ipvs: extend connection flags to 32 bits
[linux-2.6.git] / net / netfilter / ipvs / ip_vs_conn.c
1 /*
2  * IPVS         An implementation of the IP virtual server support for the
3  *              LINUX operating system.  IPVS is now implemented as a module
4  *              over the Netfilter framework. IPVS can be used to build a
5  *              high-performance and highly available server based on a
6  *              cluster of servers.
7  *
8  * 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. Many code here is taken from IP MASQ code of kernel 2.2.
20  *
21  * Changes:
22  *
23  */
24
25 #define KMSG_COMPONENT "IPVS"
26 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
27
28 #include <linux/interrupt.h>
29 #include <linux/in.h>
30 #include <linux/net.h>
31 #include <linux/kernel.h>
32 #include <linux/module.h>
33 #include <linux/vmalloc.h>
34 #include <linux/proc_fs.h>              /* for proc_net_* */
35 #include <linux/slab.h>
36 #include <linux/seq_file.h>
37 #include <linux/jhash.h>
38 #include <linux/random.h>
39
40 #include <net/net_namespace.h>
41 #include <net/ip_vs.h>
42
43
44 #ifndef CONFIG_IP_VS_TAB_BITS
45 #define CONFIG_IP_VS_TAB_BITS   12
46 #endif
47
48 /*
49  * Connection hash size. Default is what was selected at compile time.
50 */
51 int ip_vs_conn_tab_bits = CONFIG_IP_VS_TAB_BITS;
52 module_param_named(conn_tab_bits, ip_vs_conn_tab_bits, int, 0444);
53 MODULE_PARM_DESC(conn_tab_bits, "Set connections' hash size");
54
55 /* size and mask values */
56 int ip_vs_conn_tab_size;
57 int ip_vs_conn_tab_mask;
58
59 /*
60  *  Connection hash table: for input and output packets lookups of IPVS
61  */
62 static struct list_head *ip_vs_conn_tab;
63
64 /*  SLAB cache for IPVS connections */
65 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
66
67 /*  counter for current IPVS connections */
68 static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
69
70 /*  counter for no client port connections */
71 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
72
73 /* random value for IPVS connection hash */
74 static unsigned int ip_vs_conn_rnd;
75
76 /*
77  *  Fine locking granularity for big connection hash table
78  */
79 #define CT_LOCKARRAY_BITS  4
80 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
81 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
82
83 struct ip_vs_aligned_lock
84 {
85         rwlock_t        l;
86 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
87
88 /* lock array for conn table */
89 static struct ip_vs_aligned_lock
90 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
91
92 static inline void ct_read_lock(unsigned key)
93 {
94         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
95 }
96
97 static inline void ct_read_unlock(unsigned key)
98 {
99         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
100 }
101
102 static inline void ct_write_lock(unsigned key)
103 {
104         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
105 }
106
107 static inline void ct_write_unlock(unsigned key)
108 {
109         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
110 }
111
112 static inline void ct_read_lock_bh(unsigned key)
113 {
114         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
115 }
116
117 static inline void ct_read_unlock_bh(unsigned key)
118 {
119         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
120 }
121
122 static inline void ct_write_lock_bh(unsigned key)
123 {
124         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
125 }
126
127 static inline void ct_write_unlock_bh(unsigned key)
128 {
129         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
130 }
131
132
133 /*
134  *      Returns hash value for IPVS connection entry
135  */
136 static unsigned int ip_vs_conn_hashkey(int af, unsigned proto,
137                                        const union nf_inet_addr *addr,
138                                        __be16 port)
139 {
140 #ifdef CONFIG_IP_VS_IPV6
141         if (af == AF_INET6)
142                 return jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
143                                     (__force u32)port, proto, ip_vs_conn_rnd)
144                         & ip_vs_conn_tab_mask;
145 #endif
146         return jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
147                             ip_vs_conn_rnd)
148                 & ip_vs_conn_tab_mask;
149 }
150
151
152 /*
153  *      Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
154  *      returns bool success.
