kernel/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/inet.h>
  31 #include <linux/net.h>
  32 #include <linux/kernel.h>
  33 #include <linux/module.h>
  34 #include <linux/vmalloc.h>
  35 #include <linux/proc_fs.h>              /* for proc_net_* */
  36 #include <linux/slab.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/jhash.h>
  39 #include <linux/random.h>
  40
  41 #include <net/net_namespace.h>
  42 #include <net/ip_vs.h>
  43
  44
  45 #ifndef CONFIG_IP_VS_TAB_BITS
  46 #define CONFIG_IP_VS_TAB_BITS   12
  47 #endif
  48
  49 /*
  50  * Connection hash size. Default is what was selected at compile time.
  51 */
  52 static int ip_vs_conn_tab_bits = CONFIG_IP_VS_TAB_BITS;
  53 module_param_named(conn_tab_bits, ip_vs_conn_tab_bits, int, 0444);
  54 MODULE_PARM_DESC(conn_tab_bits, "Set connections' hash size");
  55
  56 /* size and mask values */
  57 int ip_vs_conn_tab_size __read_mostly;
  58 static int ip_vs_conn_tab_mask __read_mostly;
  59
  60 /*
  61  *  Connection hash table: for input and output packets lookups of IPVS
  62  */
  63 static struct hlist_head *ip_vs_conn_tab __read_mostly;
  64
  65 /*  SLAB cache for IPVS connections */
  66 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
  67
  68 /*  counter for no client port connections */
  69 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
  70
  71 /* random value for IPVS connection hash */
  72 static unsigned int ip_vs_conn_rnd __read_mostly;
  73
  74 /*
  75  *  Fine locking granularity for big connection hash table
  76  */
  77 #define CT_LOCKARRAY_BITS  5
  78 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
  79 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
  80
  81 /* We need an addrstrlen that works with or without v6 */
  82 #ifdef CONFIG_IP_VS_IPV6
  83 #define IP_VS_ADDRSTRLEN INET6_ADDRSTRLEN
  84 #else
  85 #define IP_VS_ADDRSTRLEN (8+1)
  86 #endif
  87
  88 struct ip_vs_aligned_lock
  89 {
  90         spinlock_t      l;
  91 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
  92
  93 /* lock array for conn table */
  94 static struct ip_vs_aligned_lock
  95 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
  96
  97 static inline void ct_write_lock_bh(unsigned int key)
  98 {
  99         spin_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 100 }
 101
 102 static inline void ct_write_unlock_bh(unsigned int key)
 103 {
 104         spin_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 105 }
 106
 107
 108 /*
 109  *      Returns hash value for IPVS connection entry
 110  */
 111 static unsigned int ip_vs_conn_hashkey(struct net *net, int af, unsigned int proto,
 112                                        const union nf_inet_addr *addr,
 113                                        __be16 port)
 114 {
 115 #ifdef CONFIG_IP_VS_IPV6
 116         if (af == AF_INET6)
 117                 return (jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
 118                                     (__force u32)port, proto, ip_vs_conn_rnd) ^
 119                         ((size_t)net>>8)) & ip_vs_conn_tab_mask;
 120 #endif
 121         return (jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
 122                             ip_vs_conn_rnd) ^
 123                 ((size_t)net>>8)) & ip_vs_conn_tab_mask;
 124 }
 125
 126 static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param *p,
 127                                              bool inverse)
 128 {
 129         const union nf_inet_addr *addr;
 130         __be16 port;
 131
 132         if (p->pe_data && p->pe->hashkey_raw)
 133                 return p->pe->hashkey_raw(p, ip_vs_conn_rnd, inverse) &
 134                         ip_vs_conn_tab_mask;
 135
 136         if (likely(!inverse)) {
 137                 addr = p->caddr;
 138                 port = p->cport;
 139         } else {
 140                 addr = p->vaddr;
 141                 port = p->vport;
 142         }
 143
 144         return ip_vs_conn_hashkey(p->net, p->af, p->protocol, addr, port);
 145 }
 146
 147 static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn *cp)
 148 {
 149         struct ip_vs_conn_param p;
 150
 151         ip_vs_conn_fill_param(ip_vs_conn_net(cp), cp->af, cp->protocol,
 152                               &cp->caddr, cp->cport, NULL, 0, &p);
 153
 154         if (cp->pe) {
 155                 p.pe = cp->pe;
 156                 p.pe_data = cp->pe_data;
 157                 p.pe_data_len = cp->pe_data_len;
 158         }
 159
 160         return ip_vs_conn_hashkey_param(&p, false);
 161 }
 162
 163 /*
 164  *      Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
 165  *      returns bool success.
 166  */
 167 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
 168 {
 169         unsigned int hash;
 170         int ret;
 171
 172         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
 173                 return 0;
 174
 175         /* Hash by protocol, client address and port */
 176         hash = ip_vs_conn_hashkey_conn(cp);
 177
 178         ct_write_lock_bh(hash);
 179         spin_lock(&cp->lock);
 180
 181         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
 182                 cp->flags |= IP_VS_CONN_F_HASHED;
 183                 atomic_inc(&cp->refcnt);
 184                 hlist_add_head_rcu(&cp->c_list, &ip_vs_conn_tab[hash]);
 185                 ret = 1;
 186         } else {
 187                 pr_err("%s(): request for already hashed, called from %pF\n",
 188                        __func__, __builtin_return_address(0));
 189                 ret = 0;
 190         }
 191
 192         spin_unlock(&cp->lock);
 193         ct_write_unlock_bh(hash);
 194
 195         return ret;
 196 }
 197
 198
 199 /*
 200  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
 201  *      returns bool success. Caller should hold conn reference.
