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
25 * Harald Welte don't use nfcache
29 #include <linux/module.h>
30 #include <linux/kernel.h>
32 #include <linux/tcp.h>
33 #include <linux/icmp.h>
38 #include <net/icmp.h> /* for icmp_send */
39 #include <net/route.h>
41 #include <linux/netfilter.h>
42 #include <linux/netfilter_ipv4.h>
44 #include <net/ip_vs.h>
47 EXPORT_SYMBOL(register_ip_vs_scheduler
);
48 EXPORT_SYMBOL(unregister_ip_vs_scheduler
);
49 EXPORT_SYMBOL(ip_vs_skb_replace
);
50 EXPORT_SYMBOL(ip_vs_proto_name
);
51 EXPORT_SYMBOL(ip_vs_conn_new
);
52 EXPORT_SYMBOL(ip_vs_conn_in_get
);
53 EXPORT_SYMBOL(ip_vs_conn_out_get
);
54 #ifdef CONFIG_IP_VS_PROTO_TCP
55 EXPORT_SYMBOL(ip_vs_tcp_conn_listen
);
57 EXPORT_SYMBOL(ip_vs_conn_put
);
58 #ifdef CONFIG_IP_VS_DEBUG
59 EXPORT_SYMBOL(ip_vs_get_debug_level
);
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 __be16 dport
; /* destination port to forward */
219 __be32 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_ct_in_get(iph
->protocol
, snet
, 0,
247 iph
->daddr
, ports
[1]);
249 ct
= ip_vs_ct_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
,
275 IP_VS_CONN_F_TEMPLATE
,
278 ct
= ip_vs_conn_new(iph
->protocol
,
282 IP_VS_CONN_F_TEMPLATE
,
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_ct_in_get(IPPROTO_IP
, snet
, 0,
302 htonl(svc
->fwmark
), 0);
304 ct
= ip_vs_ct_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,
329 IP_VS_CONN_F_TEMPLATE
,
332 ct
= ip_vs_conn_new(iph
->protocol
,
336 IP_VS_CONN_F_TEMPLATE
,
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
;
386 __be16 _ports
[2], *pptr
;
388 pptr
= skb_header_pointer(skb
, iph
->ihl
*4,
389 sizeof(_ports
), _ports
);
396 if (svc
->flags
& IP_VS_SVC_F_PERSISTENT
)
397 return ip_vs_sched_persist(svc
, skb
, pptr
);
400 * Non-persistent service
402 if (!svc
->fwmark
&& pptr
[1] != svc
->port
) {
404 IP_VS_ERR("Schedule: port zero only supported "
405 "in persistent services, "
406 "check your ipvs configuration\n");
410 dest
= svc
->scheduler
->schedule(svc
, skb
);
412 IP_VS_DBG(1, "Schedule: no dest found.\n");
417 * Create a connection entry.
419 cp
= ip_vs_conn_new(iph
->protocol
,
422 dest
->addr
, dest
->port
?dest
->port
:pptr
[1],
428 IP_VS_DBG(6, "Schedule fwd:%c c:%u.%u.%u.%u:%u v:%u.%u.%u.%u:%u "
429 "d:%u.%u.%u.%u:%u conn->flags:%X conn->refcnt:%d\n",
431 NIPQUAD(cp
->caddr
), ntohs(cp
->cport
),
432 NIPQUAD(cp
->vaddr
), ntohs(cp
->vport
),
433 NIPQUAD(cp
->daddr
), ntohs(cp
->dport
),
434 cp
->flags
, atomic_read(&cp
->refcnt
));
436 ip_vs_conn_stats(cp
, svc
);
442 * Pass or drop the packet.
443 * Called by ip_vs_in, when the virtual service is available but
444 * no destination is available for a new connection.
