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 * Version: $Id: ip_vs_core.c,v 1.34 2003/05/10 03:05:23 wensong Exp $
10 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
11 * Peter Kese <peter.kese@ijs.si>
12 * Julian Anastasov <ja@ssi.bg>
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public License
16 * as published by the Free Software Foundation; either version
17 * 2 of the License, or (at your option) any later version.
19 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
20 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
24 * Paul `Rusty' Russell properly handle non-linear skbs
28 #include <linux/module.h>
29 #include <linux/kernel.h>
31 #include <linux/tcp.h>
32 #include <linux/icmp.h>
37 #include <net/icmp.h> /* for icmp_send */
38 #include <net/route.h>
40 #include <linux/netfilter.h>
41 #include <linux/netfilter_ipv4.h>
43 #include <net/ip_vs.h>
46 EXPORT_SYMBOL(register_ip_vs_scheduler);
47 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
48 EXPORT_SYMBOL(ip_vs_skb_replace);
49 EXPORT_SYMBOL(ip_vs_proto_name);
50 EXPORT_SYMBOL(ip_vs_conn_new);
51 EXPORT_SYMBOL(ip_vs_conn_in_get);
52 EXPORT_SYMBOL(ip_vs_conn_out_get);
53 #ifdef CONFIG_IP_VS_PROTO_TCP
54 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
56 EXPORT_SYMBOL(ip_vs_conn_put);
57 #ifdef CONFIG_IP_VS_DEBUG
58 EXPORT_SYMBOL(ip_vs_get_debug_level);
60 EXPORT_SYMBOL(check_for_ip_vs_out);
61 EXPORT_SYMBOL(ip_vs_make_skb_writable);
64 /* ID used in ICMP lookups */
65 #define icmp_id(icmph) (((icmph)->un).echo.id)
67 const char *ip_vs_proto_name(unsigned proto)
81 sprintf(buf, "IP_%d", proto);
86 void ip_vs_init_hash_table(struct list_head *table, int rows)
89 INIT_LIST_HEAD(&table[rows]);
93 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
95 struct ip_vs_dest *dest = cp->dest;
96 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
97 spin_lock(&dest->stats.lock);
99 dest->stats.inbytes += skb->len;
100 spin_unlock(&dest->stats.lock);
102 spin_lock(&dest->svc->stats.lock);
103 dest->svc->stats.inpkts++;
104 dest->svc->stats.inbytes += skb->len;
105 spin_unlock(&dest->svc->stats.lock);
107 spin_lock(&ip_vs_stats.lock);
108 ip_vs_stats.inpkts++;
109 ip_vs_stats.inbytes += skb->len;
110 spin_unlock(&ip_vs_stats.lock);
116 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
118 struct ip_vs_dest *dest = cp->dest;
119 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
120 spin_lock(&dest->stats.lock);
121 dest->stats.outpkts++;
122 dest->stats.outbytes += skb->len;
123 spin_unlock(&dest->stats.lock);
125 spin_lock(&dest->svc->stats.lock);
126 dest->svc->stats.outpkts++;
127 dest->svc->stats.outbytes += skb->len;
128 spin_unlock(&dest->svc->stats.lock);
130 spin_lock(&ip_vs_stats.lock);
131 ip_vs_stats.outpkts++;
132 ip_vs_stats.outbytes += skb->len;
133 spin_unlock(&ip_vs_stats.lock);
139 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
141 spin_lock(&cp->dest->stats.lock);
142 cp->dest->stats.conns++;
143 spin_unlock(&cp->dest->stats.lock);
145 spin_lock(&svc->stats.lock);
147 spin_unlock(&svc->stats.lock);
149 spin_lock(&ip_vs_stats.lock);
151 spin_unlock(&ip_vs_stats.lock);
156 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
157 const struct sk_buff *skb,
158 struct ip_vs_protocol *pp)
160 if (unlikely(!pp->state_transition))
162 return pp->state_transition(cp, direction, skb, pp);
166 int ip_vs_make_skb_writable(struct sk_buff **pskb, int writable_len)
168 struct sk_buff *skb = *pskb;
170 /* skb is already used, better copy skb and its payload */
171 if (unlikely(skb_shared(skb) || skb->sk))
174 /* skb data is already used, copy it */
175 if (unlikely(skb_cloned(skb)))
178 return pskb_may_pull(skb, writable_len);
181 if (unlikely(writable_len > skb->len))
183 return !pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
186 if (unlikely(writable_len > skb->len))
188 skb = skb_copy(skb, GFP_ATOMIC);
191 BUG_ON(skb_is_nonlinear(skb));
193 /* Rest of kernel will get very unhappy if we pass it a
194 suddenly-orphaned skbuff */
196 skb_set_owner_w(skb, (*pskb)->sk);
203 * IPVS persistent scheduling function
204 * It creates a connection entry according to its template if exists,
205 * or selects a server and creates a connection entry plus a template.