155  */
156 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
157 {
158         unsigned hash;
159         int ret;
160
161         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
162                 return 0;
163
164         /* Hash by protocol, client address and port */
165         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
166
167         ct_write_lock(hash);
168         spin_lock(&cp->lock);
169
170         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
171                 list_add(&cp->c_list, &ip_vs_conn_tab[hash]);
172                 cp->flags |= IP_VS_CONN_F_HASHED;
173                 atomic_inc(&cp->refcnt);
174                 ret = 1;
175         } else {
176                 pr_err("%s(): request for already hashed, called from %pF\n",
177                        __func__, __builtin_return_address(0));
178                 ret = 0;
179         }
180
181         spin_unlock(&cp->lock);
182         ct_write_unlock(hash);
183
184         return ret;
185 }
186
187
188 /*
189  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
190  *      returns bool success.
191  */
192 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
193 {
194         unsigned hash;
195         int ret;
196
197         /* unhash it and decrease its reference counter */
198         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
199
200         ct_write_lock(hash);
201         spin_lock(&cp->lock);
202
203         if (cp->flags & IP_VS_CONN_F_HASHED) {
204                 list_del(&cp->c_list);
205                 cp->flags &= ~IP_VS_CONN_F_HASHED;
206                 atomic_dec(&cp->refcnt);
207                 ret = 1;
208         } else
209                 ret = 0;
210
211         spin_unlock(&cp->lock);
212         ct_write_unlock(hash);
213
214         return ret;
215 }
216
217
218 /*
219  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
220  *  Called for pkts coming from OUTside-to-INside.
221  *      s_addr, s_port: pkt source address (foreign host)
222  *      d_addr, d_port: pkt dest address (load balancer)
223  */
224 static inline struct ip_vs_conn *__ip_vs_conn_in_get
225 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
226  const union nf_inet_addr *d_addr, __be16 d_port)
227 {
228         unsigned hash;
229         struct ip_vs_conn *cp;
230
231         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
232
233         ct_read_lock(hash);
234
235         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
236                 if (cp->af == af &&
237                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
238                     ip_vs_addr_equal(af, d_addr, &cp->vaddr) &&
239                     s_port == cp->cport && d_port == cp->vport &&
240                     ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
241                     protocol == cp->protocol) {
242                         /* HIT */
243                         atomic_inc(&cp->refcnt);
244                         ct_read_unlock(hash);
245                         return cp;
246                 }
247         }
248
249         ct_read_unlock(hash);
250
251         return NULL;
252 }
253
254 struct ip_vs_conn *ip_vs_conn_in_get
255 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
256  const union nf_inet_addr *d_addr, __be16 d_port)
257 {
258         struct ip_vs_conn *cp;
259
260         cp = __ip_vs_conn_in_get(af, protocol, s_addr, s_port, d_addr, d_port);
261         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt))
262                 cp = __ip_vs_conn_in_get(af, protocol, s_addr, 0, d_addr,
263                                          d_port);
264
265         IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
266                       ip_vs_proto_name(protocol),
267                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
268                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
269                       cp ? "hit" : "not hit");
270
271         return cp;
272 }
273
274 struct ip_vs_conn *
275 ip_vs_conn_in_get_proto(int af, const struct sk_buff *skb,
276                         struct ip_vs_protocol *pp,
277                         const struct ip_vs_iphdr *iph,
278                         unsigned int proto_off, int inverse)
279 {
280         __be16 _ports[2], *pptr;
281
282         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
283         if (pptr == NULL)
284                 return NULL;
285
286         if (likely(!inverse))
287                 return ip_vs_conn_in_get(af, iph->protocol,
288                                          &iph->saddr, pptr[0],
289                                          &iph->daddr, pptr[1]);
290         else
291                 return ip_vs_conn_in_get(af, iph->protocol,
292                                          &iph->daddr, pptr[1],
293                                          &iph->saddr, pptr[0]);
294 }
295 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
296
297 /* Get reference to connection template */
298 struct ip_vs_conn *ip_vs_ct_in_get
299 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
300  const union nf_inet_addr *d_addr, __be16 d_port)
301 {
302         unsigned hash;
303         struct ip_vs_conn *cp;
304
305         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
306
307         ct_read_lock(hash);
308
309         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
310                 if (cp->af == af &&
311                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
312                     /* protocol should only be IPPROTO_IP if
313                      * d_addr is a fwmark */
314                     ip_vs_addr_equal(protocol == IPPROTO_IP ? AF_UNSPEC : af,
315                                      d_addr, &cp->vaddr) &&
316                     s_port == cp->cport && d_port == cp->vport &&
317                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
318                     protocol == cp->protocol) {
319                         /* HIT */
320                         atomic_inc(&cp->refcnt);
321                         goto out;
322                 }
323         }
324         cp = NULL;
325
326   out:
327         ct_read_unlock(hash);
328
329         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
330                       ip_vs_proto_name(protocol),
331                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
332                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
333                       cp ? "hit" : "not hit");
334
335         return cp;
336 }
337
338 /*
339  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
340  *  Called for pkts coming from inside-to-OUTside.
