?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
IPVS: Handle Scheduling errors.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / netfilter / ipvs / ip_vs_core.c
blob3445da6e8c95896164b1b2e5ac65864eceefb231
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 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
9 * Peter Kese <peter.kese@ijs.si>
10 * Julian Anastasov <ja@ssi.bg>
11 *
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
16 *
17 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
18 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
19 * and others.
20 *
21 * Changes:
22 * Paul `Rusty' Russell properly handle non-linear skbs
23 * Harald Welte don't use nfcache
24 *
25 */
26
27 #define KMSG_COMPONENT "IPVS"
28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
29
30 #include <linux/module.h>
31 #include <linux/kernel.h>
32 #include <linux/ip.h>
33 #include <linux/tcp.h>
34 #include <linux/sctp.h>
35 #include <linux/icmp.h>
36 #include <linux/slab.h>
37
38 #include <net/ip.h>
39 #include <net/tcp.h>
40 #include <net/udp.h>
41 #include <net/icmp.h> /* for icmp_send */
42 #include <net/route.h>
43 #include <net/ip6_checksum.h>
44
45 #include <linux/netfilter.h>
46 #include <linux/netfilter_ipv4.h>
47
48 #ifdef CONFIG_IP_VS_IPV6
49 #include <net/ipv6.h>
50 #include <linux/netfilter_ipv6.h>
51 #include <net/ip6_route.h>
52 #endif
53
54 #include <net/ip_vs.h>
55
56
57 EXPORT_SYMBOL(register_ip_vs_scheduler);
58 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
59 EXPORT_SYMBOL(ip_vs_proto_name);
60 EXPORT_SYMBOL(ip_vs_conn_new);
61 EXPORT_SYMBOL(ip_vs_conn_in_get);
62 EXPORT_SYMBOL(ip_vs_conn_out_get);
63 #ifdef CONFIG_IP_VS_PROTO_TCP
64 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
65 #endif
66 EXPORT_SYMBOL(ip_vs_conn_put);
67 #ifdef CONFIG_IP_VS_DEBUG
68 EXPORT_SYMBOL(ip_vs_get_debug_level);
69 #endif
70
71
72 /* ID used in ICMP lookups */
73 #define icmp_id(icmph) (((icmph)->un).echo.id)
74 #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier)
75
76 const char *ip_vs_proto_name(unsigned proto)
77 {
78 static char buf[20];
79
80 switch (proto) {
81 case IPPROTO_IP:
82 return "IP";
83 case IPPROTO_UDP:
84 return "UDP";
85 case IPPROTO_TCP:
86 return "TCP";
87 case IPPROTO_SCTP:
88 return "SCTP";
89 case IPPROTO_ICMP:
90 return "ICMP";
91 #ifdef CONFIG_IP_VS_IPV6
92 case IPPROTO_ICMPV6:
93 return "ICMPv6";
94 #endif
95 default:
96 sprintf(buf, "IP_%d", proto);
97 return buf;
98 }
99 }
100
101 void ip_vs_init_hash_table(struct list_head *table, int rows)
102 {
103 while (--rows >= 0)
104 INIT_LIST_HEAD(&table[rows]);
105 }
106
107 static inline void
108 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
109 {
110 struct ip_vs_dest *dest = cp->dest;
111 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
112 spin_lock(&dest->stats.lock);
113 dest->stats.ustats.inpkts++;
114 dest->stats.ustats.inbytes += skb->len;
115 spin_unlock(&dest->stats.lock);
116
117 spin_lock(&dest->svc->stats.lock);
118 dest->svc->stats.ustats.inpkts++;
119 dest->svc->stats.ustats.inbytes += skb->len;
120 spin_unlock(&dest->svc->stats.lock);
121
122 spin_lock(&ip_vs_stats.lock);
123 ip_vs_stats.ustats.inpkts++;
124 ip_vs_stats.ustats.inbytes += skb->len;
125 spin_unlock(&ip_vs_stats.lock);
126 }
127 }
128
129
130 static inline void
131 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
132 {
133 struct ip_vs_dest *dest = cp->dest;
134 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
135 spin_lock(&dest->stats.lock);
136 dest->stats.ustats.outpkts++;
137 dest->stats.ustats.outbytes += skb->len;
138 spin_unlock(&dest->stats.lock);
139
140 spin_lock(&dest->svc->stats.lock);
141 dest->svc->stats.ustats.outpkts++;
142 dest->svc->stats.ustats.outbytes += skb->len;
143 spin_unlock(&dest->svc->stats.lock);
144
145 spin_lock(&ip_vs_stats.lock);
146 ip_vs_stats.ustats.outpkts++;
147 ip_vs_stats.ustats.outbytes += skb->len;
148 spin_unlock(&ip_vs_stats.lock);
149 }
150 }
151
152
153 static inline void
154 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
155 {
156 spin_lock(&cp->dest->stats.lock);
157 cp->dest->stats.ustats.conns++;
158 spin_unlock(&cp->dest->stats.lock);
159
160 spin_lock(&svc->stats.lock);
161 svc->stats.ustats.conns++;
162 spin_unlock(&svc->stats.lock);
163
164 spin_lock(&ip_vs_stats.lock);
165 ip_vs_stats.ustats.conns++;
166 spin_unlock(&ip_vs_stats.lock);
167 }
168
169
170 static inline int
171 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
172 const struct sk_buff *skb,
173 struct ip_vs_protocol *pp)
174 {
175 if (unlikely(!pp->state_transition))
176 return 0;
177 return pp->state_transition(cp, direction, skb, pp);
178 }
179
180 static inline int
181 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
182 struct sk_buff *skb, int protocol,
183 const union nf_inet_addr *caddr, __be16 cport,
184 const union nf_inet_addr *vaddr, __be16 vport,
185 struct ip_vs_conn_param *p)
186 {
187 ip_vs_conn_fill_param(svc->af, protocol, caddr, cport, vaddr, vport, p);
188 p->pe = svc->pe;
189 if (p->pe && p->pe->fill_param)
190 return p->pe->fill_param(p, skb);
191
192 return 0;
193 }
194
195 /*
196 * IPVS persistent scheduling function
197 * It creates a connection entry according to its template if exists,
198 * or selects a server and creates a connection entry plus a template.
199 * Locking: we are svc user (svc->refcnt), so we hold all dests too
200 * Protocols supported: TCP, UDP
201 */
202 static struct ip_vs_conn *
203 ip_vs_sched_persist(struct ip_vs_service *svc,
204 struct sk_buff *skb,
205 __be16 src_port, __be16 dst_port, int *ignored)
206 {
207 struct ip_vs_conn *cp = NULL;
208 struct ip_vs_iphdr iph;
209 struct ip_vs_dest *dest;
210 struct ip_vs_conn *ct;
211 __be16 dport = 0; /* destination port to forward */
212 unsigned int flags;
213 struct ip_vs_conn_param param;
214 union nf_inet_addr snet; /* source network of the client,
215 after masking */
216
217 ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
218
219 /* Mask saddr with the netmask to adjust template granularity */
220 #ifdef CONFIG_IP_VS_IPV6
221 if (svc->af == AF_INET6)
222 ipv6_addr_prefix(&snet.in6, &iph.saddr.in6, svc->netmask);
223 else
224 #endif
225 snet.ip = iph.saddr.ip & svc->netmask;
226
227 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
228 "mnet %s\n",
229 IP_VS_DBG_ADDR(svc->af, &iph.saddr), ntohs(src_port),
230 IP_VS_DBG_ADDR(svc->af, &iph.daddr), ntohs(dst_port),
231 IP_VS_DBG_ADDR(svc->af, &snet));
232
233 /*
234 * As far as we know, FTP is a very complicated network protocol, and
235 * it uses control connection and data connections. For active FTP,
236 * FTP server initialize data connection to the client, its source port
237 * is often 20. For passive FTP, FTP server tells the clients the port
238 * that it passively listens to, and the client issues the data
239 * connection. In the tunneling or direct routing mode, the load
240 * balancer is on the client-to-server half of connection, the port
241 * number is unknown to the load balancer. So, a conn template like
242 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
243 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
244 * is created for other persistent services.
