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
= cp
->ipvs
;
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
);
122 u64_stats_update_begin(&s
->syncp
);
124 s
->cnt
.inbytes
+= skb
->len
;
125 u64_stats_update_end(&s
->syncp
);
128 svc
= rcu_dereference(dest
->svc
);
129 s
= this_cpu_ptr(svc
->stats
.cpustats
);
130 u64_stats_update_begin(&s
->syncp
);
132 s
->cnt
.inbytes
+= skb
->len
;
133 u64_stats_update_end(&s
->syncp
);
136 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
137 u64_stats_update_begin(&s
->syncp
);
139 s
->cnt
.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
= cp
->ipvs
;
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
);
156 u64_stats_update_begin(&s
->syncp
);
158 s
->cnt
.outbytes
+= skb
->len
;
159 u64_stats_update_end(&s
->syncp
);
162 svc
= rcu_dereference(dest
->svc
);
163 s
= this_cpu_ptr(svc
->stats
.cpustats
);
164 u64_stats_update_begin(&s
->syncp
);
166 s
->cnt
.outbytes
+= skb
->len
;
167 u64_stats_update_end(&s
->syncp
);
170 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
171 u64_stats_update_begin(&s
->syncp
);
173 s
->cnt
.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
= svc
->ipvs
;
183 struct ip_vs_cpu_stats
*s
;
185 s
= this_cpu_ptr(cp
->dest
->stats
.cpustats
);
186 u64_stats_update_begin(&s
->syncp
);
188 u64_stats_update_end(&s
->syncp
);
190 s
= this_cpu_ptr(svc
->stats
.cpustats
);
191 u64_stats_update_begin(&s
->syncp
);
193 u64_stats_update_end(&s
->syncp
);
195 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
196 u64_stats_update_begin(&s
->syncp
);
198 u64_stats_update_end(&s
->syncp
);
203 ip_vs_set_state(struct ip_vs_conn
*cp
, int direction
,
204 const struct sk_buff
*skb
,
205 struct ip_vs_proto_data
*pd
)
207 if (likely(pd
->pp
->state_transition
))
208 pd
->pp
->state_transition(cp
, direction
, skb
, pd
);
212 ip_vs_conn_fill_param_persist(const struct ip_vs_service
*svc
,
213 struct sk_buff
*skb
, int protocol
,
214 const union nf_inet_addr
*caddr
, __be16 cport
,
215 const union nf_inet_addr
*vaddr
, __be16 vport
,
216 struct ip_vs_conn_param
*p
)
218 ip_vs_conn_fill_param(svc
->ipvs
, svc
->af
, protocol
, caddr
, cport
, vaddr
,
220 p
->pe
= rcu_dereference(svc
->pe
);
221 if (p
->pe
&& p
->pe
->fill_param
)
222 return p
->pe
->fill_param(p
, skb
);
228 * IPVS persistent scheduling function
229 * It creates a connection entry according to its template if exists,
230 * or selects a server and creates a connection entry plus a template.
231 * Locking: we are svc user (svc->refcnt), so we hold all dests too
232 * Protocols supported: TCP, UDP
234 static struct ip_vs_conn
*
235 ip_vs_sched_persist(struct ip_vs_service
*svc
,
236 struct sk_buff
*skb
, __be16 src_port
, __be16 dst_port
,
237 int *ignored
, struct ip_vs_iphdr
*iph
)
239 struct ip_vs_conn
*cp
= NULL
;
240 struct ip_vs_dest
*dest
;
241 struct ip_vs_conn
*ct
;
242 __be16 dport
= 0; /* destination port to forward */
244 struct ip_vs_conn_param param
;
245 const union nf_inet_addr fwmark
= { .ip
= htonl(svc
->fwmark
) };
246 union nf_inet_addr snet
; /* source network of the client,
248 const union nf_inet_addr
*src_addr
, *dst_addr
;
250 if (likely(!ip_vs_iph_inverse(iph
))) {
251 src_addr
= &iph
->saddr
;
252 dst_addr
= &iph
->daddr
;
254 src_addr
= &iph
->daddr
;
255 dst_addr
= &iph
->saddr
;
259 /* Mask saddr with the netmask to adjust template granularity */
260 #ifdef CONFIG_IP_VS_IPV6
261 if (svc
->af
== AF_INET6
)
262 ipv6_addr_prefix(&snet
.in6
, &src_addr
->in6
,
263 (__force __u32
) svc
->netmask
);
266 snet
.ip
= src_addr
->ip
& svc
->netmask
;
268 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
270 IP_VS_DBG_ADDR(svc
->af
, src_addr
), ntohs(src_port
),
271 IP_VS_DBG_ADDR(svc
->af
, dst_addr
), ntohs(dst_port
),
272 IP_VS_DBG_ADDR(svc
->af
, &snet
));
275 * As far as we know, FTP is a very complicated network protocol, and
276 * it uses control connection and data connections. For active FTP,
277 * FTP server initialize data connection to the client, its source port
278 * is often 20. For passive FTP, FTP server tells the clients the port
279 * that it passively listens to, and the client issues the data
280 * connection. In the tunneling or direct routing mode, the load
281 * balancer is on the client-to-server half of connection, the port
282 * number is unknown to the load balancer. So, a conn template like
283 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
284 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
285 * is created for other persistent services.
288 int protocol
= iph
->protocol
;
289 const union nf_inet_addr
*vaddr
= dst_addr
;
292 if (dst_port
== svc
->port
) {
294 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
296 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
298 if (svc
->port
!= FTPPORT
)
301 /* Note: persistent fwmark-based services and
302 * persistent port zero service are handled here.
304 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
305 * port zero template:
306 * <protocol,caddr,0,vaddr,0,daddr,0>
309 protocol
= IPPROTO_IP
;
313 /* return *ignored = -1 so NF_DROP can be used */
314 if (ip_vs_conn_fill_param_persist(svc
, skb
, protocol
, &snet
, 0,
315 vaddr
, vport
, ¶m
) < 0) {
321 /* Check if a template already exists */
322 ct
= ip_vs_ct_in_get(¶m
);
323 if (!ct
|| !ip_vs_check_template(ct
)) {
324 struct ip_vs_scheduler
*sched
;
327 * No template found or the dest of the connection
328 * template is not available.
329 * return *ignored=0 i.e. ICMP and NF_DROP
331 sched
= rcu_dereference(svc
->scheduler
);
333 /* read svc->sched_data after svc->scheduler */
335 dest
= sched
->schedule(svc
, skb
, iph
);
340 IP_VS_DBG(1, "p-schedule: no dest found.\n");
341 kfree(param
.pe_data
);
346 if (dst_port
== svc
->port
&& svc
->port
!= FTPPORT
)
350 * This adds param.pe_data to the template,
351 * and thus param.pe_data will be destroyed
352 * when the template expires */
353 ct
= ip_vs_conn_new(¶m
, dest
->af
, &dest
->addr
, dport
,
354 IP_VS_CONN_F_TEMPLATE
, dest
, skb
->mark
);
356 kfree(param
.pe_data
);
361 ct
->timeout
= svc
->timeout
;
363 /* set destination with the found template */
365 kfree(param
.pe_data
);
369 if (dport
== svc
->port
&& dest
->port
)
372 flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
373 && iph
->protocol
== IPPROTO_UDP
) ?
374 IP_VS_CONN_F_ONE_PACKET
: 0;
377 * Create a new connection according to the template
379 ip_vs_conn_fill_param(svc
->ipvs
, svc
->af
, iph
->protocol
, src_addr
,
380 src_port
, dst_addr
, dst_port
, ¶m
);
382 cp
= ip_vs_conn_new(¶m
, dest
->af
, &dest
->addr
, dport
, flags
, dest
,
393 ip_vs_control_add(cp
, ct
);
396 ip_vs_conn_stats(cp
, svc
);
402 * IPVS main scheduling function
403 * It selects a server according to the virtual service, and
404 * creates a connection entry.
