~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/netfilter/ipvs/ip_vs_core.c

Version: ~ [ linux-5.19-rc3 ] ~ [ linux-5.18.5 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.48 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.123 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.199 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.248 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.284 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.319 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 #include <net/netns/generic.h>          /* net_generic() */
 45 
 46 #include <linux/netfilter.h>
 47 #include <linux/netfilter_ipv4.h>
 48 
 49 #ifdef CONFIG_IP_VS_IPV6
 50 #include <net/ipv6.h>
 51 #include <linux/netfilter_ipv6.h>
 52 #include <net/ip6_route.h>
 53 #endif
 54 
 55 #include <net/ip_vs.h>
 56 
 57 
 58 EXPORT_SYMBOL(register_ip_vs_scheduler);
 59 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
 60 EXPORT_SYMBOL(ip_vs_proto_name);
 61 EXPORT_SYMBOL(ip_vs_conn_new);
 62 EXPORT_SYMBOL(ip_vs_conn_in_get);
 63 EXPORT_SYMBOL(ip_vs_conn_out_get);
 64 #ifdef CONFIG_IP_VS_PROTO_TCP
 65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
 66 #endif
 67 EXPORT_SYMBOL(ip_vs_conn_put);
 68 #ifdef CONFIG_IP_VS_DEBUG
 69 EXPORT_SYMBOL(ip_vs_get_debug_level);
 70 #endif
 71 EXPORT_SYMBOL(ip_vs_new_conn_out);
 72 
 73 static unsigned int ip_vs_net_id __read_mostly;
 74 /* netns cnt used for uniqueness */
 75 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
 76 
 77 /* ID used in ICMP lookups */
 78 #define icmp_id(icmph)          (((icmph)->un).echo.id)
 79 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
 80 
 81 const char *ip_vs_proto_name(unsigned int proto)
 82 {
 83         static char buf[20];
 84 
 85         switch (proto) {
 86         case IPPROTO_IP:
 87                 return "IP";
 88         case IPPROTO_UDP:
 89                 return "UDP";
 90         case IPPROTO_TCP:
 91                 return "TCP";
 92         case IPPROTO_SCTP:
 93                 return "SCTP";
 94         case IPPROTO_ICMP:
 95                 return "ICMP";
 96 #ifdef CONFIG_IP_VS_IPV6
 97         case IPPROTO_ICMPV6:
 98                 return "ICMPv6";
 99 #endif
100         default:
101                 sprintf(buf, "IP_%u", proto);
102                 return buf;
103         }
104 }
105 
106 void ip_vs_init_hash_table(struct list_head *table, int rows)
107 {
108         while (--rows >= 0)
109                 INIT_LIST_HEAD(&table[rows]);
110 }
111 
112 static inline void
113 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
114 {
115         struct ip_vs_dest *dest = cp->dest;
116         struct netns_ipvs *ipvs = cp->ipvs;
117 
118         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
119                 struct ip_vs_cpu_stats *s;
120                 struct ip_vs_service *svc;
121 
122                 s = this_cpu_ptr(dest->stats.cpustats);
123                 u64_stats_update_begin(&s->syncp);
124                 s->cnt.inpkts++;
125                 s->cnt.inbytes += skb->len;
126                 u64_stats_update_end(&s->syncp);
127 
128                 rcu_read_lock();
129                 svc = rcu_dereference(dest->svc);
130                 s = this_cpu_ptr(svc->stats.cpustats);
131                 u64_stats_update_begin(&s->syncp);
132                 s->cnt.inpkts++;
133                 s->cnt.inbytes += skb->len;
134                 u64_stats_update_end(&s->syncp);
135                 rcu_read_unlock();
136 
137                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
138                 u64_stats_update_begin(&s->syncp);
139                 s->cnt.inpkts++;
140                 s->cnt.inbytes += skb->len;
141                 u64_stats_update_end(&s->syncp);
142         }
143 }
144 
145 
146 static inline void
147 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
148 {
149         struct ip_vs_dest *dest = cp->dest;
150         struct netns_ipvs *ipvs = cp->ipvs;
151 
152         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
153                 struct ip_vs_cpu_stats *s;
154                 struct ip_vs_service *svc;
155 
156                 s = this_cpu_ptr(dest->stats.cpustats);
157                 u64_stats_update_begin(&s->syncp);
158                 s->cnt.outpkts++;
159                 s->cnt.outbytes += skb->len;
160                 u64_stats_update_end(&s->syncp);
161 
162                 rcu_read_lock();
163                 svc = rcu_dereference(dest->svc);
164                 s = this_cpu_ptr(svc->stats.cpustats);
165                 u64_stats_update_begin(&s->syncp);
166                 s->cnt.outpkts++;
167                 s->cnt.outbytes += skb->len;
168                 u64_stats_update_end(&s->syncp);
169                 rcu_read_unlock();
170 
171                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
172                 u64_stats_update_begin(&s->syncp);
173                 s->cnt.outpkts++;
174                 s->cnt.outbytes += skb->len;
175                 u64_stats_update_end(&s->syncp);
176         }
177 }
178 
179 
180 static inline void
181 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
182 {
183         struct netns_ipvs *ipvs = svc->ipvs;
184         struct ip_vs_cpu_stats *s;
185 
186         s = this_cpu_ptr(cp->dest->stats.cpustats);
187         u64_stats_update_begin(&s->syncp);
188         s->cnt.conns++;
189         u64_stats_update_end(&s->syncp);
190 
191         s = this_cpu_ptr(svc->stats.cpustats);
192         u64_stats_update_begin(&s->syncp);
193         s->cnt.conns++;
194         u64_stats_update_end(&s->syncp);
195 
196         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
197         u64_stats_update_begin(&s->syncp);
198         s->cnt.conns++;
199         u64_stats_update_end(&s->syncp);
200 }
201 
202 
203 static inline void
204 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
205                 const struct sk_buff *skb,
206                 struct ip_vs_proto_data *pd)
207 {
208         if (likely(pd->pp->state_transition))
209                 pd->pp->state_transition(cp, direction, skb, pd);
210 }
211 
212 static inline int
213 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
214                               struct sk_buff *skb, int protocol,
215                               const union nf_inet_addr *caddr, __be16 cport,
216                               const union nf_inet_addr *vaddr, __be16 vport,
217                               struct ip_vs_conn_param *p)
218 {
219         ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
220                               vport, p);
221         p->pe = rcu_dereference(svc->pe);
222         if (p->pe && p->pe->fill_param)
223                 return p->pe->fill_param(p, skb);
224 
225         return 0;
226 }
227 
228 /*
229  *  IPVS persistent scheduling function
230  *  It creates a connection entry according to its template if exists,
231  *  or selects a server and creates a connection entry plus a template.
232  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
233  *  Protocols supported: TCP, UDP
234  */
235 static struct ip_vs_conn *
236 ip_vs_sched_persist(struct ip_vs_service *svc,
237                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
238                     int *ignored, struct ip_vs_iphdr *iph)
239 {
240         struct ip_vs_conn *cp = NULL;
241         struct ip_vs_dest *dest;
242         struct ip_vs_conn *ct;
243         __be16 dport = 0;               /* destination port to forward */
244         unsigned int flags;
245         struct ip_vs_conn_param param;
246         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
247         union nf_inet_addr snet;        /* source network of the client,
248                                            after masking */
249         const union nf_inet_addr *src_addr, *dst_addr;
250 
251         if (likely(!ip_vs_iph_inverse(iph))) {
252                 src_addr = &iph->saddr;
253                 dst_addr = &iph->daddr;
254         } else {
255                 src_addr = &iph->daddr;
256                 dst_addr = &iph->saddr;
257         }
258 
259 
260         /* Mask saddr with the netmask to adjust template granularity */
261 #ifdef CONFIG_IP_VS_IPV6
262         if (svc->af == AF_INET6)
263                 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
264                                  (__force __u32) svc->netmask);
265         else
266 #endif
267                 snet.ip = src_addr->ip & svc->netmask;
268 
269         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
270                       "mnet %s\n",
271                       IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
272                       IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
273                       IP_VS_DBG_ADDR(svc->af, &snet));
274 
275         /*
276          * As far as we know, FTP is a very complicated network protocol, and
277          * it uses control connection and data connections. For active FTP,
278          * FTP server initialize data connection to the client, its source port
279          * is often 20. For passive FTP, FTP server tells the clients the port
280          * that it passively listens to,  and the client issues the data
281          * connection. In the tunneling or direct routing mode, the load
282          * balancer is on the client-to-server half of connection, the port
283          * number is unknown to the load balancer. So, a conn template like
284          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
285          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
286          * is created for other persistent services.
287          */
288         {
289                 int protocol = iph->protocol;
290                 const union nf_inet_addr *vaddr = dst_addr;
291                 __be16 vport = 0;
292 
293                 if (dst_port == svc->port) {
294                         /* non-FTP template:
295                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
296                          * FTP template:
297                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
298                          */
299                         if (svc->port != FTPPORT)
300                                 vport = dst_port;
301                 } else {
302                         /* Note: persistent fwmark-based services and
303                          * persistent port zero service are handled here.