 202  */
 203 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
 204 {
 205         unsigned int hash;
 206         int ret;
 207
 208         /* unhash it and decrease its reference counter */
 209         hash = ip_vs_conn_hashkey_conn(cp);
 210
 211         ct_write_lock_bh(hash);
 212         spin_lock(&cp->lock);
 213
 214         if (cp->flags & IP_VS_CONN_F_HASHED) {
 215                 hlist_del_rcu(&cp->c_list);
 216                 cp->flags &= ~IP_VS_CONN_F_HASHED;
 217                 atomic_dec(&cp->refcnt);
 218                 ret = 1;
 219         } else
 220                 ret = 0;
 221
 222         spin_unlock(&cp->lock);
 223         ct_write_unlock_bh(hash);
 224
 225         return ret;
 226 }
 227
 228 /* Try to unlink ip_vs_conn from ip_vs_conn_tab.
 229  * returns bool success.
 230  */
 231 static inline bool ip_vs_conn_unlink(struct ip_vs_conn *cp)
 232 {
 233         unsigned int hash;
 234         bool ret;
 235
 236         hash = ip_vs_conn_hashkey_conn(cp);
 237
 238         ct_write_lock_bh(hash);
 239         spin_lock(&cp->lock);
 240
 241         if (cp->flags & IP_VS_CONN_F_HASHED) {
 242                 ret = false;
 243                 /* Decrease refcnt and unlink conn only if we are last user */
 244                 if (atomic_cmpxchg(&cp->refcnt, 1, 0) == 1) {
 245                         hlist_del_rcu(&cp->c_list);
 246                         cp->flags &= ~IP_VS_CONN_F_HASHED;
 247                         ret = true;
 248                 }
 249         } else
 250                 ret = atomic_read(&cp->refcnt) ? false : true;
 251
 252         spin_unlock(&cp->lock);
 253         ct_write_unlock_bh(hash);
 254
 255         return ret;
 256 }
 257
 258
 259 /*
 260  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 261  *  Called for pkts coming from OUTside-to-INside.
 262  *      p->caddr, p->cport: pkt source address (foreign host)
 263  *      p->vaddr, p->vport: pkt dest address (load balancer)
 264  */
 265 static inline struct ip_vs_conn *
 266 __ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
 267 {
 268         unsigned int hash;
 269         struct ip_vs_conn *cp;
 270
 271         hash = ip_vs_conn_hashkey_param(p, false);
 272
 273         rcu_read_lock();
 274
 275         hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
 276                 if (p->cport == cp->cport && p->vport == cp->vport &&
 277                     cp->af == p->af &&
 278                     ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
 279                     ip_vs_addr_equal(p->af, p->vaddr, &cp->vaddr) &&
 280                     ((!p->cport) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
 281                     p->protocol == cp->protocol &&
 282                     ip_vs_conn_net_eq(cp, p->net)) {
 283                         if (!__ip_vs_conn_get(cp))
 284                                 continue;
 285                         /* HIT */
 286                         rcu_read_unlock();
 287                         return cp;
 288                 }
 289         }
 290
 291         rcu_read_unlock();
 292
 293         return NULL;
 294 }
 295
 296 struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
 297 {
 298         struct ip_vs_conn *cp;
 299
 300         cp = __ip_vs_conn_in_get(p);
 301         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt)) {
 302                 struct ip_vs_conn_param cport_zero_p = *p;
 303                 cport_zero_p.cport = 0;
 304                 cp = __ip_vs_conn_in_get(&cport_zero_p);
 305         }
 306
 307         IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
 308                       ip_vs_proto_name(p->protocol),
 309                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
 310                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
 311                       cp ? "hit" : "not hit");
 312
 313         return cp;
 314 }
 315
 316 static int
 317 ip_vs_conn_fill_param_proto(int af, const struct sk_buff *skb,
 318                             const struct ip_vs_iphdr *iph,
 319                             int inverse, struct ip_vs_conn_param *p)
 320 {
 321         __be16 _ports[2], *pptr;
 322         struct net *net = skb_net(skb);
 323
 324         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 325         if (pptr == NULL)
 326                 return 1;
 327
 328         if (likely(!inverse))
 329                 ip_vs_conn_fill_param(net, af, iph->protocol, &iph->saddr,
 330                                       pptr[0], &iph->daddr, pptr[1], p);
 331         else
 332                 ip_vs_conn_fill_param(net, af, iph->protocol, &iph->daddr,
 333                                       pptr[1], &iph->saddr, pptr[0], p);
 334         return 0;
 335 }
 336
 337 struct ip_vs_conn *
 338 ip_vs_conn_in_get_proto(int af, const struct sk_buff *skb,
 339                         const struct ip_vs_iphdr *iph, int inverse)
 340 {
 341         struct ip_vs_conn_param p;
 342
 343         if (ip_vs_conn_fill_param_proto(af, skb, iph, inverse, &p))
 344                 return NULL;
 345
 346         return ip_vs_conn_in_get(&p);
 347 }
 348 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
 349
 350 /* Get reference to connection template */
 351 struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p)
 352 {
 353         unsigned int hash;
 354         struct ip_vs_conn *cp;
 355
 356         hash = ip_vs_conn_hashkey_param(p, false);
 357
 358         rcu_read_lock();
 359
 360         hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
 361                 if (unlikely(p->pe_data && p->pe->ct_match)) {
 362                         if (!ip_vs_conn_net_eq(cp, p->net))
 363                                 continue;
 364                         if (p->pe == cp->pe && p->pe->ct_match(p, cp)) {
 365                                 if (__ip_vs_conn_get(cp))
 366                                         goto out;
 367                         }
 368                         continue;
 369                 }
 370
 371                 if (cp->af == p->af &&
 372                     ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
 373                     /* protocol should only be IPPROTO_IP if
 374                      * p->vaddr is a fwmark */
 375                     ip_vs_addr_equal(p->protocol == IPPROTO_IP ? AF_UNSPEC :
 376                                      p->af, p->vaddr, &cp->vaddr) &&
 377                     p->vport == cp->vport && p->cport == cp->cport &&
 378                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
 379                     p->protocol == cp->protocol &&
 380                     ip_vs_conn_net_eq(cp, p->net)) {
 381                         if (__ip_vs_conn_get(cp))
 382                                 goto out;
 383                 }
 384         }
 385         cp = NULL;
 386
 387   out:
 388         rcu_read_unlock();
 389
 390         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
 391                       ip_vs_proto_name(p->protocol),
 392                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
 393                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
 394                       cp ? "hit" : "not hit");
 395
 396         return cp;
 397 }
 398
 399 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 400  * Called for pkts coming from inside-to-OUTside.