206 * Locking: we are svc user (svc->refcnt), so we hold all dests too
207 * Protocols supported: TCP, UDP
209 static struct ip_vs_conn *
210 ip_vs_sched_persist(struct ip_vs_service *svc,
211 const struct sk_buff *skb,
214 struct ip_vs_conn *cp = NULL;
215 struct iphdr *iph = skb->nh.iph;
216 struct ip_vs_dest *dest;
217 struct ip_vs_conn *ct;
218 __u16 dport; /* destination port to forward */
219 __u32 snet; /* source network of the client, after masking */
221 /* Mask saddr with the netmask to adjust template granularity */
222 snet = iph->saddr & svc->netmask;
224 IP_VS_DBG(6, "p-schedule: src %u.%u.%u.%u:%u dest %u.%u.%u.%u:%u "
225 "mnet %u.%u.%u.%u\n",
226 NIPQUAD(iph->saddr), ntohs(ports[0]),
227 NIPQUAD(iph->daddr), ntohs(ports[1]),
231 * As far as we know, FTP is a very complicated network protocol, and
232 * it uses control connection and data connections. For active FTP,
233 * FTP server initialize data connection to the client, its source port
234 * is often 20. For passive FTP, FTP server tells the clients the port
235 * that it passively listens to, and the client issues the data
236 * connection. In the tunneling or direct routing mode, the load
237 * balancer is on the client-to-server half of connection, the port
238 * number is unknown to the load balancer. So, a conn template like
239 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
240 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
241 * is created for other persistent services.
243 if (ports[1] == svc->port) {
244 /* Check if a template already exists */
245 if (svc->port != FTPPORT)
246 ct = ip_vs_conn_in_get(iph->protocol, snet, 0,
247 iph->daddr, ports[1]);
249 ct = ip_vs_conn_in_get(iph->protocol, snet, 0,
252 if (!ct || !ip_vs_check_template(ct)) {
254 * No template found or the dest of the connection
255 * template is not available.
257 dest = svc->scheduler->schedule(svc, skb);
259 IP_VS_DBG(1, "p-schedule: no dest found.\n");
264 * Create a template like <protocol,caddr,0,
265 * vaddr,vport,daddr,dport> for non-ftp service,
266 * and <protocol,caddr,0,vaddr,0,daddr,0>
269 if (svc->port != FTPPORT)
270 ct = ip_vs_conn_new(iph->protocol,
274 dest->addr, dest->port,
278 ct = ip_vs_conn_new(iph->protocol,
287 ct->timeout = svc->timeout;
289 /* set destination with the found template */
295 * Note: persistent fwmark-based services and persistent
296 * port zero service are handled here.
