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_bh(unsigned int key)
  90 {
  91         spin_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  92 }
  93
  94 static inline void ct_write_unlock_bh(unsigned int key)
  95 {
  96         spin_unlock_bh(&__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_bh(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_bh(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_bh(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_bh(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_bh(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_bh(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_bh(&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_bh(&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 #ifdef CONFIG_IP_VS_IPV6
 492                 if (cp->daf == AF_INET6)
 493                         cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 494                 else
 495 #endif
 496                         cp->packet_xmit = ip_vs_tunnel_xmit;
 497                 break;
 498
 499         case IP_VS_CONN_F_DROUTE:
 500                 cp->packet_xmit = ip_vs_dr_xmit;
 501                 break;
 502
 503         case IP_VS_CONN_F_LOCALNODE:
 504                 cp->packet_xmit = ip_vs_null_xmit;
 505                 break;
 506
 507         case IP_VS_CONN_F_BYPASS:
 508                 cp->packet_xmit = ip_vs_bypass_xmit;
 509                 break;
 510         }
 511 }
 512
 513 #ifdef CONFIG_IP_VS_IPV6
 514 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
 515 {
 516         switch (IP_VS_FWD_METHOD(cp)) {
 517         case IP_VS_CONN_F_MASQ:
 518                 cp->packet_xmit = ip_vs_nat_xmit_v6;
 519                 break;
 520
 521         case IP_VS_CONN_F_TUNNEL:
 522                 if (cp->daf == AF_INET6)
 523                         cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 524                 else
 525                         cp->packet_xmit = ip_vs_tunnel_xmit;
 526                 break;
 527
 528         case IP_VS_CONN_F_DROUTE:
 529                 cp->packet_xmit = ip_vs_dr_xmit_v6;
 530                 break;
 531
 532         case IP_VS_CONN_F_LOCALNODE:
 533                 cp->packet_xmit = ip_vs_null_xmit;
 534                 break;
 535
 536         case IP_VS_CONN_F_BYPASS:
 537                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
 538                 break;
 539         }
 540 }
 541 #endif
 542
 543
 544 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
 545 {
 546         return atomic_read(&dest->activeconns)
 547                 + atomic_read(&dest->inactconns);
 548 }
 549
 550 /*
 551  *      Bind a connection entry with a virtual service destination
 552  *      Called just after a new connection entry is created.
 553  */
 554 static inline void
 555 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
 556 {
 557         unsigned int conn_flags;
 558         __u32 flags;
 559
 560         /* if dest is NULL, then return directly */
 561         if (!dest)
 562                 return;
 563
 564         /* Increase the refcnt counter of the dest */
 565         ip_vs_dest_hold(dest);
 566
 567         conn_flags = atomic_read(&dest->conn_flags);
 568         if (cp->protocol != IPPROTO_UDP)
 569                 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
 570         flags = cp->flags;
 571         /* Bind with the destination and its corresponding transmitter */
 572         if (flags & IP_VS_CONN_F_SYNC) {
 573                 /* if the connection is not template and is created
 574                  * by sync, preserve the activity flag.
 575                  */
 576                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 577                         conn_flags &= ~IP_VS_CONN_F_INACTIVE;
 578                 /* connections inherit forwarding method from dest */
 579                 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
 580         }
 581         flags |= conn_flags;
 582         cp->flags = flags;
 583         cp->dest = dest;
 584
 585         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
 586                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 587                       "dest->refcnt:%d\n",
 588                       ip_vs_proto_name(cp->protocol),
 589                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 590                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 591                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
 592                       ip_vs_fwd_tag(cp), cp->state,
 593                       cp->flags, atomic_read(&cp->refcnt),
 594                       atomic_read(&dest->refcnt));
 595
 596         /* Update the connection counters */
 597         if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 598                 /* It is a normal connection, so modify the counters
 599                  * according to the flags, later the protocol can
 600                  * update them on state change
 601                  */
 602                 if (!(flags & IP_VS_CONN_F_INACTIVE))
 603                         atomic_inc(&dest->activeconns);
 604                 else
 605                         atomic_inc(&dest->inactconns);
 606         } else {
 607                 /* It is a persistent connection/template, so increase
 608                    the persistent connection counter */
 609                 atomic_inc(&dest->persistconns);
 610         }
 611
 612         if (dest->u_threshold != 0 &&
 613             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
 614                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
 615 }
 616
 617
 618 /*
 619  * Check if there is a destination for the connection, if so
 620  * bind the connection to the destination.