405 * Protocols supported: TCP, UDP
409 * 1 : protocol tried to schedule (eg. on SYN), found svc but the
410 * svc/scheduler decides that this packet should be accepted with
411 * NF_ACCEPT because it must not be scheduled.
413 * 0 : scheduler can not find destination, so try bypass or
414 * return ICMP and then NF_DROP (ip_vs_leave).
416 * -1 : scheduler tried to schedule but fatal error occurred, eg.
417 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
418 * failure such as missing Call-ID, ENOMEM on skb_linearize
419 * or pe_data. In this case we should return NF_DROP without
420 * any attempts to send ICMP with ip_vs_leave.
423 ip_vs_schedule(struct ip_vs_service
*svc
, struct sk_buff
*skb
,
424 struct ip_vs_proto_data
*pd
, int *ignored
,
425 struct ip_vs_iphdr
*iph
)
427 struct ip_vs_protocol
*pp
= pd
->pp
;
428 struct ip_vs_conn
*cp
= NULL
;
429 struct ip_vs_scheduler
*sched
;
430 struct ip_vs_dest
*dest
;
431 __be16 _ports
[2], *pptr
, cport
, vport
;
432 const void *caddr
, *vaddr
;
437 * IPv6 frags, only the first hit here.
439 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
443 if (likely(!ip_vs_iph_inverse(iph
))) {
456 * FTPDATA needs this check when using local real server.
457 * Never schedule Active FTPDATA connections from real server.
458 * For LVS-NAT they must be already created. For other methods
459 * with persistence the connection is created on SYN+ACK.
461 if (cport
== FTPDATA
) {
462 IP_VS_DBG_PKT(12, svc
->af
, pp
, skb
, iph
->off
,
463 "Not scheduling FTPDATA");
468 * Do not schedule replies from local real server.
470 if ((!skb
->dev
|| skb
->dev
->flags
& IFF_LOOPBACK
)) {
471 iph
->hdr_flags
^= IP_VS_HDR_INVERSE
;
472 cp
= pp
->conn_in_get(svc
->ipvs
, svc
->af
, skb
, iph
);
473 iph
->hdr_flags
^= IP_VS_HDR_INVERSE
;
476 IP_VS_DBG_PKT(12, svc
->af
, pp
, skb
, iph
->off
,
477 "Not scheduling reply for existing"
479 __ip_vs_conn_put(cp
);
487 if (svc
->flags
& IP_VS_SVC_F_PERSISTENT
)
488 return ip_vs_sched_persist(svc
, skb
, cport
, vport
, ignored
,
494 * Non-persistent service
496 if (!svc
->fwmark
&& vport
!= svc
->port
) {
498 pr_err("Schedule: port zero only supported "
499 "in persistent services, "
500 "check your ipvs configuration\n");
504 sched
= rcu_dereference(svc
->scheduler
);
506 /* read svc->sched_data after svc->scheduler */
508 dest
= sched
->schedule(svc
, skb
, iph
);
513 IP_VS_DBG(1, "Schedule: no dest found.\n");
517 flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
518 && iph
->protocol
== IPPROTO_UDP
) ?
519 IP_VS_CONN_F_ONE_PACKET
: 0;
522 * Create a connection entry.
525 struct ip_vs_conn_param p
;
527 ip_vs_conn_fill_param(svc
->ipvs
, svc
->af
, iph
->protocol
,
528 caddr
, cport
, vaddr
, vport
, &p
);
529 cp
= ip_vs_conn_new(&p
, dest
->af
, &dest
->addr
,
530 dest
->port
? dest
->port
: vport
,
531 flags
, dest
, skb
->mark
);
538 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
539 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
541 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
542 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
543 IP_VS_DBG_ADDR(cp
->daf
, &cp
->daddr
), ntohs(cp
->dport
),
544 cp
->flags
, atomic_read(&cp
->refcnt
));
546 ip_vs_conn_stats(cp
, svc
);
551 static inline int ip_vs_addr_is_unicast(struct net
*net
, int af
,
552 union nf_inet_addr
*addr
)
554 #ifdef CONFIG_IP_VS_IPV6
556 return ipv6_addr_type(&addr
->in6
) & IPV6_ADDR_UNICAST
;
558 return (inet_addr_type(net
, addr
->ip
) == RTN_UNICAST
);
563 * Pass or drop the packet.
564 * Called by ip_vs_in, when the virtual service is available but
565 * no destination is available for a new connection.
567 int ip_vs_leave(struct ip_vs_service
*svc
, struct sk_buff
*skb
,
568 struct ip_vs_proto_data
*pd
, struct ip_vs_iphdr
*iph
)
570 __be16 _ports
[2], *pptr
, dport
;
572 struct netns_ipvs
*ipvs
;
574 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
577 dport
= likely(!ip_vs_iph_inverse(iph
)) ? pptr
[1] : pptr
[0];
582 /* if it is fwmark-based service, the cache_bypass sysctl is up
583 and the destination is a non-local unicast, then create
584 a cache_bypass connection entry */
585 ipvs
= net_ipvs(net
);
586 if (sysctl_cache_bypass(ipvs
) && svc
->fwmark
&&
587 !(iph
->hdr_flags
& (IP_VS_HDR_INVERSE
| IP_VS_HDR_ICMP
)) &&
588 ip_vs_addr_is_unicast(net
, svc
->af
, &iph
->daddr
)) {
590 struct ip_vs_conn
*cp
;
591 unsigned int flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
&&
592 iph
->protocol
== IPPROTO_UDP
) ?
593 IP_VS_CONN_F_ONE_PACKET
: 0;
594 union nf_inet_addr daddr
= { .all
= { 0, 0, 0, 0 } };
596 /* create a new connection entry */
597 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__
);
599 struct ip_vs_conn_param p
;
600 ip_vs_conn_fill_param(svc
->ipvs
, svc
->af
, iph
->protocol
,
601 &iph
->saddr
, pptr
[0],
602 &iph
->daddr
, pptr
[1], &p
);
603 cp
= ip_vs_conn_new(&p
, svc
->af
, &daddr
, 0,
604 IP_VS_CONN_F_BYPASS
| flags
,
611 ip_vs_in_stats(cp
, skb
);
614 ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pd
);
616 /* transmit the first SYN packet */
617 ret
= cp
->packet_xmit(skb
, cp
, pd
->pp
, iph
);
618 /* do not touch skb anymore */
620 atomic_inc(&cp
->in_pkts
);
626 * When the virtual ftp service is presented, packets destined
627 * for other services on the VIP may get here (except services
628 * listed in the ipvs table), pass the packets, because it is
629 * not ipvs job to decide to drop the packets.
631 if (svc
->port
== FTPPORT
&& dport
!= FTPPORT
)
634 if (unlikely(ip_vs_iph_icmp(iph
)))
638 * Notify the client that the destination is unreachable, and
639 * release the socket buffer.
