2 * IPVS An implementation of the IP virtual server support for the
3 * LINUX operating system. IPVS is now implemented as a module
4 * over the Netfilter framework. IPVS can be used to build a
5 * high-performance and highly available server based on a
8 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
9 * Peter Kese <peter.kese@ijs.si>
10 * Julian Anastasov <ja@ssi.bg>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
17 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
18 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
22 * Paul `Rusty' Russell properly handle non-linear skbs
23 * Harald Welte don't use nfcache
27 #define KMSG_COMPONENT "IPVS"
28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
30 #include <linux/module.h>
31 #include <linux/kernel.h>
33 #include <linux/tcp.h>
34 #include <linux/sctp.h>
35 #include <linux/icmp.h>
36 #include <linux/slab.h>
41 #include <net/icmp.h> /* for icmp_send */
42 #include <net/route.h>
43 #include <net/ip6_checksum.h>
44 #include <net/netns/generic.h> /* net_generic() */
46 #include <linux/netfilter.h>
47 #include <linux/netfilter_ipv4.h>
49 #ifdef CONFIG_IP_VS_IPV6
51 #include <linux/netfilter_ipv6.h>
52 #include <net/ip6_route.h>
55 #include <net/ip_vs.h>
58 EXPORT_SYMBOL(register_ip_vs_scheduler
);
59 EXPORT_SYMBOL(unregister_ip_vs_scheduler
);
60 EXPORT_SYMBOL(ip_vs_proto_name
);
61 EXPORT_SYMBOL(ip_vs_conn_new
);
62 EXPORT_SYMBOL(ip_vs_conn_in_get
);
63 EXPORT_SYMBOL(ip_vs_conn_out_get
);
64 #ifdef CONFIG_IP_VS_PROTO_TCP
65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen
);
67 EXPORT_SYMBOL(ip_vs_conn_put
);
68 #ifdef CONFIG_IP_VS_DEBUG
69 EXPORT_SYMBOL(ip_vs_get_debug_level
);
72 static int ip_vs_net_id __read_mostly
;
73 /* netns cnt used for uniqueness */
74 static atomic_t ipvs_netns_cnt
= ATOMIC_INIT(0);
76 /* ID used in ICMP lookups */
77 #define icmp_id(icmph) (((icmph)->un).echo.id)
78 #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier)
80 const char *ip_vs_proto_name(unsigned int proto
)
95 #ifdef CONFIG_IP_VS_IPV6
100 sprintf(buf
, "IP_%u", proto
);
105 void ip_vs_init_hash_table(struct list_head
*table
, int rows
)
108 INIT_LIST_HEAD(&table
[rows
]);
112 ip_vs_in_stats(struct ip_vs_conn
*cp
, struct sk_buff
*skb
)
114 struct ip_vs_dest
*dest
= cp
->dest
;
115 struct netns_ipvs
*ipvs
= net_ipvs(skb_net(skb
));
117 if (dest
&& (dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
118 struct ip_vs_cpu_stats
*s
;
119 struct ip_vs_service
*svc
;
121 s
= this_cpu_ptr(dest
->stats
.cpustats
);
123 u64_stats_update_begin(&s
->syncp
);
124 s
->ustats
.inbytes
+= skb
->len
;
125 u64_stats_update_end(&s
->syncp
);
128 svc
= rcu_dereference(dest
->svc
);
129 s
= this_cpu_ptr(svc
->stats
.cpustats
);
131 u64_stats_update_begin(&s
->syncp
);
132 s
->ustats
.inbytes
+= skb
->len
;
133 u64_stats_update_end(&s
->syncp
);
136 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
138 u64_stats_update_begin(&s
->syncp
);
139 s
->ustats
.inbytes
+= skb
->len
;
140 u64_stats_update_end(&s
->syncp
);
146 ip_vs_out_stats(struct ip_vs_conn
*cp
, struct sk_buff
*skb
)
148 struct ip_vs_dest
*dest
= cp
->dest
;
149 struct netns_ipvs
*ipvs
= net_ipvs(skb_net(skb
));
151 if (dest
&& (dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
152 struct ip_vs_cpu_stats
*s
;
153 struct ip_vs_service
*svc
;
155 s
= this_cpu_ptr(dest
->stats
.cpustats
);
157 u64_stats_update_begin(&s
->syncp
);
158 s
->ustats
.outbytes
+= skb
->len
;
159 u64_stats_update_end(&s
->syncp
);
162 svc
= rcu_dereference(dest
->svc
);
163 s
= this_cpu_ptr(svc
->stats
.cpustats
);
165 u64_stats_update_begin(&s
->syncp
);
166 s
->ustats
.outbytes
+= skb
->len
;
167 u64_stats_update_end(&s
->syncp
);
170 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
172 u64_stats_update_begin(&s
->syncp
);
173 s
->ustats
.outbytes
+= skb
->len
;
174 u64_stats_update_end(&s
->syncp
);
180 ip_vs_conn_stats(struct ip_vs_conn
*cp
, struct ip_vs_service
*svc
)
182 struct netns_ipvs
*ipvs
= net_ipvs(svc
->net
);
183 struct ip_vs_cpu_stats
*s
;
185 s
= this_cpu_ptr(cp
->dest
->stats
.cpustats
);
188 s
= this_cpu_ptr(svc
->stats
.cpustats
);
191 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
197 ip_vs_set_state(struct ip_vs_conn
*cp
, int direction
,
198 const struct sk_buff
*skb
,
199 struct ip_vs_proto_data
*pd
)
201 if (likely(pd
->pp
->state_transition
))
202 pd
->pp
->state_transition(cp
, direction
, skb
, pd
);
206 ip_vs_conn_fill_param_persist(const struct ip_vs_service
*svc
,
207 struct sk_buff
*skb
, int protocol
,
208 const union nf_inet_addr
*caddr
, __be16 cport
,
209 const union nf_inet_addr
*vaddr
, __be16 vport
,
210 struct ip_vs_conn_param
*p
)
212 ip_vs_conn_fill_param(svc
->net
, svc
->af
, protocol
, caddr
, cport
, vaddr
,
214 p
->pe
= rcu_dereference(svc
->pe
);
215 if (p
->pe
&& p
->pe
->fill_param
)
216 return p
->pe
->fill_param(p
, skb
);
222 * IPVS persistent scheduling function
223 * It creates a connection entry according to its template if exists,
224 * or selects a server and creates a connection entry plus a template.
225 * Locking: we are svc user (svc->refcnt), so we hold all dests too
226 * Protocols supported: TCP, UDP
228 static struct ip_vs_conn
*
229 ip_vs_sched_persist(struct ip_vs_service
*svc
,
230 struct sk_buff
*skb
, __be16 src_port
, __be16 dst_port
,
231 int *ignored
, struct ip_vs_iphdr
*iph
)
233 struct ip_vs_conn
*cp
= NULL
;
234 struct ip_vs_dest
*dest
;
235 struct ip_vs_conn
*ct
;
236 __be16 dport
= 0; /* destination port to forward */
238 struct ip_vs_conn_param param
;
239 const union nf_inet_addr fwmark
= { .ip
= htonl(svc
->fwmark
) };
240 union nf_inet_addr snet
; /* source network of the client,
243 /* Mask saddr with the netmask to adjust template granularity */
244 #ifdef CONFIG_IP_VS_IPV6
245 if (svc
->af
== AF_INET6
)
246 ipv6_addr_prefix(&snet
.in6
, &iph
->saddr
.in6
,
247 (__force __u32
) svc
->netmask
);
250 snet
.ip
= iph
->saddr
.ip
& svc
->netmask
;
252 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
254 IP_VS_DBG_ADDR(svc
->af
, &iph
->saddr
), ntohs(src_port
),
255 IP_VS_DBG_ADDR(svc
->af
, &iph
->daddr
), ntohs(dst_port
),
256 IP_VS_DBG_ADDR(svc
->af
, &snet
));
259 * As far as we know, FTP is a very complicated network protocol, and
260 * it uses control connection and data connections. For active FTP,
261 * FTP server initialize data connection to the client, its source port
262 * is often 20. For passive FTP, FTP server tells the clients the port
263 * that it passively listens to, and the client issues the data
264 * connection. In the tunneling or direct routing mode, the load
265 * balancer is on the client-to-server half of connection, the port
266 * number is unknown to the load balancer. So, a conn template like
267 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
268 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
269 * is created for other persistent services.
