Merge git://git.kernel.org/pub/scm/linux/kernel/git/bunk/trivial
[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/in.h>
28 #include <linux/net.h>
29 #include <linux/kernel.h>
30 #include <linux/module.h>
31 #include <linux/vmalloc.h>
32 #include <linux/proc_fs.h>              /* for proc_net_* */
33 #include <linux/seq_file.h>
34 #include <linux/jhash.h>
35 #include <linux/random.h>
36
37 #include <net/ip_vs.h>
38
39
40 /*
41  *  Connection hash table: for input and output packets lookups of IPVS
42  */
43 static struct list_head *ip_vs_conn_tab;
44
45 /*  SLAB cache for IPVS connections */
46 static kmem_cache_t *ip_vs_conn_cachep __read_mostly;
47
48 /*  counter for current IPVS connections */
49 static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
50
51 /*  counter for no client port connections */
52 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
53
54 /* random value for IPVS connection hash */
55 static unsigned int ip_vs_conn_rnd;
56
57 /*
58  *  Fine locking granularity for big connection hash table
59  */
60 #define CT_LOCKARRAY_BITS  4
61 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
62 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
63
64 struct ip_vs_aligned_lock
65 {
66         rwlock_t        l;
67 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
68
69 /* lock array for conn table */
70 static struct ip_vs_aligned_lock
71 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
72
73 static inline void ct_read_lock(unsigned key)
74 {
75         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
76 }
77
78 static inline void ct_read_unlock(unsigned key)
79 {
80         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
81 }
82
83 static inline void ct_write_lock(unsigned key)
84 {
85         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
86 }
87
88 static inline void ct_write_unlock(unsigned key)
89 {
90         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
91 }
92
93 static inline void ct_read_lock_bh(unsigned key)
94 {
95         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
96 }
97
98 static inline void ct_read_unlock_bh(unsigned key)
99 {
100         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
101 }
102
103 static inline void ct_write_lock_bh(unsigned key)
104 {
105         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
106 }
107
108 static inline void ct_write_unlock_bh(unsigned key)
109 {
110         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
111 }
112
113
114 /*
115  *      Returns hash value for IPVS connection entry
116  */
117 static unsigned int ip_vs_conn_hashkey(unsigned proto, __u32 addr, __u16 port)
118 {
119         return jhash_3words(addr, port, proto, ip_vs_conn_rnd)
120                 & IP_VS_CONN_TAB_MASK;
121 }
122
123
124 /*
125  *      Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
126  *      returns bool success.
127  */
128 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
129 {
130         unsigned hash;
131         int ret;
132
133         /* Hash by protocol, client address and port */
134         hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport);
135
136         ct_write_lock(hash);
137
138         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
139                 list_add(&cp->c_list, &ip_vs_conn_tab[hash]);
140                 cp->flags |= IP_VS_CONN_F_HASHED;
141                 atomic_inc(&cp->refcnt);
142                 ret = 1;
143         } else {
144                 IP_VS_ERR("ip_vs_conn_hash(): request for already hashed, "
145                           "called from %p\n", __builtin_return_address(0));
146                 ret = 0;
147         }
148
149         ct_write_unlock(hash);
150
151         return ret;
152 }
153
154
155 /*
156  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
157  *      returns bool success.
158  */
159 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
160 {
161         unsigned hash;
162         int ret;
163
164         /* unhash it and decrease its reference counter */
165         hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport);
166
167         ct_write_lock(hash);
168
169         if (cp->flags & IP_VS_CONN_F_HASHED) {
170                 list_del(&cp->c_list);
171                 cp->flags &= ~IP_VS_CONN_F_HASHED;
172                 atomic_dec(&cp->refcnt);
173                 ret = 1;
174         } else
175                 ret = 0;
176
177         ct_write_unlock(hash);
178
179         return ret;
180 }
181
182
183 /*
184  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
185  *  Called for pkts coming from OUTside-to-INside.
