Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[linux-2.6] / 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         if ((cp->flags & IP_VS_CONN_F_SYNC) &&
397             (!(cp->flags & IP_VS_CONN_F_TEMPLATE)))
398                 /* if the connection is not template and is created
399                  * by sync, preserve the activity flag.
400                  */
401                 cp->flags |= atomic_read(&dest->conn_flags) &
402                              (~IP_VS_CONN_F_INACTIVE);
403         else
404                 cp->flags |= atomic_read(&dest->conn_flags);
405         cp->dest = dest;
406
407         IP_VS_DBG(7, "Bind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
408                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
409                   "dest->refcnt:%d\n",
410                   ip_vs_proto_name(cp->protocol),
411                   NIPQUAD(cp->caddr), ntohs(cp->cport),
412                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
413                   NIPQUAD(cp->daddr), ntohs(cp->dport),
414                   ip_vs_fwd_tag(cp), cp->state,
415                   cp->flags, atomic_read(&cp->refcnt),
416                   atomic_read(&dest->refcnt));
417
418         /* Update the connection counters */
419         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
420                 /* It is a normal connection, so increase the inactive
421                    connection counter because it is in TCP SYNRECV
422                    state (inactive) or other protocol inacive state */
423                 if ((cp->flags & IP_VS_CONN_F_SYNC) &&
424                     (!(cp->flags & IP_VS_CONN_F_INACTIVE)))
425                         atomic_inc(&dest->activeconns);
426                 else
427                         atomic_inc(&dest->inactconns);
428         } else {
429                 /* It is a persistent connection/template, so increase
430                    the peristent connection counter */
431                 atomic_inc(&dest->persistconns);
432         }
433
434         if (dest->u_threshold != 0 &&
435             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
436                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
437 }
438
439
440 /*
441  * Check if there is a destination for the connection, if so
442  * bind the connection to the destination.
443  */
444 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
445 {
446         struct ip_vs_dest *dest;
447
448         if ((cp) && (!cp->dest)) {
449                 dest = ip_vs_find_dest(cp->daddr, cp->dport,
450                                        cp->vaddr, cp->vport, cp->protocol);
451                 ip_vs_bind_dest(cp, dest);
452                 return dest;
453         } else
454                 return NULL;
455 }
456
457
458 /*
459  *      Unbind a connection entry with its VS destination
460  *      Called by the ip_vs_conn_expire function.
461  */
462 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
463 {
464         struct ip_vs_dest *dest = cp->dest;
465
466         if (!dest)
467                 return;
468
469         IP_VS_DBG(7, "Unbind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
470                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
471                   "dest->refcnt:%d\n",
472                   ip_vs_proto_name(cp->protocol),
473                   NIPQUAD(cp->caddr), ntohs(cp->cport),
474                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
475                   NIPQUAD(cp->daddr), ntohs(cp->dport),
476                   ip_vs_fwd_tag(cp), cp->state,
477                   cp->flags, atomic_read(&cp->refcnt),
478                   atomic_read(&dest->refcnt));
479
480         /* Update the connection counters */
481         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
482                 /* It is a normal connection, so decrease the inactconns
483                    or activeconns counter */
484                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
485                         atomic_dec(&dest->inactconns);
486                 } else {
487                         atomic_dec(&dest->activeconns);
488                 }
489         } else {
490                 /* It is a persistent connection/template, so decrease
491                    the peristent connection counter */
492                 atomic_dec(&dest->persistconns);
493         }
494
495         if (dest->l_threshold != 0) {
496                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
497                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
498         } else if (dest->u_threshold != 0) {
499                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
500                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
501         } else {
502                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
503                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
504         }
505
506         /*
507          * Simply decrease the refcnt of the dest, because the
508          * dest will be either in service's destination list
509          * or in the trash.
510          */
511         atomic_dec(&dest->refcnt);
512 }
513
514
515 /*
516  *      Checking if the destination of a connection template is available.
