?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[PATCH] v850: call init_page_count() instead of set_page_count()
[linux-2.6/verdex.git] / net / ipv4 / ipvs / ip_vs_core.c
blob3f47ad8e1cad1c7f8165b6e5c003a3f919711e4a
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 __u16 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 __u16 dport; /* destination port to forward */
219 __u32 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 __u16 _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 __u16 _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 u16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
540 {
541 return (u16) 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->nh.raw + icmp_offset);
563 struct iphdr *ciph = (struct iphdr *)(icmph + 1);
564
565 if (inout) {
566 iph->saddr = cp->vaddr;
567 ip_send_check(iph);
568 ciph->daddr = cp->vaddr;
569 ip_send_check(ciph);
570 } else {
571 iph->daddr = cp->daddr;
572 ip_send_check(iph);
573 ciph->saddr = cp->daddr;
574 ip_send_check(ciph);
575 }
576
577 /* the TCP/UDP port */
578 if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol) {
579 __u16 *ports = (void *)ciph + ciph->ihl*4;
580
581 if (inout)
582 ports[1] = cp->vport;
583 else
584 ports[0] = cp->dport;
585 }
586
587 /* And finally the ICMP checksum */
588 icmph->checksum = 0;
589 icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
590 skb->ip_summed = CHECKSUM_UNNECESSARY;
591
592 if (inout)
593 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
594 "Forwarding altered outgoing ICMP");
595 else
596 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
597 "Forwarding altered incoming ICMP");
598 }
599
600 /*
601 * Handle ICMP messages in the inside-to-outside direction (outgoing).
602 * Find any that might be relevant, check against existing connections,
603 * forward to the right destination host if relevant.
604 * Currently handles error types - unreachable, quench, ttl exceeded.
605 * (Only used in VS/NAT)
606 */
607 static int ip_vs_out_icmp(struct sk_buff **pskb, int *related)
608 {
609 struct sk_buff *skb = *pskb;
610 struct iphdr *iph;
611 struct icmphdr _icmph, *ic;
612 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
613 struct ip_vs_conn *cp;
614 struct ip_vs_protocol *pp;
615 unsigned int offset, ihl, verdict;
616
617 *related = 1;
618
619 /* reassemble IP fragments */
620 if (skb->nh.iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) {
621 skb = ip_vs_gather_frags(skb, IP_DEFRAG_VS_OUT);
622 if (!skb)
623 return NF_STOLEN;
624 *pskb = skb;
625 }
626
627 iph = skb->nh.iph;
628 offset = ihl = iph->ihl * 4;
629 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
630 if (ic == NULL)
631 return NF_DROP;
632
633 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
634 ic->type, ntohs(icmp_id(ic)),
635 NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
636
637 /*
638 * Work through seeing if this is for us.
639 * These checks are supposed to be in an order that means easy
640 * things are checked first to speed up processing.... however
641 * this means that some packets will manage to get a long way
642 * down this stack and then be rejected, but that's life.
643 */
644 if ((ic->type != ICMP_DEST_UNREACH) &&
645 (ic->type != ICMP_SOURCE_QUENCH) &&
646 (ic->type != ICMP_TIME_EXCEEDED)) {
647 *related = 0;
648 return NF_ACCEPT;
649 }
650
651 /* Now find the contained IP header */
652 offset += sizeof(_icmph);
653 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
654 if (cih == NULL)
655 return NF_ACCEPT; /* The packet looks wrong, ignore */
656
657 pp = ip_vs_proto_get(cih->protocol);
658 if (!pp)
659 return NF_ACCEPT;
660
661 /* Is the embedded protocol header present? */
662 if (unlikely(cih->frag_off & __constant_htons(IP_OFFSET) &&
663 pp->dont_defrag))
664 return NF_ACCEPT;
665
666 IP_VS_DBG_PKT(11, pp, skb, offset, "Checking outgoing ICMP for");
667
668 offset += cih->ihl * 4;
669
670 /* The embedded headers contain source and dest in reverse order */
671 cp = pp->conn_out_get(skb, pp, cih, offset, 1);
672 if (!cp)
673 return NF_ACCEPT;
674
675 verdict = NF_DROP;
676
677 if (IP_VS_FWD_METHOD(cp) != 0) {
678 IP_VS_ERR("shouldn't reach here, because the box is on the"
679 "half connection in the tun/dr module.\n");
680 }
681
682 /* Ensure the checksum is correct */
683 if (skb->ip_summed != CHECKSUM_UNNECESSARY &&
684 ip_vs_checksum_complete(skb, ihl)) {
685 /* Failed checksum! */
686 IP_VS_DBG(1, "Forward ICMP: failed checksum from %d.%d.%d.%d!\n",
687 NIPQUAD(iph->saddr));
688 goto out;
689 }
690
691 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
692 offset += 2 * sizeof(__u16);
693 if (!ip_vs_make_skb_writable(pskb, offset))
694 goto out;
695 skb = *pskb;
696
697 ip_vs_nat_icmp(skb, pp, cp, 1);
698
699 /* do the statistics and put it back */
700 ip_vs_out_stats(cp, skb);
701
702 skb->ipvs_property = 1;
703 verdict = NF_ACCEPT;
704
705 out:
706 __ip_vs_conn_put(cp);
707
708 return verdict;
709 }
710
711 static inline int is_tcp_reset(const struct sk_buff *skb)
712 {
713 struct tcphdr _tcph, *th;
714
715 th = skb_header_pointer(skb, skb->nh.iph->ihl * 4,
716 sizeof(_tcph), &_tcph);
717 if (th == NULL)
718 return 0;
719 return th->rst;
720 }
721
722 /*
723 * It is hooked at the NF_IP_FORWARD chain, used only for VS/NAT.