 621  */
 622 void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
 623 {
 624         struct ip_vs_dest *dest;
 625
 626         rcu_read_lock();
 627
 628         /* This function is only invoked by the synchronization code. We do
 629          * not currently support heterogeneous pools with synchronization,
 630          * so we can make the assumption that the svc_af is the same as the
 631          * dest_af
 632          */
 633         dest = ip_vs_find_dest(ip_vs_conn_net(cp), cp->af, cp->af, &cp->daddr,
 634                                cp->dport, &cp->vaddr, cp->vport,
 635                                cp->protocol, cp->fwmark, cp->flags);
 636         if (dest) {
 637                 struct ip_vs_proto_data *pd;
 638
 639                 spin_lock_bh(&cp->lock);
 640                 if (cp->dest) {
 641                         spin_unlock_bh(&cp->lock);
 642                         rcu_read_unlock();
 643                         return;
 644                 }
 645
 646                 /* Applications work depending on the forwarding method
 647                  * but better to reassign them always when binding dest */
 648                 if (cp->app)
 649                         ip_vs_unbind_app(cp);
 650
 651                 ip_vs_bind_dest(cp, dest);
 652                 spin_unlock_bh(&cp->lock);
 653
 654                 /* Update its packet transmitter */
 655                 cp->packet_xmit = NULL;
 656 #ifdef CONFIG_IP_VS_IPV6
 657                 if (cp->af == AF_INET6)
 658                         ip_vs_bind_xmit_v6(cp);
 659                 else
 660 #endif
 661                         ip_vs_bind_xmit(cp);
 662
 663                 pd = ip_vs_proto_data_get(ip_vs_conn_net(cp), cp->protocol);
 664                 if (pd && atomic_read(&pd->appcnt))
 665                         ip_vs_bind_app(cp, pd->pp);
 666         }
 667         rcu_read_unlock();
 668 }
 669
 670
 671 /*
 672  *      Unbind a connection entry with its VS destination
 673  *      Called by the ip_vs_conn_expire function.
 674  */
 675 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
 676 {
 677         struct ip_vs_dest *dest = cp->dest;
 678
 679         if (!dest)
 680                 return;
 681
 682         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
 683                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 684                       "dest->refcnt:%d\n",
 685                       ip_vs_proto_name(cp->protocol),
 686                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 687                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 688                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
 689                       ip_vs_fwd_tag(cp), cp->state,
 690                       cp->flags, atomic_read(&cp->refcnt),
 691                       atomic_read(&dest->refcnt));
 692
 693         /* Update the connection counters */
 694         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 695                 /* It is a normal connection, so decrease the inactconns
 696                    or activeconns counter */
 697                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
 698                         atomic_dec(&dest->inactconns);
 699                 } else {
 700                         atomic_dec(&dest->activeconns);
 701                 }
 702         } else {
 703                 /* It is a persistent connection/template, so decrease
 704                    the persistent connection counter */
 705                 atomic_dec(&dest->persistconns);
 706         }
 707
 708         if (dest->l_threshold != 0) {
 709                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
 710                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 711         } else if (dest->u_threshold != 0) {
 712                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
 713                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 714         } else {
 715                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 716                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 717         }
 718
 719         ip_vs_dest_put(dest);
 720 }
 721
 722 static int expire_quiescent_template(struct netns_ipvs *ipvs,
 723                                      struct ip_vs_dest *dest)
 724 {
 725 #ifdef CONFIG_SYSCTL
 726         return ipvs->sysctl_expire_quiescent_template &&
 727                 (atomic_read(&dest->weight) == 0);
 728 #else
 729         return 0;
 730 #endif
 731 }
 732
 733 /*
 734  *      Checking if the destination of a connection template is available.
