Merge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block
[linux-2.6] / net / ipv4 / ipvs / ip_vs_core.c
1 /*
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
6  *              cluster of servers.
7  *
8  * Version:     $Id: ip_vs_core.c,v 1.34 2003/05/10 03:05:23 wensong Exp $
9  *
10  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
11  *              Peter Kese <peter.kese@ijs.si>
12  *              Julian Anastasov <ja@ssi.bg>
13  *
14  *              This program is free software; you can redistribute it and/or
15  *              modify it under the terms of the GNU General Public License
16  *              as published by the Free Software Foundation; either version
17  *              2 of the License, or (at your option) any later version.
18  *
19  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
20  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
21  * and others.
22  *
23  * Changes:
24  *      Paul `Rusty' Russell            properly handle non-linear skbs
25  *      Harald Welte                    don't use nfcache
26  *
27  */
28
29 #include <linux/module.h>
30 #include <linux/kernel.h>
31 #include <linux/ip.h>
32 #include <linux/tcp.h>
33 #include <linux/icmp.h>
34
35 #include <net/ip.h>
36 #include <net/tcp.h>
37 #include <net/udp.h>
38 #include <net/icmp.h>                   /* for icmp_send */
39 #include <net/route.h>
40
41 #include <linux/netfilter.h>
42 #include <linux/netfilter_ipv4.h>
43
44 #include <net/ip_vs.h>
45
46
47 EXPORT_SYMBOL(register_ip_vs_scheduler);
48 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
49 EXPORT_SYMBOL(ip_vs_skb_replace);
50 EXPORT_SYMBOL(ip_vs_proto_name);
51 EXPORT_SYMBOL(ip_vs_conn_new);
52 EXPORT_SYMBOL(ip_vs_conn_in_get);
53 EXPORT_SYMBOL(ip_vs_conn_out_get);
54 #ifdef CONFIG_IP_VS_PROTO_TCP
55 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
56 #endif
57 EXPORT_SYMBOL(ip_vs_conn_put);
58 #ifdef CONFIG_IP_VS_DEBUG
59 EXPORT_SYMBOL(ip_vs_get_debug_level);
60 #endif
61
62
63 /* ID used in ICMP lookups */
64 #define icmp_id(icmph)          (((icmph)->un).echo.id)
65
66 const char *ip_vs_proto_name(unsigned proto)
67 {
68         static char buf[20];
69
70         switch (proto) {
71         case IPPROTO_IP:
72                 return "IP";
73         case IPPROTO_UDP:
74                 return "UDP";
75         case IPPROTO_TCP:
76                 return "TCP";
77         case IPPROTO_ICMP:
78                 return "ICMP";
79         default:
80                 sprintf(buf, "IP_%d", proto);
81                 return buf;
82         }
83 }
84
85 void ip_vs_init_hash_table(struct list_head *table, int rows)
86 {
87         while (--rows >= 0)
88                 INIT_LIST_HEAD(&table[rows]);
89 }
90
91 static inline void
92 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
93 {
94         struct ip_vs_dest *dest = cp->dest;
95         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
96                 spin_lock(&dest->stats.lock);
97                 dest->stats.inpkts++;
98                 dest->stats.inbytes += skb->len;
99                 spin_unlock(&dest->stats.lock);
100
101                 spin_lock(&dest->svc->stats.lock);
102                 dest->svc->stats.inpkts++;
103                 dest->svc->stats.inbytes += skb->len;
104                 spin_unlock(&dest->svc->stats.lock);
105
106                 spin_lock(&ip_vs_stats.lock);
107                 ip_vs_stats.inpkts++;
108                 ip_vs_stats.inbytes += skb->len;
109                 spin_unlock(&ip_vs_stats.lock);
110         }
111 }
112
113
114 static inline void
115 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
116 {
117         struct ip_vs_dest *dest = cp->dest;
118         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
119                 spin_lock(&dest->stats.lock);
120                 dest->stats.outpkts++;
121                 dest->stats.outbytes += skb->len;
122                 spin_unlock(&dest->stats.lock);
123
124                 spin_lock(&dest->svc->stats.lock);
125                 dest->svc->stats.outpkts++;
126                 dest->svc->stats.outbytes += skb->len;
127                 spin_unlock(&dest->svc->stats.lock);
128
129                 spin_lock(&ip_vs_stats.lock);
130                 ip_vs_stats.outpkts++;
131                 ip_vs_stats.outbytes += skb->len;
132                 spin_unlock(&ip_vs_stats.lock);
133         }
134 }
135
136
137 static inline void
138 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
139 {
140         spin_lock(&cp->dest->stats.lock);
141         cp->dest->stats.conns++;
142         spin_unlock(&cp->dest->stats.lock);
143
144         spin_lock(&svc->stats.lock);
145         svc->stats.conns++;
146         spin_unlock(&svc->stats.lock);
147
148         spin_lock(&ip_vs_stats.lock);
149         ip_vs_stats.conns++;
150         spin_unlock(&ip_vs_stats.lock);
151 }
152
153
154 static inline int
155 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
156                 const struct sk_buff *skb,
157                 struct ip_vs_protocol *pp)
158 {
159         if (unlikely(!pp->state_transition))
160                 return 0;
161         return pp->state_transition(cp, direction, skb, pp);
162 }
163
164
165 /*
166  *  IPVS persistent scheduling function
167  *  It creates a connection entry according to its template if exists,
168  *  or selects a server and creates a connection entry plus a template.