724 * Check if outgoing packet belongs to the established ip_vs_conn,
725 * rewrite addresses of the packet and send it on its way...
726 */
727 static unsigned int
728 ip_vs_out(unsigned int hooknum, struct sk_buff **pskb,
729 const struct net_device *in, const struct net_device *out,
730 int (*okfn)(struct sk_buff *))
731 {
732 struct sk_buff *skb = *pskb;
733 struct iphdr *iph;
734 struct ip_vs_protocol *pp;
735 struct ip_vs_conn *cp;
736 int ihl;
737
738 EnterFunction(11);
739
740 if (skb->ipvs_property)
741 return NF_ACCEPT;
742
743 iph = skb->nh.iph;
744 if (unlikely(iph->protocol == IPPROTO_ICMP)) {
745 int related, verdict = ip_vs_out_icmp(pskb, &related);
746
747 if (related)
748 return verdict;
749 skb = *pskb;
750 iph = skb->nh.iph;
751 }
752
753 pp = ip_vs_proto_get(iph->protocol);
754 if (unlikely(!pp))
755 return NF_ACCEPT;
756
757 /* reassemble IP fragments */
758 if (unlikely(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET) &&
759 !pp->dont_defrag)) {
760 skb = ip_vs_gather_frags(skb, IP_DEFRAG_VS_OUT);
761 if (!skb)
762 return NF_STOLEN;
763 iph = skb->nh.iph;
764 *pskb = skb;
765 }
766
767 ihl = iph->ihl << 2;
768
769 /*
770 * Check if the packet belongs to an existing entry
771 */
772 cp = pp->conn_out_get(skb, pp, iph, ihl, 0);
773
774 if (unlikely(!cp)) {
775 if (sysctl_ip_vs_nat_icmp_send &&
776 (pp->protocol == IPPROTO_TCP ||
777 pp->protocol == IPPROTO_UDP)) {
778 __u16 _ports[2], *pptr;
779
780 pptr = skb_header_pointer(skb, ihl,
781 sizeof(_ports), _ports);
782 if (pptr == NULL)
783 return NF_ACCEPT; /* Not for me */
784 if (ip_vs_lookup_real_service(iph->protocol,
785 iph->saddr, pptr[0])) {
786 /*
787 * Notify the real server: there is no
788 * existing entry if it is not RST
789 * packet or not TCP packet.
790 */
791 if (iph->protocol != IPPROTO_TCP
792 || !is_tcp_reset(skb)) {
793 icmp_send(skb,ICMP_DEST_UNREACH,
794 ICMP_PORT_UNREACH, 0);
795 return NF_DROP;
796 }
797 }
798 }
799 IP_VS_DBG_PKT(12, pp, skb, 0,
800 "packet continues traversal as normal");
801 return NF_ACCEPT;
802 }
803
804 IP_VS_DBG_PKT(11, pp, skb, 0, "Outgoing packet");
805
806 if (!ip_vs_make_skb_writable(pskb, ihl))
807 goto drop;
808
809 /* mangle the packet */
810 if (pp->snat_handler && !pp->snat_handler(pskb, pp, cp))
811 goto drop;
812 skb = *pskb;
813 skb->nh.iph->saddr = cp->vaddr;
814 ip_send_check(skb->nh.iph);
815
816 IP_VS_DBG_PKT(10, pp, skb, 0, "After SNAT");
817
818 ip_vs_out_stats(cp, skb);
819 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pp);
820 ip_vs_conn_put(cp);
821
822 skb->ipvs_property = 1;
823
824 LeaveFunction(11);
825 return NF_ACCEPT;
826
827 drop:
828 ip_vs_conn_put(cp);
829 kfree_skb(*pskb);
830 return NF_STOLEN;
831 }
832
833
834 /*
835 * Handle ICMP messages in the outside-to-inside direction (incoming).
