0a9f3c37e18d8caa939f22d40b7b66dc72df3753
[linux-2.6.git] / net / ipv4 / 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  * Version:     $Id: ip_vs_conn.c,v 1.31 2003/04/18 09:03:16 wensong Exp $
9  *
10  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
11  *              Peter Kese <peter.kese@ijs.si>
12  *              Julian Anastasov <ja@ssi.bg>
13  *
14  *              This program is free software; you can redistribute it and/or
15  *              modify it under the terms of the GNU General Public License
16  *              as published by the Free Software Foundation; either version
17  *              2 of the License, or (at your option) any later version.
18  *
19  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
20  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
21  * and others. Many code here is taken from IP MASQ code of kernel 2.2.
22  *
23  * Changes:
24  *
25  */
26
27 #include <linux/interrupt.h>
28 #include <linux/in.h>
29 #include <linux/net.h>
30 #include <linux/kernel.h>
31 #include <linux/module.h>
32 #include <linux/vmalloc.h>
33 #include <linux/proc_fs.h>              /* for proc_net_* */
34 #include <linux/seq_file.h>
35 #include <linux/jhash.h>
36 #include <linux/random.h>
37
38 #include <net/net_namespace.h>
39 #include <net/ip_vs.h>
40
41
42 /*
43  *  Connection hash table: for input and output packets lookups of IPVS
44  */
45 static struct list_head *ip_vs_conn_tab;
46
47 /*  SLAB cache for IPVS connections */
48 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
49
50 /*  counter for current IPVS connections */
51 static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
52
53 /*  counter for no client port connections */
54 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
55
56 /* random value for IPVS connection hash */
57 static unsigned int ip_vs_conn_rnd;
58
59 /*
60  *  Fine locking granularity for big connection hash table
61  */
62 #define CT_LOCKARRAY_BITS  4
63 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
64 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
65
66 struct ip_vs_aligned_lock
67 {
68         rwlock_t        l;
69 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
70
71 /* lock array for conn table */
72 static struct ip_vs_aligned_lock
73 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
74
75 static inline void ct_read_lock(unsigned key)
76 {
77         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
78 }
79
80 static inline void ct_read_unlock(unsigned key)
81 {
82         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
83 }
84
85 static inline void ct_write_lock(unsigned key)
86 {
87         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
88 }
89
90 static inline void ct_write_unlock(unsigned key)
91 {
92         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
93 }
94
95 static inline void ct_read_lock_bh(unsigned key)
96 {
97         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
98 }
99
100 static inline void ct_read_unlock_bh(unsigned key)
101 {
102         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
103 }
104
105 static inline void ct_write_lock_bh(unsigned key)
106 {
107         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
108 }
109
110 static inline void ct_write_unlock_bh(unsigned key)
111 {
112         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
113 }
114
115
116 /*
117  *      Returns hash value for IPVS connection entry
118  */
119 static unsigned int ip_vs_conn_hashkey(unsigned proto, __be32 addr, __be16 port)
120 {
121         return jhash_3words((__force u32)addr, (__force u32)port, proto, ip_vs_conn_rnd)
122                 & IP_VS_CONN_TAB_MASK;
123 }
124
125
126 /*
127  *      Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
128  *      returns bool success.
129  */
130 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
131 {
132         unsigned hash;
133         int ret;
134
135         /* Hash by protocol, client address and port */
136         hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport);
137
138         ct_write_lock(hash);
139
140         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
141                 list_add(&cp->c_list, &ip_vs_conn_tab[hash]);
142                 cp->flags |= IP_VS_CONN_F_HASHED;
143                 atomic_inc(&cp->refcnt);
144                 ret = 1;
145         } else {
146                 IP_VS_ERR("ip_vs_conn_hash(): request for already hashed, "
147                           "called from %p\n", __builtin_return_address(0));
148                 ret = 0;
149         }
150
151         ct_write_unlock(hash);
152
153         return ret;
154 }
155
156
157 /*
158  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
159  *      returns bool success.
