c97bd45975bed636c99dd55e263869f8be11a98b
[linux-2.6.git] / net / netfilter / ipvs / ip_vs_conn.c
1 /*
2  * IPVS         An implementation of the IP virtual server support for the
3  *              LINUX operating system.  IPVS is now implemented as a module
4  *              over the Netfilter framework. IPVS can be used to build a
5  *              high-performance and highly available server based on a
6  *              cluster of servers.
7  *
8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
9  *              Peter Kese <peter.kese@ijs.si>
10  *              Julian Anastasov <ja@ssi.bg>
11  *
12  *              This program is free software; you can redistribute it and/or
13  *              modify it under the terms of the GNU General Public License
14  *              as published by the Free Software Foundation; either version
15  *              2 of the License, or (at your option) any later version.
16  *
17  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
18  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
19  * and others. Many code here is taken from IP MASQ code of kernel 2.2.
20  *
21  * Changes:
22  *
23  */
24
25 #define KMSG_COMPONENT "IPVS"
26 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
27
28 #include <linux/interrupt.h>
29 #include <linux/in.h>
30 #include <linux/net.h>
31 #include <linux/kernel.h>
32 #include <linux/module.h>
33 #include <linux/vmalloc.h>
34 #include <linux/proc_fs.h>              /* for proc_net_* */
35 #include <linux/slab.h>
36 #include <linux/seq_file.h>
37 #include <linux/jhash.h>
38 #include <linux/random.h>
39
40 #include <net/net_namespace.h>
41 #include <net/ip_vs.h>
42
43
44 #ifndef CONFIG_IP_VS_TAB_BITS
45 #define CONFIG_IP_VS_TAB_BITS   12
46 #endif
47
48 /*
49  * Connection hash size. Default is what was selected at compile time.
50 */
51 static int ip_vs_conn_tab_bits = CONFIG_IP_VS_TAB_BITS;
52 module_param_named(conn_tab_bits, ip_vs_conn_tab_bits, int, 0444);
53 MODULE_PARM_DESC(conn_tab_bits, "Set connections' hash size");
54
55 /* size and mask values */
56 int ip_vs_conn_tab_size __read_mostly;
57 static int ip_vs_conn_tab_mask __read_mostly;
58
59 /*
60  *  Connection hash table: for input and output packets lookups of IPVS
61  */
62 static struct hlist_head *ip_vs_conn_tab __read_mostly;
63
64 /*  SLAB cache for IPVS connections */
65 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
66
67 /*  counter for no client port connections */
68 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
69
70 /* random value for IPVS connection hash */
71 static unsigned int ip_vs_conn_rnd __read_mostly;
72
73 /*
74  *  Fine locking granularity for big connection hash table
75  */
76 #define CT_LOCKARRAY_BITS  5
77 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
78 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
79
80 struct ip_vs_aligned_lock
81 {
82         rwlock_t        l;
83 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
84
85 /* lock array for conn table */
86 static struct ip_vs_aligned_lock
87 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
88
89 static inline void ct_read_lock(unsigned key)
90 {
91         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
92 }
93
94 static inline void ct_read_unlock(unsigned key)
95 {
96         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
97 }
98
99 static inline void ct_write_lock(unsigned key)
100 {
101         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
102 }
103
104 static inline void ct_write_unlock(unsigned key)
105 {
106         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
107 }
108
109 static inline void ct_read_lock_bh(unsigned key)
110 {
111         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
112 }
113
114 static inline void ct_read_unlock_bh(unsigned key)
115 {
116         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
117 }
118
119 static inline void ct_write_lock_bh(unsigned key)
120 {
121         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
122 }
123
124 static inline void ct_write_unlock_bh(unsigned key)
125 {
126         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
127 }
128
129
130 /*
131  *      Returns hash value for IPVS connection entry
132  */
133 static unsigned int ip_vs_conn_hashkey(struct net *net, int af, unsigned proto,
134                                        const union nf_inet_addr *addr,
135                                        __be16 port)
136 {
137 #ifdef CONFIG_IP_VS_IPV6
138         if (af == AF_INET6)
139                 return (jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
140                                     (__force u32)port, proto, ip_vs_conn_rnd) ^
141                         ((size_t)net>>8)) & ip_vs_conn_tab_mask;
142 #endif
143         return (jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
144                             ip_vs_conn_rnd) ^
145                 ((size_t)net>>8)) & ip_vs_conn_tab_mask;
146 }
147
148 static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param *p,
149                                              bool inverse)
150 {
151         const union nf_inet_addr *addr;
152         __be16 port;
153
154         if (p->pe_data && p->pe->hashkey_raw)
155                 return p->pe->hashkey_raw(p, ip_vs_conn_rnd, inverse) &
156                         ip_vs_conn_tab_mask;
157
158         if (likely(!inverse)) {
159                 addr = p->caddr;
160                 port = p->cport;
161         } else {
162                 addr = p->vaddr;
163                 port = p->vport;
164         }
165
166         return ip_vs_conn_hashkey(p->net, p->af, p->protocol, addr, port);
167 }
168
169 static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn *cp)
170 {
171         struct ip_vs_conn_param p;
172
173         ip_vs_conn_fill_param(ip_vs_conn_net(cp), cp->af, cp->protocol,
174                               &cp->caddr, cp->cport, NULL, 0, &p);
175
176         if (cp->pe) {
177                 p.pe = cp->pe;
178                 p.pe_data = cp->pe_data;
179                 p.pe_data_len = cp->pe_data_len;
180         }
181
182         return ip_vs_conn_hashkey_param(&p, false);
183 }
184
185 /*
186  *      Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
187  *      returns bool success.
