Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/kaber/nf-next-2.6
[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 /* Get reference to connection template */
275 struct ip_vs_conn *ip_vs_ct_in_get
276 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
277  const union nf_inet_addr *d_addr, __be16 d_port)
278 {
279         unsigned hash;
280         struct ip_vs_conn *cp;
281
282         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
283
284         ct_read_lock(hash);
285
286         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
287                 if (cp->af == af &&
288                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
289                     /* protocol should only be IPPROTO_IP if
290                      * d_addr is a fwmark */
291                     ip_vs_addr_equal(protocol == IPPROTO_IP ? AF_UNSPEC : af,
292                                      d_addr, &cp->vaddr) &&
293                     s_port == cp->cport && d_port == cp->vport &&
294                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
295                     protocol == cp->protocol) {
296                         /* HIT */
297                         atomic_inc(&cp->refcnt);
298                         goto out;
299                 }
300         }
301         cp = NULL;
302
303   out:
304         ct_read_unlock(hash);
305
306         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
307                       ip_vs_proto_name(protocol),
308                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
309                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
310                       cp ? "hit" : "not hit");
311
312         return cp;
313 }
314
315 /*
316  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
317  *  Called for pkts coming from inside-to-OUTside.
318  *      s_addr, s_port: pkt source address (inside host)
319  *      d_addr, d_port: pkt dest address (foreign host)
320  */
321 struct ip_vs_conn *ip_vs_conn_out_get
322 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
323  const union nf_inet_addr *d_addr, __be16 d_port)
324 {
325         unsigned hash;
326         struct ip_vs_conn *cp, *ret=NULL;
327
328         /*
329          *      Check for "full" addressed entries
330          */
331         hash = ip_vs_conn_hashkey(af, protocol, d_addr, d_port);
332
333         ct_read_lock(hash);
334
335         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
336                 if (cp->af == af &&
337                     ip_vs_addr_equal(af, d_addr, &cp->caddr) &&
338                     ip_vs_addr_equal(af, s_addr, &cp->daddr) &&
339                     d_port == cp->cport && s_port == cp->dport &&
340                     protocol == cp->protocol) {
341                         /* HIT */
342                         atomic_inc(&cp->refcnt);
343                         ret = cp;
344                         break;
345                 }
346         }
347
348         ct_read_unlock(hash);
349
350         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
351                       ip_vs_proto_name(protocol),
352                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
353                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
354                       ret ? "hit" : "not hit");
355
356         return ret;
357 }
358
359
360 /*
361  *      Put back the conn and restart its timer with its timeout
362  */
363 void ip_vs_conn_put(struct ip_vs_conn *cp)
364 {
365         unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
366                 0 : cp->timeout;
367         mod_timer(&cp->timer, jiffies+t);
368
369         __ip_vs_conn_put(cp);
370 }
371
372
373 /*
374  *      Fill a no_client_port connection with a client port number
375  */
376 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
377 {
378         if (ip_vs_conn_unhash(cp)) {
379                 spin_lock(&cp->lock);
380                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
381                         atomic_dec(&ip_vs_conn_no_cport_cnt);
382                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
383                         cp->cport = cport;
384                 }
385                 spin_unlock(&cp->lock);
386
387                 /* hash on new dport */
388                 ip_vs_conn_hash(cp);
389         }
390 }
391
392
393 /*
394  *      Bind a connection entry with the corresponding packet_xmit.
395  *      Called by ip_vs_conn_new.
