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
,
654 unsigned int hooknum
)
656 if (!sysctl_snat_reroute(skb
))
658 /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
659 if (NF_INET_LOCAL_IN
== hooknum
)
661 #ifdef CONFIG_IP_VS_IPV6
662 if (af
== AF_INET6
) {
663 struct dst_entry
*dst
= skb_dst(skb
);
665 if (dst
->dev
&& !(dst
->dev
->flags
& IFF_LOOPBACK
) &&
666 ip6_route_me_harder(skb
) != 0)
670 if (!(skb_rtable(skb
)->rt_flags
& RTCF_LOCAL
) &&
671 ip_route_me_harder(skb
, RTN_LOCAL
) != 0)
678 * Packet has been made sufficiently writable in caller
679 * - inout: 1=in->out, 0=out->in
681 void ip_vs_nat_icmp(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
682 struct ip_vs_conn
*cp
, int inout
)
684 struct iphdr
*iph
= ip_hdr(skb
);
685 unsigned int icmp_offset
= iph
->ihl
*4;
686 struct icmphdr
*icmph
= (struct icmphdr
*)(skb_network_header(skb
) +
688 struct iphdr
*ciph
= (struct iphdr
*)(icmph
+ 1);
691 iph
->saddr
= cp
->vaddr
.ip
;
693 ciph
->daddr
= cp
->vaddr
.ip
;
696 iph
->daddr
= cp
->daddr
.ip
;
698 ciph
->saddr
= cp
->daddr
.ip
;
702 /* the TCP/UDP/SCTP port */
703 if (IPPROTO_TCP
== ciph
->protocol
|| IPPROTO_UDP
== ciph
->protocol
||
704 IPPROTO_SCTP
== ciph
->protocol
) {
705 __be16
*ports
= (void *)ciph
+ ciph
->ihl
*4;
708 ports
[1] = cp
->vport
;
710 ports
[0] = cp
->dport
;
713 /* And finally the ICMP checksum */
715 icmph
->checksum
= ip_vs_checksum_complete(skb
, icmp_offset
);
716 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
719 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, (void *)ciph
- (void *)iph
,
720 "Forwarding altered outgoing ICMP");
722 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, (void *)ciph
- (void *)iph
,
723 "Forwarding altered incoming ICMP");
726 #ifdef CONFIG_IP_VS_IPV6
727 void ip_vs_nat_icmp_v6(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
728 struct ip_vs_conn
*cp
, int inout
)
730 struct ipv6hdr
*iph
= ipv6_hdr(skb
);
731 unsigned int icmp_offset
= 0;
732 unsigned int offs
= 0; /* header offset*/
734 struct icmp6hdr
*icmph
;
735 struct ipv6hdr
*ciph
;
736 unsigned short fragoffs
;
738 ipv6_find_hdr(skb
, &icmp_offset
, IPPROTO_ICMPV6
, &fragoffs
, NULL
);
739 icmph
= (struct icmp6hdr
*)(skb_network_header(skb
) + icmp_offset
);
740 offs
= icmp_offset
+ sizeof(struct icmp6hdr
);
741 ciph
= (struct ipv6hdr
*)(skb_network_header(skb
) + offs
);
743 protocol
= ipv6_find_hdr(skb
, &offs
, -1, &fragoffs
, NULL
);
746 iph
->saddr
= cp
->vaddr
.in6
;
747 ciph
->daddr
= cp
->vaddr
.in6
;
749 iph
->daddr
= cp
->daddr
.in6
;
750 ciph
->saddr
= cp
->daddr
.in6
;
753 /* the TCP/UDP/SCTP port */
754 if (!fragoffs
&& (IPPROTO_TCP
== protocol
|| IPPROTO_UDP
== protocol
||
755 IPPROTO_SCTP
== protocol
)) {
756 __be16
*ports
= (void *)(skb_network_header(skb
) + offs
);
758 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__
,
759 ntohs(inout
? ports
[1] : ports
[0]),
760 ntohs(inout
? cp
->vport
: cp
->dport
));
762 ports
[1] = cp
->vport
;
764 ports
[0] = cp
->dport
;
767 /* And finally the ICMP checksum */
768 icmph
->icmp6_cksum
= ~csum_ipv6_magic(&iph
->saddr
, &iph
->daddr
,
769 skb
->len
- icmp_offset
,
771 skb
->csum_start
= skb_network_header(skb
) - skb
->head
+ icmp_offset
;
772 skb
->csum_offset
= offsetof(struct icmp6hdr
, icmp6_cksum
);
773 skb
->ip_summed
= CHECKSUM_PARTIAL
;
776 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
,
777 (void *)ciph
- (void *)iph
,
778 "Forwarding altered outgoing ICMPv6");
780 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
,
781 (void *)ciph
- (void *)iph
,
782 "Forwarding altered incoming ICMPv6");
786 /* Handle relevant response ICMP messages - forward to the right
789 static int handle_response_icmp(int af
, struct sk_buff
*skb
,
790 union nf_inet_addr
*snet
,
791 __u8 protocol
, struct ip_vs_conn
*cp
,
792 struct ip_vs_protocol
*pp
,
793 unsigned int offset
, unsigned int ihl
,
794 unsigned int hooknum
)
796 unsigned int verdict
= NF_DROP
;
798 if (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
)
801 /* Ensure the checksum is correct */
802 if (!skb_csum_unnecessary(skb
) && ip_vs_checksum_complete(skb
, ihl
)) {
803 /* Failed checksum! */
804 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
805 IP_VS_DBG_ADDR(af
, snet
));
809 if (IPPROTO_TCP
== protocol
|| IPPROTO_UDP
== protocol
||
810 IPPROTO_SCTP
== protocol
)
811 offset
+= 2 * sizeof(__u16
);
812 if (!skb_make_writable(skb
, offset
))
815 #ifdef CONFIG_IP_VS_IPV6
817 ip_vs_nat_icmp_v6(skb
, pp
, cp
, 1);
820 ip_vs_nat_icmp(skb
, pp
, cp
, 1);
822 if (ip_vs_route_me_harder(af
, skb
, hooknum
))
825 /* do the statistics and put it back */
826 ip_vs_out_stats(cp
, skb
);
828 skb
->ipvs_property
= 1;
829 if (!(cp
->flags
& IP_VS_CONN_F_NFCT
))
832 ip_vs_update_conntrack(skb
, cp
, 0);
838 __ip_vs_conn_put(cp
);
844 * Handle ICMP messages in the inside-to-outside direction (outgoing).