188  */
189 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
190 {
191         unsigned hash;
192         int ret;
193
194         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
195                 return 0;
196
197         /* Hash by protocol, client address and port */
198         hash = ip_vs_conn_hashkey_conn(cp);
199
200         ct_write_lock(hash);
201         spin_lock(&cp->lock);
202
203         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
204                 hlist_add_head(&cp->c_list, &ip_vs_conn_tab[hash]);
205                 cp->flags |= IP_VS_CONN_F_HASHED;
206                 atomic_inc(&cp->refcnt);
207                 ret = 1;
208         } else {
209                 pr_err("%s(): request for already hashed, called from %pF\n",
210                        __func__, __builtin_return_address(0));
211                 ret = 0;
212         }
213
214         spin_unlock(&cp->lock);
215         ct_write_unlock(hash);
216
217         return ret;
218 }
219
220
221 /*
222  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
223  *      returns bool success.
224  */
225 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
226 {
227         unsigned hash;
228         int ret;
229
230         /* unhash it and decrease its reference counter */
231         hash = ip_vs_conn_hashkey_conn(cp);
232
233         ct_write_lock(hash);
234         spin_lock(&cp->lock);
235
236         if (cp->flags & IP_VS_CONN_F_HASHED) {
237                 hlist_del(&cp->c_list);
238                 cp->flags &= ~IP_VS_CONN_F_HASHED;
239                 atomic_dec(&cp->refcnt);
240                 ret = 1;
241         } else
242                 ret = 0;
243
244         spin_unlock(&cp->lock);
245         ct_write_unlock(hash);
246
247         return ret;
248 }
249
250
251 /*
252  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
253  *  Called for pkts coming from OUTside-to-INside.
254  *      p->caddr, p->cport: pkt source address (foreign host)
255  *      p->vaddr, p->vport: pkt dest address (load balancer)
256  */
257 static inline struct ip_vs_conn *
258 __ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
259 {
260         unsigned hash;
261         struct ip_vs_conn *cp;
262         struct hlist_node *n;
263
264         hash = ip_vs_conn_hashkey_param(p, false);
265
266         ct_read_lock(hash);
267
268         hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
269                 if (cp->af == p->af &&
270                     p->cport == cp->cport && p->vport == cp->vport &&
271                     ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
272                     ip_vs_addr_equal(p->af, p->vaddr, &cp->vaddr) &&
273                     ((!p->cport) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
274                     p->protocol == cp->protocol &&
275                     ip_vs_conn_net_eq(cp, p->net)) {
276                         /* HIT */
277                         atomic_inc(&cp->refcnt);
278                         ct_read_unlock(hash);
279                         return cp;
280                 }
281         }
282
283         ct_read_unlock(hash);
284
285         return NULL;
286 }
287
288 struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
289 {
290         struct ip_vs_conn *cp;
291
292         cp = __ip_vs_conn_in_get(p);
293         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt)) {
294                 struct ip_vs_conn_param cport_zero_p = *p;
295                 cport_zero_p.cport = 0;
296                 cp = __ip_vs_conn_in_get(&cport_zero_p);
297         }
298
299         IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
300                       ip_vs_proto_name(p->protocol),
301                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
302                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
303                       cp ? "hit" : "not hit");
304
305         return cp;
306 }
307
308 static int
309 ip_vs_conn_fill_param_proto(int af, const struct sk_buff *skb,
310                             const struct ip_vs_iphdr *iph,
311                             unsigned int proto_off, int inverse,
312                             struct ip_vs_conn_param *p)
313 {
314         __be16 _ports[2], *pptr;
315         struct net *net = skb_net(skb);
316
317         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
318         if (pptr == NULL)
319                 return 1;
320
321         if (likely(!inverse))
322                 ip_vs_conn_fill_param(net, af, iph->protocol, &iph->saddr,
323                                       pptr[0], &iph->daddr, pptr[1], p);
324         else
325                 ip_vs_conn_fill_param(net, af, iph->protocol, &iph->daddr,
326                                       pptr[1], &iph->saddr, pptr[0], p);
327         return 0;
328 }
329
330 struct ip_vs_conn *
331 ip_vs_conn_in_get_proto(int af, const struct sk_buff *skb,
332                         const struct ip_vs_iphdr *iph,
333                         unsigned int proto_off, int inverse)
334 {
335         struct ip_vs_conn_param p;
336
337         if (ip_vs_conn_fill_param_proto(af, skb, iph, proto_off, inverse, &p))
338                 return NULL;
339
340         return ip_vs_conn_in_get(&p);
341 }
342 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
343
344 /* Get reference to connection template */
345 struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p)
346 {
347         unsigned hash;
348         struct ip_vs_conn *cp;
349         struct hlist_node *n;
350
351         hash = ip_vs_conn_hashkey_param(p, false);
352
353         ct_read_lock(hash);
354
355         hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
356                 if (!ip_vs_conn_net_eq(cp, p->net))
357                         continue;
358                 if (p->pe_data && p->pe->ct_match) {
359                         if (p->pe == cp->pe && p->pe->ct_match(p, cp))
360                                 goto out;
361                         continue;
362                 }
363
364                 if (cp->af == p->af &&
365                     ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
366                     /* protocol should only be IPPROTO_IP if
367                      * p->vaddr is a fwmark */
368                     ip_vs_addr_equal(p->protocol == IPPROTO_IP ? AF_UNSPEC :
369                                      p->af, p->vaddr, &cp->vaddr) &&
370                     p->cport == cp->cport && p->vport == cp->vport &&
371                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
372                     p->protocol == cp->protocol)
373                         goto out;
374         }
375         cp = NULL;
376
377   out:
378         if (cp)
379                 atomic_inc(&cp->refcnt);
380         ct_read_unlock(hash);
381
382         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
383                       ip_vs_proto_name(p->protocol),
384                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
385                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
386                       cp ? "hit" : "not hit");
387
388         return cp;
389 }
390
391 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
392  * Called for pkts coming from inside-to-OUTside.