169  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
170  *  Protocols supported: TCP, UDP
171  */
172 static struct ip_vs_conn *
173 ip_vs_sched_persist(struct ip_vs_service *svc,
174                     const struct sk_buff *skb,
175                     __be16 ports[2])
176 {
177         struct ip_vs_conn *cp = NULL;
178         struct iphdr *iph = ip_hdr(skb);
179         struct ip_vs_dest *dest;
180         struct ip_vs_conn *ct;
181         __be16  dport;   /* destination port to forward */
182         __be32  snet;    /* source network of the client, after masking */
183
184         /* Mask saddr with the netmask to adjust template granularity */
185         snet = iph->saddr & svc->netmask;
186
187         IP_VS_DBG(6, "p-schedule: src %u.%u.%u.%u:%u dest %u.%u.%u.%u:%u "
188                   "mnet %u.%u.%u.%u\n",
189                   NIPQUAD(iph->saddr), ntohs(ports[0]),
190                   NIPQUAD(iph->daddr), ntohs(ports[1]),
191                   NIPQUAD(snet));
192
193         /*
194          * As far as we know, FTP is a very complicated network protocol, and
195          * it uses control connection and data connections. For active FTP,
196          * FTP server initialize data connection to the client, its source port
197          * is often 20. For passive FTP, FTP server tells the clients the port
198          * that it passively listens to,  and the client issues the data
199          * connection. In the tunneling or direct routing mode, the load
200          * balancer is on the client-to-server half of connection, the port
201          * number is unknown to the load balancer. So, a conn template like
202          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
203          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
204          * is created for other persistent services.
205          */
206         if (ports[1] == svc->port) {
207                 /* Check if a template already exists */
208                 if (svc->port != FTPPORT)
209                         ct = ip_vs_ct_in_get(iph->protocol, snet, 0,
210                                                iph->daddr, ports[1]);
211                 else
212                         ct = ip_vs_ct_in_get(iph->protocol, snet, 0,
213                                                iph->daddr, 0);
214
215                 if (!ct || !ip_vs_check_template(ct)) {
216                         /*
217                          * No template found or the dest of the connection
218                          * template is not available.
219                          */
220                         dest = svc->scheduler->schedule(svc, skb);
221                         if (dest == NULL) {
222                                 IP_VS_DBG(1, "p-schedule: no dest found.\n");
223                                 return NULL;
224                         }
225
226                         /*
227                          * Create a template like <protocol,caddr,0,
228                          * vaddr,vport,daddr,dport> for non-ftp service,
229                          * and <protocol,caddr,0,vaddr,0,daddr,0>
230                          * for ftp service.
231                          */
232                         if (svc->port != FTPPORT)
233                                 ct = ip_vs_conn_new(iph->protocol,
234                                                     snet, 0,
235                                                     iph->daddr,
236                                                     ports[1],
237                                                     dest->addr, dest->port,
238                                                     IP_VS_CONN_F_TEMPLATE,
239                                                     dest);
240                         else
241                                 ct = ip_vs_conn_new(iph->protocol,
242                                                     snet, 0,
243                                                     iph->daddr, 0,
244                                                     dest->addr, 0,
245                                                     IP_VS_CONN_F_TEMPLATE,
246                                                     dest);
247                         if (ct == NULL)
248                                 return NULL;
249
250                         ct->timeout = svc->timeout;
251                 } else {
252                         /* set destination with the found template */
253                         dest = ct->dest;
254                 }
255                 dport = dest->port;
256         } else {
257                 /*
258                  * Note: persistent fwmark-based services and persistent
259                  * port zero service are handled here.
260                  * fwmark template: <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
261                  * port zero template: <protocol,caddr,0,vaddr,0,daddr,0>
262                  */
263                 if (svc->fwmark)
264                         ct = ip_vs_ct_in_get(IPPROTO_IP, snet, 0,
265                                                htonl(svc->fwmark), 0);
266                 else
267                         ct = ip_vs_ct_in_get(iph->protocol, snet, 0,
268                                                iph->daddr, 0);
269
270                 if (!ct || !ip_vs_check_template(ct)) {
271                         /*
272                          * If it is not persistent port zero, return NULL,
273                          * otherwise create a connection template.