 735  *      If available, return 1, otherwise invalidate this connection
 736  *      template and return 0.
 737  */
 738 int ip_vs_check_template(struct ip_vs_conn *ct)
 739 {
 740         struct ip_vs_dest *dest = ct->dest;
 741         struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(ct));
 742
 743         /*
 744          * Checking the dest server status.
 745          */
 746         if ((dest == NULL) ||
 747             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
 748             expire_quiescent_template(ipvs, dest)) {
 749                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
 750                               "protocol %s s:%s:%d v:%s:%d "
 751                               "-> d:%s:%d\n",
 752                               ip_vs_proto_name(ct->protocol),
 753                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
 754                               ntohs(ct->cport),
 755                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
 756                               ntohs(ct->vport),
 757                               IP_VS_DBG_ADDR(ct->af, &ct->daddr),
 758                               ntohs(ct->dport));
 759
 760                 /*
 761                  * Invalidate the connection template
 762                  */
 763                 if (ct->vport != htons(0xffff)) {
 764                         if (ip_vs_conn_unhash(ct)) {
 765                                 ct->dport = htons(0xffff);
 766                                 ct->vport = htons(0xffff);
 767                                 ct->cport = 0;
 768                                 ip_vs_conn_hash(ct);
 769                         }
 770                 }
 771
 772                 /*
 773                  * Simply decrease the refcnt of the template,
 774                  * don't restart its timer.
 775                  */
 776                 __ip_vs_conn_put(ct);
 777                 return 0;
 778         }
 779         return 1;
 780 }
 781
 782 static void ip_vs_conn_rcu_free(struct rcu_head *head)
 783 {
 784         struct ip_vs_conn *cp = container_of(head, struct ip_vs_conn,
 785                                              rcu_head);
 786
 787         ip_vs_pe_put(cp->pe);
 788         kfree(cp->pe_data);
 789         kmem_cache_free(ip_vs_conn_cachep, cp);
 790 }
 791
 792 static void ip_vs_conn_expire(unsigned long data)
 793 {
 794         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
 795         struct net *net = ip_vs_conn_net(cp);
 796         struct netns_ipvs *ipvs = net_ipvs(net);
 797
 798         /*
 799          *      do I control anybody?
 800          */
 801         if (atomic_read(&cp->n_control))
 802                 goto expire_later;
 803
 804         /* Unlink conn if not referenced anymore */
 805         if (likely(ip_vs_conn_unlink(cp))) {
 806                 /* delete the timer if it is activated by other users */
 807                 del_timer(&cp->timer);
 808
 809                 /* does anybody control me? */
 810                 if (cp->control)
 811                         ip_vs_control_del(cp);
 812
 813                 if (cp->flags & IP_VS_CONN_F_NFCT) {
 814                         /* Do not access conntracks during subsys cleanup
 815                          * because nf_conntrack_find_get can not be used after
 816                          * conntrack cleanup for the net.