272 int protocol
= iph
->protocol
;
273 const union nf_inet_addr
*vaddr
= &iph
->daddr
;
276 if (dst_port
== svc
->port
) {
278 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
280 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
282 if (svc
->port
!= FTPPORT
)
285 /* Note: persistent fwmark-based services and
286 * persistent port zero service are handled here.
288 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
289 * port zero template:
290 * <protocol,caddr,0,vaddr,0,daddr,0>
293 protocol
= IPPROTO_IP
;
297 /* return *ignored = -1 so NF_DROP can be used */
298 if (ip_vs_conn_fill_param_persist(svc
, skb
, protocol
, &snet
, 0,
299 vaddr
, vport
, ¶m
) < 0) {
305 /* Check if a template already exists */
306 ct
= ip_vs_ct_in_get(¶m
);
307 if (!ct
|| !ip_vs_check_template(ct
)) {
308 struct ip_vs_scheduler
*sched
;
311 * No template found or the dest of the connection
312 * template is not available.
313 * return *ignored=0 i.e. ICMP and NF_DROP
315 sched
= rcu_dereference(svc
->scheduler
);
316 dest
= sched
->schedule(svc
, skb
, iph
);
318 IP_VS_DBG(1, "p-schedule: no dest found.\n");
319 kfree(param
.pe_data
);
324 if (dst_port
== svc
->port
&& svc
->port
!= FTPPORT
)
328 * This adds param.pe_data to the template,
329 * and thus param.pe_data will be destroyed
330 * when the template expires */
331 ct
= ip_vs_conn_new(¶m
, &dest
->addr
, dport
,
332 IP_VS_CONN_F_TEMPLATE
, dest
, skb
->mark
);
334 kfree(param
.pe_data
);
339 ct
->timeout
= svc
->timeout
;
341 /* set destination with the found template */
343 kfree(param
.pe_data
);
347 if (dport
== svc
->port
&& dest
->port
)
350 flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
351 && iph
->protocol
== IPPROTO_UDP
) ?
352 IP_VS_CONN_F_ONE_PACKET
: 0;
355 * Create a new connection according to the template
357 ip_vs_conn_fill_param(svc
->net
, svc
->af
, iph
->protocol
, &iph
->saddr
,
358 src_port
, &iph
->daddr
, dst_port
, ¶m
);
360 cp
= ip_vs_conn_new(¶m
, &dest
->addr
, dport
, flags
, dest
, skb
->mark
);
370 ip_vs_control_add(cp
, ct
);
373 ip_vs_conn_stats(cp
, svc
);
379 * IPVS main scheduling function
380 * It selects a server according to the virtual service, and
381 * creates a connection entry.
382 * Protocols supported: TCP, UDP
386 * 1 : protocol tried to schedule (eg. on SYN), found svc but the
387 * svc/scheduler decides that this packet should be accepted with
388 * NF_ACCEPT because it must not be scheduled.
390 * 0 : scheduler can not find destination, so try bypass or
391 * return ICMP and then NF_DROP (ip_vs_leave).
393 * -1 : scheduler tried to schedule but fatal error occurred, eg.
394 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
395 * failure such as missing Call-ID, ENOMEM on skb_linearize
396 * or pe_data. In this case we should return NF_DROP without
397 * any attempts to send ICMP with ip_vs_leave.
400 ip_vs_schedule(struct ip_vs_service
*svc
, struct sk_buff
*skb
,
401 struct ip_vs_proto_data
*pd
, int *ignored
,
402 struct ip_vs_iphdr
*iph
)
404 struct ip_vs_protocol
*pp
= pd
->pp
;
405 struct ip_vs_conn
*cp
= NULL
;
406 struct ip_vs_scheduler
*sched
;
407 struct ip_vs_dest
*dest
;
408 __be16 _ports
[2], *pptr
;
413 * IPv6 frags, only the first hit here.
415 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
420 * FTPDATA needs this check when using local real server.
421 * Never schedule Active FTPDATA connections from real server.
422 * For LVS-NAT they must be already created. For other methods
423 * with persistence the connection is created on SYN+ACK.
425 if (pptr
[0] == FTPDATA
) {
426 IP_VS_DBG_PKT(12, svc
->af
, pp
, skb
, 0,
427 "Not scheduling FTPDATA");
432 * Do not schedule replies from local real server.
434 if ((!skb
->dev
|| skb
->dev
->flags
& IFF_LOOPBACK
) &&
435 (cp
= pp
->conn_in_get(svc
->af
, skb
, iph
, 1))) {
436 IP_VS_DBG_PKT(12, svc
->af
, pp
, skb
, 0,
437 "Not scheduling reply for existing connection");
438 __ip_vs_conn_put(cp
);
445 if (svc
->flags
& IP_VS_SVC_F_PERSISTENT
)
446 return ip_vs_sched_persist(svc
, skb
, pptr
[0], pptr
[1], ignored
,
452 * Non-persistent service
454 if (!svc
->fwmark
&& pptr
[1] != svc
->port
) {
456 pr_err("Schedule: port zero only supported "
457 "in persistent services, "
458 "check your ipvs configuration\n");
462 sched
= rcu_dereference(svc
->scheduler
);
463 dest
= sched
->schedule(svc
, skb
, iph
);
465 IP_VS_DBG(1, "Schedule: no dest found.\n");
469 flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
470 && iph
->protocol
== IPPROTO_UDP
) ?
471 IP_VS_CONN_F_ONE_PACKET
: 0;
474 * Create a connection entry.
477 struct ip_vs_conn_param p
;
479 ip_vs_conn_fill_param(svc
->net
, svc
->af
, iph
->protocol
,
480 &iph
->saddr
, pptr
[0], &iph
->daddr
,
482 cp
= ip_vs_conn_new(&p
, &dest
->addr
,
483 dest
->port
? dest
->port
: pptr
[1],
484 flags
, dest
, skb
->mark
);
491 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
492 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
494 IP_VS_DBG_ADDR(svc
->af
, &cp
->caddr
), ntohs(cp
->cport
),
495 IP_VS_DBG_ADDR(svc
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
496 IP_VS_DBG_ADDR(svc
->af
, &cp
->daddr
), ntohs(cp
->dport
),
497 cp
->flags
, atomic_read(&cp
->refcnt
));
499 ip_vs_conn_stats(cp
, svc
);
505 * Pass or drop the packet.
506 * Called by ip_vs_in, when the virtual service is available but
507 * no destination is available for a new connection.