517  *      If available, return 1, otherwise invalidate this connection
518  *      template and return 0.
519  */
520 int ip_vs_check_template(struct ip_vs_conn *ct)
521 {
522         struct ip_vs_dest *dest = ct->dest;
523
524         /*
525          * Checking the dest server status.
526          */
527         if ((dest == NULL) ||
528             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
529             (sysctl_ip_vs_expire_quiescent_template &&
530              (atomic_read(&dest->weight) == 0))) {
531                 IP_VS_DBG(9, "check_template: dest not available for "
532                           "protocol %s s:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
533                           "-> d:%u.%u.%u.%u:%d\n",
534                           ip_vs_proto_name(ct->protocol),
535                           NIPQUAD(ct->caddr), ntohs(ct->cport),
536                           NIPQUAD(ct->vaddr), ntohs(ct->vport),
537                           NIPQUAD(ct->daddr), ntohs(ct->dport));
538
539                 /*
540                  * Invalidate the connection template
541                  */
542                 if (ct->vport != htons(0xffff)) {
543                         if (ip_vs_conn_unhash(ct)) {
544                                 ct->dport = htons(0xffff);
545                                 ct->vport = htons(0xffff);
546                                 ct->cport = 0;
547                                 ip_vs_conn_hash(ct);
548                         }
549                 }
550
551                 /*
552                  * Simply decrease the refcnt of the template,
553                  * don't restart its timer.
554                  */
555                 atomic_dec(&ct->refcnt);
556                 return 0;
557         }
558         return 1;
559 }
560
561 static void ip_vs_conn_expire(unsigned long data)
562 {
563         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
564
565         cp->timeout = 60*HZ;
566
567         /*
568          *      hey, I'm using it
569          */
570         atomic_inc(&cp->refcnt);
571
572         /*
573          *      do I control anybody?
574          */
575         if (atomic_read(&cp->n_control))
576                 goto expire_later;
577
578         /*
579          *      unhash it if it is hashed in the conn table
580          */
581         if (!ip_vs_conn_unhash(cp))
582                 goto expire_later;
583
584         /*
585          *      refcnt==1 implies I'm the only one referrer
586          */
587         if (likely(atomic_read(&cp->refcnt) == 1)) {
588                 /* delete the timer if it is activated by other users */
589                 if (timer_pending(&cp->timer))
590                         del_timer(&cp->timer);
591
592                 /* does anybody control me? */
593                 if (cp->control)
594                         ip_vs_control_del(cp);
595
596                 if (unlikely(cp->app != NULL))
597                         ip_vs_unbind_app(cp);
598                 ip_vs_unbind_dest(cp);
599                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
600                         atomic_dec(&ip_vs_conn_no_cport_cnt);
601                 atomic_dec(&ip_vs_conn_count);
602
603                 kmem_cache_free(ip_vs_conn_cachep, cp);
604                 return;
605         }
606
607         /* hash it back to the table */
608         ip_vs_conn_hash(cp);
609
610   expire_later:
611         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
612                   atomic_read(&cp->refcnt)-1,
613                   atomic_read(&cp->n_control));
614
615         ip_vs_conn_put(cp);
616 }
617
618
619 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
620 {
621         if (del_timer(&cp->timer))
622                 mod_timer(&cp->timer, jiffies);
623 }
624
625
626 /*
627  *      Create a new connection entry and hash it into the ip_vs_conn_tab
628  */
629 struct ip_vs_conn *
630 ip_vs_conn_new(int proto, __be32 caddr, __be16 cport, __be32 vaddr, __be16 vport,
631                __be32 daddr, __be16 dport, unsigned flags,
632                struct ip_vs_dest *dest)
633 {
634         struct ip_vs_conn *cp;
635         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
636
637         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
638         if (cp == NULL) {
639                 IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n");
640                 return NULL;
641         }
642
643         INIT_LIST_HEAD(&cp->c_list);
644         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
645         cp->protocol       = proto;
646         cp->caddr          = caddr;
647         cp->cport          = cport;
648         cp->vaddr          = vaddr;
649         cp->vport          = vport;
650         cp->daddr          = daddr;
651         cp->dport          = dport;
652         cp->flags          = flags;
653         spin_lock_init(&cp->lock);
654
655         /*
656          * Set the entry is referenced by the current thread before hashing
657          * it in the table, so that other thread run ip_vs_random_dropentry
658          * but cannot drop this entry.