160  */
161 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
162 {
163         unsigned hash;
164         int ret;
165
166         /* unhash it and decrease its reference counter */
167         hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport);
168
169         ct_write_lock(hash);
170
171         if (cp->flags & IP_VS_CONN_F_HASHED) {
172                 list_del(&cp->c_list);
173                 cp->flags &= ~IP_VS_CONN_F_HASHED;
174                 atomic_dec(&cp->refcnt);
175                 ret = 1;
176         } else
177                 ret = 0;
178
179         ct_write_unlock(hash);
180
181         return ret;
182 }
183
184
185 /*
186  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
187  *  Called for pkts coming from OUTside-to-INside.
188  *      s_addr, s_port: pkt source address (foreign host)
189  *      d_addr, d_port: pkt dest address (load balancer)
190  */
191 static inline struct ip_vs_conn *__ip_vs_conn_in_get
192 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
193 {
194         unsigned hash;
195         struct ip_vs_conn *cp;
196
197         hash = ip_vs_conn_hashkey(protocol, s_addr, s_port);
198
199         ct_read_lock(hash);
200
201         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
202                 if (s_addr==cp->caddr && s_port==cp->cport &&
203                     d_port==cp->vport && d_addr==cp->vaddr &&
204                     ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
205                     protocol==cp->protocol) {
206                         /* HIT */
207                         atomic_inc(&cp->refcnt);
208                         ct_read_unlock(hash);
209                         return cp;
210                 }
211         }
212
213         ct_read_unlock(hash);
214
215         return NULL;
216 }
217
218 struct ip_vs_conn *ip_vs_conn_in_get
219 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
220 {
221         struct ip_vs_conn *cp;
222
223         cp = __ip_vs_conn_in_get(protocol, s_addr, s_port, d_addr, d_port);
224         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt))
225                 cp = __ip_vs_conn_in_get(protocol, s_addr, 0, d_addr, d_port);
226
227         IP_VS_DBG(9, "lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
228                   ip_vs_proto_name(protocol),
229                   NIPQUAD(s_addr), ntohs(s_port),
230                   NIPQUAD(d_addr), ntohs(d_port),
231                   cp?"hit":"not hit");
232
233         return cp;
234 }
235
236 /* Get reference to connection template */
237 struct ip_vs_conn *ip_vs_ct_in_get
238 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
239 {
240         unsigned hash;
241         struct ip_vs_conn *cp;
242
243         hash = ip_vs_conn_hashkey(protocol, s_addr, s_port);
244
245         ct_read_lock(hash);
246
247         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
248                 if (s_addr==cp->caddr && s_port==cp->cport &&
249                     d_port==cp->vport && d_addr==cp->vaddr &&
250                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
251                     protocol==cp->protocol) {
252                         /* HIT */
253                         atomic_inc(&cp->refcnt);
254                         goto out;
255                 }
256         }
257         cp = NULL;
258
259   out:
260         ct_read_unlock(hash);
261
262         IP_VS_DBG(9, "template lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
263                   ip_vs_proto_name(protocol),
264                   NIPQUAD(s_addr), ntohs(s_port),
265                   NIPQUAD(d_addr), ntohs(d_port),
266                   cp?"hit":"not hit");
267
268         return cp;
269 }
270
271 /*
272  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
273  *  Called for pkts coming from inside-to-OUTside.