 817                          */
 818                         smp_rmb();
 819                         if (ipvs->enable)
 820                                 ip_vs_conn_drop_conntrack(cp);
 821                 }
 822
 823                 if (unlikely(cp->app != NULL))
 824                         ip_vs_unbind_app(cp);
 825                 ip_vs_unbind_dest(cp);
 826                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
 827                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 828                 call_rcu(&cp->rcu_head, ip_vs_conn_rcu_free);
 829                 atomic_dec(&ipvs->conn_count);
 830                 return;
 831         }
 832
 833   expire_later:
 834         IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
 835                   atomic_read(&cp->refcnt),
 836                   atomic_read(&cp->n_control));
 837
 838         atomic_inc(&cp->refcnt);
 839         cp->timeout = 60*HZ;
 840
 841         if (ipvs->sync_state & IP_VS_STATE_MASTER)
 842                 ip_vs_sync_conn(net, cp, sysctl_sync_threshold(ipvs));
 843
 844         ip_vs_conn_put(cp);
 845 }
 846
 847 /* Modify timer, so that it expires as soon as possible.
 848  * Can be called without reference only if under RCU lock.
 849  */
 850 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
 851 {
 852         /* Using mod_timer_pending will ensure the timer is not
 853          * modified after the final del_timer in ip_vs_conn_expire.
 854          */
 855         if (timer_pending(&cp->timer) &&
 856             time_after(cp->timer.expires, jiffies))
 857                 mod_timer_pending(&cp->timer, jiffies);
 858 }
 859
 860
 861 /*
 862  *      Create a new connection entry and hash it into the ip_vs_conn_tab
 863  */
 864 struct ip_vs_conn *
 865 ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af,
 866                const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
 867                struct ip_vs_dest *dest, __u32 fwmark)
 868 {
 869         struct ip_vs_conn *cp;
 870         struct netns_ipvs *ipvs = net_ipvs(p->net);
 871         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->net,
 872                                                            p->protocol);
 873
 874         cp = kmem_cache_alloc(ip_vs_conn_cachep, GFP_ATOMIC);
 875         if (cp == NULL) {
 876                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
 877                 return NULL;
 878         }
 879
 880         INIT_HLIST_NODE(&cp->c_list);
 881         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
 882         ip_vs_conn_net_set(cp, p->net);
 883         cp->af             = p->af;
 884         cp->daf            = dest_af;
 885         cp->protocol       = p->protocol;
 886         ip_vs_addr_set(p->af, &cp->caddr, p->caddr);
 887         cp->cport          = p->cport;
 888         /* proto should only be IPPROTO_IP if p->vaddr is a fwmark */
 889         ip_vs_addr_set(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
 890                        &cp->vaddr, p->vaddr);
 891         cp->vport          = p->vport;
 892         ip_vs_addr_set(cp->daf, &cp->daddr, daddr);
 893         cp->dport          = dport;
 894         cp->flags          = flags;
 895         cp->fwmark         = fwmark;
 896         if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
 897                 ip_vs_pe_get(p->pe);
 898                 cp->pe = p->pe;
 899                 cp->pe_data = p->pe_data;
 900                 cp->pe_data_len = p->pe_data_len;
 901         } else {
 902                 cp->pe = NULL;
 903                 cp->pe_data = NULL;
 904                 cp->pe_data_len = 0;
 905         }
 906         spin_lock_init(&cp->lock);
 907
 908         /*
 909          * Set the entry is referenced by the current thread before hashing
 910          * it in the table, so that other thread run ip_vs_random_dropentry
 911          * but cannot drop this entry.
 912          */
 913         atomic_set(&cp->refcnt, 1);
 914
 915         cp->control = NULL;
 916         atomic_set(&cp->n_control, 0);
 917         atomic_set(&cp->in_pkts, 0);
 918
 919         cp->packet_xmit = NULL;
 920         cp->app = NULL;
 921         cp->app_data = NULL;
 922         /* reset struct ip_vs_seq */
 923         cp->in_seq.delta = 0;
 924         cp->out_seq.delta = 0;
 925
 926         atomic_inc(&ipvs->conn_count);
 927         if (flags & IP_VS_CONN_F_NO_CPORT)
 928                 atomic_inc(&ip_vs_conn_no_cport_cnt);
 929
 930         /* Bind the connection with a destination server */
 931         cp->dest = NULL;
 932         ip_vs_bind_dest(cp, dest);
 933
 934         /* Set its state and timeout */
 935         cp->state = 0;
 936         cp->old_state = 0;
 937         cp->timeout = 3*HZ;
 938         cp->sync_endtime = jiffies & ~3UL;
 939
 940         /* Bind its packet transmitter */
 941 #ifdef CONFIG_IP_VS_IPV6
 942         if (p->af == AF_INET6)
 943                 ip_vs_bind_xmit_v6(cp);
 944         else
 945 #endif
 946                 ip_vs_bind_xmit(cp);
 947
 948         if (unlikely(pd && atomic_read(&pd->appcnt)))
 949                 ip_vs_bind_app(cp, pd->pp);
 950
 951         /*
 952          * Allow conntrack to be preserved. By default, conntrack
 953          * is created and destroyed for every packet.
