ipvs: fix synchronization on connection close
[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/seq_file.h>
36 #include <linux/jhash.h>
37 #include <linux/random.h>
38
39 #include <net/net_namespace.h>
40 #include <net/ip_vs.h>
41
42
43 /*
44  *  Connection hash table: for input and output packets lookups of IPVS
45  */
46 static struct list_head *ip_vs_conn_tab;
47
48 /*  SLAB cache for IPVS connections */
49 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
50
51 /*  counter for current IPVS connections */
52 static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
53
54 /*  counter for no client port connections */
55 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
56
57 /* random value for IPVS connection hash */
58 static unsigned int ip_vs_conn_rnd;
59
60 /*
61  *  Fine locking granularity for big connection hash table
62  */
63 #define CT_LOCKARRAY_BITS  4
64 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
65 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
66
67 struct ip_vs_aligned_lock
68 {
69         rwlock_t        l;
70 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
71
72 /* lock array for conn table */
73 static struct ip_vs_aligned_lock
74 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
75
76 static inline void ct_read_lock(unsigned key)
77 {
78         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
79 }
80
81 static inline void ct_read_unlock(unsigned key)
82 {
83         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
84 }
85
86 static inline void ct_write_lock(unsigned key)
87 {
88         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
89 }
90
91 static inline void ct_write_unlock(unsigned key)
92 {
93         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
94 }
95
96 static inline void ct_read_lock_bh(unsigned key)
97 {
98         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
99 }
100
101 static inline void ct_read_unlock_bh(unsigned key)
102 {
103         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
104 }
105
106 static inline void ct_write_lock_bh(unsigned key)
107 {
108         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
109 }
110
111 static inline void ct_write_unlock_bh(unsigned key)
112 {
113         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
114 }
115
116
117 /*
118  *      Returns hash value for IPVS connection entry
119  */
120 static unsigned int ip_vs_conn_hashkey(int af, unsigned proto,
121                                        const union nf_inet_addr *addr,
122                                        __be16 port)
123 {
124 #ifdef CONFIG_IP_VS_IPV6
125         if (af == AF_INET6)
126                 return jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
127                                     (__force u32)port, proto, ip_vs_conn_rnd)
128                         & IP_VS_CONN_TAB_MASK;
129 #endif
130         return jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
131                             ip_vs_conn_rnd)
132                 & IP_VS_CONN_TAB_MASK;
133 }
134
135
136 /*
137  *      Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
138  *      returns bool success.
139  */
140 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
141 {
142         unsigned hash;
143         int ret;
144
145         /* Hash by protocol, client address and port */
146         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
147
148         ct_write_lock(hash);
149
150         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
151                 list_add(&cp->c_list, &ip_vs_conn_tab[hash]);
152                 cp->flags |= IP_VS_CONN_F_HASHED;
153                 atomic_inc(&cp->refcnt);
154                 ret = 1;
155         } else {
156                 pr_err("%s(): request for already hashed, called from %pF\n",
157                        __func__, __builtin_return_address(0));
158                 ret = 0;
159         }
160
161         ct_write_unlock(hash);
162
163         return ret;
164 }
165
166
167 /*
168  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
169  *      returns bool success.
170  */
171 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
172 {
173         unsigned hash;
174         int ret;
175
176         /* unhash it and decrease its reference counter */
177         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
178
179         ct_write_lock(hash);
180
181         if (cp->flags & IP_VS_CONN_F_HASHED) {
182                 list_del(&cp->c_list);
183                 cp->flags &= ~IP_VS_CONN_F_HASHED;
184                 atomic_dec(&cp->refcnt);
185                 ret = 1;
186         } else
187                 ret = 0;
188
189         ct_write_unlock(hash);
190
191         return ret;
192 }
193
194
195 /*
196  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
197  *  Called for pkts coming from OUTside-to-INside.