640 * Since it is in IP layer, the TCP socket is not actually
641 * created, the TCP RST packet cannot be sent, instead that
642 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
644 #ifdef CONFIG_IP_VS_IPV6
645 if (svc
->af
== AF_INET6
) {
647 skb
->dev
= net
->loopback_dev
;
648 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
651 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
658 static int sysctl_snat_reroute(struct sk_buff
*skb
)
660 struct netns_ipvs
*ipvs
= net_ipvs(skb_net(skb
));
661 return ipvs
->sysctl_snat_reroute
;
664 static int sysctl_nat_icmp_send(struct net
*net
)
666 struct netns_ipvs
*ipvs
= net_ipvs(net
);
667 return ipvs
->sysctl_nat_icmp_send
;
670 static int sysctl_expire_nodest_conn(struct netns_ipvs
*ipvs
)
672 return ipvs
->sysctl_expire_nodest_conn
;
677 static int sysctl_snat_reroute(struct sk_buff
*skb
) { return 0; }
678 static int sysctl_nat_icmp_send(struct net
*net
) { return 0; }
679 static int sysctl_expire_nodest_conn(struct netns_ipvs
*ipvs
) { return 0; }
683 __sum16
ip_vs_checksum_complete(struct sk_buff
*skb
, int offset
)
685 return csum_fold(skb_checksum(skb
, offset
, skb
->len
- offset
, 0));
688 static inline enum ip_defrag_users
ip_vs_defrag_user(unsigned int hooknum
)
690 if (NF_INET_LOCAL_IN
== hooknum
)
691 return IP_DEFRAG_VS_IN
;
692 if (NF_INET_FORWARD
== hooknum
)
693 return IP_DEFRAG_VS_FWD
;
694 return IP_DEFRAG_VS_OUT
;
697 static inline int ip_vs_gather_frags(struct sk_buff
*skb
, u_int32_t user
)
702 err
= ip_defrag(skb
, user
);
705 ip_send_check(ip_hdr(skb
));
710 static int ip_vs_route_me_harder(int af
, struct sk_buff
*skb
,
711 unsigned int hooknum
)
713 if (!sysctl_snat_reroute(skb
))
715 /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
716 if (NF_INET_LOCAL_IN
== hooknum
)
718 #ifdef CONFIG_IP_VS_IPV6
719 if (af
== AF_INET6
) {
720 struct dst_entry
*dst
= skb_dst(skb
);
722 if (dst
->dev
&& !(dst
->dev
->flags
& IFF_LOOPBACK
) &&
723 ip6_route_me_harder(skb
) != 0)
727 if (!(skb_rtable(skb
)->rt_flags
& RTCF_LOCAL
) &&
728 ip_route_me_harder(skb
, RTN_LOCAL
) != 0)
735 * Packet has been made sufficiently writable in caller
736 * - inout: 1=in->out, 0=out->in
738 void ip_vs_nat_icmp(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
739 struct ip_vs_conn
*cp
, int inout
)
741 struct iphdr
*iph
= ip_hdr(skb
);
742 unsigned int icmp_offset
= iph
->ihl
*4;
743 struct icmphdr
*icmph
= (struct icmphdr
*)(skb_network_header(skb
) +
745 struct iphdr
*ciph
= (struct iphdr
*)(icmph
+ 1);
748 iph
->saddr
= cp
->vaddr
.ip
;
750 ciph
->daddr
= cp
->vaddr
.ip
;
753 iph
->daddr
= cp
->daddr
.ip
;
755 ciph
->saddr
= cp
->daddr
.ip
;
759 /* the TCP/UDP/SCTP port */
760 if (IPPROTO_TCP
== ciph
->protocol
|| IPPROTO_UDP
== ciph
->protocol
||
761 IPPROTO_SCTP
== ciph
->protocol
) {
762 __be16
*ports
= (void *)ciph
+ ciph
->ihl
*4;
765 ports
[1] = cp
->vport
;
767 ports
[0] = cp
->dport
;
770 /* And finally the ICMP checksum */
772 icmph
->checksum
= ip_vs_checksum_complete(skb
, icmp_offset
);
773 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
776 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, (void *)ciph
- (void *)iph
,
777 "Forwarding altered outgoing ICMP");
779 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, (void *)ciph
- (void *)iph
,
780 "Forwarding altered incoming ICMP");
783 #ifdef CONFIG_IP_VS_IPV6
784 void ip_vs_nat_icmp_v6(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
785 struct ip_vs_conn
*cp
, int inout
)
787 struct ipv6hdr
*iph
= ipv6_hdr(skb
);
788 unsigned int icmp_offset
= 0;
789 unsigned int offs
= 0; /* header offset*/
791 struct icmp6hdr
*icmph
;
792 struct ipv6hdr
*ciph
;
793 unsigned short fragoffs
;
795 ipv6_find_hdr(skb
, &icmp_offset
, IPPROTO_ICMPV6
, &fragoffs
, NULL
);
796 icmph
= (struct icmp6hdr
*)(skb_network_header(skb
) + icmp_offset
);
797 offs
= icmp_offset
+ sizeof(struct icmp6hdr
);
798 ciph
= (struct ipv6hdr
*)(skb_network_header(skb
) + offs
);
800 protocol
= ipv6_find_hdr(skb
, &offs
, -1, &fragoffs
, NULL
);
803 iph
->saddr
= cp
->vaddr
.in6
;
804 ciph
->daddr
= cp
->vaddr
.in6
;
806 iph
->daddr
= cp
->daddr
.in6
;
807 ciph
->saddr
= cp
->daddr
.in6
;
810 /* the TCP/UDP/SCTP port */
811 if (!fragoffs
&& (IPPROTO_TCP
== protocol
|| IPPROTO_UDP
== protocol
||
812 IPPROTO_SCTP
== protocol
)) {
813 __be16
*ports
= (void *)(skb_network_header(skb
) + offs
);
815 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__
,
816 ntohs(inout
? ports
[1] : ports
[0]),
817 ntohs(inout
? cp
->vport
: cp
->dport
));
819 ports
[1] = cp
->vport
;
821 ports
[0] = cp
->dport
;
824 /* And finally the ICMP checksum */
825 icmph
->icmp6_cksum
= ~csum_ipv6_magic(&iph
->saddr
, &iph
->daddr
,
826 skb
->len
- icmp_offset
,
828 skb
->csum_start
= skb_network_header(skb
) - skb
->head
+ icmp_offset
;
829 skb
->csum_offset
= offsetof(struct icmp6hdr
, icmp6_cksum
);
830 skb
->ip_summed
= CHECKSUM_PARTIAL
;
833 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
,
834 (void *)ciph
- (void *)iph
,
835 "Forwarding altered outgoing ICMPv6");
837 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
,
838 (void *)ciph
- (void *)iph
,
839 "Forwarding altered incoming ICMPv6");
843 /* Handle relevant response ICMP messages - forward to the right
846 static int handle_response_icmp(int af
, struct sk_buff
*skb
,
847 union nf_inet_addr
*snet
,
848 __u8 protocol
, struct ip_vs_conn
*cp
,
849 struct ip_vs_protocol
*pp
,
850 unsigned int offset
, unsigned int ihl
,
851 unsigned int hooknum
)
853 unsigned int verdict
= NF_DROP
;
855 if (IP_VS_FWD_METHOD(cp
) != 0) {
856 pr_err("shouldn't reach here, because the box is on the "
857 "half connection in the tun/dr module.\n");
860 /* Ensure the checksum is correct */
861 if (!skb_csum_unnecessary(skb
) && ip_vs_checksum_complete(skb
, ihl
)) {
862 /* Failed checksum! */
863 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
864 IP_VS_DBG_ADDR(af
, snet
));
868 if (IPPROTO_TCP
== protocol
|| IPPROTO_UDP
== protocol
||
869 IPPROTO_SCTP
== protocol
)
870 offset
+= 2 * sizeof(__u16
);
871 if (!skb_make_writable(skb
, offset
))
874 #ifdef CONFIG_IP_VS_IPV6
876 ip_vs_nat_icmp_v6(skb
, pp
, cp
, 1);
879 ip_vs_nat_icmp(skb
, pp
, cp
, 1);
881 if (ip_vs_route_me_harder(af
, skb
, hooknum
))
884 /* do the statistics and put it back */
885 ip_vs_out_stats(cp
, skb
);
887 skb
->ipvs_property
= 1;
888 if (!(cp
->flags
& IP_VS_CONN_F_NFCT
))
891 ip_vs_update_conntrack(skb
, cp
, 0);
895 __ip_vs_conn_put(cp
);
901 * Handle ICMP messages in the inside-to-outside direction (outgoing).
902 * Find any that might be relevant, check against existing connections.
903 * Currently handles error types - unreachable, quench, ttl exceeded.