393  *      p->caddr, p->cport: pkt source address (inside host)
394  *      p->vaddr, p->vport: pkt dest address (foreign host) */
395 struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
396 {
397         unsigned hash;
398         struct ip_vs_conn *cp, *ret=NULL;
399         struct hlist_node *n;
400
401         /*
402          *      Check for "full" addressed entries
403          */
404         hash = ip_vs_conn_hashkey_param(p, true);
405
406         ct_read_lock(hash);
407
408         hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
409                 if (cp->af == p->af &&
410                     p->vport == cp->cport && p->cport == cp->dport &&
411                     ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
412                     ip_vs_addr_equal(p->af, p->caddr, &cp->daddr) &&
413                     p->protocol == cp->protocol &&
414                     ip_vs_conn_net_eq(cp, p->net)) {
415                         /* HIT */
416                         atomic_inc(&cp->refcnt);
417                         ret = cp;
418                         break;
419                 }
420         }
421
422         ct_read_unlock(hash);
423
424         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
425                       ip_vs_proto_name(p->protocol),
426                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
427                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
428                       ret ? "hit" : "not hit");
429
430         return ret;
431 }
432
433 struct ip_vs_conn *
434 ip_vs_conn_out_get_proto(int af, const struct sk_buff *skb,
435                          const struct ip_vs_iphdr *iph,
436                          unsigned int proto_off, int inverse)
437 {
438         struct ip_vs_conn_param p;
439
440         if (ip_vs_conn_fill_param_proto(af, skb, iph, proto_off, inverse, &p))
441                 return NULL;
442
443         return ip_vs_conn_out_get(&p);
444 }
445 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
446
447 /*
448  *      Put back the conn and restart its timer with its timeout
449  */
450 void ip_vs_conn_put(struct ip_vs_conn *cp)
451 {
452         unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
453                 0 : cp->timeout;
454         mod_timer(&cp->timer, jiffies+t);
455
456         __ip_vs_conn_put(cp);
457 }
458
459
460 /*
461  *      Fill a no_client_port connection with a client port number
462  */
463 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
464 {
465         if (ip_vs_conn_unhash(cp)) {
466                 spin_lock(&cp->lock);
467                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
468                         atomic_dec(&ip_vs_conn_no_cport_cnt);
469                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
470                         cp->cport = cport;
471                 }
472                 spin_unlock(&cp->lock);
473
474                 /* hash on new dport */
475                 ip_vs_conn_hash(cp);
476         }
477 }
478
479
480 /*
481  *      Bind a connection entry with the corresponding packet_xmit.
482  *      Called by ip_vs_conn_new.
483  */
484 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
485 {
486         switch (IP_VS_FWD_METHOD(cp)) {
487         case IP_VS_CONN_F_MASQ:
488                 cp->packet_xmit = ip_vs_nat_xmit;
489                 break;
490
491         case IP_VS_CONN_F_TUNNEL:
492                 cp->packet_xmit = ip_vs_tunnel_xmit;
493                 break;
494
495         case IP_VS_CONN_F_DROUTE:
496                 cp->packet_xmit = ip_vs_dr_xmit;
497                 break;
498
499         case IP_VS_CONN_F_LOCALNODE:
500                 cp->packet_xmit = ip_vs_null_xmit;
501                 break;
502
503         case IP_VS_CONN_F_BYPASS:
504                 cp->packet_xmit = ip_vs_bypass_xmit;
505                 break;
506         }
507 }
508
509 #ifdef CONFIG_IP_VS_IPV6
510 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
511 {
512         switch (IP_VS_FWD_METHOD(cp)) {
513         case IP_VS_CONN_F_MASQ:
514                 cp->packet_xmit = ip_vs_nat_xmit_v6;
515                 break;
516
517         case IP_VS_CONN_F_TUNNEL:
518                 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
519                 break;
520
521         case IP_VS_CONN_F_DROUTE:
522                 cp->packet_xmit = ip_vs_dr_xmit_v6;
523                 break;
524
525         case IP_VS_CONN_F_LOCALNODE:
526                 cp->packet_xmit = ip_vs_null_xmit;
527                 break;
528
529         case IP_VS_CONN_F_BYPASS:
530                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
531                 break;
532         }
533 }
534 #endif
535
536
537 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
538 {
539         return atomic_read(&dest->activeconns)
540                 + atomic_read(&dest->inactconns);
541 }
542
543 /*
544  *      Bind a connection entry with a virtual service destination
545  *      Called just after a new connection entry is created.