297 * fwmark template: <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
298 * port zero template: <protocol,caddr,0,vaddr,0,daddr,0>
301 ct = ip_vs_conn_in_get(IPPROTO_IP, snet, 0,
302 htonl(svc->fwmark), 0);
304 ct = ip_vs_conn_in_get(iph->protocol, snet, 0,
307 if (!ct || !ip_vs_check_template(ct)) {
309 * If it is not persistent port zero, return NULL,
310 * otherwise create a connection template.
315 dest = svc->scheduler->schedule(svc, skb);
317 IP_VS_DBG(1, "p-schedule: no dest found.\n");
322 * Create a template according to the service
325 ct = ip_vs_conn_new(IPPROTO_IP,
327 htonl(svc->fwmark), 0,
332 ct = ip_vs_conn_new(iph->protocol,
341 ct->timeout = svc->timeout;
343 /* set destination with the found template */
350 * Create a new connection according to the template
352 cp = ip_vs_conn_new(iph->protocol,
353 iph->saddr, ports[0],
354 iph->daddr, ports[1],
366 ip_vs_control_add(cp, ct);
369 ip_vs_conn_stats(cp, svc);
375 * IPVS main scheduling function
376 * It selects a server according to the virtual service, and
377 * creates a connection entry.
378 * Protocols supported: TCP, UDP
381 ip_vs_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
383 struct ip_vs_conn *cp = NULL;
384 struct iphdr *iph = skb->nh.iph;
385 struct ip_vs_dest *dest;
388 if (skb_copy_bits(skb, iph->ihl*4, ports, sizeof(ports)) < 0)
394 if (svc->flags & IP_VS_SVC_F_PERSISTENT)
395 return ip_vs_sched_persist(svc, skb, ports);
398 * Non-persistent service
400 if (!svc->fwmark && ports[1] != svc->port) {
402 IP_VS_ERR("Schedule: port zero only supported "
403 "in persistent services, "
404 "check your ipvs configuration\n");
408 dest = svc->scheduler->schedule(svc, skb);
410 IP_VS_DBG(1, "Schedule: no dest found.\n");
415 * Create a connection entry.
417 cp = ip_vs_conn_new(iph->protocol,
418 iph->saddr, ports[0],
419 iph->daddr, ports[1],
420 dest->addr, dest->port?dest->port:ports[1],
426 IP_VS_DBG(6, "Schedule fwd:%c c:%u.%u.%u.%u:%u v:%u.%u.%u.%u:%u "
427 "d:%u.%u.%u.%u:%u flg:%X cnt:%d\n",
429 NIPQUAD(cp->caddr), ntohs(cp->cport),
430 NIPQUAD(cp->vaddr), ntohs(cp->vport),
431 NIPQUAD(cp->daddr), ntohs(cp->dport),
432 cp->flags, atomic_read(&cp->refcnt));
434 ip_vs_conn_stats(cp, svc);
440 * Pass or drop the packet.
441 * Called by ip_vs_in, when the virtual service is available but
442 * no destination is available for a new connection.
444 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
445 struct ip_vs_protocol *pp)
448 struct iphdr *iph = skb->nh.iph;
450 if (skb_copy_bits(skb, iph->ihl*4, ports, sizeof(ports)) < 0) {
451 ip_vs_service_put(svc);
455 /* if it is fwmark-based service, the cache_bypass sysctl is up
456 and the destination is RTN_UNICAST (and not local), then create
457 a cache_bypass connection entry */
458 if (sysctl_ip_vs_cache_bypass && svc->fwmark
459 && (inet_addr_type(iph->daddr) == RTN_UNICAST)) {
461 struct ip_vs_conn *cp;
463 ip_vs_service_put(svc);
465 /* create a new connection entry */
466 IP_VS_DBG(6, "ip_vs_leave: create a cache_bypass entry\n");
467 cp = ip_vs_conn_new(iph->protocol,
468 iph->saddr, ports[0],
469 iph->daddr, ports[1],
477 ip_vs_in_stats(cp, skb);
480 cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
482 /* transmit the first SYN packet */
483 ret = cp->packet_xmit(skb, cp, pp);
484 /* do not touch skb anymore */
486 atomic_inc(&cp->in_pkts);
492 * When the virtual ftp service is presented, packets destined
493 * for other services on the VIP may get here (except services
494 * listed in the ipvs table), pass the packets, because it is
495 * not ipvs job to decide to drop the packets.