905 static int ip_vs_out_icmp(struct sk_buff
*skb
, int *related
,
906 unsigned int hooknum
)
908 struct netns_ipvs
*ipvs
= net_ipvs(skb_net(skb
));
910 struct icmphdr _icmph
, *ic
;
911 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
912 struct ip_vs_iphdr ciph
;
913 struct ip_vs_conn
*cp
;
914 struct ip_vs_protocol
*pp
;
915 unsigned int offset
, ihl
;
916 union nf_inet_addr snet
;
920 /* reassemble IP fragments */
921 if (ip_is_fragment(ip_hdr(skb
))) {
922 if (ip_vs_gather_frags(skb
, ip_vs_defrag_user(hooknum
)))
927 offset
= ihl
= iph
->ihl
* 4;
928 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
932 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
933 ic
->type
, ntohs(icmp_id(ic
)),
934 &iph
->saddr
, &iph
->daddr
);
937 * Work through seeing if this is for us.
938 * These checks are supposed to be in an order that means easy
939 * things are checked first to speed up processing.... however
940 * this means that some packets will manage to get a long way
941 * down this stack and then be rejected, but that's life.
943 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
944 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
945 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
950 /* Now find the contained IP header */
951 offset
+= sizeof(_icmph
);
952 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
954 return NF_ACCEPT
; /* The packet looks wrong, ignore */
956 pp
= ip_vs_proto_get(cih
->protocol
);
960 /* Is the embedded protocol header present? */
961 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
965 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, offset
,
966 "Checking outgoing ICMP for");
968 ip_vs_fill_iph_skb_icmp(AF_INET
, skb
, offset
, true, &ciph
);
970 /* The embedded headers contain source and dest in reverse order */
971 cp
= pp
->conn_out_get(ipvs
, AF_INET
, skb
, &ciph
);
975 snet
.ip
= iph
->saddr
;
976 return handle_response_icmp(AF_INET
, skb
, &snet
, cih
->protocol
, cp
,
977 pp
, ciph
.len
, ihl
, hooknum
);
980 #ifdef CONFIG_IP_VS_IPV6
981 static int ip_vs_out_icmp_v6(struct sk_buff
*skb
, int *related
,
982 unsigned int hooknum
, struct ip_vs_iphdr
*ipvsh
)
984 struct netns_ipvs
*ipvs
= net_ipvs(skb_net(skb
));
985 struct icmp6hdr _icmph
, *ic
;
986 struct ip_vs_iphdr ciph
= {.flags
= 0, .fragoffs
= 0};/*Contained IP */
987 struct ip_vs_conn
*cp
;
988 struct ip_vs_protocol
*pp
;
989 union nf_inet_addr snet
;
993 ic
= frag_safe_skb_hp(skb
, ipvsh
->len
, sizeof(_icmph
), &_icmph
, ipvsh
);
998 * Work through seeing if this is for us.
999 * These checks are supposed to be in an order that means easy
1000 * things are checked first to speed up processing.... however
1001 * this means that some packets will manage to get a long way
1002 * down this stack and then be rejected, but that's life.
1004 if (ic
->icmp6_type
& ICMPV6_INFOMSG_MASK
) {
1008 /* Fragment header that is before ICMP header tells us that:
1009 * it's not an error message since they can't be fragmented.
1011 if (ipvsh
->flags
& IP6_FH_F_FRAG
)
1014 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1015 ic
->icmp6_type
, ntohs(icmpv6_id(ic
)),
1016 &ipvsh
->saddr
, &ipvsh
->daddr
);
1018 if (!ip_vs_fill_iph_skb_icmp(AF_INET6
, skb
, ipvsh
->len
+ sizeof(_icmph
),
1020 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1022 pp
= ip_vs_proto_get(ciph
.protocol
);
1026 /* The embedded headers contain source and dest in reverse order */
1027 cp
= pp
->conn_out_get(ipvs
, AF_INET6
, skb
, &ciph
);
1031 snet
.in6
= ciph
.saddr
.in6
;
1033 return handle_response_icmp(AF_INET6
, skb
, &snet
, ciph
.protocol
, cp
,
1034 pp
, offset
, sizeof(struct ipv6hdr
),
1040 * Check if sctp chunc is ABORT chunk
1042 static inline int is_sctp_abort(const struct sk_buff
*skb
, int nh_len
)
1044 sctp_chunkhdr_t
*sch
, schunk
;
1045 sch
= skb_header_pointer(skb
, nh_len
+ sizeof(sctp_sctphdr_t
),
1046 sizeof(schunk
), &schunk
);
1049 if (sch
->type
== SCTP_CID_ABORT
)
1054 static inline int is_tcp_reset(const struct sk_buff
*skb
, int nh_len
)
1056 struct tcphdr _tcph
, *th
;
1058 th
= skb_header_pointer(skb
, nh_len
, sizeof(_tcph
), &_tcph
);
1064 static inline bool is_new_conn(const struct sk_buff
*skb
,
1065 struct ip_vs_iphdr
*iph
)
1067 switch (iph
->protocol
) {
1069 struct tcphdr _tcph
, *th
;
1071 th
= skb_header_pointer(skb
, iph
->len
, sizeof(_tcph
), &_tcph
);
1076 case IPPROTO_SCTP
: {
1077 sctp_chunkhdr_t
*sch
, schunk
;
1079 sch
= skb_header_pointer(skb
, iph
->len
+ sizeof(sctp_sctphdr_t
),
1080 sizeof(schunk
), &schunk
);
1083 return sch
->type
== SCTP_CID_INIT
;
1090 static inline bool is_new_conn_expected(const struct ip_vs_conn
*cp
,
1091 int conn_reuse_mode
)
1093 /* Controlled (FTP DATA or persistence)? */
1097 switch (cp
->protocol
) {
1099 return (cp
->state
== IP_VS_TCP_S_TIME_WAIT
) ||
1100 ((conn_reuse_mode
& 2) &&
1101 (cp
->state
== IP_VS_TCP_S_FIN_WAIT
) &&
1102 (cp
->flags
& IP_VS_CONN_F_NOOUTPUT
));
1104 return cp
->state
== IP_VS_SCTP_S_CLOSED
;
1110 /* Handle response packets: rewrite addresses and send away...
1113 handle_response(int af
, struct sk_buff
*skb
, struct ip_vs_proto_data
*pd
,
1114 struct ip_vs_conn
*cp
, struct ip_vs_iphdr
*iph
,
1115 unsigned int hooknum
)
1117 struct ip_vs_protocol
*pp
= pd
->pp
;
1119 IP_VS_DBG_PKT(11, af
, pp
, skb
, iph
->off
, "Outgoing packet");
1121 if (!skb_make_writable(skb
, iph
->len
))
1124 /* mangle the packet */
1125 if (pp
->snat_handler
&& !pp
->snat_handler(skb
, pp
, cp
, iph
))
1128 #ifdef CONFIG_IP_VS_IPV6
1130 ipv6_hdr(skb
)->saddr
= cp
->vaddr
.in6
;
1134 ip_hdr(skb
)->saddr
= cp
->vaddr
.ip
;
1135 ip_send_check(ip_hdr(skb
));
1139 * nf_iterate does not expect change in the skb->dst->dev.
1140 * It looks like it is not fatal to enable this code for hooks
1141 * where our handlers are at the end of the chain list and
1142 * when all next handlers use skb->dst->dev and not outdev.
1143 * It will definitely route properly the inout NAT traffic
1144 * when multiple paths are used.
1147 /* For policy routing, packets originating from this
1148 * machine itself may be routed differently to packets
1149 * passing through. We want this packet to be routed as
1150 * if it came from this machine itself. So re-compute
1151 * the routing information.