446 int ip_vs_leave(struct ip_vs_service
*svc
, struct sk_buff
*skb
,
447 struct ip_vs_protocol
*pp
)
449 __be16 _ports
[2], *pptr
;
450 struct iphdr
*iph
= skb
->nh
.iph
;
452 pptr
= skb_header_pointer(skb
, iph
->ihl
*4,
453 sizeof(_ports
), _ports
);
455 ip_vs_service_put(svc
);
459 /* if it is fwmark-based service, the cache_bypass sysctl is up
460 and the destination is RTN_UNICAST (and not local), then create
461 a cache_bypass connection entry */
462 if (sysctl_ip_vs_cache_bypass
&& svc
->fwmark
463 && (inet_addr_type(iph
->daddr
) == RTN_UNICAST
)) {
465 struct ip_vs_conn
*cp
;
467 ip_vs_service_put(svc
);
469 /* create a new connection entry */
470 IP_VS_DBG(6, "ip_vs_leave: create a cache_bypass entry\n");
471 cp
= ip_vs_conn_new(iph
->protocol
,
481 ip_vs_in_stats(cp
, skb
);
484 cs
= ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pp
);
486 /* transmit the first SYN packet */
487 ret
= cp
->packet_xmit(skb
, cp
, pp
);
488 /* do not touch skb anymore */
490 atomic_inc(&cp
->in_pkts
);
496 * When the virtual ftp service is presented, packets destined
497 * for other services on the VIP may get here (except services
498 * listed in the ipvs table), pass the packets, because it is
499 * not ipvs job to decide to drop the packets.
501 if ((svc
->port
== FTPPORT
) && (pptr
[1] != FTPPORT
)) {
502 ip_vs_service_put(svc
);
506 ip_vs_service_put(svc
);
509 * Notify the client that the destination is unreachable, and
510 * release the socket buffer.
511 * Since it is in IP layer, the TCP socket is not actually
512 * created, the TCP RST packet cannot be sent, instead that
513 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
515 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
521 * It is hooked before NF_IP_PRI_NAT_SRC at the NF_IP_POST_ROUTING
522 * chain, and is used for VS/NAT.
523 * It detects packets for VS/NAT connections and sends the packets
524 * immediately. This can avoid that iptable_nat mangles the packets
527 static unsigned int ip_vs_post_routing(unsigned int hooknum
,
528 struct sk_buff
**pskb
,
529 const struct net_device
*in
,
530 const struct net_device
*out
,
531 int (*okfn
)(struct sk_buff
*))
533 if (!((*pskb
)->ipvs_property
))
535 /* The packet was sent from IPVS, exit this chain */
539 __sum16
ip_vs_checksum_complete(struct sk_buff
*skb
, int offset
)
541 return csum_fold(skb_checksum(skb
, offset
, skb
->len
- offset
, 0));
544 static inline struct sk_buff
*
545 ip_vs_gather_frags(struct sk_buff
*skb
, u_int32_t user
)
547 skb
= ip_defrag(skb
, user
);
549 ip_send_check(skb
->nh
.iph
);
554 * Packet has been made sufficiently writable in caller
555 * - inout: 1=in->out, 0=out->in
557 void ip_vs_nat_icmp(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
558 struct ip_vs_conn
*cp
, int inout
)
560 struct iphdr
*iph
= skb
->nh
.iph
;
561 unsigned int icmp_offset
= iph
->ihl
*4;
562 struct icmphdr
*icmph
= (struct icmphdr
*)(skb_network_header(skb
) +
564 struct iphdr
*ciph
= (struct iphdr
*)(icmph
+ 1);
567 iph
->saddr
= cp
->vaddr
;
569 ciph
->daddr
= cp
->vaddr
;
572 iph
->daddr
= cp
->daddr
;
574 ciph
->saddr
= cp
->daddr
;
578 /* the TCP/UDP port */
579 if (IPPROTO_TCP
== ciph
->protocol
|| IPPROTO_UDP
== ciph
->protocol
) {
580 __be16
*ports
= (void *)ciph
+ ciph
->ihl
*4;
583 ports
[1] = cp
->vport
;
585 ports
[0] = cp
->dport
;
588 /* And finally the ICMP checksum */
590 icmph
->checksum
= ip_vs_checksum_complete(skb
, icmp_offset
);
591 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
594 IP_VS_DBG_PKT(11, pp
, skb
, (void *)ciph
- (void *)iph
,
595 "Forwarding altered outgoing ICMP");
597 IP_VS_DBG_PKT(11, pp
, skb
, (void *)ciph
- (void *)iph
,
598 "Forwarding altered incoming ICMP");
602 * Handle ICMP messages in the inside-to-outside direction (outgoing).