396  */
397 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
398 {
399         switch (IP_VS_FWD_METHOD(cp)) {
400         case IP_VS_CONN_F_MASQ:
401                 cp->packet_xmit = ip_vs_nat_xmit;
402                 break;
403
404         case IP_VS_CONN_F_TUNNEL:
405                 cp->packet_xmit = ip_vs_tunnel_xmit;
406                 break;
407
408         case IP_VS_CONN_F_DROUTE:
409                 cp->packet_xmit = ip_vs_dr_xmit;
410                 break;
411
412         case IP_VS_CONN_F_LOCALNODE:
413                 cp->packet_xmit = ip_vs_null_xmit;
414                 break;
415
416         case IP_VS_CONN_F_BYPASS:
417                 cp->packet_xmit = ip_vs_bypass_xmit;
418                 break;
419         }
420 }
421
422 #ifdef CONFIG_IP_VS_IPV6
423 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
424 {
425         switch (IP_VS_FWD_METHOD(cp)) {
426         case IP_VS_CONN_F_MASQ:
427                 cp->packet_xmit = ip_vs_nat_xmit_v6;
428                 break;
429
430         case IP_VS_CONN_F_TUNNEL:
431                 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
432                 break;
433
434         case IP_VS_CONN_F_DROUTE:
435                 cp->packet_xmit = ip_vs_dr_xmit_v6;
436                 break;
437
438         case IP_VS_CONN_F_LOCALNODE:
439                 cp->packet_xmit = ip_vs_null_xmit;
440                 break;
441
442         case IP_VS_CONN_F_BYPASS:
443                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
444                 break;
445         }
446 }
447 #endif
448
449
450 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
451 {
452         return atomic_read(&dest->activeconns)
453                 + atomic_read(&dest->inactconns);
454 }
455
456 /*
457  *      Bind a connection entry with a virtual service destination
458  *      Called just after a new connection entry is created.
459  */
460 static inline void
461 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
462 {
463         /* if dest is NULL, then return directly */
464         if (!dest)
465                 return;
466
467         /* Increase the refcnt counter of the dest */
468         atomic_inc(&dest->refcnt);
469
470         /* Bind with the destination and its corresponding transmitter */
471         if ((cp->flags & IP_VS_CONN_F_SYNC) &&
472             (!(cp->flags & IP_VS_CONN_F_TEMPLATE)))
473                 /* if the connection is not template and is created
474                  * by sync, preserve the activity flag.
475                  */
476                 cp->flags |= atomic_read(&dest->conn_flags) &
477                              (~IP_VS_CONN_F_INACTIVE);
478         else
479                 cp->flags |= atomic_read(&dest->conn_flags);
480         cp->dest = dest;
481
482         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
483                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
484                       "dest->refcnt:%d\n",
485                       ip_vs_proto_name(cp->protocol),
486                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
487                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
488                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
489                       ip_vs_fwd_tag(cp), cp->state,
490                       cp->flags, atomic_read(&cp->refcnt),
491                       atomic_read(&dest->refcnt));
492
493         /* Update the connection counters */
494         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
495                 /* It is a normal connection, so increase the inactive
496                    connection counter because it is in TCP SYNRECV
497                    state (inactive) or other protocol inacive state */
498                 if ((cp->flags & IP_VS_CONN_F_SYNC) &&
499                     (!(cp->flags & IP_VS_CONN_F_INACTIVE)))
500                         atomic_inc(&dest->activeconns);
501                 else
502                         atomic_inc(&dest->inactconns);
503         } else {
504                 /* It is a persistent connection/template, so increase
505                    the peristent connection counter */
506                 atomic_inc(&dest->persistconns);
507         }
508
509         if (dest->u_threshold != 0 &&
510             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
511                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
512 }
513
514
515 /*
516  * Check if there is a destination for the connection, if so
517  * bind the connection to the destination.
518  */
519 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
520 {
521         struct ip_vs_dest *dest;
522
523         if ((cp) && (!cp->dest)) {
524                 dest = ip_vs_find_dest(cp->af, &cp->daddr, cp->dport,
525                                        &cp->vaddr, cp->vport,
526                                        cp->protocol);
527                 ip_vs_bind_dest(cp, dest);
528                 return dest;
529         } else
530                 return NULL;
531 }
532
533
534 /*
535  *      Unbind a connection entry with its VS destination
536  *      Called by the ip_vs_conn_expire function.
537  */
538 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
539 {
540         struct ip_vs_dest *dest = cp->dest;
541
542         if (!dest)
543                 return;
544
545         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
546                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
547                       "dest->refcnt:%d\n",
548                       ip_vs_proto_name(cp->protocol),
549                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
550                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
551                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
552                       ip_vs_fwd_tag(cp), cp->state,
553                       cp->flags, atomic_read(&cp->refcnt),
554                       atomic_read(&dest->refcnt));
555
556         /* Update the connection counters */
557         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
558                 /* It is a normal connection, so decrease the inactconns
559                    or activeconns counter */
560                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
561                         atomic_dec(&dest->inactconns);
562                 } else {
563                         atomic_dec(&dest->activeconns);
564                 }
565         } else {
566                 /* It is a persistent connection/template, so decrease
567                    the peristent connection counter */
568                 atomic_dec(&dest->persistconns);
569         }
570
571         if (dest->l_threshold != 0) {
572                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
573                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
574         } else if (dest->u_threshold != 0) {
575                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
576                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
577         } else {
578                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
579                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
580         }
581
582         /*
583          * Simply decrease the refcnt of the dest, because the
584          * dest will be either in service's destination list
585          * or in the trash.