1153 if (ip_vs_route_me_harder(af
, skb
, hooknum
))
1156 IP_VS_DBG_PKT(10, af
, pp
, skb
, iph
->off
, "After SNAT");
1158 ip_vs_out_stats(cp
, skb
);
1159 ip_vs_set_state(cp
, IP_VS_DIR_OUTPUT
, skb
, pd
);
1160 skb
->ipvs_property
= 1;
1161 if (!(cp
->flags
& IP_VS_CONN_F_NFCT
))
1164 ip_vs_update_conntrack(skb
, cp
, 0);
1178 * Check if outgoing packet belongs to the established ip_vs_conn.
1181 ip_vs_out(unsigned int hooknum
, struct sk_buff
*skb
, int af
)
1183 struct net
*net
= NULL
;
1184 struct netns_ipvs
*ipvs
;
1185 struct ip_vs_iphdr iph
;
1186 struct ip_vs_protocol
*pp
;
1187 struct ip_vs_proto_data
*pd
;
1188 struct ip_vs_conn
*cp
;
1192 /* Already marked as IPVS request or reply? */
1193 if (skb
->ipvs_property
)
1196 /* Bad... Do not break raw sockets */
1197 if (unlikely(skb
->sk
!= NULL
&& hooknum
== NF_INET_LOCAL_OUT
&&
1199 struct sock
*sk
= skb
->sk
;
1200 struct inet_sock
*inet
= inet_sk(skb
->sk
);
1202 if (inet
&& sk
->sk_family
== PF_INET
&& inet
->nodefrag
)
1206 if (unlikely(!skb_dst(skb
)))
1210 ipvs
= net_ipvs(net
);
1214 ip_vs_fill_iph_skb(af
, skb
, false, &iph
);
1215 #ifdef CONFIG_IP_VS_IPV6
1216 if (af
== AF_INET6
) {
1217 if (unlikely(iph
.protocol
== IPPROTO_ICMPV6
)) {
1219 int verdict
= ip_vs_out_icmp_v6(skb
, &related
,
1227 if (unlikely(iph
.protocol
== IPPROTO_ICMP
)) {
1229 int verdict
= ip_vs_out_icmp(skb
, &related
, hooknum
);
1235 pd
= ip_vs_proto_data_get(ipvs
, iph
.protocol
);
1240 /* reassemble IP fragments */
1241 #ifdef CONFIG_IP_VS_IPV6
1244 if (unlikely(ip_is_fragment(ip_hdr(skb
)) && !pp
->dont_defrag
)) {
1245 if (ip_vs_gather_frags(skb
,
1246 ip_vs_defrag_user(hooknum
)))
1249 ip_vs_fill_iph_skb(AF_INET
, skb
, false, &iph
);
1253 * Check if the packet belongs to an existing entry
1255 cp
= pp
->conn_out_get(ipvs
, af
, skb
, &iph
);
1258 return handle_response(af
, skb
, pd
, cp
, &iph
, hooknum
);
1259 if (sysctl_nat_icmp_send(net
) &&
1260 (pp
->protocol
== IPPROTO_TCP
||
1261 pp
->protocol
== IPPROTO_UDP
||
1262 pp
->protocol
== IPPROTO_SCTP
)) {
1263 __be16 _ports
[2], *pptr
;
1265 pptr
= frag_safe_skb_hp(skb
, iph
.len
,
1266 sizeof(_ports
), _ports
, &iph
);
1268 return NF_ACCEPT
; /* Not for me */
1269 if (ip_vs_has_real_service(ipvs
, af
, iph
.protocol
, &iph
.saddr
,
1272 * Notify the real server: there is no
1273 * existing entry if it is not RST
1274 * packet or not TCP packet.
1276 if ((iph
.protocol
!= IPPROTO_TCP
&&
1277 iph
.protocol
!= IPPROTO_SCTP
)
1278 || ((iph
.protocol
== IPPROTO_TCP
1279 && !is_tcp_reset(skb
, iph
.len
))
1280 || (iph
.protocol
== IPPROTO_SCTP
1281 && !is_sctp_abort(skb
,
1283 #ifdef CONFIG_IP_VS_IPV6
1284 if (af
== AF_INET6
) {
1286 skb
->dev
= net
->loopback_dev
;
1288 ICMPV6_DEST_UNREACH
,
1289 ICMPV6_PORT_UNREACH
,
1295 ICMP_PORT_UNREACH
, 0);
1300 IP_VS_DBG_PKT(12, af
, pp
, skb
, iph
.off
,
1301 "ip_vs_out: packet continues traversal as normal");
1306 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1307 * used only for VS/NAT.
1308 * Check if packet is reply for established ip_vs_conn.
1311 ip_vs_reply4(void *priv
, struct sk_buff
*skb
,
1312 const struct nf_hook_state
*state
)
1314 return ip_vs_out(state
->hook
, skb
, AF_INET
);
1318 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1319 * Check if packet is reply for established ip_vs_conn.
1322 ip_vs_local_reply4(void *priv
, struct sk_buff
*skb
,
1323 const struct nf_hook_state
*state
)
1325 return ip_vs_out(state
->hook
, skb
, AF_INET
);
1328 #ifdef CONFIG_IP_VS_IPV6
1331 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1332 * used only for VS/NAT.
1333 * Check if packet is reply for established ip_vs_conn.
1336 ip_vs_reply6(void *priv
, struct sk_buff
*skb
,
1337 const struct nf_hook_state
*state
)
1339 return ip_vs_out(state
->hook
, skb
, AF_INET6
);
1343 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1344 * Check if packet is reply for established ip_vs_conn.
1347 ip_vs_local_reply6(void *priv
, struct sk_buff
*skb
,
1348 const struct nf_hook_state
*state
)
1350 return ip_vs_out(state
->hook
, skb
, AF_INET6
);
1356 ip_vs_try_to_schedule(struct netns_ipvs
*ipvs
, int af
, struct sk_buff
*skb
,
1357 struct ip_vs_proto_data
*pd
,
1358 int *verdict
, struct ip_vs_conn
**cpp
,
1359 struct ip_vs_iphdr
*iph
)
1361 struct ip_vs_protocol
*pp
= pd
->pp
;
1363 if (!iph
->fragoffs
) {
1364 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1365 * replayed fragment zero will already have created the cp
1368 /* Schedule and create new connection entry into cpp */
1369 if (!pp
->conn_schedule(ipvs
, af
, skb
, pd
, verdict
, cpp
, iph
))
1373 if (unlikely(!*cpp
)) {
1374 /* sorry, all this trouble for a no-hit :) */
1375 IP_VS_DBG_PKT(12, af
, pp
, skb
, iph
->off
,
1376 "ip_vs_in: packet continues traversal as normal");
1377 if (iph
->fragoffs
) {
1378 /* Fragment that couldn't be mapped to a conn entry
1379 * is missing module nf_defrag_ipv6
1381 IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1382 IP_VS_DBG_PKT(7, af
, pp
, skb
, iph
->off
,
1383 "unhandled fragment");
1385 *verdict
= NF_ACCEPT
;
1393 * Handle ICMP messages in the outside-to-inside direction (incoming).
1394 * Find any that might be relevant, check against existing connections,
1395 * forward to the right destination host if relevant.
1396 * Currently handles error types - unreachable, quench, ttl exceeded.
1399 ip_vs_in_icmp(struct sk_buff
*skb
, int *related
, unsigned int hooknum
)
1401 struct net
*net
= NULL
;
1402 struct netns_ipvs
*ipvs
;
1404 struct icmphdr _icmph
, *ic
;
1405 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
1406 struct ip_vs_iphdr ciph
;
1407 struct ip_vs_conn
*cp
;
1408 struct ip_vs_protocol
*pp
;
1409 struct ip_vs_proto_data
*pd
;
1410 unsigned int offset
, offset2
, ihl
, verdict
;
1411 bool ipip
, new_cp
= false;
1415 /* reassemble IP fragments */
1416 if (ip_is_fragment(ip_hdr(skb
))) {
1417 if (ip_vs_gather_frags(skb
, ip_vs_defrag_user(hooknum
)))
1422 offset
= ihl
= iph
->ihl
* 4;
1423 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
1427 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1428 ic
->type
, ntohs(icmp_id(ic
)),
1429 &iph
->saddr
, &iph
->daddr
);
1432 * Work through seeing if this is for us.