546  */
547 static inline void
548 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
549 {
550         unsigned int conn_flags;
551
552         /* if dest is NULL, then return directly */
553         if (!dest)
554                 return;
555
556         /* Increase the refcnt counter of the dest */
557         atomic_inc(&dest->refcnt);
558
559         conn_flags = atomic_read(&dest->conn_flags);
560         if (cp->protocol != IPPROTO_UDP)
561                 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
562         /* Bind with the destination and its corresponding transmitter */
563         if (cp->flags & IP_VS_CONN_F_SYNC) {
564                 /* if the connection is not template and is created
565                  * by sync, preserve the activity flag.
566                  */
567                 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE))
568                         conn_flags &= ~IP_VS_CONN_F_INACTIVE;
569                 /* connections inherit forwarding method from dest */
570                 cp->flags &= ~IP_VS_CONN_F_FWD_MASK;
571         }
572         cp->flags |= conn_flags;
573         cp->dest = dest;
574
575         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
576                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
577                       "dest->refcnt:%d\n",
578                       ip_vs_proto_name(cp->protocol),
579                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
580                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
581                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
582                       ip_vs_fwd_tag(cp), cp->state,
583                       cp->flags, atomic_read(&cp->refcnt),
584                       atomic_read(&dest->refcnt));
585
586         /* Update the connection counters */
587         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
588                 /* It is a normal connection, so increase the inactive
589                    connection counter because it is in TCP SYNRECV
590                    state (inactive) or other protocol inacive state */
591                 if ((cp->flags & IP_VS_CONN_F_SYNC) &&
592                     (!(cp->flags & IP_VS_CONN_F_INACTIVE)))
593                         atomic_inc(&dest->activeconns);
594                 else
595                         atomic_inc(&dest->inactconns);
596         } else {
597                 /* It is a persistent connection/template, so increase
598                    the persistent connection counter */
599                 atomic_inc(&dest->persistconns);
600         }
601
602         if (dest->u_threshold != 0 &&
603             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
604                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
605 }
606
607
608 /*
609  * Check if there is a destination for the connection, if so
610  * bind the connection to the destination.
611  */
612 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
613 {
614         struct ip_vs_dest *dest;
615
616         if ((cp) && (!cp->dest)) {
617                 dest = ip_vs_find_dest(ip_vs_conn_net(cp), cp->af, &cp->daddr,
618                                        cp->dport, &cp->vaddr, cp->vport,
619                                        cp->protocol, cp->fwmark);
620                 ip_vs_bind_dest(cp, dest);
621                 return dest;
622         } else
623                 return NULL;
624 }
625
626
627 /*
628  *      Unbind a connection entry with its VS destination
629  *      Called by the ip_vs_conn_expire function.
630  */
631 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
632 {
633         struct ip_vs_dest *dest = cp->dest;
634
635         if (!dest)
636                 return;
637
638         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
639                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
640                       "dest->refcnt:%d\n",
641                       ip_vs_proto_name(cp->protocol),
642                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
643                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
644                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
645                       ip_vs_fwd_tag(cp), cp->state,
646                       cp->flags, atomic_read(&cp->refcnt),
647                       atomic_read(&dest->refcnt));
648
649         /* Update the connection counters */
650         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
651                 /* It is a normal connection, so decrease the inactconns
652                    or activeconns counter */
653                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
654                         atomic_dec(&dest->inactconns);
655                 } else {
656                         atomic_dec(&dest->activeconns);
657                 }
658         } else {
659                 /* It is a persistent connection/template, so decrease
660                    the persistent connection counter */
661                 atomic_dec(&dest->persistconns);
662         }
663
664         if (dest->l_threshold != 0) {
665                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
666                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
667         } else if (dest->u_threshold != 0) {
668                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
669                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
670         } else {
671                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
672                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
673         }
674
675         /*
676          * Simply decrease the refcnt of the dest, because the
677          * dest will be either in service's destination list
678          * or in the trash.
679          */
680         atomic_dec(&dest->refcnt);
681 }
682
683 static int expire_quiescent_template(struct netns_ipvs *ipvs,
684                                      struct ip_vs_dest *dest)
685 {
686 #ifdef CONFIG_SYSCTL
687         return ipvs->sysctl_expire_quiescent_template &&
688                 (atomic_read(&dest->weight) == 0);
689 #else
690         return 0;
691 #endif
692 }
693
694 /*
695  *      Checking if the destination of a connection template is available.
