[SK_BUFF]: Introduce skb_network_header()
[linux-2.6.git] / 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 EXPORT_SYMBOL(ip_vs_make_skb_writable);
62
63
64 /* ID used in ICMP lookups */
65 #define icmp_id(icmph)          (((icmph)->un).echo.id)
66
67 const char *ip_vs_proto_name(unsigned proto)
68 {
69         static char buf[20];
70
71         switch (proto) {
72         case IPPROTO_IP:
73                 return "IP";
74         case IPPROTO_UDP:
75                 return "UDP";
76         case IPPROTO_TCP:
77                 return "TCP";
78         case IPPROTO_ICMP:
79                 return "ICMP";
80         default:
81                 sprintf(buf, "IP_%d", proto);
82                 return buf;
83         }
84 }
85
86 void ip_vs_init_hash_table(struct list_head *table, int rows)
87 {
88         while (--rows >= 0)
89                 INIT_LIST_HEAD(&table[rows]);
90 }
91
92 static inline void
93 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
94 {
95         struct ip_vs_dest *dest = cp->dest;
96         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
97                 spin_lock(&dest->stats.lock);
98                 dest->stats.inpkts++;
99                 dest->stats.inbytes += skb->len;
100                 spin_unlock(&dest->stats.lock);
101
102                 spin_lock(&dest->svc->stats.lock);
103                 dest->svc->stats.inpkts++;
104                 dest->svc->stats.inbytes += skb->len;
105                 spin_unlock(&dest->svc->stats.lock);
106
107                 spin_lock(&ip_vs_stats.lock);
108                 ip_vs_stats.inpkts++;
109                 ip_vs_stats.inbytes += skb->len;
110                 spin_unlock(&ip_vs_stats.lock);
111         }
112 }
113
114
115 static inline void
116 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
117 {
118         struct ip_vs_dest *dest = cp->dest;
119         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
120                 spin_lock(&dest->stats.lock);
121                 dest->stats.outpkts++;
122                 dest->stats.outbytes += skb->len;
123                 spin_unlock(&dest->stats.lock);
124
125                 spin_lock(&dest->svc->stats.lock);
126                 dest->svc->stats.outpkts++;
127                 dest->svc->stats.outbytes += skb->len;
128                 spin_unlock(&dest->svc->stats.lock);
129
130                 spin_lock(&ip_vs_stats.lock);
131                 ip_vs_stats.outpkts++;
132                 ip_vs_stats.outbytes += skb->len;
133                 spin_unlock(&ip_vs_stats.lock);
134         }
135 }
136
137
138 static inline void
139 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
140 {
141         spin_lock(&cp->dest->stats.lock);
142         cp->dest->stats.conns++;
143         spin_unlock(&cp->dest->stats.lock);
144
145         spin_lock(&svc->stats.lock);
146         svc->stats.conns++;
147         spin_unlock(&svc->stats.lock);
148
149         spin_lock(&ip_vs_stats.lock);
150         ip_vs_stats.conns++;
151         spin_unlock(&ip_vs_stats.lock);
152 }
153
154
155 static inline int
156 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
157                 const struct sk_buff *skb,
158                 struct ip_vs_protocol *pp)
159 {
160         if (unlikely(!pp->state_transition))
161                 return 0;
162         return pp->state_transition(cp, direction, skb, pp);
163 }
164
165
166 int ip_vs_make_skb_writable(struct sk_buff **pskb, int writable_len)
167 {
168         struct sk_buff *skb = *pskb;
169
170         /* skb is already used, better copy skb and its payload */
171         if (unlikely(skb_shared(skb) || skb->sk))
172                 goto copy_skb;
173
174         /* skb data is already used, copy it */
175         if (unlikely(skb_cloned(skb)))
176                 goto copy_data;
177
178         return pskb_may_pull(skb, writable_len);
179
180   copy_data:
181         if (unlikely(writable_len > skb->len))
182                 return 0;
183         return !pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
184
185   copy_skb:
186         if (unlikely(writable_len > skb->len))
187                 return 0;
188         skb = skb_copy(skb, GFP_ATOMIC);
189         if (!skb)
190                 return 0;
191         BUG_ON(skb_is_nonlinear(skb));
192
193         /* Rest of kernel will get very unhappy if we pass it a
194            suddenly-orphaned skbuff */
195         if ((*pskb)->sk)
196                 skb_set_owner_w(skb, (*pskb)->sk);
197         kfree_skb(*pskb);
198         *pskb = skb;
199         return 1;
200 }
201
202 /*
203  *  IPVS persistent scheduling function
204  *  It creates a connection entry according to its template if exists,
205  *  or selects a server and creates a connection entry plus a template.