274  *      s_addr, s_port: pkt source address (inside host)
275  *      d_addr, d_port: pkt dest address (foreign host)
276  */
277 struct ip_vs_conn *ip_vs_conn_out_get
278 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
279 {
280         unsigned hash;
281         struct ip_vs_conn *cp, *ret=NULL;
282
283         /*
284          *      Check for "full" addressed entries
285          */
286         hash = ip_vs_conn_hashkey(protocol, d_addr, d_port);
287
288         ct_read_lock(hash);
289
290         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
291                 if (d_addr == cp->caddr && d_port == cp->cport &&
292                     s_port == cp->dport && s_addr == cp->daddr &&
293                     protocol == cp->protocol) {
294                         /* HIT */
295                         atomic_inc(&cp->refcnt);
296                         ret = cp;
297                         break;
298                 }
299         }
300
301         ct_read_unlock(hash);
302
303         IP_VS_DBG(9, "lookup/out %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
304                   ip_vs_proto_name(protocol),
305                   NIPQUAD(s_addr), ntohs(s_port),
306                   NIPQUAD(d_addr), ntohs(d_port),
307                   ret?"hit":"not hit");
308
309         return ret;
310 }
311
312
313 /*
314  *      Put back the conn and restart its timer with its timeout
315  */
316 void ip_vs_conn_put(struct ip_vs_conn *cp)
317 {
318         /* reset it expire in its timeout */
319         mod_timer(&cp->timer, jiffies+cp->timeout);
320
321         __ip_vs_conn_put(cp);
322 }
323
324
325 /*
326  *      Fill a no_client_port connection with a client port number
327  */
328 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
329 {
330         if (ip_vs_conn_unhash(cp)) {
331                 spin_lock(&cp->lock);
332                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
333                         atomic_dec(&ip_vs_conn_no_cport_cnt);
334                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
335                         cp->cport = cport;
336                 }
337                 spin_unlock(&cp->lock);
338
339                 /* hash on new dport */
340                 ip_vs_conn_hash(cp);
341         }
342 }
343
344
345 /*
346  *      Bind a connection entry with the corresponding packet_xmit.
347  *      Called by ip_vs_conn_new.
348  */
349 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
350 {
351         switch (IP_VS_FWD_METHOD(cp)) {
352         case IP_VS_CONN_F_MASQ:
353                 cp->packet_xmit = ip_vs_nat_xmit;
354                 break;
355
356         case IP_VS_CONN_F_TUNNEL:
357                 cp->packet_xmit = ip_vs_tunnel_xmit;
358                 break;
359
360         case IP_VS_CONN_F_DROUTE:
361                 cp->packet_xmit = ip_vs_dr_xmit;
362                 break;
363
364         case IP_VS_CONN_F_LOCALNODE:
365                 cp->packet_xmit = ip_vs_null_xmit;
366                 break;
367
368         case IP_VS_CONN_F_BYPASS:
369                 cp->packet_xmit = ip_vs_bypass_xmit;
370                 break;
371         }
372 }
373
374
375 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
376 {
377         return atomic_read(&dest->activeconns)
378                 + atomic_read(&dest->inactconns);
379 }
380
381 /*
382  *      Bind a connection entry with a virtual service destination
383  *      Called just after a new connection entry is created.
384  */
385 static inline void
386 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
387 {
388         /* if dest is NULL, then return directly */
389         if (!dest)
390                 return;
391
392         /* Increase the refcnt counter of the dest */
393         atomic_inc(&dest->refcnt);
394
395         /* Bind with the destination and its corresponding transmitter */
396         cp->flags |= atomic_read(&dest->conn_flags);
397         cp->dest = dest;
398
399         IP_VS_DBG(7, "Bind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
400                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
401                   "dest->refcnt:%d\n",
402                   ip_vs_proto_name(cp->protocol),
403                   NIPQUAD(cp->caddr), ntohs(cp->cport),
404                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
405                   NIPQUAD(cp->daddr), ntohs(cp->dport),
406                   ip_vs_fwd_tag(cp), cp->state,
407                   cp->flags, atomic_read(&cp->refcnt),
408                   atomic_read(&dest->refcnt));
409
410         /* Update the connection counters */
411         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
412                 /* It is a normal connection, so increase the inactive
413                    connection counter because it is in TCP SYNRECV
414                    state (inactive) or other protocol inacive state */
415                 atomic_inc(&dest->inactconns);
416         } else {
417                 /* It is a persistent connection/template, so increase
418                    the peristent connection counter */
419                 atomic_inc(&dest->persistconns);
420         }
421
422         if (dest->u_threshold != 0 &&
423             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
424                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
425 }
426
427
428 /*
429  * Check if there is a destination for the connection, if so
430  * bind the connection to the destination.