659          */
660         atomic_set(&cp->refcnt, 1);
661
662         atomic_set(&cp->n_control, 0);
663         atomic_set(&cp->in_pkts, 0);
664
665         atomic_inc(&ip_vs_conn_count);
666         if (flags & IP_VS_CONN_F_NO_CPORT)
667                 atomic_inc(&ip_vs_conn_no_cport_cnt);
668
669         /* Bind the connection with a destination server */
670         ip_vs_bind_dest(cp, dest);
671
672         /* Set its state and timeout */
673         cp->state = 0;
674         cp->timeout = 3*HZ;
675
676         /* Bind its packet transmitter */
677         ip_vs_bind_xmit(cp);
678
679         if (unlikely(pp && atomic_read(&pp->appcnt)))
680                 ip_vs_bind_app(cp, pp);
681
682         /* Hash it in the ip_vs_conn_tab finally */
683         ip_vs_conn_hash(cp);
684
685         return cp;
686 }
687
688
689 /*
690  *      /proc/net/ip_vs_conn entries
691  */
692 #ifdef CONFIG_PROC_FS
693
694 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
695 {
696         int idx;
697         struct ip_vs_conn *cp;
698
699         for(idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
700                 ct_read_lock_bh(idx);
701                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
702                         if (pos-- == 0) {
703                                 seq->private = &ip_vs_conn_tab[idx];
704                                 return cp;
705                         }
706                 }
707                 ct_read_unlock_bh(idx);
708         }
709
710         return NULL;
711 }
712
713 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
714 {
715         seq->private = NULL;
716         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
717 }
718
719 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
720 {
721         struct ip_vs_conn *cp = v;
722         struct list_head *e, *l = seq->private;
723         int idx;
724
725         ++*pos;
726         if (v == SEQ_START_TOKEN)
727                 return ip_vs_conn_array(seq, 0);
728
729         /* more on same hash chain? */
730         if ((e = cp->c_list.next) != l)
731                 return list_entry(e, struct ip_vs_conn, c_list);
732
733         idx = l - ip_vs_conn_tab;
734         ct_read_unlock_bh(idx);
735
736         while (++idx < IP_VS_CONN_TAB_SIZE) {
737                 ct_read_lock_bh(idx);
738                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
739                         seq->private = &ip_vs_conn_tab[idx];
740                         return cp;
741                 }
742                 ct_read_unlock_bh(idx);
743         }
744         seq->private = NULL;
745         return NULL;
746 }
747
748 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
749 {
750         struct list_head *l = seq->private;
751
752         if (l)
753                 ct_read_unlock_bh(l - ip_vs_conn_tab);
754 }
755
756 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
757 {
758
759         if (v == SEQ_START_TOKEN)
760                 seq_puts(seq,
761    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
762         else {
763                 const struct ip_vs_conn *cp = v;
764
765                 seq_printf(seq,
766                         "%-3s %08X %04X %08X %04X %08X %04X %-11s %7lu\n",
767                                 ip_vs_proto_name(cp->protocol),
768                                 ntohl(cp->caddr), ntohs(cp->cport),
769                                 ntohl(cp->vaddr), ntohs(cp->vport),
770                                 ntohl(cp->daddr), ntohs(cp->dport),
771                                 ip_vs_state_name(cp->protocol, cp->state),
772                                 (cp->timer.expires-jiffies)/HZ);
773         }
774         return 0;
775 }
776
777 static const struct seq_operations ip_vs_conn_seq_ops = {
778         .start = ip_vs_conn_seq_start,
779         .next  = ip_vs_conn_seq_next,