586          */
587         atomic_dec(&dest->refcnt);
588 }
589
590
591 /*
592  *      Checking if the destination of a connection template is available.
593  *      If available, return 1, otherwise invalidate this connection
594  *      template and return 0.
595  */
596 int ip_vs_check_template(struct ip_vs_conn *ct)
597 {
598         struct ip_vs_dest *dest = ct->dest;
599
600         /*
601          * Checking the dest server status.
602          */
603         if ((dest == NULL) ||
604             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
605             (sysctl_ip_vs_expire_quiescent_template &&
606              (atomic_read(&dest->weight) == 0))) {
607                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
608                               "protocol %s s:%s:%d v:%s:%d "
609                               "-> d:%s:%d\n",
610                               ip_vs_proto_name(ct->protocol),
611                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
612                               ntohs(ct->cport),
613                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
614                               ntohs(ct->vport),
615                               IP_VS_DBG_ADDR(ct->af, &ct->daddr),
616                               ntohs(ct->dport));
617
618                 /*
619                  * Invalidate the connection template
620                  */
621                 if (ct->vport != htons(0xffff)) {
622                         if (ip_vs_conn_unhash(ct)) {
623                                 ct->dport = htons(0xffff);
624                                 ct->vport = htons(0xffff);
625                                 ct->cport = 0;
626                                 ip_vs_conn_hash(ct);
627                         }
628                 }
629
630                 /*
631                  * Simply decrease the refcnt of the template,
632                  * don't restart its timer.
633                  */
634                 atomic_dec(&ct->refcnt);
635                 return 0;
636         }
637         return 1;
638 }
639
640 static void ip_vs_conn_expire(unsigned long data)
641 {
642         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
643
644         cp->timeout = 60*HZ;
645
646         /*
647          *      hey, I'm using it
648          */
649         atomic_inc(&cp->refcnt);
650
651         /*
652          *      do I control anybody?
653          */
654         if (atomic_read(&cp->n_control))
655                 goto expire_later;
656
657         /*
658          *      unhash it if it is hashed in the conn table
659          */
660         if (!ip_vs_conn_unhash(cp) && !(cp->flags & IP_VS_CONN_F_ONE_PACKET))
661                 goto expire_later;
662
663         /*
664          *      refcnt==1 implies I'm the only one referrer
665          */
666         if (likely(atomic_read(&cp->refcnt) == 1)) {
667                 /* delete the timer if it is activated by other users */
668                 if (timer_pending(&cp->timer))
669                         del_timer(&cp->timer);
670
671                 /* does anybody control me? */
672                 if (cp->control)
673                         ip_vs_control_del(cp);
674
675                 if (unlikely(cp->app != NULL))
676                         ip_vs_unbind_app(cp);
677                 ip_vs_unbind_dest(cp);
678                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
679                         atomic_dec(&ip_vs_conn_no_cport_cnt);
680                 atomic_dec(&ip_vs_conn_count);
681
682                 kmem_cache_free(ip_vs_conn_cachep, cp);
683                 return;
684         }
685
686         /* hash it back to the table */
687         ip_vs_conn_hash(cp);
688
689   expire_later:
690         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
691                   atomic_read(&cp->refcnt)-1,
692                   atomic_read(&cp->n_control));
693
694         ip_vs_conn_put(cp);
695 }
696
697
698 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
699 {
700         if (del_timer(&cp->timer))
701                 mod_timer(&cp->timer, jiffies);
702 }
703
704
705 /*
706  *      Create a new connection entry and hash it into the ip_vs_conn_tab
707  */
708 struct ip_vs_conn *
709 ip_vs_conn_new(int af, int proto, const union nf_inet_addr *caddr, __be16 cport,
710                const union nf_inet_addr *vaddr, __be16 vport,
711                const union nf_inet_addr *daddr, __be16 dport, unsigned flags,
712                struct ip_vs_dest *dest)
713 {
714         struct ip_vs_conn *cp;
715         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
716
717         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
718         if (cp == NULL) {
719                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
720                 return NULL;
721         }
722
723         INIT_LIST_HEAD(&cp->c_list);
724         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
725         cp->af             = af;
726         cp->protocol       = proto;
727         ip_vs_addr_copy(af, &cp->caddr, caddr);
728         cp->cport          = cport;
729         ip_vs_addr_copy(af, &cp->vaddr, vaddr);
730         cp->vport          = vport;
731         /* proto should only be IPPROTO_IP if d_addr is a fwmark */
732         ip_vs_addr_copy(proto == IPPROTO_IP ? AF_UNSPEC : af,
733                         &cp->daddr, daddr);
734         cp->dport          = dport;
735         cp->flags          = flags;
736         spin_lock_init(&cp->lock);
737
738         /*
739          * Set the entry is referenced by the current thread before hashing
740          * it in the table, so that other thread run ip_vs_random_dropentry
741          * but cannot drop this entry.