341  *      s_addr, s_port: pkt source address (inside host)
342  *      d_addr, d_port: pkt dest address (foreign host)
343  */
344 struct ip_vs_conn *ip_vs_conn_out_get
345 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
346  const union nf_inet_addr *d_addr, __be16 d_port)
347 {
348         unsigned hash;
349         struct ip_vs_conn *cp, *ret=NULL;
350
351         /*
352          *      Check for "full" addressed entries
353          */
354         hash = ip_vs_conn_hashkey(af, protocol, d_addr, d_port);
355
356         ct_read_lock(hash);
357
358         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
359                 if (cp->af == af &&
360                     ip_vs_addr_equal(af, d_addr, &cp->caddr) &&
361                     ip_vs_addr_equal(af, s_addr, &cp->daddr) &&
362                     d_port == cp->cport && s_port == cp->dport &&
363                     protocol == cp->protocol) {
364                         /* HIT */
365                         atomic_inc(&cp->refcnt);
366                         ret = cp;
367                         break;
368                 }
369         }
370
371         ct_read_unlock(hash);
372
373         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
374                       ip_vs_proto_name(protocol),
375                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
376                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
377                       ret ? "hit" : "not hit");
378
379         return ret;
380 }
381
382 struct ip_vs_conn *
383 ip_vs_conn_out_get_proto(int af, const struct sk_buff *skb,
384                          struct ip_vs_protocol *pp,
385                          const struct ip_vs_iphdr *iph,
386                          unsigned int proto_off, int inverse)
387 {
388         __be16 _ports[2], *pptr;
389
390         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
391         if (pptr == NULL)
392                 return NULL;
393
394         if (likely(!inverse))
395                 return ip_vs_conn_out_get(af, iph->protocol,
396                                           &iph->saddr, pptr[0],
397                                           &iph->daddr, pptr[1]);
398         else
399                 return ip_vs_conn_out_get(af, iph->protocol,
400                                           &iph->daddr, pptr[1],
401                                           &iph->saddr, pptr[0]);
402 }
403 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
404
405 /*
406  *      Put back the conn and restart its timer with its timeout
407  */
408 void ip_vs_conn_put(struct ip_vs_conn *cp)
409 {
410         unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
411                 0 : cp->timeout;
412         mod_timer(&cp->timer, jiffies+t);
413
414         __ip_vs_conn_put(cp);
415 }
416
417
418 /*
419  *      Fill a no_client_port connection with a client port number
420  */
421 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
422 {
423         if (ip_vs_conn_unhash(cp)) {
424                 spin_lock(&cp->lock);
425                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
426                         atomic_dec(&ip_vs_conn_no_cport_cnt);
427                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
428                         cp->cport = cport;
429                 }
430                 spin_unlock(&cp->lock);
431
432                 /* hash on new dport */
433                 ip_vs_conn_hash(cp);
434         }
435 }
436
437
438 /*
439  *      Bind a connection entry with the corresponding packet_xmit.
440  *      Called by ip_vs_conn_new.
441  */
442 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
443 {
444         switch (IP_VS_FWD_METHOD(cp)) {
445         case IP_VS_CONN_F_MASQ:
446                 cp->packet_xmit = ip_vs_nat_xmit;
447                 break;
448
449         case IP_VS_CONN_F_TUNNEL:
450                 cp->packet_xmit = ip_vs_tunnel_xmit;
451                 break;
452
453         case IP_VS_CONN_F_DROUTE:
454                 cp->packet_xmit = ip_vs_dr_xmit;
455                 break;
456
457         case IP_VS_CONN_F_LOCALNODE:
458                 cp->packet_xmit = ip_vs_null_xmit;
459                 break;
460
461         case IP_VS_CONN_F_BYPASS:
462                 cp->packet_xmit = ip_vs_bypass_xmit;
463                 break;
464         }
465 }
466
467 #ifdef CONFIG_IP_VS_IPV6
468 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
469 {
470         switch (IP_VS_FWD_METHOD(cp)) {
471         case IP_VS_CONN_F_MASQ:
472                 cp->packet_xmit = ip_vs_nat_xmit_v6;
473                 break;
474
475         case IP_VS_CONN_F_TUNNEL:
476                 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
477                 break;
478
479         case IP_VS_CONN_F_DROUTE:
480                 cp->packet_xmit = ip_vs_dr_xmit_v6;
481                 break;
482
483         case IP_VS_CONN_F_LOCALNODE:
484                 cp->packet_xmit = ip_vs_null_xmit;
485                 break;
486
487         case IP_VS_CONN_F_BYPASS:
488                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
489                 break;
490         }
491 }
492 #endif
493
494
495 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
496 {
497         return atomic_read(&dest->activeconns)
498                 + atomic_read(&dest->inactconns);
499 }
500
501 /*
502  *      Bind a connection entry with a virtual service destination
503  *      Called just after a new connection entry is created.