198  *      s_addr, s_port: pkt source address (foreign host)
199  *      d_addr, d_port: pkt dest address (load balancer)
200  */
201 static inline struct ip_vs_conn *__ip_vs_conn_in_get
202 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
203  const union nf_inet_addr *d_addr, __be16 d_port)
204 {
205         unsigned hash;
206         struct ip_vs_conn *cp;
207
208         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
209
210         ct_read_lock(hash);
211
212         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
213                 if (cp->af == af &&
214                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
215                     ip_vs_addr_equal(af, d_addr, &cp->vaddr) &&
216                     s_port == cp->cport && d_port == cp->vport &&
217                     ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
218                     protocol == cp->protocol) {
219                         /* HIT */
220                         atomic_inc(&cp->refcnt);
221                         ct_read_unlock(hash);
222                         return cp;
223                 }
224         }
225
226         ct_read_unlock(hash);
227
228         return NULL;
229 }
230
231 struct ip_vs_conn *ip_vs_conn_in_get
232 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
233  const union nf_inet_addr *d_addr, __be16 d_port)
234 {
235         struct ip_vs_conn *cp;
236
237         cp = __ip_vs_conn_in_get(af, protocol, s_addr, s_port, d_addr, d_port);
238         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt))
239                 cp = __ip_vs_conn_in_get(af, protocol, s_addr, 0, d_addr,
240                                          d_port);
241
242         IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
243                       ip_vs_proto_name(protocol),
244                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
245                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
246                       cp ? "hit" : "not hit");
247
248         return cp;
249 }
250
251 /* Get reference to connection template */
252 struct ip_vs_conn *ip_vs_ct_in_get
253 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
254  const union nf_inet_addr *d_addr, __be16 d_port)
255 {
256         unsigned hash;
257         struct ip_vs_conn *cp;
258
259         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
260
261         ct_read_lock(hash);
262
263         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
264                 if (cp->af == af &&
265                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
266                     /* protocol should only be IPPROTO_IP if
267                      * d_addr is a fwmark */
268                     ip_vs_addr_equal(protocol == IPPROTO_IP ? AF_UNSPEC : af,
269                                      d_addr, &cp->vaddr) &&
270                     s_port == cp->cport && d_port == cp->vport &&
271                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
272                     protocol == cp->protocol) {
273                         /* HIT */
274                         atomic_inc(&cp->refcnt);
275                         goto out;
276                 }
277         }
278         cp = NULL;
279
280   out:
281         ct_read_unlock(hash);
282
283         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
284                       ip_vs_proto_name(protocol),
285                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
286                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
287                       cp ? "hit" : "not hit");
288
289         return cp;
290 }
291
292 /*
293  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
294  *  Called for pkts coming from inside-to-OUTside.
295  *      s_addr, s_port: pkt source address (inside host)
296  *      d_addr, d_port: pkt dest address (foreign host)
297  */
298 struct ip_vs_conn *ip_vs_conn_out_get
299 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
300  const union nf_inet_addr *d_addr, __be16 d_port)
301 {
302         unsigned hash;
303         struct ip_vs_conn *cp, *ret=NULL;
304
305         /*
306          *      Check for "full" addressed entries
307          */
308         hash = ip_vs_conn_hashkey(af, protocol, d_addr, d_port);
309
310         ct_read_lock(hash);
311
312         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
313                 if (cp->af == af &&
314                     ip_vs_addr_equal(af, d_addr, &cp->caddr) &&
315                     ip_vs_addr_equal(af, s_addr, &cp->daddr) &&
316                     d_port == cp->cport && s_port == cp->dport &&
317                     protocol == cp->protocol) {
318                         /* HIT */
319                         atomic_inc(&cp->refcnt);
320                         ret = cp;
321                         break;
322                 }
323         }
324
325         ct_read_unlock(hash);
326
327         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
328                       ip_vs_proto_name(protocol),
329                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
330                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
331                       ret ? "hit" : "not hit");
332
333         return ret;
334 }
335
336
337 /*
338  *      Put back the conn and restart its timer with its timeout
339  */
340 void ip_vs_conn_put(struct ip_vs_conn *cp)
341 {
342         /* reset it expire in its timeout */
343         mod_timer(&cp->timer, jiffies+cp->timeout);
344
345         __ip_vs_conn_put(cp);
346 }
347
348
349 /*
350  *      Fill a no_client_port connection with a client port number
351  */
352 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
353 {
354         if (ip_vs_conn_unhash(cp)) {
355                 spin_lock(&cp->lock);
356                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
357                         atomic_dec(&ip_vs_conn_no_cport_cnt);
358                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
359                         cp->cport = cport;
360                 }
361                 spin_unlock(&cp->lock);
362
363                 /* hash on new dport */
364                 ip_vs_conn_hash(cp);
365         }
366 }
367
368
369 /*
370  *      Bind a connection entry with the corresponding packet_xmit.