206  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
207  *  Protocols supported: TCP, UDP
208  */
209 static struct ip_vs_conn *
210 ip_vs_sched_persist(struct ip_vs_service *svc,
211                     const struct sk_buff *skb,
212                     __be16 ports[2])
213 {
214         struct ip_vs_conn *cp = NULL;
215         struct iphdr *iph = skb->nh.iph;
216         struct ip_vs_dest *dest;
217         struct ip_vs_conn *ct;
218         __be16  dport;   /* destination port to forward */
219         __be32  snet;    /* source network of the client, after masking */
220
221         /* Mask saddr with the netmask to adjust template granularity */
222         snet = iph->saddr & svc->netmask;
223
224         IP_VS_DBG(6, "p-schedule: src %u.%u.%u.%u:%u dest %u.%u.%u.%u:%u "
225                   "mnet %u.%u.%u.%u\n",
226                   NIPQUAD(iph->saddr), ntohs(ports[0]),
227                   NIPQUAD(iph->daddr), ntohs(ports[1]),
228                   NIPQUAD(snet));
229
230         /*
231          * As far as we know, FTP is a very complicated network protocol, and
232          * it uses control connection and data connections. For active FTP,
233          * FTP server initialize data connection to the client, its source port
234          * is often 20. For passive FTP, FTP server tells the clients the port
235          * that it passively listens to,  and the client issues the data
236          * connection. In the tunneling or direct routing mode, the load
237          * balancer is on the client-to-server half of connection, the port
238          * number is unknown to the load balancer. So, a conn template like
239          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
240          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
241          * is created for other persistent services.
242          */
243         if (ports[1] == svc->port) {
244                 /* Check if a template already exists */
245                 if (svc->port != FTPPORT)
246                         ct = ip_vs_ct_in_get(iph->protocol, snet, 0,
247                                                iph->daddr, ports[1]);
248                 else
249                         ct = ip_vs_ct_in_get(iph->protocol, snet, 0,
250                                                iph->daddr, 0);
251
252                 if (!ct || !ip_vs_check_template(ct)) {
253                         /*
254                          * No template found or the dest of the connection
255                          * template is not available.
256                          */
257                         dest = svc->scheduler->schedule(svc, skb);
258                         if (dest == NULL) {
259                                 IP_VS_DBG(1, "p-schedule: no dest found.\n");
260                                 return NULL;
261                         }
262
263                         /*
264                          * Create a template like <protocol,caddr,0,
265                          * vaddr,vport,daddr,dport> for non-ftp service,
266                          * and <protocol,caddr,0,vaddr,0,daddr,0>
267                          * for ftp service.
268                          */
269                         if (svc->port != FTPPORT)
270                                 ct = ip_vs_conn_new(iph->protocol,
271                                                     snet, 0,
272                                                     iph->daddr,
273                                                     ports[1],
274                                                     dest->addr, dest->port,
275                                                     IP_VS_CONN_F_TEMPLATE,
276                                                     dest);
277                         else
278                                 ct = ip_vs_conn_new(iph->protocol,
279                                                     snet, 0,
280                                                     iph->daddr, 0,
281                                                     dest->addr, 0,
282                                                     IP_VS_CONN_F_TEMPLATE,
283                                                     dest);
284                         if (ct == NULL)
285                                 return NULL;
286
287                         ct->timeout = svc->timeout;
288                 } else {
289                         /* set destination with the found template */
290                         dest = ct->dest;
291                 }
292                 dport = dest->port;
293         } else {
294                 /*
295                  * Note: persistent fwmark-based services and persistent
296                  * port zero service are handled here.
297                  * fwmark template: <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
298                  * port zero template: <protocol,caddr,0,vaddr,0,daddr,0>
299                  */
300                 if (svc->fwmark)
301                         ct = ip_vs_ct_in_get(IPPROTO_IP, snet, 0,
302                                                htonl(svc->fwmark), 0);
303                 else
304                         ct = ip_vs_ct_in_get(iph->protocol, snet, 0,
305                                                iph->daddr, 0);
306
307                 if (!ct || !ip_vs_check_template(ct)) {
308                         /*
309                          * If it is not persistent port zero, return NULL,
310                          * otherwise create a connection template.
311                          */
312                         if (svc->port)
313                                 return NULL;
314
315                         dest = svc->scheduler->schedule(svc, skb);
316                         if (dest == NULL) {
317                                 IP_VS_DBG(1, "p-schedule: no dest found.\n");
318                                 return NULL;
319                         }
320
321                         /*
322                          * Create a template according to the service
323                          */
324                         if (svc->fwmark)
325                                 ct = ip_vs_conn_new(IPPROTO_IP,
326                                                     snet, 0,
327                                                     htonl(svc->fwmark), 0,
328                                                     dest->addr, 0,
329                                                     IP_VS_CONN_F_TEMPLATE,
330                                                     dest);
331                         else
332                                 ct = ip_vs_conn_new(iph->protocol,
333                                                     snet, 0,
334                                                     iph->daddr, 0,
335                                                     dest->addr, 0,
336                                                     IP_VS_CONN_F_TEMPLATE,
337                                                     dest);
338                         if (ct == NULL)
339                                 return NULL;
340
341                         ct->timeout = svc->timeout;
342                 } else {
343                         /* set destination with the found template */
344                         dest = ct->dest;
345                 }
346                 dport = ports[1];
347         }
348
349         /*
350          *    Create a new connection according to the template
351          */
352         cp = ip_vs_conn_new(iph->protocol,
353                             iph->saddr, ports[0],
354                             iph->daddr, ports[1],
355                             dest->addr, dport,
356                             0,
357                             dest);
358         if (cp == NULL) {
359                 ip_vs_conn_put(ct);
360                 return NULL;
361         }
362
363         /*
364          *    Add its control
365          */
366         ip_vs_control_add(cp, ct);
367         ip_vs_conn_put(ct);
368
369         ip_vs_conn_stats(cp, svc);
370         return cp;
371 }
372
373
374 /*
375  *  IPVS main scheduling function
376  *  It selects a server according to the virtual service, and
377  *  creates a connection entry.