274                          */
275                         if (svc->port)
276                                 return NULL;
277
278                         dest = svc->scheduler->schedule(svc, skb);
279                         if (dest == NULL) {
280                                 IP_VS_DBG(1, "p-schedule: no dest found.\n");
281                                 return NULL;
282                         }
283
284                         /*
285                          * Create a template according to the service
286                          */
287                         if (svc->fwmark)
288                                 ct = ip_vs_conn_new(IPPROTO_IP,
289                                                     snet, 0,
290                                                     htonl(svc->fwmark), 0,
291                                                     dest->addr, 0,
292                                                     IP_VS_CONN_F_TEMPLATE,
293                                                     dest);
294                         else
295                                 ct = ip_vs_conn_new(iph->protocol,
296                                                     snet, 0,
297                                                     iph->daddr, 0,
298                                                     dest->addr, 0,
299                                                     IP_VS_CONN_F_TEMPLATE,
300                                                     dest);
301                         if (ct == NULL)
302                                 return NULL;
303
304                         ct->timeout = svc->timeout;
305                 } else {
306                         /* set destination with the found template */
307                         dest = ct->dest;
308                 }
309                 dport = ports[1];
310         }
311
312         /*
313          *    Create a new connection according to the template
314          */
315         cp = ip_vs_conn_new(iph->protocol,
316                             iph->saddr, ports[0],
317                             iph->daddr, ports[1],
318                             dest->addr, dport,
319                             0,
320                             dest);
321         if (cp == NULL) {
322                 ip_vs_conn_put(ct);
323                 return NULL;
324         }
325
326         /*
327          *    Add its control
328          */
329         ip_vs_control_add(cp, ct);
330         ip_vs_conn_put(ct);
331
332         ip_vs_conn_stats(cp, svc);
333         return cp;
334 }
335
336
337 /*
338  *  IPVS main scheduling function
339  *  It selects a server according to the virtual service, and
340  *  creates a connection entry.
341  *  Protocols supported: TCP, UDP
342  */
343 struct ip_vs_conn *
344 ip_vs_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
345 {
346         struct ip_vs_conn *cp = NULL;
347         struct iphdr *iph = ip_hdr(skb);
348         struct ip_vs_dest *dest;
349         __be16 _ports[2], *pptr;
350
351         pptr = skb_header_pointer(skb, iph->ihl*4,
352                                   sizeof(_ports), _ports);
353         if (pptr == NULL)
354                 return NULL;
355
356         /*
357          *    Persistent service
358          */
359         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
360                 return ip_vs_sched_persist(svc, skb, pptr);
361
362         /*
363          *    Non-persistent service
364          */
365         if (!svc->fwmark && pptr[1] != svc->port) {
366                 if (!svc->port)
367                         IP_VS_ERR("Schedule: port zero only supported "
368                                   "in persistent services, "
369                                   "check your ipvs configuration\n");
370                 return NULL;
371         }
372
373         dest = svc->scheduler->schedule(svc, skb);
374         if (dest == NULL) {
375                 IP_VS_DBG(1, "Schedule: no dest found.\n");
376                 return NULL;
377         }
378
379         /*
380          *    Create a connection entry.
381          */
382         cp = ip_vs_conn_new(iph->protocol,
383                             iph->saddr, pptr[0],
384                             iph->daddr, pptr[1],
385                             dest->addr, dest->port?dest->port:pptr[1],
386                             0,
387                             dest);
388         if (cp == NULL)
389                 return NULL;
390
391         IP_VS_DBG(6, "Schedule fwd:%c c:%u.%u.%u.%u:%u v:%u.%u.%u.%u:%u "
392                   "d:%u.%u.%u.%u:%u conn->flags:%X conn->refcnt:%d\n",
393                   ip_vs_fwd_tag(cp),
394                   NIPQUAD(cp->caddr), ntohs(cp->cport),
395                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
396                   NIPQUAD(cp->daddr), ntohs(cp->dport),
397                   cp->flags, atomic_read(&cp->refcnt));
398
399         ip_vs_conn_stats(cp, svc);
400         return cp;
401 }
402
403
404 /*
405  *  Pass or drop the packet.
406  *  Called by ip_vs_in, when the virtual service is available but
407  *  no destination is available for a new connection.