1433 * These checks are supposed to be in an order that means easy
1434 * things are checked first to speed up processing.... however
1435 * this means that some packets will manage to get a long way
1436 * down this stack and then be rejected, but that's life.
1438 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
1439 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
1440 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
1445 /* Now find the contained IP header */
1446 offset
+= sizeof(_icmph
);
1447 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
1449 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1452 ipvs
= net_ipvs(net
);
1454 /* Special case for errors for IPIP packets */
1456 if (cih
->protocol
== IPPROTO_IPIP
) {
1457 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
)))
1459 /* Error for our IPIP must arrive at LOCAL_IN */
1460 if (!(skb_rtable(skb
)->rt_flags
& RTCF_LOCAL
))
1462 offset
+= cih
->ihl
* 4;
1463 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
1465 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1469 pd
= ip_vs_proto_data_get(ipvs
, cih
->protocol
);
1474 /* Is the embedded protocol header present? */
1475 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
1479 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, offset
,
1480 "Checking incoming ICMP for");
1483 ip_vs_fill_iph_skb_icmp(AF_INET
, skb
, offset
, !ipip
, &ciph
);
1486 /* The embedded headers contain source and dest in reverse order.
1487 * For IPIP this is error for request, not for reply.
1489 cp
= pp
->conn_in_get(ipvs
, AF_INET
, skb
, &ciph
);
1494 if (!sysctl_schedule_icmp(ipvs
))
1497 if (!ip_vs_try_to_schedule(ipvs
, AF_INET
, skb
, pd
, &v
, &cp
, &ciph
))
1504 /* Ensure the checksum is correct */
1505 if (!skb_csum_unnecessary(skb
) && ip_vs_checksum_complete(skb
, ihl
)) {
1506 /* Failed checksum! */
1507 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1513 __be32 info
= ic
->un
.gateway
;
1514 __u8 type
= ic
->type
;
1515 __u8 code
= ic
->code
;
1517 /* Update the MTU */
1518 if (ic
->type
== ICMP_DEST_UNREACH
&&
1519 ic
->code
== ICMP_FRAG_NEEDED
) {
1520 struct ip_vs_dest
*dest
= cp
->dest
;
1521 u32 mtu
= ntohs(ic
->un
.frag
.mtu
);
1522 __be16 frag_off
= cih
->frag_off
;
1524 /* Strip outer IP and ICMP, go to IPIP header */
1525 if (pskb_pull(skb
, ihl
+ sizeof(_icmph
)) == NULL
)
1527 offset2
-= ihl
+ sizeof(_icmph
);
1528 skb_reset_network_header(skb
);
1529 IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1530 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
, mtu
);
1531 ipv4_update_pmtu(skb
, dev_net(skb
->dev
),
1533 /* Client uses PMTUD? */
1534 if (!(frag_off
& htons(IP_DF
)))
1536 /* Prefer the resulting PMTU */
1538 struct ip_vs_dest_dst
*dest_dst
;
1541 dest_dst
= rcu_dereference(dest
->dest_dst
);
1543 mtu
= dst_mtu(dest_dst
->dst_cache
);
1546 if (mtu
> 68 + sizeof(struct iphdr
))
1547 mtu
-= sizeof(struct iphdr
);
1550 /* Strip outer IP, ICMP and IPIP, go to IP header of
1553 if (pskb_pull(skb
, offset2
) == NULL
)
1555 skb_reset_network_header(skb
);
1556 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1557 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1558 type
, code
, ntohl(info
));
1559 icmp_send(skb
, type
, code
, info
);
1560 /* ICMP can be shorter but anyways, account it */
1561 ip_vs_out_stats(cp
, skb
);
1565 verdict
= NF_STOLEN
;
1569 /* do the statistics and put it back */
1570 ip_vs_in_stats(cp
, skb
);
1571 if (IPPROTO_TCP
== cih
->protocol
|| IPPROTO_UDP
== cih
->protocol
||
1572 IPPROTO_SCTP
== cih
->protocol
)
1573 offset
+= 2 * sizeof(__u16
);
1574 verdict
= ip_vs_icmp_xmit(skb
, cp
, pp
, offset
, hooknum
, &ciph
);
1577 if (likely(!new_cp
))
1578 __ip_vs_conn_put(cp
);
1585 #ifdef CONFIG_IP_VS_IPV6
1586 static int ip_vs_in_icmp_v6(struct sk_buff
*skb
, int *related
,
1587 unsigned int hooknum
, struct ip_vs_iphdr
*iph
)
1589 struct net
*net
= NULL
;
1590 struct netns_ipvs
*ipvs
;
1591 struct icmp6hdr _icmph
, *ic
;
1592 struct ip_vs_iphdr ciph
= {.flags
= 0, .fragoffs
= 0};/*Contained IP */
1593 struct ip_vs_conn
*cp
;
1594 struct ip_vs_protocol
*pp
;
1595 struct ip_vs_proto_data
*pd
;
1596 unsigned int offset
, verdict
;
1597 bool new_cp
= false;
1601 ic
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_icmph
), &_icmph
, iph
);
1606 * Work through seeing if this is for us.
1607 * These checks are supposed to be in an order that means easy
1608 * things are checked first to speed up processing.... however
1609 * this means that some packets will manage to get a long way
1610 * down this stack and then be rejected, but that's life.
1612 if (ic
->icmp6_type
& ICMPV6_INFOMSG_MASK
) {
1616 /* Fragment header that is before ICMP header tells us that:
1617 * it's not an error message since they can't be fragmented.
1619 if (iph
->flags
& IP6_FH_F_FRAG
)
1622 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1623 ic
->icmp6_type
, ntohs(icmpv6_id(ic
)),
1624 &iph
->saddr
, &iph
->daddr
);
1626 offset
= iph
->len
+ sizeof(_icmph
);
1627 if (!ip_vs_fill_iph_skb_icmp(AF_INET6
, skb
, offset
, true, &ciph
))
1631 ipvs
= net_ipvs(net
);
1632 pd
= ip_vs_proto_data_get(ipvs
, ciph
.protocol
);
1637 /* Cannot handle fragmented embedded protocol */
1641 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
, offset
,
1642 "Checking incoming ICMPv6 for");
1644 /* The embedded headers contain source and dest in reverse order
1645 * if not from localhost
1647 cp
= pp
->conn_in_get(ipvs
, AF_INET6
, skb
, &ciph
);
1652 if (!sysctl_schedule_icmp(ipvs
))
1655 if (!ip_vs_try_to_schedule(ipvs
, AF_INET6
, skb
, pd
, &v
, &cp
, &ciph
))
1661 /* VS/TUN, VS/DR and LOCALNODE just let it go */
1662 if ((hooknum
== NF_INET_LOCAL_OUT
) &&
1663 (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
)) {
1664 verdict
= NF_ACCEPT
;
1668 /* do the statistics and put it back */
1669 ip_vs_in_stats(cp
, skb
);
1671 /* Need to mangle contained IPv6 header in ICMPv6 packet */
1673 if (IPPROTO_TCP
== ciph
.protocol
|| IPPROTO_UDP
== ciph
.protocol
||
1674 IPPROTO_SCTP
== ciph
.protocol
)
1675 offset
+= 2 * sizeof(__u16
); /* Also mangle ports */
1677 verdict
= ip_vs_icmp_xmit_v6(skb
, cp
, pp
, offset
, hooknum
, &ciph
);
1680 if (likely(!new_cp
))
1681 __ip_vs_conn_put(cp
);
1691 * Check if it's for virtual services, look it up,
1692 * and send it on its way...