304                          * fwmark template:
305                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
306                          * port zero template:
307                          * <protocol,caddr,0,vaddr,0,daddr,0>
308                          */
309                         if (svc->fwmark) {
310                                 protocol = IPPROTO_IP;
311                                 vaddr = &fwmark;
312                         }
313                 }
314                 /* return *ignored = -1 so NF_DROP can be used */
315                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
316                                                   vaddr, vport, &param) < 0) {
317                         *ignored = -1;
318                         return NULL;
319                 }
320         }
321 
322         /* Check if a template already exists */
323         ct = ip_vs_ct_in_get(&param);
324         if (!ct || !ip_vs_check_template(ct, NULL)) {
325                 struct ip_vs_scheduler *sched;
326 
327                 /*
328                  * No template found or the dest of the connection
329                  * template is not available.
330                  * return *ignored=0 i.e. ICMP and NF_DROP
331                  */
332                 sched = rcu_dereference(svc->scheduler);
333                 if (sched) {
334                         /* read svc->sched_data after svc->scheduler */
335                         smp_rmb();
336                         dest = sched->schedule(svc, skb, iph);
337                 } else {
338                         dest = NULL;
339                 }
340                 if (!dest) {
341                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
342                         kfree(param.pe_data);
343                         *ignored = 0;
344                         return NULL;
345                 }
346 
347                 if (dst_port == svc->port && svc->port != FTPPORT)
348                         dport = dest->port;
349 
350                 /* Create a template
351                  * This adds param.pe_data to the template,
352                  * and thus param.pe_data will be destroyed
353                  * when the template expires */
354                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
355                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
356                 if (ct == NULL) {
357                         kfree(param.pe_data);
358                         *ignored = -1;
359                         return NULL;
360                 }
361 
362                 ct->timeout = svc->timeout;
363         } else {
364                 /* set destination with the found template */
365                 dest = ct->dest;
366                 kfree(param.pe_data);
367         }
368 
369         dport = dst_port;
370         if (dport == svc->port && dest->port)
371                 dport = dest->port;
372 
373         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
374                  && iph->protocol == IPPROTO_UDP) ?
375                 IP_VS_CONN_F_ONE_PACKET : 0;
376 
377         /*
378          *    Create a new connection according to the template
379          */
380         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
381                               src_port, dst_addr, dst_port, &param);
382 
383         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
384                             skb->mark);
385         if (cp == NULL) {
386                 ip_vs_conn_put(ct);
387                 *ignored = -1;
388                 return NULL;
389         }
390 
391         /*
392          *    Add its control
393          */
394         ip_vs_control_add(cp, ct);
395         ip_vs_conn_put(ct);
396 
397         ip_vs_conn_stats(cp, svc);
398         return cp;
399 }
400 
401 
402 /*
403  *  IPVS main scheduling function
404  *  It selects a server according to the virtual service, and
405  *  creates a connection entry.
406  *  Protocols supported: TCP, UDP
407  *
408  *  Usage of *ignored
409  *
410  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
411  *       svc/scheduler decides that this packet should be accepted with
412  *       NF_ACCEPT because it must not be scheduled.
413  *
414  * 0 :   scheduler can not find destination, so try bypass or
415  *       return ICMP and then NF_DROP (ip_vs_leave).
416  *
417  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
418  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
419  *       failure such as missing Call-ID, ENOMEM on skb_linearize
420  *       or pe_data. In this case we should return NF_DROP without
421  *       any attempts to send ICMP with ip_vs_leave.
422  */
423 struct ip_vs_conn *
424 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
425                struct ip_vs_proto_data *pd, int *ignored,
426                struct ip_vs_iphdr *iph)
427 {
428         struct ip_vs_protocol *pp = pd->pp;
429         struct ip_vs_conn *cp = NULL;
430         struct ip_vs_scheduler *sched;
431         struct ip_vs_dest *dest;
432         __be16 _ports[2], *pptr, cport, vport;
433         const void *caddr, *vaddr;
434         unsigned int flags;
435 
436         *ignored = 1;
437         /*
438          * IPv6 frags, only the first hit here.
439          */
440         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
441         if (pptr == NULL)
442                 return NULL;
443 
444         if (likely(!ip_vs_iph_inverse(iph))) {
445                 cport = pptr[0];
446                 caddr = &iph->saddr;
447                 vport = pptr[1];
448                 vaddr = &iph->daddr;
449         } else {
450                 cport = pptr[1];
451                 caddr = &iph->daddr;
452                 vport = pptr[0];
453                 vaddr = &iph->saddr;
454         }
455 
456         /*
457          * FTPDATA needs this check when using local real server.
458          * Never schedule Active FTPDATA connections from real server.
459          * For LVS-NAT they must be already created. For other methods
460          * with persistence the connection is created on SYN+ACK.
461          */
462         if (cport == FTPDATA) {
463                 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
464                               "Not scheduling FTPDATA");
465                 return NULL;
466         }
467 
468         /*
469          *    Do not schedule replies from local real server.
470          */
471         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
472                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
473                 cp = pp->conn_in_get(svc->ipvs, svc->af, skb, iph);
474                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
475 
476                 if (cp) {
477                         IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
478                                       "Not scheduling reply for existing"
479                                       " connection");
480                         __ip_vs_conn_put(cp);
481                         return NULL;
482                 }
483         }
484 
485         /*
486          *    Persistent service
487          */
488         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
489                 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
490                                            iph);
491 
492         *ignored = 0;
493 
494         /*
495          *    Non-persistent service
496          */
497         if (!svc->fwmark && vport != svc->port) {
498                 if (!svc->port)
499                         pr_err("Schedule: port zero only supported "
500                                "in persistent services, "
501                                "check your ipvs configuration\n");
502                 return NULL;
503         }
504 
505         sched = rcu_dereference(svc->scheduler);
506         if (sched) {
507                 /* read svc->sched_data after svc->scheduler */
508                 smp_rmb();
509                 dest = sched->schedule(svc, skb, iph);
510         } else {
511                 dest = NULL;
512         }
513         if (dest == NULL) {
514                 IP_VS_DBG(1, "Schedule: no dest found.\n");
515                 return NULL;
516         }
517 
518         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
519                  && iph->protocol == IPPROTO_UDP) ?
520                 IP_VS_CONN_F_ONE_PACKET : 0;
521 
522         /*
523          *    Create a connection entry.
524          */
525         {
526                 struct ip_vs_conn_param p;
527 
528                 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
529                                       caddr, cport, vaddr, vport, &p);
530                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
531                                     dest->port ? dest->port : vport,
532                                     flags, dest, skb->mark);
533                 if (!cp) {
534                         *ignored = -1;
535                         return NULL;
536                 }
537         }
538 
539         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
540                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
541                       ip_vs_fwd_tag(cp),
542                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
543                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
544                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
545                       cp->flags, refcount_read(&cp->refcnt));
546 
547         ip_vs_conn_stats(cp, svc);
548         return cp;
549 }
550 
551 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
552                                         union nf_inet_addr *addr)
553 {
554 #ifdef CONFIG_IP_VS_IPV6
555         if (af == AF_INET6)
556                 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
557 #endif
558         return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
559 }
560 
561 /*
562  *  Pass or drop the packet.
563  *  Called by ip_vs_in, when the virtual service is available but
564  *  no destination is available for a new connection.
565  */
566 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
567                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
568 {
569         __be16 _ports[2], *pptr, dport;
570         struct netns_ipvs *ipvs = svc->ipvs;
571         struct net *net = ipvs->net;
572 
573         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
574         if (!pptr)
575                 return NF_DROP;
576         dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
577 
578         /* if it is fwmark-based service, the cache_bypass sysctl is up
579            and the destination is a non-local unicast, then create
580            a cache_bypass connection entry */
581         if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
582             !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
583             ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
584                 int ret;
585                 struct ip_vs_conn *cp;
586                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
587                                       iph->protocol == IPPROTO_UDP) ?
588                                       IP_VS_CONN_F_ONE_PACKET : 0;
589                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
590 
591                 /* create a new connection entry */
592                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
593                 {
594                         struct ip_vs_conn_param p;
595                         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
596                                               &iph->saddr, pptr[0],
597                                               &iph->daddr, pptr[1], &p);
598                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
599                                             IP_VS_CONN_F_BYPASS | flags,
600                                             NULL, skb->mark);
601                         if (!cp)
602                                 return NF_DROP;
603                 }
604 
605                 /* statistics */
606                 ip_vs_in_stats(cp, skb);
607 
608                 /* set state */
609                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
610 
611                 /* transmit the first SYN packet */
612                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
613                 /* do not touch skb anymore */
614 
615                 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
616                         atomic_inc(&cp->control->in_pkts);
617                 else
618                         atomic_inc(&cp->in_pkts);
619                 ip_vs_conn_put(cp);
620                 return ret;
621         }
622 
623         /*
624          * When the virtual ftp service is presented, packets destined
625          * for other services on the VIP may get here (except services
626          * listed in the ipvs table), pass the packets, because it is
627          * not ipvs job to decide to drop the packets.