186  *      s_addr, s_port: pkt source address (foreign host)
187  *      d_addr, d_port: pkt dest address (load balancer)
188  */
189 static inline struct ip_vs_conn *__ip_vs_conn_in_get
190 (int protocol, __u32 s_addr, __u16 s_port, __u32 d_addr, __u16 d_port)
191 {
192         unsigned hash;
193         struct ip_vs_conn *cp;
194
195         hash = ip_vs_conn_hashkey(protocol, s_addr, s_port);
196
197         ct_read_lock(hash);
198
199         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
200                 if (s_addr==cp->caddr && s_port==cp->cport &&
201                     d_port==cp->vport && d_addr==cp->vaddr &&
202                     ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
203                     protocol==cp->protocol) {
204                         /* HIT */
205                         atomic_inc(&cp->refcnt);
206                         ct_read_unlock(hash);
207                         return cp;
208                 }
209         }
210
211         ct_read_unlock(hash);
212
213         return NULL;
214 }
215
216 struct ip_vs_conn *ip_vs_conn_in_get
217 (int protocol, __u32 s_addr, __u16 s_port, __u32 d_addr, __u16 d_port)
218 {
219         struct ip_vs_conn *cp;
220
221         cp = __ip_vs_conn_in_get(protocol, s_addr, s_port, d_addr, d_port);
222         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt))
223                 cp = __ip_vs_conn_in_get(protocol, s_addr, 0, d_addr, d_port);
224
225         IP_VS_DBG(9, "lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
226                   ip_vs_proto_name(protocol),
227                   NIPQUAD(s_addr), ntohs(s_port),
228                   NIPQUAD(d_addr), ntohs(d_port),
229                   cp?"hit":"not hit");
230
231         return cp;
232 }
233
234 /* Get reference to connection template */
235 struct ip_vs_conn *ip_vs_ct_in_get
236 (int protocol, __u32 s_addr, __u16 s_port, __u32 d_addr, __u16 d_port)
237 {
238         unsigned hash;
239         struct ip_vs_conn *cp;
240
241         hash = ip_vs_conn_hashkey(protocol, s_addr, s_port);
242
243         ct_read_lock(hash);
244
245         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
246                 if (s_addr==cp->caddr && s_port==cp->cport &&
247                     d_port==cp->vport && d_addr==cp->vaddr &&
248                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
249                     protocol==cp->protocol) {
250                         /* HIT */
251                         atomic_inc(&cp->refcnt);
252                         goto out;
253                 }
254         }
255         cp = NULL;
256
257   out:
258         ct_read_unlock(hash);
259
260         IP_VS_DBG(9, "template lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
261                   ip_vs_proto_name(protocol),
262                   NIPQUAD(s_addr), ntohs(s_port),
263                   NIPQUAD(d_addr), ntohs(d_port),
264                   cp?"hit":"not hit");
265
266         return cp;
267 }
268
269 /*
270  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
271  *  Called for pkts coming from inside-to-OUTside.