 401  *      p->caddr, p->cport: pkt source address (inside host)
 402  *      p->vaddr, p->vport: pkt dest address (foreign host) */
 403 struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
 404 {
 405         unsigned int hash;
 406         struct ip_vs_conn *cp, *ret=NULL;
 407
 408         /*
 409          *      Check for "full" addressed entries
 410          */
 411         hash = ip_vs_conn_hashkey_param(p, true);
 412
 413         rcu_read_lock();
 414
 415         hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
 416                 if (p->vport == cp->cport && p->cport == cp->dport &&
 417                     cp->af == p->af &&
 418                     ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
 419                     ip_vs_addr_equal(p->af, p->caddr, &cp->daddr) &&
 420                     p->protocol == cp->protocol &&
 421                     ip_vs_conn_net_eq(cp, p->net)) {
 422                         if (!__ip_vs_conn_get(cp))
 423                                 continue;
 424                         /* HIT */
 425                         ret = cp;
 426                         break;
 427                 }
 428         }
 429
 430         rcu_read_unlock();
 431
 432         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
 433                       ip_vs_proto_name(p->protocol),
 434                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
 435                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
 436                       ret ? "hit" : "not hit");
 437
 438         return ret;
 439 }
 440
 441 struct ip_vs_conn *
 442 ip_vs_conn_out_get_proto(int af, const struct sk_buff *skb,
 443                          const struct ip_vs_iphdr *iph, int inverse)
 444 {
 445         struct ip_vs_conn_param p;
 446
 447         if (ip_vs_conn_fill_param_proto(af, skb, iph, inverse, &p))
 448                 return NULL;
 449
 450         return ip_vs_conn_out_get(&p);
 451 }
 452 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
 453
 454 /*
 455  *      Put back the conn and restart its timer with its timeout
 456  */
 457 void ip_vs_conn_put(struct ip_vs_conn *cp)
 458 {
 459         unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
 460                 0 : cp->timeout;
 461         mod_timer(&cp->timer, jiffies+t);
 462
 463         __ip_vs_conn_put(cp);
 464 }
 465
 466
 467 /*
 468  *      Fill a no_client_port connection with a client port number
 469  */
 470 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
 471 {
 472         if (ip_vs_conn_unhash(cp)) {
 473                 spin_lock_bh(&cp->lock);
 474                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
 475                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 476                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
 477                         cp->cport = cport;
 478                 }
 479                 spin_unlock_bh(&cp->lock);
 480
 481                 /* hash on new dport */
 482                 ip_vs_conn_hash(cp);
 483         }
 484 }
 485
 486
 487 /*
 488  *      Bind a connection entry with the corresponding packet_xmit.
 489  *      Called by ip_vs_conn_new.
 490  */
 491 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
 492 {
 493         switch (IP_VS_FWD_METHOD(cp)) {
 494         case IP_VS_CONN_F_MASQ:
 495                 cp->packet_xmit = ip_vs_nat_xmit;
 496                 break;
 497
 498         case IP_VS_CONN_F_TUNNEL:
 499 #ifdef CONFIG_IP_VS_IPV6
 500                 if (cp->daf == AF_INET6)
 501                         cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 502                 else
 503 #endif
 504                         cp->packet_xmit = ip_vs_tunnel_xmit;
 505                 break;
 506
 507         case IP_VS_CONN_F_DROUTE:
 508                 cp->packet_xmit = ip_vs_dr_xmit;
 509                 break;
 510
 511         case IP_VS_CONN_F_LOCALNODE:
 512                 cp->packet_xmit = ip_vs_null_xmit;
 513                 break;
 514
 515         case IP_VS_CONN_F_BYPASS:
 516                 cp->packet_xmit = ip_vs_bypass_xmit;
 517                 break;
 518         }
 519 }
 520
 521 #ifdef CONFIG_IP_VS_IPV6
 522 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
 523 {
 524         switch (IP_VS_FWD_METHOD(cp)) {
 525         case IP_VS_CONN_F_MASQ:
 526                 cp->packet_xmit = ip_vs_nat_xmit_v6;
 527                 break;
 528
 529         case IP_VS_CONN_F_TUNNEL:
 530                 if (cp->daf == AF_INET6)
 531                         cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 532                 else
 533                         cp->packet_xmit = ip_vs_tunnel_xmit;
 534                 break;
 535
 536         case IP_VS_CONN_F_DROUTE:
 537                 cp->packet_xmit = ip_vs_dr_xmit_v6;
 538                 break;
 539
 540         case IP_VS_CONN_F_LOCALNODE:
 541                 cp->packet_xmit = ip_vs_null_xmit;
 542                 break;
 543
 544         case IP_VS_CONN_F_BYPASS:
 545                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
 546                 break;
 547         }
 548 }
 549 #endif
 550
 551
 552 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
 553 {
 554         return atomic_read(&dest->activeconns)
 555                 + atomic_read(&dest->inactconns);
 556 }
 557
 558 /*
 559  *      Bind a connection entry with a virtual service destination
 560  *      Called just after a new connection entry is created.