509 int ip_vs_leave(struct ip_vs_service
*svc
, struct sk_buff
*skb
,
510 struct ip_vs_proto_data
*pd
, struct ip_vs_iphdr
*iph
)
512 __be16 _ports
[2], *pptr
;
515 struct netns_ipvs
*ipvs
;
519 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
527 #ifdef CONFIG_IP_VS_IPV6
528 if (svc
->af
== AF_INET6
)
529 unicast
= ipv6_addr_type(&iph
->daddr
.in6
) & IPV6_ADDR_UNICAST
;
532 unicast
= (inet_addr_type(net
, iph
->daddr
.ip
) == RTN_UNICAST
);
534 /* if it is fwmark-based service, the cache_bypass sysctl is up
535 and the destination is a non-local unicast, then create
536 a cache_bypass connection entry */
537 ipvs
= net_ipvs(net
);
538 if (ipvs
->sysctl_cache_bypass
&& svc
->fwmark
&& unicast
) {
540 struct ip_vs_conn
*cp
;
541 unsigned int flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
&&
542 iph
->protocol
== IPPROTO_UDP
) ?
543 IP_VS_CONN_F_ONE_PACKET
: 0;
544 union nf_inet_addr daddr
= { .all
= { 0, 0, 0, 0 } };
546 /* create a new connection entry */
547 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__
);
549 struct ip_vs_conn_param p
;
550 ip_vs_conn_fill_param(svc
->net
, svc
->af
, iph
->protocol
,
551 &iph
->saddr
, pptr
[0],
552 &iph
->daddr
, pptr
[1], &p
);
553 cp
= ip_vs_conn_new(&p
, &daddr
, 0,
554 IP_VS_CONN_F_BYPASS
| flags
,
561 ip_vs_in_stats(cp
, skb
);
564 ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pd
);
566 /* transmit the first SYN packet */
567 ret
= cp
->packet_xmit(skb
, cp
, pd
->pp
, iph
);
568 /* do not touch skb anymore */
570 atomic_inc(&cp
->in_pkts
);
577 * When the virtual ftp service is presented, packets destined
578 * for other services on the VIP may get here (except services
579 * listed in the ipvs table), pass the packets, because it is
580 * not ipvs job to decide to drop the packets.
582 if ((svc
->port
== FTPPORT
) && (pptr
[1] != FTPPORT
))
586 * Notify the client that the destination is unreachable, and
587 * release the socket buffer.
588 * Since it is in IP layer, the TCP socket is not actually
589 * created, the TCP RST packet cannot be sent, instead that
590 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
592 #ifdef CONFIG_IP_VS_IPV6
593 if (svc
->af
== AF_INET6
) {
595 struct net
*net_
= dev_net(skb_dst(skb
)->dev
);
597 skb
->dev
= net_
->loopback_dev
;
599 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
602 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
609 static int sysctl_snat_reroute(struct sk_buff
*skb
)
611 struct netns_ipvs
*ipvs
= net_ipvs(skb_net(skb
));
612 return ipvs
->sysctl_snat_reroute
;
615 static int sysctl_nat_icmp_send(struct net
*net
)
617 struct netns_ipvs
*ipvs
= net_ipvs(net
);
618 return ipvs
->sysctl_nat_icmp_send
;
621 static int sysctl_expire_nodest_conn(struct netns_ipvs
*ipvs
)
623 return ipvs
->sysctl_expire_nodest_conn
;
628 static int sysctl_snat_reroute(struct sk_buff
*skb
) { return 0; }
629 static int sysctl_nat_icmp_send(struct net
*net
) { return 0; }
630 static int sysctl_expire_nodest_conn(struct netns_ipvs
*ipvs
) { return 0; }
634 __sum16
ip_vs_checksum_complete(struct sk_buff
*skb
, int offset
)
636 return csum_fold(skb_checksum(skb
, offset
, skb
->len
- offset
, 0));
639 static inline enum ip_defrag_users
ip_vs_defrag_user(unsigned int hooknum
)
641 if (NF_INET_LOCAL_IN
== hooknum
)
642 return IP_DEFRAG_VS_IN
;
643 if (NF_INET_FORWARD
== hooknum
)
644 return IP_DEFRAG_VS_FWD
;
645 return IP_DEFRAG_VS_OUT
;
648 static inline int ip_vs_gather_frags(struct sk_buff
*skb
, u_int32_t user
)
653 err
= ip_defrag(skb
, user
);
656 ip_send_check(ip_hdr(skb
));
661 static int ip_vs_route_me_harder(int af
, struct sk_buff
*skb
)
663 #ifdef CONFIG_IP_VS_IPV6
664 if (af
== AF_INET6
) {
665 if (sysctl_snat_reroute(skb
) && ip6_route_me_harder(skb
) != 0)
669 if ((sysctl_snat_reroute(skb
) ||
670 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
)
795 unsigned int verdict
= NF_DROP
;
797 if (IP_VS_FWD_METHOD(cp
) != 0) {
798 pr_err("shouldn't reach here, because the box is on the "
799 "half connection in the tun/dr module.\n");
802 /* Ensure the checksum is correct */
803 if (!skb_csum_unnecessary(skb
) && ip_vs_checksum_complete(skb
, ihl
)) {
804 /* Failed checksum! */
805 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
806 IP_VS_DBG_ADDR(af
, snet
));
810 if (IPPROTO_TCP
== protocol
|| IPPROTO_UDP
== protocol
||
811 IPPROTO_SCTP
== protocol
)
812 offset
+= 2 * sizeof(__u16
);
813 if (!skb_make_writable(skb
, offset
))
816 #ifdef CONFIG_IP_VS_IPV6
818 ip_vs_nat_icmp_v6(skb
, pp
, cp
, 1);
821 ip_vs_nat_icmp(skb
, pp
, cp
, 1);
823 if (ip_vs_route_me_harder(af
, skb
))
826 /* do the statistics and put it back */
827 ip_vs_out_stats(cp
, skb
);
829 skb
->ipvs_property
= 1;
830 if (!(cp
->flags
& IP_VS_CONN_F_NFCT
))
833 ip_vs_update_conntrack(skb
, cp
, 0);
837 __ip_vs_conn_put(cp
);
843 * Handle ICMP messages in the inside-to-outside direction (outgoing).
844 * Find any that might be relevant, check against existing connections.
845 * Currently handles error types - unreachable, quench, ttl exceeded.
847 static int ip_vs_out_icmp(struct sk_buff
*skb
, int *related
,
848 unsigned int hooknum
)
851 struct icmphdr _icmph
, *ic
;
852 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
853 struct ip_vs_iphdr ciph
;
854 struct ip_vs_conn
*cp
;
855 struct ip_vs_protocol
*pp
;
856 unsigned int offset
, ihl
;
857 union nf_inet_addr snet
;
861 /* reassemble IP fragments */
862 if (ip_is_fragment(ip_hdr(skb
))) {
863 if (ip_vs_gather_frags(skb
, ip_vs_defrag_user(hooknum
)))
868 offset
= ihl
= iph
->ihl
* 4;
869 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
873 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
874 ic
->type
, ntohs(icmp_id(ic
)),
875 &iph
->saddr
, &iph
->daddr
);
878 * Work through seeing if this is for us.
879 * These checks are supposed to be in an order that means easy
880 * things are checked first to speed up processing.... however
881 * this means that some packets will manage to get a long way
882 * down this stack and then be rejected, but that's life.
