6_PRI_NAT_DST + 1,
1925 /* After mangle, schedule and forward local requests */
1927 .hook = ip_vs_local_request6,
1928 .owner = THIS_MODULE,
1930 .hooknum = NF_INET_LOCAL_OUT,
1931 .priority = NF_IP6_PRI_NAT_DST + 2,
1933 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1934 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1936 .hook = ip_vs_forward_icmp_v6,
1937 .owner = THIS_MODULE,
1939 .hooknum = NF_INET_FORWARD,
1942 /* After packet filtering, change source only for VS/NAT */
1944 .hook = ip_vs_reply6,
1945 .owner = THIS_MODULE,
1947 .hooknum = NF_INET_FORWARD,
1953 * Initialize IP Virtual Server netns mem.
1955 static int __net_init __ip_vs_init(struct net *net)
1957 struct netns_ipvs *ipvs;
1959 ipvs = net_generic(net, ip_vs_net_id);
1963 /* Hold the beast until a service is registerd */
1966 /* Counters used for creating unique names */
1967 ipvs->gen = atomic_read(&ipvs_netns_cnt);
1968 atomic_inc(&ipvs_netns_cnt);
1971 if (ip_vs_estimator_net_init(net) < 0)
1972 goto estimator_fail;
1974 if (ip_vs_control_net_init(net) < 0)
1977 if (ip_vs_protocol_net_init(net) < 0)
1980 if (ip_vs_app_net_init(net) < 0)
1983 if (ip_vs_conn_net_init(net) < 0)
1986 if (ip_vs_sync_net_init(net) < 0)
1989 printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
1990 sizeof(struct netns_ipvs), ipvs->gen);
1997 ip_vs_conn_net_cleanup(net);
1999 ip_vs_app_net_cleanup(net);
2001 ip_vs_protocol_net_cleanup(net);
2003 ip_vs_control_net_cleanup(net);
2005 ip_vs_estimator_net_cleanup(net);
2011 static void __net_exit __ip_vs_cleanup(struct net *net)
2013 ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2014 ip_vs_conn_net_cleanup(net);
2015 ip_vs_app_net_cleanup(net);
2016 ip_vs_protocol_net_cleanup(net);
2017 ip_vs_control_net_cleanup(net);
2018 ip_vs_estimator_net_cleanup(net);
2019 IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2023 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2026 net_ipvs(net)->enable = 0; /* Disable packet reception */
2028 ip_vs_sync_net_cleanup(net);
2032 static struct pernet_operations ipvs_core_ops = {
2033 .init = __ip_vs_init,
2034 .exit = __ip_vs_cleanup,
2035 .id = &ip_vs_net_id,
2036 .size = sizeof(struct netns_ipvs),
2039 static struct pernet_operations ipvs_core_dev_ops = {
2040 .exit = __ip_vs_dev_cleanup,
2044 * Initialize IP Virtual Server
2046 static int __init ip_vs_init(void)
2050 ret = ip_vs_control_init();
2052 pr_err("can't setup control.\n");
2056 ip_vs_protocol_init();
2058 ret = ip_vs_conn_init();
2060 pr_err("can't setup connection table.\n");
2061 goto cleanup_protocol;
2064 ret = register_pernet_subsys(&ipvs_core_ops); /* Alloc ip_vs struct */
2068 ret = register_pernet_device(&ipvs_core_dev_ops);
2072 ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2074 pr_err("can't register hooks.\n");
2078 ret = ip_vs_register_nl_ioctl();
2080 pr_err("can't register netlink/ioctl.\n");
2084 pr_info("ipvs loaded.\n");
2089 nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2091 unregister_pernet_device(&ipvs_core_dev_ops);
2093 unregister_pernet_subsys(&ipvs_core_ops);
2095 ip_vs_conn_cleanup();
2097 ip_vs_protocol_cleanup();
2098 ip_vs_control_cleanup();
2103 static void __exit ip_vs_cleanup(void)
2105 ip_vs_unregister_nl_ioctl();
2106 nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2107 unregister_pernet_device(&ipvs_core_dev_ops);
2108 unregister_pernet_subsys(&ipvs_core_ops); /* free ip_vs struct */
2109 ip_vs_conn_cleanup();
2110 ip_vs_protocol_cleanup();
2111 ip_vs_control_cleanup();
2112 pr_info("ipvs unloaded.\n");
2115 module_init(ip_vs_init);
2116 module_exit(ip_vs_cleanup);
2117 MODULE_LICENSE("GPL");