408  */
409 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
410                 struct ip_vs_protocol *pp)
411 {
412         __be16 _ports[2], *pptr;
413         struct iphdr *iph = ip_hdr(skb);
414
415         pptr = skb_header_pointer(skb, iph->ihl*4,
416                                   sizeof(_ports), _ports);
417         if (pptr == NULL) {
418                 ip_vs_service_put(svc);
419                 return NF_DROP;
420         }
421
422         /* if it is fwmark-based service, the cache_bypass sysctl is up
423            and the destination is RTN_UNICAST (and not local), then create
424            a cache_bypass connection entry */
425         if (sysctl_ip_vs_cache_bypass && svc->fwmark
426             && (inet_addr_type(iph->daddr) == RTN_UNICAST)) {
427                 int ret, cs;
428                 struct ip_vs_conn *cp;
429
430                 ip_vs_service_put(svc);
431
432                 /* create a new connection entry */
433                 IP_VS_DBG(6, "ip_vs_leave: create a cache_bypass entry\n");
434                 cp = ip_vs_conn_new(iph->protocol,
435                                     iph->saddr, pptr[0],
436                                     iph->daddr, pptr[1],
437                                     0, 0,
438                                     IP_VS_CONN_F_BYPASS,
439                                     NULL);
440                 if (cp == NULL)
441                         return NF_DROP;
442
443                 /* statistics */
444                 ip_vs_in_stats(cp, skb);
445
446                 /* set state */
447                 cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
448
449                 /* transmit the first SYN packet */
450                 ret = cp->packet_xmit(skb, cp, pp);
451                 /* do not touch skb anymore */
452
453                 atomic_inc(&cp->in_pkts);
454                 ip_vs_conn_put(cp);
455                 return ret;
456         }
457
458         /*
459          * When the virtual ftp service is presented, packets destined
460          * for other services on the VIP may get here (except services
461          * listed in the ipvs table), pass the packets, because it is
462          * not ipvs job to decide to drop the packets.
463          */
464         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
465                 ip_vs_service_put(svc);
466                 return NF_ACCEPT;
467         }
468
469         ip_vs_service_put(svc);
470
471         /*
472          * Notify the client that the destination is unreachable, and
473          * release the socket buffer.
474          * Since it is in IP layer, the TCP socket is not actually
475          * created, the TCP RST packet cannot be sent, instead that
476          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
477          */
478         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
479         return NF_DROP;
480 }
481
482
483 /*
484  *      It is hooked before NF_IP_PRI_NAT_SRC at the NF_IP_POST_ROUTING
485  *      chain, and is used for VS/NAT.
486  *      It detects packets for VS/NAT connections and sends the packets
487  *      immediately. This can avoid that iptable_nat mangles the packets
488  *      for VS/NAT.
489  */
490 static unsigned int ip_vs_post_routing(unsigned int hooknum,
491                                        struct sk_buff *skb,
492                                        const struct net_device *in,
493                                        const struct net_device *out,
494                                        int (*okfn)(struct sk_buff *))
495 {
496         if (!skb->ipvs_property)
497                 return NF_ACCEPT;
498         /* The packet was sent from IPVS, exit this chain */
499         return NF_STOP;
500 }
501
502 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
503 {
504         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
505 }
506
507 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
508 {
509         int err = ip_defrag(skb, user);
510
511         if (!err)
512                 ip_send_check(ip_hdr(skb));
513
514         return err;
515 }
516
517 /*
518  * Packet has been made sufficiently writable in caller
519  * - inout: 1=in->out, 0=out->in
520  */
521 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
522                     struct ip_vs_conn *cp, int inout)
523 {
524         struct iphdr *iph        = ip_hdr(skb);
525         unsigned int icmp_offset = iph->ihl*4;
526         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
527                                                       icmp_offset);
528         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
529
530         if (inout) {
531                 iph->saddr = cp->vaddr;
532                 ip_send_check(iph);
533                 ciph->daddr = cp->vaddr;
534                 ip_send_check(ciph);
535         } else {
536                 iph->daddr = cp->daddr;
537                 ip_send_check(iph);
538                 ciph->saddr = cp->daddr;
539                 ip_send_check(ciph);
540         }
541
542         /* the TCP/UDP port */
543         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol) {
544                 __be16 *ports = (void *)ciph + ciph->ihl*4;
545
546                 if (inout)
547                         ports[1] = cp->vport;
548                 else
549                         ports[0] = cp->dport;
550         }
551
552         /* And finally the ICMP checksum */
553         icmph->checksum = 0;
554         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
555         skb->ip_summed = CHECKSUM_UNNECESSARY;
556
557         if (inout)
558                 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
559                         "Forwarding altered outgoing ICMP");
560         else
561                 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
562                         "Forwarding altered incoming ICMP");
563 }
564
565 /*
566  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
567  *      Find any that might be relevant, check against existing connections,
568  *      forward to the right destination host if relevant.
569  *      Currently handles error types - unreachable, quench, ttl exceeded.
570  *      (Only used in VS/NAT)
571  */
572 static int ip_vs_out_icmp(struct sk_buff *skb, int *related)
573 {
574         struct iphdr *iph;
575         struct icmphdr  _icmph, *ic;
576         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
577         struct ip_vs_conn *cp;
578         struct ip_vs_protocol *pp;
579         unsigned int offset, ihl, verdict;
580
581         *related = 1;
582
583         /* reassemble IP fragments */
584         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
585                 if (ip_vs_gather_frags(skb, IP_DEFRAG_VS_OUT))
586                         return NF_STOLEN;
587         }
588
589         iph = ip_hdr(skb);
590         offset = ihl = iph->ihl * 4;
591         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
592         if (ic == NULL)
593                 return NF_DROP;
594
595         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
596                   ic->type, ntohs(icmp_id(ic)),
597                   NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
598
599         /*
600          * Work through seeing if this is for us.