1695 ip_vs_in(unsigned int hooknum
, struct sk_buff
*skb
, int af
)
1698 struct ip_vs_iphdr iph
;
1699 struct ip_vs_protocol
*pp
;
1700 struct ip_vs_proto_data
*pd
;
1701 struct ip_vs_conn
*cp
;
1703 struct netns_ipvs
*ipvs
;
1704 int conn_reuse_mode
;
1706 /* Already marked as IPVS request or reply? */
1707 if (skb
->ipvs_property
)
1712 * - remote client: only PACKET_HOST
1713 * - route: used for struct net when skb->dev is unset
1715 if (unlikely((skb
->pkt_type
!= PACKET_HOST
&&
1716 hooknum
!= NF_INET_LOCAL_OUT
) ||
1718 ip_vs_fill_iph_skb(af
, skb
, false, &iph
);
1719 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1720 " ignored in hook %u\n",
1721 skb
->pkt_type
, iph
.protocol
,
1722 IP_VS_DBG_ADDR(af
, &iph
.daddr
), hooknum
);
1725 /* ipvs enabled in this netns ? */
1727 ipvs
= net_ipvs(net
);
1728 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1731 ip_vs_fill_iph_skb(af
, skb
, false, &iph
);
1733 /* Bad... Do not break raw sockets */
1734 if (unlikely(skb
->sk
!= NULL
&& hooknum
== NF_INET_LOCAL_OUT
&&
1736 struct sock
*sk
= skb
->sk
;
1737 struct inet_sock
*inet
= inet_sk(skb
->sk
);
1739 if (inet
&& sk
->sk_family
== PF_INET
&& inet
->nodefrag
)
1743 #ifdef CONFIG_IP_VS_IPV6
1744 if (af
== AF_INET6
) {
1745 if (unlikely(iph
.protocol
== IPPROTO_ICMPV6
)) {
1747 int verdict
= ip_vs_in_icmp_v6(skb
, &related
, hooknum
,
1755 if (unlikely(iph
.protocol
== IPPROTO_ICMP
)) {
1757 int verdict
= ip_vs_in_icmp(skb
, &related
, hooknum
);
1763 /* Protocol supported? */
1764 pd
= ip_vs_proto_data_get(ipvs
, iph
.protocol
);
1765 if (unlikely(!pd
)) {
1766 /* The only way we'll see this packet again is if it's
1767 * encapsulated, so mark it with ipvs_property=1 so we
1768 * skip it if we're ignoring tunneled packets
1770 if (sysctl_ignore_tunneled(ipvs
))
1771 skb
->ipvs_property
= 1;
1777 * Check if the packet belongs to an existing connection entry
1779 cp
= pp
->conn_in_get(ipvs
, af
, skb
, &iph
);
1781 conn_reuse_mode
= sysctl_conn_reuse_mode(ipvs
);
1782 if (conn_reuse_mode
&& !iph
.fragoffs
&&
1783 is_new_conn(skb
, &iph
) && cp
&&
1784 ((unlikely(sysctl_expire_nodest_conn(ipvs
)) && cp
->dest
&&
1785 unlikely(!atomic_read(&cp
->dest
->weight
))) ||
1786 unlikely(is_new_conn_expected(cp
, conn_reuse_mode
)))) {
1787 if (!atomic_read(&cp
->n_control
))
1788 ip_vs_conn_expire_now(cp
);
1789 __ip_vs_conn_put(cp
);
1793 if (unlikely(!cp
)) {
1796 if (!ip_vs_try_to_schedule(ipvs
, af
, skb
, pd
, &v
, &cp
, &iph
))
1800 IP_VS_DBG_PKT(11, af
, pp
, skb
, iph
.off
, "Incoming packet");
1802 /* Check the server status */
1803 if (cp
->dest
&& !(cp
->dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
1804 /* the destination server is not available */
1806 if (sysctl_expire_nodest_conn(ipvs
)) {
1807 /* try to expire the connection immediately */
1808 ip_vs_conn_expire_now(cp
);
1810 /* don't restart its timer, and silently
1812 __ip_vs_conn_put(cp
);
1816 ip_vs_in_stats(cp
, skb
);
1817 ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pd
);
1818 if (cp
->packet_xmit
)
1819 ret
= cp
->packet_xmit(skb
, cp
, pp
, &iph
);
1820 /* do not touch skb anymore */
1822 IP_VS_DBG_RL("warning: packet_xmit is null");
1826 /* Increase its packet counter and check if it is needed
1827 * to be synchronized
1829 * Sync connection if it is about to close to
1830 * encorage the standby servers to update the connections timeout
1832 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1835 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
1836 pkts
= sysctl_sync_threshold(ipvs
);
1838 pkts
= atomic_add_return(1, &cp
->in_pkts
);
1840 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
)
1841 ip_vs_sync_conn(ipvs
, cp
, pkts
);
1848 * AF_INET handler in NF_INET_LOCAL_IN chain
1849 * Schedule and forward packets from remote clients
1852 ip_vs_remote_request4(void *priv
, struct sk_buff
*skb
,
1853 const struct nf_hook_state
*state
)
1855 return ip_vs_in(state
->hook
, skb
, AF_INET
);
1859 * AF_INET handler in NF_INET_LOCAL_OUT chain
1860 * Schedule and forward packets from local clients
1863 ip_vs_local_request4(void *priv
, struct sk_buff
*skb
,
1864 const struct nf_hook_state
*state
)
1866 return ip_vs_in(state
->hook
, skb
, AF_INET
);
1869 #ifdef CONFIG_IP_VS_IPV6
1872 * AF_INET6 handler in NF_INET_LOCAL_IN chain
1873 * Schedule and forward packets from remote clients
1876 ip_vs_remote_request6(void *priv
, struct sk_buff
*skb
,
1877 const struct nf_hook_state
*state
)
1879 return ip_vs_in(state
->hook
, skb
, AF_INET6
);
1883 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
1884 * Schedule and forward packets from local clients
1887 ip_vs_local_request6(void *priv
, struct sk_buff
*skb
,
1888 const struct nf_hook_state
*state
)
1890 return ip_vs_in(state
->hook
, skb
, AF_INET6
);
1897 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1898 * related packets destined for 0.0.0.0/0.
1899 * When fwmark-based virtual service is used, such as transparent
1900 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1901 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1902 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1903 * and send them to ip_vs_in_icmp.