245 */
246 {
247 int protocol = iph.protocol;
248 const union nf_inet_addr *vaddr = &iph.daddr;
249 const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
250 __be16 vport = 0;
251
252 if (dst_port == svc->port) {
253 /* non-FTP template:
254 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
255 * FTP template:
256 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
257 */
258 if (svc->port != FTPPORT)
259 vport = dst_port;
260 } else {
261 /* Note: persistent fwmark-based services and
262 * persistent port zero service are handled here.
263 * fwmark template:
264 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
265 * port zero template:
266 * <protocol,caddr,0,vaddr,0,daddr,0>
267 */
268 if (svc->fwmark) {
269 protocol = IPPROTO_IP;
270 vaddr = &fwmark;
271 }
272 }
273 /* return *ignored = -1 so NF_DROP can be used */
274 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
275 vaddr, vport, &param) < 0) {
276 *ignored = -1;
277 return NULL;
278 }
279 }
280
281 /* Check if a template already exists */
282 ct = ip_vs_ct_in_get(&param);
283 if (!ct || !ip_vs_check_template(ct)) {
284 /*
285 * No template found or the dest of the connection
286 * template is not available.
287 * return *ignored=0 i.e. ICMP and NF_DROP
288 */
289 dest = svc->scheduler->schedule(svc, skb);
290 if (!dest) {
291 IP_VS_DBG(1, "p-schedule: no dest found.\n");
292 kfree(param.pe_data);
293 *ignored = 0;
294 return NULL;
295 }
296
297 if (dst_port == svc->port && svc->port != FTPPORT)
298 dport = dest->port;
299
300 /* Create a template
301 * This adds param.pe_data to the template,
302 * and thus param.pe_data will be destroyed
303 * when the template expires */
304 ct = ip_vs_conn_new(&param, &dest->addr, dport,
305 IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
306 if (ct == NULL) {
307 kfree(param.pe_data);
308 *ignored = -1;
309 return NULL;
310 }
311
312 ct->timeout = svc->timeout;
313 } else {
314 /* set destination with the found template */
315 dest = ct->dest;
316 kfree(param.pe_data);
317 }
318
319 dport = dst_port;
320 if (dport == svc->port && dest->port)
321 dport = dest->port;
322
323 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
324 && iph.protocol == IPPROTO_UDP)?
325 IP_VS_CONN_F_ONE_PACKET : 0;
326
327 /*
328 * Create a new connection according to the template
329 */
330 ip_vs_conn_fill_param(svc->af, iph.protocol, &iph.saddr, src_port,
331 &iph.daddr, dst_port, &param);
332
333 cp = ip_vs_conn_new(&param, &dest->addr, dport, flags, dest, skb->mark);
334 if (cp == NULL) {
335 ip_vs_conn_put(ct);
336 *ignored = -1;
337 return NULL;
338 }
339
340 /*
341 * Add its control
342 */
343 ip_vs_control_add(cp, ct);
344 ip_vs_conn_put(ct);
345
346 ip_vs_conn_stats(cp, svc);
347 return cp;
348 }
349
350
351 /*
352 * IPVS main scheduling function
353 * It selects a server according to the virtual service, and
354 * creates a connection entry.
355 * Protocols supported: TCP, UDP
356 *
357 * Usage of *ignored
358 *
359 * 1 : protocol tried to schedule (eg. on SYN), found svc but the
360 * svc/scheduler decides that this packet should be accepted with
361 * NF_ACCEPT because it must not be scheduled.
362 *
363 * 0 : scheduler can not find destination, so try bypass or
364 * return ICMP and then NF_DROP (ip_vs_leave).
365 *
366 * -1 : scheduler tried to schedule but fatal error occurred, eg.
367 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
368 * failure such as missing Call-ID, ENOMEM on skb_linearize
369 * or pe_data. In this case we should return NF_DROP without
370 * any attempts to send ICMP with ip_vs_leave.
371 */
372 struct ip_vs_conn *
373 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
374 struct ip_vs_protocol *pp, int *ignored)
375 {
376 struct ip_vs_conn *cp = NULL;
377 struct ip_vs_iphdr iph;
378 struct ip_vs_dest *dest;
379 __be16 _ports[2], *pptr;
380 unsigned int flags;
381
382 *ignored = 1;
383 ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
384 pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
385 if (pptr == NULL)
386 return NULL;
387
388 /*
389 * FTPDATA needs this check when using local real server.
390 * Never schedule Active FTPDATA connections from real server.
391 * For LVS-NAT they must be already created. For other methods
392 * with persistence the connection is created on SYN+ACK.
393 */
394 if (pptr[0] == FTPDATA) {
395 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
396 "Not scheduling FTPDATA");
397 return NULL;
398 }
399
400 /*
401 * Do not schedule replies from local real server.
402 */
403 if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
404 (cp = pp->conn_in_get(svc->af, skb, pp, &iph, iph.len, 1))) {
405 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
406 "Not scheduling reply for existing connection");
407 __ip_vs_conn_put(cp);
408 return NULL;
409 }
410
411 /*
412 * Persistent service
413 */
414 if (svc->flags & IP_VS_SVC_F_PERSISTENT)
415 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored);
416
417 *ignored = 0;
418
419 /*
420 * Non-persistent service
421 */
422 if (!svc->fwmark && pptr[1] != svc->port) {
423 if (!svc->port)
424 pr_err("Schedule: port zero only supported "
425 "in persistent services, "
426 "check your ipvs configuration\n");
427 return NULL;
428 }
429
430 dest = svc->scheduler->schedule(svc, skb);
431 if (dest == NULL) {
432 IP_VS_DBG(1, "Schedule: no dest found.\n");
433 return NULL;
434 }
435
436 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
437 && iph.protocol == IPPROTO_UDP)?
438 IP_VS_CONN_F_ONE_PACKET : 0;
439
440 /*
441 * Create a connection entry.
442 */
443 {
444 struct ip_vs_conn_param p;
445 ip_vs_conn_fill_param(svc->af, iph.protocol, &iph.saddr,
446 pptr[0], &iph.daddr, pptr[1], &p);
447 cp = ip_vs_conn_new(&p, &dest->addr,
448 dest->port ? dest->port : pptr[1],
449 flags, dest, skb->mark);
450 if (!cp) {
451 *ignored = -1;
452 return NULL;
453 }
454 }
455
456 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
457 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
458 ip_vs_fwd_tag(cp),
459 IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
460 IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
461 IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
462 cp->flags, atomic_read(&cp->refcnt));
463
464 ip_vs_conn_stats(cp, svc);
465 return cp;
466 }
467
468
469 /*
470 * Pass or drop the packet.
471 * Called by ip_vs_in, when the virtual service is available but
472 * no destination is available for a new connection.