431  */
432 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
433 {
434         struct ip_vs_dest *dest;
435
436         if ((cp) && (!cp->dest)) {
437                 dest = ip_vs_find_dest(cp->daddr, cp->dport,
438                                        cp->vaddr, cp->vport, cp->protocol);
439                 ip_vs_bind_dest(cp, dest);
440                 return dest;
441         } else
442                 return NULL;
443 }
444
445
446 /*
447  *      Unbind a connection entry with its VS destination
448  *      Called by the ip_vs_conn_expire function.
449  */
450 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
451 {
452         struct ip_vs_dest *dest = cp->dest;
453
454         if (!dest)
455                 return;
456
457         IP_VS_DBG(7, "Unbind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
458                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
459                   "dest->refcnt:%d\n",
460                   ip_vs_proto_name(cp->protocol),
461                   NIPQUAD(cp->caddr), ntohs(cp->cport),
462                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
463                   NIPQUAD(cp->daddr), ntohs(cp->dport),
464                   ip_vs_fwd_tag(cp), cp->state,
465                   cp->flags, atomic_read(&cp->refcnt),
466                   atomic_read(&dest->refcnt));
467
468         /* Update the connection counters */
469         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
470                 /* It is a normal connection, so decrease the inactconns
471                    or activeconns counter */
472                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
473                         atomic_dec(&dest->inactconns);
474                 } else {
475                         atomic_dec(&dest->activeconns);
476                 }
477         } else {
478                 /* It is a persistent connection/template, so decrease
479                    the peristent connection counter */
480                 atomic_dec(&dest->persistconns);
481         }
482
483         if (dest->l_threshold != 0) {
484                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
485                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
486         } else if (dest->u_threshold != 0) {
487                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
488                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
489         } else {
490                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
491                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
492         }
493
494         /*
495          * Simply decrease the refcnt of the dest, because the
496          * dest will be either in service's destination list
497          * or in the trash.
498          */
499         atomic_dec(&dest->refcnt);
500 }
501
502
503 /*
504  *      Checking if the destination of a connection template is available.
505  *      If available, return 1, otherwise invalidate this connection
506  *      template and return 0.
507  */
508 int ip_vs_check_template(struct ip_vs_conn *ct)
509 {
510         struct ip_vs_dest *dest = ct->dest;
511
512         /*
513          * Checking the dest server status.
514          */
515         if ((dest == NULL) ||
516             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
517             (sysctl_ip_vs_expire_quiescent_template &&
518              (atomic_read(&dest->weight) == 0))) {
519                 IP_VS_DBG(9, "check_template: dest not available for "
520                           "protocol %s s:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
521                           "-> d:%u.%u.%u.%u:%d\n",
522                           ip_vs_proto_name(ct->protocol),
523                           NIPQUAD(ct->caddr), ntohs(ct->cport),
524                           NIPQUAD(ct->vaddr), ntohs(ct->vport),
525                           NIPQUAD(ct->daddr), ntohs(ct->dport));
526
527                 /*
528                  * Invalidate the connection template
529                  */
530                 if (ct->vport != htons(0xffff)) {
531                         if (ip_vs_conn_unhash(ct)) {
532                                 ct->dport = htons(0xffff);
533                                 ct->vport = htons(0xffff);
534                                 ct->cport = 0;
535                                 ip_vs_conn_hash(ct);
536                         }
537                 }
538
539                 /*
540                  * Simply decrease the refcnt of the template,
541                  * don't restart its timer.
542                  */
543                 atomic_dec(&ct->refcnt);
544                 return 0;
545         }
546         return 1;
547 }
548
549 static void ip_vs_conn_expire(unsigned long data)
550 {
551         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
552
553         cp->timeout = 60*HZ;
554
555         /*
556          *      hey, I'm using it
557          */
558         atomic_inc(&cp->refcnt);
559
560         /*
561          *      do I control anybody?