836 * Find any that might be relevant, check against existing connections,
837 * forward to the right destination host if relevant.
838 * Currently handles error types - unreachable, quench, ttl exceeded.
839 */
840 static int
841 ip_vs_in_icmp(struct sk_buff **pskb, int *related, unsigned int hooknum)
842 {
843 struct sk_buff *skb = *pskb;
844 struct iphdr *iph;
845 struct icmphdr _icmph, *ic;
846 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
847 struct ip_vs_conn *cp;
848 struct ip_vs_protocol *pp;
849 unsigned int offset, ihl, verdict;
850
851 *related = 1;
852
853 /* reassemble IP fragments */
854 if (skb->nh.iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) {
855 skb = ip_vs_gather_frags(skb,
856 hooknum == NF_IP_LOCAL_IN ?
857 IP_DEFRAG_VS_IN : IP_DEFRAG_VS_FWD);
858 if (!skb)
859 return NF_STOLEN;
860 *pskb = skb;
861 }
862
863 iph = skb->nh.iph;
864 offset = ihl = iph->ihl * 4;
865 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
866 if (ic == NULL)
867 return NF_DROP;
868
869 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
870 ic->type, ntohs(icmp_id(ic)),
871 NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
872
873 /*
874 * Work through seeing if this is for us.
875 * These checks are supposed to be in an order that means easy
876 * things are checked first to speed up processing.... however
877 * this means that some packets will manage to get a long way
878 * down this stack and then be rejected, but that's life.
879 */
880 if ((ic->type != ICMP_DEST_UNREACH) &&
881 (ic->type != ICMP_SOURCE_QUENCH) &&
882 (ic->type != ICMP_TIME_EXCEEDED)) {
883 *related = 0;
884 return NF_ACCEPT;
885 }
886
887 /* Now find the contained IP header */
888 offset += sizeof(_icmph);
889 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
890 if (cih == NULL)
891 return NF_ACCEPT; /* The packet looks wrong, ignore */
892
893 pp = ip_vs_proto_get(cih->protocol);
894 if (!pp)
895 return NF_ACCEPT;
896
897 /* Is the embedded protocol header present? */
898 if (unlikely(cih->frag_off & __constant_htons(IP_OFFSET) &&
899 pp->dont_defrag))
900 return NF_ACCEPT;
901
902 IP_VS_DBG_PKT(11, pp, skb, offset, "Checking incoming ICMP for");
903
904 offset += cih->ihl * 4;
905
906 /* The embedded headers contain source and dest in reverse order */
907 cp = pp->conn_in_get(skb, pp, cih, offset, 1);
908 if (!cp)
909 return NF_ACCEPT;
910
911 verdict = NF_DROP;
912
913 /* Ensure the checksum is correct */
914 if (skb->ip_summed != CHECKSUM_UNNECESSARY &&
915 ip_vs_checksum_complete(skb, ihl)) {
916 /* Failed checksum! */
917 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %d.%d.%d.%d!\n",
918 NIPQUAD(iph->saddr));
919 goto out;
920 }
921
922 /* do the statistics and put it back */
923 ip_vs_in_stats(cp, skb);
924 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
925 offset += 2 * sizeof(__u16);
926 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
927 /* do not touch skb anymore */
928
929 out:
930 __ip_vs_conn_put(cp);
931
932 return verdict;
933 }
934
935 /*
936 * Check if it's for virtual services, look it up,
937 * and send it on its way...
938 */
939 static unsigned int
940 ip_vs_in(unsigned int hooknum, struct sk_buff **pskb,
941 const struct net_device *in, const struct net_device *out,
942 int (*okfn)(struct sk_buff *))
943 {
944 struct sk_buff *skb = *pskb;
945 struct iphdr *iph;
946 struct ip_vs_protocol *pp;
947 struct ip_vs_conn *cp;
948 int ret, restart;
949 int ihl;
950
951 /*
952 * Big tappo: only PACKET_HOST (neither loopback nor mcasts)
953 * ... don't know why 1st test DOES NOT include 2nd (?)
