2 * IPVS An implementation of the IP virtual server support for the
3 * LINUX operating system. IPVS is now implemented as a module
4 * over the Netfilter framework. IPVS can be used to build a
5 * high-performance and highly available server based on a
8 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
9 * Peter Kese <peter.kese@ijs.si>
10 * Julian Anastasov <ja@ssi.bg>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
17 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
18 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
22 * Paul `Rusty' Russell properly handle non-linear skbs
23 * Harald Welte don't use nfcache
27 #define KMSG_COMPONENT "IPVS"
28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
30 #include <linux/module.h>
31 #include <linux/kernel.h>
33 #include <linux/tcp.h>
34 #include <linux/sctp.h>
35 #include <linux/icmp.h>
36 #include <linux/slab.h>
41 #include <net/icmp.h> /* for icmp_send */
42 #include <net/route.h>
43 #include <net/ip6_checksum.h>
44 #include <net/netns/generic.h> /* net_generic() */
46 #include <linux/netfilter.h>
47 #include <linux/netfilter_ipv4.h>
49 #ifdef CONFIG_IP_VS_IPV6
51 #include <linux/netfilter_ipv6.h>
52 #include <net/ip6_route.h>
55 #include <net/ip_vs.h>
58 EXPORT_SYMBOL(register_ip_vs_scheduler
);
59 EXPORT_SYMBOL(unregister_ip_vs_scheduler
);
60 EXPORT_SYMBOL(ip_vs_proto_name
);
61 EXPORT_SYMBOL(ip_vs_conn_new
);
62 EXPORT_SYMBOL(ip_vs_conn_in_get
);
63 EXPORT_SYMBOL(ip_vs_conn_out_get
);
64 #ifdef CONFIG_IP_VS_PROTO_TCP
65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen
);
67 EXPORT_SYMBOL(ip_vs_conn_put
);
68 #ifdef CONFIG_IP_VS_DEBUG
69 EXPORT_SYMBOL(ip_vs_get_debug_level
);
72 static int ip_vs_net_id __read_mostly
;
73 /* netns cnt used for uniqueness */
74 static atomic_t ipvs_netns_cnt
= ATOMIC_INIT(0);
76 /* ID used in ICMP lookups */
77 #define icmp_id(icmph) (((icmph)->un).echo.id)
78 #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier)
80 const char *ip_vs_proto_name(unsigned int proto
)
95 #ifdef CONFIG_IP_VS_IPV6
100 sprintf(buf
, "IP_%d", proto
);
105 void ip_vs_init_hash_table(struct list_head
*table
, int rows
)
108 INIT_LIST_HEAD(&table
[rows
]);
112 ip_vs_in_stats(struct ip_vs_conn
*cp
, struct sk_buff
*skb
)
114 struct ip_vs_dest
*dest
= cp
->dest
;
115 struct netns_ipvs
*ipvs
= net_ipvs(skb_net(skb
));
117 if (dest
&& (dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
118 struct ip_vs_cpu_stats
*s
;
120 s
= this_cpu_ptr(dest
->stats
.cpustats
);
122 u64_stats_update_begin(&s
->syncp
);
123 s
->ustats
.inbytes
+= skb
->len
;
124 u64_stats_update_end(&s
->syncp
);
126 s
= this_cpu_ptr(dest
->svc
->stats
.cpustats
);
128 u64_stats_update_begin(&s
->syncp
);
129 s
->ustats
.inbytes
+= skb
->len
;
130 u64_stats_update_end(&s
->syncp
);
132 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
134 u64_stats_update_begin(&s
->syncp
);
135 s
->ustats
.inbytes
+= skb
->len
;
136 u64_stats_update_end(&s
->syncp
);
142 ip_vs_out_stats(struct ip_vs_conn
*cp
, struct sk_buff
*skb
)
144 struct ip_vs_dest
*dest
= cp
->dest
;
145 struct netns_ipvs
*ipvs
= net_ipvs(skb_net(skb
));
147 if (dest
&& (dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
148 struct ip_vs_cpu_stats
*s
;
150 s
= this_cpu_ptr(dest
->stats
.cpustats
);
152 u64_stats_update_begin(&s
->syncp
);
153 s
->ustats
.outbytes
+= skb
->len
;
154 u64_stats_update_end(&s
->syncp
);
156 s
= this_cpu_ptr(dest
->svc
->stats
.cpustats
);
158 u64_stats_update_begin(&s
->syncp
);
159 s
->ustats
.outbytes
+= skb
->len
;
160 u64_stats_update_end(&s
->syncp
);
162 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
164 u64_stats_update_begin(&s
->syncp
);
165 s
->ustats
.outbytes
+= skb
->len
;
166 u64_stats_update_end(&s
->syncp
);
172 ip_vs_conn_stats(struct ip_vs_conn
*cp
, struct ip_vs_service
*svc
)
174 struct netns_ipvs
*ipvs
= net_ipvs(svc
->net
);
175 struct ip_vs_cpu_stats
*s
;
177 s
= this_cpu_ptr(cp
->dest
->stats
.cpustats
);
180 s
= this_cpu_ptr(svc
->stats
.cpustats
);
183 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
189 ip_vs_set_state(struct ip_vs_conn
*cp
, int direction
,
190 const struct sk_buff
*skb
,
191 struct ip_vs_proto_data
*pd
)
193 if (likely(pd
->pp
->state_transition
))
194 pd
->pp
->state_transition(cp
, direction
, skb
, pd
);
198 ip_vs_conn_fill_param_persist(const struct ip_vs_service
*svc
,
199 struct sk_buff
*skb
, int protocol
,
200 const union nf_inet_addr
*caddr
, __be16 cport
,
201 const union nf_inet_addr
*vaddr
, __be16 vport
,
202 struct ip_vs_conn_param
*p
)
204 ip_vs_conn_fill_param(svc
->net
, svc
->af
, protocol
, caddr
, cport
, vaddr
,
206 p
->pe
= rcu_dereference(svc
->pe
);
207 if (p
->pe
&& p
->pe
->fill_param
)
208 return p
->pe
->fill_param(p
, skb
);
214 * IPVS persistent scheduling function
215 * It creates a connection entry according to its template if exists,
216 * or selects a server and creates a connection entry plus a template.
217 * Locking: we are svc user (svc->refcnt), so we hold all dests too
218 * Protocols supported: TCP, UDP
220 static struct ip_vs_conn
*
221 ip_vs_sched_persist(struct ip_vs_service
*svc
,
222 struct sk_buff
*skb
, __be16 src_port
, __be16 dst_port
,
223 int *ignored
, struct ip_vs_iphdr
*iph
)
225 struct ip_vs_conn
*cp
= NULL
;
226 struct ip_vs_dest
*dest
;
227 struct ip_vs_conn
*ct
;
228 __be16 dport
= 0; /* destination port to forward */
230 struct ip_vs_conn_param param
;
231 const union nf_inet_addr fwmark
= { .ip
= htonl(svc
->fwmark
) };
232 union nf_inet_addr snet
; /* source network of the client,
235 /* Mask saddr with the netmask to adjust template granularity */
236 #ifdef CONFIG_IP_VS_IPV6
237 if (svc
->af
== AF_INET6
)
238 ipv6_addr_prefix(&snet
.in6
, &iph
->saddr
.in6
,
239 (__force __u32
) svc
->netmask
);
242 snet
.ip
= iph
->saddr
.ip
& svc
->netmask
;
244 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
246 IP_VS_DBG_ADDR(svc
->af
, &iph
->saddr
), ntohs(src_port
),
247 IP_VS_DBG_ADDR(svc
->af
, &iph
->daddr
), ntohs(dst_port
),
248 IP_VS_DBG_ADDR(svc
->af
, &snet
));
251 * As far as we know, FTP is a very complicated network protocol, and
252 * it uses control connection and data connections. For active FTP,
253 * FTP server initialize data connection to the client, its source port
254 * is often 20. For passive FTP, FTP server tells the clients the port
255 * that it passively listens to, and the client issues the data
256 * connection. In the tunneling or direct routing mode, the load
257 * balancer is on the client-to-server half of connection, the port
258 * number is unknown to the load balancer. So, a conn template like
259 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
260 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
261 * is created for other persistent services.