628          */
629         if (svc->port == FTPPORT && dport != FTPPORT)
630                 return NF_ACCEPT;
631 
632         if (unlikely(ip_vs_iph_icmp(iph)))
633                 return NF_DROP;
634 
635         /*
636          * Notify the client that the destination is unreachable, and
637          * release the socket buffer.
638          * Since it is in IP layer, the TCP socket is not actually
639          * created, the TCP RST packet cannot be sent, instead that
640          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
641          */
642 #ifdef CONFIG_IP_VS_IPV6
643         if (svc->af == AF_INET6) {
644                 if (!skb->dev)
645                         skb->dev = net->loopback_dev;
646                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
647         } else
648 #endif
649                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
650 
651         return NF_DROP;
652 }
653 
654 #ifdef CONFIG_SYSCTL
655 
656 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
657 {
658         return ipvs->sysctl_snat_reroute;
659 }
660 
661 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
662 {
663         return ipvs->sysctl_nat_icmp_send;
664 }
665 
666 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
667 {
668         return ipvs->sysctl_expire_nodest_conn;
669 }
670 
671 #else
672 
673 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
674 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
675 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
676 
677 #endif
678 
679 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
680 {
681         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
682 }
683 
684 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
685 {
686         if (NF_INET_LOCAL_IN == hooknum)
687                 return IP_DEFRAG_VS_IN;
688         if (NF_INET_FORWARD == hooknum)
689                 return IP_DEFRAG_VS_FWD;
690         return IP_DEFRAG_VS_OUT;
691 }
692 
693 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
694                                      struct sk_buff *skb, u_int32_t user)
695 {
696         int err;
697 
698         local_bh_disable();
699         err = ip_defrag(ipvs->net, skb, user);
700         local_bh_enable();
701         if (!err)
702                 ip_send_check(ip_hdr(skb));
703 
704         return err;
705 }
706 
707 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
708                                  struct sk_buff *skb, unsigned int hooknum)
709 {
710         if (!sysctl_snat_reroute(ipvs))
711                 return 0;
712         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
713         if (NF_INET_LOCAL_IN == hooknum)
714                 return 0;
715 #ifdef CONFIG_IP_VS_IPV6
716         if (af == AF_INET6) {
717                 struct dst_entry *dst = skb_dst(skb);
718 
719                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
720                     ip6_route_me_harder(ipvs->net, skb) != 0)
721                         return 1;
722         } else
723 #endif
724                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
725                     ip_route_me_harder(ipvs->net, skb, RTN_LOCAL) != 0)
726                         return 1;
727 
728         return 0;
729 }
730 
731 /*
732  * Packet has been made sufficiently writable in caller
733  * - inout: 1=in->out, 0=out->in
734  */
735 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
736                     struct ip_vs_conn *cp, int inout)
737 {
738         struct iphdr *iph        = ip_hdr(skb);
739         unsigned int icmp_offset = iph->ihl*4;
740         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
741                                                       icmp_offset);
742         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
743 
744         if (inout) {
745                 iph->saddr = cp->vaddr.ip;
746                 ip_send_check(iph);
747                 ciph->daddr = cp->vaddr.ip;
748                 ip_send_check(ciph);
749         } else {
750                 iph->daddr = cp->daddr.ip;
751                 ip_send_check(iph);
752                 ciph->saddr = cp->daddr.ip;
753                 ip_send_check(ciph);
754         }
755 
756         /* the TCP/UDP/SCTP port */
757         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
758             IPPROTO_SCTP == ciph->protocol) {
759                 __be16 *ports = (void *)ciph + ciph->ihl*4;
760 
761                 if (inout)
762                         ports[1] = cp->vport;
763                 else
764                         ports[0] = cp->dport;
765         }
766 
767         /* And finally the ICMP checksum */
768         icmph->checksum = 0;
769         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
770         skb->ip_summed = CHECKSUM_UNNECESSARY;
771 
772         if (inout)
773                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
774                         "Forwarding altered outgoing ICMP");
775         else
776                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
777                         "Forwarding altered incoming ICMP");
778 }
779 
780 #ifdef CONFIG_IP_VS_IPV6
781 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
782                     struct ip_vs_conn *cp, int inout)
783 {
784         struct ipv6hdr *iph      = ipv6_hdr(skb);
785         unsigned int icmp_offset = 0;
786         unsigned int offs        = 0; /* header offset*/
787         int protocol;
788         struct icmp6hdr *icmph;
789         struct ipv6hdr *ciph;
790         unsigned short fragoffs;
791 
792         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
793         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
794         offs = icmp_offset + sizeof(struct icmp6hdr);
795         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
796 
797         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
798 
799         if (inout) {
800                 iph->saddr = cp->vaddr.in6;
801                 ciph->daddr = cp->vaddr.in6;
802         } else {
803                 iph->daddr = cp->daddr.in6;
804                 ciph->saddr = cp->daddr.in6;
805         }
806 
807         /* the TCP/UDP/SCTP port */
808         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
809                           IPPROTO_SCTP == protocol)) {
810                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
811 
812                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
813                               ntohs(inout ? ports[1] : ports[0]),
814                               ntohs(inout ? cp->vport : cp->dport));
815                 if (inout)
816                         ports[1] = cp->vport;
817                 else
818                         ports[0] = cp->dport;
819         }
820 
821         /* And finally the ICMP checksum */
822         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
823                                               skb->len - icmp_offset,
824                                               IPPROTO_ICMPV6, 0);
825         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
826         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
827         skb->ip_summed = CHECKSUM_PARTIAL;
828 
829         if (inout)
830                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
831                               (void *)ciph - (void *)iph,
832                               "Forwarding altered outgoing ICMPv6");
833         else
834                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
835                               (void *)ciph - (void *)iph,
836                               "Forwarding altered incoming ICMPv6");
837 }
838 #endif
839 
840 /* Handle relevant response ICMP messages - forward to the right
841  * destination host.
842  */
843 static int handle_response_icmp(int af, struct sk_buff *skb,
844                                 union nf_inet_addr *snet,
845                                 __u8 protocol, struct ip_vs_conn *cp,
846                                 struct ip_vs_protocol *pp,
847                                 unsigned int offset, unsigned int ihl,
848                                 unsigned int hooknum)
849 {
850         unsigned int verdict = NF_DROP;
851 
852         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
853                 goto ignore_cp;
854 
855         /* Ensure the checksum is correct */
856         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
857                 /* Failed checksum! */
858                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
859                               IP_VS_DBG_ADDR(af, snet));
860                 goto out;
861         }
862 
863         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
864             IPPROTO_SCTP == protocol)
865                 offset += 2 * sizeof(__u16);
866         if (!skb_make_writable(skb, offset))
867                 goto out;
868 
869 #ifdef CONFIG_IP_VS_IPV6
870         if (af == AF_INET6)
871                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
872         else
873 #endif
874                 ip_vs_nat_icmp(skb, pp, cp, 1);
875 
876         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
877                 goto out;
878 
879         /* do the statistics and put it back */
880         ip_vs_out_stats(cp, skb);
881 
882         skb->ipvs_property = 1;
883         if (!(cp->flags & IP_VS_CONN_F_NFCT))
884                 ip_vs_notrack(skb);
885         else
886                 ip_vs_update_conntrack(skb, cp, 0);
887 
888 ignore_cp:
889         verdict = NF_ACCEPT;
890 
891 out:
892         __ip_vs_conn_put(cp);
893 
894         return verdict;
895 }
896 
897 /*
898  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
899  *      Find any that might be relevant, check against existing connections.
900  *      Currently handles error types - unreachable, quench, ttl exceeded.
901  */
902 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
903                           int *related, unsigned int hooknum)
904 {
905         struct iphdr *iph;
906         struct icmphdr  _icmph, *ic;
907         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
908         struct ip_vs_iphdr ciph;
909         struct ip_vs_conn *cp;
910         struct ip_vs_protocol *pp;
911         unsigned int offset, ihl;
912         union nf_inet_addr snet;
913 
914         *related = 1;
915 
916         /* reassemble IP fragments */
917         if (ip_is_fragment(ip_hdr(skb))) {
918                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
919                         return NF_STOLEN;
920         }
921 
922         iph = ip_hdr(skb);
923         offset = ihl = iph->ihl * 4;
924         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
925         if (ic == NULL)
926                 return NF_DROP;
927 
928         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
929                   ic->type, ntohs(icmp_id(ic)),
930                   &iph->saddr, &iph->daddr);
931 
932         /*
933          * Work through seeing if this is for us.
934          * These checks are supposed to be in an order that means easy
935          * things are checked first to speed up processing.... however
936          * this means that some packets will manage to get a long way
937          * down this stack and then be rejected, but that's life.
