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 (p->cport == cp->cport && p->vport == cp->vport &&
 269                     cp->af == p->af &&
 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 (unlikely(p->pe_data && p->pe->ct_match)) {
 354                         if (!ip_vs_conn_net_eq(cp, p->net))
 355                                 continue;
 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->vport == cp->vport && p->cport == cp->cport &&
 370                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
 371                     p->protocol == cp->protocol &&
 372                     ip_vs_conn_net_eq(cp, p->net)) {
 373                         if (__ip_vs_conn_get(cp))
 374                                 goto out;
 375                 }
 376         }
 377         cp = NULL;
 378
 379   out:
 380         rcu_read_unlock();
 381
 382         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
 383                       ip_vs_proto_name(p->protocol),
 384                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
 385                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
 386                       cp ? "hit" : "not hit");
 387
 388         return cp;
 389 }
 390
 391 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 392  * Called for pkts coming from inside-to-OUTside.
 393  *      p->caddr, p->cport: pkt source address (inside host)
 394  *      p->vaddr, p->vport: pkt dest address (foreign host) */
 395 struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
 396 {
 397         unsigned int hash;
 398         struct ip_vs_conn *cp, *ret=NULL;
 399
 400         /*
 401          *      Check for "full" addressed entries
 402          */
 403         hash = ip_vs_conn_hashkey_param(p, true);
 404
 405         rcu_read_lock();
 406
 407         hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
 408                 if (p->vport == cp->cport && p->cport == cp->dport &&
 409                     cp->af == p->af &&
 410                     ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
 411                     ip_vs_addr_equal(p->af, p->caddr, &cp->daddr) &&
 412                     p->protocol == cp->protocol &&
 413                     ip_vs_conn_net_eq(cp, p->net)) {
 414                         if (!__ip_vs_conn_get(cp))
 415                                 continue;
 416                         /* HIT */
 417                         ret = cp;
 418                         break;
 419                 }
 420         }
 421
 422         rcu_read_unlock();
 423
 424         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
 425                       ip_vs_proto_name(p->protocol),
 426                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
 427                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
 428                       ret ? "hit" : "not hit");
 429
 430         return ret;
 431 }
 432
 433 struct ip_vs_conn *
 434 ip_vs_conn_out_get_proto(int af, const struct sk_buff *skb,
 435                          const struct ip_vs_iphdr *iph, int inverse)
 436 {
 437         struct ip_vs_conn_param p;
 438
 439         if (ip_vs_conn_fill_param_proto(af, skb, iph, inverse, &p))
 440                 return NULL;
 441
 442         return ip_vs_conn_out_get(&p);
 443 }
 444 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
 445
 446 /*
 447  *      Put back the conn and restart its timer with its timeout
 448  */
 449 void ip_vs_conn_put(struct ip_vs_conn *cp)
 450 {
 451         unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
 452                 0 : cp->timeout;
 453         mod_timer(&cp->timer, jiffies+t);
 454
 455         __ip_vs_conn_put(cp);
 456 }
 457
 458
 459 /*
 460  *      Fill a no_client_port connection with a client port number
 461  */
 462 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
 463 {
 464         if (ip_vs_conn_unhash(cp)) {
 465                 spin_lock(&cp->lock);
 466                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
 467                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 468                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
 469                         cp->cport = cport;
 470                 }
 471                 spin_unlock(&cp->lock);
 472
 473                 /* hash on new dport */
 474                 ip_vs_conn_hash(cp);
 475         }
 476 }
 477
 478
 479 /*
 480  *      Bind a connection entry with the corresponding packet_xmit.
 481  *      Called by ip_vs_conn_new.
 482  */
 483 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
 484 {
 485         switch (IP_VS_FWD_METHOD(cp)) {
 486         case IP_VS_CONN_F_MASQ:
 487                 cp->packet_xmit = ip_vs_nat_xmit;
 488                 break;
 489
 490         case IP_VS_CONN_F_TUNNEL:
 491                 cp->packet_xmit = ip_vs_tunnel_xmit;
 492                 break;
 493
 494         case IP_VS_CONN_F_DROUTE:
 495                 cp->packet_xmit = ip_vs_dr_xmit;
 496                 break;
 497
 498         case IP_VS_CONN_F_LOCALNODE:
 499                 cp->packet_xmit = ip_vs_null_xmit;
 500                 break;
 501
 502         case IP_VS_CONN_F_BYPASS:
 503                 cp->packet_xmit = ip_vs_bypass_xmit;
 504                 break;
 505         }
 506 }
 507
 508 #ifdef CONFIG_IP_VS_IPV6
 509 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
 510 {
 511         switch (IP_VS_FWD_METHOD(cp)) {
 512         case IP_VS_CONN_F_MASQ:
 513                 cp->packet_xmit = ip_vs_nat_xmit_v6;
 514                 break;
 515
 516         case IP_VS_CONN_F_TUNNEL:
 517                 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 518                 break;
 519
 520         case IP_VS_CONN_F_DROUTE:
 521                 cp->packet_xmit = ip_vs_dr_xmit_v6;
 522                 break;
 523
 524         case IP_VS_CONN_F_LOCALNODE:
 525                 cp->packet_xmit = ip_vs_null_xmit;
 526                 break;
 527
 528         case IP_VS_CONN_F_BYPASS:
 529                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
 530                 break;
 531         }
 532 }
 533 #endif
 534
 535
 536 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
 537 {
 538         return atomic_read(&dest->activeconns)
 539                 + atomic_read(&dest->inactconns);
 540 }
 541
 542 /*
 543  *      Bind a connection entry with a virtual service destination
 544  *      Called just after a new connection entry is created.
 545  */
 546 static inline void
 547 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
 548 {
 549         unsigned int conn_flags;
 550         __u32 flags;
 551
 552         /* if dest is NULL, then return directly */
 553         if (!dest)
 554                 return;
 555
 556         /* Increase the refcnt counter of the dest */
 557         atomic_inc(&dest->refcnt);
 558
 559         conn_flags = atomic_read(&dest->conn_flags);
 560         if (cp->protocol != IPPROTO_UDP)
 561                 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
 562         flags = cp->flags;
 563         /* Bind with the destination and its corresponding transmitter */
 564         if (flags & IP_VS_CONN_F_SYNC) {
 565                 /* if the connection is not template and is created
 566                  * by sync, preserve the activity flag.
 567                  */
 568                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 569                         conn_flags &= ~IP_VS_CONN_F_INACTIVE;
 570                 /* connections inherit forwarding method from dest */
 571                 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
 572         }
 573         flags |= conn_flags;
 574         cp->flags = flags;
 575         cp->dest = dest;
 576
 577         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
 578                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 579                       "dest->refcnt:%d\n",
 580                       ip_vs_proto_name(cp->protocol),
 581                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 582                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 583                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
 584                       ip_vs_fwd_tag(cp), cp->state,
 585                       cp->flags, atomic_read(&cp->refcnt),
 586                       atomic_read(&dest->refcnt));
 587
 588         /* Update the connection counters */
 589         if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 590                 /* It is a normal connection, so modify the counters
 591                  * according to the flags, later the protocol can
 592                  * update them on state change
 593                  */
 594                 if (!(flags & IP_VS_CONN_F_INACTIVE))
 595                         atomic_inc(&dest->activeconns);
 596                 else
 597                         atomic_inc(&dest->inactconns);
 598         } else {
 599                 /* It is a persistent connection/template, so increase
 600                    the persistent connection counter */
 601                 atomic_inc(&dest->persistconns);
 602         }
 603
 604         if (dest->u_threshold != 0 &&
 605             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
 606                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
 607 }
 608
 609
 610 /*
 611  * Check if there is a destination for the connection, if so
 612  * bind the connection to the destination.