603 * Find any that might be relevant, check against existing connections,
604 * forward to the right destination host if relevant.
605 * Currently handles error types - unreachable, quench, ttl exceeded.
606 * (Only used in VS/NAT)
608 static int ip_vs_out_icmp(struct sk_buff
**pskb
, int *related
)
610 struct sk_buff
*skb
= *pskb
;
612 struct icmphdr _icmph
, *ic
;
613 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
614 struct ip_vs_conn
*cp
;
615 struct ip_vs_protocol
*pp
;
616 unsigned int offset
, ihl
, verdict
;
620 /* reassemble IP fragments */
621 if (skb
->nh
.iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) {
622 skb
= ip_vs_gather_frags(skb
, IP_DEFRAG_VS_OUT
);
629 offset
= ihl
= iph
->ihl
* 4;
630 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
634 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
635 ic
->type
, ntohs(icmp_id(ic
)),
636 NIPQUAD(iph
->saddr
), NIPQUAD(iph
->daddr
));
639 * Work through seeing if this is for us.
640 * These checks are supposed to be in an order that means easy
641 * things are checked first to speed up processing.... however
642 * this means that some packets will manage to get a long way
643 * down this stack and then be rejected, but that's life.
645 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
646 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
647 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
652 /* Now find the contained IP header */
653 offset
+= sizeof(_icmph
);
654 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
656 return NF_ACCEPT
; /* The packet looks wrong, ignore */
658 pp
= ip_vs_proto_get(cih
->protocol
);
662 /* Is the embedded protocol header present? */
663 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
667 IP_VS_DBG_PKT(11, pp
, skb
, offset
, "Checking outgoing ICMP for");
669 offset
+= cih
->ihl
* 4;
671 /* The embedded headers contain source and dest in reverse order */
672 cp
= pp
->conn_out_get(skb
, pp
, cih
, offset
, 1);
678 if (IP_VS_FWD_METHOD(cp
) != 0) {
679 IP_VS_ERR("shouldn't reach here, because the box is on the"
680 "half connection in the tun/dr module.\n");
683 /* Ensure the checksum is correct */
684 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
&&
685 ip_vs_checksum_complete(skb
, ihl
)) {
686 /* Failed checksum! */
687 IP_VS_DBG(1, "Forward ICMP: failed checksum from %d.%d.%d.%d!\n",
688 NIPQUAD(iph
->saddr
));
692 if (IPPROTO_TCP
== cih
->protocol
|| IPPROTO_UDP
== cih
->protocol
)
693 offset
+= 2 * sizeof(__u16
);
694 if (!ip_vs_make_skb_writable(pskb
, offset
))
698 ip_vs_nat_icmp(skb
, pp
, cp
, 1);
700 /* do the statistics and put it back */
701 ip_vs_out_stats(cp
, skb
);
703 skb
->ipvs_property
= 1;
707 __ip_vs_conn_put(cp
);
712 static inline int is_tcp_reset(const struct sk_buff
*skb
)
714 struct tcphdr _tcph
, *th
;
716 th
= skb_header_pointer(skb
, ip_hdrlen(skb
), sizeof(_tcph
), &_tcph
);
723 * It is hooked at the NF_IP_FORWARD chain, used only for VS/NAT.
724 * Check if outgoing packet belongs to the established ip_vs_conn,
725 * rewrite addresses of the packet and send it on its way...