696  *      If available, return 1, otherwise invalidate this connection
697  *      template and return 0.
698  */
699 int ip_vs_check_template(struct ip_vs_conn *ct)
700 {
701         struct ip_vs_dest *dest = ct->dest;
702         struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(ct));
703
704         /*
705          * Checking the dest server status.
706          */
707         if ((dest == NULL) ||
708             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
709             expire_quiescent_template(ipvs, dest)) {
710                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
711                               "protocol %s s:%s:%d v:%s:%d "
712                               "-> d:%s:%d\n",
713                               ip_vs_proto_name(ct->protocol),
714                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
715                               ntohs(ct->cport),
716                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
717                               ntohs(ct->vport),
718                               IP_VS_DBG_ADDR(ct->af, &ct->daddr),
719                               ntohs(ct->dport));
720
721                 /*
722                  * Invalidate the connection template
723                  */
724                 if (ct->vport != htons(0xffff)) {
725                         if (ip_vs_conn_unhash(ct)) {
726                                 ct->dport = htons(0xffff);
727                                 ct->vport = htons(0xffff);
728                                 ct->cport = 0;
729                                 ip_vs_conn_hash(ct);
730                         }
731                 }
732
733                 /*
734                  * Simply decrease the refcnt of the template,
735                  * don't restart its timer.
736                  */
737                 atomic_dec(&ct->refcnt);
738                 return 0;
739         }
740         return 1;
741 }
742
743 static void ip_vs_conn_expire(unsigned long data)
744 {
745         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
746         struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(cp));
747
748         cp->timeout = 60*HZ;
749
750         /*
751          *      hey, I'm using it
752          */
753         atomic_inc(&cp->refcnt);
754
755         /*
756          *      do I control anybody?
757          */
758         if (atomic_read(&cp->n_control))
759                 goto expire_later;
760
761         /*
762          *      unhash it if it is hashed in the conn table
763          */
764         if (!ip_vs_conn_unhash(cp) && !(cp->flags & IP_VS_CONN_F_ONE_PACKET))
765                 goto expire_later;
766
767         /*
768          *      refcnt==1 implies I'm the only one referrer
769          */
770         if (likely(atomic_read(&cp->refcnt) == 1)) {
771                 /* delete the timer if it is activated by other users */
772                 if (timer_pending(&cp->timer))
773                         del_timer(&cp->timer);
774
775                 /* does anybody control me? */
776                 if (cp->control)
777                         ip_vs_control_del(cp);
778
779                 if (cp->flags & IP_VS_CONN_F_NFCT)
780                         ip_vs_conn_drop_conntrack(cp);
781
782                 ip_vs_pe_put(cp->pe);
783                 kfree(cp->pe_data);
784                 if (unlikely(cp->app != NULL))
785                         ip_vs_unbind_app(cp);
786                 ip_vs_unbind_dest(cp);
787                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
788                         atomic_dec(&ip_vs_conn_no_cport_cnt);
789                 atomic_dec(&ipvs->conn_count);
790
791                 kmem_cache_free(ip_vs_conn_cachep, cp);
792                 return;
793         }
794
795         /* hash it back to the table */
796         ip_vs_conn_hash(cp);
797
798   expire_later:
799         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
800                   atomic_read(&cp->refcnt)-1,
801                   atomic_read(&cp->n_control));
802
803         ip_vs_conn_put(cp);
804 }
805
806
807 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
808 {
809         if (del_timer(&cp->timer))
810                 mod_timer(&cp->timer, jiffies);
811 }
812
813
814 /*
815  *      Create a new connection entry and hash it into the ip_vs_conn_tab
816  */
817 struct ip_vs_conn *
818 ip_vs_conn_new(const struct ip_vs_conn_param *p,
819                const union nf_inet_addr *daddr, __be16 dport, unsigned flags,
820                struct ip_vs_dest *dest, __u32 fwmark)
821 {
822         struct ip_vs_conn *cp;
823         struct netns_ipvs *ipvs = net_ipvs(p->net);
824         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->net,
825                                                            p->protocol);
826
827         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
828         if (cp == NULL) {
829                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
830                 return NULL;
831         }
832
833         INIT_HLIST_NODE(&cp->c_list);
834         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
835         ip_vs_conn_net_set(cp, p->net);
836         cp->af             = p->af;
837         cp->protocol       = p->protocol;
838         ip_vs_addr_copy(p->af, &cp->caddr, p->caddr);
839         cp->cport          = p->cport;
840         ip_vs_addr_copy(p->af, &cp->vaddr, p->vaddr);
841         cp->vport          = p->vport;
842         /* proto should only be IPPROTO_IP if d_addr is a fwmark */
843         ip_vs_addr_copy(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
844                         &cp->daddr, daddr);
845         cp->dport          = dport;
846         cp->flags          = flags;
847         cp->fwmark         = fwmark;
848         if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
849                 ip_vs_pe_get(p->pe);
850                 cp->pe = p->pe;
851                 cp->pe_data = p->pe_data;
852                 cp->pe_data_len = p->pe_data_len;
853         }
854         spin_lock_init(&cp->lock);
855
856         /*
857          * Set the entry is referenced by the current thread before hashing
858          * it in the table, so that other thread run ip_vs_random_dropentry
859          * but cannot drop this entry.