938          */
939         if ((ic->type != ICMP_DEST_UNREACH) &&
940             (ic->type != ICMP_SOURCE_QUENCH) &&
941             (ic->type != ICMP_TIME_EXCEEDED)) {
942                 *related = 0;
943                 return NF_ACCEPT;
944         }
945 
946         /* Now find the contained IP header */
947         offset += sizeof(_icmph);
948         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
949         if (cih == NULL)
950                 return NF_ACCEPT; /* The packet looks wrong, ignore */
951 
952         pp = ip_vs_proto_get(cih->protocol);
953         if (!pp)
954                 return NF_ACCEPT;
955 
956         /* Is the embedded protocol header present? */
957         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
958                      pp->dont_defrag))
959                 return NF_ACCEPT;
960 
961         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
962                       "Checking outgoing ICMP for");
963 
964         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
965 
966         /* The embedded headers contain source and dest in reverse order */
967         cp = pp->conn_out_get(ipvs, AF_INET, skb, &ciph);
968         if (!cp)
969                 return NF_ACCEPT;
970 
971         snet.ip = iph->saddr;
972         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
973                                     pp, ciph.len, ihl, hooknum);
974 }
975 
976 #ifdef CONFIG_IP_VS_IPV6
977 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
978                              int *related,  unsigned int hooknum,
979                              struct ip_vs_iphdr *ipvsh)
980 {
981         struct icmp6hdr _icmph, *ic;
982         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
983         struct ip_vs_conn *cp;
984         struct ip_vs_protocol *pp;
985         union nf_inet_addr snet;
986         unsigned int offset;
987 
988         *related = 1;
989         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
990         if (ic == NULL)
991                 return NF_DROP;
992 
993         /*
994          * Work through seeing if this is for us.
995          * These checks are supposed to be in an order that means easy
996          * things are checked first to speed up processing.... however
997          * this means that some packets will manage to get a long way
998          * down this stack and then be rejected, but that's life.
999          */
1000         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1001                 *related = 0;
1002                 return NF_ACCEPT;
1003         }
1004         /* Fragment header that is before ICMP header tells us that:
1005          * it's not an error message since they can't be fragmented.
1006          */
1007         if (ipvsh->flags & IP6_FH_F_FRAG)
1008                 return NF_DROP;
1009 
1010         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1011                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1012                   &ipvsh->saddr, &ipvsh->daddr);
1013 
1014         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1015                                      true, &ciph))
1016                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1017 
1018         pp = ip_vs_proto_get(ciph.protocol);
1019         if (!pp)
1020                 return NF_ACCEPT;
1021 
1022         /* The embedded headers contain source and dest in reverse order */
1023         cp = pp->conn_out_get(ipvs, AF_INET6, skb, &ciph);
1024         if (!cp)
1025                 return NF_ACCEPT;
1026 
1027         snet.in6 = ciph.saddr.in6;
1028         offset = ciph.len;
1029         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1030                                     pp, offset, sizeof(struct ipv6hdr),
1031                                     hooknum);
1032 }
1033 #endif
1034 
1035 /*
1036  * Check if sctp chunc is ABORT chunk
1037  */
1038 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1039 {
1040         sctp_chunkhdr_t *sch, schunk;
1041         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1042                         sizeof(schunk), &schunk);
1043         if (sch == NULL)
1044                 return 0;
1045         if (sch->type == SCTP_CID_ABORT)
1046                 return 1;
1047         return 0;
1048 }
1049 
1050 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1051 {
1052         struct tcphdr _tcph, *th;
1053 
1054         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1055         if (th == NULL)
1056                 return 0;
1057         return th->rst;
1058 }
1059 
1060 static inline bool is_new_conn(const struct sk_buff *skb,
1061                                struct ip_vs_iphdr *iph)
1062 {
1063         switch (iph->protocol) {
1064         case IPPROTO_TCP: {
1065                 struct tcphdr _tcph, *th;
1066 
1067                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1068                 if (th == NULL)
1069                         return false;
1070                 return th->syn;
1071         }
1072         case IPPROTO_SCTP: {
1073                 sctp_chunkhdr_t *sch, schunk;
1074 
1075                 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1076                                          sizeof(schunk), &schunk);
1077                 if (sch == NULL)
1078                         return false;
1079                 return sch->type == SCTP_CID_INIT;
1080         }
1081         default:
1082                 return false;
1083         }
1084 }
1085 
1086 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1087                                         int conn_reuse_mode)
1088 {
1089         /* Controlled (FTP DATA or persistence)? */
1090         if (cp->control)
1091                 return false;
1092 
1093         switch (cp->protocol) {
1094         case IPPROTO_TCP:
1095                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1096                        (cp->state == IP_VS_TCP_S_CLOSE) ||
1097                         ((conn_reuse_mode & 2) &&
1098                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1099                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1100         case IPPROTO_SCTP:
1101                 return cp->state == IP_VS_SCTP_S_CLOSED;
1102         default:
1103                 return false;
1104         }
1105 }
1106 
1107 /* Generic function to create new connections for outgoing RS packets
1108  *
1109  * Pre-requisites for successful connection creation:
1110  * 1) Virtual Service is NOT fwmark based:
1111  *    In fwmark-VS actual vaddr and vport are unknown to IPVS
1112  * 2) Real Server and Virtual Service were NOT configured without port:
1113  *    This is to allow match of different VS to the same RS ip-addr
1114  */
1115 struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc,
1116                                       struct ip_vs_dest *dest,
1117                                       struct sk_buff *skb,
1118                                       const struct ip_vs_iphdr *iph,
1119                                       __be16 dport,
1120                                       __be16 cport)
1121 {
1122         struct ip_vs_conn_param param;
1123         struct ip_vs_conn *ct = NULL, *cp = NULL;
1124         const union nf_inet_addr *vaddr, *daddr, *caddr;
1125         union nf_inet_addr snet;
1126         __be16 vport;
1127         unsigned int flags;
1128 
1129         EnterFunction(12);
1130         vaddr = &svc->addr;
1131         vport = svc->port;
1132         daddr = &iph->saddr;
1133         caddr = &iph->daddr;
1134 
1135         /* check pre-requisites are satisfied */
1136         if (svc->fwmark)
1137                 return NULL;
1138         if (!vport || !dport)
1139                 return NULL;
1140 
1141         /* for persistent service first create connection template */
1142         if (svc->flags & IP_VS_SVC_F_PERSISTENT) {
1143                 /* apply netmask the same way ingress-side does */
1144 #ifdef CONFIG_IP_VS_IPV6
1145                 if (svc->af == AF_INET6)
1146                         ipv6_addr_prefix(&snet.in6, &caddr->in6,
1147                                          (__force __u32)svc->netmask);
1148                 else
1149 #endif
1150                         snet.ip = caddr->ip & svc->netmask;
1151                 /* fill params and create template if not existent */
1152                 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol,
1153                                                   &snet, 0, vaddr,
1154                                                   vport, &param) < 0)
1155                         return NULL;
1156                 ct = ip_vs_ct_in_get(&param);
1157                 /* check if template exists and points to the same dest */
1158                 if (!ct || !ip_vs_check_template(ct, dest)) {
1159                         ct = ip_vs_conn_new(&param, dest->af, daddr, dport,
1160                                             IP_VS_CONN_F_TEMPLATE, dest, 0);
1161                         if (!ct) {
1162                                 kfree(param.pe_data);
1163                                 return NULL;
1164                         }
1165                         ct->timeout = svc->timeout;
1166                 } else {
1167                         kfree(param.pe_data);
1168                 }
1169         }
1170 
1171         /* connection flags */
1172         flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) &&
1173                  iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0;
1174         /* create connection */
1175         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
1176                               caddr, cport, vaddr, vport, &param);
1177         cp = ip_vs_conn_new(&param, dest->af, daddr, dport, flags, dest, 0);
1178         if (!cp) {
1179                 if (ct)
1180                         ip_vs_conn_put(ct);
1181                 return NULL;
1182         }
1183         if (ct) {
1184                 ip_vs_control_add(cp, ct);
1185                 ip_vs_conn_put(ct);
1186         }
1187         ip_vs_conn_stats(cp, svc);
1188 
1189         /* return connection (will be used to handle outgoing packet) */
1190         IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u "
1191                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
1192                       ip_vs_fwd_tag(cp),
1193                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
1194                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
1195                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
1196                       cp->flags, refcount_read(&cp->refcnt));
1197         LeaveFunction(12);
1198         return cp;
1199 }
1200 
1201 /* Handle outgoing packets which are considered requests initiated by
1202  * real servers, so that subsequent responses from external client can be
1203  * routed to the right real server.
1204  * Used also for outgoing responses in OPS mode.
1205  *
1206  * Connection management is handled by persistent-engine specific callback.
1207  */
1208 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum,
1209                                               struct netns_ipvs *ipvs,
1210                                               int af, struct sk_buff *skb,
1211                                               const struct ip_vs_iphdr *iph)
1212 {
1213         struct ip_vs_dest *dest;
1214         struct ip_vs_conn *cp = NULL;
1215         __be16 _ports[2], *pptr;
1216 
1217         if (hooknum == NF_INET_LOCAL_IN)
1218                 return NULL;
1219 
1220         pptr = frag_safe_skb_hp(skb, iph->len,
1221                                 sizeof(_ports), _ports, iph);
1222         if (!pptr)
1223                 return NULL;
1224 
1225         rcu_read_lock();
1226         dest = ip_vs_find_real_service(ipvs, af, iph->protocol,
1227                                        &iph->saddr, pptr[0]);
1228         if (dest) {
1229                 struct ip_vs_service *svc;
1230                 struct ip_vs_pe *pe;
1231 
1232                 svc = rcu_dereference(dest->svc);
1233                 if (svc) {
1234                         pe = rcu_dereference(svc->pe);
1235                         if (pe && pe->conn_out)
1236                                 cp = pe->conn_out(svc, dest, skb, iph,
1237                                                   pptr[0], pptr[1]);
1238                 }
1239         }
1240         rcu_read_unlock();
1241 
1242         return cp;
1243 }
1244 
1245 /* Handle response packets: rewrite addresses and send away...
