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 static void ip_vs_conn_expire(struct timer_list *t);
 108
 109 /*
 110  *      Returns hash value for IPVS connection entry
 111  */
 112 static unsigned int ip_vs_conn_hashkey(struct netns_ipvs *ipvs, int af, unsigned int proto,
 113                                        const union nf_inet_addr *addr,
 114                                        __be16 port)
 115 {
 116 #ifdef CONFIG_IP_VS_IPV6
 117         if (af == AF_INET6)
 118                 return (jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
 119                                     (__force u32)port, proto, ip_vs_conn_rnd) ^
 120                         ((size_t)ipvs>>8)) & ip_vs_conn_tab_mask;
 121 #endif
 122         return (jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
 123                             ip_vs_conn_rnd) ^
 124                 ((size_t)ipvs>>8)) & ip_vs_conn_tab_mask;
 125 }
 126
 127 static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param *p,
 128                                              bool inverse)
 129 {
 130         const union nf_inet_addr *addr;
 131         __be16 port;
 132
 133         if (p->pe_data && p->pe->hashkey_raw)
 134                 return p->pe->hashkey_raw(p, ip_vs_conn_rnd, inverse) &
 135                         ip_vs_conn_tab_mask;
 136
 137         if (likely(!inverse)) {
 138                 addr = p->caddr;
 139                 port = p->cport;
 140         } else {
 141                 addr = p->vaddr;
 142                 port = p->vport;
 143         }
 144
 145         return ip_vs_conn_hashkey(p->ipvs, p->af, p->protocol, addr, port);
 146 }
 147
 148 static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn *cp)
 149 {
 150         struct ip_vs_conn_param p;
 151
 152         ip_vs_conn_fill_param(cp->ipvs, cp->af, cp->protocol,
 153                               &cp->caddr, cp->cport, NULL, 0, &p);
 154
 155         if (cp->pe) {
 156                 p.pe = cp->pe;
 157                 p.pe_data = cp->pe_data;
 158                 p.pe_data_len = cp->pe_data_len;
 159         }
 160
 161         return ip_vs_conn_hashkey_param(&p, false);
 162 }
 163
 164 /*
 165  *      Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
 166  *      returns bool success.
 167  */
 168 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
 169 {
 170         unsigned int hash;
 171         int ret;
 172
 173         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
 174                 return 0;
 175
 176         /* Hash by protocol, client address and port */
 177         hash = ip_vs_conn_hashkey_conn(cp);
 178
 179         ct_write_lock_bh(hash);
 180         spin_lock(&cp->lock);
 181
 182         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
 183                 cp->flags |= IP_VS_CONN_F_HASHED;
 184                 refcount_inc(&cp->refcnt);
 185                 hlist_add_head_rcu(&cp->c_list, &ip_vs_conn_tab[hash]);
 186                 ret = 1;
 187         } else {
 188                 pr_err("%s(): request for already hashed, called from %pS\n",
 189                        __func__, __builtin_return_address(0));
 190                 ret = 0;
 191         }
 192
 193         spin_unlock(&cp->lock);
 194         ct_write_unlock_bh(hash);
 195
 196         return ret;
 197 }
 198
 199
 200 /*
 201  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
 202  *      returns bool success. Caller should hold conn reference.
 203  */
 204 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
 205 {
 206         unsigned int hash;
 207         int ret;
 208
 209         /* unhash it and decrease its reference counter */
 210         hash = ip_vs_conn_hashkey_conn(cp);
 211
 212         ct_write_lock_bh(hash);
 213         spin_lock(&cp->lock);
 214
 215         if (cp->flags & IP_VS_CONN_F_HASHED) {
 216                 hlist_del_rcu(&cp->c_list);
 217                 cp->flags &= ~IP_VS_CONN_F_HASHED;
 218                 refcount_dec(&cp->refcnt);
 219                 ret = 1;
 220         } else
 221                 ret = 0;
 222
 223         spin_unlock(&cp->lock);
 224         ct_write_unlock_bh(hash);
 225
 226         return ret;
 227 }
 228
 229 /* Try to unlink ip_vs_conn from ip_vs_conn_tab.
 230  * returns bool success.
 231  */
 232 static inline bool ip_vs_conn_unlink(struct ip_vs_conn *cp)
 233 {
 234         unsigned int hash;
 235         bool ret = false;
 236
 237         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
 238                 return refcount_dec_if_one(&cp->refcnt);
 239
 240         hash = ip_vs_conn_hashkey_conn(cp);
 241
 242         ct_write_lock_bh(hash);
 243         spin_lock(&cp->lock);
 244
 245         if (cp->flags & IP_VS_CONN_F_HASHED) {
 246                 /* Decrease refcnt and unlink conn only if we are last user */
 247                 if (refcount_dec_if_one(&cp->refcnt)) {
 248                         hlist_del_rcu(&cp->c_list);
 249                         cp->flags &= ~IP_VS_CONN_F_HASHED;
 250                         ret = true;
 251                 }
 252         }
 253
 254         spin_unlock(&cp->lock);
 255         ct_write_unlock_bh(hash);
 256
 257         return ret;
 258 }
 259
 260
 261 /*
 262  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 263  *  Called for pkts coming from OUTside-to-INside.
 264  *      p->caddr, p->cport: pkt source address (foreign host)
 265  *      p->vaddr, p->vport: pkt dest address (load balancer)
 266  */
 267 static inline struct ip_vs_conn *
 268 __ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
 269 {
 270         unsigned int hash;
 271         struct ip_vs_conn *cp;
 272
 273         hash = ip_vs_conn_hashkey_param(p, false);
 274
 275         rcu_read_lock();
 276
 277         hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
 278                 if (p->cport == cp->cport && p->vport == cp->vport &&
 279                     cp->af == p->af &&
 280                     ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
 281                     ip_vs_addr_equal(p->af, p->vaddr, &cp->vaddr) &&
 282                     ((!p->cport) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
 283                     p->protocol == cp->protocol &&
 284                     cp->ipvs == p->ipvs) {
 285                         if (!__ip_vs_conn_get(cp))
 286                                 continue;
 287                         /* HIT */
 288                         rcu_read_unlock();
 289                         return cp;
 290                 }
 291         }
 292
 293         rcu_read_unlock();
 294
 295         return NULL;
 296 }
 297
 298 struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
 299 {
 300         struct ip_vs_conn *cp;
 301
 302         cp = __ip_vs_conn_in_get(p);
 303         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt)) {
 304                 struct ip_vs_conn_param cport_zero_p = *p;
 305                 cport_zero_p.cport = 0;
 306                 cp = __ip_vs_conn_in_get(&cport_zero_p);
 307         }
 308
 309         IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
 310                       ip_vs_proto_name(p->protocol),
 311                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
 312                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
 313                       cp ? "hit" : "not hit");
 314
 315         return cp;
 316 }
 317
 318 static int
 319 ip_vs_conn_fill_param_proto(struct netns_ipvs *ipvs,
 320                             int af, const struct sk_buff *skb,
 321                             const struct ip_vs_iphdr *iph,
 322                             struct ip_vs_conn_param *p)
 323 {
 324         __be16 _ports[2], *pptr;
 325
 326         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 327         if (pptr == NULL)
 328                 return 1;
 329
 330         if (likely(!ip_vs_iph_inverse(iph)))
 331                 ip_vs_conn_fill_param(ipvs, af, iph->protocol, &iph->saddr,
 332                                       pptr[0], &iph->daddr, pptr[1], p);
 333         else
 334                 ip_vs_conn_fill_param(ipvs, af, iph->protocol, &iph->daddr,
 335                                       pptr[1], &iph->saddr, pptr[0], p);
 336         return 0;
 337 }
 338
 339 struct ip_vs_conn *
 340 ip_vs_conn_in_get_proto(struct netns_ipvs *ipvs, int af,
 341                         const struct sk_buff *skb,
 342                         const struct ip_vs_iphdr *iph)
 343 {
 344         struct ip_vs_conn_param p;
 345
 346         if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, &p))
 347                 return NULL;
 348
 349         return ip_vs_conn_in_get(&p);
 350 }
 351 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
 352
 353 /* Get reference to connection template */
 354 struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p)
 355 {
 356         unsigned int hash;
 357         struct ip_vs_conn *cp;
 358
 359         hash = ip_vs_conn_hashkey_param(p, false);
 360
 361         rcu_read_lock();
 362
 363         hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
 364                 if (unlikely(p->pe_data && p->pe->ct_match)) {
 365                         if (cp->ipvs != p->ipvs)
 366                                 continue;
 367                         if (p->pe == cp->pe && p->pe->ct_match(p, cp)) {
 368                                 if (__ip_vs_conn_get(cp))
 369                                         goto out;
 370                         }
 371                         continue;
 372                 }
 373
 374                 if (cp->af == p->af &&
 375                     ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
 376                     /* protocol should only be IPPROTO_IP if
 377                      * p->vaddr is a fwmark */
 378                     ip_vs_addr_equal(p->protocol == IPPROTO_IP ? AF_UNSPEC :
 379                                      p->af, p->vaddr, &cp->vaddr) &&
 380                     p->vport == cp->vport && p->cport == cp->cport &&
 381                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
 382                     p->protocol == cp->protocol &&
 383                     cp->ipvs == p->ipvs) {
 384                         if (__ip_vs_conn_get(cp))
 385                                 goto out;
 386                 }
 387         }
 388         cp = NULL;
 389
 390   out:
 391         rcu_read_unlock();
 392
 393         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
 394                       ip_vs_proto_name(p->protocol),
 395                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
 396                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
 397                       cp ? "hit" : "not hit");
 398
 399         return cp;
 400 }
 401
 402 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 403  * Called for pkts coming from inside-to-OUTside.