860          */
861         atomic_set(&cp->refcnt, 1);
862
863         atomic_set(&cp->n_control, 0);
864         atomic_set(&cp->in_pkts, 0);
865
866         atomic_inc(&ipvs->conn_count);
867         if (flags & IP_VS_CONN_F_NO_CPORT)
868                 atomic_inc(&ip_vs_conn_no_cport_cnt);
869
870         /* Bind the connection with a destination server */
871         ip_vs_bind_dest(cp, dest);
872
873         /* Set its state and timeout */
874         cp->state = 0;
875         cp->timeout = 3*HZ;
876
877         /* Bind its packet transmitter */
878 #ifdef CONFIG_IP_VS_IPV6
879         if (p->af == AF_INET6)
880                 ip_vs_bind_xmit_v6(cp);
881         else
882 #endif
883                 ip_vs_bind_xmit(cp);
884
885         if (unlikely(pd && atomic_read(&pd->appcnt)))
886                 ip_vs_bind_app(cp, pd->pp);
887
888         /*
889          * Allow conntrack to be preserved. By default, conntrack
890          * is created and destroyed for every packet.
891          * Sometimes keeping conntrack can be useful for
892          * IP_VS_CONN_F_ONE_PACKET too.
893          */
894
895         if (ip_vs_conntrack_enabled(ipvs))
896                 cp->flags |= IP_VS_CONN_F_NFCT;
897
898         /* Hash it in the ip_vs_conn_tab finally */
899         ip_vs_conn_hash(cp);
900
901         return cp;
902 }
903
904 /*
905  *      /proc/net/ip_vs_conn entries
906  */
907 #ifdef CONFIG_PROC_FS
908 struct ip_vs_iter_state {
909         struct seq_net_private  p;
910         struct hlist_head       *l;
911 };
912
913 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
914 {
915         int idx;
916         struct ip_vs_conn *cp;
917         struct ip_vs_iter_state *iter = seq->private;
918         struct hlist_node *n;
919
920         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
921                 ct_read_lock_bh(idx);
922                 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[idx], c_list) {
923                         if (pos-- == 0) {
924                                 iter->l = &ip_vs_conn_tab[idx];
925                                 return cp;
926                         }
927                 }
928                 ct_read_unlock_bh(idx);
929         }
930
931         return NULL;
932 }
933
934 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
935 {
936         struct ip_vs_iter_state *iter = seq->private;
937
938         iter->l = NULL;
939         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
940 }
941
942 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
943 {
944         struct ip_vs_conn *cp = v;
945         struct ip_vs_iter_state *iter = seq->private;
946         struct hlist_node *e;
947         struct hlist_head *l = iter->l;
948         int idx;
949
950         ++*pos;
951         if (v == SEQ_START_TOKEN)
952                 return ip_vs_conn_array(seq, 0);
953
954         /* more on same hash chain? */
955         if ((e = cp->c_list.next))
956                 return hlist_entry(e, struct ip_vs_conn, c_list);
957
958         idx = l - ip_vs_conn_tab;
959         ct_read_unlock_bh(idx);
960
961         while (++idx < ip_vs_conn_tab_size) {
962                 ct_read_lock_bh(idx);
963                 hlist_for_each_entry(cp, e, &ip_vs_conn_tab[idx], c_list) {
964                         iter->l = &ip_vs_conn_tab[idx];
965                         return cp;
966                 }
967                 ct_read_unlock_bh(idx);
968         }
969         iter->l = NULL;
970         return NULL;
971 }
972
973 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
974 {
975         struct ip_vs_iter_state *iter = seq->private;
976         struct hlist_head *l = iter->l;
977
978         if (l)
979                 ct_read_unlock_bh(l - ip_vs_conn_tab);
980 }
981
982 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
983 {
984
985         if (v == SEQ_START_TOKEN)
986                 seq_puts(seq,
987    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires PEName PEData\n");
988         else {
989                 const struct ip_vs_conn *cp = v;
990                 struct net *net = seq_file_net(seq);
991                 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
992                 size_t len = 0;
993
994                 if (!ip_vs_conn_net_eq(cp, net))
995                         return 0;
996                 if (cp->pe_data) {
997                         pe_data[0] = ' ';
998                         len = strlen(cp->pe->name);
999                         memcpy(pe_data + 1, cp->pe->name, len);
1000                         pe_data[len + 1] = ' ';
1001                         len += 2;
1002                         len += cp->pe->show_pe_data(cp, pe_data + len);
1003                 }
1004                 pe_data[len] = '\0';
1005
1006 #ifdef CONFIG_IP_VS_IPV6
1007                 if (cp->af == AF_INET6)
1008                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1009                                 "%pI6 %04X %-11s %7lu%s\n",
1010                                 ip_vs_proto_name(cp->protocol),
1011                                 &cp->caddr.in6, ntohs(cp->cport),
1012                                 &cp->vaddr.in6, ntohs(cp->vport),
1013                                 &cp->daddr.in6, ntohs(cp->dport),
1014                                 ip_vs_state_name(cp->protocol, cp->state),
1015                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1016                 else
1017 #endif
1018                         seq_printf(seq,
1019                                 "%-3s %08X %04X %08X %04X"
1020                                 " %08X %04X %-11s %7lu%s\n",
1021                                 ip_vs_proto_name(cp->protocol),