728 ip_vs_out(unsigned int hooknum
, struct sk_buff
**pskb
,
729 const struct net_device
*in
, const struct net_device
*out
,
730 int (*okfn
)(struct sk_buff
*))
732 struct sk_buff
*skb
= *pskb
;
734 struct ip_vs_protocol
*pp
;
735 struct ip_vs_conn
*cp
;
740 if (skb
->ipvs_property
)
744 if (unlikely(iph
->protocol
== IPPROTO_ICMP
)) {
745 int related
, verdict
= ip_vs_out_icmp(pskb
, &related
);
753 pp
= ip_vs_proto_get(iph
->protocol
);
757 /* reassemble IP fragments */
758 if (unlikely(iph
->frag_off
& htons(IP_MF
|IP_OFFSET
) &&
760 skb
= ip_vs_gather_frags(skb
, IP_DEFRAG_VS_OUT
);
770 * Check if the packet belongs to an existing entry
772 cp
= pp
->conn_out_get(skb
, pp
, iph
, ihl
, 0);
775 if (sysctl_ip_vs_nat_icmp_send
&&
776 (pp
->protocol
== IPPROTO_TCP
||
777 pp
->protocol
== IPPROTO_UDP
)) {
778 __be16 _ports
[2], *pptr
;
780 pptr
= skb_header_pointer(skb
, ihl
,
781 sizeof(_ports
), _ports
);
783 return NF_ACCEPT
; /* Not for me */
784 if (ip_vs_lookup_real_service(iph
->protocol
,
785 iph
->saddr
, pptr
[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 /* For policy routing, packets originating from this
817 * machine itself may be routed differently to packets
818 * passing through. We want this packet to be routed as
819 * if it came from this machine itself. So re-compute
820 * the routing information.
822 if (ip_route_me_harder(pskb
, RTN_LOCAL
) != 0)
826 IP_VS_DBG_PKT(10, pp
, skb
, 0, "After SNAT");
828 ip_vs_out_stats(cp
, skb
);
829 ip_vs_set_state(cp
, IP_VS_DIR_OUTPUT
, skb
, pp
);
832 skb
->ipvs_property
= 1;
845 * Handle ICMP messages in the outside-to-inside direction (incoming).
846 * Find any that might be relevant, check against existing connections,
847 * forward to the right destination host if relevant.
848 * Currently handles error types - unreachable, quench, ttl exceeded.
851 ip_vs_in_icmp(struct sk_buff
**pskb
, int *related
, unsigned int hooknum
)
853 struct sk_buff
*skb
= *pskb
;
855 struct icmphdr _icmph
, *ic
;
856 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
857 struct ip_vs_conn
*cp
;
858 struct ip_vs_protocol
*pp
;
859 unsigned int offset
, ihl
, verdict
;
863 /* reassemble IP fragments */
864 if (skb
->nh
.iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) {
865 skb
= ip_vs_gather_frags(skb
,
866 hooknum
== NF_IP_LOCAL_IN
?
867 IP_DEFRAG_VS_IN
: IP_DEFRAG_VS_FWD
);
874 offset
= ihl
= iph
->ihl
* 4;
875 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
879 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
880 ic
->type
, ntohs(icmp_id(ic
)),
881 NIPQUAD(iph
->saddr
), NIPQUAD(iph
->daddr
));
884 * Work through seeing if this is for us.
885 * These checks are supposed to be in an order that means easy
886 * things are checked first to speed up processing.... however
887 * this means that some packets will manage to get a long way
888 * down this stack and then be rejected, but that's life.
890 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
891 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
892 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
897 /* Now find the contained IP header */
898 offset
+= sizeof(_icmph
);
899 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
901 return NF_ACCEPT
; /* The packet looks wrong, ignore */
903 pp
= ip_vs_proto_get(cih
->protocol
);
907 /* Is the embedded protocol header present? */
908 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
912 IP_VS_DBG_PKT(11, pp
, skb
, offset
, "Checking incoming ICMP for");
914 offset
+= cih
->ihl
* 4;
916 /* The embedded headers contain source and dest in reverse order */
917 cp
= pp
->conn_in_get(skb
, pp
, cih
, offset
, 1);
923 /* Ensure the checksum is correct */
924 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
&&
925 ip_vs_checksum_complete(skb
, ihl
)) {
926 /* Failed checksum! */
927 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %d.%d.%d.%d!\n",
928 NIPQUAD(iph
->saddr
));
932 /* do the statistics and put it back */
933 ip_vs_in_stats(cp
, skb
);
934 if (IPPROTO_TCP
== cih
->protocol
|| IPPROTO_UDP
== cih
->protocol
)
935 offset
+= 2 * sizeof(__u16
);
936 verdict
= ip_vs_icmp_xmit(skb
, cp
, pp
, offset
);
937 /* do not touch skb anymore */
940 __ip_vs_conn_put(cp
);
946 * Check if it's for virtual services, look it up,
947 * and send it on its way...
950 ip_vs_in(unsigned int hooknum
, struct sk_buff
**pskb
,
951 const struct net_device
*in
, const struct net_device
*out
,
952 int (*okfn
)(struct sk_buff
*))
954 struct sk_buff
*skb
= *pskb
;
956 struct ip_vs_protocol
*pp
;
957 struct ip_vs_conn
*cp
;
962 * Big tappo: only PACKET_HOST (neither loopback nor mcasts)
963 * ... don't know why 1st test DOES NOT include 2nd (?)
965 if (unlikely(skb
->pkt_type
!= PACKET_HOST
966 || skb
->dev
== &loopback_dev
|| skb
->sk
)) {
967 IP_VS_DBG(12, "packet type=%d proto=%d daddr=%d.%d.%d.%d ignored\n",
969 skb
->nh
.iph
->protocol
,
970 NIPQUAD(skb
->nh
.iph
->daddr
));
975 if (unlikely(iph
->protocol
== IPPROTO_ICMP
)) {
976 int related
, verdict
= ip_vs_in_icmp(pskb
, &related
, hooknum
);
984 /* Protocol supported? */
985 pp
= ip_vs_proto_get(iph
->protocol
);
992 * Check if the packet belongs to an existing connection entry
994 cp
= pp
->conn_in_get(skb
, pp
, iph
, ihl
, 0);
999 if (!pp
->conn_schedule(skb
, pp
, &v
, &cp
))
1003 if (unlikely(!cp
)) {
1004 /* sorry, all this trouble for a no-hit :) */
1005 IP_VS_DBG_PKT(12, pp
, skb
, 0,
1006 "packet continues traversal as normal");
1010 IP_VS_DBG_PKT(11, pp
, skb
, 0, "Incoming packet");
1012 /* Check the server status */
1013 if (cp
->dest
&& !(cp
->dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
1014 /* the destination server is not available */
1016 if (sysctl_ip_vs_expire_nodest_conn
) {
1017 /* try to expire the connection immediately */
1018 ip_vs_conn_expire_now(cp
);
1020 /* don't restart its timer, and silently
1022 __ip_vs_conn_put(cp
);
1026 ip_vs_in_stats(cp
, skb
);
1027 restart
= ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pp
);
1028 if (cp
->packet_xmit
)
1029 ret
= cp
->packet_xmit(skb
, cp
, pp
);
1030 /* do not touch skb anymore */
1032 IP_VS_DBG_RL("warning: packet_xmit is null");
1036 /* increase its packet counter and check if it is needed
1037 to be synchronized */
1038 atomic_inc(&cp
->in_pkts
);
1039 if ((ip_vs_sync_state
& IP_VS_STATE_MASTER
) &&
1040 (cp
->protocol
!= IPPROTO_TCP
||
1041 cp
->state
== IP_VS_TCP_S_ESTABLISHED
) &&
1042 (atomic_read(&cp
->in_pkts
) % sysctl_ip_vs_sync_threshold
[1]
1043 == sysctl_ip_vs_sync_threshold
[0]))
1044 ip_vs_sync_conn(cp
);
1052 * It is hooked at the NF_IP_FORWARD chain, in order to catch ICMP
1053 * related packets destined for 0.0.0.0/0.
1054 * When fwmark-based virtual service is used, such as transparent
1055 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1056 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1057 * sent to ip_vs_in_icmp. So, catch them at the NF_IP_FORWARD chain
1058 * and send them to ip_vs_in_icmp.