378  *  Protocols supported: TCP, UDP
379  */
380 struct ip_vs_conn *
381 ip_vs_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
382 {
383         struct ip_vs_conn *cp = NULL;
384         struct iphdr *iph = skb->nh.iph;
385         struct ip_vs_dest *dest;
386         __be16 _ports[2], *pptr;
387
388         pptr = skb_header_pointer(skb, iph->ihl*4,
389                                   sizeof(_ports), _ports);
390         if (pptr == NULL)
391                 return NULL;
392
393         /*
394          *    Persistent service
395          */
396         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
397                 return ip_vs_sched_persist(svc, skb, pptr);
398
399         /*
400          *    Non-persistent service
401          */
402         if (!svc->fwmark && pptr[1] != svc->port) {
403                 if (!svc->port)
404                         IP_VS_ERR("Schedule: port zero only supported "
405                                   "in persistent services, "
406                                   "check your ipvs configuration\n");
407                 return NULL;
408         }
409
410         dest = svc->scheduler->schedule(svc, skb);
411         if (dest == NULL) {
412                 IP_VS_DBG(1, "Schedule: no dest found.\n");
413                 return NULL;
414         }
415
416         /*
417          *    Create a connection entry.
418          */
419         cp = ip_vs_conn_new(iph->protocol,
420                             iph->saddr, pptr[0],
421                             iph->daddr, pptr[1],
422                             dest->addr, dest->port?dest->port:pptr[1],
423                             0,
424                             dest);
425         if (cp == NULL)
426                 return NULL;
427
428         IP_VS_DBG(6, "Schedule fwd:%c c:%u.%u.%u.%u:%u v:%u.%u.%u.%u:%u "
429                   "d:%u.%u.%u.%u:%u conn->flags:%X conn->refcnt:%d\n",
430                   ip_vs_fwd_tag(cp),
431                   NIPQUAD(cp->caddr), ntohs(cp->cport),
432                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
433                   NIPQUAD(cp->daddr), ntohs(cp->dport),
434                   cp->flags, atomic_read(&cp->refcnt));
435
436         ip_vs_conn_stats(cp, svc);
437         return cp;
438 }
439
440
441 /*
442  *  Pass or drop the packet.
443  *  Called by ip_vs_in, when the virtual service is available but
444  *  no destination is available for a new connection.
445  */
446 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
447                 struct ip_vs_protocol *pp)
448 {
449         __be16 _ports[2], *pptr;
450         struct iphdr *iph = skb->nh.iph;
451
452         pptr = skb_header_pointer(skb, iph->ihl*4,
453                                   sizeof(_ports), _ports);
454         if (pptr == NULL) {
455                 ip_vs_service_put(svc);
456                 return NF_DROP;
457         }
458
459         /* if it is fwmark-based service, the cache_bypass sysctl is up
460            and the destination is RTN_UNICAST (and not local), then create
461            a cache_bypass connection entry */
462         if (sysctl_ip_vs_cache_bypass && svc->fwmark
463             && (inet_addr_type(iph->daddr) == RTN_UNICAST)) {
464                 int ret, cs;
465                 struct ip_vs_conn *cp;
466
467                 ip_vs_service_put(svc);
468
469                 /* create a new connection entry */
470                 IP_VS_DBG(6, "ip_vs_leave: create a cache_bypass entry\n");
471                 cp = ip_vs_conn_new(iph->protocol,
472                                     iph->saddr, pptr[0],
473                                     iph->daddr, pptr[1],
474                                     0, 0,
475                                     IP_VS_CONN_F_BYPASS,
476                                     NULL);
477                 if (cp == NULL)
478                         return NF_DROP;
479
480                 /* statistics */
481                 ip_vs_in_stats(cp, skb);
482
483                 /* set state */
484                 cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
485
486                 /* transmit the first SYN packet */
487                 ret = cp->packet_xmit(skb, cp, pp);
488                 /* do not touch skb anymore */
489
490                 atomic_inc(&cp->in_pkts);
491                 ip_vs_conn_put(cp);
492                 return ret;
493         }
494
495         /*
496          * When the virtual ftp service is presented, packets destined
497          * for other services on the VIP may get here (except services
498          * listed in the ipvs table), pass the packets, because it is
499          * not ipvs job to decide to drop the packets.