497 if ((svc->port == FTPPORT) && (ports[1] != FTPPORT)) {
498 ip_vs_service_put(svc);
502 ip_vs_service_put(svc);
505 * Notify the client that the destination is unreachable, and
506 * release the socket buffer.
507 * Since it is in IP layer, the TCP socket is not actually
508 * created, the TCP RST packet cannot be sent, instead that
509 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
511 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
517 * It is hooked before NF_IP_PRI_NAT_SRC at the NF_IP_POST_ROUTING
518 * chain, and is used for VS/NAT.
519 * It detects packets for VS/NAT connections and sends the packets
520 * immediately. This can avoid that iptable_nat mangles the packets
523 static unsigned int ip_vs_post_routing(unsigned int hooknum,
524 struct sk_buff **pskb,
525 const struct net_device *in,
526 const struct net_device *out,
527 int (*okfn)(struct sk_buff *))
529 if (!((*pskb)->nfcache & NFC_IPVS_PROPERTY))
532 /* The packet was sent from IPVS, exit this chain */
538 u16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
540 return (u16) csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
543 static inline struct sk_buff *
544 ip_vs_gather_frags(struct sk_buff *skb)
546 skb = ip_defrag(skb);
548 ip_send_check(skb->nh.iph);
553 * Packet has been made sufficiently writable in caller
554 * - inout: 1=in->out, 0=out->in
556 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
557 struct ip_vs_conn *cp, int inout)
559 struct iphdr *iph = skb->nh.iph;
560 unsigned int icmp_offset = iph->ihl*4;
561 struct icmphdr *icmph = (struct icmphdr *)(skb->nh.raw + icmp_offset);
562 struct iphdr *ciph = (struct iphdr *)(icmph + 1);
565 iph->saddr = cp->vaddr;
567 ciph->daddr = cp->vaddr;
570 iph->daddr = cp->daddr;
572 ciph->saddr = cp->daddr;
576 /* the TCP/UDP port */
577 if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol) {
578 __u16 *ports = (void *)ciph + ciph->ihl*4;
581 ports[1] = cp->vport;
583 ports[0] = cp->dport;
586 /* And finally the ICMP checksum */
588 icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
589 skb->ip_summed = CHECKSUM_UNNECESSARY;
592 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
593 "Forwarding altered outgoing ICMP");
595 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
596 "Forwarding altered incoming ICMP");
600 * Handle ICMP messages in the inside-to-outside direction (outgoing).
601 * Find any that might be relevant, check against existing connections,
602 * forward to the right destination host if relevant.
603 * Currently handles error types - unreachable, quench, ttl exceeded.
604 * (Only used in VS/NAT)
606 static int ip_vs_out_icmp(struct sk_buff **pskb, int *related)
608 struct sk_buff *skb = *pskb;
610 struct icmphdr icmph;
611 struct iphdr ciph; /* The ip header contained within the ICMP */
612 struct ip_vs_conn *cp;
613 struct ip_vs_protocol *pp;
614 unsigned int offset, ihl, verdict;
618 /* reassemble IP fragments */
619 if (skb->nh.iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) {
620 skb = ip_vs_gather_frags(skb);
627 offset = ihl = iph->ihl * 4;
628 if (skb_copy_bits(skb, offset, &icmph, sizeof(icmph)) < 0)
631 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
632 icmph.type, ntohs(icmp_id(&icmph)),
633 NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
636 * Work through seeing if this is for us.