742          */
743         atomic_set(&cp->refcnt, 1);
744
745         atomic_set(&cp->n_control, 0);
746         atomic_set(&cp->in_pkts, 0);
747
748         atomic_inc(&ip_vs_conn_count);
749         if (flags & IP_VS_CONN_F_NO_CPORT)
750                 atomic_inc(&ip_vs_conn_no_cport_cnt);
751
752         /* Bind the connection with a destination server */
753         ip_vs_bind_dest(cp, dest);
754
755         /* Set its state and timeout */
756         cp->state = 0;
757         cp->timeout = 3*HZ;
758
759         /* Bind its packet transmitter */
760 #ifdef CONFIG_IP_VS_IPV6
761         if (af == AF_INET6)
762                 ip_vs_bind_xmit_v6(cp);
763         else
764 #endif
765                 ip_vs_bind_xmit(cp);
766
767         if (unlikely(pp && atomic_read(&pp->appcnt)))
768                 ip_vs_bind_app(cp, pp);
769
770         /* Hash it in the ip_vs_conn_tab finally */
771         ip_vs_conn_hash(cp);
772
773         return cp;
774 }
775
776
777 /*
778  *      /proc/net/ip_vs_conn entries
779  */
780 #ifdef CONFIG_PROC_FS
781
782 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
783 {
784         int idx;
785         struct ip_vs_conn *cp;
786
787         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
788                 ct_read_lock_bh(idx);
789                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
790                         if (pos-- == 0) {
791                                 seq->private = &ip_vs_conn_tab[idx];
792                                 return cp;
793                         }
794                 }
795                 ct_read_unlock_bh(idx);
796         }
797
798         return NULL;
799 }
800
801 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
802 {
803         seq->private = NULL;
804         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
805 }
806
807 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
808 {
809         struct ip_vs_conn *cp = v;
810         struct list_head *e, *l = seq->private;
811         int idx;
812
813         ++*pos;
814         if (v == SEQ_START_TOKEN)
815                 return ip_vs_conn_array(seq, 0);
816
817         /* more on same hash chain? */
818         if ((e = cp->c_list.next) != l)
819                 return list_entry(e, struct ip_vs_conn, c_list);
820
821         idx = l - ip_vs_conn_tab;
822         ct_read_unlock_bh(idx);
823
824         while (++idx < ip_vs_conn_tab_size) {
825                 ct_read_lock_bh(idx);
826                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
827                         seq->private = &ip_vs_conn_tab[idx];
828                         return cp;
829                 }
830                 ct_read_unlock_bh(idx);
831         }
832         seq->private = NULL;
833         return NULL;
834 }
835
836 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
837 {
838         struct list_head *l = seq->private;
839
840         if (l)
841                 ct_read_unlock_bh(l - ip_vs_conn_tab);
842 }
843
844 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
845 {
846
847         if (v == SEQ_START_TOKEN)
848                 seq_puts(seq,
849    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
850         else {
851                 const struct ip_vs_conn *cp = v;
852
853 #ifdef CONFIG_IP_VS_IPV6
854                 if (cp->af == AF_INET6)
855                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %7lu\n",
856                                 ip_vs_proto_name(cp->protocol),
857                                 &cp->caddr.in6, ntohs(cp->cport),
858                                 &cp->vaddr.in6, ntohs(cp->vport),
859                                 &cp->daddr.in6, ntohs(cp->dport),
860                                 ip_vs_state_name(cp->protocol, cp->state),
861                                 (cp->timer.expires-jiffies)/HZ);
862                 else
863 #endif
864                         seq_printf(seq,
865                                 "%-3s %08X %04X %08X %04X"
866                                 " %08X %04X %-11s %7lu\n",
867                                 ip_vs_proto_name(cp->protocol),
868                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
869                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
870                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
871                                 ip_vs_state_name(cp->protocol, cp->state),
872                                 (cp->timer.expires-jiffies)/HZ);
873         }
874         return 0;
875 }
876
877 static const struct seq_operations ip_vs_conn_seq_ops = {
878         .start = ip_vs_conn_seq_start,