1246  */
1247 static unsigned int
1248 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1249                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1250                 unsigned int hooknum)
1251 {
1252         struct ip_vs_protocol *pp = pd->pp;
1253 
1254         IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1255 
1256         if (!skb_make_writable(skb, iph->len))
1257                 goto drop;
1258 
1259         /* mangle the packet */
1260         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1261                 goto drop;
1262 
1263 #ifdef CONFIG_IP_VS_IPV6
1264         if (af == AF_INET6)
1265                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1266         else
1267 #endif
1268         {
1269                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1270                 ip_send_check(ip_hdr(skb));
1271         }
1272 
1273         /*
1274          * nf_iterate does not expect change in the skb->dst->dev.
1275          * It looks like it is not fatal to enable this code for hooks
1276          * where our handlers are at the end of the chain list and
1277          * when all next handlers use skb->dst->dev and not outdev.
1278          * It will definitely route properly the inout NAT traffic
1279          * when multiple paths are used.
1280          */
1281 
1282         /* For policy routing, packets originating from this
1283          * machine itself may be routed differently to packets
1284          * passing through.  We want this packet to be routed as
1285          * if it came from this machine itself.  So re-compute
1286          * the routing information.
1287          */
1288         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1289                 goto drop;
1290 
1291         IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1292 
1293         ip_vs_out_stats(cp, skb);
1294         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1295         skb->ipvs_property = 1;
1296         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1297                 ip_vs_notrack(skb);
1298         else
1299                 ip_vs_update_conntrack(skb, cp, 0);
1300         ip_vs_conn_put(cp);
1301 
1302         LeaveFunction(11);
1303         return NF_ACCEPT;
1304 
1305 drop:
1306         ip_vs_conn_put(cp);
1307         kfree_skb(skb);
1308         LeaveFunction(11);
1309         return NF_STOLEN;
1310 }
1311 
1312 /*
1313  *      Check if outgoing packet belongs to the established ip_vs_conn.
1314  */
1315 static unsigned int
1316 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1317 {
1318         struct ip_vs_iphdr iph;
1319         struct ip_vs_protocol *pp;
1320         struct ip_vs_proto_data *pd;
1321         struct ip_vs_conn *cp;
1322         struct sock *sk;
1323 
1324         EnterFunction(11);
1325 
1326         /* Already marked as IPVS request or reply? */
1327         if (skb->ipvs_property)
1328                 return NF_ACCEPT;
1329 
1330         sk = skb_to_full_sk(skb);
1331         /* Bad... Do not break raw sockets */
1332         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1333                      af == AF_INET)) {
1334 
1335                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1336                         return NF_ACCEPT;
1337         }
1338 
1339         if (unlikely(!skb_dst(skb)))
1340                 return NF_ACCEPT;
1341 
1342         if (!ipvs->enable)
1343                 return NF_ACCEPT;
1344 
1345         ip_vs_fill_iph_skb(af, skb, false, &iph);
1346 #ifdef CONFIG_IP_VS_IPV6
1347         if (af == AF_INET6) {
1348                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1349                         int related;
1350                         int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1351                                                         hooknum, &iph);
1352 
1353                         if (related)
1354                                 return verdict;
1355                 }
1356         } else
1357 #endif
1358                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1359                         int related;
1360                         int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1361 
1362                         if (related)
1363                                 return verdict;
1364                 }
1365 
1366         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1367         if (unlikely(!pd))
1368                 return NF_ACCEPT;
1369         pp = pd->pp;
1370 
1371         /* reassemble IP fragments */
1372 #ifdef CONFIG_IP_VS_IPV6
1373         if (af == AF_INET)
1374 #endif
1375                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1376                         if (ip_vs_gather_frags(ipvs, skb,
1377                                                ip_vs_defrag_user(hooknum)))
1378                                 return NF_STOLEN;
1379 
1380                         ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1381                 }
1382 
1383         /*
1384          * Check if the packet belongs to an existing entry
1385          */
1386         cp = pp->conn_out_get(ipvs, af, skb, &iph);
1387 
1388         if (likely(cp)) {
1389                 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
1390                         goto ignore_cp;
1391                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1392         }
1393 
1394         /* Check for real-server-started requests */
1395         if (atomic_read(&ipvs->conn_out_counter)) {
1396                 /* Currently only for UDP:
1397                  * connection oriented protocols typically use
1398                  * ephemeral ports for outgoing connections, so
1399                  * related incoming responses would not match any VS
1400                  */
1401                 if (pp->protocol == IPPROTO_UDP) {
1402                         cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph);
1403                         if (likely(cp))
1404                                 return handle_response(af, skb, pd, cp, &iph,
1405                                                        hooknum);
1406                 }
1407         }
1408 
1409         if (sysctl_nat_icmp_send(ipvs) &&
1410             (pp->protocol == IPPROTO_TCP ||
1411              pp->protocol == IPPROTO_UDP ||
1412              pp->protocol == IPPROTO_SCTP)) {
1413                 __be16 _ports[2], *pptr;
1414 
1415                 pptr = frag_safe_skb_hp(skb, iph.len,
1416                                          sizeof(_ports), _ports, &iph);
1417                 if (pptr == NULL)
1418                         return NF_ACCEPT;       /* Not for me */
1419                 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1420                                            pptr[0])) {
1421                         /*
1422                          * Notify the real server: there is no
1423                          * existing entry if it is not RST
1424                          * packet or not TCP packet.
1425                          */
1426                         if ((iph.protocol != IPPROTO_TCP &&
1427                              iph.protocol != IPPROTO_SCTP)
1428                              || ((iph.protocol == IPPROTO_TCP
1429                                   && !is_tcp_reset(skb, iph.len))
1430                                  || (iph.protocol == IPPROTO_SCTP
1431                                         && !is_sctp_abort(skb,
1432                                                 iph.len)))) {
1433 #ifdef CONFIG_IP_VS_IPV6
1434                                 if (af == AF_INET6) {
1435                                         if (!skb->dev)
1436                                                 skb->dev = ipvs->net->loopback_dev;
1437                                         icmpv6_send(skb,
1438                                                     ICMPV6_DEST_UNREACH,
1439                                                     ICMPV6_PORT_UNREACH,
1440                                                     0);
1441                                 } else
1442 #endif
1443                                         icmp_send(skb,
1444                                                   ICMP_DEST_UNREACH,
1445                                                   ICMP_PORT_UNREACH, 0);
1446                                 return NF_DROP;
1447                         }
1448                 }
1449         }
1450 
1451 out:
1452         IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1453                       "ip_vs_out: packet continues traversal as normal");
1454         return NF_ACCEPT;
1455 
1456 ignore_cp:
1457         __ip_vs_conn_put(cp);
1458         goto out;
1459 }
1460 
1461 /*
1462  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1463  *      used only for VS/NAT.
1464  *      Check if packet is reply for established ip_vs_conn.
1465  */
1466 static unsigned int
1467 ip_vs_reply4(void *priv, struct sk_buff *skb,
1468              const struct nf_hook_state *state)
1469 {
1470         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1471 }
1472 
1473 /*
1474  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1475  *      Check if packet is reply for established ip_vs_conn.
1476  */
1477 static unsigned int
1478 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1479                    const struct nf_hook_state *state)
1480 {
1481         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1482 }
1483 
1484 #ifdef CONFIG_IP_VS_IPV6
1485 
1486 /*
1487  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1488  *      used only for VS/NAT.
1489  *      Check if packet is reply for established ip_vs_conn.
1490  */
1491 static unsigned int
1492 ip_vs_reply6(void *priv, struct sk_buff *skb,
1493              const struct nf_hook_state *state)
1494 {
1495         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1496 }
1497 
1498 /*
1499  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1500  *      Check if packet is reply for established ip_vs_conn.