371  *      Called by ip_vs_conn_new.
372  */
373 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
374 {
375         switch (IP_VS_FWD_METHOD(cp)) {
376         case IP_VS_CONN_F_MASQ:
377                 cp->packet_xmit = ip_vs_nat_xmit;
378                 break;
379
380         case IP_VS_CONN_F_TUNNEL:
381                 cp->packet_xmit = ip_vs_tunnel_xmit;
382                 break;
383
384         case IP_VS_CONN_F_DROUTE:
385                 cp->packet_xmit = ip_vs_dr_xmit;
386                 break;
387
388         case IP_VS_CONN_F_LOCALNODE:
389                 cp->packet_xmit = ip_vs_null_xmit;
390                 break;
391
392         case IP_VS_CONN_F_BYPASS:
393                 cp->packet_xmit = ip_vs_bypass_xmit;
394                 break;
395         }
396 }
397
398 #ifdef CONFIG_IP_VS_IPV6
399 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
400 {
401         switch (IP_VS_FWD_METHOD(cp)) {
402         case IP_VS_CONN_F_MASQ:
403                 cp->packet_xmit = ip_vs_nat_xmit_v6;
404                 break;
405
406         case IP_VS_CONN_F_TUNNEL:
407                 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
408                 break;
409
410         case IP_VS_CONN_F_DROUTE:
411                 cp->packet_xmit = ip_vs_dr_xmit_v6;
412                 break;
413
414         case IP_VS_CONN_F_LOCALNODE:
415                 cp->packet_xmit = ip_vs_null_xmit;
416                 break;
417
418         case IP_VS_CONN_F_BYPASS:
419                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
420                 break;
421         }
422 }
423 #endif
424
425
426 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
427 {
428         return atomic_read(&dest->activeconns)
429                 + atomic_read(&dest->inactconns);
430 }
431
432 /*
433  *      Bind a connection entry with a virtual service destination
434  *      Called just after a new connection entry is created.
435  */
436 static inline void
437 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
438 {
439         /* if dest is NULL, then return directly */
440         if (!dest)
441                 return;
442
443         /* Increase the refcnt counter of the dest */
444         atomic_inc(&dest->refcnt);
445
446         /* Bind with the destination and its corresponding transmitter */
447         if ((cp->flags & IP_VS_CONN_F_SYNC) &&
448             (!(cp->flags & IP_VS_CONN_F_TEMPLATE)))
449                 /* if the connection is not template and is created
450                  * by sync, preserve the activity flag.