264 int protocol
= iph
->protocol
;
265 const union nf_inet_addr
*vaddr
= &iph
->daddr
;
268 if (dst_port
== svc
->port
) {
270 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
272 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
274 if (svc
->port
!= FTPPORT
)
277 /* Note: persistent fwmark-based services and
278 * persistent port zero service are handled here.
280 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
281 * port zero template:
282 * <protocol,caddr,0,vaddr,0,daddr,0>
285 protocol
= IPPROTO_IP
;
289 /* return *ignored = -1 so NF_DROP can be used */
290 if (ip_vs_conn_fill_param_persist(svc
, skb
, protocol
, &snet
, 0,
291 vaddr
, vport
, ¶m
) < 0) {
297 /* Check if a template already exists */
298 ct
= ip_vs_ct_in_get(¶m
);
299 if (!ct
|| !ip_vs_check_template(ct
)) {
300 struct ip_vs_scheduler
*sched
;
303 * No template found or the dest of the connection
304 * template is not available.
305 * return *ignored=0 i.e. ICMP and NF_DROP
307 sched
= rcu_dereference(svc
->scheduler
);
308 dest
= sched
->schedule(svc
, skb
);
310 IP_VS_DBG(1, "p-schedule: no dest found.\n");
311 kfree(param
.pe_data
);
316 if (dst_port
== svc
->port
&& svc
->port
!= FTPPORT
)
320 * This adds param.pe_data to the template,
321 * and thus param.pe_data will be destroyed
322 * when the template expires */
323 ct
= ip_vs_conn_new(¶m
, &dest
->addr
, dport
,
324 IP_VS_CONN_F_TEMPLATE
, dest
, skb
->mark
);
326 kfree(param
.pe_data
);
331 ct
->timeout
= svc
->timeout
;
333 /* set destination with the found template */
335 kfree(param
.pe_data
);
339 if (dport
== svc
->port
&& dest
->port
)
342 flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
343 && iph
->protocol
== IPPROTO_UDP
) ?
344 IP_VS_CONN_F_ONE_PACKET
: 0;
347 * Create a new connection according to the template
349 ip_vs_conn_fill_param(svc
->net
, svc
->af
, iph
->protocol
, &iph
->saddr
,
350 src_port
, &iph
->daddr
, dst_port
, ¶m
);
352 cp
= ip_vs_conn_new(¶m
, &dest
->addr
, dport
, flags
, dest
, skb
->mark
);
362 ip_vs_control_add(cp
, ct
);
365 ip_vs_conn_stats(cp
, svc
);
371 * IPVS main scheduling function
372 * It selects a server according to the virtual service, and
373 * creates a connection entry.
374 * Protocols supported: TCP, UDP
378 * 1 : protocol tried to schedule (eg. on SYN), found svc but the
379 * svc/scheduler decides that this packet should be accepted with
380 * NF_ACCEPT because it must not be scheduled.
382 * 0 : scheduler can not find destination, so try bypass or
383 * return ICMP and then NF_DROP (ip_vs_leave).
385 * -1 : scheduler tried to schedule but fatal error occurred, eg.
386 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
387 * failure such as missing Call-ID, ENOMEM on skb_linearize
388 * or pe_data. In this case we should return NF_DROP without
389 * any attempts to send ICMP with ip_vs_leave.
392 ip_vs_schedule(struct ip_vs_service
*svc
, struct sk_buff
*skb
,
393 struct ip_vs_proto_data
*pd
, int *ignored
,
394 struct ip_vs_iphdr
*iph
)
396 struct ip_vs_protocol
*pp
= pd
->pp
;
397 struct ip_vs_conn
*cp
= NULL
;
398 struct ip_vs_scheduler
*sched
;
399 struct ip_vs_dest
*dest
;
400 __be16 _ports
[2], *pptr
;
405 * IPv6 frags, only the first hit here.
407 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
412 * FTPDATA needs this check when using local real server.
413 * Never schedule Active FTPDATA connections from real server.
414 * For LVS-NAT they must be already created. For other methods
415 * with persistence the connection is created on SYN+ACK.
417 if (pptr
[0] == FTPDATA
) {
418 IP_VS_DBG_PKT(12, svc
->af
, pp
, skb
, 0,
419 "Not scheduling FTPDATA");
424 * Do not schedule replies from local real server.
426 if ((!skb
->dev
|| skb
->dev
->flags
& IFF_LOOPBACK
) &&
427 (cp
= pp
->conn_in_get(svc
->af
, skb
, iph
, 1))) {
428 IP_VS_DBG_PKT(12, svc
->af
, pp
, skb
, 0,
429 "Not scheduling reply for existing connection");
430 __ip_vs_conn_put(cp
);
437 if (svc
->flags
& IP_VS_SVC_F_PERSISTENT
)
438 return ip_vs_sched_persist(svc
, skb
, pptr
[0], pptr
[1], ignored
,
444 * Non-persistent service
446 if (!svc
->fwmark
&& pptr
[1] != svc
->port
) {
448 pr_err("Schedule: port zero only supported "
449 "in persistent services, "
450 "check your ipvs configuration\n");
454 sched
= rcu_dereference(svc
->scheduler
);
455 dest
= sched
->schedule(svc
, skb
);
457 IP_VS_DBG(1, "Schedule: no dest found.\n");
461 flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
462 && iph
->protocol
== IPPROTO_UDP
) ?
463 IP_VS_CONN_F_ONE_PACKET
: 0;
466 * Create a connection entry.
469 struct ip_vs_conn_param p
;
471 ip_vs_conn_fill_param(svc
->net
, svc
->af
, iph
->protocol
,
472 &iph
->saddr
, pptr
[0], &iph
->daddr
,
474 cp
= ip_vs_conn_new(&p
, &dest
->addr
,
475 dest
->port
? dest
->port
: pptr
[1],
476 flags
, dest
, skb
->mark
);
483 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
484 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
486 IP_VS_DBG_ADDR(svc
->af
, &cp
->caddr
), ntohs(cp
->cport
),
487 IP_VS_DBG_ADDR(svc
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
488 IP_VS_DBG_ADDR(svc
->af
, &cp
->daddr
), ntohs(cp
->dport
),
489 cp
->flags
, atomic_read(&cp
->refcnt
));
491 ip_vs_conn_stats(cp
, svc
);
497 * Pass or drop the packet.
498 * Called by ip_vs_in, when the virtual service is available but
499 * no destination is available for a new connection.
501 int ip_vs_leave(struct ip_vs_service
*svc
, struct sk_buff
*skb
,
502 struct ip_vs_proto_data
*pd
, struct ip_vs_iphdr
*iph
)
504 __be16 _ports
[2], *pptr
;
507 struct netns_ipvs
*ipvs
;
511 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
519 #ifdef CONFIG_IP_VS_IPV6
520 if (svc
->af
== AF_INET6
)
521 unicast
= ipv6_addr_type(&iph
->daddr
.in6
) & IPV6_ADDR_UNICAST
;
524 unicast
= (inet_addr_type(net
, iph
->daddr
.ip
) == RTN_UNICAST
);
526 /* if it is fwmark-based service, the cache_bypass sysctl is up
527 and the destination is a non-local unicast, then create
528 a cache_bypass connection entry */
529 ipvs
= net_ipvs(net
);
530 if (ipvs
->sysctl_cache_bypass
&& svc
->fwmark
&& unicast
) {
532 struct ip_vs_conn
*cp
;
533 unsigned int flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
&&
534 iph
->protocol
== IPPROTO_UDP
) ?