884 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
885 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
886 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
891 /* Now find the contained IP header */
892 offset
+= sizeof(_icmph
);
893 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
895 return NF_ACCEPT
; /* The packet looks wrong, ignore */
897 pp
= ip_vs_proto_get(cih
->protocol
);
901 /* Is the embedded protocol header present? */
902 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
906 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, offset
,
907 "Checking outgoing ICMP for");
909 ip_vs_fill_ip4hdr(cih
, &ciph
);
911 /* The embedded headers contain source and dest in reverse order */
912 cp
= pp
->conn_out_get(AF_INET
, skb
, &ciph
, 1);
916 snet
.ip
= iph
->saddr
;
917 return handle_response_icmp(AF_INET
, skb
, &snet
, cih
->protocol
, cp
,
921 #ifdef CONFIG_IP_VS_IPV6
922 static int ip_vs_out_icmp_v6(struct sk_buff
*skb
, int *related
,
923 unsigned int hooknum
, struct ip_vs_iphdr
*ipvsh
)
925 struct icmp6hdr _icmph
, *ic
;
926 struct ipv6hdr _ip6h
, *ip6h
; /* The ip header contained within ICMP */
927 struct ip_vs_iphdr ciph
= {.flags
= 0, .fragoffs
= 0};/*Contained IP */
928 struct ip_vs_conn
*cp
;
929 struct ip_vs_protocol
*pp
;
930 union nf_inet_addr snet
;
931 unsigned int writable
;
934 ic
= frag_safe_skb_hp(skb
, ipvsh
->len
, sizeof(_icmph
), &_icmph
, ipvsh
);
939 * Work through seeing if this is for us.
940 * These checks are supposed to be in an order that means easy
941 * things are checked first to speed up processing.... however
942 * this means that some packets will manage to get a long way
943 * down this stack and then be rejected, but that's life.
945 if (ic
->icmp6_type
& ICMPV6_INFOMSG_MASK
) {
949 /* Fragment header that is before ICMP header tells us that:
950 * it's not an error message since they can't be fragmented.
952 if (ipvsh
->flags
& IP6_FH_F_FRAG
)
955 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
956 ic
->icmp6_type
, ntohs(icmpv6_id(ic
)),
957 &ipvsh
->saddr
, &ipvsh
->daddr
);
959 /* Now find the contained IP header */
960 ciph
.len
= ipvsh
->len
+ sizeof(_icmph
);
961 ip6h
= skb_header_pointer(skb
, ciph
.len
, sizeof(_ip6h
), &_ip6h
);
963 return NF_ACCEPT
; /* The packet looks wrong, ignore */
964 ciph
.saddr
.in6
= ip6h
->saddr
; /* conn_out_get() handles reverse order */
965 ciph
.daddr
.in6
= ip6h
->daddr
;
966 /* skip possible IPv6 exthdrs of contained IPv6 packet */
967 ciph
.protocol
= ipv6_find_hdr(skb
, &ciph
.len
, -1, &ciph
.fragoffs
, NULL
);
968 if (ciph
.protocol
< 0)
969 return NF_ACCEPT
; /* Contained IPv6 hdr looks wrong, ignore */
971 pp
= ip_vs_proto_get(ciph
.protocol
);
975 /* The embedded headers contain source and dest in reverse order */
976 cp
= pp
->conn_out_get(AF_INET6
, skb
, &ciph
, 1);
980 snet
.in6
= ciph
.saddr
.in6
;
982 return handle_response_icmp(AF_INET6
, skb
, &snet
, ciph
.protocol
, cp
,
983 pp
, writable
, sizeof(struct ipv6hdr
));
988 * Check if sctp chunc is ABORT chunk
990 static inline int is_sctp_abort(const struct sk_buff
*skb
, int nh_len
)
992 sctp_chunkhdr_t
*sch
, schunk
;
993 sch
= skb_header_pointer(skb
, nh_len
+ sizeof(sctp_sctphdr_t
),
994 sizeof(schunk
), &schunk
);
997 if (sch
->type
== SCTP_CID_ABORT
)
1002 static inline int is_tcp_reset(const struct sk_buff
*skb
, int nh_len
)
1004 struct tcphdr _tcph
, *th
;
1006 th
= skb_header_pointer(skb
, nh_len
, sizeof(_tcph
), &_tcph
);
1012 static inline bool is_new_conn(const struct sk_buff
*skb
,
1013 struct ip_vs_iphdr
*iph
)
1015 switch (iph
->protocol
) {
1017 struct tcphdr _tcph
, *th
;
1019 th
= skb_header_pointer(skb
, iph
->len
, sizeof(_tcph
), &_tcph
);
1024 case IPPROTO_SCTP
: {
1025 sctp_chunkhdr_t
*sch
, schunk
;
1027 sch
= skb_header_pointer(skb
, iph
->len
+ sizeof(sctp_sctphdr_t
),
1028 sizeof(schunk
), &schunk
);
1031 return sch
->type
== SCTP_CID_INIT
;
1038 /* Handle response packets: rewrite addresses and send away...
1041 handle_response(int af
, struct sk_buff
*skb
, struct ip_vs_proto_data
*pd
,
1042 struct ip_vs_conn
*cp
, struct ip_vs_iphdr
*iph
)
1044 struct ip_vs_protocol
*pp
= pd
->pp
;
1046 IP_VS_DBG_PKT(11, af
, pp
, skb
, 0, "Outgoing packet");
1048 if (!skb_make_writable(skb
, iph
->len
))
1051 /* mangle the packet */
1052 if (pp
->snat_handler
&& !pp
->snat_handler(skb
, pp
, cp
, iph
))
1055 #ifdef CONFIG_IP_VS_IPV6
1057 ipv6_hdr(skb
)->saddr
= cp
->vaddr
.in6
;
1061 ip_hdr(skb
)->saddr
= cp
->vaddr
.ip
;
1062 ip_send_check(ip_hdr(skb
));
1066 * nf_iterate does not expect change in the skb->dst->dev.
1067 * It looks like it is not fatal to enable this code for hooks
1068 * where our handlers are at the end of the chain list and
1069 * when all next handlers use skb->dst->dev and not outdev.
1070 * It will definitely route properly the inout NAT traffic
1071 * when multiple paths are used.
1074 /* For policy routing, packets originating from this
1075 * machine itself may be routed differently to packets
1076 * passing through. We want this packet to be routed as
1077 * if it came from this machine itself. So re-compute
1078 * the routing information.
1080 if (ip_vs_route_me_harder(af
, skb
))
1083 IP_VS_DBG_PKT(10, af
, pp
, skb
, 0, "After SNAT");
1085 ip_vs_out_stats(cp
, skb
);
1086 ip_vs_set_state(cp
, IP_VS_DIR_OUTPUT
, skb
, pd
);
1087 skb
->ipvs_property
= 1;
1088 if (!(cp
->flags
& IP_VS_CONN_F_NFCT
))
1091 ip_vs_update_conntrack(skb
, cp
, 0);
1105 * Check if outgoing packet belongs to the established ip_vs_conn.