473 */
474 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
475 struct ip_vs_protocol *pp)
476 {
477 __be16 _ports[2], *pptr;
478 struct ip_vs_iphdr iph;
479 int unicast;
480 ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
481
482 pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
483 if (pptr == NULL) {
484 ip_vs_service_put(svc);
485 return NF_DROP;
486 }
487
488 #ifdef CONFIG_IP_VS_IPV6
489 if (svc->af == AF_INET6)
490 unicast = ipv6_addr_type(&iph.daddr.in6) & IPV6_ADDR_UNICAST;
491 else
492 #endif
493 unicast = (inet_addr_type(&init_net, iph.daddr.ip) == RTN_UNICAST);
494
495 /* if it is fwmark-based service, the cache_bypass sysctl is up
496 and the destination is a non-local unicast, then create
497 a cache_bypass connection entry */
498 if (sysctl_ip_vs_cache_bypass && svc->fwmark && unicast) {
499 int ret, cs;
500 struct ip_vs_conn *cp;
501 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
502 iph.protocol == IPPROTO_UDP)?
503 IP_VS_CONN_F_ONE_PACKET : 0;
504 union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } };
505
506 ip_vs_service_put(svc);
507
508 /* create a new connection entry */
509 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
510 {
511 struct ip_vs_conn_param p;
512 ip_vs_conn_fill_param(svc->af, iph.protocol,
513 &iph.saddr, pptr[0],
514 &iph.daddr, pptr[1], &p);
515 cp = ip_vs_conn_new(&p, &daddr, 0,
516 IP_VS_CONN_F_BYPASS | flags,
517 NULL, skb->mark);
518 if (!cp)
519 return NF_DROP;
520 }
521
522 /* statistics */
523 ip_vs_in_stats(cp, skb);
524
525 /* set state */
526 cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
527
528 /* transmit the first SYN packet */
529 ret = cp->packet_xmit(skb, cp, pp);
530 /* do not touch skb anymore */
531
532 atomic_inc(&cp->in_pkts);
533 ip_vs_conn_put(cp);
534 return ret;
535 }
536
537 /*
538 * When the virtual ftp service is presented, packets destined
539 * for other services on the VIP may get here (except services
540 * listed in the ipvs table), pass the packets, because it is
541 * not ipvs job to decide to drop the packets.
542 */
543 if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
544 ip_vs_service_put(svc);
545 return NF_ACCEPT;
546 }
547
548 ip_vs_service_put(svc);
549
550 /*
551 * Notify the client that the destination is unreachable, and
552 * release the socket buffer.
553 * Since it is in IP layer, the TCP socket is not actually
554 * created, the TCP RST packet cannot be sent, instead that
555 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
556 */
557 #ifdef CONFIG_IP_VS_IPV6
558 if (svc->af == AF_INET6) {
559 if (!skb->dev) {
560 struct net *net = dev_net(skb_dst(skb)->dev);
561
562 skb->dev = net->loopback_dev;
563 }
564 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
565 } else
566 #endif
567 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
568
569 return NF_DROP;
570 }
571
572 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
573 {
574 return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
575 }
576
577 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
578 {
579 if (NF_INET_LOCAL_IN == hooknum)
580 return IP_DEFRAG_VS_IN;
581 if (NF_INET_FORWARD == hooknum)
582 return IP_DEFRAG_VS_FWD;
583 return IP_DEFRAG_VS_OUT;
584 }
585
586 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
587 {
588 int err = ip_defrag(skb, user);
589
590 if (!err)
591 ip_send_check(ip_hdr(skb));
592
593 return err;
594 }
595
596 #ifdef CONFIG_IP_VS_IPV6
597 static inline int ip_vs_gather_frags_v6(struct sk_buff *skb, u_int32_t user)
598 {
599 /* TODO IPv6: Find out what to do here for IPv6 */
600 return 0;
601 }
602 #endif
603
604 /*
605 * Packet has been made sufficiently writable in caller
606 * - inout: 1=in->out, 0=out->in
607 */
608 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
609 struct ip_vs_conn *cp, int inout)
610 {
611 struct iphdr *iph = ip_hdr(skb);
612 unsigned int icmp_offset = iph->ihl*4;
613 struct icmphdr *icmph = (struct icmphdr *)(skb_network_header(skb) +
614 icmp_offset);
615 struct iphdr *ciph = (struct iphdr *)(icmph + 1);
616
617 if (inout) {
618 iph->saddr = cp->vaddr.ip;
619 ip_send_check(iph);
620 ciph->daddr = cp->vaddr.ip;
621 ip_send_check(ciph);
622 } else {
623 iph->daddr = cp->daddr.ip;
624 ip_send_check(iph);
625 ciph->saddr = cp->daddr.ip;
626 ip_send_check(ciph);
627 }
628
629 /* the TCP/UDP/SCTP port */
630 if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
631 IPPROTO_SCTP == ciph->protocol) {
632 __be16 *ports = (void *)ciph + ciph->ihl*4;
633
634 if (inout)
635 ports[1] = cp->vport;
636 else
637 ports[0] = cp->dport;
638 }
639
640 /* And finally the ICMP checksum */
641 icmph->checksum = 0;
642 icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
643 skb->ip_summed = CHECKSUM_UNNECESSARY;
644
645 if (inout)
646 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
647 "Forwarding altered outgoing ICMP");
648 else
649 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
650 "Forwarding altered incoming ICMP");
651 }
652
653 #ifdef CONFIG_IP_VS_IPV6
654 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
655 struct ip_vs_conn *cp, int inout)
656 {
657 struct ipv6hdr *iph = ipv6_hdr(skb);
658 unsigned int icmp_offset = sizeof(struct ipv6hdr);
659 struct icmp6hdr *icmph = (struct icmp6hdr *)(skb_network_header(skb) +
660 icmp_offset);
661 struct ipv6hdr *ciph = (struct ipv6hdr *)(icmph + 1);
662
663 if (inout) {
664 iph->saddr = cp->vaddr.in6;
665 ciph->daddr = cp->vaddr.in6;
666 } else {
667 iph->daddr = cp->daddr.in6;
668 ciph->saddr = cp->daddr.in6;
669 }
670
671 /* the TCP/UDP/SCTP port */
672 if (IPPROTO_TCP == ciph->nexthdr || IPPROTO_UDP == ciph->nexthdr ||
673 IPPROTO_SCTP == ciph->nexthdr) {
674 __be16 *ports = (void *)ciph + sizeof(struct ipv6hdr);
675
676 if (inout)
677 ports[1] = cp->vport;
678 else
679 ports[0] = cp->dport;
680 }
681
682 /* And finally the ICMP checksum */
683 icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
684 skb->len - icmp_offset,
685 IPPROTO_ICMPV6, 0);
686 skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
687 skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
688 skb->ip_summed = CHECKSUM_PARTIAL;
689
690 if (inout)
691 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
692 (void *)ciph - (void *)iph,
693 "Forwarding altered outgoing ICMPv6");
694 else
695 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
696 (void *)ciph - (void *)iph,
697 "Forwarding altered incoming ICMPv6");
698 }
699 #endif
700
701 /* Handle relevant response ICMP messages - forward to the right
702 * destination host. Used for NAT and local client.