 954          * Sometimes keeping conntrack can be useful for
 955          * IP_VS_CONN_F_ONE_PACKET too.
 956          */
 957
 958         if (ip_vs_conntrack_enabled(ipvs))
 959                 cp->flags |= IP_VS_CONN_F_NFCT;
 960
 961         /* Hash it in the ip_vs_conn_tab finally */
 962         ip_vs_conn_hash(cp);
 963
 964         return cp;
 965 }
 966
 967 /*
 968  *      /proc/net/ip_vs_conn entries
 969  */
 970 #ifdef CONFIG_PROC_FS
 971 struct ip_vs_iter_state {
 972         struct seq_net_private  p;
 973         struct hlist_head       *l;
 974 };
 975
 976 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
 977 {
 978         int idx;
 979         struct ip_vs_conn *cp;
 980         struct ip_vs_iter_state *iter = seq->private;
 981
 982         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
 983                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
 984                         /* __ip_vs_conn_get() is not needed by
 985                          * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
 986                          */
 987                         if (pos-- == 0) {
 988                                 iter->l = &ip_vs_conn_tab[idx];
 989                                 return cp;
 990                         }
 991                 }
 992                 cond_resched_rcu();
 993         }
 994
 995         return NULL;
 996 }
 997
 998 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
 999         __acquires(RCU)
1000 {
1001         struct ip_vs_iter_state *iter = seq->private;
1002
1003         iter->l = NULL;
1004         rcu_read_lock();
1005         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
1006 }
1007
1008 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1009 {
1010         struct ip_vs_conn *cp = v;
1011         struct ip_vs_iter_state *iter = seq->private;
1012         struct hlist_node *e;
1013         struct hlist_head *l = iter->l;
1014         int idx;
1015
1016         ++*pos;
1017         if (v == SEQ_START_TOKEN)
1018                 return ip_vs_conn_array(seq, 0);
1019
1020         /* more on same hash chain? */
1021         e = rcu_dereference(hlist_next_rcu(&cp->c_list));
1022         if (e)
1023                 return hlist_entry(e, struct ip_vs_conn, c_list);
1024
1025         idx = l - ip_vs_conn_tab;
1026         while (++idx < ip_vs_conn_tab_size) {
1027                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1028                         iter->l = &ip_vs_conn_tab[idx];
1029                         return cp;
1030                 }
1031                 cond_resched_rcu();
1032         }
1033         iter->l = NULL;
1034         return NULL;
1035 }
1036
1037 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1038         __releases(RCU)
1039 {
1040         rcu_read_unlock();
1041 }
1042
1043 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1044 {
1045
1046         if (v == SEQ_START_TOKEN)
1047                 seq_puts(seq,
1048    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires PEName PEData\n");
1049         else {
1050                 const struct ip_vs_conn *cp = v;
1051                 struct net *net = seq_file_net(seq);
1052                 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1053                 size_t len = 0;
1054
1055                 if (!ip_vs_conn_net_eq(cp, net))
1056                         return 0;
1057                 if (cp->pe_data) {
1058                         pe_data[0] = ' ';
1059                         len = strlen(cp->pe->name);
1060                         memcpy(pe_data + 1, cp->pe->name, len);
1061                         pe_data[len + 1] = ' ';
1062                         len += 2;
1063                         len += cp->pe->show_pe_data(cp, pe_data + len);
1064                 }
1065                 pe_data[len] = '\0';
1066
1067 #ifdef CONFIG_IP_VS_IPV6
1068                 if (cp->af == AF_INET6)
1069                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1070                                 "%pI6 %04X %-11s %7lu%s\n",
1071                                 ip_vs_proto_name(cp->protocol),
1072                                 &cp->caddr.in6, ntohs(cp->cport),
1073                                 &cp->vaddr.in6, ntohs(cp->vport),
1074                                 &cp->daddr.in6, ntohs(cp->dport),
1075                                 ip_vs_state_name(cp->protocol, cp->state),
1076                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1077                 else