1108 ip_vs_out(unsigned int hooknum
, struct sk_buff
*skb
, int af
)
1110 struct net
*net
= NULL
;
1111 struct ip_vs_iphdr iph
;
1112 struct ip_vs_protocol
*pp
;
1113 struct ip_vs_proto_data
*pd
;
1114 struct ip_vs_conn
*cp
;
1118 /* Already marked as IPVS request or reply? */
1119 if (skb
->ipvs_property
)
1122 /* Bad... Do not break raw sockets */
1123 if (unlikely(skb
->sk
!= NULL
&& hooknum
== NF_INET_LOCAL_OUT
&&
1125 struct sock
*sk
= skb
->sk
;
1126 struct inet_sock
*inet
= inet_sk(skb
->sk
);
1128 if (inet
&& sk
->sk_family
== PF_INET
&& inet
->nodefrag
)
1132 if (unlikely(!skb_dst(skb
)))
1136 if (!net_ipvs(net
)->enable
)
1139 ip_vs_fill_iph_skb(af
, skb
, &iph
);
1140 #ifdef CONFIG_IP_VS_IPV6
1141 if (af
== AF_INET6
) {
1142 if (unlikely(iph
.protocol
== IPPROTO_ICMPV6
)) {
1144 int verdict
= ip_vs_out_icmp_v6(skb
, &related
,
1152 if (unlikely(iph
.protocol
== IPPROTO_ICMP
)) {
1154 int verdict
= ip_vs_out_icmp(skb
, &related
, hooknum
);
1160 pd
= ip_vs_proto_data_get(net
, iph
.protocol
);
1165 /* reassemble IP fragments */
1166 #ifdef CONFIG_IP_VS_IPV6
1169 if (unlikely(ip_is_fragment(ip_hdr(skb
)) && !pp
->dont_defrag
)) {
1170 if (ip_vs_gather_frags(skb
,
1171 ip_vs_defrag_user(hooknum
)))
1174 ip_vs_fill_ip4hdr(skb_network_header(skb
), &iph
);
1178 * Check if the packet belongs to an existing entry
1180 cp
= pp
->conn_out_get(af
, skb
, &iph
, 0);
1183 return handle_response(af
, skb
, pd
, cp
, &iph
);
1184 if (sysctl_nat_icmp_send(net
) &&
1185 (pp
->protocol
== IPPROTO_TCP
||
1186 pp
->protocol
== IPPROTO_UDP
||
1187 pp
->protocol
== IPPROTO_SCTP
)) {
1188 __be16 _ports
[2], *pptr
;
1190 pptr
= frag_safe_skb_hp(skb
, iph
.len
,
1191 sizeof(_ports
), _ports
, &iph
);
1193 return NF_ACCEPT
; /* Not for me */
1194 if (ip_vs_has_real_service(net
, af
, iph
.protocol
, &iph
.saddr
,
1197 * Notify the real server: there is no
1198 * existing entry if it is not RST
1199 * packet or not TCP packet.
1201 if ((iph
.protocol
!= IPPROTO_TCP
&&
1202 iph
.protocol
!= IPPROTO_SCTP
)
1203 || ((iph
.protocol
== IPPROTO_TCP
1204 && !is_tcp_reset(skb
, iph
.len
))
1205 || (iph
.protocol
== IPPROTO_SCTP
1206 && !is_sctp_abort(skb
,
1208 #ifdef CONFIG_IP_VS_IPV6
1209 if (af
== AF_INET6
) {
1211 skb
->dev
= net
->loopback_dev
;
1213 ICMPV6_DEST_UNREACH
,
1214 ICMPV6_PORT_UNREACH
,
1220 ICMP_PORT_UNREACH
, 0);
1225 IP_VS_DBG_PKT(12, af
, pp
, skb
, 0,
1226 "ip_vs_out: packet continues traversal as normal");
1231 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1232 * used only for VS/NAT.
1233 * Check if packet is reply for established ip_vs_conn.
1236 ip_vs_reply4(const struct nf_hook_ops
*ops
, struct sk_buff
*skb
,
1237 const struct net_device
*in
, const struct net_device
*out
,
1238 int (*okfn
)(struct sk_buff
*))
1240 return ip_vs_out(ops
->hooknum
, skb
, AF_INET
);
1244 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1245 * Check if packet is reply for established ip_vs_conn.
1248 ip_vs_local_reply4(const struct nf_hook_ops
*ops
, struct sk_buff
*skb
,
1249 const struct net_device
*in
, const struct net_device
*out
,
1250 int (*okfn
)(struct sk_buff
*))
1252 return ip_vs_out(ops
->hooknum
, skb
, AF_INET
);
1255 #ifdef CONFIG_IP_VS_IPV6
1258 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1259 * used only for VS/NAT.
1260 * Check if packet is reply for established ip_vs_conn.
1263 ip_vs_reply6(const struct nf_hook_ops
*ops
, struct sk_buff
*skb
,
1264 const struct net_device
*in
, const struct net_device
*out
,
1265 int (*okfn
)(struct sk_buff
*))
1267 return ip_vs_out(ops
->hooknum
, skb
, AF_INET6
);
1271 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1272 * Check if packet is reply for established ip_vs_conn.
1275 ip_vs_local_reply6(const struct nf_hook_ops
*ops
, struct sk_buff
*skb
,
1276 const struct net_device
*in
, const struct net_device
*out
,
1277 int (*okfn
)(struct sk_buff
*))
1279 return ip_vs_out(ops
->hooknum
, skb
, AF_INET6
);
1285 * Handle ICMP messages in the outside-to-inside direction (incoming).
1286 * Find any that might be relevant, check against existing connections,
1287 * forward to the right destination host if relevant.
1288 * Currently handles error types - unreachable, quench, ttl exceeded.
1291 ip_vs_in_icmp(struct sk_buff
*skb
, int *related
, unsigned int hooknum
)
1293 struct net
*net
= NULL
;
1295 struct icmphdr _icmph
, *ic
;
1296 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
1297 struct ip_vs_iphdr ciph
;
1298 struct ip_vs_conn
*cp
;
1299 struct ip_vs_protocol
*pp
;
1300 struct ip_vs_proto_data
*pd
;
1301 unsigned int offset
, offset2
, ihl
, verdict
;
1306 /* reassemble IP fragments */
1307 if (ip_is_fragment(ip_hdr(skb
))) {
1308 if (ip_vs_gather_frags(skb
, ip_vs_defrag_user(hooknum
)))
1313 offset
= ihl
= iph
->ihl
* 4;
1314 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
1318 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1319 ic
->type
, ntohs(icmp_id(ic
)),
1320 &iph
->saddr
, &iph
->daddr
);
1323 * Work through seeing if this is for us.
1324 * These checks are supposed to be in an order that means easy
1325 * things are checked first to speed up processing.... however
1326 * this means that some packets will manage to get a long way
1327 * down this stack and then be rejected, but that's life.
1329 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
1330 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
1331 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
1336 /* Now find the contained IP header */
1337 offset
+= sizeof(_icmph
);
1338 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
1340 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1344 /* Special case for errors for IPIP packets */
1346 if (cih
->protocol
== IPPROTO_IPIP
) {
1347 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
)))
1349 /* Error for our IPIP must arrive at LOCAL_IN */
1350 if (!(skb_rtable(skb
)->rt_flags
& RTCF_LOCAL
))
1352 offset
+= cih
->ihl
* 4;
1353 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
1355 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1359 pd
= ip_vs_proto_data_get(net
, cih
->protocol
);
1364 /* Is the embedded protocol header present? */
1365 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
1369 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, offset
,
1370 "Checking incoming ICMP for");
1373 ip_vs_fill_ip4hdr(cih
, &ciph
);
1376 /* The embedded headers contain source and dest in reverse order.
1377 * For IPIP this is error for request, not for reply.