504  */
505 static inline void
506 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
507 {
508         unsigned int conn_flags;
509
510         /* if dest is NULL, then return directly */
511         if (!dest)
512                 return;
513
514         /* Increase the refcnt counter of the dest */
515         atomic_inc(&dest->refcnt);
516
517         conn_flags = atomic_read(&dest->conn_flags);
518         if (cp->protocol != IPPROTO_UDP)
519                 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
520         /* Bind with the destination and its corresponding transmitter */
521         if (cp->flags & IP_VS_CONN_F_SYNC) {
522                 /* if the connection is not template and is created
523                  * by sync, preserve the activity flag.
524                  */
525                 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE))
526                         conn_flags &= ~IP_VS_CONN_F_INACTIVE;
527         }
528         cp->flags |= conn_flags;
529         cp->dest = dest;
530
531         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
532                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
533                       "dest->refcnt:%d\n",
534                       ip_vs_proto_name(cp->protocol),
535                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
536                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
537                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
538                       ip_vs_fwd_tag(cp), cp->state,
539                       cp->flags, atomic_read(&cp->refcnt),
540                       atomic_read(&dest->refcnt));
541
542         /* Update the connection counters */
543         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
544                 /* It is a normal connection, so increase the inactive
545                    connection counter because it is in TCP SYNRECV
546                    state (inactive) or other protocol inacive state */
547                 if ((cp->flags & IP_VS_CONN_F_SYNC) &&
548                     (!(cp->flags & IP_VS_CONN_F_INACTIVE)))
549                         atomic_inc(&dest->activeconns);
550                 else
551                         atomic_inc(&dest->inactconns);
552         } else {
553                 /* It is a persistent connection/template, so increase
554                    the peristent connection counter */
555                 atomic_inc(&dest->persistconns);
556         }
557
558         if (dest->u_threshold != 0 &&
559             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
560                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
561 }
562
563
564 /*
565  * Check if there is a destination for the connection, if so
566  * bind the connection to the destination.
567  */
568 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
569 {
570         struct ip_vs_dest *dest;
571
572         if ((cp) && (!cp->dest)) {
573                 dest = ip_vs_find_dest(cp->af, &cp->daddr, cp->dport,
574                                        &cp->vaddr, cp->vport,
575                                        cp->protocol);
576                 ip_vs_bind_dest(cp, dest);
577                 return dest;
578         } else
579                 return NULL;
580 }
581
582
583 /*
584  *      Unbind a connection entry with its VS destination
585  *      Called by the ip_vs_conn_expire function.
586  */
587 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
588 {
589         struct ip_vs_dest *dest = cp->dest;
590
591         if (!dest)
592                 return;
593
594         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
595                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
596                       "dest->refcnt:%d\n",
597                       ip_vs_proto_name(cp->protocol),
598                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
599                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
600                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
601                       ip_vs_fwd_tag(cp), cp->state,
602                       cp->flags, atomic_read(&cp->refcnt),
603                       atomic_read(&dest->refcnt));
604
605         /* Update the connection counters */
606         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
607                 /* It is a normal connection, so decrease the inactconns
608                    or activeconns counter */
609                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
610                         atomic_dec(&dest->inactconns);
611                 } else {
612                         atomic_dec(&dest->activeconns);
613                 }
614         } else {
615                 /* It is a persistent connection/template, so decrease
616                    the peristent connection counter */
617                 atomic_dec(&dest->persistconns);
618         }
619
620         if (dest->l_threshold != 0) {
621                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
622                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
623         } else if (dest->u_threshold != 0) {
624                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
625                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
626         } else {
627                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
628                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
629         }
630
631         /*
632          * Simply decrease the refcnt of the dest, because the
633          * dest will be either in service's destination list
634          * or in the trash.