1078 #endif
1079                         seq_printf(seq,
1080                                 "%-3s %08X %04X %08X %04X"
1081                                 " %08X %04X %-11s %7lu%s\n",
1082                                 ip_vs_proto_name(cp->protocol),
1083                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1084                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1085                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
1086                                 ip_vs_state_name(cp->protocol, cp->state),
1087                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1088         }
1089         return 0;
1090 }
1091
1092 static const struct seq_operations ip_vs_conn_seq_ops = {
1093         .start = ip_vs_conn_seq_start,
1094         .next  = ip_vs_conn_seq_next,
1095         .stop  = ip_vs_conn_seq_stop,
1096         .show  = ip_vs_conn_seq_show,
1097 };
1098
1099 static int ip_vs_conn_open(struct inode *inode, struct file *file)
1100 {
1101         return seq_open_net(inode, file, &ip_vs_conn_seq_ops,
1102                             sizeof(struct ip_vs_iter_state));
1103 }
1104
1105 static const struct file_operations ip_vs_conn_fops = {
1106         .owner   = THIS_MODULE,
1107         .open    = ip_vs_conn_open,
1108         .read    = seq_read,
1109         .llseek  = seq_lseek,
1110         .release = seq_release_net,
1111 };
1112
1113 static const char *ip_vs_origin_name(unsigned int flags)
1114 {
1115         if (flags & IP_VS_CONN_F_SYNC)
1116                 return "SYNC";
1117         else
1118                 return "LOCAL";
1119 }
1120
1121 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1122 {
1123
1124         if (v == SEQ_START_TOKEN)
1125                 seq_puts(seq,
1126    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
1127         else {
1128                 const struct ip_vs_conn *cp = v;
1129                 struct net *net = seq_file_net(seq);
1130
1131                 if (!ip_vs_conn_net_eq(cp, net))
1132                         return 0;
1133
1134 #ifdef CONFIG_IP_VS_IPV6
1135                 if (cp->af == AF_INET6)
1136                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
1137                                 ip_vs_proto_name(cp->protocol),
1138                                 &cp->caddr.in6, ntohs(cp->cport),
1139                                 &cp->vaddr.in6, ntohs(cp->vport),
1140                                 &cp->daddr.in6, ntohs(cp->dport),
1141                                 ip_vs_state_name(cp->protocol, cp->state),
1142                                 ip_vs_origin_name(cp->flags),
1143                                 (cp->timer.expires-jiffies)/HZ);
1144                 else
1145 #endif
1146                         seq_printf(seq,
1147                                 "%-3s %08X %04X %08X %04X "
1148                                 "%08X %04X %-11s %-6s %7lu\n",
1149                                 ip_vs_proto_name(cp->protocol),
1150                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1151                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1152                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
1153                                 ip_vs_state_name(cp->protocol, cp->state),
1154                                 ip_vs_origin_name(cp->flags),
1155                                 (cp->timer.expires-jiffies)/HZ);
1156         }
1157         return 0;
1158 }
1159
1160 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1161         .start = ip_vs_conn_seq_start,
1162         .next  = ip_vs_conn_seq_next,
1163         .stop  = ip_vs_conn_seq_stop,
1164         .show  = ip_vs_conn_sync_seq_show,
1165 };
1166
1167 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
1168 {
1169         return seq_open_net(inode, file, &ip_vs_conn_sync_seq_ops,
1170                             sizeof(struct ip_vs_iter_state));
1171 }
1172
1173 static const struct file_operations ip_vs_conn_sync_fops = {
1174         .owner   = THIS_MODULE,
1175         .open    = ip_vs_conn_sync_open,
1176         .read    = seq_read,
1177         .llseek  = seq_lseek,
1178         .release = seq_release_net,
1179 };
1180
1181 #endif
1182
1183
1184 /*
1185  *      Randomly drop connection entries before running out of memory
1186  */
1187 static inline int todrop_entry(struct ip_vs_conn *cp)
1188 {
1189         /*
1190          * The drop rate array needs tuning for real environments.