1379 cp
= pp
->conn_in_get(AF_INET
, skb
, &ciph
, ipip
? 0 : 1);
1385 /* Ensure the checksum is correct */
1386 if (!skb_csum_unnecessary(skb
) && ip_vs_checksum_complete(skb
, ihl
)) {
1387 /* Failed checksum! */
1388 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1394 __be32 info
= ic
->un
.gateway
;
1395 __u8 type
= ic
->type
;
1396 __u8 code
= ic
->code
;
1398 /* Update the MTU */
1399 if (ic
->type
== ICMP_DEST_UNREACH
&&
1400 ic
->code
== ICMP_FRAG_NEEDED
) {
1401 struct ip_vs_dest
*dest
= cp
->dest
;
1402 u32 mtu
= ntohs(ic
->un
.frag
.mtu
);
1403 __be16 frag_off
= cih
->frag_off
;
1405 /* Strip outer IP and ICMP, go to IPIP header */
1406 if (pskb_pull(skb
, ihl
+ sizeof(_icmph
)) == NULL
)
1408 offset2
-= ihl
+ sizeof(_icmph
);
1409 skb_reset_network_header(skb
);
1410 IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1411 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
, mtu
);
1412 ipv4_update_pmtu(skb
, dev_net(skb
->dev
),
1414 /* Client uses PMTUD? */
1415 if (!(frag_off
& htons(IP_DF
)))
1417 /* Prefer the resulting PMTU */
1419 struct ip_vs_dest_dst
*dest_dst
;
1422 dest_dst
= rcu_dereference(dest
->dest_dst
);
1424 mtu
= dst_mtu(dest_dst
->dst_cache
);
1427 if (mtu
> 68 + sizeof(struct iphdr
))
1428 mtu
-= sizeof(struct iphdr
);
1431 /* Strip outer IP, ICMP and IPIP, go to IP header of
1434 if (pskb_pull(skb
, offset2
) == NULL
)
1436 skb_reset_network_header(skb
);
1437 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1438 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1439 type
, code
, ntohl(info
));
1440 icmp_send(skb
, type
, code
, info
);
1441 /* ICMP can be shorter but anyways, account it */
1442 ip_vs_out_stats(cp
, skb
);
1446 verdict
= NF_STOLEN
;
1450 /* do the statistics and put it back */
1451 ip_vs_in_stats(cp
, skb
);
1452 if (IPPROTO_TCP
== cih
->protocol
|| IPPROTO_UDP
== cih
->protocol
||
1453 IPPROTO_SCTP
== cih
->protocol
)
1454 offset
+= 2 * sizeof(__u16
);
1455 verdict
= ip_vs_icmp_xmit(skb
, cp
, pp
, offset
, hooknum
, &ciph
);
1458 __ip_vs_conn_put(cp
);
1463 #ifdef CONFIG_IP_VS_IPV6
1464 static int ip_vs_in_icmp_v6(struct sk_buff
*skb
, int *related
,
1465 unsigned int hooknum
, struct ip_vs_iphdr
*iph
)
1467 struct net
*net
= NULL
;
1468 struct ipv6hdr _ip6h
, *ip6h
;
1469 struct icmp6hdr _icmph
, *ic
;
1470 struct ip_vs_iphdr ciph
= {.flags
= 0, .fragoffs
= 0};/*Contained IP */
1471 struct ip_vs_conn
*cp
;
1472 struct ip_vs_protocol
*pp
;
1473 struct ip_vs_proto_data
*pd
;
1474 unsigned int offs_ciph
, writable
, verdict
;
1478 ic
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_icmph
), &_icmph
, iph
);
1483 * Work through seeing if this is for us.
1484 * These checks are supposed to be in an order that means easy
1485 * things are checked first to speed up processing.... however
1486 * this means that some packets will manage to get a long way
1487 * down this stack and then be rejected, but that's life.
1489 if (ic
->icmp6_type
& ICMPV6_INFOMSG_MASK
) {
1493 /* Fragment header that is before ICMP header tells us that:
1494 * it's not an error message since they can't be fragmented.
1496 if (iph
->flags
& IP6_FH_F_FRAG
)
1499 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1500 ic
->icmp6_type
, ntohs(icmpv6_id(ic
)),
1501 &iph
->saddr
, &iph
->daddr
);
1503 /* Now find the contained IP header */
1504 ciph
.len
= iph
->len
+ sizeof(_icmph
);
1505 offs_ciph
= ciph
.len
; /* Save ip header offset */
1506 ip6h
= skb_header_pointer(skb
, ciph
.len
, sizeof(_ip6h
), &_ip6h
);
1508 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1509 ciph
.saddr
.in6
= ip6h
->saddr
; /* conn_in_get() handles reverse order */
1510 ciph
.daddr
.in6
= ip6h
->daddr
;
1511 /* skip possible IPv6 exthdrs of contained IPv6 packet */
1512 ciph
.protocol
= ipv6_find_hdr(skb
, &ciph
.len
, -1, &ciph
.fragoffs
, NULL
);
1513 if (ciph
.protocol
< 0)
1514 return NF_ACCEPT
; /* Contained IPv6 hdr looks wrong, ignore */
1517 pd
= ip_vs_proto_data_get(net
, ciph
.protocol
);
1522 /* Cannot handle fragmented embedded protocol */
1526 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
, offs_ciph
,
1527 "Checking incoming ICMPv6 for");
1529 /* The embedded headers contain source and dest in reverse order
1530 * if not from localhost
1532 cp
= pp
->conn_in_get(AF_INET6
, skb
, &ciph
,
1533 (hooknum
== NF_INET_LOCAL_OUT
) ? 0 : 1);
1537 /* VS/TUN, VS/DR and LOCALNODE just let it go */
1538 if ((hooknum
== NF_INET_LOCAL_OUT
) &&
1539 (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
)) {
1540 __ip_vs_conn_put(cp
);
1544 /* do the statistics and put it back */
1545 ip_vs_in_stats(cp
, skb
);
1547 /* Need to mangle contained IPv6 header in ICMPv6 packet */
1548 writable
= ciph
.len
;
1549 if (IPPROTO_TCP
== ciph
.protocol
|| IPPROTO_UDP
== ciph
.protocol
||
1550 IPPROTO_SCTP
== ciph
.protocol
)
1551 writable
+= 2 * sizeof(__u16
); /* Also mangle ports */
1553 verdict
= ip_vs_icmp_xmit_v6(skb
, cp
, pp
, writable
, hooknum
, &ciph
);
1555 __ip_vs_conn_put(cp
);
1563 * Check if it's for virtual services, look it up,
1564 * and send it on its way...