845 * Find any that might be relevant, check against existing connections.
846 * Currently handles error types - unreachable, quench, ttl exceeded.
848 static int ip_vs_out_icmp(struct sk_buff
*skb
, int *related
,
849 unsigned int hooknum
)
852 struct icmphdr _icmph
, *ic
;
853 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
854 struct ip_vs_iphdr ciph
;
855 struct ip_vs_conn
*cp
;
856 struct ip_vs_protocol
*pp
;
857 unsigned int offset
, ihl
;
858 union nf_inet_addr snet
;
862 /* reassemble IP fragments */
863 if (ip_is_fragment(ip_hdr(skb
))) {
864 if (ip_vs_gather_frags(skb
, ip_vs_defrag_user(hooknum
)))
869 offset
= ihl
= iph
->ihl
* 4;
870 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
874 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
875 ic
->type
, ntohs(icmp_id(ic
)),
876 &iph
->saddr
, &iph
->daddr
);
879 * Work through seeing if this is for us.
880 * These checks are supposed to be in an order that means easy
881 * things are checked first to speed up processing.... however
882 * this means that some packets will manage to get a long way
883 * down this stack and then be rejected, but that's life.
885 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
886 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
887 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
892 /* Now find the contained IP header */
893 offset
+= sizeof(_icmph
);
894 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
896 return NF_ACCEPT
; /* The packet looks wrong, ignore */
898 pp
= ip_vs_proto_get(cih
->protocol
);
902 /* Is the embedded protocol header present? */
903 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
907 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, offset
,
908 "Checking outgoing ICMP for");
910 ip_vs_fill_ip4hdr(cih
, &ciph
);
912 /* The embedded headers contain source and dest in reverse order */
913 cp
= pp
->conn_out_get(AF_INET
, skb
, &ciph
, 1);
917 snet
.ip
= iph
->saddr
;
918 return handle_response_icmp(AF_INET
, skb
, &snet
, cih
->protocol
, cp
,
919 pp
, ciph
.len
, ihl
, hooknum
);
922 #ifdef CONFIG_IP_VS_IPV6
923 static int ip_vs_out_icmp_v6(struct sk_buff
*skb
, int *related
,
924 unsigned int hooknum
, struct ip_vs_iphdr
*ipvsh
)
926 struct icmp6hdr _icmph
, *ic
;
927 struct ipv6hdr _ip6h
, *ip6h
; /* The ip header contained within ICMP */
928 struct ip_vs_iphdr ciph
= {.flags
= 0, .fragoffs
= 0};/*Contained IP */
929 struct ip_vs_conn
*cp
;
930 struct ip_vs_protocol
*pp
;
931 union nf_inet_addr snet
;
932 unsigned int writable
;
935 ic
= frag_safe_skb_hp(skb
, ipvsh
->len
, sizeof(_icmph
), &_icmph
, ipvsh
);
940 * Work through seeing if this is for us.
941 * These checks are supposed to be in an order that means easy
942 * things are checked first to speed up processing.... however
943 * this means that some packets will manage to get a long way
944 * down this stack and then be rejected, but that's life.
946 if (ic
->icmp6_type
& ICMPV6_INFOMSG_MASK
) {
950 /* Fragment header that is before ICMP header tells us that:
951 * it's not an error message since they can't be fragmented.