 561  */
 562 static inline void
 563 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
 564 {
 565         unsigned int conn_flags;
 566         __u32 flags;
 567
 568         /* if dest is NULL, then return directly */
 569         if (!dest)
 570                 return;
 571
 572         /* Increase the refcnt counter of the dest */
 573         ip_vs_dest_hold(dest);
 574
 575         conn_flags = atomic_read(&dest->conn_flags);
 576         if (cp->protocol != IPPROTO_UDP)
 577                 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
 578         flags = cp->flags;
 579         /* Bind with the destination and its corresponding transmitter */
 580         if (flags & IP_VS_CONN_F_SYNC) {
 581                 /* if the connection is not template and is created
 582                  * by sync, preserve the activity flag.
 583                  */
 584                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 585                         conn_flags &= ~IP_VS_CONN_F_INACTIVE;
 586                 /* connections inherit forwarding method from dest */
 587                 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
 588         }
 589         flags |= conn_flags;
 590         cp->flags = flags;
 591         cp->dest = dest;
 592
 593         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
 594                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 595                       "dest->refcnt:%d\n",
 596                       ip_vs_proto_name(cp->protocol),
 597                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 598                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 599                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 600                       ip_vs_fwd_tag(cp), cp->state,
 601                       cp->flags, atomic_read(&cp->refcnt),
 602                       atomic_read(&dest->refcnt));
 603
 604         /* Update the connection counters */
 605         if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 606                 /* It is a normal connection, so modify the counters
 607                  * according to the flags, later the protocol can
 608                  * update them on state change
 609                  */
 610                 if (!(flags & IP_VS_CONN_F_INACTIVE))
 611                         atomic_inc(&dest->activeconns);
 612                 else
 613                         atomic_inc(&dest->inactconns);
 614         } else {
 615                 /* It is a persistent connection/template, so increase
 616                    the persistent connection counter */
 617                 atomic_inc(&dest->persistconns);
 618         }
 619
 620         if (dest->u_threshold != 0 &&
 621             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
 622                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
 623 }
 624
 625
 626 /*
 627  * Check if there is a destination for the connection, if so
 628  * bind the connection to the destination.
 629  */
 630 void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
 631 {
 632         struct ip_vs_dest *dest;
 633
 634         rcu_read_lock();
 635
 636         /* This function is only invoked by the synchronization code. We do
 637          * not currently support heterogeneous pools with synchronization,
 638          * so we can make the assumption that the svc_af is the same as the
 639          * dest_af
 640          */
 641         dest = ip_vs_find_dest(ip_vs_conn_net(cp), cp->af, cp->af, &cp->daddr,
 642                                cp->dport, &cp->vaddr, cp->vport,
 643                                cp->protocol, cp->fwmark, cp->flags);
 644         if (dest) {
 645                 struct ip_vs_proto_data *pd;
 646
 647                 spin_lock_bh(&cp->lock);
 648                 if (cp->dest) {
 649                         spin_unlock_bh(&cp->lock);
 650                         rcu_read_unlock();
 651                         return;
 652                 }
 653
 654                 /* Applications work depending on the forwarding method
 655                  * but better to reassign them always when binding dest */
 656                 if (cp->app)
 657                         ip_vs_unbind_app(cp);
 658
 659                 ip_vs_bind_dest(cp, dest);
 660                 spin_unlock_bh(&cp->lock);
 661
 662                 /* Update its packet transmitter */
 663                 cp->packet_xmit = NULL;
 664 #ifdef CONFIG_IP_VS_IPV6
 665                 if (cp->af == AF_INET6)
 666                         ip_vs_bind_xmit_v6(cp);
 667                 else
 668 #endif
 669                         ip_vs_bind_xmit(cp);
 670
 671                 pd = ip_vs_proto_data_get(ip_vs_conn_net(cp), cp->protocol);
 672                 if (pd && atomic_read(&pd->appcnt))
 673                         ip_vs_bind_app(cp, pd->pp);
 674         }
 675         rcu_read_unlock();
 676 }
 677
 678
 679 /*
 680  *      Unbind a connection entry with its VS destination
 681  *      Called by the ip_vs_conn_expire function.
 682  */
 683 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
 684 {
 685         struct ip_vs_dest *dest = cp->dest;
 686
 687         if (!dest)
 688                 return;
 689
 690         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
 691                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 692                       "dest->refcnt:%d\n",
 693                       ip_vs_proto_name(cp->protocol),
 694                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 695                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 696                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 697                       ip_vs_fwd_tag(cp), cp->state,
 698                       cp->flags, atomic_read(&cp->refcnt),
 699                       atomic_read(&dest->refcnt));
 700
 701         /* Update the connection counters */
 702         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 703                 /* It is a normal connection, so decrease the inactconns
 704                    or activeconns counter */
 705                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
 706                         atomic_dec(&dest->inactconns);
 707                 } else {
 708                         atomic_dec(&dest->activeconns);
 709                 }
 710         } else {
 711                 /* It is a persistent connection/template, so decrease
 712                    the persistent connection counter */
 713                 atomic_dec(&dest->persistconns);
 714         }
 715
 716         if (dest->l_threshold != 0) {
 717                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
 718                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 719         } else if (dest->u_threshold != 0) {
 720                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
 721                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 722         } else {
 723                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 724                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 725         }
 726
 727         ip_vs_dest_put(dest);
 728 }
 729
 730 static int expire_quiescent_template(struct netns_ipvs *ipvs,
 731                                      struct ip_vs_dest *dest)
 732 {
 733 #ifdef CONFIG_SYSCTL
 734         return ipvs->sysctl_expire_quiescent_template &&
 735                 (atomic_read(&dest->weight) == 0);
 736 #else
 737         return 0;
 738 #endif
 739 }
 740
 741 /*
 742  *      Checking if the destination of a connection template is available.
 743  *      If available, return 1, otherwise invalidate this connection
 744  *      template and return 0.
 745  */
 746 int ip_vs_check_template(struct ip_vs_conn *ct)
 747 {
 748         struct ip_vs_dest *dest = ct->dest;
 749         struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(ct));
 750
 751         /*
 752          * Checking the dest server status.