1501  */
1502 static unsigned int
1503 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1504                    const struct nf_hook_state *state)
1505 {
1506         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1507 }
1508 
1509 #endif
1510 
1511 static unsigned int
1512 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1513                       struct ip_vs_proto_data *pd,
1514                       int *verdict, struct ip_vs_conn **cpp,
1515                       struct ip_vs_iphdr *iph)
1516 {
1517         struct ip_vs_protocol *pp = pd->pp;
1518 
1519         if (!iph->fragoffs) {
1520                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1521                  * replayed fragment zero will already have created the cp
1522                  */
1523 
1524                 /* Schedule and create new connection entry into cpp */
1525                 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1526                         return 0;
1527         }
1528 
1529         if (unlikely(!*cpp)) {
1530                 /* sorry, all this trouble for a no-hit :) */
1531                 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1532                               "ip_vs_in: packet continues traversal as normal");
1533                 if (iph->fragoffs) {
1534                         /* Fragment that couldn't be mapped to a conn entry
1535                          * is missing module nf_defrag_ipv6
1536                          */
1537                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1538                         IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1539                                       "unhandled fragment");
1540                 }
1541                 *verdict = NF_ACCEPT;
1542                 return 0;
1543         }
1544 
1545         return 1;
1546 }
1547 
1548 /*
1549  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1550  *      Find any that might be relevant, check against existing connections,
1551  *      forward to the right destination host if relevant.
1552  *      Currently handles error types - unreachable, quench, ttl exceeded.
1553  */
1554 static int
1555 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1556               unsigned int hooknum)
1557 {
1558         struct iphdr *iph;
1559         struct icmphdr  _icmph, *ic;
1560         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1561         struct ip_vs_iphdr ciph;
1562         struct ip_vs_conn *cp;
1563         struct ip_vs_protocol *pp;
1564         struct ip_vs_proto_data *pd;
1565         unsigned int offset, offset2, ihl, verdict;
1566         bool ipip, new_cp = false;
1567 
1568         *related = 1;
1569 
1570         /* reassemble IP fragments */
1571         if (ip_is_fragment(ip_hdr(skb))) {
1572                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1573                         return NF_STOLEN;
1574         }
1575 
1576         iph = ip_hdr(skb);
1577         offset = ihl = iph->ihl * 4;
1578         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1579         if (ic == NULL)
1580                 return NF_DROP;
1581 
1582         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1583                   ic->type, ntohs(icmp_id(ic)),
1584                   &iph->saddr, &iph->daddr);
1585 
1586         /*
1587          * Work through seeing if this is for us.
1588          * These checks are supposed to be in an order that means easy
1589          * things are checked first to speed up processing.... however
1590          * this means that some packets will manage to get a long way
1591          * down this stack and then be rejected, but that's life.
1592          */
1593         if ((ic->type != ICMP_DEST_UNREACH) &&
1594             (ic->type != ICMP_SOURCE_QUENCH) &&
1595             (ic->type != ICMP_TIME_EXCEEDED)) {
1596                 *related = 0;
1597                 return NF_ACCEPT;
1598         }
1599 
1600         /* Now find the contained IP header */
1601         offset += sizeof(_icmph);
1602         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1603         if (cih == NULL)
1604                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1605 
1606         /* Special case for errors for IPIP packets */
1607         ipip = false;
1608         if (cih->protocol == IPPROTO_IPIP) {
1609                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1610                         return NF_ACCEPT;
1611                 /* Error for our IPIP must arrive at LOCAL_IN */
1612                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1613                         return NF_ACCEPT;
1614                 offset += cih->ihl * 4;
1615                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1616                 if (cih == NULL)
1617                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1618                 ipip = true;
1619         }
1620 
1621         pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1622         if (!pd)
1623                 return NF_ACCEPT;
1624         pp = pd->pp;
1625 
1626         /* Is the embedded protocol header present? */
1627         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1628                      pp->dont_defrag))
1629                 return NF_ACCEPT;
1630 
1631         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1632                       "Checking incoming ICMP for");
1633 
1634         offset2 = offset;
1635         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1636         offset = ciph.len;
1637 
1638         /* The embedded headers contain source and dest in reverse order.
1639          * For IPIP this is error for request, not for reply.
1640          */
1641         cp = pp->conn_in_get(ipvs, AF_INET, skb, &ciph);
1642 
1643         if (!cp) {
1644                 int v;
1645 
1646                 if (!sysctl_schedule_icmp(ipvs))
1647                         return NF_ACCEPT;
1648 
1649                 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1650                         return v;
1651                 new_cp = true;
1652         }
1653 
1654         verdict = NF_DROP;
1655 
1656         /* Ensure the checksum is correct */
1657         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1658                 /* Failed checksum! */
1659                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1660                           &iph->saddr);
1661                 goto out;
1662         }
1663 
1664         if (ipip) {
1665                 __be32 info = ic->un.gateway;
1666                 __u8 type = ic->type;
1667                 __u8 code = ic->code;
1668 
1669                 /* Update the MTU */
1670                 if (ic->type == ICMP_DEST_UNREACH &&
1671                     ic->code == ICMP_FRAG_NEEDED) {
1672                         struct ip_vs_dest *dest = cp->dest;
1673                         u32 mtu = ntohs(ic->un.frag.mtu);
1674                         __be16 frag_off = cih->frag_off;
1675 
1676                         /* Strip outer IP and ICMP, go to IPIP header */
1677                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1678                                 goto ignore_ipip;
1679                         offset2 -= ihl + sizeof(_icmph);
1680                         skb_reset_network_header(skb);
1681                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1682                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1683                         ipv4_update_pmtu(skb, ipvs->net,
1684                                          mtu, 0, 0, 0, 0);
1685                         /* Client uses PMTUD? */
1686                         if (!(frag_off & htons(IP_DF)))
1687                                 goto ignore_ipip;
1688                         /* Prefer the resulting PMTU */
1689                         if (dest) {
1690                                 struct ip_vs_dest_dst *dest_dst;
1691 
1692                                 rcu_read_lock();
1693                                 dest_dst = rcu_dereference(dest->dest_dst);
1694                                 if (dest_dst)
1695                                         mtu = dst_mtu(dest_dst->dst_cache);
1696                                 rcu_read_unlock();
1697                         }
1698                         if (mtu > 68 + sizeof(struct iphdr))
1699                                 mtu -= sizeof(struct iphdr);
1700                         info = htonl(mtu);
1701                 }
1702                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1703                  * original request.
1704                  */
1705                 if (pskb_pull(skb, offset2) == NULL)
1706                         goto ignore_ipip;
1707                 skb_reset_network_header(skb);
1708                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1709                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1710                         type, code, ntohl(info));
1711                 icmp_send(skb, type, code, info);
1712                 /* ICMP can be shorter but anyways, account it */
1713                 ip_vs_out_stats(cp, skb);
1714 
1715 ignore_ipip:
1716                 consume_skb(skb);
1717                 verdict = NF_STOLEN;
1718                 goto out;
1719         }
1720 
1721         /* do the statistics and put it back */
1722         ip_vs_in_stats(cp, skb);
1723         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1724             IPPROTO_SCTP == cih->protocol)
1725                 offset += 2 * sizeof(__u16);
1726         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1727 
1728 out:
1729         if (likely(!new_cp))
1730                 __ip_vs_conn_put(cp);
1731         else
1732                 ip_vs_conn_put(cp);
1733 
1734         return verdict;
1735 }
1736 
1737 #ifdef CONFIG_IP_VS_IPV6
1738 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1739                             int *related, unsigned int hooknum,
1740                             struct ip_vs_iphdr *iph)
1741 {
1742         struct icmp6hdr _icmph, *ic;
1743         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1744         struct ip_vs_conn *cp;
1745         struct ip_vs_protocol *pp;
1746         struct ip_vs_proto_data *pd;
1747         unsigned int offset, verdict;
1748         bool new_cp = false;
1749 
1750         *related = 1;
1751 
1752         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1753         if (ic == NULL)
1754                 return NF_DROP;
1755 
1756         /*
1757          * Work through seeing if this is for us.
1758          * These checks are supposed to be in an order that means easy
1759          * things are checked first to speed up processing.... however
1760          * this means that some packets will manage to get a long way
1761          * down this stack and then be rejected, but that's life.
1762          */
1763         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1764                 *related = 0;
1765                 return NF_ACCEPT;
1766         }
1767         /* Fragment header that is before ICMP header tells us that:
1768          * it's not an error message since they can't be fragmented.
1769          */
1770         if (iph->flags & IP6_FH_F_FRAG)
1771                 return NF_DROP;
1772 
1773         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1774                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1775                   &iph->saddr, &iph->daddr);
1776 
1777         offset = iph->len + sizeof(_icmph);
1778         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1779                 return NF_ACCEPT;
1780 
1781         pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1782         if (!pd)
1783                 return NF_ACCEPT;
1784         pp = pd->pp;
1785 
1786         /* Cannot handle fragmented embedded protocol */
1787         if (ciph.fragoffs)
1788                 return NF_ACCEPT;
1789 
1790         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1791                       "Checking incoming ICMPv6 for");
1792 
1793         /* The embedded headers contain source and dest in reverse order
1794          * if not from localhost
1795          */
1796         cp = pp->conn_in_get(ipvs, AF_INET6, skb, &ciph);
1797 
1798         if (!cp) {
1799                 int v;
1800 
1801                 if (!sysctl_schedule_icmp(ipvs))
1802                         return NF_ACCEPT;
1803 
1804                 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1805                         return v;
1806 
1807                 new_cp = true;
1808         }
1809 
1810         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1811         if ((hooknum == NF_INET_LOCAL_OUT) &&
1812             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1813                 verdict = NF_ACCEPT;
1814                 goto out;
1815         }
1816 
1817         /* do the statistics and put it back */
1818         ip_vs_in_stats(cp, skb);
1819 
1820         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1821         offset = ciph.len;
1822         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1823             IPPROTO_SCTP == ciph.protocol)
1824                 offset += 2 * sizeof(__u16); /* Also mangle ports */
1825 
1826         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1827 
1828 out:
1829         if (likely(!new_cp))
1830                 __ip_vs_conn_put(cp);
1831         else
1832                 ip_vs_conn_put(cp);
1833 
1834         return verdict;
1835 }
1836 #endif
1837 
1838 
1839 /*
1840  *      Check if it's for virtual services, look it up,
1841  *      and send it on its way...