635          */
636         atomic_dec(&dest->refcnt);
637 }
638
639
640 /*
641  *      Checking if the destination of a connection template is available.
642  *      If available, return 1, otherwise invalidate this connection
643  *      template and return 0.
644  */
645 int ip_vs_check_template(struct ip_vs_conn *ct)
646 {
647         struct ip_vs_dest *dest = ct->dest;
648
649         /*
650          * Checking the dest server status.
651          */
652         if ((dest == NULL) ||
653             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
654             (sysctl_ip_vs_expire_quiescent_template &&
655              (atomic_read(&dest->weight) == 0))) {
656                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
657                               "protocol %s s:%s:%d v:%s:%d "
658                               "-> d:%s:%d\n",
659                               ip_vs_proto_name(ct->protocol),
660                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
661                               ntohs(ct->cport),
662                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
663                               ntohs(ct->vport),
664                               IP_VS_DBG_ADDR(ct->af, &ct->daddr),
665                               ntohs(ct->dport));
666
667                 /*
668                  * Invalidate the connection template
669                  */
670                 if (ct->vport != htons(0xffff)) {
671                         if (ip_vs_conn_unhash(ct)) {
672                                 ct->dport = htons(0xffff);
673                                 ct->vport = htons(0xffff);
674                                 ct->cport = 0;
675                                 ip_vs_conn_hash(ct);
676                         }
677                 }
678
679                 /*
680                  * Simply decrease the refcnt of the template,
681                  * don't restart its timer.
682                  */
683                 atomic_dec(&ct->refcnt);
684                 return 0;
685         }
686         return 1;
687 }
688
689 static void ip_vs_conn_expire(unsigned long data)
690 {
691         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
692
693         cp->timeout = 60*HZ;
694
695         /*
696          *      hey, I'm using it
697          */
698         atomic_inc(&cp->refcnt);
699
700         /*
701          *      do I control anybody?
702          */
703         if (atomic_read(&cp->n_control))
704                 goto expire_later;
705
706         /*
707          *      unhash it if it is hashed in the conn table
708          */
709         if (!ip_vs_conn_unhash(cp) && !(cp->flags & IP_VS_CONN_F_ONE_PACKET))
710                 goto expire_later;
711
712         /*
713          *      refcnt==1 implies I'm the only one referrer
714          */
715         if (likely(atomic_read(&cp->refcnt) == 1)) {
716                 /* delete the timer if it is activated by other users */
717                 if (timer_pending(&cp->timer))
718                         del_timer(&cp->timer);
719
720                 /* does anybody control me? */
721                 if (cp->control)
722                         ip_vs_control_del(cp);
723
724                 if (unlikely(cp->app != NULL))
725                         ip_vs_unbind_app(cp);
726                 ip_vs_unbind_dest(cp);
727                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
728                         atomic_dec(&ip_vs_conn_no_cport_cnt);
729                 atomic_dec(&ip_vs_conn_count);
730
731                 kmem_cache_free(ip_vs_conn_cachep, cp);
732                 return;
733         }
734
735         /* hash it back to the table */
736         ip_vs_conn_hash(cp);
737
738   expire_later:
739         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
740                   atomic_read(&cp->refcnt)-1,
741                   atomic_read(&cp->n_control));
742
743         ip_vs_conn_put(cp);
744 }
745
746
747 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
748 {
749         if (del_timer(&cp->timer))
750                 mod_timer(&cp->timer, jiffies);
751 }
752
753
754 /*
755  *      Create a new connection entry and hash it into the ip_vs_conn_tab
756  */
757 struct ip_vs_conn *
758 ip_vs_conn_new(int af, int proto, const union nf_inet_addr *caddr, __be16 cport,
759                const union nf_inet_addr *vaddr, __be16 vport,
760                const union nf_inet_addr *daddr, __be16 dport, unsigned flags,
761                struct ip_vs_dest *dest)
762 {
763         struct ip_vs_conn *cp;
764         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
765
766         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
767         if (cp == NULL) {
768                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
769                 return NULL;
770         }
771
772         INIT_LIST_HEAD(&cp->c_list);
773         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
774         cp->af             = af;
775         cp->protocol       = proto;
776         ip_vs_addr_copy(af, &cp->caddr, caddr);
777         cp->cport          = cport;
778         ip_vs_addr_copy(af, &cp->vaddr, vaddr);
779         cp->vport          = vport;
780         /* proto should only be IPPROTO_IP if d_addr is a fwmark */
781         ip_vs_addr_copy(proto == IPPROTO_IP ? AF_UNSPEC : af,
782                         &cp->daddr, daddr);
783         cp->dport          = dport;
784         cp->flags          = flags;
785         spin_lock_init(&cp->lock);
786
787         /*
788          * Set the entry is referenced by the current thread before hashing
789          * it in the table, so that other thread run ip_vs_random_dropentry
790          * but cannot drop this entry.