451                  */
452                 cp->flags |= atomic_read(&dest->conn_flags) &
453                              (~IP_VS_CONN_F_INACTIVE);
454         else
455                 cp->flags |= atomic_read(&dest->conn_flags);
456         cp->dest = dest;
457
458         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
459                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
460                       "dest->refcnt:%d\n",
461                       ip_vs_proto_name(cp->protocol),
462                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
463                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
464                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
465                       ip_vs_fwd_tag(cp), cp->state,
466                       cp->flags, atomic_read(&cp->refcnt),
467                       atomic_read(&dest->refcnt));
468
469         /* Update the connection counters */
470         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
471                 /* It is a normal connection, so increase the inactive
472                    connection counter because it is in TCP SYNRECV
473                    state (inactive) or other protocol inacive state */
474                 if ((cp->flags & IP_VS_CONN_F_SYNC) &&
475                     (!(cp->flags & IP_VS_CONN_F_INACTIVE)))
476                         atomic_inc(&dest->activeconns);
477                 else
478                         atomic_inc(&dest->inactconns);
479         } else {
480                 /* It is a persistent connection/template, so increase
481                    the peristent connection counter */
482                 atomic_inc(&dest->persistconns);
483         }
484
485         if (dest->u_threshold != 0 &&
486             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
487                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
488 }
489
490
491 /*
492  * Check if there is a destination for the connection, if so
493  * bind the connection to the destination.
494  */
495 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
496 {
497         struct ip_vs_dest *dest;
498
499         if ((cp) && (!cp->dest)) {
500                 dest = ip_vs_find_dest(cp->af, &cp->daddr, cp->dport,
501                                        &cp->vaddr, cp->vport,
502                                        cp->protocol);
503                 ip_vs_bind_dest(cp, dest);
504                 return dest;
505         } else
506                 return NULL;
507 }
508
509
510 /*
511  *      Unbind a connection entry with its VS destination
512  *      Called by the ip_vs_conn_expire function.
513  */
514 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
515 {
516         struct ip_vs_dest *dest = cp->dest;
517
518         if (!dest)
519                 return;
520
521         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
522                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
523                       "dest->refcnt:%d\n",
524                       ip_vs_proto_name(cp->protocol),
525                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
526                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
527                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
528                       ip_vs_fwd_tag(cp), cp->state,
529                       cp->flags, atomic_read(&cp->refcnt),
530                       atomic_read(&dest->refcnt));
531
532         /* Update the connection counters */
533         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
534                 /* It is a normal connection, so decrease the inactconns
535                    or activeconns counter */
536                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
537                         atomic_dec(&dest->inactconns);
538                 } else {
539                         atomic_dec(&dest->activeconns);
540                 }
541         } else {
542                 /* It is a persistent connection/template, so decrease
543                    the peristent connection counter */
544                 atomic_dec(&dest->persistconns);
545         }
546
547         if (dest->l_threshold != 0) {
548                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
549                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
550         } else if (dest->u_threshold != 0) {
551                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
552                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
553         } else {
554                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
555                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
556         }
557
558         /*
559          * Simply decrease the refcnt of the dest, because the
560          * dest will be either in service's destination list
561          * or in the trash.
562          */
563         atomic_dec(&dest->refcnt);
564 }
565
566
567 /*
568  *      Checking if the destination of a connection template is available.
569  *      If available, return 1, otherwise invalidate this connection
570  *      template and return 0.
571  */
572 int ip_vs_check_template(struct ip_vs_conn *ct)
573 {
574         struct ip_vs_dest *dest = ct->dest;
575
576         /*
577          * Checking the dest server status.
578          */
579         if ((dest == NULL) ||
580             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
581             (sysctl_ip_vs_expire_quiescent_template &&
582              (atomic_read(&dest->weight) == 0))) {
583                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
584                               "protocol %s s:%s:%d v:%s:%d "
585                               "-> d:%s:%d\n",
586                               ip_vs_proto_name(ct->protocol),
587                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
588                               ntohs(ct->cport),
589                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
590                               ntohs(ct->vport),
591                               IP_VS_DBG_ADDR(ct->af, &ct->daddr),
592                               ntohs(ct->dport));
593
594                 /*
595                  * Invalidate the connection template
596                  */
597                 if (ct->vport != htons(0xffff)) {
598                         if (ip_vs_conn_unhash(ct)) {
599                                 ct->dport = htons(0xffff);
600                                 ct->vport = htons(0xffff);
601                                 ct->cport = 0;
602                                 ip_vs_conn_hash(ct);
603                         }
604                 }
605
606                 /*
607                  * Simply decrease the refcnt of the template,
608                  * don't restart its timer.