1842  */
1843 static unsigned int
1844 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1845 {
1846         struct ip_vs_iphdr iph;
1847         struct ip_vs_protocol *pp;
1848         struct ip_vs_proto_data *pd;
1849         struct ip_vs_conn *cp;
1850         int ret, pkts;
1851         int conn_reuse_mode;
1852         struct sock *sk;
1853 
1854         /* Already marked as IPVS request or reply? */
1855         if (skb->ipvs_property)
1856                 return NF_ACCEPT;
1857 
1858         /*
1859          *      Big tappo:
1860          *      - remote client: only PACKET_HOST
1861          *      - route: used for struct net when skb->dev is unset
1862          */
1863         if (unlikely((skb->pkt_type != PACKET_HOST &&
1864                       hooknum != NF_INET_LOCAL_OUT) ||
1865                      !skb_dst(skb))) {
1866                 ip_vs_fill_iph_skb(af, skb, false, &iph);
1867                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1868                               " ignored in hook %u\n",
1869                               skb->pkt_type, iph.protocol,
1870                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1871                 return NF_ACCEPT;
1872         }
1873         /* ipvs enabled in this netns ? */
1874         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1875                 return NF_ACCEPT;
1876 
1877         ip_vs_fill_iph_skb(af, skb, false, &iph);
1878 
1879         /* Bad... Do not break raw sockets */
1880         sk = skb_to_full_sk(skb);
1881         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1882                      af == AF_INET)) {
1883 
1884                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1885                         return NF_ACCEPT;
1886         }
1887 
1888 #ifdef CONFIG_IP_VS_IPV6
1889         if (af == AF_INET6) {
1890                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1891                         int related;
1892                         int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
1893                                                        hooknum, &iph);
1894 
1895                         if (related)
1896                                 return verdict;
1897                 }
1898         } else
1899 #endif
1900                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1901                         int related;
1902                         int verdict = ip_vs_in_icmp(ipvs, skb, &related,
1903                                                     hooknum);
1904 
1905                         if (related)
1906                                 return verdict;
1907                 }
1908 
1909         /* Protocol supported? */
1910         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1911         if (unlikely(!pd)) {
1912                 /* The only way we'll see this packet again is if it's
1913                  * encapsulated, so mark it with ipvs_property=1 so we
1914                  * skip it if we're ignoring tunneled packets
1915                  */
1916                 if (sysctl_ignore_tunneled(ipvs))
1917                         skb->ipvs_property = 1;
1918 
1919                 return NF_ACCEPT;
1920         }
1921         pp = pd->pp;
1922         /*
1923          * Check if the packet belongs to an existing connection entry
1924          */
1925         cp = pp->conn_in_get(ipvs, af, skb, &iph);
1926 
1927         conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1928         if (conn_reuse_mode && !iph.fragoffs && is_new_conn(skb, &iph) && cp) {
1929                 bool uses_ct = false, resched = false;
1930 
1931                 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1932                     unlikely(!atomic_read(&cp->dest->weight))) {
1933                         resched = true;
1934                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1935                 } else if (is_new_conn_expected(cp, conn_reuse_mode)) {
1936                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1937                         if (!atomic_read(&cp->n_control)) {
1938                                 resched = true;
1939                         } else {
1940                                 /* Do not reschedule controlling connection
1941                                  * that uses conntrack while it is still
1942                                  * referenced by controlled connection(s).
1943                                  */
1944                                 resched = !uses_ct;
1945                         }
1946                 }
1947 
1948                 if (resched) {
1949                         if (!atomic_read(&cp->n_control))
1950                                 ip_vs_conn_expire_now(cp);
1951                         __ip_vs_conn_put(cp);
1952                         if (uses_ct)
1953                                 return NF_DROP;
1954                         cp = NULL;
1955                 }
1956         }
1957 
1958         if (unlikely(!cp)) {
1959                 int v;
1960 
1961                 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
1962                         return v;
1963         }
1964 
1965         IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
1966 
1967         /* Check the server status */
1968         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1969                 /* the destination server is not available */
1970 
1971                 if (sysctl_expire_nodest_conn(ipvs)) {
1972                         /* try to expire the connection immediately */
1973                         ip_vs_conn_expire_now(cp);
1974                 }
1975                 /* don't restart its timer, and silently
1976                    drop the packet. */
1977                 __ip_vs_conn_put(cp);
1978                 return NF_DROP;
1979         }
1980 
1981         ip_vs_in_stats(cp, skb);
1982         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1983         if (cp->packet_xmit)
1984                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1985                 /* do not touch skb anymore */
1986         else {
1987                 IP_VS_DBG_RL("warning: packet_xmit is null");
1988                 ret = NF_ACCEPT;
1989         }
1990 
1991         /* Increase its packet counter and check if it is needed
1992          * to be synchronized
1993          *
1994          * Sync connection if it is about to close to
1995          * encorage the standby servers to update the connections timeout
1996          *
1997          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1998          */
1999 
2000         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
2001                 pkts = sysctl_sync_threshold(ipvs);
2002         else
2003                 pkts = atomic_add_return(1, &cp->in_pkts);
2004 
2005         if (ipvs->sync_state & IP_VS_STATE_MASTER)
2006                 ip_vs_sync_conn(ipvs, cp, pkts);
2007         else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
2008                 /* increment is done inside ip_vs_sync_conn too */
2009                 atomic_inc(&cp->control->in_pkts);
2010 
2011         ip_vs_conn_put(cp);
2012         return ret;
2013 }
2014 
2015 /*
2016  *      AF_INET handler in NF_INET_LOCAL_IN chain
2017  *      Schedule and forward packets from remote clients
2018  */
2019 static unsigned int
2020 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
2021                       const struct nf_hook_state *state)
2022 {
2023         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2024 }
2025 
2026 /*
2027  *      AF_INET handler in NF_INET_LOCAL_OUT chain
2028  *      Schedule and forward packets from local clients
2029  */
2030 static unsigned int
2031 ip_vs_local_request4(void *priv, struct sk_buff *skb,
2032                      const struct nf_hook_state *state)
2033 {
2034         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2035 }
2036 
2037 #ifdef CONFIG_IP_VS_IPV6
2038 
2039 /*
2040  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
2041  *      Schedule and forward packets from remote clients
2042  */
2043 static unsigned int
2044 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
2045                       const struct nf_hook_state *state)
2046 {
2047         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2048 }
2049 
2050 /*
2051  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
2052  *      Schedule and forward packets from local clients
2053  */
2054 static unsigned int
2055 ip_vs_local_request6(void *priv, struct sk_buff *skb,
2056                      const struct nf_hook_state *state)
2057 {
2058         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2059 }
2060 
2061 #endif
2062 
2063 
2064 /*
2065  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
2066  *      related packets destined for 0.0.0.0/0.
2067  *      When fwmark-based virtual service is used, such as transparent
2068  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
2069  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
2070  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
2071  *      and send them to ip_vs_in_icmp.