637 * These checks are supposed to be in an order that means easy
638 * things are checked first to speed up processing.... however
639 * this means that some packets will manage to get a long way
640 * down this stack and then be rejected, but that's life.
642 if ((icmph.type != ICMP_DEST_UNREACH) &&
643 (icmph.type != ICMP_SOURCE_QUENCH) &&
644 (icmph.type != ICMP_TIME_EXCEEDED)) {
649 /* Now find the contained IP header */
650 offset += sizeof(icmph);
651 if (skb_copy_bits(skb, offset, &ciph, sizeof(ciph)) < 0)
652 return NF_ACCEPT; /* The packet looks wrong, ignore */
654 pp = ip_vs_proto_get(ciph.protocol);
658 /* Is the embedded protocol header present? */
659 if (unlikely(ciph.frag_off & __constant_htons(IP_OFFSET) &&
663 IP_VS_DBG_PKT(11, pp, skb, offset, "Checking outgoing ICMP for");
665 offset += ciph.ihl * 4;
667 /* The embedded headers contain source and dest in reverse order */
668 cp = pp->conn_out_get(skb, pp, &ciph, offset, 1);
674 if (IP_VS_FWD_METHOD(cp) != 0) {
675 IP_VS_ERR("shouldn't reach here, because the box is on the"
676 "half connection in the tun/dr module.\n");
679 /* Ensure the checksum is correct */
680 if (skb->ip_summed != CHECKSUM_UNNECESSARY &&
681 ip_vs_checksum_complete(skb, ihl)) {
682 /* Failed checksum! */
683 IP_VS_DBG(1, "Forward ICMP: failed checksum from %d.%d.%d.%d!\n",
684 NIPQUAD(iph->saddr));
688 if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol)
689 offset += 2 * sizeof(__u16);
690 if (!ip_vs_make_skb_writable(pskb, offset))
694 ip_vs_nat_icmp(skb, pp, cp, 1);
696 /* do the statistics and put it back */
697 ip_vs_out_stats(cp, skb);
699 skb->nfcache |= NFC_IPVS_PROPERTY;
703 __ip_vs_conn_put(cp);
708 static inline int is_tcp_reset(const struct sk_buff *skb)
712 if (skb_copy_bits(skb, skb->nh.iph->ihl * 4, &tcph, sizeof(tcph)) < 0)
718 * It is hooked at the NF_IP_FORWARD chain, used only for VS/NAT.
719 * Check if outgoing packet belongs to the established ip_vs_conn,
720 * rewrite addresses of the packet and send it on its way...
723 ip_vs_out(unsigned int hooknum, struct sk_buff **pskb,
724 const struct net_device *in, const struct net_device *out,
725 int (*okfn)(struct sk_buff *))
727 struct sk_buff *skb = *pskb;
729 struct ip_vs_protocol *pp;
730 struct ip_vs_conn *cp;
735 if (skb->nfcache & NFC_IPVS_PROPERTY)
738 if (skb->ip_summed == CHECKSUM_HW) {
739 if (skb_checksum_help(pskb, (out == NULL)))
746 if (unlikely(iph->protocol == IPPROTO_ICMP)) {
747 int related, verdict = ip_vs_out_icmp(pskb, &related);
755 pp = ip_vs_proto_get(iph->protocol);
759 /* reassemble IP fragments */
760 if (unlikely(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET) &&
762 skb = ip_vs_gather_frags(skb);
772 * Check if the packet belongs to an existing entry
774 cp = pp->conn_out_get(skb, pp, iph, ihl, 0);
777 if (sysctl_ip_vs_nat_icmp_send &&
778 (pp->protocol == IPPROTO_TCP ||
779 pp->protocol == IPPROTO_UDP)) {
782 if (skb_copy_bits(skb, ihl, ports, sizeof(ports)) < 0)
783 return NF_ACCEPT; /* Not for me */
784 if (ip_vs_lookup_real_service(iph->protocol,
785 iph->saddr, ports[0])) {
787 * Notify the real server: there is no
788 * existing entry if it is not RST
789 * packet or not TCP packet.