953 if (ipvsh
->flags
& IP6_FH_F_FRAG
)
956 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
957 ic
->icmp6_type
, ntohs(icmpv6_id(ic
)),
958 &ipvsh
->saddr
, &ipvsh
->daddr
);
960 /* Now find the contained IP header */
961 ciph
.len
= ipvsh
->len
+ sizeof(_icmph
);
962 ip6h
= skb_header_pointer(skb
, ciph
.len
, sizeof(_ip6h
), &_ip6h
);
964 return NF_ACCEPT
; /* The packet looks wrong, ignore */
965 ciph
.saddr
.in6
= ip6h
->saddr
; /* conn_out_get() handles reverse order */
966 ciph
.daddr
.in6
= ip6h
->daddr
;
967 /* skip possible IPv6 exthdrs of contained IPv6 packet */
968 ciph
.protocol
= ipv6_find_hdr(skb
, &ciph
.len
, -1, &ciph
.fragoffs
, NULL
);
969 if (ciph
.protocol
< 0)
970 return NF_ACCEPT
; /* Contained IPv6 hdr looks wrong, ignore */
972 pp
= ip_vs_proto_get(ciph
.protocol
);
976 /* The embedded headers contain source and dest in reverse order */
977 cp
= pp
->conn_out_get(AF_INET6
, skb
, &ciph
, 1);
981 snet
.in6
= ciph
.saddr
.in6
;
983 return handle_response_icmp(AF_INET6
, skb
, &snet
, ciph
.protocol
, cp
,
984 pp
, writable
, sizeof(struct ipv6hdr
),
990 * Check if sctp chunc is ABORT chunk
992 static inline int is_sctp_abort(const struct sk_buff
*skb
, int nh_len
)
994 sctp_chunkhdr_t
*sch
, schunk
;
995 sch
= skb_header_pointer(skb
, nh_len
+ sizeof(sctp_sctphdr_t
),
996 sizeof(schunk
), &schunk
);
999 if (sch
->type
== SCTP_CID_ABORT
)
1004 static inline int is_tcp_reset(const struct sk_buff
*skb
, int nh_len
)
1006 struct tcphdr _tcph
, *th
;
1008 th
= skb_header_pointer(skb
, nh_len
, sizeof(_tcph
), &_tcph
);
1014 static inline bool is_new_conn(const struct sk_buff
*skb
,
1015 struct ip_vs_iphdr
*iph
)
1017 switch (iph
->protocol
) {
1019 struct tcphdr _tcph
, *th
;
1021 th
= skb_header_pointer(skb
, iph
->len
, sizeof(_tcph
), &_tcph
);
1026 case IPPROTO_SCTP
: {
1027 sctp_chunkhdr_t
*sch
, schunk
;
1029 sch
= skb_header_pointer(skb
, iph
->len
+ sizeof(sctp_sctphdr_t
),
1030 sizeof(schunk
), &schunk
);
1033 return sch
->type
== SCTP_CID_INIT
;
1040 /* Handle response packets: rewrite addresses and send away...
1043 handle_response(int af
, struct sk_buff
*skb
, struct ip_vs_proto_data
*pd
,
1044 struct ip_vs_conn
*cp
, struct ip_vs_iphdr
*iph
,
1045 unsigned int hooknum
)
1047 struct ip_vs_protocol
*pp
= pd
->pp
;
1049 IP_VS_DBG_PKT(11, af
, pp
, skb
, 0, "Outgoing packet");
1051 if (!skb_make_writable(skb
, iph
->len
))
1054 /* mangle the packet */
1055 if (pp
->snat_handler
&& !pp
->snat_handler(skb
, pp
, cp
, iph
))
1058 #ifdef CONFIG_IP_VS_IPV6
1060 ipv6_hdr(skb
)->saddr
= cp
->vaddr
.in6
;
1064 ip_hdr(skb
)->saddr
= cp
->vaddr
.ip
;
1065 ip_send_check(ip_hdr(skb
));
1069 * nf_iterate does not expect change in the skb->dst->dev.
1070 * It looks like it is not fatal to enable this code for hooks
1071 * where our handlers are at the end of the chain list and
1072 * when all next handlers use skb->dst->dev and not outdev.
1073 * It will definitely route properly the inout NAT traffic
1074 * when multiple paths are used.
1077 /* For policy routing, packets originating from this
1078 * machine itself may be routed differently to packets
1079 * passing through. We want this packet to be routed as
1080 * if it came from this machine itself. So re-compute
1081 * the routing information.
1083 if (ip_vs_route_me_harder(af
, skb
, hooknum
))
1086 IP_VS_DBG_PKT(10, af
, pp
, skb
, 0, "After SNAT");
1088 ip_vs_out_stats(cp
, skb
);
1089 ip_vs_set_state(cp
, IP_VS_DIR_OUTPUT
, skb
, pd
);
1090 skb
->ipvs_property
= 1;
1091 if (!(cp
->flags
& IP_VS_CONN_F_NFCT
))
1094 ip_vs_update_conntrack(skb
, cp
, 0);
1108 * Check if outgoing packet belongs to the established ip_vs_conn.
1111 ip_vs_out(unsigned int hooknum
, struct sk_buff
*skb
, int af
)
1113 struct net
*net
= NULL
;
1114 struct ip_vs_iphdr iph
;
1115 struct ip_vs_protocol
*pp
;
1116 struct ip_vs_proto_data
*pd
;
1117 struct ip_vs_conn
*cp
;
1121 /* Already marked as IPVS request or reply? */
1122 if (skb
->ipvs_property
)
1125 /* Bad... Do not break raw sockets */
1126 if (unlikely(skb
->sk
!= NULL
&& hooknum
== NF_INET_LOCAL_OUT
&&
1128 struct sock
*sk
= skb
->sk
;
1129 struct inet_sock
*inet
= inet_sk(skb
->sk
);
1131 if (inet
&& sk
->sk_family
== PF_INET
&& inet
->nodefrag
)
1135 if (unlikely(!skb_dst(skb
)))
1139 if (!net_ipvs(net
)->enable
)
1142 ip_vs_fill_iph_skb(af
, skb
, &iph
);
1143 #ifdef CONFIG_IP_VS_IPV6
1144 if (af
== AF_INET6
) {
1145 if (unlikely(iph
.protocol
== IPPROTO_ICMPV6
)) {
1147 int verdict
= ip_vs_out_icmp_v6(skb
, &related
,
1155 if (unlikely(iph
.protocol
== IPPROTO_ICMP
)) {
1157 int verdict
= ip_vs_out_icmp(skb
, &related
, hooknum
);
1163 pd
= ip_vs_proto_data_get(net
, iph
.protocol
);
1168 /* reassemble IP fragments */
1169 #ifdef CONFIG_IP_VS_IPV6
1172 if (unlikely(ip_is_fragment(ip_hdr(skb
)) && !pp
->dont_defrag
)) {
1173 if (ip_vs_gather_frags(skb
,
1174 ip_vs_defrag_user(hooknum
)))
1177 ip_vs_fill_ip4hdr(skb_network_header(skb
), &iph
);
1181 * Check if the packet belongs to an existing entry
1183 cp
= pp
->conn_out_get(af
, skb
, &iph
, 0);
1186 if (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
)
1188 return handle_response(af
, skb
, pd
, cp
, &iph
, hooknum
);
1190 if (sysctl_nat_icmp_send(net
) &&
1191 (pp
->protocol
== IPPROTO_TCP
||
1192 pp
->protocol
== IPPROTO_UDP
||
1193 pp
->protocol
== IPPROTO_SCTP
)) {
1194 __be16 _ports
[2], *pptr
;
1196 pptr
= frag_safe_skb_hp(skb
, iph
.len
,
1197 sizeof(_ports
), _ports
, &iph
);
1199 return NF_ACCEPT
; /* Not for me */
1200 if (ip_vs_has_real_service(net
, af
, iph
.protocol
, &iph
.saddr
,
1203 * Notify the real server: there is no
1204 * existing entry if it is not RST
1205 * packet or not TCP packet.
