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