272  *      s_addr, s_port: pkt source address (inside host)
273  *      d_addr, d_port: pkt dest address (foreign host)
274  */
275 struct ip_vs_conn *ip_vs_conn_out_get
276 (int protocol, __u32 s_addr, __u16 s_port, __u32 d_addr, __u16 d_port)
277 {
278         unsigned hash;
279         struct ip_vs_conn *cp, *ret=NULL;
280
281         /*
282          *      Check for "full" addressed entries
283          */
284         hash = ip_vs_conn_hashkey(protocol, d_addr, d_port);
285
286         ct_read_lock(hash);
287
288         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
289                 if (d_addr == cp->caddr && d_port == cp->cport &&
290                     s_port == cp->dport && s_addr == cp->daddr &&
291                     protocol == cp->protocol) {
292                         /* HIT */
293                         atomic_inc(&cp->refcnt);
294                         ret = cp;
295                         break;
296                 }
297         }
298
299         ct_read_unlock(hash);
300
301         IP_VS_DBG(9, "lookup/out %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
302                   ip_vs_proto_name(protocol),
303                   NIPQUAD(s_addr), ntohs(s_port),
304                   NIPQUAD(d_addr), ntohs(d_port),
305                   ret?"hit":"not hit");
306
307         return ret;
308 }
309
310
311 /*
312  *      Put back the conn and restart its timer with its timeout
313  */
314 void ip_vs_conn_put(struct ip_vs_conn *cp)
315 {
316         /* reset it expire in its timeout */
317         mod_timer(&cp->timer, jiffies+cp->timeout);
318
319         __ip_vs_conn_put(cp);
320 }
321
322
323 /*
324  *      Fill a no_client_port connection with a client port number
325  */
326 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __u16 cport)
327 {
328         if (ip_vs_conn_unhash(cp)) {
329                 spin_lock(&cp->lock);
330                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
331                         atomic_dec(&ip_vs_conn_no_cport_cnt);
332                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
333                         cp->cport = cport;
334                 }
335                 spin_unlock(&cp->lock);
336
337                 /* hash on new dport */
338                 ip_vs_conn_hash(cp);
339         }
340 }
341
342
343 /*
344  *      Bind a connection entry with the corresponding packet_xmit.
345  *      Called by ip_vs_conn_new.
346  */
347 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
348 {
349         switch (IP_VS_FWD_METHOD(cp)) {
350         case IP_VS_CONN_F_MASQ:
351                 cp->packet_xmit = ip_vs_nat_xmit;
352                 break;
353
354         case IP_VS_CONN_F_TUNNEL:
355                 cp->packet_xmit = ip_vs_tunnel_xmit;
356                 break;
357
358         case IP_VS_CONN_F_DROUTE:
359                 cp->packet_xmit = ip_vs_dr_xmit;
360                 break;
361
362         case IP_VS_CONN_F_LOCALNODE:
363                 cp->packet_xmit = ip_vs_null_xmit;
364                 break;
365
366         case IP_VS_CONN_F_BYPASS:
367                 cp->packet_xmit = ip_vs_bypass_xmit;
368                 break;
369         }
370 }
371
372
373 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
374 {
375         return atomic_read(&dest->activeconns)
376                 + atomic_read(&dest->inactconns);
377 }
378
379 /*
380  *      Bind a connection entry with a virtual service destination
381  *      Called just after a new connection entry is created.
382  */
383 static inline void
384 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
385 {
386         /* if dest is NULL, then return directly */
387         if (!dest)
388                 return;
389
390         /* Increase the refcnt counter of the dest */
391         atomic_inc(&dest->refcnt);
392
393         /* Bind with the destination and its corresponding transmitter */
394         cp->flags |= atomic_read(&dest->conn_flags);
395         cp->dest = dest;
396
397         IP_VS_DBG(7, "Bind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
398                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
399                   "dest->refcnt:%d\n",
400                   ip_vs_proto_name(cp->protocol),
401                   NIPQUAD(cp->caddr), ntohs(cp->cport),
402                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
403                   NIPQUAD(cp->daddr), ntohs(cp->dport),
404                   ip_vs_fwd_tag(cp), cp->state,
405                   cp->flags, atomic_read(&cp->refcnt),
406                   atomic_read(&dest->refcnt));
407
408         /* Update the connection counters */
409         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
410                 /* It is a normal connection, so increase the inactive
411                    connection counter because it is in TCP SYNRECV
412                    state (inactive) or other protocol inacive state */
413                 atomic_inc(&dest->inactconns);
414         } else {
415                 /* It is a persistent connection/template, so increase
416                    the peristent connection counter */
417                 atomic_inc(&dest->persistconns);
418         }
419
420         if (dest->u_threshold != 0 &&
421             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
422                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
423 }
424
425
426 /*
427  *      Unbind a connection entry with its VS destination
428  *      Called by the ip_vs_conn_expire function.