562          */
563         if (atomic_read(&cp->n_control))
564                 goto expire_later;
565
566         /*
567          *      unhash it if it is hashed in the conn table
568          */
569         if (!ip_vs_conn_unhash(cp))
570                 goto expire_later;
571
572         /*
573          *      refcnt==1 implies I'm the only one referrer
574          */
575         if (likely(atomic_read(&cp->refcnt) == 1)) {
576                 /* delete the timer if it is activated by other users */
577                 if (timer_pending(&cp->timer))
578                         del_timer(&cp->timer);
579
580                 /* does anybody control me? */
581                 if (cp->control)
582                         ip_vs_control_del(cp);
583
584                 if (unlikely(cp->app != NULL))
585                         ip_vs_unbind_app(cp);
586                 ip_vs_unbind_dest(cp);
587                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
588                         atomic_dec(&ip_vs_conn_no_cport_cnt);
589                 atomic_dec(&ip_vs_conn_count);
590
591                 kmem_cache_free(ip_vs_conn_cachep, cp);
592                 return;
593         }
594
595         /* hash it back to the table */
596         ip_vs_conn_hash(cp);
597
598   expire_later:
599         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
600                   atomic_read(&cp->refcnt)-1,
601                   atomic_read(&cp->n_control));
602
603         ip_vs_conn_put(cp);
604 }
605
606
607 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
608 {
609         if (del_timer(&cp->timer))
610                 mod_timer(&cp->timer, jiffies);
611 }
612
613
614 /*
615  *      Create a new connection entry and hash it into the ip_vs_conn_tab
616  */
617 struct ip_vs_conn *
618 ip_vs_conn_new(int proto, __be32 caddr, __be16 cport, __be32 vaddr, __be16 vport,
619                __be32 daddr, __be16 dport, unsigned flags,
620                struct ip_vs_dest *dest)
621 {
622         struct ip_vs_conn *cp;
623         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
624
625         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
626         if (cp == NULL) {
627                 IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n");
628                 return NULL;
629         }
630
631         INIT_LIST_HEAD(&cp->c_list);
632         init_timer(&cp->timer);
633         cp->timer.data     = (unsigned long)cp;
634         cp->timer.function = ip_vs_conn_expire;
635         cp->protocol       = proto;
636         cp->caddr          = caddr;
637         cp->cport          = cport;
638         cp->vaddr          = vaddr;
639         cp->vport          = vport;
640         cp->daddr          = daddr;
641         cp->dport          = dport;
642         cp->flags          = flags;
643         spin_lock_init(&cp->lock);
644
645         /*
646          * Set the entry is referenced by the current thread before hashing
647          * it in the table, so that other thread run ip_vs_random_dropentry
648          * but cannot drop this entry.
649          */
650         atomic_set(&cp->refcnt, 1);
651
652         atomic_set(&cp->n_control, 0);
653         atomic_set(&cp->in_pkts, 0);
654
655         atomic_inc(&ip_vs_conn_count);
656         if (flags & IP_VS_CONN_F_NO_CPORT)
657                 atomic_inc(&ip_vs_conn_no_cport_cnt);
658
659         /* Bind the connection with a destination server */
660         ip_vs_bind_dest(cp, dest);
661
662         /* Set its state and timeout */
663         cp->state = 0;
664         cp->timeout = 3*HZ;
665
666         /* Bind its packet transmitter */
667         ip_vs_bind_xmit(cp);
668
669         if (unlikely(pp && atomic_read(&pp->appcnt)))
670                 ip_vs_bind_app(cp, pp);
671
672         /* Hash it in the ip_vs_conn_tab finally */
673         ip_vs_conn_hash(cp);
674
675         return cp;
676 }
677
678
679 /*
680  *      /proc/net/ip_vs_conn entries
681  */
682 #ifdef CONFIG_PROC_FS
683
684 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
685 {
686         int idx;
687         struct ip_vs_conn *cp;
688
689         for(idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
690                 ct_read_lock_bh(idx);
691                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
692                         if (pos-- == 0) {
693                                 seq->private = &ip_vs_conn_tab[idx];
694                                 return cp;
695                         }
696                 }