535 IP_VS_CONN_F_ONE_PACKET
: 0;
536 union nf_inet_addr daddr
= { .all
= { 0, 0, 0, 0 } };
538 /* create a new connection entry */
539 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__
);
541 struct ip_vs_conn_param p
;
542 ip_vs_conn_fill_param(svc
->net
, svc
->af
, iph
->protocol
,
543 &iph
->saddr
, pptr
[0],
544 &iph
->daddr
, pptr
[1], &p
);
545 cp
= ip_vs_conn_new(&p
, &daddr
, 0,
546 IP_VS_CONN_F_BYPASS
| flags
,
553 ip_vs_in_stats(cp
, skb
);
556 ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pd
);
558 /* transmit the first SYN packet */
559 ret
= cp
->packet_xmit(skb
, cp
, pd
->pp
, iph
);
560 /* do not touch skb anymore */
562 atomic_inc(&cp
->in_pkts
);
569 * When the virtual ftp service is presented, packets destined
570 * for other services on the VIP may get here (except services
571 * listed in the ipvs table), pass the packets, because it is
572 * not ipvs job to decide to drop the packets.
574 if ((svc
->port
== FTPPORT
) && (pptr
[1] != FTPPORT
))
578 * Notify the client that the destination is unreachable, and
579 * release the socket buffer.
580 * Since it is in IP layer, the TCP socket is not actually
581 * created, the TCP RST packet cannot be sent, instead that
582 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
584 #ifdef CONFIG_IP_VS_IPV6
585 if (svc
->af
== AF_INET6
) {
587 struct net
*net_
= dev_net(skb_dst(skb
)->dev
);
589 skb
->dev
= net_
->loopback_dev
;
591 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
594 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
601 static int sysctl_snat_reroute(struct sk_buff
*skb
)
603 struct netns_ipvs
*ipvs
= net_ipvs(skb_net(skb
));
604 return ipvs
->sysctl_snat_reroute
;
607 static int sysctl_nat_icmp_send(struct net
*net
)
609 struct netns_ipvs
*ipvs
= net_ipvs(net
);
610 return ipvs
->sysctl_nat_icmp_send
;
613 static int sysctl_expire_nodest_conn(struct netns_ipvs
*ipvs
)
615 return ipvs
->sysctl_expire_nodest_conn
;
620 static int sysctl_snat_reroute(struct sk_buff
*skb
) { return 0; }
621 static int sysctl_nat_icmp_send(struct net
*net
) { return 0; }
622 static int sysctl_expire_nodest_conn(struct netns_ipvs
*ipvs
) { return 0; }
626 __sum16
ip_vs_checksum_complete(struct sk_buff
*skb
, int offset
)
628 return csum_fold(skb_checksum(skb
, offset
, skb
->len
- offset
, 0));
631 static inline enum ip_defrag_users
ip_vs_defrag_user(unsigned int hooknum
)
633 if (NF_INET_LOCAL_IN
== hooknum
)
634 return IP_DEFRAG_VS_IN
;
635 if (NF_INET_FORWARD
== hooknum
)
636 return IP_DEFRAG_VS_FWD
;
637 return IP_DEFRAG_VS_OUT
;
640 static inline int ip_vs_gather_frags(struct sk_buff
*skb
, u_int32_t user
)
645 err
= ip_defrag(skb
, user
);
648 ip_send_check(ip_hdr(skb
));
653 static int ip_vs_route_me_harder(int af
, struct sk_buff
*skb
)
655 #ifdef CONFIG_IP_VS_IPV6
656 if (af
== AF_INET6
) {
657 if (sysctl_snat_reroute(skb
) && ip6_route_me_harder(skb
) != 0)
661 if ((sysctl_snat_reroute(skb
) ||
662 skb_rtable(skb
)->rt_flags
& RTCF_LOCAL
) &&
663 ip_route_me_harder(skb
, RTN_LOCAL
) != 0)
670 * Packet has been made sufficiently writable in caller
671 * - inout: 1=in->out, 0=out->in
673 void ip_vs_nat_icmp(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
674 struct ip_vs_conn
*cp
, int inout
)
676 struct iphdr
*iph
= ip_hdr(skb
);
677 unsigned int icmp_offset
= iph
->ihl
*4;
678 struct icmphdr
*icmph
= (struct icmphdr
*)(skb_network_header(skb
) +
680 struct iphdr
*ciph
= (struct iphdr
*)(icmph
+ 1);
683 iph
->saddr
= cp
->vaddr
.ip
;
685 ciph
->daddr
= cp
->vaddr
.ip
;
688 iph
->daddr
= cp
->daddr
.ip
;
690 ciph
->saddr
= cp
->daddr
.ip
;
694 /* the TCP/UDP/SCTP port */
695 if (IPPROTO_TCP
== ciph
->protocol
|| IPPROTO_UDP
== ciph
->protocol
||
696 IPPROTO_SCTP
== ciph
->protocol
) {
697 __be16
*ports
= (void *)ciph
+ ciph
->ihl
*4;
700 ports
[1] = cp
->vport
;
702 ports
[0] = cp
->dport
;
705 /* And finally the ICMP checksum */
707 icmph
->checksum
= ip_vs_checksum_complete(skb
, icmp_offset
);
708 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
711 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, (void *)ciph
- (void *)iph
,
712 "Forwarding altered outgoing ICMP");
714 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, (void *)ciph
- (void *)iph
,
715 "Forwarding altered incoming ICMP");
718 #ifdef CONFIG_IP_VS_IPV6
719 void ip_vs_nat_icmp_v6(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
720 struct ip_vs_conn
*cp
, int inout
)
722 struct ipv6hdr
*iph
= ipv6_hdr(skb
);
723 unsigned int icmp_offset
= 0;
724 unsigned int offs
= 0; /* header offset*/
726 struct icmp6hdr
*icmph
;
727 struct ipv6hdr
*ciph
;
728 unsigned short fragoffs
;
730 ipv6_find_hdr(skb
, &icmp_offset
, IPPROTO_ICMPV6
, &fragoffs
, NULL
);
731 icmph
= (struct icmp6hdr
*)(skb_network_header(skb
) + icmp_offset
);
732 offs
= icmp_offset
+ sizeof(struct icmp6hdr
);
733 ciph
= (struct ipv6hdr
*)(skb_network_header(skb
) + offs
);
735 protocol
= ipv6_find_hdr(skb
, &offs
, -1, &fragoffs
, NULL
);
738 iph
->saddr
= cp
->vaddr
.in6
;
739 ciph
->daddr
= cp
->vaddr
.in6
;
741 iph
->daddr
= cp
->daddr
.in6
;
742 ciph
->saddr
= cp
->daddr
.in6
;
745 /* the TCP/UDP/SCTP port */
746 if (!fragoffs
&& (IPPROTO_TCP
== protocol
|| IPPROTO_UDP
== protocol
||
747 IPPROTO_SCTP
== protocol
)) {
748 __be16
*ports
= (void *)(skb_network_header(skb
) + offs
);
750 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__
,
751 ntohs(inout
? ports
[1] : ports
[0]),
752 ntohs(inout
? cp
->vport
: cp
->dport
));
754 ports
[1] = cp
->vport
;
756 ports
[0] = cp
->dport
;
759 /* And finally the ICMP checksum */
760 icmph
->icmp6_cksum
= ~csum_ipv6_magic(&iph
->saddr
, &iph
->daddr
,
761 skb
->len
- icmp_offset
,
763 skb
->csum_start
= skb_network_header(skb
) - skb
->head
+ icmp_offset
;
764 skb
->csum_offset
= offsetof(struct icmp6hdr
, icmp6_cksum
);
765 skb
->ip_summed
= CHECKSUM_PARTIAL
;
768 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
,
769 (void *)ciph
- (void *)iph
,
770 "Forwarding altered outgoing ICMPv6");
772 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
,
773 (void *)ciph
- (void *)iph
,
774 "Forwarding altered incoming ICMPv6");
778 /* Handle relevant response ICMP messages - forward to the right
781 static int handle_response_icmp(int af
, struct sk_buff
*skb
,
782 union nf_inet_addr
*snet
,
783 __u8 protocol
, struct ip_vs_conn
*cp
,
784 struct ip_vs_protocol
*pp
,
785 unsigned int offset
, unsigned int ihl
)
787 unsigned int verdict
= NF_DROP
;
789 if (IP_VS_FWD_METHOD(cp
) != 0) {
790 pr_err("shouldn't reach here, because the box is on the "
791 "half connection in the tun/dr module.\n");
794 /* Ensure the checksum is correct */
795 if (!skb_csum_unnecessary(skb
) && ip_vs_checksum_complete(skb
, ihl
)) {
796 /* Failed checksum! */
797 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
798 IP_VS_DBG_ADDR(af
, snet
));
802 if (IPPROTO_TCP
== protocol
|| IPPROTO_UDP
== protocol
||
803 IPPROTO_SCTP
== protocol
)
804 offset
+= 2 * sizeof(__u16
);
805 if (!skb_make_writable(skb
, offset
))
808 #ifdef CONFIG_IP_VS_IPV6
810 ip_vs_nat_icmp_v6(skb
, pp
, cp
, 1);
813 ip_vs_nat_icmp(skb
, pp
, cp
, 1);
815 if (ip_vs_route_me_harder(af
, skb
))
818 /* do the statistics and put it back */
819 ip_vs_out_stats(cp
, skb
);
821 skb
->ipvs_property
= 1;
822 if (!(cp
->flags
& IP_VS_CONN_F_NFCT
))
825 ip_vs_update_conntrack(skb
, cp
, 0);
829 __ip_vs_conn_put(cp
);
835 * Handle ICMP messages in the inside-to-outside direction (outgoing).