429  */
430 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
431 {
432         struct ip_vs_dest *dest = cp->dest;
433
434         if (!dest)
435                 return;
436
437         IP_VS_DBG(7, "Unbind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
438                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
439                   "dest->refcnt:%d\n",
440                   ip_vs_proto_name(cp->protocol),
441                   NIPQUAD(cp->caddr), ntohs(cp->cport),
442                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
443                   NIPQUAD(cp->daddr), ntohs(cp->dport),
444                   ip_vs_fwd_tag(cp), cp->state,
445                   cp->flags, atomic_read(&cp->refcnt),
446                   atomic_read(&dest->refcnt));
447
448         /* Update the connection counters */
449         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
450                 /* It is a normal connection, so decrease the inactconns
451                    or activeconns counter */
452                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
453                         atomic_dec(&dest->inactconns);
454                 } else {
455                         atomic_dec(&dest->activeconns);
456                 }
457         } else {
458                 /* It is a persistent connection/template, so decrease
459                    the peristent connection counter */
460                 atomic_dec(&dest->persistconns);
461         }
462
463         if (dest->l_threshold != 0) {
464                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
465                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
466         } else if (dest->u_threshold != 0) {
467                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
468                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
469         } else {
470                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
471                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
472         }
473
474         /*
475          * Simply decrease the refcnt of the dest, because the
476          * dest will be either in service's destination list
477          * or in the trash.
478          */
479         atomic_dec(&dest->refcnt);
480 }
481
482
483 /*
484  *      Checking if the destination of a connection template is available.
485  *      If available, return 1, otherwise invalidate this connection
486  *      template and return 0.
487  */
488 int ip_vs_check_template(struct ip_vs_conn *ct)
489 {
490         struct ip_vs_dest *dest = ct->dest;
491
492         /*
493          * Checking the dest server status.
494          */
495         if ((dest == NULL) ||
496             !(dest->flags & IP_VS_DEST_F_AVAILABLE) || 
497             (sysctl_ip_vs_expire_quiescent_template && 
498              (atomic_read(&dest->weight) == 0))) {
499                 IP_VS_DBG(9, "check_template: dest not available for "
500                           "protocol %s s:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
501                           "-> d:%u.%u.%u.%u:%d\n",
502                           ip_vs_proto_name(ct->protocol),
503                           NIPQUAD(ct->caddr), ntohs(ct->cport),
504                           NIPQUAD(ct->vaddr), ntohs(ct->vport),
505                           NIPQUAD(ct->daddr), ntohs(ct->dport));
506
507                 /*
508                  * Invalidate the connection template
509                  */
510                 if (ct->vport != 65535) {
511                         if (ip_vs_conn_unhash(ct)) {
512                                 ct->dport = 65535;
513                                 ct->vport = 65535;
514                                 ct->cport = 0;
515                                 ip_vs_conn_hash(ct);
516                         }
517                 }
518
519                 /*
520                  * Simply decrease the refcnt of the template,
521                  * don't restart its timer.
522                  */
523                 atomic_dec(&ct->refcnt);
524                 return 0;
525         }
526         return 1;
527 }
528
529 static void ip_vs_conn_expire(unsigned long data)
530 {
531         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
532
533         cp->timeout = 60*HZ;
534
535         /*
536          *      hey, I'm using it
537          */
538         atomic_inc(&cp->refcnt);
539
540         /*
541          *      do I control anybody?