791          */
792         atomic_set(&cp->refcnt, 1);
793
794         atomic_set(&cp->n_control, 0);
795         atomic_set(&cp->in_pkts, 0);
796
797         atomic_inc(&ip_vs_conn_count);
798         if (flags & IP_VS_CONN_F_NO_CPORT)
799                 atomic_inc(&ip_vs_conn_no_cport_cnt);
800
801         /* Bind the connection with a destination server */
802         ip_vs_bind_dest(cp, dest);
803
804         /* Set its state and timeout */
805         cp->state = 0;
806         cp->timeout = 3*HZ;
807
808         /* Bind its packet transmitter */
809 #ifdef CONFIG_IP_VS_IPV6
810         if (af == AF_INET6)
811                 ip_vs_bind_xmit_v6(cp);
812         else
813 #endif
814                 ip_vs_bind_xmit(cp);
815
816         if (unlikely(pp && atomic_read(&pp->appcnt)))
817                 ip_vs_bind_app(cp, pp);
818
819         /* Hash it in the ip_vs_conn_tab finally */
820         ip_vs_conn_hash(cp);
821
822         return cp;
823 }
824
825
826 /*
827  *      /proc/net/ip_vs_conn entries
828  */
829 #ifdef CONFIG_PROC_FS
830
831 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
832 {
833         int idx;
834         struct ip_vs_conn *cp;
835
836         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
837                 ct_read_lock_bh(idx);
838                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
839                         if (pos-- == 0) {
840                                 seq->private = &ip_vs_conn_tab[idx];
841                                 return cp;
842                         }
843                 }
844                 ct_read_unlock_bh(idx);
845         }
846
847         return NULL;
848 }
849
850 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
851 {
852         seq->private = NULL;
853         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
854 }
855
856 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
857 {
858         struct ip_vs_conn *cp = v;
859         struct list_head *e, *l = seq->private;
860         int idx;
861
862         ++*pos;
863         if (v == SEQ_START_TOKEN)
864                 return ip_vs_conn_array(seq, 0);
865
866         /* more on same hash chain? */
867         if ((e = cp->c_list.next) != l)
868                 return list_entry(e, struct ip_vs_conn, c_list);
869
870         idx = l - ip_vs_conn_tab;
871         ct_read_unlock_bh(idx);
872
873         while (++idx < ip_vs_conn_tab_size) {
874                 ct_read_lock_bh(idx);
875                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
876                         seq->private = &ip_vs_conn_tab[idx];
877                         return cp;
878                 }
879                 ct_read_unlock_bh(idx);
880         }
881         seq->private = NULL;
882         return NULL;
883 }
884
885 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
886 {
887         struct list_head *l = seq->private;
888
889         if (l)
890                 ct_read_unlock_bh(l - ip_vs_conn_tab);
891 }
892
893 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
894 {
895
896         if (v == SEQ_START_TOKEN)
897                 seq_puts(seq,
898    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
899         else {
900                 const struct ip_vs_conn *cp = v;
901
902 #ifdef CONFIG_IP_VS_IPV6
903                 if (cp->af == AF_INET6)
904                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %7lu\n",
905                                 ip_vs_proto_name(cp->protocol),
906                                 &cp->caddr.in6, ntohs(cp->cport),
907                                 &cp->vaddr.in6, ntohs(cp->vport),
908                                 &cp->daddr.in6, ntohs(cp->dport),
909                                 ip_vs_state_name(cp->protocol, cp->state),
910                                 (cp->timer.expires-jiffies)/HZ);
911                 else
912 #endif
913                         seq_printf(seq,
914                                 "%-3s %08X %04X %08X %04X"
915                                 " %08X %04X %-11s %7lu\n",
916                                 ip_vs_proto_name(cp->protocol),
917                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
918                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
919                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
920                                 ip_vs_state_name(cp->protocol, cp->state),
921                                 (cp->timer.expires-jiffies)/HZ);
922         }
923         return 0;
924 }
925
926 static const struct seq_operations ip_vs_conn_seq_ops = {
927         .start = ip_vs_conn_seq_start,
928         .next  = ip_vs_conn_seq_next,