601          * These checks are supposed to be in an order that means easy
602          * things are checked first to speed up processing.... however
603          * this means that some packets will manage to get a long way
604          * down this stack and then be rejected, but that's life.
605          */
606         if ((ic->type != ICMP_DEST_UNREACH) &&
607             (ic->type != ICMP_SOURCE_QUENCH) &&
608             (ic->type != ICMP_TIME_EXCEEDED)) {
609                 *related = 0;
610                 return NF_ACCEPT;
611         }
612
613         /* Now find the contained IP header */
614         offset += sizeof(_icmph);
615         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
616         if (cih == NULL)
617                 return NF_ACCEPT; /* The packet looks wrong, ignore */
618
619         pp = ip_vs_proto_get(cih->protocol);
620         if (!pp)
621                 return NF_ACCEPT;
622
623         /* Is the embedded protocol header present? */
624         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
625                      pp->dont_defrag))
626                 return NF_ACCEPT;
627
628         IP_VS_DBG_PKT(11, pp, skb, offset, "Checking outgoing ICMP for");
629
630         offset += cih->ihl * 4;
631
632         /* The embedded headers contain source and dest in reverse order */
633         cp = pp->conn_out_get(skb, pp, cih, offset, 1);
634         if (!cp)
635                 return NF_ACCEPT;
636
637         verdict = NF_DROP;
638
639         if (IP_VS_FWD_METHOD(cp) != 0) {
640                 IP_VS_ERR("shouldn't reach here, because the box is on the"
641                           "half connection in the tun/dr module.\n");
642         }
643
644         /* Ensure the checksum is correct */
645         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
646                 /* Failed checksum! */
647                 IP_VS_DBG(1, "Forward ICMP: failed checksum from %d.%d.%d.%d!\n",
648                           NIPQUAD(iph->saddr));
649                 goto out;
650         }
651
652         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
653                 offset += 2 * sizeof(__u16);
654         if (!skb_make_writable(skb, offset))
655                 goto out;
656
657         ip_vs_nat_icmp(skb, pp, cp, 1);
658
659         /* do the statistics and put it back */
660         ip_vs_out_stats(cp, skb);
661
662         skb->ipvs_property = 1;
663         verdict = NF_ACCEPT;
664
665   out:
666         __ip_vs_conn_put(cp);
667
668         return verdict;
669 }
670
671 static inline int is_tcp_reset(const struct sk_buff *skb)
672 {
673         struct tcphdr _tcph, *th;
674
675         th = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_tcph), &_tcph);
676         if (th == NULL)
677                 return 0;
678         return th->rst;
679 }
680
681 /*
682  *      It is hooked at the NF_IP_FORWARD chain, used only for VS/NAT.
683  *      Check if outgoing packet belongs to the established ip_vs_conn,
684  *      rewrite addresses of the packet and send it on its way...
685  */
686 static unsigned int
687 ip_vs_out(unsigned int hooknum, struct sk_buff *skb,
688           const struct net_device *in, const struct net_device *out,
689           int (*okfn)(struct sk_buff *))
690 {
691         struct iphdr    *iph;
692         struct ip_vs_protocol *pp;
693         struct ip_vs_conn *cp;
694         int ihl;
695
696         EnterFunction(11);
697
698         if (skb->ipvs_property)
699                 return NF_ACCEPT;
700
701         iph = ip_hdr(skb);
702         if (unlikely(iph->protocol == IPPROTO_ICMP)) {
703                 int related, verdict = ip_vs_out_icmp(skb, &related);
704
705                 if (related)
706                         return verdict;
707                 iph = ip_hdr(skb);
708         }
709
710         pp = ip_vs_proto_get(iph->protocol);
711         if (unlikely(!pp))
712                 return NF_ACCEPT;
713
714         /* reassemble IP fragments */
715         if (unlikely(iph->frag_off & htons(IP_MF|IP_OFFSET) &&
716                      !pp->dont_defrag)) {
717                 if (ip_vs_gather_frags(skb, IP_DEFRAG_VS_OUT))
718                         return NF_STOLEN;
719                 iph = ip_hdr(skb);
720         }
721
722         ihl = iph->ihl << 2;
723
724         /*
725          * Check if the packet belongs to an existing entry
726          */
727         cp = pp->conn_out_get(skb, pp, iph, ihl, 0);
728
729         if (unlikely(!cp)) {
730                 if (sysctl_ip_vs_nat_icmp_send &&
731                     (pp->protocol == IPPROTO_TCP ||
732                      pp->protocol == IPPROTO_UDP)) {
733                         __be16 _ports[2], *pptr;
734
735                         pptr = skb_header_pointer(skb, ihl,
736                                                   sizeof(_ports), _ports);
737                         if (pptr == NULL)
738                                 return NF_ACCEPT;       /* Not for me */
739                         if (ip_vs_lookup_real_service(iph->protocol,
740                                                       iph->saddr, pptr[0])) {
741                                 /*
742                                  * Notify the real server: there is no
743                                  * existing entry if it is not RST
744                                  * packet or not TCP packet.