703 */
704 static int handle_response_icmp(int af, struct sk_buff *skb,
705 union nf_inet_addr *snet,
706 __u8 protocol, struct ip_vs_conn *cp,
707 struct ip_vs_protocol *pp,
708 unsigned int offset, unsigned int ihl)
709 {
710 unsigned int verdict = NF_DROP;
711
712 if (IP_VS_FWD_METHOD(cp) != 0) {
713 pr_err("shouldn't reach here, because the box is on the "
714 "half connection in the tun/dr module.\n");
715 }
716
717 /* Ensure the checksum is correct */
718 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
719 /* Failed checksum! */
720 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
721 IP_VS_DBG_ADDR(af, snet));
722 goto out;
723 }
724
725 if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
726 IPPROTO_SCTP == protocol)
727 offset += 2 * sizeof(__u16);
728 if (!skb_make_writable(skb, offset))
729 goto out;
730
731 #ifdef CONFIG_IP_VS_IPV6
732 if (af == AF_INET6)
733 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
734 else
735 #endif
736 ip_vs_nat_icmp(skb, pp, cp, 1);
737
738 #ifdef CONFIG_IP_VS_IPV6
739 if (af == AF_INET6) {
740 if (sysctl_ip_vs_snat_reroute && ip6_route_me_harder(skb) != 0)
741 goto out;
742 } else
743 #endif
744 if ((sysctl_ip_vs_snat_reroute ||
745 skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
746 ip_route_me_harder(skb, RTN_LOCAL) != 0)
747 goto out;
748
749 /* do the statistics and put it back */
750 ip_vs_out_stats(cp, skb);
751
752 skb->ipvs_property = 1;
753 if (!(cp->flags & IP_VS_CONN_F_NFCT))
754 ip_vs_notrack(skb);
755 else
756 ip_vs_update_conntrack(skb, cp, 0);
757 verdict = NF_ACCEPT;
758
759 out:
760 __ip_vs_conn_put(cp);
761
762 return verdict;
763 }
764
765 /*
766 * Handle ICMP messages in the inside-to-outside direction (outgoing).
767 * Find any that might be relevant, check against existing connections.
768 * Currently handles error types - unreachable, quench, ttl exceeded.
769 */
770 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
771 unsigned int hooknum)
772 {
773 struct iphdr *iph;
774 struct icmphdr _icmph, *ic;
775 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
776 struct ip_vs_iphdr ciph;
777 struct ip_vs_conn *cp;
778 struct ip_vs_protocol *pp;
779 unsigned int offset, ihl;
780 union nf_inet_addr snet;
781
782 *related = 1;
783
784 /* reassemble IP fragments */
785 if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
786 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
787 return NF_STOLEN;
788 }
789
790 iph = ip_hdr(skb);
791 offset = ihl = iph->ihl * 4;
792 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
793 if (ic == NULL)
794 return NF_DROP;
795
796 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
797 ic->type, ntohs(icmp_id(ic)),
798 &iph->saddr, &iph->daddr);
799
800 /*
801 * Work through seeing if this is for us.
802 * These checks are supposed to be in an order that means easy
803 * things are checked first to speed up processing.... however
804 * this means that some packets will manage to get a long way
805 * down this stack and then be rejected, but that's life.
806 */
807 if ((ic->type != ICMP_DEST_UNREACH) &&
808 (ic->type != ICMP_SOURCE_QUENCH) &&
809 (ic->type != ICMP_TIME_EXCEEDED)) {
810 *related = 0;
811 return NF_ACCEPT;
812 }
813
814 /* Now find the contained IP header */
815 offset += sizeof(_icmph);
816 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
817 if (cih == NULL)
818 return NF_ACCEPT; /* The packet looks wrong, ignore */
819
820 pp = ip_vs_proto_get(cih->protocol);
821 if (!pp)
822 return NF_ACCEPT;
823
824 /* Is the embedded protocol header present? */
825 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
826 pp->dont_defrag))
827 return NF_ACCEPT;
828
829 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
830 "Checking outgoing ICMP for");
831
832 offset += cih->ihl * 4;
833
834 ip_vs_fill_iphdr(AF_INET, cih, &ciph);
835 /* The embedded headers contain source and dest in reverse order */
836 cp = pp->conn_out_get(AF_INET, skb, pp, &ciph, offset, 1);
837 if (!cp)
838 return NF_ACCEPT;
839
840 snet.ip = iph->saddr;
841 return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
842 pp, offset, ihl);
843 }
844
845 #ifdef CONFIG_IP_VS_IPV6
846 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
847 unsigned int hooknum)
848 {
849 struct ipv6hdr *iph;
850 struct icmp6hdr _icmph, *ic;
851 struct ipv6hdr _ciph, *cih; /* The ip header contained
852 within the ICMP */
853 struct ip_vs_iphdr ciph;
854 struct ip_vs_conn *cp;
855 struct ip_vs_protocol *pp;
856 unsigned int offset;
857 union nf_inet_addr snet;
858
859 *related = 1;
860
861 /* reassemble IP fragments */
862 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
863 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
864 return NF_STOLEN;
865 }
866
867 iph = ipv6_hdr(skb);
868 offset = sizeof(struct ipv6hdr);
869 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
870 if (ic == NULL)
871 return NF_DROP;
872
873 IP_VS_DBG(12, "Outgoing ICMPv6 (%d,%d) %pI6->%pI6\n",
874 ic->icmp6_type, ntohs(icmpv6_id(ic)),
875 &iph->saddr, &iph->daddr);
876
877 /*
878 * Work through seeing if this is for us.
879 * These checks are supposed to be in an order that means easy
880 * things are checked first to speed up processing.... however
881 * this means that some packets will manage to get a long way
882 * down this stack and then be rejected, but that's life.
883 */
884 if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
885 (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
886 (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
887 *related = 0;
888 return NF_ACCEPT;
889 }
890
891 /* Now find the contained IP header */
892 offset += sizeof(_icmph);
893 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
894 if (cih == NULL)
895 return NF_ACCEPT; /* The packet looks wrong, ignore */
896
897 pp = ip_vs_proto_get(cih->nexthdr);
898 if (!pp)
899 return NF_ACCEPT;
900
901 /* Is the embedded protocol header present? */
902 /* TODO: we don't support fragmentation at the moment anyways */
903 if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
904 return NF_ACCEPT;
905
906 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
907 "Checking outgoing ICMPv6 for");
908
909 offset += sizeof(struct ipv6hdr);
910
911 ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
912 /* The embedded headers contain source and dest in reverse order */
913 cp = pp->conn_out_get(AF_INET6, skb, pp, &ciph, offset, 1);
914 if (!cp)
915 return NF_ACCEPT;
916
917 ipv6_addr_copy(&snet.in6, &iph->saddr);
918 return handle_response_icmp(AF_INET6, skb, &snet, cih->nexthdr, cp,
919 pp, offset, sizeof(struct ipv6hdr));
920 }
921 #endif
922
923 /*
924 * Check if sctp chunc is ABORT chunk
925 */
926 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
927 {
928 sctp_chunkhdr_t *sch, schunk;
929 sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
930 sizeof(schunk), &schunk);
931 if (sch == NULL)
932 return 0;
933 if (sch->type == SCTP_CID_ABORT)
934 return 1;
935 return 0;
936 }
937
938 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
939 {
940 struct tcphdr _tcph, *th;
941
942 th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
943 if (th == NULL)
944 return 0;
945 return th->rst;
946 }
947
948 /* Handle response packets: rewrite addresses and send away...
949 * Used for NAT and local client.