1207 if ((iph
.protocol
!= IPPROTO_TCP
&&
1208 iph
.protocol
!= IPPROTO_SCTP
)
1209 || ((iph
.protocol
== IPPROTO_TCP
1210 && !is_tcp_reset(skb
, iph
.len
))
1211 || (iph
.protocol
== IPPROTO_SCTP
1212 && !is_sctp_abort(skb
,
1214 #ifdef CONFIG_IP_VS_IPV6
1215 if (af
== AF_INET6
) {
1217 skb
->dev
= net
->loopback_dev
;
1219 ICMPV6_DEST_UNREACH
,
1220 ICMPV6_PORT_UNREACH
,
1226 ICMP_PORT_UNREACH
, 0);
1233 IP_VS_DBG_PKT(12, af
, pp
, skb
, 0,
1234 "ip_vs_out: packet continues traversal as normal");
1238 __ip_vs_conn_put(cp
);
1243 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1244 * used only for VS/NAT.
1245 * Check if packet is reply for established ip_vs_conn.
1248 ip_vs_reply4(unsigned int hooknum
, struct sk_buff
*skb
,
1249 const struct net_device
*in
, const struct net_device
*out
,
1250 int (*okfn
)(struct sk_buff
*))
1252 return ip_vs_out(hooknum
, skb
, AF_INET
);
1256 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1257 * Check if packet is reply for established ip_vs_conn.
1260 ip_vs_local_reply4(unsigned int hooknum
, struct sk_buff
*skb
,
1261 const struct net_device
*in
, const struct net_device
*out
,
1262 int (*okfn
)(struct sk_buff
*))
1264 return ip_vs_out(hooknum
, skb
, AF_INET
);
1267 #ifdef CONFIG_IP_VS_IPV6
1270 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1271 * used only for VS/NAT.
1272 * Check if packet is reply for established ip_vs_conn.
1275 ip_vs_reply6(unsigned int hooknum
, struct sk_buff
*skb
,
1276 const struct net_device
*in
, const struct net_device
*out
,
1277 int (*okfn
)(struct sk_buff
*))
1279 return ip_vs_out(hooknum
, skb
, AF_INET6
);
1283 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1284 * Check if packet is reply for established ip_vs_conn.
1287 ip_vs_local_reply6(unsigned int hooknum
, struct sk_buff
*skb
,
1288 const struct net_device
*in
, const struct net_device
*out
,
1289 int (*okfn
)(struct sk_buff
*))
1291 return ip_vs_out(hooknum
, skb
, AF_INET6
);
1297 * Handle ICMP messages in the outside-to-inside direction (incoming).
1298 * Find any that might be relevant, check against existing connections,
1299 * forward to the right destination host if relevant.
1300 * Currently handles error types - unreachable, quench, ttl exceeded.
1303 ip_vs_in_icmp(struct sk_buff
*skb
, int *related
, unsigned int hooknum
)
1305 struct net
*net
= NULL
;
1307 struct icmphdr _icmph
, *ic
;
1308 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
1309 struct ip_vs_iphdr ciph
;
1310 struct ip_vs_conn
*cp
;
1311 struct ip_vs_protocol
*pp
;
1312 struct ip_vs_proto_data
*pd
;
1313 unsigned int offset
, offset2
, ihl
, verdict
;
1318 /* reassemble IP fragments */
1319 if (ip_is_fragment(ip_hdr(skb
))) {
1320 if (ip_vs_gather_frags(skb
, ip_vs_defrag_user(hooknum
)))
1325 offset
= ihl
= iph
->ihl
* 4;
1326 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
1330 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1331 ic
->type
, ntohs(icmp_id(ic
)),
1332 &iph
->saddr
, &iph
->daddr
);
1335 * Work through seeing if this is for us.
1336 * These checks are supposed to be in an order that means easy
1337 * things are checked first to speed up processing.... however
1338 * this means that some packets will manage to get a long way
1339 * down this stack and then be rejected, but that's life.
1341 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
1342 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
1343 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
1348 /* Now find the contained IP header */
1349 offset
+= sizeof(_icmph
);
1350 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
1352 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1356 /* Special case for errors for IPIP packets */
1358 if (cih
->protocol
== IPPROTO_IPIP
) {
1359 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
)))
1361 /* Error for our IPIP must arrive at LOCAL_IN */
1362 if (!(skb_rtable(skb
)->rt_flags
& RTCF_LOCAL
))
1364 offset
+= cih
->ihl
* 4;
1365 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
1367 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1371 pd
= ip_vs_proto_data_get(net
, cih
->protocol
);
1376 /* Is the embedded protocol header present? */
1377 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
1381 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, offset
,
1382 "Checking incoming ICMP for");
1385 ip_vs_fill_ip4hdr(cih
, &ciph
);
1388 /* The embedded headers contain source and dest in reverse order.