609                  */
610                 atomic_dec(&ct->refcnt);
611                 return 0;
612         }
613         return 1;
614 }
615
616 static void ip_vs_conn_expire(unsigned long data)
617 {
618         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
619
620         cp->timeout = 60*HZ;
621
622         /*
623          *      hey, I'm using it
624          */
625         atomic_inc(&cp->refcnt);
626
627         /*
628          *      do I control anybody?
629          */
630         if (atomic_read(&cp->n_control))
631                 goto expire_later;
632
633         /*
634          *      unhash it if it is hashed in the conn table
635          */
636         if (!ip_vs_conn_unhash(cp))
637                 goto expire_later;
638
639         /*
640          *      refcnt==1 implies I'm the only one referrer
641          */
642         if (likely(atomic_read(&cp->refcnt) == 1)) {
643                 /* delete the timer if it is activated by other users */
644                 if (timer_pending(&cp->timer))
645                         del_timer(&cp->timer);
646
647                 /* does anybody control me? */
648                 if (cp->control)
649                         ip_vs_control_del(cp);
650
651                 if (unlikely(cp->app != NULL))
652                         ip_vs_unbind_app(cp);
653                 ip_vs_unbind_dest(cp);
654                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
655                         atomic_dec(&ip_vs_conn_no_cport_cnt);
656                 atomic_dec(&ip_vs_conn_count);
657
658                 kmem_cache_free(ip_vs_conn_cachep, cp);
659                 return;
660         }
661
662         /* hash it back to the table */
663         ip_vs_conn_hash(cp);
664
665   expire_later:
666         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
667                   atomic_read(&cp->refcnt)-1,
668                   atomic_read(&cp->n_control));
669
670         ip_vs_conn_put(cp);
671 }
672
673
674 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
675 {
676         if (del_timer(&cp->timer))
677                 mod_timer(&cp->timer, jiffies);
678 }
679
680
681 /*
682  *      Create a new connection entry and hash it into the ip_vs_conn_tab
683  */
684 struct ip_vs_conn *
685 ip_vs_conn_new(int af, int proto, const union nf_inet_addr *caddr, __be16 cport,
686                const union nf_inet_addr *vaddr, __be16 vport,
687                const union nf_inet_addr *daddr, __be16 dport, unsigned flags,
688                struct ip_vs_dest *dest)
689 {
690         struct ip_vs_conn *cp;
691         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
692
693         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
694         if (cp == NULL) {
695                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
696                 return NULL;
697         }
698
699         INIT_LIST_HEAD(&cp->c_list);
700         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
701         cp->af             = af;
702         cp->protocol       = proto;
703         ip_vs_addr_copy(af, &cp->caddr, caddr);
704         cp->cport          = cport;
705         ip_vs_addr_copy(af, &cp->vaddr, vaddr);
706         cp->vport          = vport;
707         /* proto should only be IPPROTO_IP if d_addr is a fwmark */
708         ip_vs_addr_copy(proto == IPPROTO_IP ? AF_UNSPEC : af,
709                         &cp->daddr, daddr);
710         cp->dport          = dport;
711         cp->flags          = flags;
712         spin_lock_init(&cp->lock);
713
714         /*
715          * Set the entry is referenced by the current thread before hashing
716          * it in the table, so that other thread run ip_vs_random_dropentry
717          * but cannot drop this entry.