929         .stop  = ip_vs_conn_seq_stop,
930         .show  = ip_vs_conn_seq_show,
931 };
932
933 static int ip_vs_conn_open(struct inode *inode, struct file *file)
934 {
935         return seq_open(file, &ip_vs_conn_seq_ops);
936 }
937
938 static const struct file_operations ip_vs_conn_fops = {
939         .owner   = THIS_MODULE,
940         .open    = ip_vs_conn_open,
941         .read    = seq_read,
942         .llseek  = seq_lseek,
943         .release = seq_release,
944 };
945
946 static const char *ip_vs_origin_name(unsigned flags)
947 {
948         if (flags & IP_VS_CONN_F_SYNC)
949                 return "SYNC";
950         else
951                 return "LOCAL";
952 }
953
954 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
955 {
956
957         if (v == SEQ_START_TOKEN)
958                 seq_puts(seq,
959    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
960         else {
961                 const struct ip_vs_conn *cp = v;
962
963 #ifdef CONFIG_IP_VS_IPV6
964                 if (cp->af == AF_INET6)
965                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
966                                 ip_vs_proto_name(cp->protocol),
967                                 &cp->caddr.in6, ntohs(cp->cport),
968                                 &cp->vaddr.in6, ntohs(cp->vport),
969                                 &cp->daddr.in6, ntohs(cp->dport),
970                                 ip_vs_state_name(cp->protocol, cp->state),
971                                 ip_vs_origin_name(cp->flags),
972                                 (cp->timer.expires-jiffies)/HZ);
973                 else
974 #endif
975                         seq_printf(seq,
976                                 "%-3s %08X %04X %08X %04X "
977                                 "%08X %04X %-11s %-6s %7lu\n",
978                                 ip_vs_proto_name(cp->protocol),
979                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
980                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
981                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
982                                 ip_vs_state_name(cp->protocol, cp->state),
983                                 ip_vs_origin_name(cp->flags),
984                                 (cp->timer.expires-jiffies)/HZ);
985         }
986         return 0;
987 }
988
989 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
990         .start = ip_vs_conn_seq_start,
991         .next  = ip_vs_conn_seq_next,
992         .stop  = ip_vs_conn_seq_stop,
993         .show  = ip_vs_conn_sync_seq_show,
994 };
995
996 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
997 {
998         return seq_open(file, &ip_vs_conn_sync_seq_ops);
999 }
1000
1001 static const struct file_operations ip_vs_conn_sync_fops = {
1002         .owner   = THIS_MODULE,
1003         .open    = ip_vs_conn_sync_open,
1004         .read    = seq_read,
1005         .llseek  = seq_lseek,
1006         .release = seq_release,
1007 };
1008
1009 #endif
1010
1011
1012 /*
1013  *      Randomly drop connection entries before running out of memory
1014  */
1015 static inline int todrop_entry(struct ip_vs_conn *cp)
1016 {
1017         /*
1018          * The drop rate array needs tuning for real environments.
1019          * Called from timer bh only => no locking
1020          */
1021         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1022         static char todrop_counter[9] = {0};
1023         int i;
1024
1025         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1026            This will leave enough time for normal connection to get
1027            through. */
1028         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1029                 return 0;
1030
1031         /* Don't drop the entry if its number of incoming packets is not
1032            located in [0, 8] */
1033         i = atomic_read(&cp->in_pkts);
1034         if (i > 8 || i < 0) return 0;
1035
1036         if (!todrop_rate[i]) return 0;
1037         if (--todrop_counter[i] > 0) return 0;
1038
1039         todrop_counter[i] = todrop_rate[i];
1040         return 1;
1041 }
1042
1043 /* Called from keventd and must protect itself from softirqs */
1044 void ip_vs_random_dropentry(void)
1045 {
1046         int idx;
1047         struct ip_vs_conn *cp;
1048
1049         /*
1050          * Randomly scan 1/32 of the whole table every second
1051          */
1052         for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1053                 unsigned hash = net_random() & ip_vs_conn_tab_mask;
1054
1055                 /*
1056                  *  Lock is actually needed in this loop.