697                 ct_read_unlock_bh(idx);
698         }
699
700         return NULL;
701 }
702
703 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
704 {
705         seq->private = NULL;
706         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
707 }
708
709 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
710 {
711         struct ip_vs_conn *cp = v;
712         struct list_head *e, *l = seq->private;
713         int idx;
714
715         ++*pos;
716         if (v == SEQ_START_TOKEN)
717                 return ip_vs_conn_array(seq, 0);
718
719         /* more on same hash chain? */
720         if ((e = cp->c_list.next) != l)
721                 return list_entry(e, struct ip_vs_conn, c_list);
722
723         idx = l - ip_vs_conn_tab;
724         ct_read_unlock_bh(idx);
725
726         while (++idx < IP_VS_CONN_TAB_SIZE) {
727                 ct_read_lock_bh(idx);
728                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
729                         seq->private = &ip_vs_conn_tab[idx];
730                         return cp;
731                 }
732                 ct_read_unlock_bh(idx);
733         }
734         seq->private = NULL;
735         return NULL;
736 }
737
738 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
739 {
740         struct list_head *l = seq->private;
741
742         if (l)
743                 ct_read_unlock_bh(l - ip_vs_conn_tab);
744 }
745
746 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
747 {
748
749         if (v == SEQ_START_TOKEN)
750                 seq_puts(seq,
751    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
752         else {
753                 const struct ip_vs_conn *cp = v;
754
755                 seq_printf(seq,
756                         "%-3s %08X %04X %08X %04X %08X %04X %-11s %7lu\n",
757                                 ip_vs_proto_name(cp->protocol),
758                                 ntohl(cp->caddr), ntohs(cp->cport),
759                                 ntohl(cp->vaddr), ntohs(cp->vport),
760                                 ntohl(cp->daddr), ntohs(cp->dport),
761                                 ip_vs_state_name(cp->protocol, cp->state),
762                                 (cp->timer.expires-jiffies)/HZ);
763         }
764         return 0;
765 }
766
767 static const struct seq_operations ip_vs_conn_seq_ops = {
768         .start = ip_vs_conn_seq_start,
769         .next  = ip_vs_conn_seq_next,
770         .stop  = ip_vs_conn_seq_stop,
771         .show  = ip_vs_conn_seq_show,
772 };
773
774 static int ip_vs_conn_open(struct inode *inode, struct file *file)
775 {
776         return seq_open(file, &ip_vs_conn_seq_ops);
777 }
778
779 static const struct file_operations ip_vs_conn_fops = {
780         .owner   = THIS_MODULE,
781         .open    = ip_vs_conn_open,
782         .read    = seq_read,
783         .llseek  = seq_lseek,
784         .release = seq_release,
785 };
786 #endif
787
788
789 /*
790  *      Randomly drop connection entries before running out of memory
791  */
792 static inline int todrop_entry(struct ip_vs_conn *cp)
793 {
794         /*
795          * The drop rate array needs tuning for real environments.
796          * Called from timer bh only => no locking
797          */
798         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
799         static char todrop_counter[9] = {0};
800         int i;
801
802         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
803            This will leave enough time for normal connection to get
804            through. */
805         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
806                 return 0;
807
808         /* Don't drop the entry if its number of incoming packets is not
809            located in [0, 8] */
810         i = atomic_read(&cp->in_pkts);
811         if (i > 8 || i < 0) return 0;
812
813         if (!todrop_rate[i]) return 0;
814         if (--todrop_counter[i] > 0) return 0;
815
816         todrop_counter[i] = todrop_rate[i];
817         return 1;
818 }
819
820 /* Called from keventd and must protect itself from softirqs */
821 void ip_vs_random_dropentry(void)
822 {
823         int idx;
824         struct ip_vs_conn *cp;
825
826         /*
827          * Randomly scan 1/32 of the whole table every second
828          */
829         for (idx = 0; idx < (IP_VS_CONN_TAB_SIZE>>5); idx++) {
830                 unsigned hash = net_random() & IP_VS_CONN_TAB_MASK;
831
832                 /*
833                  *  Lock is actually needed in this loop.