500          */
501         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
502                 ip_vs_service_put(svc);
503                 return NF_ACCEPT;
504         }
505
506         ip_vs_service_put(svc);
507
508         /*
509          * Notify the client that the destination is unreachable, and
510          * release the socket buffer.
511          * Since it is in IP layer, the TCP socket is not actually
512          * created, the TCP RST packet cannot be sent, instead that
513          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
514          */
515         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
516         return NF_DROP;
517 }
518
519
520 /*
521  *      It is hooked before NF_IP_PRI_NAT_SRC at the NF_IP_POST_ROUTING
522  *      chain, and is used for VS/NAT.
523  *      It detects packets for VS/NAT connections and sends the packets
524  *      immediately. This can avoid that iptable_nat mangles the packets
525  *      for VS/NAT.
526  */
527 static unsigned int ip_vs_post_routing(unsigned int hooknum,
528                                        struct sk_buff **pskb,
529                                        const struct net_device *in,
530                                        const struct net_device *out,
531                                        int (*okfn)(struct sk_buff *))
532 {
533         if (!((*pskb)->ipvs_property))
534                 return NF_ACCEPT;
535         /* The packet was sent from IPVS, exit this chain */
536         return NF_STOP;
537 }
538
539 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
540 {
541         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
542 }
543
544 static inline struct sk_buff *
545 ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
546 {
547         skb = ip_defrag(skb, user);
548         if (skb)
549                 ip_send_check(skb->nh.iph);
550         return skb;
551 }
552
553 /*
554  * Packet has been made sufficiently writable in caller
555  * - inout: 1=in->out, 0=out->in
556  */
557 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
558                     struct ip_vs_conn *cp, int inout)
559 {
560         struct iphdr *iph        = skb->nh.iph;
561         unsigned int icmp_offset = iph->ihl*4;
562         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
563                                                       icmp_offset);
564         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
565
566         if (inout) {
567                 iph->saddr = cp->vaddr;
568                 ip_send_check(iph);
569                 ciph->daddr = cp->vaddr;
570                 ip_send_check(ciph);
571         } else {
572                 iph->daddr = cp->daddr;
573                 ip_send_check(iph);
574                 ciph->saddr = cp->daddr;
575                 ip_send_check(ciph);
576         }
577
578         /* the TCP/UDP port */
579         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol) {
580                 __be16 *ports = (void *)ciph + ciph->ihl*4;
581
582                 if (inout)
583                         ports[1] = cp->vport;
584                 else
585                         ports[0] = cp->dport;
586         }
587
588         /* And finally the ICMP checksum */
589         icmph->checksum = 0;
590         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
591         skb->ip_summed = CHECKSUM_UNNECESSARY;
592
593         if (inout)
594                 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
595                         "Forwarding altered outgoing ICMP");
596         else
597                 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
598                         "Forwarding altered incoming ICMP");
599 }
600
601 /*
602  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
603  *      Find any that might be relevant, check against existing connections,
604  *      forward to the right destination host if relevant.
605  *      Currently handles error types - unreachable, quench, ttl exceeded.
606  *      (Only used in VS/NAT)
607  */
608 static int ip_vs_out_icmp(struct sk_buff **pskb, int *related)
609 {
610         struct sk_buff *skb = *pskb;
611         struct iphdr *iph;
612         struct icmphdr  _icmph, *ic;
613         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
614         struct ip_vs_conn *cp;
615         struct ip_vs_protocol *pp;
616         unsigned int offset, ihl, verdict;
617
618         *related = 1;
619
620         /* reassemble IP fragments */
621         if (skb->nh.iph->frag_off & htons(IP_MF|IP_OFFSET)) {
622                 skb = ip_vs_gather_frags(skb, IP_DEFRAG_VS_OUT);
623                 if (!skb)
624                         return NF_STOLEN;
625                 *pskb = skb;
626         }
627
628         iph = skb->nh.iph;
629         offset = ihl = iph->ihl * 4;
630         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
631         if (ic == NULL)
632                 return NF_DROP;
633
634         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
635                   ic->type, ntohs(icmp_id(ic)),
636                   NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
637
638         /*
639          * Work through seeing if this is for us.
640          * These checks are supposed to be in an order that means easy
641          * things are checked first to speed up processing.... however
642          * this means that some packets will manage to get a long way
643          * down this stack and then be rejected, but that's life.