791 if (iph->protocol != IPPROTO_TCP
792 || !is_tcp_reset(skb)) {
793 icmp_send(skb,ICMP_DEST_UNREACH,
794 ICMP_PORT_UNREACH, 0);
799 IP_VS_DBG_PKT(12, pp, skb, 0,
800 "packet continues traversal as normal");
804 IP_VS_DBG_PKT(11, pp, skb, 0, "Outgoing packet");
806 if (!ip_vs_make_skb_writable(pskb, ihl))
809 /* mangle the packet */
810 if (pp->snat_handler && !pp->snat_handler(pskb, pp, cp))
813 skb->nh.iph->saddr = cp->vaddr;
814 ip_send_check(skb->nh.iph);
816 IP_VS_DBG_PKT(10, pp, skb, 0, "After SNAT");
818 ip_vs_out_stats(cp, skb);
819 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pp);
822 skb->nfcache |= NFC_IPVS_PROPERTY;
835 * Check if the packet is for VS/NAT connections, then send it
837 * Called by ip_fw_compact to detect packets for VS/NAT before
838 * they are changed by ipchains masquerading code.
841 check_for_ip_vs_out(struct sk_buff **pskb, int (*okfn)(struct sk_buff *))
845 ret = ip_vs_out(NF_IP_FORWARD, pskb, NULL, NULL, NULL);
846 if (ret != NF_ACCEPT) {
849 /* send the packet immediately if it is already mangled
851 if ((*pskb)->nfcache & NFC_IPVS_PROPERTY) {
860 * Handle ICMP messages in the outside-to-inside direction (incoming).
861 * Find any that might be relevant, check against existing connections,
862 * forward to the right destination host if relevant.
863 * Currently handles error types - unreachable, quench, ttl exceeded.
865 static int ip_vs_in_icmp(struct sk_buff **pskb, int *related)
867 struct sk_buff *skb = *pskb;
869 struct icmphdr icmph;
870 struct iphdr ciph; /* The ip header contained within the ICMP */
871 struct ip_vs_conn *cp;
872 struct ip_vs_protocol *pp;
873 unsigned int offset, ihl, verdict;
877 /* reassemble IP fragments */
878 if (skb->nh.iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) {
879 skb = ip_vs_gather_frags(skb);
886 offset = ihl = iph->ihl * 4;
887 if (skb_copy_bits(skb, offset, &icmph, sizeof(icmph)) < 0)
890 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
891 icmph.type, ntohs(icmp_id(&icmph)),
892 NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
895 * Work through seeing if this is for us.
896 * These checks are supposed to be in an order that means easy
897 * things are checked first to speed up processing.... however
898 * this means that some packets will manage to get a long way
899 * down this stack and then be rejected, but that's life.
901 if ((icmph.type != ICMP_DEST_UNREACH) &&
902 (icmph.type != ICMP_SOURCE_QUENCH) &&
903 (icmph.type != ICMP_TIME_EXCEEDED)) {
908 /* Now find the contained IP header */
909 offset += sizeof(icmph);
910 if (skb_copy_bits(skb, offset, &ciph, sizeof(ciph)) < 0)
911 return NF_ACCEPT; /* The packet looks wrong, ignore */
913 pp = ip_vs_proto_get(ciph.protocol);
917 /* Is the embedded protocol header present? */
918 if (unlikely(ciph.frag_off & __constant_htons(IP_OFFSET) &&
922 IP_VS_DBG_PKT(11, pp, skb, offset, "Checking incoming ICMP for");
924 offset += ciph.ihl * 4;
926 /* The embedded headers contain source and dest in reverse order */
927 cp = pp->conn_in_get(skb, pp, &ciph, offset, 1);
933 /* Ensure the checksum is correct */
934 if (skb->ip_summed != CHECKSUM_UNNECESSARY &&
935 ip_vs_checksum_complete(skb, ihl)) {
936 /* Failed checksum! */
937 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %d.%d.%d.%d!\n",
938 NIPQUAD(iph->saddr));
942 /* do the statistics and put it back */
943 ip_vs_in_stats(cp, skb);
944 if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol)
945 offset += 2 * sizeof(__u16);
946 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
947 /* do not touch skb anymore */
950 __ip_vs_conn_put(cp);
956 * Check if it's for virtual services, look it up,
957 * and send it on its way...
960 ip_vs_in(unsigned int hooknum, struct sk_buff **pskb,
961 const struct net_device *in, const struct net_device *out,
962 int (*okfn)(struct sk_buff *))
964 struct sk_buff *skb = *pskb;
966 struct ip_vs_protocol *pp;
967 struct ip_vs_conn *cp;
972 * Big tappo: only PACKET_HOST (neither loopback nor mcasts)
973 * ... don't know why 1st test DOES NOT include 2nd (?)
975 if (unlikely(skb->pkt_type != PACKET_HOST
976 || skb->dev == &loopback_dev || skb->sk)) {
977 IP_VS_DBG(12, "packet type=%d proto=%d daddr=%d.%d.%d.%d ignored\n",
979 skb->nh.iph->protocol,
980 NIPQUAD(skb->nh.iph->daddr));
984 if (skb->ip_summed == CHECKSUM_HW) {
985 if (skb_checksum_help(pskb, (out == NULL)))
992 if (unlikely(iph->protocol == IPPROTO_ICMP)) {
993 int related, verdict = ip_vs_in_icmp(pskb, &related);
1001 /* Protocol supported? */
1002 pp = ip_vs_proto_get(iph->protocol);
1006 ihl = iph->ihl << 2;
1009 * Check if the packet belongs to an existing connection entry
1011 cp = pp->conn_in_get(skb, pp, iph, ihl, 0);
1013 if (unlikely(!cp)) {
1016 if (!pp->conn_schedule(skb, pp, &v, &cp))
1020 if (unlikely(!cp)) {
1021 /* sorry, all this trouble for a no-hit :) */
1022 IP_VS_DBG_PKT(12, pp, skb, 0,
1023 "packet continues traversal as normal");
1027 IP_VS_DBG_PKT(11, pp, skb, 0, "Incoming packet");
1029 /* Check the server status */
1030 if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1031 /* the destination server is not availabe */
1033 if (sysctl_ip_vs_expire_nodest_conn) {
1034 /* try to expire the connection immediately */
1035 ip_vs_conn_expire_now(cp);
1037 /* don't restart its timer, and silently
1039 __ip_vs_conn_put(cp);
1044 ip_vs_in_stats(cp, skb);
1045 restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
1046 if (cp->packet_xmit)
1047 ret = cp->packet_xmit(skb, cp, pp);
1048 /* do not touch skb anymore */
1050 IP_VS_DBG_RL("warning: packet_xmit is null");
1054 /* increase its packet counter and check if it is needed
1055 to be synchronized */
1056 atomic_inc(&cp->in_pkts);
1057 if ((ip_vs_sync_state & IP_VS_STATE_MASTER) &&
1058 (cp->protocol != IPPROTO_TCP ||
1059 cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1060 (atomic_read(&cp->in_pkts) % sysctl_ip_vs_sync_threshold[1]
1061 == sysctl_ip_vs_sync_threshold[0]))
1062 ip_vs_sync_conn(cp);
1070 * It is hooked at the NF_IP_FORWARD chain, in order to catch ICMP
1071 * related packets destined for 0.0.0.0/0.
1072 * When fwmark-based virtual service is used, such as transparent
1073 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1074 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1075 * sent to ip_vs_in_icmp. So, catch them at the NF_IP_FORWARD chain
1076 * and send them to ip_vs_in_icmp.