834                  */
835                 ct_write_lock_bh(hash);
836
837                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
838                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
839                                 /* connection template */
840                                 continue;
841
842                         if (cp->protocol == IPPROTO_TCP) {
843                                 switch(cp->state) {
844                                 case IP_VS_TCP_S_SYN_RECV:
845                                 case IP_VS_TCP_S_SYNACK:
846                                         break;
847
848                                 case IP_VS_TCP_S_ESTABLISHED:
849                                         if (todrop_entry(cp))
850                                                 break;
851                                         continue;
852
853                                 default:
854                                         continue;
855                                 }
856                         } else {
857                                 if (!todrop_entry(cp))
858                                         continue;
859                         }
860
861                         IP_VS_DBG(4, "del connection\n");
862                         ip_vs_conn_expire_now(cp);
863                         if (cp->control) {
864                                 IP_VS_DBG(4, "del conn template\n");
865                                 ip_vs_conn_expire_now(cp->control);
866                         }
867                 }
868                 ct_write_unlock_bh(hash);
869         }
870 }
871
872
873 /*
874  *      Flush all the connection entries in the ip_vs_conn_tab
875  */
876 static void ip_vs_conn_flush(void)
877 {
878         int idx;
879         struct ip_vs_conn *cp;
880
881   flush_again:
882         for (idx=0; idx<IP_VS_CONN_TAB_SIZE; idx++) {
883                 /*
884                  *  Lock is actually needed in this loop.
885                  */
886                 ct_write_lock_bh(idx);
887
888                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
889
890                         IP_VS_DBG(4, "del connection\n");
891                         ip_vs_conn_expire_now(cp);
892                         if (cp->control) {
893                                 IP_VS_DBG(4, "del conn template\n");
894                                 ip_vs_conn_expire_now(cp->control);
895                         }
896                 }
897                 ct_write_unlock_bh(idx);
898         }
899
900         /* the counter may be not NULL, because maybe some conn entries
901            are run by slow timer handler or unhashed but still referred */
902         if (atomic_read(&ip_vs_conn_count) != 0) {
903                 schedule();
904                 goto flush_again;
905         }
906 }
907
908
909 int ip_vs_conn_init(void)
910 {
911         int idx;
912
913         /*
914          * Allocate the connection hash table and initialize its list heads
915          */
916         ip_vs_conn_tab = vmalloc(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head));
917         if (!ip_vs_conn_tab)
918                 return -ENOMEM;
919
920         /* Allocate ip_vs_conn slab cache */
921         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
922                                               sizeof(struct ip_vs_conn), 0,
923                                               SLAB_HWCACHE_ALIGN, NULL);
924         if (!ip_vs_conn_cachep) {
925                 vfree(ip_vs_conn_tab);
926                 return -ENOMEM;
927         }
928
929         IP_VS_INFO("Connection hash table configured "
930                    "(size=%d, memory=%ldKbytes)\n",
931                    IP_VS_CONN_TAB_SIZE,
932                    (long)(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head))/1024);
933         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
934                   sizeof(struct ip_vs_conn));
935
936         for (idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
937                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
938         }
939
940         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
941                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
942         }
943
944         proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops);
945
946         /* calculate the random value for connection hash */
947         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
948
949         return 0;
950 }
951
952
953 void ip_vs_conn_cleanup(void)
954 {
955         /* flush all the connection entries first */
956         ip_vs_conn_flush();
957
958         /* Release the empty cache */
959         kmem_cache_destroy(ip_vs_conn_cachep);
960         proc_net_remove(&init_net, "ip_vs_conn");
961         vfree(ip_vs_conn_tab);
962 }