 753          */
 754         if ((dest == NULL) ||
 755             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
 756             expire_quiescent_template(ipvs, dest)) {
 757                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
 758                               "protocol %s s:%s:%d v:%s:%d "
 759                               "-> d:%s:%d\n",
 760                               ip_vs_proto_name(ct->protocol),
 761                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
 762                               ntohs(ct->cport),
 763                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
 764                               ntohs(ct->vport),
 765                               IP_VS_DBG_ADDR(ct->daf, &ct->daddr),
 766                               ntohs(ct->dport));
 767
 768                 /*
 769                  * Invalidate the connection template
 770                  */
 771                 if (ct->vport != htons(0xffff)) {
 772                         if (ip_vs_conn_unhash(ct)) {
 773                                 ct->dport = htons(0xffff);
 774                                 ct->vport = htons(0xffff);
 775                                 ct->cport = 0;
 776                                 ip_vs_conn_hash(ct);
 777                         }
 778                 }
 779
 780                 /*
 781                  * Simply decrease the refcnt of the template,
 782                  * don't restart its timer.
 783                  */
 784                 __ip_vs_conn_put(ct);
 785                 return 0;
 786         }
 787         return 1;
 788 }
 789
 790 static void ip_vs_conn_rcu_free(struct rcu_head *head)
 791 {
 792         struct ip_vs_conn *cp = container_of(head, struct ip_vs_conn,
 793                                              rcu_head);
 794
 795         ip_vs_pe_put(cp->pe);
 796         kfree(cp->pe_data);
 797         kmem_cache_free(ip_vs_conn_cachep, cp);
 798 }
 799
 800 static void ip_vs_conn_expire(unsigned long data)
 801 {
 802         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
 803         struct net *net = ip_vs_conn_net(cp);
 804         struct netns_ipvs *ipvs = net_ipvs(net);
 805
 806         /*
 807          *      do I control anybody?
 808          */
 809         if (atomic_read(&cp->n_control))
 810                 goto expire_later;
 811
 812         /* Unlink conn if not referenced anymore */
 813         if (likely(ip_vs_conn_unlink(cp))) {
 814                 /* delete the timer if it is activated by other users */
 815                 del_timer(&cp->timer);
 816
 817                 /* does anybody control me? */
 818                 if (cp->control)
 819                         ip_vs_control_del(cp);
 820
 821                 if (cp->flags & IP_VS_CONN_F_NFCT) {
 822                         /* Do not access conntracks during subsys cleanup
 823                          * because nf_conntrack_find_get can not be used after
 824                          * conntrack cleanup for the net.
 825                          */
 826                         smp_rmb();
 827                         if (ipvs->enable)
 828                                 ip_vs_conn_drop_conntrack(cp);
 829                 }
 830
 831                 if (unlikely(cp->app != NULL))
 832                         ip_vs_unbind_app(cp);
 833                 ip_vs_unbind_dest(cp);
 834                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
 835                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 836                 call_rcu(&cp->rcu_head, ip_vs_conn_rcu_free);
 837                 atomic_dec(&ipvs->conn_count);
 838                 return;
 839         }
 840
 841   expire_later:
 842         IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
 843                   atomic_read(&cp->refcnt),
 844                   atomic_read(&cp->n_control));
 845
 846         atomic_inc(&cp->refcnt);
 847         cp->timeout = 60*HZ;
 848
 849         if (ipvs->sync_state & IP_VS_STATE_MASTER)
 850                 ip_vs_sync_conn(net, cp, sysctl_sync_threshold(ipvs));
 851
 852         ip_vs_conn_put(cp);
 853 }
 854
 855 /* Modify timer, so that it expires as soon as possible.
 856  * Can be called without reference only if under RCU lock.
 857  */
 858 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
 859 {
 860         /* Using mod_timer_pending will ensure the timer is not
 861          * modified after the final del_timer in ip_vs_conn_expire.
 862          */
 863         if (timer_pending(&cp->timer) &&
 864             time_after(cp->timer.expires, jiffies))
 865                 mod_timer_pending(&cp->timer, jiffies);
 866 }
 867
 868
 869 /*
 870  *      Create a new connection entry and hash it into the ip_vs_conn_tab
 871  */
 872 struct ip_vs_conn *
 873 ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af,
 874                const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
 875                struct ip_vs_dest *dest, __u32 fwmark)
 876 {
 877         struct ip_vs_conn *cp;
 878         struct netns_ipvs *ipvs = net_ipvs(p->net);
 879         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->net,
 880                                                            p->protocol);
 881
 882         cp = kmem_cache_alloc(ip_vs_conn_cachep, GFP_ATOMIC);
 883         if (cp == NULL) {
 884                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
 885                 return NULL;
 886         }
 887
 888         INIT_HLIST_NODE(&cp->c_list);
 889         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
 890         ip_vs_conn_net_set(cp, p->net);
 891         cp->af             = p->af;
 892         cp->daf            = dest_af;
 893         cp->protocol       = p->protocol;
 894         ip_vs_addr_set(p->af, &cp->caddr, p->caddr);
 895         cp->cport          = p->cport;
 896         /* proto should only be IPPROTO_IP if p->vaddr is a fwmark */
 897         ip_vs_addr_set(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
 898                        &cp->vaddr, p->vaddr);
 899         cp->vport          = p->vport;
 900         ip_vs_addr_set(cp->daf, &cp->daddr, daddr);
 901         cp->dport          = dport;
 902         cp->flags          = flags;
 903         cp->fwmark         = fwmark;
 904         if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
 905                 ip_vs_pe_get(p->pe);
 906                 cp->pe = p->pe;
 907                 cp->pe_data = p->pe_data;
 908                 cp->pe_data_len = p->pe_data_len;
 909         } else {
 910                 cp->pe = NULL;
 911                 cp->pe_data = NULL;
 912                 cp->pe_data_len = 0;
 913         }
 914         spin_lock_init(&cp->lock);
 915
 916         /*
 917          * Set the entry is referenced by the current thread before hashing
 918          * it in the table, so that other thread run ip_vs_random_dropentry
 919          * but cannot drop this entry.