1567 ip_vs_in(unsigned int hooknum
, struct sk_buff
*skb
, int af
)
1570 struct ip_vs_iphdr iph
;
1571 struct ip_vs_protocol
*pp
;
1572 struct ip_vs_proto_data
*pd
;
1573 struct ip_vs_conn
*cp
;
1575 struct netns_ipvs
*ipvs
;
1577 /* Already marked as IPVS request or reply? */
1578 if (skb
->ipvs_property
)
1583 * - remote client: only PACKET_HOST
1584 * - route: used for struct net when skb->dev is unset
1586 if (unlikely((skb
->pkt_type
!= PACKET_HOST
&&
1587 hooknum
!= NF_INET_LOCAL_OUT
) ||
1589 ip_vs_fill_iph_skb(af
, skb
, &iph
);
1590 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1591 " ignored in hook %u\n",
1592 skb
->pkt_type
, iph
.protocol
,
1593 IP_VS_DBG_ADDR(af
, &iph
.daddr
), hooknum
);
1596 /* ipvs enabled in this netns ? */
1598 ipvs
= net_ipvs(net
);
1599 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1602 ip_vs_fill_iph_skb(af
, skb
, &iph
);
1604 /* Bad... Do not break raw sockets */
1605 if (unlikely(skb
->sk
!= NULL
&& hooknum
== NF_INET_LOCAL_OUT
&&
1607 struct sock
*sk
= skb
->sk
;
1608 struct inet_sock
*inet
= inet_sk(skb
->sk
);
1610 if (inet
&& sk
->sk_family
== PF_INET
&& inet
->nodefrag
)
1614 #ifdef CONFIG_IP_VS_IPV6
1615 if (af
== AF_INET6
) {
1616 if (unlikely(iph
.protocol
== IPPROTO_ICMPV6
)) {
1618 int verdict
= ip_vs_in_icmp_v6(skb
, &related
, hooknum
,
1626 if (unlikely(iph
.protocol
== IPPROTO_ICMP
)) {
1628 int verdict
= ip_vs_in_icmp(skb
, &related
, hooknum
);
1634 /* Protocol supported? */
1635 pd
= ip_vs_proto_data_get(net
, iph
.protocol
);
1640 * Check if the packet belongs to an existing connection entry
1642 cp
= pp
->conn_in_get(af
, skb
, &iph
, 0);
1644 if (unlikely(sysctl_expire_nodest_conn(ipvs
)) && cp
&& cp
->dest
&&
1645 unlikely(!atomic_read(&cp
->dest
->weight
)) && !iph
.fragoffs
&&
1646 is_new_conn(skb
, &iph
)) {
1647 ip_vs_conn_expire_now(cp
);
1648 __ip_vs_conn_put(cp
);
1652 if (unlikely(!cp
) && !iph
.fragoffs
) {
1653 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1654 * replayed fragment zero will already have created the cp
1658 /* Schedule and create new connection entry into &cp */
1659 if (!pp
->conn_schedule(af
, skb
, pd
, &v
, &cp
, &iph
))
1663 if (unlikely(!cp
)) {
1664 /* sorry, all this trouble for a no-hit :) */
1665 IP_VS_DBG_PKT(12, af
, pp
, skb
, 0,
1666 "ip_vs_in: packet continues traversal as normal");
1668 /* Fragment that couldn't be mapped to a conn entry
1669 * is missing module nf_defrag_ipv6
1671 IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1672 IP_VS_DBG_PKT(7, af
, pp
, skb
, 0, "unhandled fragment");
1677 IP_VS_DBG_PKT(11, af
, pp
, skb
, 0, "Incoming packet");
1678 /* Check the server status */
1679 if (cp
->dest
&& !(cp
->dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
1680 /* the destination server is not available */
1682 if (sysctl_expire_nodest_conn(ipvs
)) {
1683 /* try to expire the connection immediately */
1684 ip_vs_conn_expire_now(cp
);
1686 /* don't restart its timer, and silently
1688 __ip_vs_conn_put(cp
);
1692 ip_vs_in_stats(cp
, skb
);
1693 ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pd
);
1694 if (cp
->packet_xmit
)
1695 ret
= cp
->packet_xmit(skb
, cp
, pp
, &iph
);
1696 /* do not touch skb anymore */
1698 IP_VS_DBG_RL("warning: packet_xmit is null");
1702 /* Increase its packet counter and check if it is needed
1703 * to be synchronized
1705 * Sync connection if it is about to close to
1706 * encorage the standby servers to update the connections timeout
1708 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1711 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
1712 pkts
= sysctl_sync_threshold(ipvs
);
1714 pkts
= atomic_add_return(1, &cp
->in_pkts
);
1716 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
)
1717 ip_vs_sync_conn(net
, cp
, pkts
);
1724 * AF_INET handler in NF_INET_LOCAL_IN chain
1725 * Schedule and forward packets from remote clients
1728 ip_vs_remote_request4(const struct nf_hook_ops
*ops
, struct sk_buff
*skb
,
1729 const struct net_device
*in
,
1730 const struct net_device
*out
,
1731 int (*okfn
)(struct sk_buff
*))
1733 return ip_vs_in(ops
->hooknum
, skb
, AF_INET
);
1737 * AF_INET handler in NF_INET_LOCAL_OUT chain
1738 * Schedule and forward packets from local clients
1741 ip_vs_local_request4(const struct nf_hook_ops
*ops
, struct sk_buff
*skb
,
1742 const struct net_device
*in
, const struct net_device
*out
,
1743 int (*okfn
)(struct sk_buff
*))
1745 return ip_vs_in(ops
->hooknum
, skb
, AF_INET
);
1748 #ifdef CONFIG_IP_VS_IPV6
1751 * AF_INET6 handler in NF_INET_LOCAL_IN chain
1752 * Schedule and forward packets from remote clients
1755 ip_vs_remote_request6(const struct nf_hook_ops
*ops
, struct sk_buff
*skb
,
1756 const struct net_device
*in
,
1757 const struct net_device
*out
,
1758 int (*okfn
)(struct sk_buff
*))
1760 return ip_vs_in(ops
->hooknum
, skb
, AF_INET6
);
1764 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
1765 * Schedule and forward packets from local clients
1768 ip_vs_local_request6(const struct nf_hook_ops
*ops
, struct sk_buff
*skb
,
1769 const struct net_device
*in
, const struct net_device
*out
,
1770 int (*okfn
)(struct sk_buff
*))
1772 return ip_vs_in(ops
->hooknum
, skb
, AF_INET6
);
1779 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1780 * related packets destined for 0.0.0.0/0.
1781 * When fwmark-based virtual service is used, such as transparent
1782 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1783 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1784 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1785 * and send them to ip_vs_in_icmp.