542          */
543         if (atomic_read(&cp->n_control))
544                 goto expire_later;
545
546         /*
547          *      unhash it if it is hashed in the conn table
548          */
549         if (!ip_vs_conn_unhash(cp))
550                 goto expire_later;
551
552         /*
553          *      refcnt==1 implies I'm the only one referrer
554          */
555         if (likely(atomic_read(&cp->refcnt) == 1)) {
556                 /* delete the timer if it is activated by other users */
557                 if (timer_pending(&cp->timer))
558                         del_timer(&cp->timer);
559
560                 /* does anybody control me? */
561                 if (cp->control)
562                         ip_vs_control_del(cp);
563
564                 if (unlikely(cp->app != NULL))
565                         ip_vs_unbind_app(cp);
566                 ip_vs_unbind_dest(cp);
567                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
568                         atomic_dec(&ip_vs_conn_no_cport_cnt);
569                 atomic_dec(&ip_vs_conn_count);
570
571                 kmem_cache_free(ip_vs_conn_cachep, cp);
572                 return;
573         }
574
575         /* hash it back to the table */
576         ip_vs_conn_hash(cp);
577
578   expire_later:
579         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
580                   atomic_read(&cp->refcnt)-1,
581                   atomic_read(&cp->n_control));
582
583         ip_vs_conn_put(cp);
584 }
585
586
587 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
588 {
589         if (del_timer(&cp->timer))
590                 mod_timer(&cp->timer, jiffies);
591 }
592
593
594 /*
595  *      Create a new connection entry and hash it into the ip_vs_conn_tab
596  */
597 struct ip_vs_conn *
598 ip_vs_conn_new(int proto, __u32 caddr, __u16 cport, __u32 vaddr, __u16 vport,
599                __u32 daddr, __u16 dport, unsigned flags,
600                struct ip_vs_dest *dest)
601 {
602         struct ip_vs_conn *cp;
603         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
604
605         cp = kmem_cache_alloc(ip_vs_conn_cachep, GFP_ATOMIC);
606         if (cp == NULL) {
607                 IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n");
608                 return NULL;
609         }
610
611         memset(cp, 0, sizeof(*cp));
612         INIT_LIST_HEAD(&cp->c_list);
613         init_timer(&cp->timer);
614         cp->timer.data     = (unsigned long)cp;
615         cp->timer.function = ip_vs_conn_expire;
616         cp->protocol       = proto;
617         cp->caddr          = caddr;
618         cp->cport          = cport;
619         cp->vaddr          = vaddr;
620         cp->vport          = vport;
621         cp->daddr          = daddr;
622         cp->dport          = dport;
623         cp->flags          = flags;
624         spin_lock_init(&cp->lock);
625
626         /*
627          * Set the entry is referenced by the current thread before hashing
628          * it in the table, so that other thread run ip_vs_random_dropentry
629          * but cannot drop this entry.
630          */
631         atomic_set(&cp->refcnt, 1);
632
633         atomic_set(&cp->n_control, 0);
634         atomic_set(&cp->in_pkts, 0);
635
636         atomic_inc(&ip_vs_conn_count);
637         if (flags & IP_VS_CONN_F_NO_CPORT)
638                 atomic_inc(&ip_vs_conn_no_cport_cnt);
639
640         /* Bind the connection with a destination server */
641         ip_vs_bind_dest(cp, dest);
642
643         /* Set its state and timeout */
644         cp->state = 0;
645         cp->timeout = 3*HZ;
646
647         /* Bind its packet transmitter */
648         ip_vs_bind_xmit(cp);
649
650         if (unlikely(pp && atomic_read(&pp->appcnt)))
651                 ip_vs_bind_app(cp, pp);
652
653         /* Hash it in the ip_vs_conn_tab finally */
654         ip_vs_conn_hash(cp);
655
656         return cp;
657 }
658
659
660 /*
661  *      /proc/net/ip_vs_conn entries
662  */
663 #ifdef CONFIG_PROC_FS
664
665 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
666 {
667         int idx;
668         struct ip_vs_conn *cp;
669         
670         for(idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
671                 ct_read_lock_bh(idx);
672                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
673                         if (pos-- == 0) {
674                                 seq->private = &ip_vs_conn_tab[idx];
675                                 return cp;
676                         }