1057                  */
1058                 ct_write_lock_bh(hash);
1059
1060                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
1061                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1062                                 /* connection template */
1063                                 continue;
1064
1065                         if (cp->protocol == IPPROTO_TCP) {
1066                                 switch(cp->state) {
1067                                 case IP_VS_TCP_S_SYN_RECV:
1068                                 case IP_VS_TCP_S_SYNACK:
1069                                         break;
1070
1071                                 case IP_VS_TCP_S_ESTABLISHED:
1072                                         if (todrop_entry(cp))
1073                                                 break;
1074                                         continue;
1075
1076                                 default:
1077                                         continue;
1078                                 }
1079                         } else {
1080                                 if (!todrop_entry(cp))
1081                                         continue;
1082                         }
1083
1084                         IP_VS_DBG(4, "del connection\n");
1085                         ip_vs_conn_expire_now(cp);
1086                         if (cp->control) {
1087                                 IP_VS_DBG(4, "del conn template\n");
1088                                 ip_vs_conn_expire_now(cp->control);
1089                         }
1090                 }
1091                 ct_write_unlock_bh(hash);
1092         }
1093 }
1094
1095
1096 /*
1097  *      Flush all the connection entries in the ip_vs_conn_tab
1098  */
1099 static void ip_vs_conn_flush(void)
1100 {
1101         int idx;
1102         struct ip_vs_conn *cp;
1103
1104   flush_again:
1105         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1106                 /*
1107                  *  Lock is actually needed in this loop.
1108                  */
1109                 ct_write_lock_bh(idx);
1110
1111                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
1112
1113                         IP_VS_DBG(4, "del connection\n");
1114                         ip_vs_conn_expire_now(cp);
1115                         if (cp->control) {
1116                                 IP_VS_DBG(4, "del conn template\n");
1117                                 ip_vs_conn_expire_now(cp->control);
1118                         }
1119                 }
1120                 ct_write_unlock_bh(idx);
1121         }
1122
1123         /* the counter may be not NULL, because maybe some conn entries
1124            are run by slow timer handler or unhashed but still referred */
1125         if (atomic_read(&ip_vs_conn_count) != 0) {
1126                 schedule();
1127                 goto flush_again;
1128         }
1129 }
1130
1131
1132 int __init ip_vs_conn_init(void)
1133 {
1134         int idx;
1135
1136         /* Compute size and mask */
1137         ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1138         ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1139
1140         /*
1141          * Allocate the connection hash table and initialize its list heads
1142          */
1143         ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size *
1144                                  sizeof(struct list_head));
1145         if (!ip_vs_conn_tab)
1146                 return -ENOMEM;
1147
1148         /* Allocate ip_vs_conn slab cache */
1149         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1150                                               sizeof(struct ip_vs_conn), 0,
1151                                               SLAB_HWCACHE_ALIGN, NULL);
1152         if (!ip_vs_conn_cachep) {
1153                 vfree(ip_vs_conn_tab);
1154                 return -ENOMEM;
1155         }
1156
1157         pr_info("Connection hash table configured "
1158                 "(size=%d, memory=%ldKbytes)\n",
1159                 ip_vs_conn_tab_size,
1160                 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1161         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1162                   sizeof(struct ip_vs_conn));
1163
1164         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1165                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
1166         }
1167
1168         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1169                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1170         }
1171
1172         proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops);
1173         proc_net_fops_create(&init_net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops);
1174
1175         /* calculate the random value for connection hash */
1176         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1177
1178         return 0;
1179 }
1180
1181
1182 void ip_vs_conn_cleanup(void)
1183 {
1184         /* flush all the connection entries first */
1185         ip_vs_conn_flush();
1186
1187         /* Release the empty cache */
1188         kmem_cache_destroy(ip_vs_conn_cachep);
1189         proc_net_remove(&init_net, "ip_vs_conn");
1190         proc_net_remove(&init_net, "ip_vs_conn_sync");
1191         vfree(ip_vs_conn_tab);
1192 }