836 * Find any that might be relevant, check against existing connections.
837 * Currently handles error types - unreachable, quench, ttl exceeded.
839 static int ip_vs_out_icmp(struct sk_buff
*skb
, int *related
,
840 unsigned int hooknum
)
843 struct icmphdr _icmph
, *ic
;
844 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
845 struct ip_vs_iphdr ciph
;
846 struct ip_vs_conn
*cp
;
847 struct ip_vs_protocol
*pp
;
848 unsigned int offset
, ihl
;
849 union nf_inet_addr snet
;
853 /* reassemble IP fragments */
854 if (ip_is_fragment(ip_hdr(skb
))) {
855 if (ip_vs_gather_frags(skb
, ip_vs_defrag_user(hooknum
)))
860 offset
= ihl
= iph
->ihl
* 4;
861 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
865 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
866 ic
->type
, ntohs(icmp_id(ic
)),
867 &iph
->saddr
, &iph
->daddr
);
870 * Work through seeing if this is for us.
871 * These checks are supposed to be in an order that means easy
872 * things are checked first to speed up processing.... however
873 * this means that some packets will manage to get a long way
874 * down this stack and then be rejected, but that's life.
876 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
877 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
878 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
883 /* Now find the contained IP header */
884 offset
+= sizeof(_icmph
);
885 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
887 return NF_ACCEPT
; /* The packet looks wrong, ignore */
889 pp
= ip_vs_proto_get(cih
->protocol
);
893 /* Is the embedded protocol header present? */
894 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
898 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, offset
,
899 "Checking outgoing ICMP for");
901 ip_vs_fill_ip4hdr(cih
, &ciph
);
903 /* The embedded headers contain source and dest in reverse order */
904 cp
= pp
->conn_out_get(AF_INET
, skb
, &ciph
, 1);
908 snet
.ip
= iph
->saddr
;
909 return handle_response_icmp(AF_INET
, skb
, &snet
, cih
->protocol
, cp
,
913 #ifdef CONFIG_IP_VS_IPV6
914 static int ip_vs_out_icmp_v6(struct sk_buff
*skb
, int *related
,
915 unsigned int hooknum
, struct ip_vs_iphdr
*ipvsh
)
917 struct icmp6hdr _icmph
, *ic
;
918 struct ipv6hdr _ip6h
, *ip6h
; /* The ip header contained within ICMP */
919 struct ip_vs_iphdr ciph
= {.flags
= 0, .fragoffs
= 0};/*Contained IP */
920 struct ip_vs_conn
*cp
;
921 struct ip_vs_protocol
*pp
;
922 union nf_inet_addr snet
;
923 unsigned int writable
;
926 ic
= frag_safe_skb_hp(skb
, ipvsh
->len
, sizeof(_icmph
), &_icmph
, ipvsh
);
931 * Work through seeing if this is for us.
932 * These checks are supposed to be in an order that means easy
933 * things are checked first to speed up processing.... however
934 * this means that some packets will manage to get a long way
935 * down this stack and then be rejected, but that's life.
937 if (ic
->icmp6_type
& ICMPV6_INFOMSG_MASK
) {
941 /* Fragment header that is before ICMP header tells us that:
942 * it's not an error message since they can't be fragmented.
944 if (ipvsh
->flags
& IP6_FH_F_FRAG
)
947 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
948 ic
->icmp6_type
, ntohs(icmpv6_id(ic
)),
949 &ipvsh
->saddr
, &ipvsh
->daddr
);
951 /* Now find the contained IP header */
952 ciph
.len
= ipvsh
->len
+ sizeof(_icmph
);
953 ip6h
= skb_header_pointer(skb
, ciph
.len
, sizeof(_ip6h
), &_ip6h
);
955 return NF_ACCEPT
; /* The packet looks wrong, ignore */
956 ciph
.saddr
.in6
= ip6h
->saddr
; /* conn_out_get() handles reverse order */
957 ciph
.daddr
.in6
= ip6h
->daddr
;
958 /* skip possible IPv6 exthdrs of contained IPv6 packet */
959 ciph
.protocol
= ipv6_find_hdr(skb
, &ciph
.len
, -1, &ciph
.fragoffs
, NULL
);
960 if (ciph
.protocol
< 0)
961 return NF_ACCEPT
; /* Contained IPv6 hdr looks wrong, ignore */
963 pp
= ip_vs_proto_get(ciph
.protocol
);
967 /* The embedded headers contain source and dest in reverse order */
968 cp
= pp
->conn_out_get(AF_INET6
, skb
, &ciph
, 1);
972 snet
.in6
= ciph
.saddr
.in6
;
974 return handle_response_icmp(AF_INET6
, skb
, &snet
, ciph
.protocol
, cp
,
975 pp
, writable
, sizeof(struct ipv6hdr
));
980 * Check if sctp chunc is ABORT chunk
982 static inline int is_sctp_abort(const struct sk_buff
*skb
, int nh_len
)
984 sctp_chunkhdr_t
*sch
, schunk
;
985 sch
= skb_header_pointer(skb
, nh_len
+ sizeof(sctp_sctphdr_t
),
986 sizeof(schunk
), &schunk
);
989 if (sch
->type
== SCTP_CID_ABORT
)
994 static inline int is_tcp_reset(const struct sk_buff
*skb
, int nh_len
)
996 struct tcphdr _tcph
, *th
;
998 th
= skb_header_pointer(skb
, nh_len
, sizeof(_tcph
), &_tcph
);
1004 static inline bool is_new_conn(const struct sk_buff
*skb
,
1005 struct ip_vs_iphdr
*iph
)
1007 switch (iph
->protocol
) {
1009 struct tcphdr _tcph
, *th
;
1011 th
= skb_header_pointer(skb
, iph
->len
, sizeof(_tcph
), &_tcph
);
1016 case IPPROTO_SCTP
: {
1017 sctp_chunkhdr_t
*sch
, schunk
;
1019 sch
= skb_header_pointer(skb
, iph
->len
+ sizeof(sctp_sctphdr_t
),
1020 sizeof(schunk
), &schunk
);
1023 return sch
->type
== SCTP_CID_INIT
;
1030 /* Handle response packets: rewrite addresses and send away...