1389 * For IPIP this is error for request, not for reply.
1391 cp
= pp
->conn_in_get(AF_INET
, skb
, &ciph
, ipip
? 0 : 1);
1397 /* Ensure the checksum is correct */
1398 if (!skb_csum_unnecessary(skb
) && ip_vs_checksum_complete(skb
, ihl
)) {
1399 /* Failed checksum! */
1400 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1406 __be32 info
= ic
->un
.gateway
;
1407 __u8 type
= ic
->type
;
1408 __u8 code
= ic
->code
;
1410 /* Update the MTU */
1411 if (ic
->type
== ICMP_DEST_UNREACH
&&
1412 ic
->code
== ICMP_FRAG_NEEDED
) {
1413 struct ip_vs_dest
*dest
= cp
->dest
;
1414 u32 mtu
= ntohs(ic
->un
.frag
.mtu
);
1415 __be16 frag_off
= cih
->frag_off
;
1417 /* Strip outer IP and ICMP, go to IPIP header */
1418 if (pskb_pull(skb
, ihl
+ sizeof(_icmph
)) == NULL
)
1420 offset2
-= ihl
+ sizeof(_icmph
);
1421 skb_reset_network_header(skb
);
1422 IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1423 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
, mtu
);
1424 ipv4_update_pmtu(skb
, dev_net(skb
->dev
),
1426 /* Client uses PMTUD? */
1427 if (!(frag_off
& htons(IP_DF
)))
1429 /* Prefer the resulting PMTU */
1431 struct ip_vs_dest_dst
*dest_dst
;
1434 dest_dst
= rcu_dereference(dest
->dest_dst
);
1436 mtu
= dst_mtu(dest_dst
->dst_cache
);
1439 if (mtu
> 68 + sizeof(struct iphdr
))
1440 mtu
-= sizeof(struct iphdr
);
1443 /* Strip outer IP, ICMP and IPIP, go to IP header of
1446 if (pskb_pull(skb
, offset2
) == NULL
)
1448 skb_reset_network_header(skb
);
1449 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1450 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1451 type
, code
, ntohl(info
));
1452 icmp_send(skb
, type
, code
, info
);
1453 /* ICMP can be shorter but anyways, account it */
1454 ip_vs_out_stats(cp
, skb
);
1458 verdict
= NF_STOLEN
;
1462 /* do the statistics and put it back */
1463 ip_vs_in_stats(cp
, skb
);
1464 if (IPPROTO_TCP
== cih
->protocol
|| IPPROTO_UDP
== cih
->protocol
||
1465 IPPROTO_SCTP
== cih
->protocol
)
1466 offset
+= 2 * sizeof(__u16
);
1467 verdict
= ip_vs_icmp_xmit(skb
, cp
, pp
, offset
, hooknum
, &ciph
);
1470 __ip_vs_conn_put(cp
);
1475 #ifdef CONFIG_IP_VS_IPV6
1476 static int ip_vs_in_icmp_v6(struct sk_buff
*skb
, int *related
,
1477 unsigned int hooknum
, struct ip_vs_iphdr
*iph
)
1479 struct net
*net
= NULL
;
1480 struct ipv6hdr _ip6h
, *ip6h
;
1481 struct icmp6hdr _icmph
, *ic
;
1482 struct ip_vs_iphdr ciph
= {.flags
= 0, .fragoffs
= 0};/*Contained IP */
1483 struct ip_vs_conn
*cp
;
1484 struct ip_vs_protocol
*pp
;
1485 struct ip_vs_proto_data
*pd
;
1486 unsigned int offs_ciph
, writable
, verdict
;
1490 ic
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_icmph
), &_icmph
, iph
);
1495 * Work through seeing if this is for us.
1496 * These checks are supposed to be in an order that means easy
1497 * things are checked first to speed up processing.... however
1498 * this means that some packets will manage to get a long way
1499 * down this stack and then be rejected, but that's life.
1501 if (ic
->icmp6_type
& ICMPV6_INFOMSG_MASK
) {
1505 /* Fragment header that is before ICMP header tells us that:
1506 * it's not an error message since they can't be fragmented.
1508 if (iph
->flags
& IP6_FH_F_FRAG
)
1511 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1512 ic
->icmp6_type
, ntohs(icmpv6_id(ic
)),
1513 &iph
->saddr
, &iph
->daddr
);
1515 /* Now find the contained IP header */
1516 ciph
.len
= iph
->len
+ sizeof(_icmph
);
1517 offs_ciph
= ciph
.len
; /* Save ip header offset */
1518 ip6h
= skb_header_pointer(skb
, ciph
.len
, sizeof(_ip6h
), &_ip6h
);
1520 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1521 ciph
.saddr
.in6
= ip6h
->saddr
; /* conn_in_get() handles reverse order */
1522 ciph
.daddr
.in6
= ip6h
->daddr
;
1523 /* skip possible IPv6 exthdrs of contained IPv6 packet */
1524 ciph
.protocol
= ipv6_find_hdr(skb
, &ciph
.len
, -1, &ciph
.fragoffs
, NULL
);
1525 if (ciph
.protocol
< 0)
1526 return NF_ACCEPT
; /* Contained IPv6 hdr looks wrong, ignore */
1529 pd
= ip_vs_proto_data_get(net
, ciph
.protocol
);
1534 /* Cannot handle fragmented embedded protocol */
1538 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
, offs_ciph
,
1539 "Checking incoming ICMPv6 for");
1541 /* The embedded headers contain source and dest in reverse order
1542 * if not from localhost
1544 cp
= pp
->conn_in_get(AF_INET6
, skb
, &ciph
,
1545 (hooknum
== NF_INET_LOCAL_OUT
) ? 0 : 1);
1549 /* VS/TUN, VS/DR and LOCALNODE just let it go */
1550 if ((hooknum
== NF_INET_LOCAL_OUT
) &&
1551 (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
)) {
1552 __ip_vs_conn_put(cp
);
1556 /* do the statistics and put it back */
1557 ip_vs_in_stats(cp
, skb
);
1559 /* Need to mangle contained IPv6 header in ICMPv6 packet */
1560 writable
= ciph
.len
;
1561 if (IPPROTO_TCP
== ciph
.protocol
|| IPPROTO_UDP
== ciph
.protocol
||
1562 IPPROTO_SCTP
== ciph
.protocol
)
1563 writable
+= 2 * sizeof(__u16
); /* Also mangle ports */
1565 verdict
= ip_vs_icmp_xmit_v6(skb
, cp
, pp
, writable
, hooknum
, &ciph
);
1567 __ip_vs_conn_put(cp
);
1575 * Check if it's for virtual services, look it up,
1576 * and send it on its way...