950 */
951 static unsigned int
952 handle_response(int af, struct sk_buff *skb, struct ip_vs_protocol *pp,
953 struct ip_vs_conn *cp, int ihl)
954 {
955 IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
956
957 if (!skb_make_writable(skb, ihl))
958 goto drop;
959
960 /* mangle the packet */
961 if (pp->snat_handler && !pp->snat_handler(skb, pp, cp))
962 goto drop;
963
964 #ifdef CONFIG_IP_VS_IPV6
965 if (af == AF_INET6)
966 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
967 else
968 #endif
969 {
970 ip_hdr(skb)->saddr = cp->vaddr.ip;
971 ip_send_check(ip_hdr(skb));
972 }
973
974 /*
975 * nf_iterate does not expect change in the skb->dst->dev.
976 * It looks like it is not fatal to enable this code for hooks
977 * where our handlers are at the end of the chain list and
978 * when all next handlers use skb->dst->dev and not outdev.
979 * It will definitely route properly the inout NAT traffic
980 * when multiple paths are used.
981 */
982
983 /* For policy routing, packets originating from this
984 * machine itself may be routed differently to packets
985 * passing through. We want this packet to be routed as
986 * if it came from this machine itself. So re-compute
987 * the routing information.
988 */
989 #ifdef CONFIG_IP_VS_IPV6
990 if (af == AF_INET6) {
991 if (sysctl_ip_vs_snat_reroute && ip6_route_me_harder(skb) != 0)
992 goto drop;
993 } else
994 #endif
995 if ((sysctl_ip_vs_snat_reroute ||
996 skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
997 ip_route_me_harder(skb, RTN_LOCAL) != 0)
998 goto drop;
999
1000 IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1001
1002 ip_vs_out_stats(cp, skb);
1003 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pp);
1004 skb->ipvs_property = 1;
1005 if (!(cp->flags & IP_VS_CONN_F_NFCT))
1006 ip_vs_notrack(skb);
1007 else
1008 ip_vs_update_conntrack(skb, cp, 0);
1009 ip_vs_conn_put(cp);
1010
1011 LeaveFunction(11);
1012 return NF_ACCEPT;
1013
1014 drop:
1015 ip_vs_conn_put(cp);
1016 kfree_skb(skb);
1017 LeaveFunction(11);
1018 return NF_STOLEN;
1019 }
1020
1021 /*
1022 * Check if outgoing packet belongs to the established ip_vs_conn.
1023 */
1024 static unsigned int
1025 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1026 {
1027 struct ip_vs_iphdr iph;
1028 struct ip_vs_protocol *pp;
1029 struct ip_vs_conn *cp;
1030
1031 EnterFunction(11);
1032
1033 /* Already marked as IPVS request or reply? */
1034 if (skb->ipvs_property)
1035 return NF_ACCEPT;
1036
1037 /* Bad... Do not break raw sockets */
1038 if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1039 af == AF_INET)) {
1040 struct sock *sk = skb->sk;
1041 struct inet_sock *inet = inet_sk(skb->sk);
1042
1043 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1044 return NF_ACCEPT;
1045 }
1046
1047 if (unlikely(!skb_dst(skb)))
1048 return NF_ACCEPT;
1049
1050 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1051 #ifdef CONFIG_IP_VS_IPV6
1052 if (af == AF_INET6) {
1053 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1054 int related;
1055 int verdict = ip_vs_out_icmp_v6(skb, &related,
1056 hooknum);
1057
1058 if (related)
1059 return verdict;
1060 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1061 }
1062 } else
1063 #endif
1064 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1065 int related;
1066 int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1067
1068 if (related)
1069 return verdict;
1070 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1071 }
1072
1073 pp = ip_vs_proto_get(iph.protocol);
1074 if (unlikely(!pp))
1075 return NF_ACCEPT;
1076
1077 /* reassemble IP fragments */
1078 #ifdef CONFIG_IP_VS_IPV6
1079 if (af == AF_INET6) {
1080 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1081 if (ip_vs_gather_frags_v6(skb,
1082 ip_vs_defrag_user(hooknum)))
1083 return NF_STOLEN;
1084 }
1085
1086 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1087 } else
1088 #endif
1089 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_MF|IP_OFFSET) &&
1090 !pp->dont_defrag)) {
1091 if (ip_vs_gather_frags(skb,
1092 ip_vs_defrag_user(hooknum)))
1093 return NF_STOLEN;
1094
1095 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1096 }
1097
1098 /*
1099 * Check if the packet belongs to an existing entry
1100 */
1101 cp = pp->conn_out_get(af, skb, pp, &iph, iph.len, 0);
1102
1103 if (likely(cp))
1104 return handle_response(af, skb, pp, cp, iph.len);
1105 if (sysctl_ip_vs_nat_icmp_send &&
1106 (pp->protocol == IPPROTO_TCP ||
1107 pp->protocol == IPPROTO_UDP ||
1108 pp->protocol == IPPROTO_SCTP)) {
1109 __be16 _ports[2], *pptr;
1110
1111 pptr = skb_header_pointer(skb, iph.len,
1112 sizeof(_ports), _ports);
1113 if (pptr == NULL)
1114 return NF_ACCEPT; /* Not for me */
1115 if (ip_vs_lookup_real_service(af, iph.protocol,
1116 &iph.saddr,
1117 pptr[0])) {
1118 /*
1119 * Notify the real server: there is no
1120 * existing entry if it is not RST
1121 * packet or not TCP packet.
1122 */
1123 if ((iph.protocol != IPPROTO_TCP &&
1124 iph.protocol != IPPROTO_SCTP)
1125 || ((iph.protocol == IPPROTO_TCP
1126 && !is_tcp_reset(skb, iph.len))
1127 || (iph.protocol == IPPROTO_SCTP
1128 && !is_sctp_abort(skb,
1129 iph.len)))) {
1130 #ifdef CONFIG_IP_VS_IPV6
1131 if (af == AF_INET6) {
1132 struct net *net =
1133 dev_net(skb_dst(skb)->dev);
1134
1135 if (!skb->dev)
1136 skb->dev = net->loopback_dev;
1137 icmpv6_send(skb,
1138 ICMPV6_DEST_UNREACH,
1139 ICMPV6_PORT_UNREACH,
1140 0);
1141 } else
1142 #endif
1143 icmp_send(skb,
1144 ICMP_DEST_UNREACH,
1145 ICMP_PORT_UNREACH, 0);
1146 return NF_DROP;
1147 }
1148 }
1149 }
1150 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1151 "ip_vs_out: packet continues traversal as normal");
1152 return NF_ACCEPT;
1153 }
1154
1155 /*
1156 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1157 * used only for VS/NAT.
1158 * Check if packet is reply for established ip_vs_conn.
1159 */
1160 static unsigned int
1161 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1162 const struct net_device *in, const struct net_device *out,
1163 int (*okfn)(struct sk_buff *))
1164 {
1165 return ip_vs_out(hooknum, skb, AF_INET);
1166 }
1167
1168 /*
1169 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1170 * Check if packet is reply for established ip_vs_conn.
1171 */
1172 static unsigned int
1173 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1174 const struct net_device *in, const struct net_device *out,
1175 int (*okfn)(struct sk_buff *))
1176 {
1177 unsigned int verdict;
1178
1179 /* Disable BH in LOCAL_OUT until all places are fixed */
1180 local_bh_disable();
1181 verdict = ip_vs_out(hooknum, skb, AF_INET);
1182 local_bh_enable();
1183 return verdict;
1184 }
1185
1186 #ifdef CONFIG_IP_VS_IPV6
1187
1188 /*
1189 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1190 * used only for VS/NAT.
1191 * Check if packet is reply for established ip_vs_conn.
1192 */
1193 static unsigned int
1194 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1195 const struct net_device *in, const struct net_device *out,
1196 int (*okfn)(struct sk_buff *))
1197 {
1198 return ip_vs_out(hooknum, skb, AF_INET6);
1199 }
1200
1201 /*
1202 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1203 * Check if packet is reply for established ip_vs_conn.
1204 */
1205 static unsigned int
1206 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1207 const struct net_device *in, const struct net_device *out,
1208 int (*okfn)(struct sk_buff *))
1209 {
1210 unsigned int verdict;
1211
1212 /* Disable BH in LOCAL_OUT until all places are fixed */
1213 local_bh_disable();
1214 verdict = ip_vs_out(hooknum, skb, AF_INET6);
1215 local_bh_enable();
1216 return verdict;
1217 }
1218
1219 #endif
1220
1221 /*
1222 * Handle ICMP messages in the outside-to-inside direction (incoming).
1223 * Find any that might be relevant, check against existing connections,
1224 * forward to the right destination host if relevant.
1225 * Currently handles error types - unreachable, quench, ttl exceeded.
1226 */
1227 static int
1228 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1229 {
1230 struct iphdr *iph;
1231 struct icmphdr _icmph, *ic;
1232 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */
1233 struct ip_vs_iphdr ciph;
1234 struct ip_vs_conn *cp;
1235 struct ip_vs_protocol *pp;
1236 unsigned int offset, ihl, verdict;
1237 union nf_inet_addr snet;
1238
1239 *related = 1;
1240
1241 /* reassemble IP fragments */
1242 if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
1243 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1244 return NF_STOLEN;
1245 }
1246
1247 iph = ip_hdr(skb);
1248 offset = ihl = iph->ihl * 4;
1249 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1250 if (ic == NULL)
1251 return NF_DROP;
1252
1253 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1254 ic->type, ntohs(icmp_id(ic)),
1255 &iph->saddr, &iph->daddr);
1256
1257 /*
1258 * Work through seeing if this is for us.
1259 * These checks are supposed to be in an order that means easy
1260 * things are checked first to speed up processing.... however
1261 * this means that some packets will manage to get a long way
1262 * down this stack and then be rejected, but that's life.
1263 */
1264 if ((ic->type != ICMP_DEST_UNREACH) &&
1265 (ic->type != ICMP_SOURCE_QUENCH) &&
1266 (ic->type != ICMP_TIME_EXCEEDED)) {
1267 *related = 0;
1268 return NF_ACCEPT;
1269 }
1270
1271 /* Now find the contained IP header */
1272 offset += sizeof(_icmph);
1273 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1274 if (cih == NULL)
1275 return NF_ACCEPT; /* The packet looks wrong, ignore */
1276
1277 pp = ip_vs_proto_get(cih->protocol);
1278 if (!pp)
1279 return NF_ACCEPT;
1280
1281 /* Is the embedded protocol header present? */
1282 if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1283 pp->dont_defrag))
1284 return NF_ACCEPT;
1285
1286 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1287 "Checking incoming ICMP for");
1288
1289 offset += cih->ihl * 4;
1290
1291 ip_vs_fill_iphdr(AF_INET, cih, &ciph);
1292 /* The embedded headers contain source and dest in reverse order */
1293 cp = pp->conn_in_get(AF_INET, skb, pp, &ciph, offset, 1);
1294 if (!cp) {
1295 /* The packet could also belong to a local client */
1296 cp = pp->conn_out_get(AF_INET, skb, pp, &ciph, offset, 1);
1297 if (cp) {
1298 snet.ip = iph->saddr;
1299 return handle_response_icmp(AF_INET, skb, &snet,
1300 cih->protocol, cp, pp,
1301 offset, ihl);
1302 }
1303 return NF_ACCEPT;
1304 }
1305
1306 verdict = NF_DROP;
1307
1308 /* Ensure the checksum is correct */
1309 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1310 /* Failed checksum! */
1311 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1312 &iph->saddr);
1313 goto out;
1314 }
1315
1316 /* do the statistics and put it back */
1317 ip_vs_in_stats(cp, skb);
1318 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1319 offset += 2 * sizeof(__u16);
1320 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
1321 /* LOCALNODE from FORWARD hook is not supported */
1322 if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD &&
1323 skb_rtable(skb)->rt_flags & RTCF_LOCAL) {
1324 IP_VS_DBG(1, "%s(): "
1325 "local delivery to %pI4 but in FORWARD\n",
1326 __func__, &skb_rtable(skb)->rt_dst);
1327 verdict = NF_DROP;
1328 }
1329
1330 out:
1331 __ip_vs_conn_put(cp);
1332
1333 return verdict;
1334 }
1335
1336 #ifdef CONFIG_IP_VS_IPV6
1337 static int
1338 ip_vs_in_icmp_v6(struct sk_buff *skb, int *related, unsigned int hooknum)
1339 {
1340 struct ipv6hdr *iph;
1341 struct icmp6hdr _icmph, *ic;
1342 struct ipv6hdr _ciph, *cih; /* The ip header contained
1343 within the ICMP */
1344 struct ip_vs_iphdr ciph;
1345 struct ip_vs_conn *cp;
1346 struct ip_vs_protocol *pp;
1347 unsigned int offset, verdict;
1348 union nf_inet_addr snet;
1349 struct rt6_info *rt;
1350
1351 *related = 1;
1352
1353 /* reassemble IP fragments */
1354 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1355 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
1356 return NF_STOLEN;
1357 }
1358
1359 iph = ipv6_hdr(skb);
1360 offset = sizeof(struct ipv6hdr);
1361 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1362 if (ic == NULL)
1363 return NF_DROP;
1364
1365 IP_VS_DBG(12, "Incoming ICMPv6 (%d,%d) %pI6->%pI6\n",
1366 ic->icmp6_type, ntohs(icmpv6_id(ic)),
1367 &iph->saddr, &iph->daddr);
1368
1369 /*
1370 * Work through seeing if this is for us.
1371 * These checks are supposed to be in an order that means easy
1372 * things are checked first to speed up processing.... however
1373 * this means that some packets will manage to get a long way
1374 * down this stack and then be rejected, but that's life.
1375 */
1376 if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
1377 (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
1378 (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
1379 *related = 0;
1380 return NF_ACCEPT;
1381 }
1382
1383 /* Now find the contained IP header */
1384 offset += sizeof(_icmph);
1385 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1386 if (cih == NULL)
1387 return NF_ACCEPT; /* The packet looks wrong, ignore */
1388
1389 pp = ip_vs_proto_get(cih->nexthdr);
1390 if (!pp)
1391 return NF_ACCEPT;
1392
1393 /* Is the embedded protocol header present? */
1394 /* TODO: we don't support fragmentation at the moment anyways */
1395 if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
1396 return NF_ACCEPT;
1397
1398 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1399 "Checking incoming ICMPv6 for");
1400
1401 offset += sizeof(struct ipv6hdr);
1402
1403 ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
1404 /* The embedded headers contain source and dest in reverse order */
1405 cp = pp->conn_in_get(AF_INET6, skb, pp, &ciph, offset, 1);
1406 if (!cp) {
1407 /* The packet could also belong to a local client */
1408 cp = pp->conn_out_get(AF_INET6, skb, pp, &ciph, offset, 1);
1409 if (cp) {
1410 ipv6_addr_copy(&snet.in6, &iph->saddr);
1411 return handle_response_icmp(AF_INET6, skb, &snet,
1412 cih->nexthdr,
1413 cp, pp, offset,
1414 sizeof(struct ipv6hdr));
1415 }
1416 return NF_ACCEPT;
1417 }
1418
1419 verdict = NF_DROP;
1420
1421 /* do the statistics and put it back */
1422 ip_vs_in_stats(cp, skb);
1423 if (IPPROTO_TCP == cih->nexthdr || IPPROTO_UDP == cih->nexthdr ||
1424 IPPROTO_SCTP == cih->nexthdr)
1425 offset += 2 * sizeof(__u16);
1426 verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset);
1427 /* LOCALNODE from FORWARD hook is not supported */
1428 if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD &&
1429 (rt = (struct rt6_info *) skb_dst(skb)) &&
1430 rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK) {
1431 IP_VS_DBG(1, "%s(): "
1432 "local delivery to %pI6 but in FORWARD\n",
1433 __func__, &rt->rt6i_dst);
1434 verdict = NF_DROP;
1435 }
1436
1437 __ip_vs_conn_put(cp);
1438
1439 return verdict;
1440 }
1441 #endif
1442
1443
1444 /*
1445 * Check if it's for virtual services, look it up,
1446 * and send it on its way...