879         .next  = ip_vs_conn_seq_next,
880         .stop  = ip_vs_conn_seq_stop,
881         .show  = ip_vs_conn_seq_show,
882 };
883
884 static int ip_vs_conn_open(struct inode *inode, struct file *file)
885 {
886         return seq_open(file, &ip_vs_conn_seq_ops);
887 }
888
889 static const struct file_operations ip_vs_conn_fops = {
890         .owner   = THIS_MODULE,
891         .open    = ip_vs_conn_open,
892         .read    = seq_read,
893         .llseek  = seq_lseek,
894         .release = seq_release,
895 };
896
897 static const char *ip_vs_origin_name(unsigned flags)
898 {
899         if (flags & IP_VS_CONN_F_SYNC)
900                 return "SYNC";
901         else
902                 return "LOCAL";
903 }
904
905 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
906 {
907
908         if (v == SEQ_START_TOKEN)
909                 seq_puts(seq,
910    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
911         else {
912                 const struct ip_vs_conn *cp = v;
913
914 #ifdef CONFIG_IP_VS_IPV6
915                 if (cp->af == AF_INET6)
916                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
917                                 ip_vs_proto_name(cp->protocol),
918                                 &cp->caddr.in6, ntohs(cp->cport),
919                                 &cp->vaddr.in6, ntohs(cp->vport),
920                                 &cp->daddr.in6, ntohs(cp->dport),
921                                 ip_vs_state_name(cp->protocol, cp->state),
922                                 ip_vs_origin_name(cp->flags),
923                                 (cp->timer.expires-jiffies)/HZ);
924                 else
925 #endif
926                         seq_printf(seq,
927                                 "%-3s %08X %04X %08X %04X "
928                                 "%08X %04X %-11s %-6s %7lu\n",
929                                 ip_vs_proto_name(cp->protocol),
930                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
931                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
932                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
933                                 ip_vs_state_name(cp->protocol, cp->state),
934                                 ip_vs_origin_name(cp->flags),
935                                 (cp->timer.expires-jiffies)/HZ);
936         }
937         return 0;
938 }
939
940 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
941         .start = ip_vs_conn_seq_start,
942         .next  = ip_vs_conn_seq_next,
943         .stop  = ip_vs_conn_seq_stop,
944         .show  = ip_vs_conn_sync_seq_show,
945 };
946
947 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
948 {
949         return seq_open(file, &ip_vs_conn_sync_seq_ops);
950 }
951
952 static const struct file_operations ip_vs_conn_sync_fops = {
953         .owner   = THIS_MODULE,
954         .open    = ip_vs_conn_sync_open,
955         .read    = seq_read,
956         .llseek  = seq_lseek,
957         .release = seq_release,
958 };
959
960 #endif
961
962
963 /*
964  *      Randomly drop connection entries before running out of memory
965  */
966 static inline int todrop_entry(struct ip_vs_conn *cp)
967 {
968         /*
969          * The drop rate array needs tuning for real environments.
970          * Called from timer bh only => no locking
971          */
972         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
973         static char todrop_counter[9] = {0};
974         int i;
975
976         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
977            This will leave enough time for normal connection to get
978            through. */
979         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
980                 return 0;
981
982         /* Don't drop the entry if its number of incoming packets is not
983            located in [0, 8] */
984         i = atomic_read(&cp->in_pkts);
985         if (i > 8 || i < 0) return 0;
986
987         if (!todrop_rate[i]) return 0;
988         if (--todrop_counter[i] > 0) return 0;
989
990         todrop_counter[i] = todrop_rate[i];
991         return 1;
992 }
993
994 /* Called from keventd and must protect itself from softirqs */
995 void ip_vs_random_dropentry(void)
996 {
997         int idx;
998         struct ip_vs_conn *cp;
999
1000         /*
1001          * Randomly scan 1/32 of the whole table every second
1002          */
1003         for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1004                 unsigned hash = net_random() & ip_vs_conn_tab_mask;
1005
1006                 /*
1007                  *  Lock is actually needed in this loop.