1191          * Called from timer bh only => no locking
1192          */
1193         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1194         static char todrop_counter[9] = {0};
1195         int i;
1196
1197         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1198            This will leave enough time for normal connection to get
1199            through. */
1200         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1201                 return 0;
1202
1203         /* Don't drop the entry if its number of incoming packets is not
1204            located in [0, 8] */
1205         i = atomic_read(&cp->in_pkts);
1206         if (i > 8 || i < 0) return 0;
1207
1208         if (!todrop_rate[i]) return 0;
1209         if (--todrop_counter[i] > 0) return 0;
1210
1211         todrop_counter[i] = todrop_rate[i];
1212         return 1;
1213 }
1214
1215 /* Called from keventd and must protect itself from softirqs */
1216 void ip_vs_random_dropentry(struct net *net)
1217 {
1218         int idx;
1219         struct ip_vs_conn *cp, *cp_c;
1220
1221         rcu_read_lock();
1222         /*
1223          * Randomly scan 1/32 of the whole table every second
1224          */
1225         for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1226                 unsigned int hash = prandom_u32() & ip_vs_conn_tab_mask;
1227
1228                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
1229                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1230                                 /* connection template */
1231                                 continue;
1232                         if (!ip_vs_conn_net_eq(cp, net))
1233                                 continue;
1234                         if (cp->protocol == IPPROTO_TCP) {
1235                                 switch(cp->state) {
1236                                 case IP_VS_TCP_S_SYN_RECV:
1237                                 case IP_VS_TCP_S_SYNACK:
1238                                         break;
1239
1240                                 case IP_VS_TCP_S_ESTABLISHED:
1241                                         if (todrop_entry(cp))
1242                                                 break;
1243                                         continue;
1244
1245                                 default:
1246                                         continue;
1247                                 }
1248                         } else if (cp->protocol == IPPROTO_SCTP) {
1249                                 switch (cp->state) {
1250                                 case IP_VS_SCTP_S_INIT1:
1251                                 case IP_VS_SCTP_S_INIT:
1252                                         break;
1253                                 case IP_VS_SCTP_S_ESTABLISHED:
1254                                         if (todrop_entry(cp))
1255                                                 break;
1256                                         continue;
1257                                 default:
1258                                         continue;
1259                                 }
1260                         } else {
1261                                 if (!todrop_entry(cp))
1262                                         continue;
1263                         }
1264
1265                         IP_VS_DBG(4, "del connection\n");
1266                         ip_vs_conn_expire_now(cp);
1267                         cp_c = cp->control;
1268                         /* cp->control is valid only with reference to cp */
1269                         if (cp_c && __ip_vs_conn_get(cp)) {
1270                                 IP_VS_DBG(4, "del conn template\n");
1271                                 ip_vs_conn_expire_now(cp_c);
1272                                 __ip_vs_conn_put(cp);
1273                         }
1274                 }
1275                 cond_resched_rcu();
1276         }
1277         rcu_read_unlock();
1278 }
1279
1280
1281 /*
1282  *      Flush all the connection entries in the ip_vs_conn_tab
1283  */
1284 static void ip_vs_conn_flush(struct net *net)
1285 {
1286         int idx;