718          */
719         atomic_set(&cp->refcnt, 1);
720
721         atomic_set(&cp->n_control, 0);
722         atomic_set(&cp->in_pkts, 0);
723
724         atomic_inc(&ip_vs_conn_count);
725         if (flags & IP_VS_CONN_F_NO_CPORT)
726                 atomic_inc(&ip_vs_conn_no_cport_cnt);
727
728         /* Bind the connection with a destination server */
729         ip_vs_bind_dest(cp, dest);
730
731         /* Set its state and timeout */
732         cp->state = 0;
733         cp->timeout = 3*HZ;
734
735         /* Bind its packet transmitter */
736 #ifdef CONFIG_IP_VS_IPV6
737         if (af == AF_INET6)
738                 ip_vs_bind_xmit_v6(cp);
739         else
740 #endif
741                 ip_vs_bind_xmit(cp);
742
743         if (unlikely(pp && atomic_read(&pp->appcnt)))
744                 ip_vs_bind_app(cp, pp);
745
746         /* Hash it in the ip_vs_conn_tab finally */
747         ip_vs_conn_hash(cp);
748
749         return cp;
750 }
751
752
753 /*
754  *      /proc/net/ip_vs_conn entries
755  */
756 #ifdef CONFIG_PROC_FS
757
758 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
759 {
760         int idx;
761         struct ip_vs_conn *cp;
762
763         for(idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
764                 ct_read_lock_bh(idx);
765                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
766                         if (pos-- == 0) {
767                                 seq->private = &ip_vs_conn_tab[idx];
768                                 return cp;
769                         }
770                 }
771                 ct_read_unlock_bh(idx);
772         }
773
774         return NULL;
775 }
776
777 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
778 {
779         seq->private = NULL;
780         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
781 }
782
783 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
784 {
785         struct ip_vs_conn *cp = v;
786         struct list_head *e, *l = seq->private;
787         int idx;
788
789         ++*pos;
790         if (v == SEQ_START_TOKEN)
791                 return ip_vs_conn_array(seq, 0);
792
793         /* more on same hash chain? */
794         if ((e = cp->c_list.next) != l)
795                 return list_entry(e, struct ip_vs_conn, c_list);
796
797         idx = l - ip_vs_conn_tab;
798         ct_read_unlock_bh(idx);
799
800         while (++idx < IP_VS_CONN_TAB_SIZE) {
801                 ct_read_lock_bh(idx);
802                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
803                         seq->private = &ip_vs_conn_tab[idx];
804                         return cp;
805                 }
806                 ct_read_unlock_bh(idx);
807         }
808         seq->private = NULL;
809         return NULL;
810 }
811
812 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
813 {
814         struct list_head *l = seq->private;
815
816         if (l)
817                 ct_read_unlock_bh(l - ip_vs_conn_tab);
818 }
819
820 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
821 {
822
823         if (v == SEQ_START_TOKEN)
824                 seq_puts(seq,
825    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
826         else {
827                 const struct ip_vs_conn *cp = v;
828
829 #ifdef CONFIG_IP_VS_IPV6
830                 if (cp->af == AF_INET6)
831                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %7lu\n",
832                                 ip_vs_proto_name(cp->protocol),
833                                 &cp->caddr.in6, ntohs(cp->cport),
834                                 &cp->vaddr.in6, ntohs(cp->vport),
835                                 &cp->daddr.in6, ntohs(cp->dport),
836                                 ip_vs_state_name(cp->protocol, cp->state),
837                                 (cp->timer.expires-jiffies)/HZ);
838                 else
839 #endif
840                         seq_printf(seq,
841                                 "%-3s %08X %04X %08X %04X"
842                                 " %08X %04X %-11s %7lu\n",
843                                 ip_vs_proto_name(cp->protocol),
844                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
845                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
846                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
847                                 ip_vs_state_name(cp->protocol, cp->state),
848                                 (cp->timer.expires-jiffies)/HZ);
849         }
850         return 0;
851 }
852
853 static const struct seq_operations ip_vs_conn_seq_ops = {
854         .start = ip_vs_conn_seq_start,