745                                  */
746                                 if (iph->protocol != IPPROTO_TCP
747                                     || !is_tcp_reset(skb)) {
748                                         icmp_send(skb,ICMP_DEST_UNREACH,
749                                                   ICMP_PORT_UNREACH, 0);
750                                         return NF_DROP;
751                                 }
752                         }
753                 }
754                 IP_VS_DBG_PKT(12, pp, skb, 0,
755                               "packet continues traversal as normal");
756                 return NF_ACCEPT;
757         }
758
759         IP_VS_DBG_PKT(11, pp, skb, 0, "Outgoing packet");
760
761         if (!skb_make_writable(skb, ihl))
762                 goto drop;
763
764         /* mangle the packet */
765         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp))
766                 goto drop;
767         ip_hdr(skb)->saddr = cp->vaddr;
768         ip_send_check(ip_hdr(skb));
769
770         /* For policy routing, packets originating from this
771          * machine itself may be routed differently to packets
772          * passing through.  We want this packet to be routed as
773          * if it came from this machine itself.  So re-compute
774          * the routing information.
775          */
776         if (ip_route_me_harder(skb, RTN_LOCAL) != 0)
777                 goto drop;
778
779         IP_VS_DBG_PKT(10, pp, skb, 0, "After SNAT");
780
781         ip_vs_out_stats(cp, skb);
782         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pp);
783         ip_vs_conn_put(cp);
784
785         skb->ipvs_property = 1;
786
787         LeaveFunction(11);
788         return NF_ACCEPT;
789
790   drop:
791         ip_vs_conn_put(cp);
792         kfree_skb(skb);
793         return NF_STOLEN;
794 }
795
796
797 /*
798  *      Handle ICMP messages in the outside-to-inside direction (incoming).
799  *      Find any that might be relevant, check against existing connections,
800  *      forward to the right destination host if relevant.
801  *      Currently handles error types - unreachable, quench, ttl exceeded.
802  */
803 static int
804 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
805 {
806         struct iphdr *iph;
807         struct icmphdr  _icmph, *ic;
808         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
809         struct ip_vs_conn *cp;
810         struct ip_vs_protocol *pp;
811         unsigned int offset, ihl, verdict;
812
813         *related = 1;
814
815         /* reassemble IP fragments */
816         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
817                 if (ip_vs_gather_frags(skb, hooknum == NF_IP_LOCAL_IN ?
818                                             IP_DEFRAG_VS_IN : IP_DEFRAG_VS_FWD))
819                         return NF_STOLEN;
820         }
821
822         iph = ip_hdr(skb);
823         offset = ihl = iph->ihl * 4;
824         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
825         if (ic == NULL)
826                 return NF_DROP;
827
828         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
829                   ic->type, ntohs(icmp_id(ic)),
830                   NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
831
832         /*
833          * Work through seeing if this is for us.
834          * These checks are supposed to be in an order that means easy
835          * things are checked first to speed up processing.... however
836          * this means that some packets will manage to get a long way
837          * down this stack and then be rejected, but that's life.
838          */
839         if ((ic->type != ICMP_DEST_UNREACH) &&
840             (ic->type != ICMP_SOURCE_QUENCH) &&
841             (ic->type != ICMP_TIME_EXCEEDED)) {
842                 *related = 0;
843                 return NF_ACCEPT;
844         }
845
846         /* Now find the contained IP header */
847         offset += sizeof(_icmph);
848         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
849         if (cih == NULL)
850                 return NF_ACCEPT; /* The packet looks wrong, ignore */
851
852         pp = ip_vs_proto_get(cih->protocol);
853         if (!pp)
854                 return NF_ACCEPT;
855
856         /* Is the embedded protocol header present? */
857         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
858                      pp->dont_defrag))
859                 return NF_ACCEPT;
860
861         IP_VS_DBG_PKT(11, pp, skb, offset, "Checking incoming ICMP for");
862
863         offset += cih->ihl * 4;
864
865         /* The embedded headers contain source and dest in reverse order */
866         cp = pp->conn_in_get(skb, pp, cih, offset, 1);
867         if (!cp)
868                 return NF_ACCEPT;
869
870         verdict = NF_DROP;
871
872         /* Ensure the checksum is correct */
873         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
874                 /* Failed checksum! */
875                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %d.%d.%d.%d!\n",
876                           NIPQUAD(iph->saddr));
877                 goto out;
878         }
879
880         /* do the statistics and put it back */
881         ip_vs_in_stats(cp, skb);
882         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
883                 offset += 2 * sizeof(__u16);
884         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
885         /* do not touch skb anymore */
886
887   out:
888         __ip_vs_conn_put(cp);
889
890         return verdict;
891 }
892
893 /*
894  *      Check if it's for virtual services, look it up,
895  *      and send it on its way...