1447 */
1448 static unsigned int
1449 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1450 {
1451 struct ip_vs_iphdr iph;
1452 struct ip_vs_protocol *pp;
1453 struct ip_vs_conn *cp;
1454 int ret, restart, pkts;
1455
1456 /* Already marked as IPVS request or reply? */
1457 if (skb->ipvs_property)
1458 return NF_ACCEPT;
1459
1460 /*
1461 * Big tappo:
1462 * - remote client: only PACKET_HOST
1463 * - route: used for struct net when skb->dev is unset
1464 */
1465 if (unlikely((skb->pkt_type != PACKET_HOST &&
1466 hooknum != NF_INET_LOCAL_OUT) ||
1467 !skb_dst(skb))) {
1468 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1469 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1470 " ignored in hook %u\n",
1471 skb->pkt_type, iph.protocol,
1472 IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1473 return NF_ACCEPT;
1474 }
1475 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1476
1477 /* Bad... Do not break raw sockets */
1478 if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1479 af == AF_INET)) {
1480 struct sock *sk = skb->sk;
1481 struct inet_sock *inet = inet_sk(skb->sk);
1482
1483 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1484 return NF_ACCEPT;
1485 }
1486
1487 #ifdef CONFIG_IP_VS_IPV6
1488 if (af == AF_INET6) {
1489 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1490 int related;
1491 int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
1492
1493 if (related)
1494 return verdict;
1495 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1496 }
1497 } else
1498 #endif
1499 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1500 int related;
1501 int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1502
1503 if (related)
1504 return verdict;
1505 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1506 }
1507
1508 /* Protocol supported? */
1509 pp = ip_vs_proto_get(iph.protocol);
1510 if (unlikely(!pp))
1511 return NF_ACCEPT;
1512
1513 /*
1514 * Check if the packet belongs to an existing connection entry
1515 */
1516 cp = pp->conn_in_get(af, skb, pp, &iph, iph.len, 0);
1517
1518 if (unlikely(!cp)) {
1519 int v;
1520
1521 if (!pp->conn_schedule(af, skb, pp, &v, &cp))
1522 return v;
1523 }
1524
1525 if (unlikely(!cp)) {
1526 /* sorry, all this trouble for a no-hit :) */
1527 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1528 "ip_vs_in: packet continues traversal as normal");
1529 return NF_ACCEPT;
1530 }
1531
1532 IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1533
1534 /* Check the server status */
1535 if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1536 /* the destination server is not available */
1537
1538 if (sysctl_ip_vs_expire_nodest_conn) {
1539 /* try to expire the connection immediately */
1540 ip_vs_conn_expire_now(cp);
1541 }
1542 /* don't restart its timer, and silently
1543 drop the packet. */
1544 __ip_vs_conn_put(cp);
1545 return NF_DROP;
1546 }
1547
1548 ip_vs_in_stats(cp, skb);
1549 restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
1550 if (cp->packet_xmit)
1551 ret = cp->packet_xmit(skb, cp, pp);
1552 /* do not touch skb anymore */
1553 else {
1554 IP_VS_DBG_RL("warning: packet_xmit is null");
1555 ret = NF_ACCEPT;
1556 }
1557
1558 /* Increase its packet counter and check if it is needed
1559 * to be synchronized
1560 *
1561 * Sync connection if it is about to close to
1562 * encorage the standby servers to update the connections timeout
1563 */
1564 pkts = atomic_add_return(1, &cp->in_pkts);
1565 if (af == AF_INET && (ip_vs_sync_state & IP_VS_STATE_MASTER) &&
1566 cp->protocol == IPPROTO_SCTP) {
1567 if ((cp->state == IP_VS_SCTP_S_ESTABLISHED &&
1568 (pkts % sysctl_ip_vs_sync_threshold[1]
1569 == sysctl_ip_vs_sync_threshold[0])) ||
1570 (cp->old_state != cp->state &&
1571 ((cp->state == IP_VS_SCTP_S_CLOSED) ||
1572 (cp->state == IP_VS_SCTP_S_SHUT_ACK_CLI) ||
1573 (cp->state == IP_VS_SCTP_S_SHUT_ACK_SER)))) {
1574 ip_vs_sync_conn(cp);
1575 goto out;
1576 }
1577 }
1578
1579 /* Keep this block last: TCP and others with pp->num_states <= 1 */
1580 else if (af == AF_INET &&
1581 (ip_vs_sync_state & IP_VS_STATE_MASTER) &&
1582 (((cp->protocol != IPPROTO_TCP ||
1583 cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1584 (pkts % sysctl_ip_vs_sync_threshold[1]
1585 == sysctl_ip_vs_sync_threshold[0])) ||
1586 ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) &&
1587 ((cp->state == IP_VS_TCP_S_FIN_WAIT) ||
1588 (cp->state == IP_VS_TCP_S_CLOSE) ||
1589 (cp->state == IP_VS_TCP_S_CLOSE_WAIT) ||
1590 (cp->state == IP_VS_TCP_S_TIME_WAIT)))))
1591 ip_vs_sync_conn(cp);
1592 out:
1593 cp->old_state = cp->state;
1594
1595 ip_vs_conn_put(cp);
1596 return ret;
1597 }
1598
1599 /*
1600 * AF_INET handler in NF_INET_LOCAL_IN chain
1601 * Schedule and forward packets from remote clients
1602 */
1603 static unsigned int
1604 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1605 const struct net_device *in,
1606 const struct net_device *out,
1607 int (*okfn)(struct sk_buff *))
1608 {
1609 return ip_vs_in(hooknum, skb, AF_INET);
1610 }
1611
1612 /*
1613 * AF_INET handler in NF_INET_LOCAL_OUT chain
1614 * Schedule and forward packets from local clients
1615 */
1616 static unsigned int
1617 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1618 const struct net_device *in, const struct net_device *out,
1619 int (*okfn)(struct sk_buff *))
1620 {
1621 unsigned int verdict;
1622
1623 /* Disable BH in LOCAL_OUT until all places are fixed */
1624 local_bh_disable();
1625 verdict = ip_vs_in(hooknum, skb, AF_INET);
1626 local_bh_enable();
1627 return verdict;
1628 }
1629
1630 #ifdef CONFIG_IP_VS_IPV6
1631
1632 /*
1633 * AF_INET6 handler in NF_INET_LOCAL_IN chain
1634 * Schedule and forward packets from remote clients
1635 */
1636 static unsigned int
1637 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1638 const struct net_device *in,
1639 const struct net_device *out,
1640 int (*okfn)(struct sk_buff *))
1641 {
1642 return ip_vs_in(hooknum, skb, AF_INET6);
1643 }
1644
1645 /*
1646 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
1647 * Schedule and forward packets from local clients
1648 */
1649 static unsigned int
1650 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1651 const struct net_device *in, const struct net_device *out,
1652 int (*okfn)(struct sk_buff *))
1653 {
1654 unsigned int verdict;
1655
1656 /* Disable BH in LOCAL_OUT until all places are fixed */
1657 local_bh_disable();
1658 verdict = ip_vs_in(hooknum, skb, AF_INET6);
1659 local_bh_enable();
1660 return verdict;
1661 }
1662
1663 #endif
1664
1665
1666 /*
1667 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1668 * related packets destined for 0.0.0.0/0.
