Merge branch 'for-linus' of git://repo.or.cz/cris-mirror
[kernel.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 #include <linux/interrupt.h>
26 #include <linux/in.h>
27 #include <linux/net.h>
28 #include <linux/kernel.h>
29 #include <linux/module.h>
30 #include <linux/vmalloc.h>
31 #include <linux/proc_fs.h>              /* for proc_net_* */
32 #include <linux/seq_file.h>
33 #include <linux/jhash.h>
34 #include <linux/random.h>
35
36 #include <net/net_namespace.h>
37 #include <net/ip_vs.h>
38
39
40 /*
41  *  Connection hash table: for input and output packets lookups of IPVS
42  */
43 static struct list_head *ip_vs_conn_tab;
44
45 /*  SLAB cache for IPVS connections */
46 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
47
48 /*  counter for current IPVS connections */
49 static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
50
51 /*  counter for no client port connections */
52 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
53
54 /* random value for IPVS connection hash */
55 static unsigned int ip_vs_conn_rnd;
56
57 /*
58  *  Fine locking granularity for big connection hash table
59  */
60 #define CT_LOCKARRAY_BITS  4
61 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
62 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
63
64 struct ip_vs_aligned_lock
65 {
66         rwlock_t        l;
67 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
68
69 /* lock array for conn table */
70 static struct ip_vs_aligned_lock
71 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
72
73 static inline void ct_read_lock(unsigned key)
74 {
75         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
76 }
77
78 static inline void ct_read_unlock(unsigned key)
79 {
80         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
81 }
82
83 static inline void ct_write_lock(unsigned key)
84 {
85         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
86 }
87
88 static inline void ct_write_unlock(unsigned key)
89 {
90         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
91 }
92
93 static inline void ct_read_lock_bh(unsigned key)
94 {
95         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
96 }
97
98 static inline void ct_read_unlock_bh(unsigned key)
99 {
100         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
101 }
102
103 static inline void ct_write_lock_bh(unsigned key)
104 {
105         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
106 }
107
108 static inline void ct_write_unlock_bh(unsigned key)
109 {
110         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
111 }
112
113
114 /*
115  *      Returns hash value for IPVS connection entry
116  */
117 static unsigned int ip_vs_conn_hashkey(int af, unsigned proto,
118                                        const union nf_inet_addr *addr,
119                                        __be16 port)
120 {
121 #ifdef CONFIG_IP_VS_IPV6
122         if (af == AF_INET6)
123                 return jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
124                                     (__force u32)port, proto, ip_vs_conn_rnd)
125                         & IP_VS_CONN_TAB_MASK;
126 #endif
127         return jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
128                             ip_vs_conn_rnd)
129                 & IP_VS_CONN_TAB_MASK;
130 }
131
132
133 /*
134  *      Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
135  *      returns bool success.
136  */
137 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
138 {
139         unsigned hash;
140         int ret;
141
142         /* Hash by protocol, client address and port */
143         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
144
145         ct_write_lock(hash);
146
147         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
148                 list_add(&cp->c_list, &ip_vs_conn_tab[hash]);
149                 cp->flags |= IP_VS_CONN_F_HASHED;
150                 atomic_inc(&cp->refcnt);
151                 ret = 1;
152         } else {
153                 IP_VS_ERR("ip_vs_conn_hash(): request for already hashed, "
154                           "called from %p\n", __builtin_return_address(0));
155                 ret = 0;
156         }
157
158         ct_write_unlock(hash);
159
160         return ret;
161 }
162
163
164 /*
165  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
166  *      returns bool success.
167  */
168 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
169 {
170         unsigned hash;
171         int ret;
172
173         /* unhash it and decrease its reference counter */
174         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
175
176         ct_write_lock(hash);
177
178         if (cp->flags & IP_VS_CONN_F_HASHED) {
179                 list_del(&cp->c_list);
180                 cp->flags &= ~IP_VS_CONN_F_HASHED;
181                 atomic_dec(&cp->refcnt);
182                 ret = 1;
183         } else
184                 ret = 0;
185
186         ct_write_unlock(hash);
187
188         return ret;
189 }
190
191
192 /*
193  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
194  *  Called for pkts coming from OUTside-to-INside.
195  *      s_addr, s_port: pkt source address (foreign host)
196  *      d_addr, d_port: pkt dest address (load balancer)
197  */
198 static inline struct ip_vs_conn *__ip_vs_conn_in_get
199 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
200  const union nf_inet_addr *d_addr, __be16 d_port)
201 {
202         unsigned hash;
203         struct ip_vs_conn *cp;
204
205         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
206
207         ct_read_lock(hash);
208
209         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
210                 if (cp->af == af &&
211                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
212                     ip_vs_addr_equal(af, d_addr, &cp->vaddr) &&
213                     s_port == cp->cport && d_port == cp->vport &&
214                     ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
215                     protocol == cp->protocol) {
216                         /* HIT */
217                         atomic_inc(&cp->refcnt);
218                         ct_read_unlock(hash);
219                         return cp;
220                 }
221         }
222
223         ct_read_unlock(hash);
224
225         return NULL;
226 }
227
228 struct ip_vs_conn *ip_vs_conn_in_get
229 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
230  const union nf_inet_addr *d_addr, __be16 d_port)
231 {
232         struct ip_vs_conn *cp;
233
234         cp = __ip_vs_conn_in_get(af, protocol, s_addr, s_port, d_addr, d_port);
235         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt))
236                 cp = __ip_vs_conn_in_get(af, protocol, s_addr, 0, d_addr,
237                                          d_port);
238
239         IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
240                       ip_vs_proto_name(protocol),
241                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
242                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
243                       cp ? "hit" : "not hit");
244
245         return cp;
246 }
247
248 /* Get reference to connection template */
249 struct ip_vs_conn *ip_vs_ct_in_get
250 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
251  const union nf_inet_addr *d_addr, __be16 d_port)
252 {
253         unsigned hash;
254         struct ip_vs_conn *cp;
255
256         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
257
258         ct_read_lock(hash);
259
260         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
261                 if (cp->af == af &&
262                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
263                     /* protocol should only be IPPROTO_IP if
264                      * d_addr is a fwmark */
265                     ip_vs_addr_equal(protocol == IPPROTO_IP ? AF_UNSPEC : af,
266                                      d_addr, &cp->vaddr) &&
267                     s_port == cp->cport && d_port == cp->vport &&
268                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
269                     protocol == cp->protocol) {
270                         /* HIT */
271                         atomic_inc(&cp->refcnt);
272                         goto out;
273                 }
274         }
275         cp = NULL;
276
277   out:
278         ct_read_unlock(hash);
279
280         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
281                       ip_vs_proto_name(protocol),
282                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
283                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
284                       cp ? "hit" : "not hit");
285
286         return cp;
287 }
288
289 /*
290  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
291  *  Called for pkts coming from inside-to-OUTside.
292  *      s_addr, s_port: pkt source address (inside host)
293  *      d_addr, d_port: pkt dest address (foreign host)
294  */
295 struct ip_vs_conn *ip_vs_conn_out_get
296 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
297  const union nf_inet_addr *d_addr, __be16 d_port)
298 {
299         unsigned hash;
300         struct ip_vs_conn *cp, *ret=NULL;
301
302         /*
303          *      Check for "full" addressed entries
304          */
305         hash = ip_vs_conn_hashkey(af, protocol, d_addr, d_port);
306
307         ct_read_lock(hash);
308
309         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
310                 if (cp->af == af &&
311                     ip_vs_addr_equal(af, d_addr, &cp->caddr) &&
312                     ip_vs_addr_equal(af, s_addr, &cp->daddr) &&
313                     d_port == cp->cport && s_port == cp->dport &&
314                     protocol == cp->protocol) {
315                         /* HIT */
316                         atomic_inc(&cp->refcnt);
317                         ret = cp;
318                         break;
319                 }
320         }
321
322         ct_read_unlock(hash);
323
324         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
325                       ip_vs_proto_name(protocol),
326                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
327                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
328                       ret ? "hit" : "not hit");
329
330         return ret;
331 }
332
333
334 /*
335  *      Put back the conn and restart its timer with its timeout
336  */
337 void ip_vs_conn_put(struct ip_vs_conn *cp)
338 {
339         /* reset it expire in its timeout */
340         mod_timer(&cp->timer, jiffies+cp->timeout);
341
342         __ip_vs_conn_put(cp);
343 }
344
345
346 /*
347  *      Fill a no_client_port connection with a client port number
348  */
349 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
350 {
351         if (ip_vs_conn_unhash(cp)) {
352                 spin_lock(&cp->lock);
353                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
354                         atomic_dec(&ip_vs_conn_no_cport_cnt);
355                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
356                         cp->cport = cport;
357                 }
358                 spin_unlock(&cp->lock);
359
360                 /* hash on new dport */
361                 ip_vs_conn_hash(cp);
362         }
363 }
364
365
366 /*
367  *      Bind a connection entry with the corresponding packet_xmit.
368  *      Called by ip_vs_conn_new.
369  */
370 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
371 {
372         switch (IP_VS_FWD_METHOD(cp)) {
373         case IP_VS_CONN_F_MASQ:
374                 cp->packet_xmit = ip_vs_nat_xmit;
375                 break;
376
377         case IP_VS_CONN_F_TUNNEL:
378                 cp->packet_xmit = ip_vs_tunnel_xmit;
379                 break;
380
381         case IP_VS_CONN_F_DROUTE:
382                 cp->packet_xmit = ip_vs_dr_xmit;
383                 break;
384
385         case IP_VS_CONN_F_LOCALNODE:
386                 cp->packet_xmit = ip_vs_null_xmit;
387                 break;
388
389         case IP_VS_CONN_F_BYPASS:
390                 cp->packet_xmit = ip_vs_bypass_xmit;
391                 break;
392         }
393 }
394
395 #ifdef CONFIG_IP_VS_IPV6
396 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
397 {
398         switch (IP_VS_FWD_METHOD(cp)) {
399         case IP_VS_CONN_F_MASQ:
400                 cp->packet_xmit = ip_vs_nat_xmit_v6;
401                 break;
402
403         case IP_VS_CONN_F_TUNNEL:
404                 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
405                 break;
406
407         case IP_VS_CONN_F_DROUTE:
408                 cp->packet_xmit = ip_vs_dr_xmit_v6;
409                 break;
410
411         case IP_VS_CONN_F_LOCALNODE:
412                 cp->packet_xmit = ip_vs_null_xmit;
413                 break;
414
415         case IP_VS_CONN_F_BYPASS:
416                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
417                 break;
418         }
419 }
420 #endif
421
422
423 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
424 {
425         return atomic_read(&dest->activeconns)
426                 + atomic_read(&dest->inactconns);
427 }
428
429 /*
430  *      Bind a connection entry with a virtual service destination
431  *      Called just after a new connection entry is created.
432  */
433 static inline void
434 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
435 {
436         /* if dest is NULL, then return directly */
437         if (!dest)
438                 return;
439
440         /* Increase the refcnt counter of the dest */
441         atomic_inc(&dest->refcnt);
442
443         /* Bind with the destination and its corresponding transmitter */
444         if ((cp->flags & IP_VS_CONN_F_SYNC) &&
445             (!(cp->flags & IP_VS_CONN_F_TEMPLATE)))
446                 /* if the connection is not template and is created
447                  * by sync, preserve the activity flag.
448                  */
449                 cp->flags |= atomic_read(&dest->conn_flags) &
450                              (~IP_VS_CONN_F_INACTIVE);
451         else
452                 cp->flags |= atomic_read(&dest->conn_flags);
453         cp->dest = dest;
454
455         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
456                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
457                       "dest->refcnt:%d\n",
458                       ip_vs_proto_name(cp->protocol),
459                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
460                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
461                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
462                       ip_vs_fwd_tag(cp), cp->state,
463                       cp->flags, atomic_read(&cp->refcnt),
464                       atomic_read(&dest->refcnt));
465
466         /* Update the connection counters */
467         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
468                 /* It is a normal connection, so increase the inactive
469                    connection counter because it is in TCP SYNRECV
470                    state (inactive) or other protocol inacive state */
471                 if ((cp->flags & IP_VS_CONN_F_SYNC) &&
472                     (!(cp->flags & IP_VS_CONN_F_INACTIVE)))
473                         atomic_inc(&dest->activeconns);
474                 else
475                         atomic_inc(&dest->inactconns);
476         } else {
477                 /* It is a persistent connection/template, so increase
478                    the peristent connection counter */
479                 atomic_inc(&dest->persistconns);
480         }
481
482         if (dest->u_threshold != 0 &&
483             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
484                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
485 }
486
487
488 /*
489  * Check if there is a destination for the connection, if so
490  * bind the connection to the destination.
491  */
492 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
493 {
494         struct ip_vs_dest *dest;
495
496         if ((cp) && (!cp->dest)) {
497                 dest = ip_vs_find_dest(cp->af, &cp->daddr, cp->dport,
498                                        &cp->vaddr, cp->vport,
499                                        cp->protocol);
500                 ip_vs_bind_dest(cp, dest);
501                 return dest;
502         } else
503                 return NULL;
504 }
505
506
507 /*
508  *      Unbind a connection entry with its VS destination
509  *      Called by the ip_vs_conn_expire function.
510  */
511 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
512 {
513         struct ip_vs_dest *dest = cp->dest;
514
515         if (!dest)
516                 return;
517
518         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
519                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
520                       "dest->refcnt:%d\n",
521                       ip_vs_proto_name(cp->protocol),
522                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
523                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
524                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
525                       ip_vs_fwd_tag(cp), cp->state,
526                       cp->flags, atomic_read(&cp->refcnt),
527                       atomic_read(&dest->refcnt));
528
529         /* Update the connection counters */
530         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
531                 /* It is a normal connection, so decrease the inactconns
532                    or activeconns counter */
533                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
534                         atomic_dec(&dest->inactconns);
535                 } else {
536                         atomic_dec(&dest->activeconns);
537                 }
538         } else {
539                 /* It is a persistent connection/template, so decrease
540                    the peristent connection counter */
541                 atomic_dec(&dest->persistconns);
542         }
543
544         if (dest->l_threshold != 0) {
545                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
546                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
547         } else if (dest->u_threshold != 0) {
548                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
549                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
550         } else {
551                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
552                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
553         }
554
555         /*
556          * Simply decrease the refcnt of the dest, because the
557          * dest will be either in service's destination list
558          * or in the trash.
559          */
560         atomic_dec(&dest->refcnt);
561 }
562
563
564 /*
565  *      Checking if the destination of a connection template is available.
566  *      If available, return 1, otherwise invalidate this connection
567  *      template and return 0.
568  */
569 int ip_vs_check_template(struct ip_vs_conn *ct)
570 {
571         struct ip_vs_dest *dest = ct->dest;
572
573         /*
574          * Checking the dest server status.
575          */
576         if ((dest == NULL) ||
577             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
578             (sysctl_ip_vs_expire_quiescent_template &&
579              (atomic_read(&dest->weight) == 0))) {
580                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
581                               "protocol %s s:%s:%d v:%s:%d "
582                               "-> d:%s:%d\n",
583                               ip_vs_proto_name(ct->protocol),
584                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
585                               ntohs(ct->cport),
586                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
587                               ntohs(ct->vport),
588                               IP_VS_DBG_ADDR(ct->af, &ct->daddr),
589                               ntohs(ct->dport));
590
591                 /*
592                  * Invalidate the connection template
593                  */
594                 if (ct->vport != htons(0xffff)) {
595                         if (ip_vs_conn_unhash(ct)) {
596                                 ct->dport = htons(0xffff);
597                                 ct->vport = htons(0xffff);
598                                 ct->cport = 0;
599                                 ip_vs_conn_hash(ct);
600                         }
601                 }
602
603                 /*
604                  * Simply decrease the refcnt of the template,
605                  * don't restart its timer.
606                  */
607                 atomic_dec(&ct->refcnt);
608                 return 0;
609         }
610         return 1;
611 }
612
613 static void ip_vs_conn_expire(unsigned long data)
614 {
615         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
616
617         cp->timeout = 60*HZ;
618
619         /*
620          *      hey, I'm using it
621          */
622         atomic_inc(&cp->refcnt);
623
624         /*
625          *      do I control anybody?
626          */
627         if (atomic_read(&cp->n_control))
628                 goto expire_later;
629
630         /*
631          *      unhash it if it is hashed in the conn table
632          */
633         if (!ip_vs_conn_unhash(cp))
634                 goto expire_later;
635
636         /*
637          *      refcnt==1 implies I'm the only one referrer
638          */
639         if (likely(atomic_read(&cp->refcnt) == 1)) {
640                 /* delete the timer if it is activated by other users */
641                 if (timer_pending(&cp->timer))
642                         del_timer(&cp->timer);
643
644                 /* does anybody control me? */
645                 if (cp->control)
646                         ip_vs_control_del(cp);
647
648                 if (unlikely(cp->app != NULL))
649                         ip_vs_unbind_app(cp);
650                 ip_vs_unbind_dest(cp);
651                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
652                         atomic_dec(&ip_vs_conn_no_cport_cnt);
653                 atomic_dec(&ip_vs_conn_count);
654
655                 kmem_cache_free(ip_vs_conn_cachep, cp);
656                 return;
657         }
658
659         /* hash it back to the table */
660         ip_vs_conn_hash(cp);
661
662   expire_later:
663         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
664                   atomic_read(&cp->refcnt)-1,
665                   atomic_read(&cp->n_control));
666
667         ip_vs_conn_put(cp);
668 }
669
670
671 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
672 {
673         if (del_timer(&cp->timer))
674                 mod_timer(&cp->timer, jiffies);
675 }
676
677
678 /*
679  *      Create a new connection entry and hash it into the ip_vs_conn_tab
680  */
681 struct ip_vs_conn *
682 ip_vs_conn_new(int af, int proto, const union nf_inet_addr *caddr, __be16 cport,
683                const union nf_inet_addr *vaddr, __be16 vport,
684                const union nf_inet_addr *daddr, __be16 dport, unsigned flags,
685                struct ip_vs_dest *dest)
686 {
687         struct ip_vs_conn *cp;
688         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
689
690         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
691         if (cp == NULL) {
692                 IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n");
693                 return NULL;
694         }
695
696         INIT_LIST_HEAD(&cp->c_list);
697         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
698         cp->af             = af;
699         cp->protocol       = proto;
700         ip_vs_addr_copy(af, &cp->caddr, caddr);
701         cp->cport          = cport;
702         ip_vs_addr_copy(af, &cp->vaddr, vaddr);
703         cp->vport          = vport;
704         /* proto should only be IPPROTO_IP if d_addr is a fwmark */
705         ip_vs_addr_copy(proto == IPPROTO_IP ? AF_UNSPEC : af,
706                         &cp->daddr, daddr);
707         cp->dport          = dport;
708         cp->flags          = flags;
709         spin_lock_init(&cp->lock);
710
711         /*
712          * Set the entry is referenced by the current thread before hashing
713          * it in the table, so that other thread run ip_vs_random_dropentry
714          * but cannot drop this entry.
715          */
716         atomic_set(&cp->refcnt, 1);
717
718         atomic_set(&cp->n_control, 0);
719         atomic_set(&cp->in_pkts, 0);
720
721         atomic_inc(&ip_vs_conn_count);
722         if (flags & IP_VS_CONN_F_NO_CPORT)
723                 atomic_inc(&ip_vs_conn_no_cport_cnt);
724
725         /* Bind the connection with a destination server */
726         ip_vs_bind_dest(cp, dest);
727
728         /* Set its state and timeout */
729         cp->state = 0;
730         cp->timeout = 3*HZ;
731
732         /* Bind its packet transmitter */
733 #ifdef CONFIG_IP_VS_IPV6
734         if (af == AF_INET6)
735                 ip_vs_bind_xmit_v6(cp);
736         else
737 #endif
738                 ip_vs_bind_xmit(cp);
739
740         if (unlikely(pp && atomic_read(&pp->appcnt)))
741                 ip_vs_bind_app(cp, pp);
742
743         /* Hash it in the ip_vs_conn_tab finally */
744         ip_vs_conn_hash(cp);
745
746         return cp;
747 }
748
749
750 /*
751  *      /proc/net/ip_vs_conn entries
752  */
753 #ifdef CONFIG_PROC_FS
754
755 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
756 {
757         int idx;
758         struct ip_vs_conn *cp;
759
760         for(idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
761                 ct_read_lock_bh(idx);
762                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
763                         if (pos-- == 0) {
764                                 seq->private = &ip_vs_conn_tab[idx];
765                                 return cp;
766                         }
767                 }
768                 ct_read_unlock_bh(idx);
769         }
770
771         return NULL;
772 }
773
774 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
775 {
776         seq->private = NULL;
777         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
778 }
779
780 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
781 {
782         struct ip_vs_conn *cp = v;
783         struct list_head *e, *l = seq->private;
784         int idx;
785
786         ++*pos;
787         if (v == SEQ_START_TOKEN)
788                 return ip_vs_conn_array(seq, 0);
789
790         /* more on same hash chain? */
791         if ((e = cp->c_list.next) != l)
792                 return list_entry(e, struct ip_vs_conn, c_list);
793
794         idx = l - ip_vs_conn_tab;
795         ct_read_unlock_bh(idx);
796
797         while (++idx < IP_VS_CONN_TAB_SIZE) {
798                 ct_read_lock_bh(idx);
799                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
800                         seq->private = &ip_vs_conn_tab[idx];
801                         return cp;
802                 }
803                 ct_read_unlock_bh(idx);
804         }
805         seq->private = NULL;
806         return NULL;
807 }
808
809 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
810 {
811         struct list_head *l = seq->private;
812
813         if (l)
814                 ct_read_unlock_bh(l - ip_vs_conn_tab);
815 }
816
817 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
818 {
819
820         if (v == SEQ_START_TOKEN)
821                 seq_puts(seq,
822    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
823         else {
824                 const struct ip_vs_conn *cp = v;
825
826 #ifdef CONFIG_IP_VS_IPV6
827                 if (cp->af == AF_INET6)
828                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %7lu\n",
829                                 ip_vs_proto_name(cp->protocol),
830                                 &cp->caddr.in6, ntohs(cp->cport),
831                                 &cp->vaddr.in6, ntohs(cp->vport),
832                                 &cp->daddr.in6, ntohs(cp->dport),
833                                 ip_vs_state_name(cp->protocol, cp->state),
834                                 (cp->timer.expires-jiffies)/HZ);
835                 else
836 #endif
837                         seq_printf(seq,
838                                 "%-3s %08X %04X %08X %04X"
839                                 " %08X %04X %-11s %7lu\n",
840                                 ip_vs_proto_name(cp->protocol),
841                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
842                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
843                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
844                                 ip_vs_state_name(cp->protocol, cp->state),
845                                 (cp->timer.expires-jiffies)/HZ);
846         }
847         return 0;
848 }
849
850 static const struct seq_operations ip_vs_conn_seq_ops = {
851         .start = ip_vs_conn_seq_start,
852         .next  = ip_vs_conn_seq_next,
853         .stop  = ip_vs_conn_seq_stop,
854         .show  = ip_vs_conn_seq_show,
855 };
856
857 static int ip_vs_conn_open(struct inode *inode, struct file *file)
858 {
859         return seq_open(file, &ip_vs_conn_seq_ops);
860 }
861
862 static const struct file_operations ip_vs_conn_fops = {
863         .owner   = THIS_MODULE,
864         .open    = ip_vs_conn_open,
865         .read    = seq_read,
866         .llseek  = seq_lseek,
867         .release = seq_release,
868 };
869
870 static const char *ip_vs_origin_name(unsigned flags)
871 {
872         if (flags & IP_VS_CONN_F_SYNC)
873                 return "SYNC";
874         else
875                 return "LOCAL";
876 }
877
878 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
879 {
880
881         if (v == SEQ_START_TOKEN)
882                 seq_puts(seq,
883    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
884         else {
885                 const struct ip_vs_conn *cp = v;
886
887 #ifdef CONFIG_IP_VS_IPV6
888                 if (cp->af == AF_INET6)
889                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
890                                 ip_vs_proto_name(cp->protocol),
891                                 &cp->caddr.in6, ntohs(cp->cport),
892                                 &cp->vaddr.in6, ntohs(cp->vport),
893                                 &cp->daddr.in6, ntohs(cp->dport),
894                                 ip_vs_state_name(cp->protocol, cp->state),
895                                 ip_vs_origin_name(cp->flags),
896                                 (cp->timer.expires-jiffies)/HZ);
897                 else
898 #endif
899                         seq_printf(seq,
900                                 "%-3s %08X %04X %08X %04X "
901                                 "%08X %04X %-11s %-6s %7lu\n",
902                                 ip_vs_proto_name(cp->protocol),
903                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
904                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
905                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
906                                 ip_vs_state_name(cp->protocol, cp->state),
907                                 ip_vs_origin_name(cp->flags),
908                                 (cp->timer.expires-jiffies)/HZ);
909         }
910         return 0;
911 }
912
913 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
914         .start = ip_vs_conn_seq_start,
915         .next  = ip_vs_conn_seq_next,
916         .stop  = ip_vs_conn_seq_stop,
917         .show  = ip_vs_conn_sync_seq_show,
918 };
919
920 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
921 {
922         return seq_open(file, &ip_vs_conn_sync_seq_ops);
923 }
924
925 static const struct file_operations ip_vs_conn_sync_fops = {
926         .owner   = THIS_MODULE,
927         .open    = ip_vs_conn_sync_open,
928         .read    = seq_read,
929         .llseek  = seq_lseek,
930         .release = seq_release,
931 };
932
933 #endif
934
935
936 /*
937  *      Randomly drop connection entries before running out of memory
938  */
939 static inline int todrop_entry(struct ip_vs_conn *cp)
940 {
941         /*
942          * The drop rate array needs tuning for real environments.
943          * Called from timer bh only => no locking
944          */
945         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
946         static char todrop_counter[9] = {0};
947         int i;
948
949         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
950            This will leave enough time for normal connection to get
951            through. */
952         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
953                 return 0;
954
955         /* Don't drop the entry if its number of incoming packets is not
956            located in [0, 8] */
957         i = atomic_read(&cp->in_pkts);
958         if (i > 8 || i < 0) return 0;
959
960         if (!todrop_rate[i]) return 0;
961         if (--todrop_counter[i] > 0) return 0;
962
963         todrop_counter[i] = todrop_rate[i];
964         return 1;
965 }
966
967 /* Called from keventd and must protect itself from softirqs */
968 void ip_vs_random_dropentry(void)
969 {
970         int idx;
971         struct ip_vs_conn *cp;
972
973         /*
974          * Randomly scan 1/32 of the whole table every second
975          */
976         for (idx = 0; idx < (IP_VS_CONN_TAB_SIZE>>5); idx++) {
977                 unsigned hash = net_random() & IP_VS_CONN_TAB_MASK;
978
979                 /*
980                  *  Lock is actually needed in this loop.
981                  */
982                 ct_write_lock_bh(hash);
983
984                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
985                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
986                                 /* connection template */
987                                 continue;
988
989                         if (cp->protocol == IPPROTO_TCP) {
990                                 switch(cp->state) {
991                                 case IP_VS_TCP_S_SYN_RECV:
992                                 case IP_VS_TCP_S_SYNACK:
993                                         break;
994
995                                 case IP_VS_TCP_S_ESTABLISHED:
996                                         if (todrop_entry(cp))
997                                                 break;
998                                         continue;
999
1000                                 default:
1001                                         continue;
1002                                 }
1003                         } else {
1004                                 if (!todrop_entry(cp))
1005                                         continue;
1006                         }
1007
1008                         IP_VS_DBG(4, "del connection\n");
1009                         ip_vs_conn_expire_now(cp);
1010                         if (cp->control) {
1011                                 IP_VS_DBG(4, "del conn template\n");
1012                                 ip_vs_conn_expire_now(cp->control);
1013                         }
1014                 }
1015                 ct_write_unlock_bh(hash);
1016         }
1017 }
1018
1019
1020 /*
1021  *      Flush all the connection entries in the ip_vs_conn_tab
1022  */
1023 static void ip_vs_conn_flush(void)
1024 {
1025         int idx;
1026         struct ip_vs_conn *cp;
1027
1028   flush_again:
1029         for (idx=0; idx<IP_VS_CONN_TAB_SIZE; idx++) {
1030                 /*
1031                  *  Lock is actually needed in this loop.
1032                  */
1033                 ct_write_lock_bh(idx);
1034
1035                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
1036
1037                         IP_VS_DBG(4, "del connection\n");
1038                         ip_vs_conn_expire_now(cp);
1039                         if (cp->control) {
1040                                 IP_VS_DBG(4, "del conn template\n");
1041                                 ip_vs_conn_expire_now(cp->control);
1042                         }
1043                 }
1044                 ct_write_unlock_bh(idx);
1045         }
1046
1047         /* the counter may be not NULL, because maybe some conn entries
1048            are run by slow timer handler or unhashed but still referred */
1049         if (atomic_read(&ip_vs_conn_count) != 0) {
1050                 schedule();
1051                 goto flush_again;
1052         }
1053 }
1054
1055
1056 int __init ip_vs_conn_init(void)
1057 {
1058         int idx;
1059
1060         /*
1061          * Allocate the connection hash table and initialize its list heads
1062          */
1063         ip_vs_conn_tab = vmalloc(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head));
1064         if (!ip_vs_conn_tab)
1065                 return -ENOMEM;
1066
1067         /* Allocate ip_vs_conn slab cache */
1068         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1069                                               sizeof(struct ip_vs_conn), 0,
1070                                               SLAB_HWCACHE_ALIGN, NULL);
1071         if (!ip_vs_conn_cachep) {
1072                 vfree(ip_vs_conn_tab);
1073                 return -ENOMEM;
1074         }
1075
1076         IP_VS_INFO("Connection hash table configured "
1077                    "(size=%d, memory=%ldKbytes)\n",
1078                    IP_VS_CONN_TAB_SIZE,
1079                    (long)(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head))/1024);
1080         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1081                   sizeof(struct ip_vs_conn));
1082
1083         for (idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
1084                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
1085         }
1086
1087         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1088                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1089         }
1090
1091         proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops);
1092         proc_net_fops_create(&init_net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops);
1093
1094         /* calculate the random value for connection hash */
1095         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1096
1097         return 0;
1098 }
1099
1100
1101 void ip_vs_conn_cleanup(void)
1102 {
1103         /* flush all the connection entries first */
1104         ip_vs_conn_flush();
1105
1106         /* Release the empty cache */
1107         kmem_cache_destroy(ip_vs_conn_cachep);
1108         proc_net_remove(&init_net, "ip_vs_conn");
1109         proc_net_remove(&init_net, "ip_vs_conn_sync");
1110         vfree(ip_vs_conn_tab);
1111 }