2072  */
2073 static unsigned int
2074 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
2075                    const struct nf_hook_state *state)
2076 {
2077         int r;
2078         struct netns_ipvs *ipvs = net_ipvs(state->net);
2079 
2080         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
2081                 return NF_ACCEPT;
2082 
2083         /* ipvs enabled in this netns ? */
2084         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2085                 return NF_ACCEPT;
2086 
2087         return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
2088 }
2089 
2090 #ifdef CONFIG_IP_VS_IPV6
2091 static unsigned int
2092 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
2093                       const struct nf_hook_state *state)
2094 {
2095         int r;
2096         struct netns_ipvs *ipvs = net_ipvs(state->net);
2097         struct ip_vs_iphdr iphdr;
2098 
2099         ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
2100         if (iphdr.protocol != IPPROTO_ICMPV6)
2101                 return NF_ACCEPT;
2102 
2103         /* ipvs enabled in this netns ? */
2104         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2105                 return NF_ACCEPT;
2106 
2107         return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
2108 }
2109 #endif
2110 
2111 
2112 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
2113         /* After packet filtering, change source only for VS/NAT */
2114         {
2115                 .hook           = ip_vs_reply4,
2116                 .pf             = NFPROTO_IPV4,
2117                 .hooknum        = NF_INET_LOCAL_IN,
2118                 .priority       = NF_IP_PRI_NAT_SRC - 2,
2119         },
2120         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2121          * or VS/NAT(change destination), so that filtering rules can be
2122          * applied to IPVS. */
2123         {
2124                 .hook           = ip_vs_remote_request4,
2125                 .pf             = NFPROTO_IPV4,
2126                 .hooknum        = NF_INET_LOCAL_IN,
2127                 .priority       = NF_IP_PRI_NAT_SRC - 1,
2128         },
2129         /* Before ip_vs_in, change source only for VS/NAT */
2130         {
2131                 .hook           = ip_vs_local_reply4,
2132                 .pf             = NFPROTO_IPV4,
2133                 .hooknum        = NF_INET_LOCAL_OUT,
2134                 .priority       = NF_IP_PRI_NAT_DST + 1,
2135         },
2136         /* After mangle, schedule and forward local requests */
2137         {
2138                 .hook           = ip_vs_local_request4,
2139                 .pf             = NFPROTO_IPV4,
2140                 .hooknum        = NF_INET_LOCAL_OUT,
2141                 .priority       = NF_IP_PRI_NAT_DST + 2,
2142         },
2143         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2144          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2145         {
2146                 .hook           = ip_vs_forward_icmp,
2147                 .pf             = NFPROTO_IPV4,
2148                 .hooknum        = NF_INET_FORWARD,
2149                 .priority       = 99,
2150         },
2151         /* After packet filtering, change source only for VS/NAT */
2152         {
2153                 .hook           = ip_vs_reply4,
2154                 .pf             = NFPROTO_IPV4,
2155                 .hooknum        = NF_INET_FORWARD,
2156                 .priority       = 100,
2157         },
2158 #ifdef CONFIG_IP_VS_IPV6
2159         /* After packet filtering, change source only for VS/NAT */
2160         {
2161                 .hook           = ip_vs_reply6,
2162                 .pf             = NFPROTO_IPV6,
2163                 .hooknum        = NF_INET_LOCAL_IN,
2164                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
2165         },
2166         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2167          * or VS/NAT(change destination), so that filtering rules can be
2168          * applied to IPVS. */
2169         {
2170                 .hook           = ip_vs_remote_request6,
2171                 .pf             = NFPROTO_IPV6,
2172                 .hooknum        = NF_INET_LOCAL_IN,
2173                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
2174         },
2175         /* Before ip_vs_in, change source only for VS/NAT */
2176         {
2177                 .hook           = ip_vs_local_reply6,
2178                 .pf             = NFPROTO_IPV6,
2179                 .hooknum        = NF_INET_LOCAL_OUT,
2180                 .priority       = NF_IP6_PRI_NAT_DST + 1,
2181         },
2182         /* After mangle, schedule and forward local requests */
2183         {
2184                 .hook           = ip_vs_local_request6,
2185                 .pf             = NFPROTO_IPV6,
2186                 .hooknum        = NF_INET_LOCAL_OUT,
2187                 .priority       = NF_IP6_PRI_NAT_DST + 2,
2188         },
2189         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2190          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2191         {
2192                 .hook           = ip_vs_forward_icmp_v6,
2193                 .pf             = NFPROTO_IPV6,
2194                 .hooknum        = NF_INET_FORWARD,
2195                 .priority       = 99,
2196         },
2197         /* After packet filtering, change source only for VS/NAT */
2198         {
2199                 .hook           = ip_vs_reply6,
2200                 .pf             = NFPROTO_IPV6,
2201                 .hooknum        = NF_INET_FORWARD,
2202                 .priority       = 100,
2203         },
2204 #endif
2205 };
2206 /*
2207  *      Initialize IP Virtual Server netns mem.
2208  */
2209 static int __net_init __ip_vs_init(struct net *net)
2210 {
2211         struct netns_ipvs *ipvs;
2212         int ret;
2213 
2214         ipvs = net_generic(net, ip_vs_net_id);
2215         if (ipvs == NULL)
2216                 return -ENOMEM;
2217 
2218         /* Hold the beast until a service is registerd */
2219         ipvs->enable = 0;
2220         ipvs->net = net;
2221         /* Counters used for creating unique names */
2222         ipvs->gen = atomic_read(&ipvs_netns_cnt);
2223         atomic_inc(&ipvs_netns_cnt);
2224         net->ipvs = ipvs;
2225 
2226         if (ip_vs_estimator_net_init(ipvs) < 0)
2227                 goto estimator_fail;
2228 
2229         if (ip_vs_control_net_init(ipvs) < 0)
2230                 goto control_fail;
2231 
2232         if (ip_vs_protocol_net_init(ipvs) < 0)
2233                 goto protocol_fail;
2234 
2235         if (ip_vs_app_net_init(ipvs) < 0)
2236                 goto app_fail;
2237 
2238         if (ip_vs_conn_net_init(ipvs) < 0)
2239                 goto conn_fail;
2240 
2241         if (ip_vs_sync_net_init(ipvs) < 0)
2242                 goto sync_fail;
2243 
2244         ret = nf_register_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2245         if (ret < 0)
2246                 goto hook_fail;
2247 
2248         return 0;
2249 /*
2250  * Error handling
2251  */
2252 
2253 hook_fail:
2254         ip_vs_sync_net_cleanup(ipvs);
2255 sync_fail:
2256         ip_vs_conn_net_cleanup(ipvs);
2257 conn_fail:
2258         ip_vs_app_net_cleanup(ipvs);
2259 app_fail:
2260         ip_vs_protocol_net_cleanup(ipvs);
2261 protocol_fail:
2262         ip_vs_control_net_cleanup(ipvs);
2263 control_fail:
2264         ip_vs_estimator_net_cleanup(ipvs);
2265 estimator_fail:
2266         net->ipvs = NULL;
2267         return -ENOMEM;
2268 }
2269 
2270 static void __net_exit __ip_vs_cleanup(struct net *net)
2271 {
2272         struct netns_ipvs *ipvs = net_ipvs(net);
2273 
2274         nf_unregister_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2275         ip_vs_service_net_cleanup(ipvs);        /* ip_vs_flush() with locks */
2276         ip_vs_conn_net_cleanup(ipvs);
2277         ip_vs_app_net_cleanup(ipvs);
2278         ip_vs_protocol_net_cleanup(ipvs);
2279         ip_vs_control_net_cleanup(ipvs);
2280         ip_vs_estimator_net_cleanup(ipvs);
2281         IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2282         net->ipvs = NULL;
2283 }
2284 
2285 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2286 {
2287         struct netns_ipvs *ipvs = net_ipvs(net);
2288         EnterFunction(2);
2289         ipvs->enable = 0;       /* Disable packet reception */
2290         smp_wmb();
2291         ip_vs_sync_net_cleanup(ipvs);
2292         LeaveFunction(2);
2293 }
2294 
2295 static struct pernet_operations ipvs_core_ops = {
2296         .init = __ip_vs_init,
2297         .exit = __ip_vs_cleanup,
2298         .id   = &ip_vs_net_id,
2299         .size = sizeof(struct netns_ipvs),
2300 };
2301 
2302 static struct pernet_operations ipvs_core_dev_ops = {
2303         .exit = __ip_vs_dev_cleanup,
2304 };
2305 
2306 /*
2307  *      Initialize IP Virtual Server
2308  */
2309 static int __init ip_vs_init(void)
2310 {
2311         int ret;
2312 
2313         ret = ip_vs_control_init();
2314         if (ret < 0) {
2315                 pr_err("can't setup control.\n");
2316                 goto exit;
2317         }
2318 
2319         ip_vs_protocol_init();
2320 
2321         ret = ip_vs_conn_init();
2322         if (ret < 0) {
2323                 pr_err("can't setup connection table.\n");
2324                 goto cleanup_protocol;
2325         }
2326 
2327         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2328         if (ret < 0)
2329                 goto cleanup_conn;
2330 
2331         ret = register_pernet_device(&ipvs_core_dev_ops);
2332         if (ret < 0)
2333                 goto cleanup_sub;
2334 
2335         ret = ip_vs_register_nl_ioctl();
2336         if (ret < 0) {
2337                 pr_err("can't register netlink/ioctl.\n");
2338                 goto cleanup_dev;
2339         }
2340 
2341         pr_info("ipvs loaded.\n");
2342 
2343         return ret;
2344 
2345 cleanup_dev:
2346         unregister_pernet_device(&ipvs_core_dev_ops);
2347 cleanup_sub:
2348         unregister_pernet_subsys(&ipvs_core_ops);
2349 cleanup_conn:
2350         ip_vs_conn_cleanup();
2351 cleanup_protocol:
2352         ip_vs_protocol_cleanup();
2353         ip_vs_control_cleanup();
2354 exit:
2355         return ret;
2356 }
2357 
2358 static void __exit ip_vs_cleanup(void)
2359 {
2360         ip_vs_unregister_nl_ioctl();
2361         unregister_pernet_device(&ipvs_core_dev_ops);
2362         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2363         ip_vs_conn_cleanup();
2364         ip_vs_protocol_cleanup();
2365         ip_vs_control_cleanup();
2366         pr_info("ipvs unloaded.\n");
2367 }
2368 
2369 module_init(ip_vs_init);
2370 module_exit(ip_vs_cleanup);
2371 MODULE_LICENSE("GPL");
2372 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp