some kmalloc/memset ->kzalloc (tree wide)
[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/ip_vs.h>
39
40
41 /*
42  *  Connection hash table: for input and output packets lookups of IPVS
43  */
44 static struct list_head *ip_vs_conn_tab;
45
46 /*  SLAB cache for IPVS connections */
47 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
48
49 /*  counter for current IPVS connections */
50 static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
51
52 /*  counter for no client port connections */
53 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
54
55 /* random value for IPVS connection hash */
56 static unsigned int ip_vs_conn_rnd;
57
58 /*
59  *  Fine locking granularity for big connection hash table
60  */
61 #define CT_LOCKARRAY_BITS  4
62 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
63 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
64
65 struct ip_vs_aligned_lock
66 {
67         rwlock_t        l;
68 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
69
70 /* lock array for conn table */
71 static struct ip_vs_aligned_lock
72 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
73
74 static inline void ct_read_lock(unsigned key)
75 {
76         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
77 }
78
79 static inline void ct_read_unlock(unsigned key)
80 {
81         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
82 }
83
84 static inline void ct_write_lock(unsigned key)
85 {
86         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
87 }
88
89 static inline void ct_write_unlock(unsigned key)
90 {
91         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
92 }
93
94 static inline void ct_read_lock_bh(unsigned key)
95 {
96         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
97 }
98
99 static inline void ct_read_unlock_bh(unsigned key)
100 {
101         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
102 }
103
104 static inline void ct_write_lock_bh(unsigned key)
105 {
106         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
107 }
108
109 static inline void ct_write_unlock_bh(unsigned key)
110 {
111         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
112 }
113
114
115 /*
116  *      Returns hash value for IPVS connection entry
117  */
118 static unsigned int ip_vs_conn_hashkey(unsigned proto, __be32 addr, __be16 port)
119 {
120         return jhash_3words((__force u32)addr, (__force u32)port, proto, ip_vs_conn_rnd)
121                 & IP_VS_CONN_TAB_MASK;
122 }
123
124
125 /*
126  *      Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
127  *      returns bool success.
128  */
129 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
130 {
131         unsigned hash;
132         int ret;
133
134         /* Hash by protocol, client address and port */
135         hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport);
136
137         ct_write_lock(hash);
138
139         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
140                 list_add(&cp->c_list, &ip_vs_conn_tab[hash]);
141                 cp->flags |= IP_VS_CONN_F_HASHED;
142                 atomic_inc(&cp->refcnt);
143                 ret = 1;
144         } else {
145                 IP_VS_ERR("ip_vs_conn_hash(): request for already hashed, "
146                           "called from %p\n", __builtin_return_address(0));
147                 ret = 0;
148         }
149
150         ct_write_unlock(hash);
151
152         return ret;
153 }
154
155
156 /*
157  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
158  *      returns bool success.
159  */
160 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
161 {
162         unsigned hash;
163         int ret;
164
165         /* unhash it and decrease its reference counter */
166         hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport);
167
168         ct_write_lock(hash);
169
170         if (cp->flags & IP_VS_CONN_F_HASHED) {
171                 list_del(&cp->c_list);
172                 cp->flags &= ~IP_VS_CONN_F_HASHED;
173                 atomic_dec(&cp->refcnt);
174                 ret = 1;
175         } else
176                 ret = 0;
177
178         ct_write_unlock(hash);
179
180         return ret;
181 }
182
183
184 /*
185  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
186  *  Called for pkts coming from OUTside-to-INside.