1033 handle_response(int af
, struct sk_buff
*skb
, struct ip_vs_proto_data
*pd
,
1034 struct ip_vs_conn
*cp
, struct ip_vs_iphdr
*iph
)
1036 struct ip_vs_protocol
*pp
= pd
->pp
;
1038 IP_VS_DBG_PKT(11, af
, pp
, skb
, 0, "Outgoing packet");
1040 if (!skb_make_writable(skb
, iph
->len
))
1043 /* mangle the packet */
1044 if (pp
->snat_handler
&& !pp
->snat_handler(skb
, pp
, cp
, iph
))
1047 #ifdef CONFIG_IP_VS_IPV6
1049 ipv6_hdr(skb
)->saddr
= cp
->vaddr
.in6
;
1053 ip_hdr(skb
)->saddr
= cp
->vaddr
.ip
;
1054 ip_send_check(ip_hdr(skb
));
1058 * nf_iterate does not expect change in the skb->dst->dev.
1059 * It looks like it is not fatal to enable this code for hooks
1060 * where our handlers are at the end of the chain list and
1061 * when all next handlers use skb->dst->dev and not outdev.
1062 * It will definitely route properly the inout NAT traffic
1063 * when multiple paths are used.
1066 /* For policy routing, packets originating from this
1067 * machine itself may be routed differently to packets
1068 * passing through. We want this packet to be routed as
1069 * if it came from this machine itself. So re-compute
1070 * the routing information.
1072 if (ip_vs_route_me_harder(af
, skb
))
1075 IP_VS_DBG_PKT(10, af
, pp
, skb
, 0, "After SNAT");
1077 ip_vs_out_stats(cp
, skb
);
1078 ip_vs_set_state(cp
, IP_VS_DIR_OUTPUT
, skb
, pd
);
1079 skb
->ipvs_property
= 1;
1080 if (!(cp
->flags
& IP_VS_CONN_F_NFCT
))
1083 ip_vs_update_conntrack(skb
, cp
, 0);
1097 * Check if outgoing packet belongs to the established ip_vs_conn.
1100 ip_vs_out(unsigned int hooknum
, struct sk_buff
*skb
, int af
)
1102 struct net
*net
= NULL
;
1103 struct ip_vs_iphdr iph
;
1104 struct ip_vs_protocol
*pp
;
1105 struct ip_vs_proto_data
*pd
;
1106 struct ip_vs_conn
*cp
;
1110 /* Already marked as IPVS request or reply? */
1111 if (skb
->ipvs_property
)
1114 /* Bad... Do not break raw sockets */
1115 if (unlikely(skb
->sk
!= NULL
&& hooknum
== NF_INET_LOCAL_OUT
&&
1117 struct sock
*sk
= skb
->sk
;
1118 struct inet_sock
*inet
= inet_sk(skb
->sk
);
1120 if (inet
&& sk
->sk_family
== PF_INET
&& inet
->nodefrag
)
1124 if (unlikely(!skb_dst(skb
)))
1128 if (!net_ipvs(net
)->enable
)
1131 ip_vs_fill_iph_skb(af
, skb
, &iph
);
1132 #ifdef CONFIG_IP_VS_IPV6
1133 if (af
== AF_INET6
) {
1134 if (unlikely(iph
.protocol
== IPPROTO_ICMPV6
)) {
1136 int verdict
= ip_vs_out_icmp_v6(skb
, &related
,
1144 if (unlikely(iph
.protocol
== IPPROTO_ICMP
)) {
1146 int verdict
= ip_vs_out_icmp(skb
, &related
, hooknum
);
1152 pd
= ip_vs_proto_data_get(net
, iph
.protocol
);
1157 /* reassemble IP fragments */
1158 #ifdef CONFIG_IP_VS_IPV6
1161 if (unlikely(ip_is_fragment(ip_hdr(skb
)) && !pp
->dont_defrag
)) {
1162 if (ip_vs_gather_frags(skb
,
1163 ip_vs_defrag_user(hooknum
)))
1166 ip_vs_fill_ip4hdr(skb_network_header(skb
), &iph
);
1170 * Check if the packet belongs to an existing entry
1172 cp
= pp
->conn_out_get(af
, skb
, &iph
, 0);
1175 return handle_response(af
, skb
, pd
, cp
, &iph
);
1176 if (sysctl_nat_icmp_send(net
) &&
1177 (pp
->protocol
== IPPROTO_TCP
||
1178 pp
->protocol
== IPPROTO_UDP
||
1179 pp
->protocol
== IPPROTO_SCTP
)) {
1180 __be16 _ports
[2], *pptr
;
1182 pptr
= frag_safe_skb_hp(skb
, iph
.len
,
1183 sizeof(_ports
), _ports
, &iph
);
1185 return NF_ACCEPT
; /* Not for me */
1186 if (ip_vs_has_real_service(net
, af
, iph
.protocol
, &iph
.saddr
,
1189 * Notify the real server: there is no
1190 * existing entry if it is not RST
1191 * packet or not TCP packet.
1193 if ((iph
.protocol
!= IPPROTO_TCP
&&
1194 iph
.protocol
!= IPPROTO_SCTP
)
1195 || ((iph
.protocol
== IPPROTO_TCP
1196 && !is_tcp_reset(skb
, iph
.len
))
1197 || (iph
.protocol
== IPPROTO_SCTP
1198 && !is_sctp_abort(skb
,
1200 #ifdef CONFIG_IP_VS_IPV6
1201 if (af
== AF_INET6
) {
1203 skb
->dev
= net
->loopback_dev
;
1205 ICMPV6_DEST_UNREACH
,
1206 ICMPV6_PORT_UNREACH
,
1212 ICMP_PORT_UNREACH
, 0);
1217 IP_VS_DBG_PKT(12, af
, pp
, skb
, 0,
1218 "ip_vs_out: packet continues traversal as normal");
1223 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1224 * used only for VS/NAT.
1225 * Check if packet is reply for established ip_vs_conn.
1228 ip_vs_reply4(unsigned int hooknum
, struct sk_buff
*skb
,
1229 const struct net_device
*in
, const struct net_device
*out
,
1230 int (*okfn
)(struct sk_buff
*))
1232 return ip_vs_out(hooknum
, skb
, AF_INET
);
1236 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1237 * Check if packet is reply for established ip_vs_conn.
1240 ip_vs_local_reply4(unsigned int hooknum
, struct sk_buff
*skb
,
1241 const struct net_device
*in
, const struct net_device
*out
,
1242 int (*okfn
)(struct sk_buff
*))
1244 return ip_vs_out(hooknum
, skb
, AF_INET
);
1247 #ifdef CONFIG_IP_VS_IPV6
1250 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1251 * used only for VS/NAT.
1252 * Check if packet is reply for established ip_vs_conn.
1255 ip_vs_reply6(unsigned int hooknum
, struct sk_buff
*skb
,
1256 const struct net_device
*in
, const struct net_device
*out
,
1257 int (*okfn
)(struct sk_buff
*))
1259 return ip_vs_out(hooknum
, skb
, AF_INET6
);
1263 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1264 * Check if packet is reply for established ip_vs_conn.
1267 ip_vs_local_reply6(unsigned int hooknum
, struct sk_buff
*skb
,
1268 const struct net_device
*in
, const struct net_device
*out
,
1269 int (*okfn
)(struct sk_buff
*))
1271 return ip_vs_out(hooknum
, skb
, AF_INET6
);
1277 * Handle ICMP messages in the outside-to-inside direction (incoming).
1278 * Find any that might be relevant, check against existing connections,
1279 * forward to the right destination host if relevant.
1280 * Currently handles error types - unreachable, quench, ttl exceeded.