644          */
645         if ((ic->type != ICMP_DEST_UNREACH) &&
646             (ic->type != ICMP_SOURCE_QUENCH) &&
647             (ic->type != ICMP_TIME_EXCEEDED)) {
648                 *related = 0;
649                 return NF_ACCEPT;
650         }
651
652         /* Now find the contained IP header */
653         offset += sizeof(_icmph);
654         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
655         if (cih == NULL)
656                 return NF_ACCEPT; /* The packet looks wrong, ignore */
657
658         pp = ip_vs_proto_get(cih->protocol);
659         if (!pp)
660                 return NF_ACCEPT;
661
662         /* Is the embedded protocol header present? */
663         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
664                      pp->dont_defrag))
665                 return NF_ACCEPT;
666
667         IP_VS_DBG_PKT(11, pp, skb, offset, "Checking outgoing ICMP for");
668
669         offset += cih->ihl * 4;
670
671         /* The embedded headers contain source and dest in reverse order */
672         cp = pp->conn_out_get(skb, pp, cih, offset, 1);
673         if (!cp)
674                 return NF_ACCEPT;
675
676         verdict = NF_DROP;
677
678         if (IP_VS_FWD_METHOD(cp) != 0) {
679                 IP_VS_ERR("shouldn't reach here, because the box is on the"
680                           "half connection in the tun/dr module.\n");
681         }
682
683         /* Ensure the checksum is correct */
684         if (skb->ip_summed != CHECKSUM_UNNECESSARY &&
685             ip_vs_checksum_complete(skb, ihl)) {
686                 /* Failed checksum! */
687                 IP_VS_DBG(1, "Forward ICMP: failed checksum from %d.%d.%d.%d!\n",
688                           NIPQUAD(iph->saddr));
689                 goto out;
690         }
691
692         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
693                 offset += 2 * sizeof(__u16);
694         if (!ip_vs_make_skb_writable(pskb, offset))
695                 goto out;
696         skb = *pskb;
697
698         ip_vs_nat_icmp(skb, pp, cp, 1);
699
700         /* do the statistics and put it back */
701         ip_vs_out_stats(cp, skb);
702
703         skb->ipvs_property = 1;
704         verdict = NF_ACCEPT;
705
706   out:
707         __ip_vs_conn_put(cp);
708
709         return verdict;
710 }
711
712 static inline int is_tcp_reset(const struct sk_buff *skb)
713 {
714         struct tcphdr _tcph, *th;
715
716         th = skb_header_pointer(skb, skb->nh.iph->ihl * 4,
717                                 sizeof(_tcph), &_tcph);
718         if (th == NULL)
719                 return 0;
720         return th->rst;
721 }
722
723 /*
724  *      It is hooked at the NF_IP_FORWARD chain, used only for VS/NAT.
725  *      Check if outgoing packet belongs to the established ip_vs_conn,
726  *      rewrite addresses of the packet and send it on its way...
727  */
728 static unsigned int
729 ip_vs_out(unsigned int hooknum, struct sk_buff **pskb,
730           const struct net_device *in, const struct net_device *out,
731           int (*okfn)(struct sk_buff *))
732 {
733         struct sk_buff  *skb = *pskb;
734         struct iphdr    *iph;
735         struct ip_vs_protocol *pp;
736         struct ip_vs_conn *cp;
737         int ihl;
738
739         EnterFunction(11);
740
741         if (skb->ipvs_property)
742                 return NF_ACCEPT;
743
744         iph = skb->nh.iph;
745         if (unlikely(iph->protocol == IPPROTO_ICMP)) {
746                 int related, verdict = ip_vs_out_icmp(pskb, &related);
747
748                 if (related)
749                         return verdict;
750                 skb = *pskb;
751                 iph = skb->nh.iph;
752         }
753
754         pp = ip_vs_proto_get(iph->protocol);
755         if (unlikely(!pp))
756                 return NF_ACCEPT;
757
758         /* reassemble IP fragments */
759         if (unlikely(iph->frag_off & htons(IP_MF|IP_OFFSET) &&
760                      !pp->dont_defrag)) {
761                 skb = ip_vs_gather_frags(skb, IP_DEFRAG_VS_OUT);
762                 if (!skb)
763                         return NF_STOLEN;
764                 iph = skb->nh.iph;
765                 *pskb = skb;
766         }
767
768         ihl = iph->ihl << 2;
769
770         /*
771          * Check if the packet belongs to an existing entry
772          */
773         cp = pp->conn_out_get(skb, pp, iph, ihl, 0);
774
775         if (unlikely(!cp)) {
776                 if (sysctl_ip_vs_nat_icmp_send &&
777                     (pp->protocol == IPPROTO_TCP ||
778                      pp->protocol == IPPROTO_UDP)) {
779                         __be16 _ports[2], *pptr;
780
781                         pptr = skb_header_pointer(skb, ihl,
782                                                   sizeof(_ports), _ports);
783                         if (pptr == NULL)
784                                 return NF_ACCEPT;       /* Not for me */
785                         if (ip_vs_lookup_real_service(iph->protocol,
786                                                       iph->saddr, pptr[0])) {
787                                 /*
788                                  * Notify the real server: there is no
789                                  * existing entry if it is not RST
790                                  * packet or not TCP packet.