855         .next  = ip_vs_conn_seq_next,
856         .stop  = ip_vs_conn_seq_stop,
857         .show  = ip_vs_conn_seq_show,
858 };
859
860 static int ip_vs_conn_open(struct inode *inode, struct file *file)
861 {
862         return seq_open(file, &ip_vs_conn_seq_ops);
863 }
864
865 static const struct file_operations ip_vs_conn_fops = {
866         .owner   = THIS_MODULE,
867         .open    = ip_vs_conn_open,
868         .read    = seq_read,
869         .llseek  = seq_lseek,
870         .release = seq_release,
871 };
872
873 static const char *ip_vs_origin_name(unsigned flags)
874 {
875         if (flags & IP_VS_CONN_F_SYNC)
876                 return "SYNC";
877         else
878                 return "LOCAL";
879 }
880
881 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
882 {
883
884         if (v == SEQ_START_TOKEN)
885                 seq_puts(seq,
886    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
887         else {
888                 const struct ip_vs_conn *cp = v;
889
890 #ifdef CONFIG_IP_VS_IPV6
891                 if (cp->af == AF_INET6)
892                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
893                                 ip_vs_proto_name(cp->protocol),
894                                 &cp->caddr.in6, ntohs(cp->cport),
895                                 &cp->vaddr.in6, ntohs(cp->vport),
896                                 &cp->daddr.in6, ntohs(cp->dport),
897                                 ip_vs_state_name(cp->protocol, cp->state),
898                                 ip_vs_origin_name(cp->flags),
899                                 (cp->timer.expires-jiffies)/HZ);
900                 else
901 #endif
902                         seq_printf(seq,
903                                 "%-3s %08X %04X %08X %04X "
904                                 "%08X %04X %-11s %-6s %7lu\n",
905                                 ip_vs_proto_name(cp->protocol),
906                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
907                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
908                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
909                                 ip_vs_state_name(cp->protocol, cp->state),
910                                 ip_vs_origin_name(cp->flags),
911                                 (cp->timer.expires-jiffies)/HZ);
912         }
913         return 0;
914 }
915
916 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
917         .start = ip_vs_conn_seq_start,
918         .next  = ip_vs_conn_seq_next,
919         .stop  = ip_vs_conn_seq_stop,
920         .show  = ip_vs_conn_sync_seq_show,
921 };
922
923 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
924 {
925         return seq_open(file, &ip_vs_conn_sync_seq_ops);
926 }
927
928 static const struct file_operations ip_vs_conn_sync_fops = {
929         .owner   = THIS_MODULE,
930         .open    = ip_vs_conn_sync_open,
931         .read    = seq_read,
932         .llseek  = seq_lseek,
933         .release = seq_release,
934 };
935
936 #endif
937
938
939 /*
940  *      Randomly drop connection entries before running out of memory
941  */
942 static inline int todrop_entry(struct ip_vs_conn *cp)
943 {
944         /*
945          * The drop rate array needs tuning for real environments.
946          * Called from timer bh only => no locking
947          */
948         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
949         static char todrop_counter[9] = {0};
950         int i;
951
952         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
953            This will leave enough time for normal connection to get
954            through. */
955         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
956                 return 0;
957
958         /* Don't drop the entry if its number of incoming packets is not
959            located in [0, 8] */
960         i = atomic_read(&cp->in_pkts);
961         if (i > 8 || i < 0) return 0;
962
963         if (!todrop_rate[i]) return 0;
964         if (--todrop_counter[i] > 0) return 0;
965
966         todrop_counter[i] = todrop_rate[i];
967         return 1;
968 }
969
970 /* Called from keventd and must protect itself from softirqs */
971 void ip_vs_random_dropentry(void)
972 {
973         int idx;
974         struct ip_vs_conn *cp;
975
976         /*
977          * Randomly scan 1/32 of the whole table every second
978          */
979         for (idx = 0; idx < (IP_VS_CONN_TAB_SIZE>>5); idx++) {
980                 unsigned hash = net_random() & IP_VS_CONN_TAB_MASK;
981
982                 /*
983                  *  Lock is actually needed in this loop.