1283 ip_vs_in_icmp(struct sk_buff
*skb
, int *related
, unsigned int hooknum
)
1285 struct net
*net
= NULL
;
1287 struct icmphdr _icmph
, *ic
;
1288 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
1289 struct ip_vs_iphdr ciph
;
1290 struct ip_vs_conn
*cp
;
1291 struct ip_vs_protocol
*pp
;
1292 struct ip_vs_proto_data
*pd
;
1293 unsigned int offset
, offset2
, ihl
, verdict
;
1298 /* reassemble IP fragments */
1299 if (ip_is_fragment(ip_hdr(skb
))) {
1300 if (ip_vs_gather_frags(skb
, ip_vs_defrag_user(hooknum
)))
1305 offset
= ihl
= iph
->ihl
* 4;
1306 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
1310 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1311 ic
->type
, ntohs(icmp_id(ic
)),
1312 &iph
->saddr
, &iph
->daddr
);
1315 * Work through seeing if this is for us.
1316 * These checks are supposed to be in an order that means easy
1317 * things are checked first to speed up processing.... however
1318 * this means that some packets will manage to get a long way
1319 * down this stack and then be rejected, but that's life.
1321 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
1322 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
1323 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
1328 /* Now find the contained IP header */
1329 offset
+= sizeof(_icmph
);
1330 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
1332 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1336 /* Special case for errors for IPIP packets */
1338 if (cih
->protocol
== IPPROTO_IPIP
) {
1339 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
)))
1341 /* Error for our IPIP must arrive at LOCAL_IN */
1342 if (!(skb_rtable(skb
)->rt_flags
& RTCF_LOCAL
))
1344 offset
+= cih
->ihl
* 4;
1345 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
1347 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1351 pd
= ip_vs_proto_data_get(net
, cih
->protocol
);
1356 /* Is the embedded protocol header present? */
1357 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
1361 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, offset
,
1362 "Checking incoming ICMP for");
1365 ip_vs_fill_ip4hdr(cih
, &ciph
);
1368 /* The embedded headers contain source and dest in reverse order.
1369 * For IPIP this is error for request, not for reply.
1371 cp
= pp
->conn_in_get(AF_INET
, skb
, &ciph
, ipip
? 0 : 1);
1377 /* Ensure the checksum is correct */
1378 if (!skb_csum_unnecessary(skb
) && ip_vs_checksum_complete(skb
, ihl
)) {
1379 /* Failed checksum! */
1380 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1386 __be32 info
= ic
->un
.gateway
;
1387 __u8 type
= ic
->type
;
1388 __u8 code
= ic
->code
;
1390 /* Update the MTU */
1391 if (ic
->type
== ICMP_DEST_UNREACH
&&
1392 ic
->code
== ICMP_FRAG_NEEDED
) {
1393 struct ip_vs_dest
*dest
= cp
->dest
;
1394 u32 mtu
= ntohs(ic
->un
.frag
.mtu
);
1395 __be16 frag_off
= cih
->frag_off
;
1397 /* Strip outer IP and ICMP, go to IPIP header */
1398 if (pskb_pull(skb
, ihl
+ sizeof(_icmph
)) == NULL
)
1400 offset2
-= ihl
+ sizeof(_icmph
);
1401 skb_reset_network_header(skb
);
1402 IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1403 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
, mtu
);
1404 ipv4_update_pmtu(skb
, dev_net(skb
->dev
),
1406 /* Client uses PMTUD? */
1407 if (!(frag_off
& htons(IP_DF
)))
1409 /* Prefer the resulting PMTU */
1411 struct ip_vs_dest_dst
*dest_dst
;
1414 dest_dst
= rcu_dereference(dest
->dest_dst
);
1416 mtu
= dst_mtu(dest_dst
->dst_cache
);
1419 if (mtu
> 68 + sizeof(struct iphdr
))
1420 mtu
-= sizeof(struct iphdr
);
1423 /* Strip outer IP, ICMP and IPIP, go to IP header of
1426 if (pskb_pull(skb
, offset2
) == NULL
)
1428 skb_reset_network_header(skb
);
1429 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1430 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1431 type
, code
, ntohl(info
));
1432 icmp_send(skb
, type
, code
, info
);
1433 /* ICMP can be shorter but anyways, account it */
1434 ip_vs_out_stats(cp
, skb
);
1438 verdict
= NF_STOLEN
;
1442 /* do the statistics and put it back */
1443 ip_vs_in_stats(cp
, skb
);
1444 if (IPPROTO_TCP
== cih
->protocol
|| IPPROTO_UDP
== cih
->protocol
||
1445 IPPROTO_SCTP
== cih
->protocol
)
1446 offset
+= 2 * sizeof(__u16
);
1447 verdict
= ip_vs_icmp_xmit(skb
, cp
, pp
, offset
, hooknum
, &ciph
);
1450 __ip_vs_conn_put(cp
);
1455 #ifdef CONFIG_IP_VS_IPV6
1456 static int ip_vs_in_icmp_v6(struct sk_buff
*skb
, int *related
,
1457 unsigned int hooknum
, struct ip_vs_iphdr
*iph
)
1459 struct net
*net
= NULL
;
1460 struct ipv6hdr _ip6h
, *ip6h
;
1461 struct icmp6hdr _icmph
, *ic
;
1462 struct ip_vs_iphdr ciph
= {.flags
= 0, .fragoffs
= 0};/*Contained IP */
1463 struct ip_vs_conn
*cp
;
1464 struct ip_vs_protocol
*pp
;
1465 struct ip_vs_proto_data
*pd
;
1466 unsigned int offs_ciph
, writable
, verdict
;
1470 ic
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_icmph
), &_icmph
, iph
);
1475 * Work through seeing if this is for us.
1476 * These checks are supposed to be in an order that means easy
1477 * things are checked first to speed up processing.... however
1478 * this means that some packets will manage to get a long way
1479 * down this stack and then be rejected, but that's life.
1481 if (ic
->icmp6_type
& ICMPV6_INFOMSG_MASK
) {
1485 /* Fragment header that is before ICMP header tells us that:
1486 * it's not an error message since they can't be fragmented.
1488 if (iph
->flags
& IP6_FH_F_FRAG
)
1491 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1492 ic
->icmp6_type
, ntohs(icmpv6_id(ic
)),
1493 &iph
->saddr
, &iph
->daddr
);
1495 /* Now find the contained IP header */
1496 ciph
.len
= iph
->len
+ sizeof(_icmph
);
1497 offs_ciph
= ciph
.len
; /* Save ip header offset */
1498 ip6h
= skb_header_pointer(skb
, ciph
.len
, sizeof(_ip6h
), &_ip6h
);
1500 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1501 ciph
.saddr
.in6
= ip6h
->saddr
; /* conn_in_get() handles reverse order */
1502 ciph
.daddr
.in6
= ip6h
->daddr
;
1503 /* skip possible IPv6 exthdrs of contained IPv6 packet */
1504 ciph
.protocol
= ipv6_find_hdr(skb
, &ciph
.len
, -1, &ciph
.fragoffs
, NULL
);
1505 if (ciph
.protocol
< 0)
1506 return NF_ACCEPT
; /* Contained IPv6 hdr looks wrong, ignore */
1509 pd
= ip_vs_proto_data_get(net
, ciph
.protocol
);
1514 /* Cannot handle fragmented embedded protocol */
1518 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
, offs_ciph
,
1519 "Checking incoming ICMPv6 for");
1521 /* The embedded headers contain source and dest in reverse order
1522 * if not from localhost
1524 cp
= pp
->conn_in_get(AF_INET6
, skb
, &ciph
,
1525 (hooknum
== NF_INET_LOCAL_OUT
) ? 0 : 1);
1529 /* VS/TUN, VS/DR and LOCALNODE just let it go */
1530 if ((hooknum
== NF_INET_LOCAL_OUT
) &&
1531 (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
)) {
1532 __ip_vs_conn_put(cp
);
1536 /* do the statistics and put it back */
1537 ip_vs_in_stats(cp
, skb
);
1539 /* Need to mangle contained IPv6 header in ICMPv6 packet */
1540 writable
= ciph
.len
;
1541 if (IPPROTO_TCP
== ciph
.protocol
|| IPPROTO_UDP
== ciph
.protocol
||
1542 IPPROTO_SCTP
== ciph
.protocol
)
1543 writable
+= 2 * sizeof(__u16
); /* Also mangle ports */
1545 verdict
= ip_vs_icmp_xmit_v6(skb
, cp
, pp
, writable
, hooknum
, &ciph
);
1547 __ip_vs_conn_put(cp
);
1555 * Check if it's for virtual services, look it up,
1556 * and send it on its way...
1559 ip_vs_in(unsigned int hooknum
, struct sk_buff
*skb
, int af
)
1562 struct ip_vs_iphdr iph
;
1563 struct ip_vs_protocol
*pp
;
1564 struct ip_vs_proto_data
*pd
;
1565 struct ip_vs_conn
*cp
;
1567 struct netns_ipvs
*ipvs
;
1569 /* Already marked as IPVS request or reply? */
1570 if (skb
->ipvs_property
)
1575 * - remote client: only PACKET_HOST
1576 * - route: used for struct net when skb->dev is unset
1578 if (unlikely((skb
->pkt_type
!= PACKET_HOST
&&
1579 hooknum
!= NF_INET_LOCAL_OUT
) ||
1581 ip_vs_fill_iph_skb(af
, skb
, &iph
);
1582 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1583 " ignored in hook %u\n",
1584 skb
->pkt_type
, iph
.protocol
,
1585 IP_VS_DBG_ADDR(af
, &iph
.daddr
), hooknum
);
1588 /* ipvs enabled in this netns ? */
1590 ipvs
= net_ipvs(net
);
1591 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1594 ip_vs_fill_iph_skb(af
, skb
, &iph
);
1596 /* Bad... Do not break raw sockets */
1597 if (unlikely(skb
->sk
!= NULL
&& hooknum
== NF_INET_LOCAL_OUT
&&
1599 struct sock
*sk
= skb
->sk
;
1600 struct inet_sock
*inet
= inet_sk(skb
->sk
);
1602 if (inet
&& sk
->sk_family
== PF_INET
&& inet
->nodefrag
)
1606 #ifdef CONFIG_IP_VS_IPV6
1607 if (af
== AF_INET6
) {
1608 if (unlikely(iph
.protocol
== IPPROTO_ICMPV6
)) {
1610 int verdict
= ip_vs_in_icmp_v6(skb
, &related
, hooknum
,
1618 if (unlikely(iph
.protocol
== IPPROTO_ICMP
)) {
1620 int verdict
= ip_vs_in_icmp(skb
, &related
, hooknum
);
1626 /* Protocol supported? */
1627 pd
= ip_vs_proto_data_get(net
, iph
.protocol
);
1632 * Check if the packet belongs to an existing connection entry
1634 cp
= pp
->conn_in_get(af
, skb
, &iph
, 0);
1636 if (unlikely(sysctl_expire_nodest_conn(ipvs
)) && cp
&& cp
->dest
&&
1637 unlikely(!atomic_read(&cp
->dest
->weight
)) && !iph
.fragoffs
&&
1638 is_new_conn(skb
, &iph
)) {
1639 ip_vs_conn_expire_now(cp
);
1640 __ip_vs_conn_put(cp
);
1644 if (unlikely(!cp
) && !iph
.fragoffs
) {
1645 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1646 * replayed fragment zero will already have created the cp
1650 /* Schedule and create new connection entry into &cp */
1651 if (!pp
->conn_schedule(af
, skb
, pd
, &v
, &cp
, &iph
))
1655 if (unlikely(!cp
)) {
1656 /* sorry, all this trouble for a no-hit :) */
1657 IP_VS_DBG_PKT(12, af
, pp
, skb
, 0,
1658 "ip_vs_in: packet continues traversal as normal");
1660 /* Fragment that couldn't be mapped to a conn entry
1661 * is missing module nf_defrag_ipv6
1663 IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1664 IP_VS_DBG_PKT(7, af
, pp
, skb
, 0, "unhandled fragment");
1669 IP_VS_DBG_PKT(11, af
, pp
, skb
, 0, "Incoming packet");
1670 /* Check the server status */
1671 if (cp
->dest
&& !(cp
->dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
1672 /* the destination server is not available */
1674 if (sysctl_expire_nodest_conn(ipvs
)) {
1675 /* try to expire the connection immediately */
1676 ip_vs_conn_expire_now(cp
);
1678 /* don't restart its timer, and silently
1680 __ip_vs_conn_put(cp
);
1684 ip_vs_in_stats(cp
, skb
);
1685 ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pd
);
1686 if (cp
->packet_xmit
)
1687 ret
= cp
->packet_xmit(skb
, cp
, pp
, &iph
);
1688 /* do not touch skb anymore */
1690 IP_VS_DBG_RL("warning: packet_xmit is null");
1694 /* Increase its packet counter and check if it is needed
1695 * to be synchronized
1697 * Sync connection if it is about to close to
1698 * encorage the standby servers to update the connections timeout
1700 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1703 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
1704 pkts
= sysctl_sync_threshold(ipvs
);
1706 pkts
= atomic_add_return(1, &cp
->in_pkts
);
1708 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
)
1709 ip_vs_sync_conn(net
, cp
, pkts
);
1716 * AF_INET handler in NF_INET_LOCAL_IN chain
1717 * Schedule and forward packets from remote clients
1720 ip_vs_remote_request4(unsigned int hooknum
, struct sk_buff
*skb
,
1721 const struct net_device
*in
,
1722 const struct net_device
*out
,
1723 int (*okfn
)(struct sk_buff
*))
1725 return ip_vs_in(hooknum
, skb
, AF_INET
);
1729 * AF_INET handler in NF_INET_LOCAL_OUT chain
1730 * Schedule and forward packets from local clients
1733 ip_vs_local_request4(unsigned int hooknum
, struct sk_buff
*skb
,
1734 const struct net_device
*in
, const struct net_device
*out
,
1735 int (*okfn
)(struct sk_buff
*))
1737 return ip_vs_in(hooknum
, skb
, AF_INET
);
1740 #ifdef CONFIG_IP_VS_IPV6
1743 * AF_INET6 handler in NF_INET_LOCAL_IN chain
1744 * Schedule and forward packets from remote clients
1747 ip_vs_remote_request6(unsigned int hooknum
, struct sk_buff
*skb
,
1748 const struct net_device
*in
,
1749 const struct net_device
*out
,
1750 int (*okfn
)(struct sk_buff
*))
1752 return ip_vs_in(hooknum
, skb
, AF_INET6
);
1756 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
1757 * Schedule and forward packets from local clients
1760 ip_vs_local_request6(unsigned int hooknum
, struct sk_buff
*skb
,
1761 const struct net_device
*in
, const struct net_device
*out
,
1762 int (*okfn
)(struct sk_buff
*))
1764 return ip_vs_in(hooknum
, skb
, AF_INET6
);
1771 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1772 * related packets destined for 0.0.0.0/0.
1773 * When fwmark-based virtual service is used, such as transparent
1774 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1775 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1776 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1777 * and send them to ip_vs_in_icmp.
1780 ip_vs_forward_icmp(unsigned int hooknum
, struct sk_buff
*skb
,
1781 const struct net_device
*in
, const struct net_device
*out
,
1782 int (*okfn
)(struct sk_buff
*))
1786 struct netns_ipvs
*ipvs
;
1788 if (ip_hdr(skb
)->protocol
!= IPPROTO_ICMP
)
1791 /* ipvs enabled in this netns ? */
1793 ipvs
= net_ipvs(net
);
1794 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1797 return ip_vs_in_icmp(skb
, &r
, hooknum
);
1800 #ifdef CONFIG_IP_VS_IPV6
1802 ip_vs_forward_icmp_v6(unsigned int hooknum
, struct sk_buff
*skb
,
1803 const struct net_device
*in
, const struct net_device
*out
,
1804 int (*okfn
)(struct sk_buff
*))
1808 struct netns_ipvs
*ipvs
;
1809 struct ip_vs_iphdr iphdr
;
1811 ip_vs_fill_iph_skb(AF_INET6
, skb
, &iphdr
);
1812 if (iphdr
.protocol
!= IPPROTO_ICMPV6
)
1815 /* ipvs enabled in this netns ? */
1817 ipvs
= net_ipvs(net
);
1818 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1821 return ip_vs_in_icmp_v6(skb
, &r
, hooknum
, &iphdr
);
1826 static struct nf_hook_ops ip_vs_ops
[] __read_mostly
= {
1827 /* After packet filtering, change source only for VS/NAT */
1829 .hook
= ip_vs_reply4
,
1830 .owner
= THIS_MODULE
,
1832 .hooknum
= NF_INET_LOCAL_IN
,
1833 .priority
= NF_IP_PRI_NAT_SRC
- 2,
1835 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1836 * or VS/NAT(change destination), so that filtering rules can be
1837 * applied to IPVS. */
1839 .hook
= ip_vs_remote_request4
,
1840 .owner
= THIS_MODULE
,
1842 .hooknum
= NF_INET_LOCAL_IN
,
1843 .priority
= NF_IP_PRI_NAT_SRC
- 1,
1845 /* Before ip_vs_in, change source only for VS/NAT */
1847 .hook
= ip_vs_local_reply4
,
1848 .owner
= THIS_MODULE
,
1850 .hooknum
= NF_INET_LOCAL_OUT
,
1851 .priority
= NF_IP_PRI_NAT_DST
+ 1,
1853 /* After mangle, schedule and forward local requests */
1855 .hook
= ip_vs_local_request4
,
1856 .owner
= THIS_MODULE
,
1858 .hooknum
= NF_INET_LOCAL_OUT
,
1859 .priority
= NF_IP_PRI_NAT_DST
+ 2,
1861 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1862 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1864 .hook
= ip_vs_forward_icmp
,
1865 .owner
= THIS_MODULE
,
1867 .hooknum
= NF_INET_FORWARD
,
1870 /* After packet filtering, change source only for VS/NAT */
1872 .hook
= ip_vs_reply4
,
1873 .owner
= THIS_MODULE
,
1875 .hooknum
= NF_INET_FORWARD
,
1878 #ifdef CONFIG_IP_VS_IPV6
1879 /* After packet filtering, change source only for VS/NAT */
1881 .hook
= ip_vs_reply6
,
1882 .owner
= THIS_MODULE
,
1884 .hooknum
= NF_INET_LOCAL_IN
,
1885 .priority
= NF_IP6_PRI_NAT_SRC
- 2,
1887 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1888 * or VS/NAT(change destination), so that filtering rules can be
1889 * applied to IPVS. */
1891 .hook
= ip_vs_remote_request6
,
1892 .owner
= THIS_MODULE
,
1894 .hooknum
= NF_INET_LOCAL_IN
,
1895 .priority
= NF_IP6_PRI_NAT_SRC
- 1,
1897 /* Before ip_vs_in, change source only for VS/NAT */
1899 .hook
= ip_vs_local_reply6
,
1900 .owner
= THIS_MODULE
,
1902 .hooknum
= NF_INET_LOCAL_OUT
,
1903 .priority
= NF_IP6_PRI_NAT_DST
+ 1,
1905 /* After mangle, schedule and forward local requests */
1907 .hook
= ip_vs_local_request6
,
1908 .owner
= THIS_MODULE
,
1910 .hooknum
= NF_INET_LOCAL_OUT
,
1911 .priority
= NF_IP6_PRI_NAT_DST
+ 2,
1913 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1914 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1916 .hook
= ip_vs_forward_icmp_v6
,
1917 .owner
= THIS_MODULE
,
1919 .hooknum
= NF_INET_FORWARD
,
1922 /* After packet filtering, change source only for VS/NAT */
1924 .hook
= ip_vs_reply6
,
1925 .owner
= THIS_MODULE
,
1927 .hooknum
= NF_INET_FORWARD
,
1933 * Initialize IP Virtual Server netns mem.
1935 static int __net_init
__ip_vs_init(struct net
*net
)
1937 struct netns_ipvs
*ipvs
;
1939 ipvs
= net_generic(net
, ip_vs_net_id
);
1943 /* Hold the beast until a service is registerd */
1946 /* Counters used for creating unique names */
1947 ipvs
->gen
= atomic_read(&ipvs_netns_cnt
);
1948 atomic_inc(&ipvs_netns_cnt
);
1951 if (ip_vs_estimator_net_init(net
) < 0)
1952 goto estimator_fail
;
1954 if (ip_vs_control_net_init(net
) < 0)
1957 if (ip_vs_protocol_net_init(net
) < 0)
1960 if (ip_vs_app_net_init(net
) < 0)
1963 if (ip_vs_conn_net_init(net
) < 0)
1966 if (ip_vs_sync_net_init(net
) < 0)
1969 printk(KERN_INFO
"IPVS: Creating netns size=%zu id=%d\n",
1970 sizeof(struct netns_ipvs
), ipvs
->gen
);
1977 ip_vs_conn_net_cleanup(net
);
1979 ip_vs_app_net_cleanup(net
);
1981 ip_vs_protocol_net_cleanup(net
);
1983 ip_vs_control_net_cleanup(net
);
1985 ip_vs_estimator_net_cleanup(net
);
1991 static void __net_exit
__ip_vs_cleanup(struct net
*net
)
1993 ip_vs_service_net_cleanup(net
); /* ip_vs_flush() with locks */
1994 ip_vs_conn_net_cleanup(net
);
1995 ip_vs_app_net_cleanup(net
);
1996 ip_vs_protocol_net_cleanup(net
);
1997 ip_vs_control_net_cleanup(net
);
1998 ip_vs_estimator_net_cleanup(net
);
1999 IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net
)->gen
);
2003 static void __net_exit
__ip_vs_dev_cleanup(struct net
*net
)
2006 net_ipvs(net
)->enable
= 0; /* Disable packet reception */
2008 ip_vs_sync_net_cleanup(net
);
2012 static struct pernet_operations ipvs_core_ops
= {
2013 .init
= __ip_vs_init
,
2014 .exit
= __ip_vs_cleanup
,
2015 .id
= &ip_vs_net_id
,
2016 .size
= sizeof(struct netns_ipvs
),
2019 static struct pernet_operations ipvs_core_dev_ops
= {
2020 .exit
= __ip_vs_dev_cleanup
,
2024 * Initialize IP Virtual Server
2026 static int __init
ip_vs_init(void)
2030 ret
= ip_vs_control_init();
2032 pr_err("can't setup control.\n");
2036 ip_vs_protocol_init();
2038 ret
= ip_vs_conn_init();
2040 pr_err("can't setup connection table.\n");
2041 goto cleanup_protocol
;
2044 ret
= register_pernet_subsys(&ipvs_core_ops
); /* Alloc ip_vs struct */
2048 ret
= register_pernet_device(&ipvs_core_dev_ops
);
2052 ret
= nf_register_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2054 pr_err("can't register hooks.\n");
2058 ret
= ip_vs_register_nl_ioctl();
2060 pr_err("can't register netlink/ioctl.\n");
2064 pr_info("ipvs loaded.\n");
2069 nf_unregister_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2071 unregister_pernet_device(&ipvs_core_dev_ops
);
2073 unregister_pernet_subsys(&ipvs_core_ops
);
2075 ip_vs_conn_cleanup();
2077 ip_vs_protocol_cleanup();
2078 ip_vs_control_cleanup();
2083 static void __exit
ip_vs_cleanup(void)
2085 ip_vs_unregister_nl_ioctl();
2086 nf_unregister_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2087 unregister_pernet_device(&ipvs_core_dev_ops
);
2088 unregister_pernet_subsys(&ipvs_core_ops
); /* free ip_vs struct */
2089 ip_vs_conn_cleanup();
2090 ip_vs_protocol_cleanup();
2091 ip_vs_control_cleanup();
2092 pr_info("ipvs unloaded.\n");
2095 module_init(ip_vs_init
);
2096 module_exit(ip_vs_cleanup
);
2097 MODULE_LICENSE("GPL");