1579 ip_vs_in(unsigned int hooknum
, struct sk_buff
*skb
, int af
)
1582 struct ip_vs_iphdr iph
;
1583 struct ip_vs_protocol
*pp
;
1584 struct ip_vs_proto_data
*pd
;
1585 struct ip_vs_conn
*cp
;
1587 struct netns_ipvs
*ipvs
;
1589 /* Already marked as IPVS request or reply? */
1590 if (skb
->ipvs_property
)
1595 * - remote client: only PACKET_HOST
1596 * - route: used for struct net when skb->dev is unset
1598 if (unlikely((skb
->pkt_type
!= PACKET_HOST
&&
1599 hooknum
!= NF_INET_LOCAL_OUT
) ||
1601 ip_vs_fill_iph_skb(af
, skb
, &iph
);
1602 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1603 " ignored in hook %u\n",
1604 skb
->pkt_type
, iph
.protocol
,
1605 IP_VS_DBG_ADDR(af
, &iph
.daddr
), hooknum
);
1608 /* ipvs enabled in this netns ? */
1610 ipvs
= net_ipvs(net
);
1611 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1614 ip_vs_fill_iph_skb(af
, skb
, &iph
);
1616 /* Bad... Do not break raw sockets */
1617 if (unlikely(skb
->sk
!= NULL
&& hooknum
== NF_INET_LOCAL_OUT
&&
1619 struct sock
*sk
= skb
->sk
;
1620 struct inet_sock
*inet
= inet_sk(skb
->sk
);
1622 if (inet
&& sk
->sk_family
== PF_INET
&& inet
->nodefrag
)
1626 #ifdef CONFIG_IP_VS_IPV6
1627 if (af
== AF_INET6
) {
1628 if (unlikely(iph
.protocol
== IPPROTO_ICMPV6
)) {
1630 int verdict
= ip_vs_in_icmp_v6(skb
, &related
, hooknum
,
1638 if (unlikely(iph
.protocol
== IPPROTO_ICMP
)) {
1640 int verdict
= ip_vs_in_icmp(skb
, &related
, hooknum
);
1646 /* Protocol supported? */
1647 pd
= ip_vs_proto_data_get(net
, iph
.protocol
);
1652 * Check if the packet belongs to an existing connection entry
1654 cp
= pp
->conn_in_get(af
, skb
, &iph
, 0);
1656 if (unlikely(sysctl_expire_nodest_conn(ipvs
)) && cp
&& cp
->dest
&&
1657 unlikely(!atomic_read(&cp
->dest
->weight
)) && !iph
.fragoffs
&&
1658 is_new_conn(skb
, &iph
)) {
1659 ip_vs_conn_expire_now(cp
);
1660 __ip_vs_conn_put(cp
);
1664 if (unlikely(!cp
) && !iph
.fragoffs
) {
1665 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1666 * replayed fragment zero will already have created the cp
1670 /* Schedule and create new connection entry into &cp */
1671 if (!pp
->conn_schedule(af
, skb
, pd
, &v
, &cp
, &iph
))
1675 if (unlikely(!cp
)) {
1676 /* sorry, all this trouble for a no-hit :) */
1677 IP_VS_DBG_PKT(12, af
, pp
, skb
, 0,
1678 "ip_vs_in: packet continues traversal as normal");
1680 /* Fragment that couldn't be mapped to a conn entry
1681 * is missing module nf_defrag_ipv6
1683 IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1684 IP_VS_DBG_PKT(7, af
, pp
, skb
, 0, "unhandled fragment");
1689 IP_VS_DBG_PKT(11, af
, pp
, skb
, 0, "Incoming packet");
1690 /* Check the server status */
1691 if (cp
->dest
&& !(cp
->dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
1692 /* the destination server is not available */
1694 if (sysctl_expire_nodest_conn(ipvs
)) {
1695 /* try to expire the connection immediately */
1696 ip_vs_conn_expire_now(cp
);
1698 /* don't restart its timer, and silently
1700 __ip_vs_conn_put(cp
);
1704 ip_vs_in_stats(cp
, skb
);
1705 ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pd
);
1706 if (cp
->packet_xmit
)
1707 ret
= cp
->packet_xmit(skb
, cp
, pp
, &iph
);
1708 /* do not touch skb anymore */
1710 IP_VS_DBG_RL("warning: packet_xmit is null");
1714 /* Increase its packet counter and check if it is needed
1715 * to be synchronized
1717 * Sync connection if it is about to close to
1718 * encorage the standby servers to update the connections timeout
1720 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1723 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
1724 pkts
= sysctl_sync_threshold(ipvs
);
1726 pkts
= atomic_add_return(1, &cp
->in_pkts
);
1728 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
)
1729 ip_vs_sync_conn(net
, cp
, pkts
);
1736 * AF_INET handler in NF_INET_LOCAL_IN chain
1737 * Schedule and forward packets from remote clients
1740 ip_vs_remote_request4(unsigned int hooknum
, struct sk_buff
*skb
,
1741 const struct net_device
*in
,
1742 const struct net_device
*out
,
1743 int (*okfn
)(struct sk_buff
*))
1745 return ip_vs_in(hooknum
, skb
, AF_INET
);
1749 * AF_INET handler in NF_INET_LOCAL_OUT chain
1750 * Schedule and forward packets from local clients
1753 ip_vs_local_request4(unsigned int hooknum
, struct sk_buff
*skb
,
1754 const struct net_device
*in
, const struct net_device
*out
,
1755 int (*okfn
)(struct sk_buff
*))
1757 return ip_vs_in(hooknum
, skb
, AF_INET
);
1760 #ifdef CONFIG_IP_VS_IPV6
1763 * AF_INET6 handler in NF_INET_LOCAL_IN chain
1764 * Schedule and forward packets from remote clients