1287         struct ip_vs_conn *cp, *cp_c;
1288         struct netns_ipvs *ipvs = net_ipvs(net);
1289
1290 flush_again:
1291         rcu_read_lock();
1292         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1293
1294                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1295                         if (!ip_vs_conn_net_eq(cp, net))
1296                                 continue;
1297                         IP_VS_DBG(4, "del connection\n");
1298                         ip_vs_conn_expire_now(cp);
1299                         cp_c = cp->control;
1300                         /* cp->control is valid only with reference to cp */
1301                         if (cp_c && __ip_vs_conn_get(cp)) {
1302                                 IP_VS_DBG(4, "del conn template\n");
1303                                 ip_vs_conn_expire_now(cp_c);
1304                                 __ip_vs_conn_put(cp);
1305                         }
1306                 }
1307                 cond_resched_rcu();
1308         }
1309         rcu_read_unlock();
1310
1311         /* the counter may be not NULL, because maybe some conn entries
1312            are run by slow timer handler or unhashed but still referred */
1313         if (atomic_read(&ipvs->conn_count) != 0) {
1314                 schedule();
1315                 goto flush_again;
1316         }
1317 }
1318 /*
1319  * per netns init and exit
1320  */
1321 int __net_init ip_vs_conn_net_init(struct net *net)
1322 {
1323         struct netns_ipvs *ipvs = net_ipvs(net);
1324
1325         atomic_set(&ipvs->conn_count, 0);
1326
1327         proc_create("ip_vs_conn", 0, net->proc_net, &ip_vs_conn_fops);
1328         proc_create("ip_vs_conn_sync", 0, net->proc_net, &ip_vs_conn_sync_fops);
1329         return 0;
1330 }
1331
1332 void __net_exit ip_vs_conn_net_cleanup(struct net *net)
1333 {
1334         /* flush all the connection entries first */
1335         ip_vs_conn_flush(net);
1336         remove_proc_entry("ip_vs_conn", net->proc_net);
1337         remove_proc_entry("ip_vs_conn_sync", net->proc_net);
1338 }
1339
1340 int __init ip_vs_conn_init(void)
1341 {
1342         int idx;
1343
1344         /* Compute size and mask */
1345         ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1346         ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1347
1348         /*
1349          * Allocate the connection hash table and initialize its list heads
1350          */
1351         ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size * sizeof(*ip_vs_conn_tab));
1352         if (!ip_vs_conn_tab)
1353                 return -ENOMEM;
1354
1355         /* Allocate ip_vs_conn slab cache */
1356         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1357                                               sizeof(struct ip_vs_conn), 0,
1358                                               SLAB_HWCACHE_ALIGN, NULL);
1359         if (!ip_vs_conn_cachep) {
1360                 vfree(ip_vs_conn_tab);
1361                 return -ENOMEM;
1362         }
1363
1364         pr_info("Connection hash table configured "
1365                 "(size=%d, memory=%ldKbytes)\n",
1366                 ip_vs_conn_tab_size,
1367                 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1368         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1369                   sizeof(struct ip_vs_conn));
1370
1371         for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1372                 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1373
1374         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1375                 spin_lock_init(&__ip_vs_conntbl_lock_array[idx].l);
1376         }
1377
1378         /* calculate the random value for connection hash */
1379         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1380
1381         return 0;
1382 }
1383
1384 void ip_vs_conn_cleanup(void)
1385 {
1386         /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1387         rcu_barrier();
1388         /* Release the empty cache */
1389         kmem_cache_destroy(ip_vs_conn_cachep);
1390         vfree(ip_vs_conn_tab);
1391 }