 920          */
 921         atomic_set(&cp->refcnt, 1);
 922
 923         cp->control = NULL;
 924         atomic_set(&cp->n_control, 0);
 925         atomic_set(&cp->in_pkts, 0);
 926
 927         cp->packet_xmit = NULL;
 928         cp->app = NULL;
 929         cp->app_data = NULL;
 930         /* reset struct ip_vs_seq */
 931         cp->in_seq.delta = 0;
 932         cp->out_seq.delta = 0;
 933
 934         atomic_inc(&ipvs->conn_count);
 935         if (flags & IP_VS_CONN_F_NO_CPORT)
 936                 atomic_inc(&ip_vs_conn_no_cport_cnt);
 937
 938         /* Bind the connection with a destination server */
 939         cp->dest = NULL;
 940         ip_vs_bind_dest(cp, dest);
 941
 942         /* Set its state and timeout */
 943         cp->state = 0;
 944         cp->old_state = 0;
 945         cp->timeout = 3*HZ;
 946         cp->sync_endtime = jiffies & ~3UL;
 947
 948         /* Bind its packet transmitter */
 949 #ifdef CONFIG_IP_VS_IPV6
 950         if (p->af == AF_INET6)
 951                 ip_vs_bind_xmit_v6(cp);
 952         else
 953 #endif
 954                 ip_vs_bind_xmit(cp);
 955
 956         if (unlikely(pd && atomic_read(&pd->appcnt)))
 957                 ip_vs_bind_app(cp, pd->pp);
 958
 959         /*
 960          * Allow conntrack to be preserved. By default, conntrack
 961          * is created and destroyed for every packet.
 962          * Sometimes keeping conntrack can be useful for
 963          * IP_VS_CONN_F_ONE_PACKET too.
 964          */
 965
 966         if (ip_vs_conntrack_enabled(ipvs))
 967                 cp->flags |= IP_VS_CONN_F_NFCT;
 968
 969         /* Hash it in the ip_vs_conn_tab finally */
 970         ip_vs_conn_hash(cp);
 971
 972         return cp;
 973 }
 974
 975 /*
 976  *      /proc/net/ip_vs_conn entries
 977  */
 978 #ifdef CONFIG_PROC_FS
 979 struct ip_vs_iter_state {
 980         struct seq_net_private  p;
 981         struct hlist_head       *l;
 982 };
 983
 984 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
 985 {
 986         int idx;
 987         struct ip_vs_conn *cp;
 988         struct ip_vs_iter_state *iter = seq->private;
 989
 990         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
 991                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
 992                         /* __ip_vs_conn_get() is not needed by
 993                          * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
 994                          */
 995                         if (pos-- == 0) {
 996                                 iter->l = &ip_vs_conn_tab[idx];
 997                                 return cp;
 998                         }
 999                 }
1000                 cond_resched_rcu();
1001         }
1002
1003         return NULL;
1004 }
1005
1006 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
1007         __acquires(RCU)
1008 {
1009         struct ip_vs_iter_state *iter = seq->private;
1010
1011         iter->l = NULL;
1012         rcu_read_lock();
1013         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
1014 }
1015
1016 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1017 {
1018         struct ip_vs_conn *cp = v;
1019         struct ip_vs_iter_state *iter = seq->private;
1020         struct hlist_node *e;
1021         struct hlist_head *l = iter->l;
1022         int idx;
1023
1024         ++*pos;
1025         if (v == SEQ_START_TOKEN)
1026                 return ip_vs_conn_array(seq, 0);
1027
1028         /* more on same hash chain? */
1029         e = rcu_dereference(hlist_next_rcu(&cp->c_list));
1030         if (e)
1031                 return hlist_entry(e, struct ip_vs_conn, c_list);
1032
1033         idx = l - ip_vs_conn_tab;
1034         while (++idx < ip_vs_conn_tab_size) {
1035                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1036                         iter->l = &ip_vs_conn_tab[idx];
1037                         return cp;
1038                 }
1039                 cond_resched_rcu();
1040         }
1041         iter->l = NULL;
1042         return NULL;
1043 }
1044
1045 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1046         __releases(RCU)
1047 {
1048         rcu_read_unlock();
1049 }
1050
1051 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1052 {
1053
1054         if (v == SEQ_START_TOKEN)
1055                 seq_puts(seq,
1056    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires PEName PEData\n");
1057         else {
1058                 const struct ip_vs_conn *cp = v;
1059                 struct net *net = seq_file_net(seq);
1060                 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1061                 size_t len = 0;
1062                 char dbuf[IP_VS_ADDRSTRLEN];
1063
1064                 if (!ip_vs_conn_net_eq(cp, net))
1065                         return 0;
1066                 if (cp->pe_data) {
1067                         pe_data[0] = ' ';
1068                         len = strlen(cp->pe->name);
1069                         memcpy(pe_data + 1, cp->pe->name, len);
1070                         pe_data[len + 1] = ' ';
1071                         len += 2;
1072                         len += cp->pe->show_pe_data(cp, pe_data + len);
1073                 }
1074                 pe_data[len] = '\0';
1075
1076 #ifdef CONFIG_IP_VS_IPV6
1077                 if (cp->daf == AF_INET6)
1078                         snprintf(dbuf, sizeof(dbuf), "%pI6", &cp->daddr.in6);
1079                 else
1080 #endif
1081                         snprintf(dbuf, sizeof(dbuf), "%08X",