896  */
897 static unsigned int
898 ip_vs_in(unsigned int hooknum, struct sk_buff *skb,
899          const struct net_device *in, const struct net_device *out,
900          int (*okfn)(struct sk_buff *))
901 {
902         struct iphdr    *iph;
903         struct ip_vs_protocol *pp;
904         struct ip_vs_conn *cp;
905         int ret, restart;
906         int ihl;
907
908         /*
909          *      Big tappo: only PACKET_HOST (neither loopback nor mcasts)
910          *      ... don't know why 1st test DOES NOT include 2nd (?)
911          */
912         if (unlikely(skb->pkt_type != PACKET_HOST
913                      || skb->dev->flags & IFF_LOOPBACK || skb->sk)) {
914                 IP_VS_DBG(12, "packet type=%d proto=%d daddr=%d.%d.%d.%d ignored\n",
915                           skb->pkt_type,
916                           ip_hdr(skb)->protocol,
917                           NIPQUAD(ip_hdr(skb)->daddr));
918                 return NF_ACCEPT;
919         }
920
921         iph = ip_hdr(skb);
922         if (unlikely(iph->protocol == IPPROTO_ICMP)) {
923                 int related, verdict = ip_vs_in_icmp(skb, &related, hooknum);
924
925                 if (related)
926                         return verdict;
927                 iph = ip_hdr(skb);
928         }
929
930         /* Protocol supported? */
931         pp = ip_vs_proto_get(iph->protocol);
932         if (unlikely(!pp))
933                 return NF_ACCEPT;
934
935         ihl = iph->ihl << 2;
936
937         /*
938          * Check if the packet belongs to an existing connection entry
939          */
940         cp = pp->conn_in_get(skb, pp, iph, ihl, 0);
941
942         if (unlikely(!cp)) {
943                 int v;
944
945                 if (!pp->conn_schedule(skb, pp, &v, &cp))
946                         return v;
947         }
948
949         if (unlikely(!cp)) {
950                 /* sorry, all this trouble for a no-hit :) */
951                 IP_VS_DBG_PKT(12, pp, skb, 0,
952                               "packet continues traversal as normal");
953                 return NF_ACCEPT;
954         }
955
956         IP_VS_DBG_PKT(11, pp, skb, 0, "Incoming packet");
957
958         /* Check the server status */
959         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
960                 /* the destination server is not available */
961
962                 if (sysctl_ip_vs_expire_nodest_conn) {
963                         /* try to expire the connection immediately */
964                         ip_vs_conn_expire_now(cp);
965                 }
966                 /* don't restart its timer, and silently
967                    drop the packet. */
968                 __ip_vs_conn_put(cp);
969                 return NF_DROP;
970         }
971
972         ip_vs_in_stats(cp, skb);
973         restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
974         if (cp->packet_xmit)
975                 ret = cp->packet_xmit(skb, cp, pp);
976                 /* do not touch skb anymore */
977         else {
978                 IP_VS_DBG_RL("warning: packet_xmit is null");
979                 ret = NF_ACCEPT;
980         }
981
982         /* Increase its packet counter and check if it is needed
983          * to be synchronized
984          *
985          * Sync connection if it is about to close to
986          * encorage the standby servers to update the connections timeout
987          */
988         atomic_inc(&cp->in_pkts);
989         if ((ip_vs_sync_state & IP_VS_STATE_MASTER) &&
990             (((cp->protocol != IPPROTO_TCP ||
991                cp->state == IP_VS_TCP_S_ESTABLISHED) &&
992               (atomic_read(&cp->in_pkts) % sysctl_ip_vs_sync_threshold[1]
993                == sysctl_ip_vs_sync_threshold[0])) ||
994              ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) &&
995               ((cp->state == IP_VS_TCP_S_FIN_WAIT) ||
996                (cp->state == IP_VS_TCP_S_CLOSE)))))
997                 ip_vs_sync_conn(cp);
998         cp->old_state = cp->state;
999
1000         ip_vs_conn_put(cp);
1001         return ret;
1002 }
1003
1004
1005 /*
1006  *      It is hooked at the NF_IP_FORWARD chain, in order to catch ICMP
1007  *      related packets destined for 0.0.0.0/0.
1008  *      When fwmark-based virtual service is used, such as transparent
1009  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1010  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1011  *      sent to ip_vs_in_icmp. So, catch them at the NF_IP_FORWARD chain
1012  *      and send them to ip_vs_in_icmp.