187  *      s_addr, s_port: pkt source address (foreign host)
188  *      d_addr, d_port: pkt dest address (load balancer)
189  */
190 static inline struct ip_vs_conn *__ip_vs_conn_in_get
191 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
192 {
193         unsigned hash;
194         struct ip_vs_conn *cp;
195
196         hash = ip_vs_conn_hashkey(protocol, s_addr, s_port);
197
198         ct_read_lock(hash);
199
200         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
201                 if (s_addr==cp->caddr && s_port==cp->cport &&
202                     d_port==cp->vport && d_addr==cp->vaddr &&
203                     ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
204                     protocol==cp->protocol) {
205                         /* HIT */
206                         atomic_inc(&cp->refcnt);
207                         ct_read_unlock(hash);
208                         return cp;
209                 }
210         }
211
212         ct_read_unlock(hash);
213
214         return NULL;
215 }
216
217 struct ip_vs_conn *ip_vs_conn_in_get
218 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
219 {
220         struct ip_vs_conn *cp;
221
222         cp = __ip_vs_conn_in_get(protocol, s_addr, s_port, d_addr, d_port);
223         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt))
224                 cp = __ip_vs_conn_in_get(protocol, s_addr, 0, d_addr, d_port);
225
226         IP_VS_DBG(9, "lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
227                   ip_vs_proto_name(protocol),
228                   NIPQUAD(s_addr), ntohs(s_port),
229                   NIPQUAD(d_addr), ntohs(d_port),
230                   cp?"hit":"not hit");
231
232         return cp;
233 }
234
235 /* Get reference to connection template */
236 struct ip_vs_conn *ip_vs_ct_in_get
237 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
238 {
239         unsigned hash;
240         struct ip_vs_conn *cp;
241
242         hash = ip_vs_conn_hashkey(protocol, s_addr, s_port);
243
244         ct_read_lock(hash);
245
246         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
247                 if (s_addr==cp->caddr && s_port==cp->cport &&
248                     d_port==cp->vport && d_addr==cp->vaddr &&
249                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
250                     protocol==cp->protocol) {
251                         /* HIT */
252                         atomic_inc(&cp->refcnt);
253                         goto out;
254                 }
255         }
256         cp = NULL;
257
258   out:
259         ct_read_unlock(hash);
260
261         IP_VS_DBG(9, "template lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
262                   ip_vs_proto_name(protocol),
263                   NIPQUAD(s_addr), ntohs(s_port),
264                   NIPQUAD(d_addr), ntohs(d_port),
265                   cp?"hit":"not hit");
266
267         return cp;
268 }
269
270 /*
271  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
272  *  Called for pkts coming from inside-to-OUTside.
273  *      s_addr, s_port: pkt source address (inside host)
274  *      d_addr, d_port: pkt dest address (foreign host)
275  */
276 struct ip_vs_conn *ip_vs_conn_out_get
277 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
278 {
279         unsigned hash;
280         struct ip_vs_conn *cp, *ret=NULL;
281
282         /*
283          *      Check for "full" addressed entries
284          */
285         hash = ip_vs_conn_hashkey(protocol, d_addr, d_port);
286
287         ct_read_lock(hash);
288
289         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
290                 if (d_addr == cp->caddr && d_port == cp->cport &&
291                     s_port == cp->dport && s_addr == cp->daddr &&
292                     protocol == cp->protocol) {
293                         /* HIT */
294                         atomic_inc(&cp->refcnt);
295                         ret = cp;
296                         break;
297                 }
298         }
299
300         ct_read_unlock(hash);
301
302         IP_VS_DBG(9, "lookup/out %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
303                   ip_vs_proto_name(protocol),
304                   NIPQUAD(s_addr), ntohs(s_port),
305                   NIPQUAD(d_addr), ntohs(d_port),
306                   ret?"hit":"not hit");
307
308         return ret;
309 }
310
311
312 /*
313  *      Put back the conn and restart its timer with its timeout
314  */
315 void ip_vs_conn_put(struct ip_vs_conn *cp)
316 {
317         /* reset it expire in its timeout */
318         mod_timer(&cp->timer, jiffies+cp->timeout);
319
320         __ip_vs_conn_put(cp);
321 }
322
323
324 /*
325  *      Fill a no_client_port connection with a client port number
326  */
327 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
328 {
329         if (ip_vs_conn_unhash(cp)) {
330                 spin_lock(&cp->lock);
331                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
332                         atomic_dec(&ip_vs_conn_no_cport_cnt);
333                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
334                         cp->cport = cport;
335                 }
336                 spin_unlock(&cp->lock);
337
338                 /* hash on new dport */
339                 ip_vs_conn_hash(cp);
340         }
341 }
342
343
344 /*
345  *      Bind a connection entry with the corresponding packet_xmit.