 613  */
 614 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
 615 {
 616         struct ip_vs_dest *dest;
 617
 618         dest = ip_vs_find_dest(ip_vs_conn_net(cp), cp->af, &cp->daddr,
 619                                cp->dport, &cp->vaddr, cp->vport,
 620                                cp->protocol, cp->fwmark, cp->flags);
 621         if (dest) {
 622                 struct ip_vs_proto_data *pd;
 623
 624                 spin_lock(&cp->lock);
 625                 if (cp->dest) {
 626                         spin_unlock(&cp->lock);
 627                         return dest;
 628                 }
 629
 630                 /* Applications work depending on the forwarding method
 631                  * but better to reassign them always when binding dest */
 632                 if (cp->app)
 633                         ip_vs_unbind_app(cp);
 634
 635                 ip_vs_bind_dest(cp, dest);
 636                 spin_unlock(&cp->lock);
 637
 638                 /* Update its packet transmitter */
 639                 cp->packet_xmit = NULL;
 640 #ifdef CONFIG_IP_VS_IPV6
 641                 if (cp->af == AF_INET6)
 642                         ip_vs_bind_xmit_v6(cp);
 643                 else
 644 #endif
 645                         ip_vs_bind_xmit(cp);
 646
 647                 pd = ip_vs_proto_data_get(ip_vs_conn_net(cp), cp->protocol);
 648                 if (pd && atomic_read(&pd->appcnt))
 649                         ip_vs_bind_app(cp, pd->pp);
 650         }
 651         return dest;
 652 }
 653
 654
 655 /*
 656  *      Unbind a connection entry with its VS destination
 657  *      Called by the ip_vs_conn_expire function.
 658  */
 659 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
 660 {
 661         struct ip_vs_dest *dest = cp->dest;
 662
 663         if (!dest)
 664                 return;
 665
 666         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
 667                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 668                       "dest->refcnt:%d\n",
 669                       ip_vs_proto_name(cp->protocol),
 670                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 671                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 672                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
 673                       ip_vs_fwd_tag(cp), cp->state,
 674                       cp->flags, atomic_read(&cp->refcnt),
 675                       atomic_read(&dest->refcnt));
 676
 677         /* Update the connection counters */
 678         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 679                 /* It is a normal connection, so decrease the inactconns
 680                    or activeconns counter */
 681                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
 682                         atomic_dec(&dest->inactconns);
 683                 } else {
 684                         atomic_dec(&dest->activeconns);
 685                 }
 686         } else {
 687                 /* It is a persistent connection/template, so decrease
 688                    the persistent connection counter */
 689                 atomic_dec(&dest->persistconns);
 690         }
 691
 692         if (dest->l_threshold != 0) {
 693                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
 694                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 695         } else if (dest->u_threshold != 0) {
 696                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
 697                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 698         } else {
 699                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 700                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 701         }
 702
 703         /*
 704          * Simply decrease the refcnt of the dest, because the
 705          * dest will be either in service's destination list
 706          * or in the trash.
 707          */
 708         atomic_dec(&dest->refcnt);
 709 }
 710
 711 static int expire_quiescent_template(struct netns_ipvs *ipvs,
 712                                      struct ip_vs_dest *dest)
 713 {
 714 #ifdef CONFIG_SYSCTL
 715         return ipvs->sysctl_expire_quiescent_template &&
 716                 (atomic_read(&dest->weight) == 0);
 717 #else
 718         return 0;
 719 #endif
 720 }
 721
 722 /*
 723  *      Checking if the destination of a connection template is available.
 724  *      If available, return 1, otherwise invalidate this connection
 725  *      template and return 0.
 726  */
 727 int ip_vs_check_template(struct ip_vs_conn *ct)
 728 {
 729         struct ip_vs_dest *dest = ct->dest;
 730         struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(ct));
 731
 732         /*
 733          * Checking the dest server status.
 734          */
 735         if ((dest == NULL) ||
 736             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
 737             expire_quiescent_template(ipvs, dest)) {
 738                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
 739                               "protocol %s s:%s:%d v:%s:%d "
 740                               "-> d:%s:%d\n",
 741                               ip_vs_proto_name(ct->protocol),
 742                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
 743                               ntohs(ct->cport),
 744                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
 745                               ntohs(ct->vport),
 746                               IP_VS_DBG_ADDR(ct->af, &ct->daddr),
 747                               ntohs(ct->dport));
 748
 749                 /*
 750                  * Invalidate the connection template
 751                  */
 752                 if (ct->vport != htons(0xffff)) {
 753                         if (ip_vs_conn_unhash(ct)) {
 754                                 ct->dport = htons(0xffff);
 755                                 ct->vport = htons(0xffff);
 756                                 ct->cport = 0;
 757                                 ip_vs_conn_hash(ct);
 758                         }
 759                 }
 760
 761                 /*
 762                  * Simply decrease the refcnt of the template,
 763                  * don't restart its timer.
 764                  */
 765                 __ip_vs_conn_put(ct);
 766                 return 0;
 767         }
 768         return 1;
 769 }
 770
 771 static void ip_vs_conn_rcu_free(struct rcu_head *head)
 772 {
 773         struct ip_vs_conn *cp = container_of(head, struct ip_vs_conn,
 774                                              rcu_head);
 775
 776         ip_vs_pe_put(cp->pe);
 777         kfree(cp->pe_data);
 778         kmem_cache_free(ip_vs_conn_cachep, cp);
 779 }
 780
 781 static void ip_vs_conn_expire(unsigned long data)
 782 {
 783         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
 784         struct net *net = ip_vs_conn_net(cp);
 785         struct netns_ipvs *ipvs = net_ipvs(net);
 786
 787         /*
 788          *      do I control anybody?
 789          */
 790         if (atomic_read(&cp->n_control))
 791                 goto expire_later;
 792
 793         /* Unlink conn if not referenced anymore */
 794         if (likely(ip_vs_conn_unlink(cp))) {
 795                 /* delete the timer if it is activated by other users */
 796                 del_timer(&cp->timer);
 797
 798                 /* does anybody control me? */
 799                 if (cp->control)
 800                         ip_vs_control_del(cp);
 801
 802                 if (cp->flags & IP_VS_CONN_F_NFCT) {
 803                         ip_vs_conn_drop_conntrack(cp);
 804                         /* Do not access conntracks during subsys cleanup
 805                          * because nf_conntrack_find_get can not be used after
 806                          * conntrack cleanup for the net.
 807                          */
 808                         smp_rmb();
 809                         if (ipvs->enable)
 810                                 ip_vs_conn_drop_conntrack(cp);
 811                 }
 812
 813                 if (unlikely(cp->app != NULL))
 814                         ip_vs_unbind_app(cp);
 815                 ip_vs_unbind_dest(cp);
 816                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
 817                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 818                 call_rcu(&cp->rcu_head, ip_vs_conn_rcu_free);
 819                 atomic_dec(&ipvs->conn_count);
 820                 return;
 821         }
 822
 823   expire_later:
 824         IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
 825                   atomic_read(&cp->refcnt),
 826                   atomic_read(&cp->n_control));
 827
 828         atomic_inc(&cp->refcnt);
 829         cp->timeout = 60*HZ;
 830
 831         if (ipvs->sync_state & IP_VS_STATE_MASTER)
 832                 ip_vs_sync_conn(net, cp, sysctl_sync_threshold(ipvs));
 833
 834         ip_vs_conn_put(cp);
 835 }
 836
 837 /* Modify timer, so that it expires as soon as possible.
 838  * Can be called without reference only if under RCU lock.
 839  */
 840 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
 841 {
 842         /* Using mod_timer_pending will ensure the timer is not
 843          * modified after the final del_timer in ip_vs_conn_expire.
 844          */
 845         if (timer_pending(&cp->timer) &&
 846             time_after(cp->timer.expires, jiffies))
 847                 mod_timer_pending(&cp->timer, jiffies);
 848 }
 849
 850
 851 /*
 852  *      Create a new connection entry and hash it into the ip_vs_conn_tab
 853  */
 854 struct ip_vs_conn *
 855 ip_vs_conn_new(const struct ip_vs_conn_param *p,
 856                const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
 857                struct ip_vs_dest *dest, __u32 fwmark)
 858 {
 859         struct ip_vs_conn *cp;
 860         struct netns_ipvs *ipvs = net_ipvs(p->net);
 861         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->net,
 862                                                            p->protocol);
 863
 864         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
 865         if (cp == NULL) {
 866                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
 867                 return NULL;
 868         }
 869
 870         INIT_HLIST_NODE(&cp->c_list);
 871         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
 872         ip_vs_conn_net_set(cp, p->net);
 873         cp->af             = p->af;
 874         cp->protocol       = p->protocol;
 875         ip_vs_addr_copy(p->af, &cp->caddr, p->caddr);
 876         cp->cport          = p->cport;
 877         ip_vs_addr_copy(p->af, &cp->vaddr, p->vaddr);
 878         cp->vport          = p->vport;
 879         /* proto should only be IPPROTO_IP if d_addr is a fwmark */
 880         ip_vs_addr_copy(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
 881                         &cp->daddr, daddr);
 882         cp->dport          = dport;
 883         cp->flags          = flags;
 884         cp->fwmark         = fwmark;
 885         if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
 886                 ip_vs_pe_get(p->pe);
 887                 cp->pe = p->pe;
 888                 cp->pe_data = p->pe_data;
 889                 cp->pe_data_len = p->pe_data_len;
 890         }
 891         spin_lock_init(&cp->lock);
 892
 893         /*
 894          * Set the entry is referenced by the current thread before hashing
 895          * it in the table, so that other thread run ip_vs_random_dropentry
 896          * but cannot drop this entry.
 897          */
 898         atomic_set(&cp->refcnt, 1);
 899
 900         atomic_set(&cp->n_control, 0);
 901         atomic_set(&cp->in_pkts, 0);
 902
 903         atomic_inc(&ipvs->conn_count);
 904         if (flags & IP_VS_CONN_F_NO_CPORT)
 905                 atomic_inc(&ip_vs_conn_no_cport_cnt);
 906
 907         /* Bind the connection with a destination server */
 908         ip_vs_bind_dest(cp, dest);
 909
 910         /* Set its state and timeout */
 911         cp->state = 0;
 912         cp->timeout = 3*HZ;
 913         cp->sync_endtime = jiffies & ~3UL;
 914
 915         /* Bind its packet transmitter */
 916 #ifdef CONFIG_IP_VS_IPV6
 917         if (p->af == AF_INET6)
 918                 ip_vs_bind_xmit_v6(cp);
 919         else
 920 #endif
 921                 ip_vs_bind_xmit(cp);
 922
 923         if (unlikely(pd && atomic_read(&pd->appcnt)))
 924                 ip_vs_bind_app(cp, pd->pp);
 925
 926         /*
 927          * Allow conntrack to be preserved. By default, conntrack
 928          * is created and destroyed for every packet.
 929          * Sometimes keeping conntrack can be useful for
 930          * IP_VS_CONN_F_ONE_PACKET too.
 931          */
 932
 933         if (ip_vs_conntrack_enabled(ipvs))
 934                 cp->flags |= IP_VS_CONN_F_NFCT;
 935
 936         /* Hash it in the ip_vs_conn_tab finally */
 937         ip_vs_conn_hash(cp);
 938
 939         return cp;
 940 }
 941
 942 /*
 943  *      /proc/net/ip_vs_conn entries
 944  */
 945 #ifdef CONFIG_PROC_FS
 946 struct ip_vs_iter_state {
 947         struct seq_net_private  p;
 948         struct hlist_head       *l;
 949 };
 950
 951 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
 952 {
 953         int idx;
 954         struct ip_vs_conn *cp;
 955         struct ip_vs_iter_state *iter = seq->private;
 956
 957         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
 958                 rcu_read_lock();
 959                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
 960                         /* __ip_vs_conn_get() is not needed by
 961                          * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
 962                          */
 963                         if (pos-- == 0) {
 964                                 iter->l = &ip_vs_conn_tab[idx];
 965                                 return cp;
 966                         }
 967                 }
 968                 rcu_read_unlock();
 969         }
 970
 971         return NULL;
 972 }
 973
 974 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
 975 {
 976         struct ip_vs_iter_state *iter = seq->private;
 977
 978         iter->l = NULL;
 979         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
 980 }
 981
 982 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 983 {
 984         struct ip_vs_conn *cp = v;
 985         struct ip_vs_iter_state *iter = seq->private;
 986         struct hlist_node *e;
 987         struct hlist_head *l = iter->l;
 988         int idx;
 989
 990         ++*pos;
 991         if (v == SEQ_START_TOKEN)
 992                 return ip_vs_conn_array(seq, 0);
 993
 994         /* more on same hash chain? */
 995         e = rcu_dereference(hlist_next_rcu(&cp->c_list));
 996         if (e)
 997                 return hlist_entry(e, struct ip_vs_conn, c_list);
 998         rcu_read_unlock();
 999
1000         idx = l - ip_vs_conn_tab;
1001         while (++idx < ip_vs_conn_tab_size) {
1002                 rcu_read_lock();
1003                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1004                         iter->l = &ip_vs_conn_tab[idx];
1005                         return cp;
1006                 }
1007                 rcu_read_unlock();
1008         }
1009         iter->l = NULL;
1010         return NULL;
1011 }
1012
1013 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1014 {
1015         struct ip_vs_iter_state *iter = seq->private;
1016         struct hlist_head *l = iter->l;
1017
1018         if (l)
1019                 rcu_read_unlock();
1020 }
1021
1022 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1023 {
1024
1025         if (v == SEQ_START_TOKEN)
1026                 seq_puts(seq,
1027    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires PEName PEData\n");
1028         else {
1029                 const struct ip_vs_conn *cp = v;
1030                 struct net *net = seq_file_net(seq);
1031                 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1032                 size_t len = 0;
1033
1034                 if (!ip_vs_conn_net_eq(cp, net))
1035                         return 0;
1036                 if (cp->pe_data) {
1037                         pe_data[0] = ' ';
1038                         len = strlen(cp->pe->name);
1039                         memcpy(pe_data + 1, cp->pe->name, len);
1040                         pe_data[len + 1] = ' ';
1041                         len += 2;
1042                         len += cp->pe->show_pe_data(cp, pe_data + len);
1043                 }
1044                 pe_data[len] = '\0';
1045
1046 #ifdef CONFIG_IP_VS_IPV6
1047                 if (cp->af == AF_INET6)
1048                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1049                                 "%pI6 %04X %-11s %7lu%s\n",
1050                                 ip_vs_proto_name(cp->protocol),
1051                                 &cp->caddr.in6, ntohs(cp->cport),
1052                                 &cp->vaddr.in6, ntohs(cp->vport),
1053                                 &cp->daddr.in6, ntohs(cp->dport),
1054                                 ip_vs_state_name(cp->protocol, cp->state),
1055                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1056                 else
1057 #endif
1058                         seq_printf(seq,
1059                                 "%-3s %08X %04X %08X %04X"
1060                                 " %08X %04X %-11s %7lu%s\n",
1061                                 ip_vs_proto_name(cp->protocol),
1062                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1063                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1064                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
1065                                 ip_vs_state_name(cp->protocol, cp->state),
1066                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1067         }
1068         return 0;
1069 }
1070
1071 static const struct seq_operations ip_vs_conn_seq_ops = {
1072         .start = ip_vs_conn_seq_start,
1073         .next  = ip_vs_conn_seq_next,
1074         .stop  = ip_vs_conn_seq_stop,
1075         .show  = ip_vs_conn_seq_show,
1076 };
1077
1078 static int ip_vs_conn_open(struct inode *inode, struct file *file)
1079 {
1080         return seq_open_net(inode, file, &ip_vs_conn_seq_ops,
1081                             sizeof(struct ip_vs_iter_state));
1082 }
1083
1084 static const struct file_operations ip_vs_conn_fops = {
1085         .owner   = THIS_MODULE,
1086         .open    = ip_vs_conn_open,
1087         .read    = seq_read,
1088         .llseek  = seq_lseek,
1089         .release = seq_release_net,
1090 };
1091
1092 static const char *ip_vs_origin_name(unsigned int flags)
1093 {
1094         if (flags & IP_VS_CONN_F_SYNC)
1095                 return "SYNC";
1096         else
1097                 return "LOCAL";
1098 }
1099
1100 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1101 {
1102
1103         if (v == SEQ_START_TOKEN)
1104                 seq_puts(seq,
1105    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
1106         else {
1107                 const struct ip_vs_conn *cp = v;
1108                 struct net *net = seq_file_net(seq);
1109
1110                 if (!ip_vs_conn_net_eq(cp, net))
1111                         return 0;
1112
1113 #ifdef CONFIG_IP_VS_IPV6
1114                 if (cp->af == AF_INET6)
1115                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
1116                                 ip_vs_proto_name(cp->protocol),
1117                                 &cp->caddr.in6, ntohs(cp->cport),
1118                                 &cp->vaddr.in6, ntohs(cp->vport),
1119                                 &cp->daddr.in6, ntohs(cp->dport),
1120                                 ip_vs_state_name(cp->protocol, cp->state),
1121                                 ip_vs_origin_name(cp->flags),
1122                                 (cp->timer.expires-jiffies)/HZ);
1123                 else
1124 #endif
1125                         seq_printf(seq,
1126                                 "%-3s %08X %04X %08X %04X "
1127                                 "%08X %04X %-11s %-6s %7lu\n",
1128                                 ip_vs_proto_name(cp->protocol),
1129                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1130                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1131                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
1132                                 ip_vs_state_name(cp->protocol, cp->state),
1133                                 ip_vs_origin_name(cp->flags),
1134                                 (cp->timer.expires-jiffies)/HZ);
1135         }
1136         return 0;
1137 }
1138
1139 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1140         .start = ip_vs_conn_seq_start,
1141         .next  = ip_vs_conn_seq_next,
1142         .stop  = ip_vs_conn_seq_stop,
1143         .show  = ip_vs_conn_sync_seq_show,
1144 };
1145
1146 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
1147 {
1148         return seq_open_net(inode, file, &ip_vs_conn_sync_seq_ops,
1149                             sizeof(struct ip_vs_iter_state));
1150 }
1151
1152 static const struct file_operations ip_vs_conn_sync_fops = {
1153         .owner   = THIS_MODULE,
1154         .open    = ip_vs_conn_sync_open,
1155         .read    = seq_read,
1156         .llseek  = seq_lseek,
1157         .release = seq_release_net,
1158 };
1159
1160 #endif
1161
1162
1163 /*
1164  *      Randomly drop connection entries before running out of memory
1165  */
1166 static inline int todrop_entry(struct ip_vs_conn *cp)
1167 {
1168         /*
1169          * The drop rate array needs tuning for real environments.
1170          * Called from timer bh only => no locking
1171          */
1172         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1173         static char todrop_counter[9] = {0};
1174         int i;
1175
1176         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1177            This will leave enough time for normal connection to get
1178            through. */
1179         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1180                 return 0;
1181
1182         /* Don't drop the entry if its number of incoming packets is not
1183            located in [0, 8] */
1184         i = atomic_read(&cp->in_pkts);
1185         if (i > 8 || i < 0) return 0;
1186
1187         if (!todrop_rate[i]) return 0;
1188         if (--todrop_counter[i] > 0) return 0;
1189
1190         todrop_counter[i] = todrop_rate[i];
1191         return 1;
1192 }
1193
1194 /* Called from keventd and must protect itself from softirqs */
1195 void ip_vs_random_dropentry(struct net *net)
1196 {
1197         int idx;
1198         struct ip_vs_conn *cp, *cp_c;
1199
1200         /*
1201          * Randomly scan 1/32 of the whole table every second
1202          */
1203         for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1204                 unsigned int hash = net_random() & ip_vs_conn_tab_mask;
1205
1206                 /*
1207                  *  Lock is actually needed in this loop.
1208                  */
1209                 rcu_read_lock();
1210
1211                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
1212                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1213                                 /* connection template */
1214                                 continue;
1215                         if (!ip_vs_conn_net_eq(cp, net))
1216                                 continue;
1217                         if (cp->protocol == IPPROTO_TCP) {
1218                                 switch(cp->state) {
1219                                 case IP_VS_TCP_S_SYN_RECV:
1220                                 case IP_VS_TCP_S_SYNACK:
1221                                         break;
1222
1223                                 case IP_VS_TCP_S_ESTABLISHED:
1224                                         if (todrop_entry(cp))
1225                                                 break;
1226                                         continue;
1227
1228                                 default:
1229                                         continue;
1230                                 }
1231                         } else {
1232                                 if (!todrop_entry(cp))
1233                                         continue;
1234                         }
1235
1236                         IP_VS_DBG(4, "del connection\n");
1237                         ip_vs_conn_expire_now(cp);
1238                         cp_c = cp->control;
1239                         /* cp->control is valid only with reference to cp */
1240                         if (cp_c && __ip_vs_conn_get(cp)) {
1241                                 IP_VS_DBG(4, "del conn template\n");
1242                                 ip_vs_conn_expire_now(cp_c);
1243                                 __ip_vs_conn_put(cp);
1244                         }
1245                 }
1246                 rcu_read_unlock();
1247         }
1248 }
1249
1250
1251 /*
1252  *      Flush all the connection entries in the ip_vs_conn_tab
1253  */
1254 static void ip_vs_conn_flush(struct net *net)
1255 {
1256         int idx;
1257         struct ip_vs_conn *cp, *cp_c;
1258         struct netns_ipvs *ipvs = net_ipvs(net);
1259
1260 flush_again:
1261         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1262                 /*
1263                  *  Lock is actually needed in this loop.