1767 ip_vs_remote_request6(unsigned int hooknum
, struct sk_buff
*skb
,
1768 const struct net_device
*in
,
1769 const struct net_device
*out
,
1770 int (*okfn
)(struct sk_buff
*))
1772 return ip_vs_in(hooknum
, skb
, AF_INET6
);
1776 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
1777 * Schedule and forward packets from local clients
1780 ip_vs_local_request6(unsigned int hooknum
, struct sk_buff
*skb
,
1781 const struct net_device
*in
, const struct net_device
*out
,
1782 int (*okfn
)(struct sk_buff
*))
1784 return ip_vs_in(hooknum
, skb
, AF_INET6
);
1791 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1792 * related packets destined for 0.0.0.0/0.
1793 * When fwmark-based virtual service is used, such as transparent
1794 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1795 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1796 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1797 * and send them to ip_vs_in_icmp.
1800 ip_vs_forward_icmp(unsigned int hooknum
, struct sk_buff
*skb
,
1801 const struct net_device
*in
, const struct net_device
*out
,
1802 int (*okfn
)(struct sk_buff
*))
1806 struct netns_ipvs
*ipvs
;
1808 if (ip_hdr(skb
)->protocol
!= IPPROTO_ICMP
)
1811 /* ipvs enabled in this netns ? */
1813 ipvs
= net_ipvs(net
);
1814 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1817 return ip_vs_in_icmp(skb
, &r
, hooknum
);
1820 #ifdef CONFIG_IP_VS_IPV6
1822 ip_vs_forward_icmp_v6(unsigned int hooknum
, struct sk_buff
*skb
,
1823 const struct net_device
*in
, const struct net_device
*out
,
1824 int (*okfn
)(struct sk_buff
*))
1828 struct netns_ipvs
*ipvs
;
1829 struct ip_vs_iphdr iphdr
;
1831 ip_vs_fill_iph_skb(AF_INET6
, skb
, &iphdr
);
1832 if (iphdr
.protocol
!= IPPROTO_ICMPV6
)
1835 /* ipvs enabled in this netns ? */
1837 ipvs
= net_ipvs(net
);
1838 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1841 return ip_vs_in_icmp_v6(skb
, &r
, hooknum
, &iphdr
);
1846 static struct nf_hook_ops ip_vs_ops
[] __read_mostly
= {
1847 /* After packet filtering, change source only for VS/NAT */
1849 .hook
= ip_vs_reply4
,
1850 .owner
= THIS_MODULE
,
1852 .hooknum
= NF_INET_LOCAL_IN
,
1853 .priority
= NF_IP_PRI_NAT_SRC
- 2,
1855 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1856 * or VS/NAT(change destination), so that filtering rules can be
1857 * applied to IPVS. */
1859 .hook
= ip_vs_remote_request4
,
1860 .owner
= THIS_MODULE
,
1862 .hooknum
= NF_INET_LOCAL_IN
,
1863 .priority
= NF_IP_PRI_NAT_SRC
- 1,
1865 /* Before ip_vs_in, change source only for VS/NAT */
1867 .hook
= ip_vs_local_reply4
,
1868 .owner
= THIS_MODULE
,
1870 .hooknum
= NF_INET_LOCAL_OUT
,
1871 .priority
= NF_IP_PRI_NAT_DST
+ 1,
1873 /* After mangle, schedule and forward local requests */
1875 .hook
= ip_vs_local_request4
,
1876 .owner
= THIS_MODULE
,
1878 .hooknum
= NF_INET_LOCAL_OUT
,
1879 .priority
= NF_IP_PRI_NAT_DST
+ 2,
1881 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1882 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1884 .hook
= ip_vs_forward_icmp
,
1885 .owner
= THIS_MODULE
,
1887 .hooknum
= NF_INET_FORWARD
,
1890 /* After packet filtering, change source only for VS/NAT */
1892 .hook
= ip_vs_reply4
,
1893 .owner
= THIS_MODULE
,
1895 .hooknum
= NF_INET_FORWARD
,
1898 #ifdef CONFIG_IP_VS_IPV6
1899 /* After packet filtering, change source only for VS/NAT */
1901 .hook
= ip_vs_reply6
,
1902 .owner
= THIS_MODULE
,
1904 .hooknum
= NF_INET_LOCAL_IN
,
1905 .priority
= NF_IP6_PRI_NAT_SRC
- 2,
1907 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1908 * or VS/NAT(change destination), so that filtering rules can be
1909 * applied to IPVS. */
1911 .hook
= ip_vs_remote_request6
,
1912 .owner
= THIS_MODULE
,
1914 .hooknum
= NF_INET_LOCAL_IN
,
1915 .priority
= NF_IP6_PRI_NAT_SRC
- 1,
1917 /* Before ip_vs_in, change source only for VS/NAT */
1919 .hook
= ip_vs_local_reply6
,
1920 .owner
= THIS_MODULE
,
1922 .hooknum
= NF_INET_LOCAL_OUT
,
1923 .priority
= NF_IP6_PRI_NAT_DST
+ 1,
1925 /* After mangle, schedule and forward local requests */
1927 .hook
= ip_vs_local_request6
,
1928 .owner
= THIS_MODULE
,
1930 .hooknum
= NF_INET_LOCAL_OUT
,
1931 .priority
= NF_IP6_PRI_NAT_DST
+ 2,
1933 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1934 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1936 .hook
= ip_vs_forward_icmp_v6
,
1937 .owner
= THIS_MODULE
,
1939 .hooknum
= NF_INET_FORWARD
,
1942 /* After packet filtering, change source only for VS/NAT */
1944 .hook
= ip_vs_reply6
,
1945 .owner
= THIS_MODULE
,
1947 .hooknum
= NF_INET_FORWARD
,
1953 * Initialize IP Virtual Server netns mem.