346  *      Called by ip_vs_conn_new.
347  */
348 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
349 {
350         switch (IP_VS_FWD_METHOD(cp)) {
351         case IP_VS_CONN_F_MASQ:
352                 cp->packet_xmit = ip_vs_nat_xmit;
353                 break;
354
355         case IP_VS_CONN_F_TUNNEL:
356                 cp->packet_xmit = ip_vs_tunnel_xmit;
357                 break;
358
359         case IP_VS_CONN_F_DROUTE:
360                 cp->packet_xmit = ip_vs_dr_xmit;
361                 break;
362
363         case IP_VS_CONN_F_LOCALNODE:
364                 cp->packet_xmit = ip_vs_null_xmit;
365                 break;
366
367         case IP_VS_CONN_F_BYPASS:
368                 cp->packet_xmit = ip_vs_bypass_xmit;
369                 break;
370         }
371 }
372
373
374 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
375 {
376         return atomic_read(&dest->activeconns)
377                 + atomic_read(&dest->inactconns);
378 }
379
380 /*
381  *      Bind a connection entry with a virtual service destination
382  *      Called just after a new connection entry is created.
383  */
384 static inline void
385 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
386 {
387         /* if dest is NULL, then return directly */
388         if (!dest)
389                 return;
390
391         /* Increase the refcnt counter of the dest */
392         atomic_inc(&dest->refcnt);
393
394         /* Bind with the destination and its corresponding transmitter */
395         cp->flags |= atomic_read(&dest->conn_flags);
396         cp->dest = dest;
397
398         IP_VS_DBG(7, "Bind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
399                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
400                   "dest->refcnt:%d\n",
401                   ip_vs_proto_name(cp->protocol),
402                   NIPQUAD(cp->caddr), ntohs(cp->cport),
403                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
404                   NIPQUAD(cp->daddr), ntohs(cp->dport),
405                   ip_vs_fwd_tag(cp), cp->state,
406                   cp->flags, atomic_read(&cp->refcnt),
407                   atomic_read(&dest->refcnt));
408
409         /* Update the connection counters */
410         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
411                 /* It is a normal connection, so increase the inactive
412                    connection counter because it is in TCP SYNRECV
413                    state (inactive) or other protocol inacive state */
414                 atomic_inc(&dest->inactconns);
415         } else {
416                 /* It is a persistent connection/template, so increase
417                    the peristent connection counter */
418                 atomic_inc(&dest->persistconns);
419         }
420
421         if (dest->u_threshold != 0 &&
422             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
423                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
424 }
425
426
427 /*
428  *      Unbind a connection entry with its VS destination
429  *      Called by the ip_vs_conn_expire function.
430  */
431 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
432 {
433         struct ip_vs_dest *dest = cp->dest;
434
435         if (!dest)
436                 return;
437
438         IP_VS_DBG(7, "Unbind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
439                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
440                   "dest->refcnt:%d\n",
441                   ip_vs_proto_name(cp->protocol),
442                   NIPQUAD(cp->caddr), ntohs(cp->cport),
443                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
444                   NIPQUAD(cp->daddr), ntohs(cp->dport),
445                   ip_vs_fwd_tag(cp), cp->state,
446                   cp->flags, atomic_read(&cp->refcnt),
447                   atomic_read(&dest->refcnt));
448
449         /* Update the connection counters */
450         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
451                 /* It is a normal connection, so decrease the inactconns
452                    or activeconns counter */
453                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
454                         atomic_dec(&dest->inactconns);
455                 } else {
456                         atomic_dec(&dest->activeconns);
457                 }
458         } else {
459                 /* It is a persistent connection/template, so decrease
460                    the peristent connection counter */
461                 atomic_dec(&dest->persistconns);
462         }
463
464         if (dest->l_threshold != 0) {
465                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
466                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
467         } else if (dest->u_threshold != 0) {
468                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
469                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
470         } else {
471                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
472                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
473         }
474
475         /*
476          * Simply decrease the refcnt of the dest, because the
477          * dest will be either in service's destination list
478          * or in the trash.