1906 ip_vs_forward_icmp(void *priv
, struct sk_buff
*skb
,
1907 const struct nf_hook_state
*state
)
1910 struct netns_ipvs
*ipvs
;
1912 if (ip_hdr(skb
)->protocol
!= IPPROTO_ICMP
)
1915 /* ipvs enabled in this netns ? */
1916 ipvs
= net_ipvs(state
->net
);
1917 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1920 return ip_vs_in_icmp(skb
, &r
, state
->hook
);
1923 #ifdef CONFIG_IP_VS_IPV6
1925 ip_vs_forward_icmp_v6(void *priv
, struct sk_buff
*skb
,
1926 const struct nf_hook_state
*state
)
1929 struct netns_ipvs
*ipvs
;
1930 struct ip_vs_iphdr iphdr
;
1932 ip_vs_fill_iph_skb(AF_INET6
, skb
, false, &iphdr
);
1933 if (iphdr
.protocol
!= IPPROTO_ICMPV6
)
1936 /* ipvs enabled in this netns ? */
1937 ipvs
= net_ipvs(state
->net
);
1938 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1941 return ip_vs_in_icmp_v6(skb
, &r
, state
->hook
, &iphdr
);
1946 static struct nf_hook_ops ip_vs_ops
[] __read_mostly
= {
1947 /* After packet filtering, change source only for VS/NAT */
1949 .hook
= ip_vs_reply4
,
1950 .owner
= THIS_MODULE
,
1952 .hooknum
= NF_INET_LOCAL_IN
,
1953 .priority
= NF_IP_PRI_NAT_SRC
- 2,
1955 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1956 * or VS/NAT(change destination), so that filtering rules can be
1957 * applied to IPVS. */
1959 .hook
= ip_vs_remote_request4
,
1960 .owner
= THIS_MODULE
,
1962 .hooknum
= NF_INET_LOCAL_IN
,
1963 .priority
= NF_IP_PRI_NAT_SRC
- 1,
1965 /* Before ip_vs_in, change source only for VS/NAT */
1967 .hook
= ip_vs_local_reply4
,
1968 .owner
= THIS_MODULE
,
1970 .hooknum
= NF_INET_LOCAL_OUT
,
1971 .priority
= NF_IP_PRI_NAT_DST
+ 1,
1973 /* After mangle, schedule and forward local requests */
1975 .hook
= ip_vs_local_request4
,
1976 .owner
= THIS_MODULE
,
1978 .hooknum
= NF_INET_LOCAL_OUT
,
1979 .priority
= NF_IP_PRI_NAT_DST
+ 2,
1981 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1982 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1984 .hook
= ip_vs_forward_icmp
,
1985 .owner
= THIS_MODULE
,
1987 .hooknum
= NF_INET_FORWARD
,
1990 /* After packet filtering, change source only for VS/NAT */
1992 .hook
= ip_vs_reply4
,
1993 .owner
= THIS_MODULE
,
1995 .hooknum
= NF_INET_FORWARD
,
1998 #ifdef CONFIG_IP_VS_IPV6
1999 /* After packet filtering, change source only for VS/NAT */
2001 .hook
= ip_vs_reply6
,
2002 .owner
= THIS_MODULE
,
2004 .hooknum
= NF_INET_LOCAL_IN
,
2005 .priority
= NF_IP6_PRI_NAT_SRC
- 2,
2007 /* After packet filtering, forward packet through VS/DR, VS/TUN,
2008 * or VS/NAT(change destination), so that filtering rules can be
2009 * applied to IPVS. */
2011 .hook
= ip_vs_remote_request6
,
2012 .owner
= THIS_MODULE
,
2014 .hooknum
= NF_INET_LOCAL_IN
,
2015 .priority
= NF_IP6_PRI_NAT_SRC
- 1,
2017 /* Before ip_vs_in, change source only for VS/NAT */
2019 .hook
= ip_vs_local_reply6
,
2020 .owner
= THIS_MODULE
,
2022 .hooknum
= NF_INET_LOCAL_OUT
,
2023 .priority
= NF_IP6_PRI_NAT_DST
+ 1,
2025 /* After mangle, schedule and forward local requests */
2027 .hook
= ip_vs_local_request6
,
2028 .owner
= THIS_MODULE
,
2030 .hooknum
= NF_INET_LOCAL_OUT
,
2031 .priority
= NF_IP6_PRI_NAT_DST
+ 2,
2033 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2034 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2036 .hook
= ip_vs_forward_icmp_v6
,
2037 .owner
= THIS_MODULE
,
2039 .hooknum
= NF_INET_FORWARD
,
2042 /* After packet filtering, change source only for VS/NAT */
2044 .hook
= ip_vs_reply6
,
2045 .owner
= THIS_MODULE
,
2047 .hooknum
= NF_INET_FORWARD
,
2053 * Initialize IP Virtual Server netns mem.
2055 static int __net_init
__ip_vs_init(struct net
*net
)
2057 struct netns_ipvs
*ipvs
;
2059 ipvs
= net_generic(net
, ip_vs_net_id
);
2063 /* Hold the beast until a service is registerd */
2066 /* Counters used for creating unique names */
2067 ipvs
->gen
= atomic_read(&ipvs_netns_cnt
);
2068 atomic_inc(&ipvs_netns_cnt
);
2071 if (ip_vs_estimator_net_init(ipvs
) < 0)
2072 goto estimator_fail
;
2074 if (ip_vs_control_net_init(ipvs
) < 0)
2077 if (ip_vs_protocol_net_init(net
) < 0)
2080 if (ip_vs_app_net_init(ipvs
) < 0)
2083 if (ip_vs_conn_net_init(ipvs
) < 0)
2086 if (ip_vs_sync_net_init(ipvs
) < 0)
2089 printk(KERN_INFO
"IPVS: Creating netns size=%zu id=%d\n",
2090 sizeof(struct netns_ipvs
), ipvs
->gen
);
2097 ip_vs_conn_net_cleanup(ipvs
);
2099 ip_vs_app_net_cleanup(ipvs
);
2101 ip_vs_protocol_net_cleanup(net
);
2103 ip_vs_control_net_cleanup(ipvs
);
2105 ip_vs_estimator_net_cleanup(ipvs
);
2111 static void __net_exit
__ip_vs_cleanup(struct net
*net
)
2113 struct netns_ipvs
*ipvs
= net_ipvs(net
);
2115 ip_vs_service_net_cleanup(ipvs
); /* ip_vs_flush() with locks */
2116 ip_vs_conn_net_cleanup(ipvs
);
2117 ip_vs_app_net_cleanup(ipvs
);
2118 ip_vs_protocol_net_cleanup(net
);
2119 ip_vs_control_net_cleanup(ipvs
);
2120 ip_vs_estimator_net_cleanup(ipvs
);
2121 IP_VS_DBG(2, "ipvs netns %d released\n", ipvs
->gen
);
2125 static void __net_exit
__ip_vs_dev_cleanup(struct net
*net
)
2127 struct netns_ipvs
*ipvs
= net_ipvs(net
);
2129 ipvs
->enable
= 0; /* Disable packet reception */
2131 ip_vs_sync_net_cleanup(ipvs
);
2135 static struct pernet_operations ipvs_core_ops
= {
2136 .init
= __ip_vs_init
,
2137 .exit
= __ip_vs_cleanup
,
2138 .id
= &ip_vs_net_id
,
2139 .size
= sizeof(struct netns_ipvs
),
2142 static struct pernet_operations ipvs_core_dev_ops
= {
2143 .exit
= __ip_vs_dev_cleanup
,
2147 * Initialize IP Virtual Server
2149 static int __init
ip_vs_init(void)
2153 ret
= ip_vs_control_init();
2155 pr_err("can't setup control.\n");
2159 ip_vs_protocol_init();
2161 ret
= ip_vs_conn_init();
2163 pr_err("can't setup connection table.\n");
2164 goto cleanup_protocol
;
2167 ret
= register_pernet_subsys(&ipvs_core_ops
); /* Alloc ip_vs struct */
2171 ret
= register_pernet_device(&ipvs_core_dev_ops
);
2175 ret
= nf_register_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2177 pr_err("can't register hooks.\n");
2181 ret
= ip_vs_register_nl_ioctl();
2183 pr_err("can't register netlink/ioctl.\n");
2187 pr_info("ipvs loaded.\n");
2192 nf_unregister_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2194 unregister_pernet_device(&ipvs_core_dev_ops
);
2196 unregister_pernet_subsys(&ipvs_core_ops
);
2198 ip_vs_conn_cleanup();
2200 ip_vs_protocol_cleanup();
2201 ip_vs_control_cleanup();
2206 static void __exit
ip_vs_cleanup(void)
2208 ip_vs_unregister_nl_ioctl();
2209 nf_unregister_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2210 unregister_pernet_device(&ipvs_core_dev_ops
);
2211 unregister_pernet_subsys(&ipvs_core_ops
); /* free ip_vs struct */
2212 ip_vs_conn_cleanup();
2213 ip_vs_protocol_cleanup();
2214 ip_vs_control_cleanup();
2215 pr_info("ipvs unloaded.\n");
2218 module_init(ip_vs_init
);
2219 module_exit(ip_vs_cleanup
);
2220 MODULE_LICENSE("GPL");