1079 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff **pskb,
1080 const struct net_device *in, const struct net_device *out,
1081 int (*okfn)(struct sk_buff *))
1085 if ((*pskb)->nh.iph->protocol != IPPROTO_ICMP)
1088 return ip_vs_in_icmp(pskb, &r);
1092 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1093 or VS/NAT(change destination), so that filtering rules can be
1095 static struct nf_hook_ops ip_vs_in_ops = {
1097 .owner = THIS_MODULE,
1099 .hooknum = NF_IP_LOCAL_IN,
1103 /* After packet filtering, change source only for VS/NAT */
1104 static struct nf_hook_ops ip_vs_out_ops = {
1106 .owner = THIS_MODULE,
1108 .hooknum = NF_IP_FORWARD,
1112 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1113 destined for 0.0.0.0/0, which is for incoming IPVS connections */
1114 static struct nf_hook_ops ip_vs_forward_icmp_ops = {
1115 .hook = ip_vs_forward_icmp,
1116 .owner = THIS_MODULE,
1118 .hooknum = NF_IP_FORWARD,
1122 /* Before the netfilter connection tracking, exit from POST_ROUTING */
1123 static struct nf_hook_ops ip_vs_post_routing_ops = {
1124 .hook = ip_vs_post_routing,
1125 .owner = THIS_MODULE,
1127 .hooknum = NF_IP_POST_ROUTING,
1128 .priority = NF_IP_PRI_NAT_SRC-1,
1133 * Initialize IP Virtual Server
1135 static int __init ip_vs_init(void)
1139 ret = ip_vs_control_init();
1141 IP_VS_ERR("can't setup control.\n");
1142 goto cleanup_nothing;
1145 ip_vs_protocol_init();
1147 ret = ip_vs_app_init();
1149 IP_VS_ERR("can't setup application helper.\n");
1150 goto cleanup_protocol;
1153 ret = ip_vs_conn_init();
1155 IP_VS_ERR("can't setup connection table.\n");
1159 ret = nf_register_hook(&ip_vs_in_ops);
1161 IP_VS_ERR("can't register in hook.\n");
1165 ret = nf_register_hook(&ip_vs_out_ops);
1167 IP_VS_ERR("can't register out hook.\n");
1170 ret = nf_register_hook(&ip_vs_post_routing_ops);
1172 IP_VS_ERR("can't register post_routing hook.\n");
1173 goto cleanup_outops;
1175 ret = nf_register_hook(&ip_vs_forward_icmp_ops);
1177 IP_VS_ERR("can't register forward_icmp hook.\n");
1178 goto cleanup_postroutingops;
1181 IP_VS_INFO("ipvs loaded.\n");
1184 cleanup_postroutingops:
1185 nf_unregister_hook(&ip_vs_post_routing_ops);
1187 nf_unregister_hook(&ip_vs_out_ops);
1189 nf_unregister_hook(&ip_vs_in_ops);
1191 ip_vs_conn_cleanup();
1193 ip_vs_app_cleanup();
1195 ip_vs_protocol_cleanup();
1196 ip_vs_control_cleanup();
1201 static void __exit ip_vs_cleanup(void)
1203 nf_unregister_hook(&ip_vs_forward_icmp_ops);
1204 nf_unregister_hook(&ip_vs_post_routing_ops);
1205 nf_unregister_hook(&ip_vs_out_ops);
1206 nf_unregister_hook(&ip_vs_in_ops);
1207 ip_vs_conn_cleanup();
1208 ip_vs_app_cleanup();
1209 ip_vs_protocol_cleanup();
1210 ip_vs_control_cleanup();
1211 IP_VS_INFO("ipvs unloaded.\n");
1214 module_init(ip_vs_init);
1215 module_exit(ip_vs_cleanup);
1216 MODULE_LICENSE("GPL");