479          */
480         atomic_dec(&dest->refcnt);
481 }
482
483
484 /*
485  *      Checking if the destination of a connection template is available.
486  *      If available, return 1, otherwise invalidate this connection
487  *      template and return 0.
488  */
489 int ip_vs_check_template(struct ip_vs_conn *ct)
490 {
491         struct ip_vs_dest *dest = ct->dest;
492
493         /*
494          * Checking the dest server status.
495          */
496         if ((dest == NULL) ||
497             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
498             (sysctl_ip_vs_expire_quiescent_template &&
499              (atomic_read(&dest->weight) == 0))) {
500                 IP_VS_DBG(9, "check_template: dest not available for "
501                           "protocol %s s:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
502                           "-> d:%u.%u.%u.%u:%d\n",
503                           ip_vs_proto_name(ct->protocol),
504                           NIPQUAD(ct->caddr), ntohs(ct->cport),
505                           NIPQUAD(ct->vaddr), ntohs(ct->vport),
506                           NIPQUAD(ct->daddr), ntohs(ct->dport));
507
508                 /*
509                  * Invalidate the connection template
510                  */
511                 if (ct->vport != htons(0xffff)) {
512                         if (ip_vs_conn_unhash(ct)) {
513                                 ct->dport = htons(0xffff);
514                                 ct->vport = htons(0xffff);
515                                 ct->cport = 0;
516                                 ip_vs_conn_hash(ct);
517                         }
518                 }
519
520                 /*
521                  * Simply decrease the refcnt of the template,
522                  * don't restart its timer.
523                  */
524                 atomic_dec(&ct->refcnt);
525                 return 0;
526         }
527         return 1;
528 }
529
530 static void ip_vs_conn_expire(unsigned long data)
531 {
532         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
533
534         cp->timeout = 60*HZ;
535
536         /*
537          *      hey, I'm using it
538          */
539         atomic_inc(&cp->refcnt);
540
541         /*
542          *      do I control anybody?
543          */
544         if (atomic_read(&cp->n_control))
545                 goto expire_later;
546
547         /*
548          *      unhash it if it is hashed in the conn table
549          */
550         if (!ip_vs_conn_unhash(cp))
551                 goto expire_later;
552
553         /*
554          *      refcnt==1 implies I'm the only one referrer
555          */
556         if (likely(atomic_read(&cp->refcnt) == 1)) {
557                 /* delete the timer if it is activated by other users */
558                 if (timer_pending(&cp->timer))
559                         del_timer(&cp->timer);
560
561                 /* does anybody control me? */
562                 if (cp->control)
563                         ip_vs_control_del(cp);
564
565                 if (unlikely(cp->app != NULL))
566                         ip_vs_unbind_app(cp);
567                 ip_vs_unbind_dest(cp);
568                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
569                         atomic_dec(&ip_vs_conn_no_cport_cnt);
570                 atomic_dec(&ip_vs_conn_count);
571
572                 kmem_cache_free(ip_vs_conn_cachep, cp);
573                 return;
574         }
575
576         /* hash it back to the table */
577         ip_vs_conn_hash(cp);
578
579   expire_later:
580         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
581                   atomic_read(&cp->refcnt)-1,
582                   atomic_read(&cp->n_control));
583
584         ip_vs_conn_put(cp);
585 }
586
587
588 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
589 {
590         if (del_timer(&cp->timer))
591                 mod_timer(&cp->timer, jiffies);
592 }
593
594
595 /*
596  *      Create a new connection entry and hash it into the ip_vs_conn_tab
597  */
598 struct ip_vs_conn *
599 ip_vs_conn_new(int proto, __be32 caddr, __be16 cport, __be32 vaddr, __be16 vport,
600                __be32 daddr, __be16 dport, unsigned flags,
601                struct ip_vs_dest *dest)
602 {
603         struct ip_vs_conn *cp;
604         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
605
606         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
607         if (cp == NULL) {
608                 IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n");
609                 return NULL;
610         }
611
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 const 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 const 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 }