1061 ip_vs_forward_icmp(unsigned int hooknum
, struct sk_buff
**pskb
,
1062 const struct net_device
*in
, const struct net_device
*out
,
1063 int (*okfn
)(struct sk_buff
*))
1067 if ((*pskb
)->nh
.iph
->protocol
!= IPPROTO_ICMP
)
1070 return ip_vs_in_icmp(pskb
, &r
, hooknum
);
1074 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1075 or VS/NAT(change destination), so that filtering rules can be
1077 static struct nf_hook_ops ip_vs_in_ops
= {
1079 .owner
= THIS_MODULE
,
1081 .hooknum
= NF_IP_LOCAL_IN
,
1085 /* After packet filtering, change source only for VS/NAT */
1086 static struct nf_hook_ops ip_vs_out_ops
= {
1088 .owner
= THIS_MODULE
,
1090 .hooknum
= NF_IP_FORWARD
,
1094 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1095 destined for 0.0.0.0/0, which is for incoming IPVS connections */
1096 static struct nf_hook_ops ip_vs_forward_icmp_ops
= {
1097 .hook
= ip_vs_forward_icmp
,
1098 .owner
= THIS_MODULE
,
1100 .hooknum
= NF_IP_FORWARD
,
1104 /* Before the netfilter connection tracking, exit from POST_ROUTING */
1105 static struct nf_hook_ops ip_vs_post_routing_ops
= {
1106 .hook
= ip_vs_post_routing
,
1107 .owner
= THIS_MODULE
,
1109 .hooknum
= NF_IP_POST_ROUTING
,
1110 .priority
= NF_IP_PRI_NAT_SRC
-1,
1115 * Initialize IP Virtual Server
1117 static int __init
ip_vs_init(void)
1121 ret
= ip_vs_control_init();
1123 IP_VS_ERR("can't setup control.\n");
1124 goto cleanup_nothing
;
1127 ip_vs_protocol_init();
1129 ret
= ip_vs_app_init();
1131 IP_VS_ERR("can't setup application helper.\n");
1132 goto cleanup_protocol
;
1135 ret
= ip_vs_conn_init();
1137 IP_VS_ERR("can't setup connection table.\n");
1141 ret
= nf_register_hook(&ip_vs_in_ops
);
1143 IP_VS_ERR("can't register in hook.\n");
1147 ret
= nf_register_hook(&ip_vs_out_ops
);
1149 IP_VS_ERR("can't register out hook.\n");
1152 ret
= nf_register_hook(&ip_vs_post_routing_ops
);
1154 IP_VS_ERR("can't register post_routing hook.\n");
1155 goto cleanup_outops
;
1157 ret
= nf_register_hook(&ip_vs_forward_icmp_ops
);
1159 IP_VS_ERR("can't register forward_icmp hook.\n");
1160 goto cleanup_postroutingops
;
1163 IP_VS_INFO("ipvs loaded.\n");
1166 cleanup_postroutingops
:
1167 nf_unregister_hook(&ip_vs_post_routing_ops
);
1169 nf_unregister_hook(&ip_vs_out_ops
);
1171 nf_unregister_hook(&ip_vs_in_ops
);
1173 ip_vs_conn_cleanup();
1175 ip_vs_app_cleanup();
1177 ip_vs_protocol_cleanup();
1178 ip_vs_control_cleanup();
1183 static void __exit
ip_vs_cleanup(void)
1185 nf_unregister_hook(&ip_vs_forward_icmp_ops
);
1186 nf_unregister_hook(&ip_vs_post_routing_ops
);
1187 nf_unregister_hook(&ip_vs_out_ops
);
1188 nf_unregister_hook(&ip_vs_in_ops
);
1189 ip_vs_conn_cleanup();
1190 ip_vs_app_cleanup();
1191 ip_vs_protocol_cleanup();
1192 ip_vs_control_cleanup();
1193 IP_VS_INFO("ipvs unloaded.\n");
1196 module_init(ip_vs_init
);
1197 module_exit(ip_vs_cleanup
);
1198 MODULE_LICENSE("GPL");