1788 ip_vs_forward_icmp(const struct nf_hook_ops
*ops
, struct sk_buff
*skb
,
1789 const struct net_device
*in
, const struct net_device
*out
,
1790 int (*okfn
)(struct sk_buff
*))
1794 struct netns_ipvs
*ipvs
;
1796 if (ip_hdr(skb
)->protocol
!= IPPROTO_ICMP
)
1799 /* ipvs enabled in this netns ? */
1801 ipvs
= net_ipvs(net
);
1802 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1805 return ip_vs_in_icmp(skb
, &r
, ops
->hooknum
);
1808 #ifdef CONFIG_IP_VS_IPV6
1810 ip_vs_forward_icmp_v6(const struct nf_hook_ops
*ops
, struct sk_buff
*skb
,
1811 const struct net_device
*in
, const struct net_device
*out
,
1812 int (*okfn
)(struct sk_buff
*))
1816 struct netns_ipvs
*ipvs
;
1817 struct ip_vs_iphdr iphdr
;
1819 ip_vs_fill_iph_skb(AF_INET6
, skb
, &iphdr
);
1820 if (iphdr
.protocol
!= IPPROTO_ICMPV6
)
1823 /* ipvs enabled in this netns ? */
1825 ipvs
= net_ipvs(net
);
1826 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1829 return ip_vs_in_icmp_v6(skb
, &r
, ops
->hooknum
, &iphdr
);
1834 static struct nf_hook_ops ip_vs_ops
[] __read_mostly
= {
1835 /* After packet filtering, change source only for VS/NAT */
1837 .hook
= ip_vs_reply4
,
1838 .owner
= THIS_MODULE
,
1840 .hooknum
= NF_INET_LOCAL_IN
,
1841 .priority
= NF_IP_PRI_NAT_SRC
- 2,
1843 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1844 * or VS/NAT(change destination), so that filtering rules can be
1845 * applied to IPVS. */
1847 .hook
= ip_vs_remote_request4
,
1848 .owner
= THIS_MODULE
,
1850 .hooknum
= NF_INET_LOCAL_IN
,
1851 .priority
= NF_IP_PRI_NAT_SRC
- 1,
1853 /* Before ip_vs_in, change source only for VS/NAT */
1855 .hook
= ip_vs_local_reply4
,
1856 .owner
= THIS_MODULE
,
1858 .hooknum
= NF_INET_LOCAL_OUT
,
1859 .priority
= NF_IP_PRI_NAT_DST
+ 1,
1861 /* After mangle, schedule and forward local requests */
1863 .hook
= ip_vs_local_request4
,
1864 .owner
= THIS_MODULE
,
1866 .hooknum
= NF_INET_LOCAL_OUT
,
1867 .priority
= NF_IP_PRI_NAT_DST
+ 2,
1869 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1870 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1872 .hook
= ip_vs_forward_icmp
,
1873 .owner
= THIS_MODULE
,
1875 .hooknum
= NF_INET_FORWARD
,
1878 /* After packet filtering, change source only for VS/NAT */
1880 .hook
= ip_vs_reply4
,
1881 .owner
= THIS_MODULE
,
1883 .hooknum
= NF_INET_FORWARD
,
1886 #ifdef CONFIG_IP_VS_IPV6
1887 /* After packet filtering, change source only for VS/NAT */
1889 .hook
= ip_vs_reply6
,
1890 .owner
= THIS_MODULE
,
1892 .hooknum
= NF_INET_LOCAL_IN
,
1893 .priority
= NF_IP6_PRI_NAT_SRC
- 2,
1895 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1896 * or VS/NAT(change destination), so that filtering rules can be
1897 * applied to IPVS. */
1899 .hook
= ip_vs_remote_request6
,
1900 .owner
= THIS_MODULE
,
1902 .hooknum
= NF_INET_LOCAL_IN
,
1903 .priority
= NF_IP6_PRI_NAT_SRC
- 1,
1905 /* Before ip_vs_in, change source only for VS/NAT */
1907 .hook
= ip_vs_local_reply6
,
1908 .owner
= THIS_MODULE
,
1910 .hooknum
= NF_INET_LOCAL_OUT
,
1911 .priority
= NF_IP6_PRI_NAT_DST
+ 1,
1913 /* After mangle, schedule and forward local requests */
1915 .hook
= ip_vs_local_request6
,
1916 .owner
= THIS_MODULE
,
1918 .hooknum
= NF_INET_LOCAL_OUT
,
1919 .priority
= NF_IP6_PRI_NAT_DST
+ 2,
1921 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1922 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1924 .hook
= ip_vs_forward_icmp_v6
,
1925 .owner
= THIS_MODULE
,
1927 .hooknum
= NF_INET_FORWARD
,
1930 /* After packet filtering, change source only for VS/NAT */
1932 .hook
= ip_vs_reply6
,
1933 .owner
= THIS_MODULE
,
1935 .hooknum
= NF_INET_FORWARD
,
1941 * Initialize IP Virtual Server netns mem.
1943 static int __net_init
__ip_vs_init(struct net
*net
)
1945 struct netns_ipvs
*ipvs
;
1947 ipvs
= net_generic(net
, ip_vs_net_id
);
1951 /* Hold the beast until a service is registerd */
1954 /* Counters used for creating unique names */
1955 ipvs
->gen
= atomic_read(&ipvs_netns_cnt
);
1956 atomic_inc(&ipvs_netns_cnt
);
1959 if (ip_vs_estimator_net_init(net
) < 0)
1960 goto estimator_fail
;
1962 if (ip_vs_control_net_init(net
) < 0)
1965 if (ip_vs_protocol_net_init(net
) < 0)
1968 if (ip_vs_app_net_init(net
) < 0)
1971 if (ip_vs_conn_net_init(net
) < 0)
1974 if (ip_vs_sync_net_init(net
) < 0)
1977 printk(KERN_INFO
"IPVS: Creating netns size=%zu id=%d\n",
1978 sizeof(struct netns_ipvs
), ipvs
->gen
);
1985 ip_vs_conn_net_cleanup(net
);
1987 ip_vs_app_net_cleanup(net
);
1989 ip_vs_protocol_net_cleanup(net
);
1991 ip_vs_control_net_cleanup(net
);
1993 ip_vs_estimator_net_cleanup(net
);
1999 static void __net_exit
__ip_vs_cleanup(struct net
*net
)
2001 ip_vs_service_net_cleanup(net
); /* ip_vs_flush() with locks */
2002 ip_vs_conn_net_cleanup(net
);
2003 ip_vs_app_net_cleanup(net
);
2004 ip_vs_protocol_net_cleanup(net
);
2005 ip_vs_control_net_cleanup(net
);
2006 ip_vs_estimator_net_cleanup(net
);
2007 IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net
)->gen
);
2011 static void __net_exit
__ip_vs_dev_cleanup(struct net
*net
)
2014 net_ipvs(net
)->enable
= 0; /* Disable packet reception */
2016 ip_vs_sync_net_cleanup(net
);
2020 static struct pernet_operations ipvs_core_ops
= {
2021 .init
= __ip_vs_init
,
2022 .exit
= __ip_vs_cleanup
,
2023 .id
= &ip_vs_net_id
,
2024 .size
= sizeof(struct netns_ipvs
),
2027 static struct pernet_operations ipvs_core_dev_ops
= {
2028 .exit
= __ip_vs_dev_cleanup
,
2032 * Initialize IP Virtual Server
2034 static int __init
ip_vs_init(void)
2038 ret
= ip_vs_control_init();
2040 pr_err("can't setup control.\n");
2044 ip_vs_protocol_init();
2046 ret
= ip_vs_conn_init();
2048 pr_err("can't setup connection table.\n");
2049 goto cleanup_protocol
;
2052 ret
= register_pernet_subsys(&ipvs_core_ops
); /* Alloc ip_vs struct */
2056 ret
= register_pernet_device(&ipvs_core_dev_ops
);
2060 ret
= nf_register_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2062 pr_err("can't register hooks.\n");
2066 ret
= ip_vs_register_nl_ioctl();
2068 pr_err("can't register netlink/ioctl.\n");
2072 pr_info("ipvs loaded.\n");
2077 nf_unregister_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2079 unregister_pernet_device(&ipvs_core_dev_ops
);
2081 unregister_pernet_subsys(&ipvs_core_ops
);
2083 ip_vs_conn_cleanup();
2085 ip_vs_protocol_cleanup();
2086 ip_vs_control_cleanup();
2091 static void __exit
ip_vs_cleanup(void)
2093 ip_vs_unregister_nl_ioctl();
2094 nf_unregister_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2095 unregister_pernet_device(&ipvs_core_dev_ops
);
2096 unregister_pernet_subsys(&ipvs_core_ops
); /* free ip_vs struct */
2097 ip_vs_conn_cleanup();
2098 ip_vs_protocol_cleanup();
2099 ip_vs_control_cleanup();
2100 pr_info("ipvs unloaded.\n");
2103 module_init(ip_vs_init
);
2104 module_exit(ip_vs_cleanup
);
2105 MODULE_LICENSE("GPL");