791                                  */
792                                 if (iph->protocol != IPPROTO_TCP
793                                     || !is_tcp_reset(skb)) {
794                                         icmp_send(skb,ICMP_DEST_UNREACH,
795                                                   ICMP_PORT_UNREACH, 0);
796                                         return NF_DROP;
797                                 }
798                         }
799                 }
800                 IP_VS_DBG_PKT(12, pp, skb, 0,
801                               "packet continues traversal as normal");
802                 return NF_ACCEPT;
803         }
804
805         IP_VS_DBG_PKT(11, pp, skb, 0, "Outgoing packet");
806
807         if (!ip_vs_make_skb_writable(pskb, ihl))
808                 goto drop;
809
810         /* mangle the packet */
811         if (pp->snat_handler && !pp->snat_handler(pskb, pp, cp))
812                 goto drop;
813         skb = *pskb;
814         skb->nh.iph->saddr = cp->vaddr;
815         ip_send_check(skb->nh.iph);
816
817         /* For policy routing, packets originating from this
818          * machine itself may be routed differently to packets
819          * passing through.  We want this packet to be routed as
820          * if it came from this machine itself.  So re-compute
821          * the routing information.
822          */
823         if (ip_route_me_harder(pskb, RTN_LOCAL) != 0)
824                 goto drop;
825         skb = *pskb;
826
827         IP_VS_DBG_PKT(10, pp, skb, 0, "After SNAT");
828
829         ip_vs_out_stats(cp, skb);
830         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pp);
831         ip_vs_conn_put(cp);
832
833         skb->ipvs_property = 1;
834
835         LeaveFunction(11);
836         return NF_ACCEPT;
837
838   drop:
839         ip_vs_conn_put(cp);
840         kfree_skb(*pskb);
841         return NF_STOLEN;
842 }
843
844
845 /*
846  *      Handle ICMP messages in the outside-to-inside direction (incoming).
847  *      Find any that might be relevant, check against existing connections,
848  *      forward to the right destination host if relevant.
849  *      Currently handles error types - unreachable, quench, ttl exceeded.
850  */
851 static int
852 ip_vs_in_icmp(struct sk_buff **pskb, int *related, unsigned int hooknum)
853 {
854         struct sk_buff *skb = *pskb;
855         struct iphdr *iph;
856         struct icmphdr  _icmph, *ic;
857         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
858         struct ip_vs_conn *cp;
859         struct ip_vs_protocol *pp;
860         unsigned int offset, ihl, verdict;
861
862         *related = 1;
863
864         /* reassemble IP fragments */
865         if (skb->nh.iph->frag_off & htons(IP_MF|IP_OFFSET)) {
866                 skb = ip_vs_gather_frags(skb,
867                                          hooknum == NF_IP_LOCAL_IN ?
868                                          IP_DEFRAG_VS_IN : IP_DEFRAG_VS_FWD);
869                 if (!skb)
870                         return NF_STOLEN;
871                 *pskb = skb;
872         }
873
874         iph = skb->nh.iph;
875         offset = ihl = iph->ihl * 4;
876         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
877         if (ic == NULL)
878                 return NF_DROP;
879
880         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
881                   ic->type, ntohs(icmp_id(ic)),
882                   NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
883
884         /*
885          * Work through seeing if this is for us.
886          * These checks are supposed to be in an order that means easy
887          * things are checked first to speed up processing.... however
888          * this means that some packets will manage to get a long way
889          * down this stack and then be rejected, but that's life.
890          */
891         if ((ic->type != ICMP_DEST_UNREACH) &&
892             (ic->type != ICMP_SOURCE_QUENCH) &&
893             (ic->type != ICMP_TIME_EXCEEDED)) {
894                 *related = 0;
895                 return NF_ACCEPT;
896         }
897
898         /* Now find the contained IP header */
899         offset += sizeof(_icmph);
900         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
901         if (cih == NULL)
902                 return NF_ACCEPT; /* The packet looks wrong, ignore */
903
904         pp = ip_vs_proto_get(cih->protocol);
905         if (!pp)
906                 return NF_ACCEPT;
907
908         /* Is the embedded protocol header present? */
909         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
910                      pp->dont_defrag))
911                 return NF_ACCEPT;
912
913         IP_VS_DBG_PKT(11, pp, skb, offset, "Checking incoming ICMP for");
914
915         offset += cih->ihl * 4;
916
917         /* The embedded headers contain source and dest in reverse order */
918         cp = pp->conn_in_get(skb, pp, cih, offset, 1);
919         if (!cp)
920                 return NF_ACCEPT;
921
922         verdict = NF_DROP;
923
924         /* Ensure the checksum is correct */
925         if (skb->ip_summed != CHECKSUM_UNNECESSARY &&
926             ip_vs_checksum_complete(skb, ihl)) {
927                 /* Failed checksum! */
928                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %d.%d.%d.%d!\n",
929                           NIPQUAD(iph->saddr));
930                 goto out;
931         }
932
933         /* do the statistics and put it back */
934         ip_vs_in_stats(cp, skb);
935         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
936                 offset += 2 * sizeof(__u16);
937         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
938         /* do not touch skb anymore */
939
940   out:
941         __ip_vs_conn_put(cp);
942
943         return verdict;
944 }
945
946 /*
947  *      Check if it's for virtual services, look it up,
948  *      and send it on its way...