677                 }
678                 ct_read_unlock_bh(idx);
679         }
680
681         return NULL;
682 }
683
684 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
685 {
686         seq->private = NULL;
687         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
688 }
689
690 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
691 {
692         struct ip_vs_conn *cp = v;
693         struct list_head *e, *l = seq->private;
694         int idx;
695
696         ++*pos;
697         if (v == SEQ_START_TOKEN) 
698                 return ip_vs_conn_array(seq, 0);
699
700         /* more on same hash chain? */
701         if ((e = cp->c_list.next) != l)
702                 return list_entry(e, struct ip_vs_conn, c_list);
703
704         idx = l - ip_vs_conn_tab;
705         ct_read_unlock_bh(idx);
706
707         while (++idx < IP_VS_CONN_TAB_SIZE) {
708                 ct_read_lock_bh(idx);
709                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
710                         seq->private = &ip_vs_conn_tab[idx];
711                         return cp;
712                 }       
713                 ct_read_unlock_bh(idx);
714         }
715         seq->private = NULL;
716         return NULL;
717 }
718
719 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
720 {
721         struct list_head *l = seq->private;
722
723         if (l)
724                 ct_read_unlock_bh(l - ip_vs_conn_tab);
725 }
726
727 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
728 {
729
730         if (v == SEQ_START_TOKEN)
731                 seq_puts(seq,
732    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
733         else {
734                 const struct ip_vs_conn *cp = v;
735
736                 seq_printf(seq,
737                         "%-3s %08X %04X %08X %04X %08X %04X %-11s %7lu\n",
738                                 ip_vs_proto_name(cp->protocol),
739                                 ntohl(cp->caddr), ntohs(cp->cport),
740                                 ntohl(cp->vaddr), ntohs(cp->vport),
741                                 ntohl(cp->daddr), ntohs(cp->dport),
742                                 ip_vs_state_name(cp->protocol, cp->state),
743                                 (cp->timer.expires-jiffies)/HZ);
744         }
745         return 0;
746 }
747
748 static struct seq_operations ip_vs_conn_seq_ops = {
749         .start = ip_vs_conn_seq_start,
750         .next  = ip_vs_conn_seq_next,
751         .stop  = ip_vs_conn_seq_stop,
752         .show  = ip_vs_conn_seq_show,
753 };
754
755 static int ip_vs_conn_open(struct inode *inode, struct file *file)
756 {
757         return seq_open(file, &ip_vs_conn_seq_ops);
758 }
759
760 static struct file_operations ip_vs_conn_fops = {
761         .owner   = THIS_MODULE,
762         .open    = ip_vs_conn_open,
763         .read    = seq_read,
764         .llseek  = seq_lseek,
765         .release = seq_release,
766 };
767 #endif
768
769
770 /*
771  *      Randomly drop connection entries before running out of memory
772  */
773 static inline int todrop_entry(struct ip_vs_conn *cp)
774 {
775         /*
776          * The drop rate array needs tuning for real environments.
777          * Called from timer bh only => no locking
778          */
779         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
780         static char todrop_counter[9] = {0};
781         int i;
782
783         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
784            This will leave enough time for normal connection to get
785            through. */
786         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
787                 return 0;
788
789         /* Don't drop the entry if its number of incoming packets is not
790            located in [0, 8] */
791         i = atomic_read(&cp->in_pkts);
792         if (i > 8 || i < 0) return 0;
793
794         if (!todrop_rate[i]) return 0;
795         if (--todrop_counter[i] > 0) return 0;
796
797         todrop_counter[i] = todrop_rate[i];
798         return 1;
799 }
800
801 /* Called from keventd and must protect itself from softirqs */
802 void ip_vs_random_dropentry(void)
803 {
804         int idx;
805         struct ip_vs_conn *cp;
806
807         /*
808          * Randomly scan 1/32 of the whole table every second
809          */
810         for (idx = 0; idx < (IP_VS_CONN_TAB_SIZE>>5); idx++) {
811                 unsigned hash = net_random() & IP_VS_CONN_TAB_MASK;
812
813                 /*
814                  *  Lock is actually needed in this loop.