1669 * When fwmark-based virtual service is used, such as transparent
1670 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
1671 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1672 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1673 * and send them to ip_vs_in_icmp.
1674 */
1675 static unsigned int
1676 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1677 const struct net_device *in, const struct net_device *out,
1678 int (*okfn)(struct sk_buff *))
1679 {
1680 int r;
1681
1682 if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1683 return NF_ACCEPT;
1684
1685 return ip_vs_in_icmp(skb, &r, hooknum);
1686 }
1687
1688 #ifdef CONFIG_IP_VS_IPV6
1689 static unsigned int
1690 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1691 const struct net_device *in, const struct net_device *out,
1692 int (*okfn)(struct sk_buff *))
1693 {
1694 int r;
1695
1696 if (ipv6_hdr(skb)->nexthdr != IPPROTO_ICMPV6)
1697 return NF_ACCEPT;
1698
1699 return ip_vs_in_icmp_v6(skb, &r, hooknum);
1700 }
1701 #endif
1702
1703
1704 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1705 /* After packet filtering, change source only for VS/NAT */
1706 {
1707 .hook = ip_vs_reply4,
1708 .owner = THIS_MODULE,
1709 .pf = PF_INET,
1710 .hooknum = NF_INET_LOCAL_IN,
1711 .priority = 99,
1712 },
1713 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1714 * or VS/NAT(change destination), so that filtering rules can be
1715 * applied to IPVS. */
1716 {
1717 .hook = ip_vs_remote_request4,
1718 .owner = THIS_MODULE,
1719 .pf = PF_INET,
1720 .hooknum = NF_INET_LOCAL_IN,
1721 .priority = 101,
1722 },
1723 /* Before ip_vs_in, change source only for VS/NAT */
1724 {
1725 .hook = ip_vs_local_reply4,
1726 .owner = THIS_MODULE,
1727 .pf = PF_INET,
1728 .hooknum = NF_INET_LOCAL_OUT,
1729 .priority = -99,
1730 },
1731 /* After mangle, schedule and forward local requests */
1732 {
1733 .hook = ip_vs_local_request4,
1734 .owner = THIS_MODULE,
1735 .pf = PF_INET,
1736 .hooknum = NF_INET_LOCAL_OUT,
1737 .priority = -98,
1738 },
1739 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1740 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1741 {
1742 .hook = ip_vs_forward_icmp,
1743 .owner = THIS_MODULE,
1744 .pf = PF_INET,
1745 .hooknum = NF_INET_FORWARD,
1746 .priority = 99,
1747 },
1748 /* After packet filtering, change source only for VS/NAT */
1749 {
1750 .hook = ip_vs_reply4,
1751 .owner = THIS_MODULE,
1752 .pf = PF_INET,
1753 .hooknum = NF_INET_FORWARD,
1754 .priority = 100,
1755 },
1756 #ifdef CONFIG_IP_VS_IPV6
1757 /* After packet filtering, change source only for VS/NAT */
1758 {
1759 .hook = ip_vs_reply6,
1760 .owner = THIS_MODULE,
1761 .pf = PF_INET6,
1762 .hooknum = NF_INET_LOCAL_IN,
1763 .priority = 99,
1764 },
1765 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1766 * or VS/NAT(change destination), so that filtering rules can be
1767 * applied to IPVS. */
1768 {
1769 .hook = ip_vs_remote_request6,
1770 .owner = THIS_MODULE,
1771 .pf = PF_INET6,
1772 .hooknum = NF_INET_LOCAL_IN,
1773 .priority = 101,
1774 },
1775 /* Before ip_vs_in, change source only for VS/NAT */
1776 {
1777 .hook = ip_vs_local_reply6,
1778 .owner = THIS_MODULE,
1779 .pf = PF_INET,
1780 .hooknum = NF_INET_LOCAL_OUT,
1781 .priority = -99,
1782 },
1783 /* After mangle, schedule and forward local requests */
1784 {
1785 .hook = ip_vs_local_request6,
1786 .owner = THIS_MODULE,
1787 .pf = PF_INET6,
1788 .hooknum = NF_INET_LOCAL_OUT,
1789 .priority = -98,
1790 },
1791 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1792 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1793 {
1794 .hook = ip_vs_forward_icmp_v6,
1795 .owner = THIS_MODULE,
1796 .pf = PF_INET6,
1797 .hooknum = NF_INET_FORWARD,
1798 .priority = 99,
1799 },
1800 /* After packet filtering, change source only for VS/NAT */
1801 {
1802 .hook = ip_vs_reply6,
1803 .owner = THIS_MODULE,
1804 .pf = PF_INET6,
1805 .hooknum = NF_INET_FORWARD,
1806 .priority = 100,
1807 },
1808 #endif
1809 };
1810
1811
1812 /*
1813 * Initialize IP Virtual Server
1814 */
1815 static int __init ip_vs_init(void)
1816 {
1817 int ret;
1818
1819 ip_vs_estimator_init();
1820
1821 ret = ip_vs_control_init();
1822 if (ret < 0) {
1823 pr_err("can't setup control.\n");
1824 goto cleanup_estimator;
1825 }
1826
1827 ip_vs_protocol_init();
1828
1829 ret = ip_vs_app_init();
1830 if (ret < 0) {
1831 pr_err("can't setup application helper.\n");
1832 goto cleanup_protocol;
1833 }
1834
1835 ret = ip_vs_conn_init();
1836 if (ret < 0) {
1837 pr_err("can't setup connection table.\n");
1838 goto cleanup_app;
1839 }
1840
1841 ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1842 if (ret < 0) {
1843 pr_err("can't register hooks.\n");
1844 goto cleanup_conn;
1845 }
1846
1847 pr_info("ipvs loaded.\n");
1848 return ret;
1849
1850 cleanup_conn:
1851 ip_vs_conn_cleanup();
1852 cleanup_app:
1853 ip_vs_app_cleanup();
1854 cleanup_protocol:
1855 ip_vs_protocol_cleanup();
1856 ip_vs_control_cleanup();
1857 cleanup_estimator:
1858 ip_vs_estimator_cleanup();
1859 return ret;
1860 }
1861
1862 static void __exit ip_vs_cleanup(void)
1863 {
1864 nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1865 ip_vs_conn_cleanup();
1866 ip_vs_app_cleanup();
1867 ip_vs_protocol_cleanup();
1868 ip_vs_control_cleanup();
1869 ip_vs_estimator_cleanup();
1870 pr_info("ipvs unloaded.\n");
1871 }
1872
1873 module_init(ip_vs_init);
1874 module_exit(ip_vs_cleanup);
1875 MODULE_LICENSE("GPL");
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree