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         spinlock_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_write_lock(unsigned int key)
  90 {
  91         spin_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  92 }
  93
  94 static inline void ct_write_unlock(unsigned int key)
  95 {
  96         spin_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  97 }
  98
  99
 100 /*
 101  *      Returns hash value for IPVS connection entry
 102  */
 103 static unsigned int ip_vs_conn_hashkey(struct net *net, int af, unsigned int proto,
 104                                        const union nf_inet_addr *addr,
 105                                        __be16 port)
 106 {
 107 #ifdef CONFIG_IP_VS_IPV6
 108         if (af == AF_INET6)
 109                 return (jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
 110                                     (__force u32)port, proto, ip_vs_conn_rnd) ^
 111                         ((size_t)net>>8)) & ip_vs_conn_tab_mask;
 112 #endif
 113         return (jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
 114                             ip_vs_conn_rnd) ^
 115                 ((size_t)net>>8)) & ip_vs_conn_tab_mask;
 116 }
 117
 118 static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param *p,
 119                                              bool inverse)
 120 {
 121         const union nf_inet_addr *addr;
 122         __be16 port;
 123
 124         if (p->pe_data && p->pe->hashkey_raw)
 125                 return p->pe->hashkey_raw(p, ip_vs_conn_rnd, inverse) &
 126                         ip_vs_conn_tab_mask;
 127
 128         if (likely(!inverse)) {
 129                 addr = p->caddr;
 130                 port = p->cport;
 131         } else {
 132                 addr = p->vaddr;
 133                 port = p->vport;
 134         }
 135
 136         return ip_vs_conn_hashkey(p->net, p->af, p->protocol, addr, port);
 137 }
 138
 139 static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn *cp)
 140 {
 141         struct ip_vs_conn_param p;
 142
 143         ip_vs_conn_fill_param(ip_vs_conn_net(cp), cp->af, cp->protocol,
 144                               &cp->caddr, cp->cport, NULL, 0, &p);
 145
 146         if (cp->pe) {
 147                 p.pe = cp->pe;
 148                 p.pe_data = cp->pe_data;
 149                 p.pe_data_len = cp->pe_data_len;
 150         }
 151
 152         return ip_vs_conn_hashkey_param(&p, false);
 153 }
 154
 155 /*
 156  *      Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
 157  *      returns bool success.
 158  */
 159 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
 160 {
 161         unsigned int hash;
 162         int ret;
 163
 164         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
 165                 return 0;
 166
 167         /* Hash by protocol, client address and port */
 168         hash = ip_vs_conn_hashkey_conn(cp);
 169
 170         ct_write_lock(hash);
 171         spin_lock(&cp->lock);
 172
 173         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
 174                 cp->flags |= IP_VS_CONN_F_HASHED;
 175                 atomic_inc(&cp->refcnt);
 176                 hlist_add_head_rcu(&cp->c_list, &ip_vs_conn_tab[hash]);
 177                 ret = 1;
 178         } else {
 179                 pr_err("%s(): request for already hashed, called from %pF\n",
 180                        __func__, __builtin_return_address(0));
 181                 ret = 0;
 182         }
 183
 184         spin_unlock(&cp->lock);
 185         ct_write_unlock(hash);
 186
 187         return ret;
 188 }
 189
 190
 191 /*
 192  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
 193  *      returns bool success. Caller should hold conn reference.
 194  */
 195 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
 196 {
 197         unsigned int hash;
 198         int ret;
 199
 200         /* unhash it and decrease its reference counter */
 201         hash = ip_vs_conn_hashkey_conn(cp);
 202
 203         ct_write_lock(hash);
 204         spin_lock(&cp->lock);
 205
 206         if (cp->flags & IP_VS_CONN_F_HASHED) {
 207                 hlist_del_rcu(&cp->c_list);
 208                 cp->flags &= ~IP_VS_CONN_F_HASHED;
 209                 atomic_dec(&cp->refcnt);
 210                 ret = 1;
 211         } else
 212                 ret = 0;
 213
 214         spin_unlock(&cp->lock);
 215         ct_write_unlock(hash);
 216
 217         return ret;
 218 }
 219
 220 /* Try to unlink ip_vs_conn from ip_vs_conn_tab.
 221  * returns bool success.
 222  */
 223 static inline bool ip_vs_conn_unlink(struct ip_vs_conn *cp)
 224 {
 225         unsigned int hash;
 226         bool ret;
 227
 228         hash = ip_vs_conn_hashkey_conn(cp);
 229
 230         ct_write_lock(hash);
 231         spin_lock(&cp->lock);
 232
 233         if (cp->flags & IP_VS_CONN_F_HASHED) {
 234                 ret = false;
 235                 /* Decrease refcnt and unlink conn only if we are last user */
 236                 if (atomic_cmpxchg(&cp->refcnt, 1, 0) == 1) {
 237                         hlist_del_rcu(&cp->c_list);
 238                         cp->flags &= ~IP_VS_CONN_F_HASHED;
 239                         ret = true;
 240                 }
 241         } else
 242                 ret = atomic_read(&cp->refcnt) ? false : true;
 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 int hash;
 261         struct ip_vs_conn *cp;
 262
 263         hash = ip_vs_conn_hashkey_param(p, false);
 264
 265         rcu_read_lock();
 266
 267         hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
 268                 if (cp->af == p->af &&
 269                     p->cport == cp->cport && p->vport == cp->vport &&
 270                     ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
 271                     ip_vs_addr_equal(p->af, p->vaddr, &cp->vaddr) &&
 272                     ((!p->cport) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
 273                     p->protocol == cp->protocol &&
 274                     ip_vs_conn_net_eq(cp, p->net)) {
 275                         if (!__ip_vs_conn_get(cp))
 276                                 continue;
 277                         /* HIT */
 278                         rcu_read_unlock();
 279                         return cp;
 280                 }
 281         }
 282
 283         rcu_read_unlock();
 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                             int inverse, struct ip_vs_conn_param *p)
 312 {
 313         __be16 _ports[2], *pptr;
 314         struct net *net = skb_net(skb);
 315
 316         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 317         if (pptr == NULL)
 318                 return 1;
 319
 320         if (likely(!inverse))
 321                 ip_vs_conn_fill_param(net, af, iph->protocol, &iph->saddr,
 322                                       pptr[0], &iph->daddr, pptr[1], p);
 323         else
 324                 ip_vs_conn_fill_param(net, af, iph->protocol, &iph->daddr,
 325                                       pptr[1], &iph->saddr, pptr[0], p);
 326         return 0;
 327 }
 328
 329 struct ip_vs_conn *
 330 ip_vs_conn_in_get_proto(int af, const struct sk_buff *skb,
 331                         const struct ip_vs_iphdr *iph, int inverse)
 332 {
 333         struct ip_vs_conn_param p;
 334
 335         if (ip_vs_conn_fill_param_proto(af, skb, iph, inverse, &p))
 336                 return NULL;
 337
 338         return ip_vs_conn_in_get(&p);
 339 }
 340 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
 341
 342 /* Get reference to connection template */
 343 struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p)
 344 {
 345         unsigned int hash;
 346         struct ip_vs_conn *cp;
 347
 348         hash = ip_vs_conn_hashkey_param(p, false);
 349
 350         rcu_read_lock();
 351
 352         hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
 353                 if (!ip_vs_conn_net_eq(cp, p->net))
 354                         continue;
 355                 if (p->pe_data && p->pe->ct_match) {
 356                         if (p->pe == cp->pe && p->pe->ct_match(p, cp)) {
 357                                 if (__ip_vs_conn_get(cp))
 358                                         goto out;
 359                         }
 360                         continue;
 361                 }
 362
 363                 if (cp->af == p->af &&
 364                     ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
 365                     /* protocol should only be IPPROTO_IP if
 366                      * p->vaddr is a fwmark */
 367                     ip_vs_addr_equal(p->protocol == IPPROTO_IP ? AF_UNSPEC :
 368                                      p->af, p->vaddr, &cp->vaddr) &&
 369                     p->cport == cp->cport && p->vport == cp->vport &&
 370                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
 371                     p->protocol == cp->protocol) {
 372                         if (__ip_vs_conn_get(cp))
 373                                 goto out;
 374                 }
 375         }
 376         cp = NULL;
 377
 378   out:
 379         rcu_read_unlock();
 380
 381         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
 382                       ip_vs_proto_name(p->protocol),
 383                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
 384                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
 385                       cp ? "hit" : "not hit");
 386
 387         return cp;
 388 }
 389
 390 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 391  * Called for pkts coming from inside-to-OUTside.
 392  *      p->caddr, p->cport: pkt source address (inside host)
 393  *      p->vaddr, p->vport: pkt dest address (foreign host) */
 394 struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
 395 {
 396         unsigned int hash;
 397         struct ip_vs_conn *cp, *ret=NULL;
 398
 399         /*
 400          *      Check for "full" addressed entries
 401          */
 402         hash = ip_vs_conn_hashkey_param(p, true);
 403
 404         rcu_read_lock();
 405
 406         hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
 407                 if (cp->af == p->af &&
 408                     p->vport == cp->cport && p->cport == cp->dport &&
 409                     ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
 410                     ip_vs_addr_equal(p->af, p->caddr, &cp->daddr) &&
 411                     p->protocol == cp->protocol &&
 412                     ip_vs_conn_net_eq(cp, p->net)) {
 413                         if (!__ip_vs_conn_get(cp))
 414                                 continue;
 415                         /* HIT */
 416                         ret = cp;
 417                         break;
 418                 }
 419         }
 420
 421         rcu_read_unlock();
 422
 423         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
 424                       ip_vs_proto_name(p->protocol),
 425                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
 426                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
 427                       ret ? "hit" : "not hit");
 428
 429         return ret;
 430 }
 431
 432 struct ip_vs_conn *
 433 ip_vs_conn_out_get_proto(int af, const struct sk_buff *skb,
 434                          const struct ip_vs_iphdr *iph, int inverse)
 435 {
 436         struct ip_vs_conn_param p;
 437
 438         if (ip_vs_conn_fill_param_proto(af, skb, iph, inverse, &p))
 439                 return NULL;
 440
 441         return ip_vs_conn_out_get(&p);
 442 }
 443 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
 444
 445 /*
 446  *      Put back the conn and restart its timer with its timeout
 447  */
 448 void ip_vs_conn_put(struct ip_vs_conn *cp)
 449 {
 450         unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
 451                 0 : cp->timeout;
 452         mod_timer(&cp->timer, jiffies+t);
 453
 454         __ip_vs_conn_put(cp);
 455 }
 456
 457
 458 /*
 459  *      Fill a no_client_port connection with a client port number
 460  */
 461 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
 462 {
 463         if (ip_vs_conn_unhash(cp)) {
 464                 spin_lock(&cp->lock);
 465                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
 466                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 467                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
 468                         cp->cport = cport;
 469                 }
 470                 spin_unlock(&cp->lock);
 471
 472                 /* hash on new dport */
 473                 ip_vs_conn_hash(cp);
 474         }
 475 }
 476
 477
 478 /*
 479  *      Bind a connection entry with the corresponding packet_xmit.
 480  *      Called by ip_vs_conn_new.
 481  */
 482 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
 483 {
 484         switch (IP_VS_FWD_METHOD(cp)) {
 485         case IP_VS_CONN_F_MASQ:
 486                 cp->packet_xmit = ip_vs_nat_xmit;
 487                 break;
 488
 489         case IP_VS_CONN_F_TUNNEL:
 490                 cp->packet_xmit = ip_vs_tunnel_xmit;
 491                 break;
 492
 493         case IP_VS_CONN_F_DROUTE:
 494                 cp->packet_xmit = ip_vs_dr_xmit;
 495                 break;
 496
 497         case IP_VS_CONN_F_LOCALNODE:
 498                 cp->packet_xmit = ip_vs_null_xmit;
 499                 break;
 500
 501         case IP_VS_CONN_F_BYPASS:
 502                 cp->packet_xmit = ip_vs_bypass_xmit;
 503                 break;
 504         }
 505 }
 506
 507 #ifdef CONFIG_IP_VS_IPV6
 508 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
 509 {
 510         switch (IP_VS_FWD_METHOD(cp)) {
 511         case IP_VS_CONN_F_MASQ:
 512                 cp->packet_xmit = ip_vs_nat_xmit_v6;
 513                 break;
 514
 515         case IP_VS_CONN_F_TUNNEL:
 516                 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 517                 break;
 518
 519         case IP_VS_CONN_F_DROUTE:
 520                 cp->packet_xmit = ip_vs_dr_xmit_v6;
 521                 break;
 522
 523         case IP_VS_CONN_F_LOCALNODE:
 524                 cp->packet_xmit = ip_vs_null_xmit;
 525                 break;
 526
 527         case IP_VS_CONN_F_BYPASS:
 528                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
 529                 break;
 530         }
 531 }
 532 #endif
 533
 534
 535 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
 536 {
 537         return atomic_read(&dest->activeconns)
 538                 + atomic_read(&dest->inactconns);
 539 }
 540
 541 /*
 542  *      Bind a connection entry with a virtual service destination
 543  *      Called just after a new connection entry is created.
 544  */
 545 static inline void
 546 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
 547 {
 548         unsigned int conn_flags;
 549         __u32 flags;
 550
 551         /* if dest is NULL, then return directly */
 552         if (!dest)
 553                 return;
 554
 555         /* Increase the refcnt counter of the dest */
 556         atomic_inc(&dest->refcnt);
 557
 558         conn_flags = atomic_read(&dest->conn_flags);
 559         if (cp->protocol != IPPROTO_UDP)
 560                 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
 561         flags = cp->flags;
 562         /* Bind with the destination and its corresponding transmitter */
 563         if (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 (!(flags & IP_VS_CONN_F_TEMPLATE))
 568                         conn_flags &= ~IP_VS_CONN_F_INACTIVE;
 569                 /* connections inherit forwarding method from dest */
 570                 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
 571         }
 572         flags |= conn_flags;
 573         cp->flags = flags;
 574         cp->dest = dest;
 575
 576         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
 577                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 578                       "dest->refcnt:%d\n",
 579                       ip_vs_proto_name(cp->protocol),
 580                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 581                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 582                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
 583                       ip_vs_fwd_tag(cp), cp->state,
 584                       cp->flags, atomic_read(&cp->refcnt),
 585                       atomic_read(&dest->refcnt));
 586
 587         /* Update the connection counters */
 588         if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 589                 /* It is a normal connection, so modify the counters
 590                  * according to the flags, later the protocol can
 591                  * update them on state change
 592                  */
 593                 if (!(flags & IP_VS_CONN_F_INACTIVE))
 594                         atomic_inc(&dest->activeconns);
 595                 else
 596                         atomic_inc(&dest->inactconns);
 597         } else {
 598                 /* It is a persistent connection/template, so increase
 599                    the persistent connection counter */
 600                 atomic_inc(&dest->persistconns);
 601         }
 602
 603         if (dest->u_threshold != 0 &&
 604             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
 605                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
 606 }
 607
 608
 609 /*
 610  * Check if there is a destination for the connection, if so
 611  * bind the connection to the destination.
 612  */
 613 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
 614 {
 615         struct ip_vs_dest *dest;
 616
 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, cp->flags);
 620         if (dest) {
 621                 struct ip_vs_proto_data *pd;
 622
 623                 spin_lock(&cp->lock);
 624                 if (cp->dest) {
 625                         spin_unlock(&cp->lock);
 626                         return dest;
 627                 }
 628
 629                 /* Applications work depending on the forwarding method
 630                  * but better to reassign them always when binding dest */
 631                 if (cp->app)
 632                         ip_vs_unbind_app(cp);
 633
 634                 ip_vs_bind_dest(cp, dest);
 635                 spin_unlock(&cp->lock);
 636
 637                 /* Update its packet transmitter */
 638                 cp->packet_xmit = NULL;
 639 #ifdef CONFIG_IP_VS_IPV6
 640                 if (cp->af == AF_INET6)
 641                         ip_vs_bind_xmit_v6(cp);
 642                 else
 643 #endif
 644                         ip_vs_bind_xmit(cp);
 645
 646                 pd = ip_vs_proto_data_get(ip_vs_conn_net(cp), cp->protocol);
 647                 if (pd && atomic_read(&pd->appcnt))
 648                         ip_vs_bind_app(cp, pd->pp);
 649         }
 650         return dest;
 651 }
 652
 653
 654 /*
 655  *      Unbind a connection entry with its VS destination
 656  *      Called by the ip_vs_conn_expire function.
 657  */
 658 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
 659 {
 660         struct ip_vs_dest *dest = cp->dest;
 661
 662         if (!dest)
 663                 return;
 664
 665         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
 666                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 667                       "dest->refcnt:%d\n",
 668                       ip_vs_proto_name(cp->protocol),
 669                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 670                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 671                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
 672                       ip_vs_fwd_tag(cp), cp->state,
 673                       cp->flags, atomic_read(&cp->refcnt),
 674                       atomic_read(&dest->refcnt));
 675
 676         /* Update the connection counters */
 677         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 678                 /* It is a normal connection, so decrease the inactconns
 679                    or activeconns counter */
 680                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
 681                         atomic_dec(&dest->inactconns);
 682                 } else {
 683                         atomic_dec(&dest->activeconns);
 684                 }
 685         } else {
 686                 /* It is a persistent connection/template, so decrease
 687                    the persistent connection counter */
 688                 atomic_dec(&dest->persistconns);
 689         }
 690
 691         if (dest->l_threshold != 0) {
 692                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
 693                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 694         } else if (dest->u_threshold != 0) {
 695                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
 696                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 697         } else {
 698                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 699                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 700         }
 701
 702         /*
 703          * Simply decrease the refcnt of the dest, because the
 704          * dest will be either in service's destination list
 705          * or in the trash.
 706          */
 707         atomic_dec(&dest->refcnt);
 708 }
 709
 710 static int expire_quiescent_template(struct netns_ipvs *ipvs,
 711                                      struct ip_vs_dest *dest)
 712 {
 713 #ifdef CONFIG_SYSCTL
 714         return ipvs->sysctl_expire_quiescent_template &&
 715                 (atomic_read(&dest->weight) == 0);
 716 #else
 717         return 0;
 718 #endif
 719 }
 720
 721 /*
 722  *      Checking if the destination of a connection template is available.
 723  *      If available, return 1, otherwise invalidate this connection
 724  *      template and return 0.
 725  */
 726 int ip_vs_check_template(struct ip_vs_conn *ct)
 727 {
 728         struct ip_vs_dest *dest = ct->dest;
 729         struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(ct));
 730
 731         /*
 732          * Checking the dest server status.
 733          */
 734         if ((dest == NULL) ||
 735             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
 736             expire_quiescent_template(ipvs, dest)) {
 737                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
 738                               "protocol %s s:%s:%d v:%s:%d "
 739                               "-> d:%s:%d\n",
 740                               ip_vs_proto_name(ct->protocol),
 741                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
 742                               ntohs(ct->cport),
 743                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
 744                               ntohs(ct->vport),
 745                               IP_VS_DBG_ADDR(ct->af, &ct->daddr),
 746                               ntohs(ct->dport));
 747
 748                 /*
 749                  * Invalidate the connection template
 750                  */
 751                 if (ct->vport != htons(0xffff)) {
 752                         if (ip_vs_conn_unhash(ct)) {
 753                                 ct->dport = htons(0xffff);
 754                                 ct->vport = htons(0xffff);
 755                                 ct->cport = 0;
 756                                 ip_vs_conn_hash(ct);
 757                         }
 758                 }
 759
 760                 /*
 761                  * Simply decrease the refcnt of the template,
 762                  * don't restart its timer.
 763                  */
 764                 __ip_vs_conn_put(ct);
 765                 return 0;
 766         }
 767         return 1;
 768 }
 769
 770 static void ip_vs_conn_rcu_free(struct rcu_head *head)
 771 {
 772         struct ip_vs_conn *cp = container_of(head, struct ip_vs_conn,
 773                                              rcu_head);
 774
 775         ip_vs_pe_put(cp->pe);
 776         kfree(cp->pe_data);
 777         kmem_cache_free(ip_vs_conn_cachep, cp);
 778 }
 779
 780 static void ip_vs_conn_expire(unsigned long data)
 781 {
 782         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
 783         struct net *net = ip_vs_conn_net(cp);
 784         struct netns_ipvs *ipvs = net_ipvs(net);
 785
 786         /*
 787          *      do I control anybody?
 788          */
 789         if (atomic_read(&cp->n_control))
 790                 goto expire_later;
 791
 792         /* Unlink conn if not referenced anymore */
 793         if (likely(ip_vs_conn_unlink(cp))) {
 794                 /* delete the timer if it is activated by other users */
 795                 del_timer(&cp->timer);
 796
 797                 /* does anybody control me? */
 798                 if (cp->control)
 799                         ip_vs_control_del(cp);
 800
 801                 if (cp->flags & IP_VS_CONN_F_NFCT) {
 802                         ip_vs_conn_drop_conntrack(cp);
 803                         /* Do not access conntracks during subsys cleanup
 804                          * because nf_conntrack_find_get can not be used after
 805                          * conntrack cleanup for the net.
 806                          */
 807                         smp_rmb();
 808                         if (ipvs->enable)
 809                                 ip_vs_conn_drop_conntrack(cp);
 810                 }
 811
 812                 if (unlikely(cp->app != NULL))
 813                         ip_vs_unbind_app(cp);
 814                 ip_vs_unbind_dest(cp);
 815                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
 816                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 817                 call_rcu(&cp->rcu_head, ip_vs_conn_rcu_free);
 818                 atomic_dec(&ipvs->conn_count);
 819                 return;
 820         }
 821
 822   expire_later:
 823         IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
 824                   atomic_read(&cp->refcnt),
 825                   atomic_read(&cp->n_control));
 826
 827         atomic_inc(&cp->refcnt);
 828         cp->timeout = 60*HZ;
 829
 830         if (ipvs->sync_state & IP_VS_STATE_MASTER)
 831                 ip_vs_sync_conn(net, cp, sysctl_sync_threshold(ipvs));
 832
 833         ip_vs_conn_put(cp);
 834 }
 835
 836 /* Modify timer, so that it expires as soon as possible.
 837  * Can be called without reference only if under RCU lock.
 838  */
 839 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
 840 {
 841         /* Using mod_timer_pending will ensure the timer is not
 842          * modified after the final del_timer in ip_vs_conn_expire.
 843          */
 844         if (timer_pending(&cp->timer) &&
 845             time_after(cp->timer.expires, jiffies))
 846                 mod_timer_pending(&cp->timer, jiffies);
 847 }
 848
 849
 850 /*
 851  *      Create a new connection entry and hash it into the ip_vs_conn_tab
 852  */
 853 struct ip_vs_conn *
 854 ip_vs_conn_new(const struct ip_vs_conn_param *p,
 855                const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
 856                struct ip_vs_dest *dest, __u32 fwmark)
 857 {
 858         struct ip_vs_conn *cp;
 859         struct netns_ipvs *ipvs = net_ipvs(p->net);
 860         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->net,
 861                                                            p->protocol);
 862
 863         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
 864         if (cp == NULL) {
 865                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
 866                 return NULL;
 867         }
 868
 869         INIT_HLIST_NODE(&cp->c_list);
 870         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
 871         ip_vs_conn_net_set(cp, p->net);
 872         cp->af             = p->af;
 873         cp->protocol       = p->protocol;
 874         ip_vs_addr_copy(p->af, &cp->caddr, p->caddr);
 875         cp->cport          = p->cport;
 876         ip_vs_addr_copy(p->af, &cp->vaddr, p->vaddr);
 877         cp->vport          = p->vport;
 878         /* proto should only be IPPROTO_IP if d_addr is a fwmark */
 879         ip_vs_addr_copy(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
 880                         &cp->daddr, daddr);
 881         cp->dport          = dport;
 882         cp->flags          = flags;
 883         cp->fwmark         = fwmark;
 884         if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
 885                 ip_vs_pe_get(p->pe);
 886                 cp->pe = p->pe;
 887                 cp->pe_data = p->pe_data;
 888                 cp->pe_data_len = p->pe_data_len;
 889         }
 890         spin_lock_init(&cp->lock);
 891
 892         /*
 893          * Set the entry is referenced by the current thread before hashing
 894          * it in the table, so that other thread run ip_vs_random_dropentry
 895          * but cannot drop this entry.
 896          */
 897         atomic_set(&cp->refcnt, 1);
 898
 899         atomic_set(&cp->n_control, 0);
 900         atomic_set(&cp->in_pkts, 0);
 901
 902         atomic_inc(&ipvs->conn_count);
 903         if (flags & IP_VS_CONN_F_NO_CPORT)
 904                 atomic_inc(&ip_vs_conn_no_cport_cnt);
 905
 906         /* Bind the connection with a destination server */
 907         ip_vs_bind_dest(cp, dest);
 908
 909         /* Set its state and timeout */
 910         cp->state = 0;
 911         cp->timeout = 3*HZ;
 912         cp->sync_endtime = jiffies & ~3UL;
 913
 914         /* Bind its packet transmitter */
 915 #ifdef CONFIG_IP_VS_IPV6
 916         if (p->af == AF_INET6)
 917                 ip_vs_bind_xmit_v6(cp);
 918         else
 919 #endif
 920                 ip_vs_bind_xmit(cp);
 921
 922         if (unlikely(pd && atomic_read(&pd->appcnt)))
 923                 ip_vs_bind_app(cp, pd->pp);
 924
 925         /*
 926          * Allow conntrack to be preserved. By default, conntrack
 927          * is created and destroyed for every packet.
 928          * Sometimes keeping conntrack can be useful for
 929          * IP_VS_CONN_F_ONE_PACKET too.
 930          */
 931
 932         if (ip_vs_conntrack_enabled(ipvs))
 933                 cp->flags |= IP_VS_CONN_F_NFCT;
 934
 935         /* Hash it in the ip_vs_conn_tab finally */
 936         ip_vs_conn_hash(cp);
 937
 938         return cp;
 939 }
 940
 941 /*
 942  *      /proc/net/ip_vs_conn entries
 943  */
 944 #ifdef CONFIG_PROC_FS
 945 struct ip_vs_iter_state {
 946         struct seq_net_private  p;
 947         struct hlist_head       *l;
 948 };
 949
 950 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
 951 {
 952         int idx;
 953         struct ip_vs_conn *cp;
 954         struct ip_vs_iter_state *iter = seq->private;
 955
 956         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
 957                 rcu_read_lock();
 958                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
 959                         /* __ip_vs_conn_get() is not needed by
 960                          * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
 961                          */
 962                         if (pos-- == 0) {
 963                                 iter->l = &ip_vs_conn_tab[idx];
 964                                 return cp;
 965                         }
 966                 }
 967                 rcu_read_unlock();
 968         }
 969
 970         return NULL;
 971 }
 972
 973 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
 974 {
 975         struct ip_vs_iter_state *iter = seq->private;
 976
 977         iter->l = NULL;
 978         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
 979 }
 980
 981 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 982 {
 983         struct ip_vs_conn *cp = v;
 984         struct ip_vs_iter_state *iter = seq->private;
 985         struct hlist_node *e;
 986         struct hlist_head *l = iter->l;
 987         int idx;
 988
 989         ++*pos;
 990         if (v == SEQ_START_TOKEN)
 991                 return ip_vs_conn_array(seq, 0);
 992
 993         /* more on same hash chain? */
 994         e = rcu_dereference(hlist_next_rcu(&cp->c_list));
 995         if (e)
 996                 return hlist_entry(e, struct ip_vs_conn, c_list);
 997         rcu_read_unlock();
 998
 999         idx = l - ip_vs_conn_tab;
1000         while (++idx < ip_vs_conn_tab_size) {
1001                 rcu_read_lock();
1002                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1003                         iter->l = &ip_vs_conn_tab[idx];
1004                         return cp;
1005                 }
1006                 rcu_read_unlock();
1007         }
1008         iter->l = NULL;
1009         return NULL;
1010 }
1011
1012 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1013 {
1014         struct ip_vs_iter_state *iter = seq->private;
1015         struct hlist_head *l = iter->l;
1016
1017         if (l)
1018                 rcu_read_unlock();
1019 }
1020
1021 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1022 {
1023
1024         if (v == SEQ_START_TOKEN)
1025                 seq_puts(seq,
1026    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires PEName PEData\n");
1027         else {
1028                 const struct ip_vs_conn *cp = v;
1029                 struct net *net = seq_file_net(seq);
1030                 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1031                 size_t len = 0;
1032
1033                 if (!ip_vs_conn_net_eq(cp, net))
1034                         return 0;
1035                 if (cp->pe_data) {
1036                         pe_data[0] = ' ';
1037                         len = strlen(cp->pe->name);
1038                         memcpy(pe_data + 1, cp->pe->name, len);
1039                         pe_data[len + 1] = ' ';
1040                         len += 2;
1041                         len += cp->pe->show_pe_data(cp, pe_data + len);
1042                 }
1043                 pe_data[len] = '\0';
1044
1045 #ifdef CONFIG_IP_VS_IPV6
1046                 if (cp->af == AF_INET6)
1047                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1048                                 "%pI6 %04X %-11s %7lu%s\n",
1049                                 ip_vs_proto_name(cp->protocol),
1050                                 &cp->caddr.in6, ntohs(cp->cport),
1051                                 &cp->vaddr.in6, ntohs(cp->vport),
1052                                 &cp->daddr.in6, ntohs(cp->dport),
1053                                 ip_vs_state_name(cp->protocol, cp->state),
1054                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1055                 else
1056 #endif
1057                         seq_printf(seq,
1058                                 "%-3s %08X %04X %08X %04X"
1059                                 " %08X %04X %-11s %7lu%s\n",
1060                                 ip_vs_proto_name(cp->protocol),
1061                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1062                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1063                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
1064                                 ip_vs_state_name(cp->protocol, cp->state),
1065                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1066         }
1067         return 0;
1068 }
1069
1070 static const struct seq_operations ip_vs_conn_seq_ops = {
1071         .start = ip_vs_conn_seq_start,
1072         .next  = ip_vs_conn_seq_next,
1073         .stop  = ip_vs_conn_seq_stop,
1074         .show  = ip_vs_conn_seq_show,
1075 };
1076
1077 static int ip_vs_conn_open(struct inode *inode, struct file *file)
1078 {
1079         return seq_open_net(inode, file, &ip_vs_conn_seq_ops,
1080                             sizeof(struct ip_vs_iter_state));
1081 }
1082
1083 static const struct file_operations ip_vs_conn_fops = {
1084         .owner   = THIS_MODULE,
1085         .open    = ip_vs_conn_open,
1086         .read    = seq_read,
1087         .llseek  = seq_lseek,
1088         .release = seq_release_net,
1089 };
1090
1091 static const char *ip_vs_origin_name(unsigned int flags)
1092 {
1093         if (flags & IP_VS_CONN_F_SYNC)
1094                 return "SYNC";
1095         else
1096                 return "LOCAL";
1097 }
1098
1099 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1100 {
1101
1102         if (v == SEQ_START_TOKEN)
1103                 seq_puts(seq,
1104    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
1105         else {
1106                 const struct ip_vs_conn *cp = v;
1107                 struct net *net = seq_file_net(seq);
1108
1109                 if (!ip_vs_conn_net_eq(cp, net))
1110                         return 0;
1111
1112 #ifdef CONFIG_IP_VS_IPV6
1113                 if (cp->af == AF_INET6)
1114                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
1115                                 ip_vs_proto_name(cp->protocol),
1116                                 &cp->caddr.in6, ntohs(cp->cport),
1117                                 &cp->vaddr.in6, ntohs(cp->vport),
1118                                 &cp->daddr.in6, ntohs(cp->dport),
1119                                 ip_vs_state_name(cp->protocol, cp->state),
1120                                 ip_vs_origin_name(cp->flags),
1121                                 (cp->timer.expires-jiffies)/HZ);
1122                 else
1123 #endif
1124                         seq_printf(seq,
1125                                 "%-3s %08X %04X %08X %04X "
1126                                 "%08X %04X %-11s %-6s %7lu\n",
1127                                 ip_vs_proto_name(cp->protocol),
1128                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1129                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1130                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
1131                                 ip_vs_state_name(cp->protocol, cp->state),
1132                                 ip_vs_origin_name(cp->flags),
1133                                 (cp->timer.expires-jiffies)/HZ);
1134         }
1135         return 0;
1136 }
1137
1138 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1139         .start = ip_vs_conn_seq_start,
1140         .next  = ip_vs_conn_seq_next,
1141         .stop  = ip_vs_conn_seq_stop,
1142         .show  = ip_vs_conn_sync_seq_show,
1143 };
1144
1145 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
1146 {
1147         return seq_open_net(inode, file, &ip_vs_conn_sync_seq_ops,
1148                             sizeof(struct ip_vs_iter_state));
1149 }
1150
1151 static const struct file_operations ip_vs_conn_sync_fops = {
1152         .owner   = THIS_MODULE,
1153         .open    = ip_vs_conn_sync_open,
1154         .read    = seq_read,
1155         .llseek  = seq_lseek,
1156         .release = seq_release_net,
1157 };
1158
1159 #endif
1160
1161
1162 /*
1163  *      Randomly drop connection entries before running out of memory
1164  */
1165 static inline int todrop_entry(struct ip_vs_conn *cp)
1166 {
1167         /*
1168          * The drop rate array needs tuning for real environments.
1169          * Called from timer bh only => no locking
1170          */
1171         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1172         static char todrop_counter[9] = {0};
1173         int i;
1174
1175         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1176            This will leave enough time for normal connection to get
1177            through. */
1178         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1179                 return 0;
1180
1181         /* Don't drop the entry if its number of incoming packets is not
1182            located in [0, 8] */
1183         i = atomic_read(&cp->in_pkts);
1184         if (i > 8 || i < 0) return 0;
1185
1186         if (!todrop_rate[i]) return 0;
1187         if (--todrop_counter[i] > 0) return 0;
1188
1189         todrop_counter[i] = todrop_rate[i];
1190         return 1;
1191 }
1192
1193 /* Called from keventd and must protect itself from softirqs */
1194 void ip_vs_random_dropentry(struct net *net)
1195 {
1196         int idx;
1197         struct ip_vs_conn *cp, *cp_c;
1198
1199         /*
1200          * Randomly scan 1/32 of the whole table every second
1201          */
1202         for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1203                 unsigned int hash = net_random() & ip_vs_conn_tab_mask;
1204
1205                 /*
1206                  *  Lock is actually needed in this loop.
1207                  */
1208                 rcu_read_lock();
1209
1210                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
1211                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1212                                 /* connection template */
1213                                 continue;
1214                         if (!ip_vs_conn_net_eq(cp, net))
1215                                 continue;
1216                         if (cp->protocol == IPPROTO_TCP) {
1217                                 switch(cp->state) {
1218                                 case IP_VS_TCP_S_SYN_RECV:
1219                                 case IP_VS_TCP_S_SYNACK:
1220                                         break;
1221
1222                                 case IP_VS_TCP_S_ESTABLISHED:
1223                                         if (todrop_entry(cp))
1224                                                 break;
1225                                         continue;
1226
1227                                 default:
1228                                         continue;
1229                                 }
1230                         } else {
1231                                 if (!todrop_entry(cp))
1232                                         continue;
1233                         }
1234
1235                         IP_VS_DBG(4, "del connection\n");
1236                         ip_vs_conn_expire_now(cp);
1237                         cp_c = cp->control;
1238                         /* cp->control is valid only with reference to cp */
1239                         if (cp_c && __ip_vs_conn_get(cp)) {
1240                                 IP_VS_DBG(4, "del conn template\n");
1241                                 ip_vs_conn_expire_now(cp_c);
1242                                 __ip_vs_conn_put(cp);
1243                         }
1244                 }
1245                 rcu_read_unlock();
1246         }
1247 }
1248
1249
1250 /*
1251  *      Flush all the connection entries in the ip_vs_conn_tab
1252  */
1253 static void ip_vs_conn_flush(struct net *net)
1254 {
1255         int idx;
1256         struct ip_vs_conn *cp, *cp_c;
1257         struct netns_ipvs *ipvs = net_ipvs(net);
1258
1259 flush_again:
1260         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1261                 /*
1262                  *  Lock is actually needed in this loop.
1263                  */
1264                 rcu_read_lock();
1265
1266                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1267                         if (!ip_vs_conn_net_eq(cp, net))
1268                                 continue;
1269                         IP_VS_DBG(4, "del connection\n");
1270                         ip_vs_conn_expire_now(cp);
1271                         cp_c = cp->control;
1272                         /* cp->control is valid only with reference to cp */
1273                         if (cp_c && __ip_vs_conn_get(cp)) {
1274                                 IP_VS_DBG(4, "del conn template\n");
1275                                 ip_vs_conn_expire_now(cp_c);
1276                                 __ip_vs_conn_put(cp);
1277                         }
1278                 }
1279                 rcu_read_unlock();
1280         }
1281
1282         /* the counter may be not NULL, because maybe some conn entries
1283            are run by slow timer handler or unhashed but still referred */
1284         if (atomic_read(&ipvs->conn_count) != 0) {
1285                 schedule();
1286                 goto flush_again;
1287         }
1288 }
1289 /*
1290  * per netns init and exit
1291  */
1292 int __net_init ip_vs_conn_net_init(struct net *net)
1293 {
1294         struct netns_ipvs *ipvs = net_ipvs(net);
1295
1296         atomic_set(&ipvs->conn_count, 0);
1297
1298         proc_create("ip_vs_conn", 0, net->proc_net, &ip_vs_conn_fops);
1299         proc_create("ip_vs_conn_sync", 0, net->proc_net, &ip_vs_conn_sync_fops);
1300         return 0;
1301 }
1302
1303 void __net_exit ip_vs_conn_net_cleanup(struct net *net)
1304 {
1305         /* flush all the connection entries first */
1306         ip_vs_conn_flush(net);
1307         remove_proc_entry("ip_vs_conn", net->proc_net);
1308         remove_proc_entry("ip_vs_conn_sync", net->proc_net);
1309 }
1310
1311 int __init ip_vs_conn_init(void)
1312 {
1313         int idx;
1314
1315         /* Compute size and mask */
1316         ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1317         ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1318
1319         /*
1320          * Allocate the connection hash table and initialize its list heads
1321          */
1322         ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size * sizeof(*ip_vs_conn_tab));
1323         if (!ip_vs_conn_tab)
1324                 return -ENOMEM;
1325
1326         /* Allocate ip_vs_conn slab cache */
1327         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1328                                               sizeof(struct ip_vs_conn), 0,
1329                                               SLAB_HWCACHE_ALIGN, NULL);
1330         if (!ip_vs_conn_cachep) {
1331                 vfree(ip_vs_conn_tab);
1332                 return -ENOMEM;
1333         }
1334
1335         pr_info("Connection hash table configured "
1336                 "(size=%d, memory=%ldKbytes)\n",
1337                 ip_vs_conn_tab_size,
1338                 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1339         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1340                   sizeof(struct ip_vs_conn));
1341
1342         for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1343                 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1344
1345         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1346                 spin_lock_init(&__ip_vs_conntbl_lock_array[idx].l);
1347         }
1348
1349         /* calculate the random value for connection hash */
1350         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1351
1352         return 0;
1353 }
1354
1355 void ip_vs_conn_cleanup(void)
1356 {
1357         /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1358         rcu_barrier();
1359         /* Release the empty cache */
1360         kmem_cache_destroy(ip_vs_conn_cachep);
1361         vfree(ip_vs_conn_tab);
1362 }