954 */
955 if (unlikely(skb->pkt_type != PACKET_HOST
956 || skb->dev == &loopback_dev || skb->sk)) {
957 IP_VS_DBG(12, "packet type=%d proto=%d daddr=%d.%d.%d.%d ignored\n",
958 skb->pkt_type,
959 skb->nh.iph->protocol,
960 NIPQUAD(skb->nh.iph->daddr));
961 return NF_ACCEPT;
962 }
963
964 iph = skb->nh.iph;
965 if (unlikely(iph->protocol == IPPROTO_ICMP)) {
966 int related, verdict = ip_vs_in_icmp(pskb, &related, hooknum);
967
968 if (related)
969 return verdict;
970 skb = *pskb;
971 iph = skb->nh.iph;
972 }
973
974 /* Protocol supported? */
975 pp = ip_vs_proto_get(iph->protocol);
976 if (unlikely(!pp))
977 return NF_ACCEPT;
978
979 ihl = iph->ihl << 2;
980
981 /*
982 * Check if the packet belongs to an existing connection entry
983 */
984 cp = pp->conn_in_get(skb, pp, iph, ihl, 0);
985
986 if (unlikely(!cp)) {
987 int v;
988
989 if (!pp->conn_schedule(skb, pp, &v, &cp))
990 return v;
991 }
992
993 if (unlikely(!cp)) {
994 /* sorry, all this trouble for a no-hit :) */
995 IP_VS_DBG_PKT(12, pp, skb, 0,
996 "packet continues traversal as normal");
997 return NF_ACCEPT;
998 }
999
1000 IP_VS_DBG_PKT(11, pp, skb, 0, "Incoming packet");
1001
1002 /* Check the server status */
1003 if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1004 /* the destination server is not available */
1005
1006 if (sysctl_ip_vs_expire_nodest_conn) {
1007 /* try to expire the connection immediately */
1008 ip_vs_conn_expire_now(cp);
1009 }
1010 /* don't restart its timer, and silently
1011 drop the packet. */
1012 __ip_vs_conn_put(cp);
1013 return NF_DROP;
1014 }
1015
1016 ip_vs_in_stats(cp, skb);
1017 restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
1018 if (cp->packet_xmit)
1019 ret = cp->packet_xmit(skb, cp, pp);
1020 /* do not touch skb anymore */
1021 else {
1022 IP_VS_DBG_RL("warning: packet_xmit is null");
1023 ret = NF_ACCEPT;
1024 }
1025
1026 /* increase its packet counter and check if it is needed
1027 to be synchronized */
1028 atomic_inc(&cp->in_pkts);
1029 if ((ip_vs_sync_state & IP_VS_STATE_MASTER) &&
1030 (cp->protocol != IPPROTO_TCP ||
1031 cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1032 (atomic_read(&cp->in_pkts) % sysctl_ip_vs_sync_threshold[1]
1033 == sysctl_ip_vs_sync_threshold[0]))
1034 ip_vs_sync_conn(cp);
1035
1036 ip_vs_conn_put(cp);
1037 return ret;
1038 }
1039
1040
1041 /*
1042 * It is hooked at the NF_IP_FORWARD chain, in order to catch ICMP
1043 * related packets destined for 0.0.0.0/0.
1044 * When fwmark-based virtual service is used, such as transparent
1045 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1046 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1047 * sent to ip_vs_in_icmp. So, catch them at the NF_IP_FORWARD chain
1048 * and send them to ip_vs_in_icmp.