949  */
950 static unsigned int
951 ip_vs_in(unsigned int hooknum, struct sk_buff **pskb,
952          const struct net_device *in, const struct net_device *out,
953          int (*okfn)(struct sk_buff *))
954 {
955         struct sk_buff  *skb = *pskb;
956         struct iphdr    *iph;
957         struct ip_vs_protocol *pp;
958         struct ip_vs_conn *cp;
959         int ret, restart;
960         int ihl;
961
962         /*
963          *      Big tappo: only PACKET_HOST (neither loopback nor mcasts)
964          *      ... don't know why 1st test DOES NOT include 2nd (?)
965          */
966         if (unlikely(skb->pkt_type != PACKET_HOST
967                      || skb->dev == &loopback_dev || skb->sk)) {
968                 IP_VS_DBG(12, "packet type=%d proto=%d daddr=%d.%d.%d.%d ignored\n",
969                           skb->pkt_type,
970                           skb->nh.iph->protocol,
971                           NIPQUAD(skb->nh.iph->daddr));
972                 return NF_ACCEPT;
973         }
974
975         iph = skb->nh.iph;
976         if (unlikely(iph->protocol == IPPROTO_ICMP)) {
977                 int related, verdict = ip_vs_in_icmp(pskb, &related, hooknum);
978
979                 if (related)
980                         return verdict;
981                 skb = *pskb;
982                 iph = skb->nh.iph;
983         }
984
985         /* Protocol supported? */
986         pp = ip_vs_proto_get(iph->protocol);
987         if (unlikely(!pp))
988                 return NF_ACCEPT;
989
990         ihl = iph->ihl << 2;
991
992         /*
993          * Check if the packet belongs to an existing connection entry
994          */
995         cp = pp->conn_in_get(skb, pp, iph, ihl, 0);
996
997         if (unlikely(!cp)) {
998                 int v;
999
1000                 if (!pp->conn_schedule(skb, pp, &v, &cp))
1001                         return v;
1002         }
1003
1004         if (unlikely(!cp)) {
1005                 /* sorry, all this trouble for a no-hit :) */
1006                 IP_VS_DBG_PKT(12, pp, skb, 0,
1007                               "packet continues traversal as normal");
1008                 return NF_ACCEPT;
1009         }
1010
1011         IP_VS_DBG_PKT(11, pp, skb, 0, "Incoming packet");
1012
1013         /* Check the server status */
1014         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1015                 /* the destination server is not available */
1016
1017                 if (sysctl_ip_vs_expire_nodest_conn) {
1018                         /* try to expire the connection immediately */
1019                         ip_vs_conn_expire_now(cp);
1020                 }
1021                 /* don't restart its timer, and silently
1022                    drop the packet. */
1023                 __ip_vs_conn_put(cp);
1024                 return NF_DROP;
1025         }
1026
1027         ip_vs_in_stats(cp, skb);
1028         restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
1029         if (cp->packet_xmit)
1030                 ret = cp->packet_xmit(skb, cp, pp);
1031                 /* do not touch skb anymore */
1032         else {
1033                 IP_VS_DBG_RL("warning: packet_xmit is null");
1034                 ret = NF_ACCEPT;
1035         }
1036
1037         /* increase its packet counter and check if it is needed
1038            to be synchronized */
1039         atomic_inc(&cp->in_pkts);
1040         if ((ip_vs_sync_state & IP_VS_STATE_MASTER) &&
1041             (cp->protocol != IPPROTO_TCP ||
1042              cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1043             (atomic_read(&cp->in_pkts) % sysctl_ip_vs_sync_threshold[1]
1044              == sysctl_ip_vs_sync_threshold[0]))
1045                 ip_vs_sync_conn(cp);
1046
1047         ip_vs_conn_put(cp);
1048         return ret;
1049 }
1050
1051
1052 /*
1053  *      It is hooked at the NF_IP_FORWARD chain, in order to catch ICMP
1054  *      related packets destined for 0.0.0.0/0.
1055  *      When fwmark-based virtual service is used, such as transparent
1056  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1057  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1058  *      sent to ip_vs_in_icmp. So, catch them at the NF_IP_FORWARD chain
1059  *      and send them to ip_vs_in_icmp.