1955 static int __net_init
__ip_vs_init(struct net
*net
)
1957 struct netns_ipvs
*ipvs
;
1959 ipvs
= net_generic(net
, ip_vs_net_id
);
1963 /* Hold the beast until a service is registerd */
1966 /* Counters used for creating unique names */
1967 ipvs
->gen
= atomic_read(&ipvs_netns_cnt
);
1968 atomic_inc(&ipvs_netns_cnt
);
1971 if (ip_vs_estimator_net_init(net
) < 0)
1972 goto estimator_fail
;
1974 if (ip_vs_control_net_init(net
) < 0)
1977 if (ip_vs_protocol_net_init(net
) < 0)
1980 if (ip_vs_app_net_init(net
) < 0)
1983 if (ip_vs_conn_net_init(net
) < 0)
1986 if (ip_vs_sync_net_init(net
) < 0)
1989 printk(KERN_INFO
"IPVS: Creating netns size=%zu id=%d\n",
1990 sizeof(struct netns_ipvs
), ipvs
->gen
);
1997 ip_vs_conn_net_cleanup(net
);
1999 ip_vs_app_net_cleanup(net
);
2001 ip_vs_protocol_net_cleanup(net
);
2003 ip_vs_control_net_cleanup(net
);
2005 ip_vs_estimator_net_cleanup(net
);
2011 static void __net_exit
__ip_vs_cleanup(struct net
*net
)
2013 ip_vs_service_net_cleanup(net
); /* ip_vs_flush() with locks */
2014 ip_vs_conn_net_cleanup(net
);
2015 ip_vs_app_net_cleanup(net
);
2016 ip_vs_protocol_net_cleanup(net
);
2017 ip_vs_control_net_cleanup(net
);
2018 ip_vs_estimator_net_cleanup(net
);
2019 IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net
)->gen
);
2023 static void __net_exit
__ip_vs_dev_cleanup(struct net
*net
)
2026 net_ipvs(net
)->enable
= 0; /* Disable packet reception */
2028 ip_vs_sync_net_cleanup(net
);
2032 static struct pernet_operations ipvs_core_ops
= {
2033 .init
= __ip_vs_init
,
2034 .exit
= __ip_vs_cleanup
,
2035 .id
= &ip_vs_net_id
,
2036 .size
= sizeof(struct netns_ipvs
),
2039 static struct pernet_operations ipvs_core_dev_ops
= {
2040 .exit
= __ip_vs_dev_cleanup
,
2044 * Initialize IP Virtual Server
2046 static int __init
ip_vs_init(void)
2050 ret
= ip_vs_control_init();
2052 pr_err("can't setup control.\n");
2056 ip_vs_protocol_init();
2058 ret
= ip_vs_conn_init();
2060 pr_err("can't setup connection table.\n");
2061 goto cleanup_protocol
;
2064 ret
= register_pernet_subsys(&ipvs_core_ops
); /* Alloc ip_vs struct */
2068 ret
= register_pernet_device(&ipvs_core_dev_ops
);
2072 ret
= nf_register_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2074 pr_err("can't register hooks.\n");
2078 ret
= ip_vs_register_nl_ioctl();
2080 pr_err("can't register netlink/ioctl.\n");
2084 pr_info("ipvs loaded.\n");
2089 nf_unregister_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2091 unregister_pernet_device(&ipvs_core_dev_ops
);
2093 unregister_pernet_subsys(&ipvs_core_ops
);
2095 ip_vs_conn_cleanup();
2097 ip_vs_protocol_cleanup();
2098 ip_vs_control_cleanup();
2103 static void __exit
ip_vs_cleanup(void)
2105 ip_vs_unregister_nl_ioctl();
2106 nf_unregister_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2107 unregister_pernet_device(&ipvs_core_dev_ops
);
2108 unregister_pernet_subsys(&ipvs_core_ops
); /* free ip_vs struct */
2109 ip_vs_conn_cleanup();
2110 ip_vs_protocol_cleanup();
2111 ip_vs_control_cleanup();
2112 pr_info("ipvs unloaded.\n");
2115 module_init(ip_vs_init
);
2116 module_exit(ip_vs_cleanup
);
2117 MODULE_LICENSE("GPL");