984                  */
985                 ct_write_lock_bh(hash);
986
987                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
988                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
989                                 /* connection template */
990                                 continue;
991
992                         if (cp->protocol == IPPROTO_TCP) {
993                                 switch(cp->state) {
994                                 case IP_VS_TCP_S_SYN_RECV:
995                                 case IP_VS_TCP_S_SYNACK:
996                                         break;
997
998                                 case IP_VS_TCP_S_ESTABLISHED:
999                                         if (todrop_entry(cp))
1000                                                 break;
1001                                         continue;
1002
1003                                 default:
1004                                         continue;
1005                                 }
1006                         } else {
1007                                 if (!todrop_entry(cp))
1008                                         continue;
1009                         }
1010
1011                         IP_VS_DBG(4, "del connection\n");
1012                         ip_vs_conn_expire_now(cp);
1013                         if (cp->control) {
1014                                 IP_VS_DBG(4, "del conn template\n");
1015                                 ip_vs_conn_expire_now(cp->control);
1016                         }
1017                 }
1018                 ct_write_unlock_bh(hash);
1019         }
1020 }
1021
1022
1023 /*
1024  *      Flush all the connection entries in the ip_vs_conn_tab
1025  */
1026 static void ip_vs_conn_flush(void)
1027 {
1028         int idx;
1029         struct ip_vs_conn *cp;
1030
1031   flush_again:
1032         for (idx=0; idx<IP_VS_CONN_TAB_SIZE; idx++) {
1033                 /*
1034                  *  Lock is actually needed in this loop.
1035                  */
1036                 ct_write_lock_bh(idx);
1037
1038                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
1039
1040                         IP_VS_DBG(4, "del connection\n");
1041                         ip_vs_conn_expire_now(cp);
1042                         if (cp->control) {
1043                                 IP_VS_DBG(4, "del conn template\n");
1044                                 ip_vs_conn_expire_now(cp->control);
1045                         }
1046                 }
1047                 ct_write_unlock_bh(idx);
1048         }
1049
1050         /* the counter may be not NULL, because maybe some conn entries
1051            are run by slow timer handler or unhashed but still referred */
1052         if (atomic_read(&ip_vs_conn_count) != 0) {
1053                 schedule();
1054                 goto flush_again;
1055         }
1056 }
1057
1058
1059 int __init ip_vs_conn_init(void)
1060 {
1061         int idx;
1062
1063         /*
1064          * Allocate the connection hash table and initialize its list heads
1065          */
1066         ip_vs_conn_tab = vmalloc(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head));
1067         if (!ip_vs_conn_tab)
1068                 return -ENOMEM;
1069
1070         /* Allocate ip_vs_conn slab cache */
1071         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1072                                               sizeof(struct ip_vs_conn), 0,
1073                                               SLAB_HWCACHE_ALIGN, NULL);
1074         if (!ip_vs_conn_cachep) {
1075                 vfree(ip_vs_conn_tab);
1076                 return -ENOMEM;
1077         }
1078
1079         pr_info("Connection hash table configured "
1080                 "(size=%d, memory=%ldKbytes)\n",
1081                 IP_VS_CONN_TAB_SIZE,
1082                 (long)(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head))/1024);
1083         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1084                   sizeof(struct ip_vs_conn));
1085
1086         for (idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
1087                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
1088         }
1089
1090         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1091                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1092         }
1093
1094         proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops);
1095         proc_net_fops_create(&init_net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops);
1096
1097         /* calculate the random value for connection hash */
1098         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1099
1100         return 0;
1101 }
1102
1103
1104 void ip_vs_conn_cleanup(void)
1105 {
1106         /* flush all the connection entries first */
1107         ip_vs_conn_flush();
1108
1109         /* Release the empty cache */
1110         kmem_cache_destroy(ip_vs_conn_cachep);
1111         proc_net_remove(&init_net, "ip_vs_conn");
1112         proc_net_remove(&init_net, "ip_vs_conn_sync");
1113         vfree(ip_vs_conn_tab);
1114 }