1022                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1023                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1024                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
1025                                 ip_vs_state_name(cp->protocol, cp->state),
1026                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1027         }
1028         return 0;
1029 }
1030
1031 static const struct seq_operations ip_vs_conn_seq_ops = {
1032         .start = ip_vs_conn_seq_start,
1033         .next  = ip_vs_conn_seq_next,
1034         .stop  = ip_vs_conn_seq_stop,
1035         .show  = ip_vs_conn_seq_show,
1036 };
1037
1038 static int ip_vs_conn_open(struct inode *inode, struct file *file)
1039 {
1040         return seq_open_net(inode, file, &ip_vs_conn_seq_ops,
1041                             sizeof(struct ip_vs_iter_state));
1042 }
1043
1044 static const struct file_operations ip_vs_conn_fops = {
1045         .owner   = THIS_MODULE,
1046         .open    = ip_vs_conn_open,
1047         .read    = seq_read,
1048         .llseek  = seq_lseek,
1049         .release = seq_release,
1050 };
1051
1052 static const char *ip_vs_origin_name(unsigned flags)
1053 {
1054         if (flags & IP_VS_CONN_F_SYNC)
1055                 return "SYNC";
1056         else
1057                 return "LOCAL";
1058 }
1059
1060 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1061 {
1062
1063         if (v == SEQ_START_TOKEN)
1064                 seq_puts(seq,
1065    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
1066         else {
1067                 const struct ip_vs_conn *cp = v;
1068                 struct net *net = seq_file_net(seq);
1069
1070                 if (!ip_vs_conn_net_eq(cp, net))
1071                         return 0;
1072
1073 #ifdef CONFIG_IP_VS_IPV6
1074                 if (cp->af == AF_INET6)
1075                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
1076                                 ip_vs_proto_name(cp->protocol),
1077                                 &cp->caddr.in6, ntohs(cp->cport),
1078                                 &cp->vaddr.in6, ntohs(cp->vport),
1079                                 &cp->daddr.in6, ntohs(cp->dport),
1080                                 ip_vs_state_name(cp->protocol, cp->state),
1081                                 ip_vs_origin_name(cp->flags),
1082                                 (cp->timer.expires-jiffies)/HZ);
1083                 else
1084 #endif
1085                         seq_printf(seq,
1086                                 "%-3s %08X %04X %08X %04X "
1087                                 "%08X %04X %-11s %-6s %7lu\n",
1088                                 ip_vs_proto_name(cp->protocol),
1089                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1090                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1091                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
1092                                 ip_vs_state_name(cp->protocol, cp->state),
1093                                 ip_vs_origin_name(cp->flags),
1094                                 (cp->timer.expires-jiffies)/HZ);
1095         }
1096         return 0;
1097 }
1098
1099 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1100         .start = ip_vs_conn_seq_start,
1101         .next  = ip_vs_conn_seq_next,
1102         .stop  = ip_vs_conn_seq_stop,
1103         .show  = ip_vs_conn_sync_seq_show,
1104 };
1105
1106 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
1107 {
1108         return seq_open_net(inode, file, &ip_vs_conn_sync_seq_ops,
1109                             sizeof(struct ip_vs_iter_state));
1110 }
1111
1112 static const struct file_operations ip_vs_conn_sync_fops = {
1113         .owner   = THIS_MODULE,
1114         .open    = ip_vs_conn_sync_open,
1115         .read    = seq_read,
1116         .llseek  = seq_lseek,
1117         .release = seq_release,
1118 };
1119
1120 #endif
1121
1122
1123 /*
1124  *      Randomly drop connection entries before running out of memory
1125  */
1126 static inline int todrop_entry(struct ip_vs_conn *cp)
1127 {
1128         /*
1129          * The drop rate array needs tuning for real environments.
1130          * Called from timer bh only => no locking
1131          */
1132         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1133         static char todrop_counter[9] = {0};
1134         int i;
1135
1136         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1137            This will leave enough time for normal connection to get
1138            through. */
1139         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1140                 return 0;
1141
1142         /* Don't drop the entry if its number of incoming packets is not
1143            located in [0, 8] */
1144         i = atomic_read(&cp->in_pkts);
1145         if (i > 8 || i < 0) return 0;
1146
1147         if (!todrop_rate[i]) return 0;
1148         if (--todrop_counter[i] > 0) return 0;
1149
1150         todrop_counter[i] = todrop_rate[i];
1151         return 1;
1152 }
1153
1154 /* Called from keventd and must protect itself from softirqs */
1155 void ip_vs_random_dropentry(struct net *net)
1156 {
1157         int idx;
1158         struct ip_vs_conn *cp;
1159
1160         /*
1161          * Randomly scan 1/32 of the whole table every second
1162          */
1163         for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1164                 unsigned hash = net_random() & ip_vs_conn_tab_mask;
1165                 struct hlist_node *n;
1166
1167                 /*
1168                  *  Lock is actually needed in this loop.