1049 */
1050 static unsigned int
1051 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff **pskb,
1052 const struct net_device *in, const struct net_device *out,
1053 int (*okfn)(struct sk_buff *))
1054 {
1055 int r;
1056
1057 if ((*pskb)->nh.iph->protocol != IPPROTO_ICMP)
1058 return NF_ACCEPT;
1059
1060 return ip_vs_in_icmp(pskb, &r, hooknum);
1061 }
1062
1063
1064 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1065 or VS/NAT(change destination), so that filtering rules can be
1066 applied to IPVS. */
1067 static struct nf_hook_ops ip_vs_in_ops = {
1068 .hook = ip_vs_in,
1069 .owner = THIS_MODULE,
1070 .pf = PF_INET,
1071 .hooknum = NF_IP_LOCAL_IN,
1072 .priority = 100,
1073 };
1074
1075 /* After packet filtering, change source only for VS/NAT */
1076 static struct nf_hook_ops ip_vs_out_ops = {
1077 .hook = ip_vs_out,
1078 .owner = THIS_MODULE,
1079 .pf = PF_INET,
1080 .hooknum = NF_IP_FORWARD,
1081 .priority = 100,
1082 };
1083
1084 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1085 destined for 0.0.0.0/0, which is for incoming IPVS connections */
1086 static struct nf_hook_ops ip_vs_forward_icmp_ops = {
1087 .hook = ip_vs_forward_icmp,
1088 .owner = THIS_MODULE,
1089 .pf = PF_INET,
1090 .hooknum = NF_IP_FORWARD,
1091 .priority = 99,
1092 };
1093
1094 /* Before the netfilter connection tracking, exit from POST_ROUTING */
1095 static struct nf_hook_ops ip_vs_post_routing_ops = {
1096 .hook = ip_vs_post_routing,
1097 .owner = THIS_MODULE,
1098 .pf = PF_INET,
1099 .hooknum = NF_IP_POST_ROUTING,
1100 .priority = NF_IP_PRI_NAT_SRC-1,
1101 };
1102
1103
1104 /*
1105 * Initialize IP Virtual Server
1106 */
1107 static int __init ip_vs_init(void)
1108 {
1109 int ret;
1110
1111 ret = ip_vs_control_init();
1112 if (ret < 0) {
1113 IP_VS_ERR("can't setup control.\n");
1114 goto cleanup_nothing;
1115 }
1116
1117 ip_vs_protocol_init();
1118
1119 ret = ip_vs_app_init();
1120 if (ret < 0) {
1121 IP_VS_ERR("can't setup application helper.\n");
1122 goto cleanup_protocol;
1123 }
1124
1125 ret = ip_vs_conn_init();
1126 if (ret < 0) {
1127 IP_VS_ERR("can't setup connection table.\n");
1128 goto cleanup_app;
1129 }
1130
1131 ret = nf_register_hook(&ip_vs_in_ops);
1132 if (ret < 0) {
1133 IP_VS_ERR("can't register in hook.\n");
1134 goto cleanup_conn;
1135 }
1136
1137 ret = nf_register_hook(&ip_vs_out_ops);
1138 if (ret < 0) {
1139 IP_VS_ERR("can't register out hook.\n");
1140 goto cleanup_inops;
1141 }
1142 ret = nf_register_hook(&ip_vs_post_routing_ops);
1143 if (ret < 0) {
1144 IP_VS_ERR("can't register post_routing hook.\n");
1145 goto cleanup_outops;
1146 }
1147 ret = nf_register_hook(&ip_vs_forward_icmp_ops);
1148 if (ret < 0) {
1149 IP_VS_ERR("can't register forward_icmp hook.\n");
1150 goto cleanup_postroutingops;
1151 }
1152
1153 IP_VS_INFO("ipvs loaded.\n");
1154 return ret;
1155
1156 cleanup_postroutingops:
1157 nf_unregister_hook(&ip_vs_post_routing_ops);
1158 cleanup_outops:
1159 nf_unregister_hook(&ip_vs_out_ops);
1160 cleanup_inops:
1161 nf_unregister_hook(&ip_vs_in_ops);
1162 cleanup_conn:
1163 ip_vs_conn_cleanup();
1164 cleanup_app:
1165 ip_vs_app_cleanup();
1166 cleanup_protocol:
1167 ip_vs_protocol_cleanup();
1168 ip_vs_control_cleanup();
1169 cleanup_nothing:
1170 return ret;
1171 }
1172
1173 static void __exit ip_vs_cleanup(void)
1174 {
1175 nf_unregister_hook(&ip_vs_forward_icmp_ops);
1176 nf_unregister_hook(&ip_vs_post_routing_ops);
1177 nf_unregister_hook(&ip_vs_out_ops);
1178 nf_unregister_hook(&ip_vs_in_ops);
1179 ip_vs_conn_cleanup();
1180 ip_vs_app_cleanup();
1181 ip_vs_protocol_cleanup();
1182 ip_vs_control_cleanup();
1183 IP_VS_INFO("ipvs unloaded.\n");
1184 }
1185
1186 module_init(ip_vs_init);
1187 module_exit(ip_vs_cleanup);
1188 MODULE_LICENSE("GPL");
Verdex Pro support