1008                  */
1009                 ct_write_lock_bh(hash);
1010
1011                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
1012                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1013                                 /* connection template */
1014                                 continue;
1015
1016                         if (cp->protocol == IPPROTO_TCP) {
1017                                 switch(cp->state) {
1018                                 case IP_VS_TCP_S_SYN_RECV:
1019                                 case IP_VS_TCP_S_SYNACK:
1020                                         break;
1021
1022                                 case IP_VS_TCP_S_ESTABLISHED:
1023                                         if (todrop_entry(cp))
1024                                                 break;
1025                                         continue;
1026
1027                                 default:
1028                                         continue;
1029                                 }
1030                         } else {
1031                                 if (!todrop_entry(cp))
1032                                         continue;
1033                         }
1034
1035                         IP_VS_DBG(4, "del connection\n");
1036                         ip_vs_conn_expire_now(cp);
1037                         if (cp->control) {
1038                                 IP_VS_DBG(4, "del conn template\n");
1039                                 ip_vs_conn_expire_now(cp->control);
1040                         }
1041                 }
1042                 ct_write_unlock_bh(hash);
1043         }
1044 }
1045
1046
1047 /*
1048  *      Flush all the connection entries in the ip_vs_conn_tab
1049  */
1050 static void ip_vs_conn_flush(void)
1051 {
1052         int idx;
1053         struct ip_vs_conn *cp;
1054
1055   flush_again:
1056         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1057                 /*
1058                  *  Lock is actually needed in this loop.
1059                  */
1060                 ct_write_lock_bh(idx);
1061
1062                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
1063
1064                         IP_VS_DBG(4, "del connection\n");
1065                         ip_vs_conn_expire_now(cp);
1066                         if (cp->control) {
1067                                 IP_VS_DBG(4, "del conn template\n");
1068                                 ip_vs_conn_expire_now(cp->control);
1069                         }
1070                 }
1071                 ct_write_unlock_bh(idx);
1072         }
1073
1074         /* the counter may be not NULL, because maybe some conn entries
1075            are run by slow timer handler or unhashed but still referred */
1076         if (atomic_read(&ip_vs_conn_count) != 0) {
1077                 schedule();
1078                 goto flush_again;
1079         }
1080 }
1081
1082
1083 int __init ip_vs_conn_init(void)
1084 {
1085         int idx;
1086
1087         /* Compute size and mask */
1088         ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1089         ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1090
1091         /*
1092          * Allocate the connection hash table and initialize its list heads
1093          */
1094         ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size *
1095                                  sizeof(struct list_head));
1096         if (!ip_vs_conn_tab)
1097                 return -ENOMEM;
1098
1099         /* Allocate ip_vs_conn slab cache */
1100         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1101                                               sizeof(struct ip_vs_conn), 0,
1102                                               SLAB_HWCACHE_ALIGN, NULL);
1103         if (!ip_vs_conn_cachep) {
1104                 vfree(ip_vs_conn_tab);
1105                 return -ENOMEM;
1106         }
1107
1108         pr_info("Connection hash table configured "
1109                 "(size=%d, memory=%ldKbytes)\n",
1110                 ip_vs_conn_tab_size,
1111                 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1112         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1113                   sizeof(struct ip_vs_conn));
1114
1115         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1116                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
1117         }
1118
1119         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1120                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1121         }
1122
1123         proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops);
1124         proc_net_fops_create(&init_net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops);
1125
1126         /* calculate the random value for connection hash */
1127         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1128
1129         return 0;
1130 }
1131
1132
1133 void ip_vs_conn_cleanup(void)
1134 {
1135         /* flush all the connection entries first */
1136         ip_vs_conn_flush();
1137
1138         /* Release the empty cache */
1139         kmem_cache_destroy(ip_vs_conn_cachep);
1140         proc_net_remove(&init_net, "ip_vs_conn");
1141         proc_net_remove(&init_net, "ip_vs_conn_sync");
1142         vfree(ip_vs_conn_tab);
1143 }