1169                  */
1170                 ct_write_lock_bh(hash);
1171
1172                 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
1173                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1174                                 /* connection template */
1175                                 continue;
1176                         if (!ip_vs_conn_net_eq(cp, net))
1177                                 continue;
1178                         if (cp->protocol == IPPROTO_TCP) {
1179                                 switch(cp->state) {
1180                                 case IP_VS_TCP_S_SYN_RECV:
1181                                 case IP_VS_TCP_S_SYNACK:
1182                                         break;
1183
1184                                 case IP_VS_TCP_S_ESTABLISHED:
1185                                         if (todrop_entry(cp))
1186                                                 break;
1187                                         continue;
1188
1189                                 default:
1190                                         continue;
1191                                 }
1192                         } else {
1193                                 if (!todrop_entry(cp))
1194                                         continue;
1195                         }
1196
1197                         IP_VS_DBG(4, "del connection\n");
1198                         ip_vs_conn_expire_now(cp);
1199                         if (cp->control) {
1200                                 IP_VS_DBG(4, "del conn template\n");
1201                                 ip_vs_conn_expire_now(cp->control);
1202                         }
1203                 }
1204                 ct_write_unlock_bh(hash);
1205         }
1206 }
1207
1208
1209 /*
1210  *      Flush all the connection entries in the ip_vs_conn_tab
1211  */
1212 static void ip_vs_conn_flush(struct net *net)
1213 {
1214         int idx;
1215         struct ip_vs_conn *cp;
1216         struct netns_ipvs *ipvs = net_ipvs(net);
1217
1218 flush_again:
1219         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1220                 struct hlist_node *n;
1221
1222                 /*
1223                  *  Lock is actually needed in this loop.
1224                  */
1225                 ct_write_lock_bh(idx);
1226
1227                 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[idx], c_list) {
1228                         if (!ip_vs_conn_net_eq(cp, net))
1229                                 continue;
1230                         IP_VS_DBG(4, "del connection\n");
1231                         ip_vs_conn_expire_now(cp);
1232                         if (cp->control) {
1233                                 IP_VS_DBG(4, "del conn template\n");
1234                                 ip_vs_conn_expire_now(cp->control);
1235                         }
1236                 }
1237                 ct_write_unlock_bh(idx);
1238         }
1239
1240         /* the counter may be not NULL, because maybe some conn entries
1241            are run by slow timer handler or unhashed but still referred */
1242         if (atomic_read(&ipvs->conn_count) != 0) {
1243                 schedule();
1244                 goto flush_again;
1245         }
1246 }
1247 /*
1248  * per netns init and exit
1249  */
1250 int __net_init __ip_vs_conn_init(struct net *net)
1251 {
1252         struct netns_ipvs *ipvs = net_ipvs(net);
1253
1254         atomic_set(&ipvs->conn_count, 0);
1255
1256         proc_net_fops_create(net, "ip_vs_conn", 0, &ip_vs_conn_fops);
1257         proc_net_fops_create(net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops);
1258         return 0;
1259 }
1260
1261 static void __net_exit __ip_vs_conn_cleanup(struct net *net)
1262 {
1263         /* flush all the connection entries first */
1264         ip_vs_conn_flush(net);
1265         proc_net_remove(net, "ip_vs_conn");
1266         proc_net_remove(net, "ip_vs_conn_sync");
1267 }
1268 static struct pernet_operations ipvs_conn_ops = {
1269         .init = __ip_vs_conn_init,
1270         .exit = __ip_vs_conn_cleanup,
1271 };
1272
1273 int __init ip_vs_conn_init(void)
1274 {
1275         int idx;
1276         int retc;
1277
1278         /* Compute size and mask */
1279         ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1280         ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1281
1282         /*
1283          * Allocate the connection hash table and initialize its list heads
1284          */
1285         ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size * sizeof(*ip_vs_conn_tab));
1286         if (!ip_vs_conn_tab)
1287                 return -ENOMEM;
1288
1289         /* Allocate ip_vs_conn slab cache */
1290         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1291                                               sizeof(struct ip_vs_conn), 0,
1292                                               SLAB_HWCACHE_ALIGN, NULL);
1293         if (!ip_vs_conn_cachep) {
1294                 vfree(ip_vs_conn_tab);
1295                 return -ENOMEM;
1296         }
1297
1298         pr_info("Connection hash table configured "
1299                 "(size=%d, memory=%ldKbytes)\n",
1300                 ip_vs_conn_tab_size,
1301                 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1302         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1303                   sizeof(struct ip_vs_conn));
1304
1305         for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1306                 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1307
1308         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1309                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1310         }
1311
1312         retc = register_pernet_subsys(&ipvs_conn_ops);
1313
1314         /* calculate the random value for connection hash */
1315         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1316
1317         return retc;
1318 }
1319
1320 void ip_vs_conn_cleanup(void)
1321 {
1322         unregister_pernet_subsys(&ipvs_conn_ops);
1323         /* Release the empty cache */
1324         kmem_cache_destroy(ip_vs_conn_cachep);
1325         vfree(ip_vs_conn_tab);
1326 }