1013  */
1014 static unsigned int
1015 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1016                    const struct net_device *in, const struct net_device *out,
1017                    int (*okfn)(struct sk_buff *))
1018 {
1019         int r;
1020
1021         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1022                 return NF_ACCEPT;
1023
1024         return ip_vs_in_icmp(skb, &r, hooknum);
1025 }
1026
1027
1028 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1029    or VS/NAT(change destination), so that filtering rules can be
1030    applied to IPVS. */
1031 static struct nf_hook_ops ip_vs_in_ops = {
1032         .hook           = ip_vs_in,
1033         .owner          = THIS_MODULE,
1034         .pf             = PF_INET,
1035         .hooknum        = NF_IP_LOCAL_IN,
1036         .priority       = 100,
1037 };
1038
1039 /* After packet filtering, change source only for VS/NAT */
1040 static struct nf_hook_ops ip_vs_out_ops = {
1041         .hook           = ip_vs_out,
1042         .owner          = THIS_MODULE,
1043         .pf             = PF_INET,
1044         .hooknum        = NF_IP_FORWARD,
1045         .priority       = 100,
1046 };
1047
1048 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1049    destined for 0.0.0.0/0, which is for incoming IPVS connections */
1050 static struct nf_hook_ops ip_vs_forward_icmp_ops = {
1051         .hook           = ip_vs_forward_icmp,
1052         .owner          = THIS_MODULE,
1053         .pf             = PF_INET,
1054         .hooknum        = NF_IP_FORWARD,
1055         .priority       = 99,
1056 };
1057
1058 /* Before the netfilter connection tracking, exit from POST_ROUTING */
1059 static struct nf_hook_ops ip_vs_post_routing_ops = {
1060         .hook           = ip_vs_post_routing,
1061         .owner          = THIS_MODULE,
1062         .pf             = PF_INET,
1063         .hooknum        = NF_IP_POST_ROUTING,
1064         .priority       = NF_IP_PRI_NAT_SRC-1,
1065 };
1066
1067
1068 /*
1069  *      Initialize IP Virtual Server
1070  */
1071 static int __init ip_vs_init(void)
1072 {
1073         int ret;
1074
1075         ret = ip_vs_control_init();
1076         if (ret < 0) {
1077                 IP_VS_ERR("can't setup control.\n");
1078                 goto cleanup_nothing;
1079         }
1080
1081         ip_vs_protocol_init();
1082
1083         ret = ip_vs_app_init();
1084         if (ret < 0) {
1085                 IP_VS_ERR("can't setup application helper.\n");
1086                 goto cleanup_protocol;
1087         }
1088
1089         ret = ip_vs_conn_init();
1090         if (ret < 0) {
1091                 IP_VS_ERR("can't setup connection table.\n");
1092                 goto cleanup_app;
1093         }
1094
1095         ret = nf_register_hook(&ip_vs_in_ops);
1096         if (ret < 0) {
1097                 IP_VS_ERR("can't register in hook.\n");
1098                 goto cleanup_conn;
1099         }
1100
1101         ret = nf_register_hook(&ip_vs_out_ops);
1102         if (ret < 0) {
1103                 IP_VS_ERR("can't register out hook.\n");
1104                 goto cleanup_inops;
1105         }
1106         ret = nf_register_hook(&ip_vs_post_routing_ops);
1107         if (ret < 0) {
1108                 IP_VS_ERR("can't register post_routing hook.\n");
1109                 goto cleanup_outops;
1110         }
1111         ret = nf_register_hook(&ip_vs_forward_icmp_ops);
1112         if (ret < 0) {
1113                 IP_VS_ERR("can't register forward_icmp hook.\n");
1114                 goto cleanup_postroutingops;
1115         }
1116
1117         IP_VS_INFO("ipvs loaded.\n");
1118         return ret;
1119
1120   cleanup_postroutingops:
1121         nf_unregister_hook(&ip_vs_post_routing_ops);
1122   cleanup_outops:
1123         nf_unregister_hook(&ip_vs_out_ops);
1124   cleanup_inops:
1125         nf_unregister_hook(&ip_vs_in_ops);
1126   cleanup_conn:
1127         ip_vs_conn_cleanup();
1128   cleanup_app:
1129         ip_vs_app_cleanup();
1130   cleanup_protocol:
1131         ip_vs_protocol_cleanup();
1132         ip_vs_control_cleanup();
1133   cleanup_nothing:
1134         return ret;
1135 }
1136
1137 static void __exit ip_vs_cleanup(void)
1138 {
1139         nf_unregister_hook(&ip_vs_forward_icmp_ops);
1140         nf_unregister_hook(&ip_vs_post_routing_ops);
1141         nf_unregister_hook(&ip_vs_out_ops);
1142         nf_unregister_hook(&ip_vs_in_ops);
1143         ip_vs_conn_cleanup();
1144         ip_vs_app_cleanup();
1145         ip_vs_protocol_cleanup();
1146         ip_vs_control_cleanup();
1147         IP_VS_INFO("ipvs unloaded.\n");
1148 }
1149
1150 module_init(ip_vs_init);
1151 module_exit(ip_vs_cleanup);
1152 MODULE_LICENSE("GPL");