780         .stop  = ip_vs_conn_seq_stop,
781         .show  = ip_vs_conn_seq_show,
782 };
783
784 static int ip_vs_conn_open(struct inode *inode, struct file *file)
785 {
786         return seq_open(file, &ip_vs_conn_seq_ops);
787 }
788
789 static const struct file_operations ip_vs_conn_fops = {
790         .owner   = THIS_MODULE,
791         .open    = ip_vs_conn_open,
792         .read    = seq_read,
793         .llseek  = seq_lseek,
794         .release = seq_release,
795 };
796
797 static const char *ip_vs_origin_name(unsigned flags)
798 {
799         if (flags & IP_VS_CONN_F_SYNC)
800                 return "SYNC";
801         else
802                 return "LOCAL";
803 }
804
805 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
806 {
807
808         if (v == SEQ_START_TOKEN)
809                 seq_puts(seq,
810    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
811         else {
812                 const struct ip_vs_conn *cp = v;
813
814                 seq_printf(seq,
815                         "%-3s %08X %04X %08X %04X %08X %04X %-11s %-6s %7lu\n",
816                                 ip_vs_proto_name(cp->protocol),
817                                 ntohl(cp->caddr), ntohs(cp->cport),
818                                 ntohl(cp->vaddr), ntohs(cp->vport),
819                                 ntohl(cp->daddr), ntohs(cp->dport),
820                                 ip_vs_state_name(cp->protocol, cp->state),
821                                 ip_vs_origin_name(cp->flags),
822                                 (cp->timer.expires-jiffies)/HZ);
823         }
824         return 0;
825 }
826
827 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
828         .start = ip_vs_conn_seq_start,
829         .next  = ip_vs_conn_seq_next,
830         .stop  = ip_vs_conn_seq_stop,
831         .show  = ip_vs_conn_sync_seq_show,
832 };
833
834 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
835 {
836         return seq_open(file, &ip_vs_conn_sync_seq_ops);
837 }
838
839 static const struct file_operations ip_vs_conn_sync_fops = {
840         .owner   = THIS_MODULE,
841         .open    = ip_vs_conn_sync_open,
842         .read    = seq_read,
843         .llseek  = seq_lseek,
844         .release = seq_release,
845 };
846
847 #endif
848
849
850 /*
851  *      Randomly drop connection entries before running out of memory
852  */
853 static inline int todrop_entry(struct ip_vs_conn *cp)
854 {
855         /*
856          * The drop rate array needs tuning for real environments.
857          * Called from timer bh only => no locking
858          */
859         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
860         static char todrop_counter[9] = {0};
861         int i;
862
863         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
864            This will leave enough time for normal connection to get
865            through. */
866         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
867                 return 0;
868
869         /* Don't drop the entry if its number of incoming packets is not
870            located in [0, 8] */
871         i = atomic_read(&cp->in_pkts);
872         if (i > 8 || i < 0) return 0;
873
874         if (!todrop_rate[i]) return 0;
875         if (--todrop_counter[i] > 0) return 0;
876
877         todrop_counter[i] = todrop_rate[i];
878         return 1;
879 }
880
881 /* Called from keventd and must protect itself from softirqs */
882 void ip_vs_random_dropentry(void)
883 {
884         int idx;
885         struct ip_vs_conn *cp;
886
887         /*
888          * Randomly scan 1/32 of the whole table every second
889          */
890         for (idx = 0; idx < (IP_VS_CONN_TAB_SIZE>>5); idx++) {
891                 unsigned hash = net_random() & IP_VS_CONN_TAB_MASK;
892
893                 /*
894                  *  Lock is actually needed in this loop.