1060  */
1061 static unsigned int
1062 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff **pskb,
1063                    const struct net_device *in, const struct net_device *out,
1064                    int (*okfn)(struct sk_buff *))
1065 {
1066         int r;
1067
1068         if ((*pskb)->nh.iph->protocol != IPPROTO_ICMP)
1069                 return NF_ACCEPT;
1070
1071         return ip_vs_in_icmp(pskb, &r, hooknum);
1072 }
1073
1074
1075 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1076    or VS/NAT(change destination), so that filtering rules can be
1077    applied to IPVS. */
1078 static struct nf_hook_ops ip_vs_in_ops = {
1079         .hook           = ip_vs_in,
1080         .owner          = THIS_MODULE,
1081         .pf             = PF_INET,
1082         .hooknum        = NF_IP_LOCAL_IN,
1083         .priority       = 100,
1084 };
1085
1086 /* After packet filtering, change source only for VS/NAT */
1087 static struct nf_hook_ops ip_vs_out_ops = {
1088         .hook           = ip_vs_out,
1089         .owner          = THIS_MODULE,
1090         .pf             = PF_INET,
1091         .hooknum        = NF_IP_FORWARD,
1092         .priority       = 100,
1093 };
1094
1095 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1096    destined for 0.0.0.0/0, which is for incoming IPVS connections */
1097 static struct nf_hook_ops ip_vs_forward_icmp_ops = {
1098         .hook           = ip_vs_forward_icmp,
1099         .owner          = THIS_MODULE,
1100         .pf             = PF_INET,
1101         .hooknum        = NF_IP_FORWARD,
1102         .priority       = 99,
1103 };
1104
1105 /* Before the netfilter connection tracking, exit from POST_ROUTING */
1106 static struct nf_hook_ops ip_vs_post_routing_ops = {
1107         .hook           = ip_vs_post_routing,
1108         .owner          = THIS_MODULE,
1109         .pf             = PF_INET,
1110         .hooknum        = NF_IP_POST_ROUTING,
1111         .priority       = NF_IP_PRI_NAT_SRC-1,
1112 };
1113
1114
1115 /*
1116  *      Initialize IP Virtual Server
1117  */
1118 static int __init ip_vs_init(void)
1119 {
1120         int ret;
1121
1122         ret = ip_vs_control_init();
1123         if (ret < 0) {
1124                 IP_VS_ERR("can't setup control.\n");
1125                 goto cleanup_nothing;
1126         }
1127
1128         ip_vs_protocol_init();
1129
1130         ret = ip_vs_app_init();
1131         if (ret < 0) {
1132                 IP_VS_ERR("can't setup application helper.\n");
1133                 goto cleanup_protocol;
1134         }
1135
1136         ret = ip_vs_conn_init();
1137         if (ret < 0) {
1138                 IP_VS_ERR("can't setup connection table.\n");
1139                 goto cleanup_app;
1140         }
1141
1142         ret = nf_register_hook(&ip_vs_in_ops);
1143         if (ret < 0) {
1144                 IP_VS_ERR("can't register in hook.\n");
1145                 goto cleanup_conn;
1146         }
1147
1148         ret = nf_register_hook(&ip_vs_out_ops);
1149         if (ret < 0) {
1150                 IP_VS_ERR("can't register out hook.\n");
1151                 goto cleanup_inops;
1152         }
1153         ret = nf_register_hook(&ip_vs_post_routing_ops);
1154         if (ret < 0) {
1155                 IP_VS_ERR("can't register post_routing hook.\n");
1156                 goto cleanup_outops;
1157         }
1158         ret = nf_register_hook(&ip_vs_forward_icmp_ops);
1159         if (ret < 0) {
1160                 IP_VS_ERR("can't register forward_icmp hook.\n");
1161                 goto cleanup_postroutingops;
1162         }
1163
1164         IP_VS_INFO("ipvs loaded.\n");
1165         return ret;
1166
1167   cleanup_postroutingops:
1168         nf_unregister_hook(&ip_vs_post_routing_ops);
1169   cleanup_outops:
1170         nf_unregister_hook(&ip_vs_out_ops);
1171   cleanup_inops:
1172         nf_unregister_hook(&ip_vs_in_ops);
1173   cleanup_conn:
1174         ip_vs_conn_cleanup();
1175   cleanup_app:
1176         ip_vs_app_cleanup();
1177   cleanup_protocol:
1178         ip_vs_protocol_cleanup();
1179         ip_vs_control_cleanup();
1180   cleanup_nothing:
1181         return ret;
1182 }
1183
1184 static void __exit ip_vs_cleanup(void)
1185 {
1186         nf_unregister_hook(&ip_vs_forward_icmp_ops);
1187         nf_unregister_hook(&ip_vs_post_routing_ops);
1188         nf_unregister_hook(&ip_vs_out_ops);
1189         nf_unregister_hook(&ip_vs_in_ops);
1190         ip_vs_conn_cleanup();
1191         ip_vs_app_cleanup();
1192         ip_vs_protocol_cleanup();
1193         ip_vs_control_cleanup();
1194         IP_VS_INFO("ipvs unloaded.\n");
1195 }
1196
1197 module_init(ip_vs_init);
1198 module_exit(ip_vs_cleanup);
1199 MODULE_LICENSE("GPL");