1082                                  ntohl(cp->daddr.ip));
1083
1084 #ifdef CONFIG_IP_VS_IPV6
1085                 if (cp->af == AF_INET6)
1086                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1087                                 "%s %04X %-11s %7lu%s\n",
1088                                 ip_vs_proto_name(cp->protocol),
1089                                 &cp->caddr.in6, ntohs(cp->cport),
1090                                 &cp->vaddr.in6, ntohs(cp->vport),
1091                                 dbuf, ntohs(cp->dport),
1092                                 ip_vs_state_name(cp->protocol, cp->state),
1093                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1094                 else
1095 #endif
1096                         seq_printf(seq,
1097                                 "%-3s %08X %04X %08X %04X"
1098                                 " %s %04X %-11s %7lu%s\n",
1099                                 ip_vs_proto_name(cp->protocol),
1100                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1101                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1102                                 dbuf, ntohs(cp->dport),
1103                                 ip_vs_state_name(cp->protocol, cp->state),
1104                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1105         }
1106         return 0;
1107 }
1108
1109 static const struct seq_operations ip_vs_conn_seq_ops = {
1110         .start = ip_vs_conn_seq_start,
1111         .next  = ip_vs_conn_seq_next,
1112         .stop  = ip_vs_conn_seq_stop,
1113         .show  = ip_vs_conn_seq_show,
1114 };
1115
1116 static int ip_vs_conn_open(struct inode *inode, struct file *file)
1117 {
1118         return seq_open_net(inode, file, &ip_vs_conn_seq_ops,
1119                             sizeof(struct ip_vs_iter_state));
1120 }
1121
1122 static const struct file_operations ip_vs_conn_fops = {
1123         .owner   = THIS_MODULE,
1124         .open    = ip_vs_conn_open,
1125         .read    = seq_read,
1126         .llseek  = seq_lseek,
1127         .release = seq_release_net,
1128 };
1129
1130 static const char *ip_vs_origin_name(unsigned int flags)
1131 {
1132         if (flags & IP_VS_CONN_F_SYNC)
1133                 return "SYNC";
1134         else
1135                 return "LOCAL";
1136 }
1137
1138 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1139 {
1140         char dbuf[IP_VS_ADDRSTRLEN];
1141
1142         if (v == SEQ_START_TOKEN)
1143                 seq_puts(seq,
1144    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
1145         else {
1146                 const struct ip_vs_conn *cp = v;
1147                 struct net *net = seq_file_net(seq);
1148
1149                 if (!ip_vs_conn_net_eq(cp, net))
1150                         return 0;
1151
1152 #ifdef CONFIG_IP_VS_IPV6
1153                 if (cp->daf == AF_INET6)
1154                         snprintf(dbuf, sizeof(dbuf), "%pI6", &cp->daddr.in6);
1155                 else
1156 #endif
1157                         snprintf(dbuf, sizeof(dbuf), "%08X",
1158                                  ntohl(cp->daddr.ip));
1159
1160 #ifdef CONFIG_IP_VS_IPV6
1161                 if (cp->af == AF_INET6)
1162                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1163                                 "%s %04X %-11s %-6s %7lu\n",
1164                                 ip_vs_proto_name(cp->protocol),
1165                                 &cp->caddr.in6, ntohs(cp->cport),
1166                                 &cp->vaddr.in6, ntohs(cp->vport),
1167                                 dbuf, ntohs(cp->dport),
1168                                 ip_vs_state_name(cp->protocol, cp->state),
1169                                 ip_vs_origin_name(cp->flags),
1170                                 (cp->timer.expires-jiffies)/HZ);
1171                 else
1172 #endif
1173                         seq_printf(seq,
1174                                 "%-3s %08X %04X %08X %04X "
1175                                 "%s %04X %-11s %-6s %7lu\n",
1176                                 ip_vs_proto_name(cp->protocol),
1177                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1178                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1179                                 dbuf, ntohs(cp->dport),
1180                                 ip_vs_state_name(cp->protocol, cp->state),
1181                                 ip_vs_origin_name(cp->flags),
1182                                 (cp->timer.expires-jiffies)/HZ);
1183         }
1184         return 0;
1185 }
1186
1187 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1188         .start = ip_vs_conn_seq_start,
1189         .next  = ip_vs_conn_seq_next,
1190         .stop  = ip_vs_conn_seq_stop,
1191         .show  = ip_vs_conn_sync_seq_show,
1192 };
1193
1194 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
1195 {
1196         return seq_open_net(inode, file, &ip_vs_conn_sync_seq_ops,
1197                             sizeof(struct ip_vs_iter_state));
1198 }
1199
1200 static const struct file_operations ip_vs_conn_sync_fops = {
1201         .owner   = THIS_MODULE,
1202         .open    = ip_vs_conn_sync_open,
1203         .read    = seq_read,
1204         .llseek  = seq_lseek,
1205         .release = seq_release_net,
1206 };
1207
1208 #endif
1209
1210
1211 /*
1212  *      Randomly drop connection entries before running out of memory
1213  */
1214 static inline int todrop_entry(struct ip_vs_conn *cp)
1215 {
1216         /*
1217          * The drop rate array needs tuning for real environments.
1218          * Called from timer bh only => no locking
1219          */
1220         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1221         static char todrop_counter[9] = {0};
1222         int i;
1223
1224         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1225            This will leave enough time for normal connection to get
1226            through. */
1227         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1228                 return 0;
1229
1230         /* Don't drop the entry if its number of incoming packets is not
1231            located in [0, 8] */
1232         i = atomic_read(&cp->in_pkts);
1233         if (i > 8 || i < 0) return 0;
1234
1235         if (!todrop_rate[i]) return 0;
1236         if (--todrop_counter[i] > 0) return 0;
1237
1238         todrop_counter[i] = todrop_rate[i];
1239         return 1;
1240 }
1241
1242 /* Called from keventd and must protect itself from softirqs */
1243 void ip_vs_random_dropentry(struct net *net)
1244 {
1245         int idx;
1246         struct ip_vs_conn *cp, *cp_c;
1247
1248         rcu_read_lock();
1249         /*
1250          * Randomly scan 1/32 of the whole table every second
1251          */
1252         for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1253                 unsigned int hash = prandom_u32() & ip_vs_conn_tab_mask;
1254
1255                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
1256                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1257                                 /* connection template */
1258                                 continue;
1259                         if (!ip_vs_conn_net_eq(cp, net))
1260                                 continue;
1261                         if (cp->protocol == IPPROTO_TCP) {
1262                                 switch(cp->state) {
1263                                 case IP_VS_TCP_S_SYN_RECV:
1264                                 case IP_VS_TCP_S_SYNACK:
1265                                         break;
1266
1267                                 case IP_VS_TCP_S_ESTABLISHED:
1268                                         if (todrop_entry(cp))
1269                                                 break;
1270                                         continue;
1271
1272                                 default:
1273                                         continue;
1274                                 }
1275                         } else if (cp->protocol == IPPROTO_SCTP) {
1276                                 switch (cp->state) {
1277                                 case IP_VS_SCTP_S_INIT1:
1278                                 case IP_VS_SCTP_S_INIT:
1279                                         break;
1280                                 case IP_VS_SCTP_S_ESTABLISHED:
1281                                         if (todrop_entry(cp))
1282                                                 break;
1283                                         continue;
1284                                 default:
1285                                         continue;
1286                                 }
1287                         } else {
1288                                 if (!todrop_entry(cp))
1289                                         continue;
1290                         }
1291
1292                         IP_VS_DBG(4, "del connection\n");
1293                         ip_vs_conn_expire_now(cp);
1294                         cp_c = cp->control;
1295                         /* cp->control is valid only with reference to cp */
1296                         if (cp_c && __ip_vs_conn_get(cp)) {
1297                                 IP_VS_DBG(4, "del conn template\n");
1298                                 ip_vs_conn_expire_now(cp_c);
1299                                 __ip_vs_conn_put(cp);
1300                         }
1301                 }
1302                 cond_resched_rcu();
1303         }
1304         rcu_read_unlock();
1305 }
1306
1307
1308 /*
1309  *      Flush all the connection entries in the ip_vs_conn_tab
1310  */
1311 static void ip_vs_conn_flush(struct net *net)
1312 {
1313         int idx;
1314         struct ip_vs_conn *cp, *cp_c;
1315         struct netns_ipvs *ipvs = net_ipvs(net);
1316
1317 flush_again:
1318         rcu_read_lock();
1319         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1320
1321                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1322                         if (!ip_vs_conn_net_eq(cp, net))
1323                                 continue;
1324                         IP_VS_DBG(4, "del connection\n");
1325                         ip_vs_conn_expire_now(cp);
1326                         cp_c = cp->control;
1327                         /* cp->control is valid only with reference to cp */
1328                         if (cp_c && __ip_vs_conn_get(cp)) {
1329                                 IP_VS_DBG(4, "del conn template\n");
1330                                 ip_vs_conn_expire_now(cp_c);
1331                                 __ip_vs_conn_put(cp);
1332                         }
1333                 }
1334                 cond_resched_rcu();
1335         }
1336         rcu_read_unlock();
1337
1338         /* the counter may be not NULL, because maybe some conn entries
1339            are run by slow timer handler or unhashed but still referred */
1340         if (atomic_read(&ipvs->conn_count) != 0) {
1341                 schedule();
1342                 goto flush_again;
1343         }
1344 }
1345 /*
1346  * per netns init and exit
1347  */
1348 int __net_init ip_vs_conn_net_init(struct net *net)
1349 {
1350         struct netns_ipvs *ipvs = net_ipvs(net);
1351
1352         atomic_set(&ipvs->conn_count, 0);
1353
1354         proc_create("ip_vs_conn", 0, net->proc_net, &ip_vs_conn_fops);
1355         proc_create("ip_vs_conn_sync", 0, net->proc_net, &ip_vs_conn_sync_fops);
1356         return 0;
1357 }
1358
1359 void __net_exit ip_vs_conn_net_cleanup(struct net *net)
1360 {
1361         /* flush all the connection entries first */
1362         ip_vs_conn_flush(net);
1363         remove_proc_entry("ip_vs_conn", net->proc_net);
1364         remove_proc_entry("ip_vs_conn_sync", net->proc_net);
1365 }
1366
1367 int __init ip_vs_conn_init(void)
1368 {
1369         int idx;
1370
1371         /* Compute size and mask */
1372         ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1373         ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1374
1375         /*
1376          * Allocate the connection hash table and initialize its list heads
1377          */
1378         ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size * sizeof(*ip_vs_conn_tab));
1379         if (!ip_vs_conn_tab)
1380                 return -ENOMEM;
1381
1382         /* Allocate ip_vs_conn slab cache */
1383         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1384                                               sizeof(struct ip_vs_conn), 0,
1385                                               SLAB_HWCACHE_ALIGN, NULL);
1386         if (!ip_vs_conn_cachep) {
1387                 vfree(ip_vs_conn_tab);
1388                 return -ENOMEM;
1389         }
1390
1391         pr_info("Connection hash table configured "
1392                 "(size=%d, memory=%ldKbytes)\n",
1393                 ip_vs_conn_tab_size,
1394                 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1395         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1396                   sizeof(struct ip_vs_conn));
1397
1398         for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1399                 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1400
1401         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1402                 spin_lock_init(&__ip_vs_conntbl_lock_array[idx].l);
1403         }
1404
1405         /* calculate the random value for connection hash */
1406         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1407
1408         return 0;
1409 }
1410
1411 void ip_vs_conn_cleanup(void)
1412 {
1413         /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1414         rcu_barrier();
1415         /* Release the empty cache */
1416         kmem_cache_destroy(ip_vs_conn_cachep);
1417         vfree(ip_vs_conn_tab);
1418 }