895                  */
896                 ct_write_lock_bh(hash);
897
898                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
899                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
900                                 /* connection template */
901                                 continue;
902
903                         if (cp->protocol == IPPROTO_TCP) {
904                                 switch(cp->state) {
905                                 case IP_VS_TCP_S_SYN_RECV:
906                                 case IP_VS_TCP_S_SYNACK:
907                                         break;
908
909                                 case IP_VS_TCP_S_ESTABLISHED:
910                                         if (todrop_entry(cp))
911                                                 break;
912                                         continue;
913
914                                 default:
915                                         continue;
916                                 }
917                         } else {
918                                 if (!todrop_entry(cp))
919                                         continue;
920                         }
921
922                         IP_VS_DBG(4, "del connection\n");
923                         ip_vs_conn_expire_now(cp);
924                         if (cp->control) {
925                                 IP_VS_DBG(4, "del conn template\n");
926                                 ip_vs_conn_expire_now(cp->control);
927                         }
928                 }
929                 ct_write_unlock_bh(hash);
930         }
931 }
932
933
934 /*
935  *      Flush all the connection entries in the ip_vs_conn_tab
936  */
937 static void ip_vs_conn_flush(void)
938 {
939         int idx;
940         struct ip_vs_conn *cp;
941
942   flush_again:
943         for (idx=0; idx<IP_VS_CONN_TAB_SIZE; idx++) {
944                 /*
945                  *  Lock is actually needed in this loop.
946                  */
947                 ct_write_lock_bh(idx);
948
949                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
950
951                         IP_VS_DBG(4, "del connection\n");
952                         ip_vs_conn_expire_now(cp);
953                         if (cp->control) {
954                                 IP_VS_DBG(4, "del conn template\n");
955                                 ip_vs_conn_expire_now(cp->control);
956                         }
957                 }
958                 ct_write_unlock_bh(idx);
959         }
960
961         /* the counter may be not NULL, because maybe some conn entries
962            are run by slow timer handler or unhashed but still referred */
963         if (atomic_read(&ip_vs_conn_count) != 0) {
964                 schedule();
965                 goto flush_again;
966         }
967 }
968
969
970 int ip_vs_conn_init(void)
971 {
972         int idx;
973
974         /*
975          * Allocate the connection hash table and initialize its list heads
976          */
977         ip_vs_conn_tab = vmalloc(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head));
978         if (!ip_vs_conn_tab)
979                 return -ENOMEM;
980
981         /* Allocate ip_vs_conn slab cache */
982         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
983                                               sizeof(struct ip_vs_conn), 0,
984                                               SLAB_HWCACHE_ALIGN, NULL);
985         if (!ip_vs_conn_cachep) {
986                 vfree(ip_vs_conn_tab);
987                 return -ENOMEM;
988         }
989
990         IP_VS_INFO("Connection hash table configured "
991                    "(size=%d, memory=%ldKbytes)\n",
992                    IP_VS_CONN_TAB_SIZE,
993                    (long)(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head))/1024);
994         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
995                   sizeof(struct ip_vs_conn));
996
997         for (idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
998                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
999         }
1000
1001         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1002                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1003         }
1004
1005         proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops);
1006         proc_net_fops_create(&init_net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops);
1007
1008         /* calculate the random value for connection hash */
1009         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1010
1011         return 0;
1012 }
1013
1014
1015 void ip_vs_conn_cleanup(void)
1016 {
1017         /* flush all the connection entries first */
1018         ip_vs_conn_flush();
1019
1020         /* Release the empty cache */
1021         kmem_cache_destroy(ip_vs_conn_cachep);
1022         proc_net_remove(&init_net, "ip_vs_conn");
1023         proc_net_remove(&init_net, "ip_vs_conn_sync");
1024         vfree(ip_vs_conn_tab);
1025 }