1264                  */
1265                 rcu_read_lock();
1266
1267                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1268                         if (!ip_vs_conn_net_eq(cp, net))
1269                                 continue;
1270                         IP_VS_DBG(4, "del connection\n");
1271                         ip_vs_conn_expire_now(cp);
1272                         cp_c = cp->control;
1273                         /* cp->control is valid only with reference to cp */
1274                         if (cp_c && __ip_vs_conn_get(cp)) {
1275                                 IP_VS_DBG(4, "del conn template\n");
1276                                 ip_vs_conn_expire_now(cp_c);
1277                                 __ip_vs_conn_put(cp);
1278                         }
1279                 }
1280                 rcu_read_unlock();
1281         }
1282
1283         /* the counter may be not NULL, because maybe some conn entries
1284            are run by slow timer handler or unhashed but still referred */
1285         if (atomic_read(&ipvs->conn_count) != 0) {
1286                 schedule();
1287                 goto flush_again;
1288         }
1289 }
1290 /*
1291  * per netns init and exit
1292  */
1293 int __net_init ip_vs_conn_net_init(struct net *net)
1294 {
1295         struct netns_ipvs *ipvs = net_ipvs(net);
1296
1297         atomic_set(&ipvs->conn_count, 0);
1298
1299         proc_create("ip_vs_conn", 0, net->proc_net, &ip_vs_conn_fops);
1300         proc_create("ip_vs_conn_sync", 0, net->proc_net, &ip_vs_conn_sync_fops);
1301         return 0;
1302 }
1303
1304 void __net_exit ip_vs_conn_net_cleanup(struct net *net)
1305 {
1306         /* flush all the connection entries first */
1307         ip_vs_conn_flush(net);
1308         remove_proc_entry("ip_vs_conn", net->proc_net);
1309         remove_proc_entry("ip_vs_conn_sync", net->proc_net);
1310 }
1311
1312 int __init ip_vs_conn_init(void)
1313 {
1314         int idx;
1315
1316         /* Compute size and mask */
1317         ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1318         ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1319
1320         /*
1321          * Allocate the connection hash table and initialize its list heads
1322          */
1323         ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size * sizeof(*ip_vs_conn_tab));
1324         if (!ip_vs_conn_tab)
1325                 return -ENOMEM;
1326
1327         /* Allocate ip_vs_conn slab cache */
1328         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1329                                               sizeof(struct ip_vs_conn), 0,
1330                                               SLAB_HWCACHE_ALIGN, NULL);
1331         if (!ip_vs_conn_cachep) {
1332                 vfree(ip_vs_conn_tab);
1333                 return -ENOMEM;
1334         }
1335
1336         pr_info("Connection hash table configured "
1337                 "(size=%d, memory=%ldKbytes)\n",
1338                 ip_vs_conn_tab_size,
1339                 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1340         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1341                   sizeof(struct ip_vs_conn));
1342
1343         for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1344                 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1345
1346         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1347                 spin_lock_init(&__ip_vs_conntbl_lock_array[idx].l);
1348         }
1349
1350         /* calculate the random value for connection hash */
1351         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1352
1353         return 0;
1354 }
1355
1356 void ip_vs_conn_cleanup(void)
1357 {
1358         /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1359         rcu_barrier();
1360         /* Release the empty cache */
1361         kmem_cache_destroy(ip_vs_conn_cachep);
1362         vfree(ip_vs_conn_tab);
1363 }