815                  */
816                 ct_write_lock_bh(hash);
817
818                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
819                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
820                                 /* connection template */
821                                 continue;
822
823                         if (cp->protocol == IPPROTO_TCP) {
824                                 switch(cp->state) {
825                                 case IP_VS_TCP_S_SYN_RECV:
826                                 case IP_VS_TCP_S_SYNACK:
827                                         break;
828
829                                 case IP_VS_TCP_S_ESTABLISHED:
830                                         if (todrop_entry(cp))
831                                                 break;
832                                         continue;
833
834                                 default:
835                                         continue;
836                                 }
837                         } else {
838                                 if (!todrop_entry(cp))
839                                         continue;
840                         }
841
842                         IP_VS_DBG(4, "del connection\n");
843                         ip_vs_conn_expire_now(cp);
844                         if (cp->control) {
845                                 IP_VS_DBG(4, "del conn template\n");
846                                 ip_vs_conn_expire_now(cp->control);
847                         }
848                 }
849                 ct_write_unlock_bh(hash);
850         }
851 }
852
853
854 /*
855  *      Flush all the connection entries in the ip_vs_conn_tab
856  */
857 static void ip_vs_conn_flush(void)
858 {
859         int idx;
860         struct ip_vs_conn *cp;
861
862   flush_again:
863         for (idx=0; idx<IP_VS_CONN_TAB_SIZE; idx++) {
864                 /*
865                  *  Lock is actually needed in this loop.
866                  */
867                 ct_write_lock_bh(idx);
868
869                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
870
871                         IP_VS_DBG(4, "del connection\n");
872                         ip_vs_conn_expire_now(cp);
873                         if (cp->control) {
874                                 IP_VS_DBG(4, "del conn template\n");
875                                 ip_vs_conn_expire_now(cp->control);
876                         }
877                 }
878                 ct_write_unlock_bh(idx);
879         }
880
881         /* the counter may be not NULL, because maybe some conn entries
882            are run by slow timer handler or unhashed but still referred */
883         if (atomic_read(&ip_vs_conn_count) != 0) {
884                 schedule();
885                 goto flush_again;
886         }
887 }
888
889
890 int ip_vs_conn_init(void)
891 {
892         int idx;
893
894         /*
895          * Allocate the connection hash table and initialize its list heads
896          */
897         ip_vs_conn_tab = vmalloc(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head));
898         if (!ip_vs_conn_tab)
899                 return -ENOMEM;
900
901         /* Allocate ip_vs_conn slab cache */
902         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
903                                               sizeof(struct ip_vs_conn), 0,
904                                               SLAB_HWCACHE_ALIGN, NULL, NULL);
905         if (!ip_vs_conn_cachep) {
906                 vfree(ip_vs_conn_tab);
907                 return -ENOMEM;
908         }
909
910         IP_VS_INFO("Connection hash table configured "
911                    "(size=%d, memory=%ldKbytes)\n",
912                    IP_VS_CONN_TAB_SIZE,
913                    (long)(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head))/1024);
914         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
915                   sizeof(struct ip_vs_conn));
916
917         for (idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
918                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
919         }
920
921         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
922                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
923         }
924
925         proc_net_fops_create("ip_vs_conn", 0, &ip_vs_conn_fops);
926
927         /* calculate the random value for connection hash */
928         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
929
930         return 0;
931 }
932
933
934 void ip_vs_conn_cleanup(void)
935 {
936         /* flush all the connection entries first */
937         ip_vs_conn_flush();
938
939         /* Release the empty cache */
940         kmem_cache_destroy(ip_vs_conn_cachep);
941         proc_net_remove("ip_vs_conn");
942         vfree(ip_vs_conn_tab);
943 }