 404  *      p->caddr, p->cport: pkt source address (inside host)
 405  *      p->vaddr, p->vport: pkt dest address (foreign host) */
 406 struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
 407 {
 408         unsigned int hash;
 409         struct ip_vs_conn *cp, *ret=NULL;
 410
 411         /*
 412          *      Check for "full" addressed entries
 413          */
 414         hash = ip_vs_conn_hashkey_param(p, true);
 415
 416         rcu_read_lock();
 417
 418         hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
 419                 if (p->vport == cp->cport && p->cport == cp->dport &&
 420                     cp->af == p->af &&
 421                     ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
 422                     ip_vs_addr_equal(p->af, p->caddr, &cp->daddr) &&
 423                     p->protocol == cp->protocol &&
 424                     cp->ipvs == p->ipvs) {
 425                         if (!__ip_vs_conn_get(cp))
 426                                 continue;
 427                         /* HIT */
 428                         ret = cp;
 429                         break;
 430                 }
 431         }
 432
 433         rcu_read_unlock();
 434
 435         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
 436                       ip_vs_proto_name(p->protocol),
 437                       IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
 438                       IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
 439                       ret ? "hit" : "not hit");
 440
 441         return ret;
 442 }
 443
 444 struct ip_vs_conn *
 445 ip_vs_conn_out_get_proto(struct netns_ipvs *ipvs, int af,
 446                          const struct sk_buff *skb,
 447                          const struct ip_vs_iphdr *iph)
 448 {
 449         struct ip_vs_conn_param p;
 450
 451         if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, &p))
 452                 return NULL;
 453
 454         return ip_vs_conn_out_get(&p);
 455 }
 456 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
 457
 458 /*
 459  *      Put back the conn and restart its timer with its timeout
 460  */
 461 static void __ip_vs_conn_put_timer(struct ip_vs_conn *cp)
 462 {
 463         unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
 464                 0 : cp->timeout;
 465         mod_timer(&cp->timer, jiffies+t);
 466
 467         __ip_vs_conn_put(cp);
 468 }
 469
 470 void ip_vs_conn_put(struct ip_vs_conn *cp)
 471 {
 472         if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) &&
 473             (refcount_read(&cp->refcnt) == 1) &&
 474             !timer_pending(&cp->timer))
 475                 /* expire connection immediately */
 476                 ip_vs_conn_expire(&cp->timer);
 477         else
 478                 __ip_vs_conn_put_timer(cp);
 479 }
 480
 481 /*
 482  *      Fill a no_client_port connection with a client port number
 483  */
 484 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
 485 {
 486         if (ip_vs_conn_unhash(cp)) {
 487                 spin_lock_bh(&cp->lock);
 488                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
 489                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 490                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
 491                         cp->cport = cport;
 492                 }
 493                 spin_unlock_bh(&cp->lock);
 494
 495                 /* hash on new dport */
 496                 ip_vs_conn_hash(cp);
 497         }
 498 }
 499
 500
 501 /*
 502  *      Bind a connection entry with the corresponding packet_xmit.
 503  *      Called by ip_vs_conn_new.
 504  */
 505 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
 506 {
 507         switch (IP_VS_FWD_METHOD(cp)) {
 508         case IP_VS_CONN_F_MASQ:
 509                 cp->packet_xmit = ip_vs_nat_xmit;
 510                 break;
 511
 512         case IP_VS_CONN_F_TUNNEL:
 513 #ifdef CONFIG_IP_VS_IPV6
 514                 if (cp->daf == AF_INET6)
 515                         cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 516                 else
 517 #endif
 518                         cp->packet_xmit = ip_vs_tunnel_xmit;
 519                 break;
 520
 521         case IP_VS_CONN_F_DROUTE:
 522                 cp->packet_xmit = ip_vs_dr_xmit;
 523                 break;
 524
 525         case IP_VS_CONN_F_LOCALNODE:
 526                 cp->packet_xmit = ip_vs_null_xmit;
 527                 break;
 528
 529         case IP_VS_CONN_F_BYPASS:
 530                 cp->packet_xmit = ip_vs_bypass_xmit;
 531                 break;
 532         }
 533 }
 534
 535 #ifdef CONFIG_IP_VS_IPV6
 536 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
 537 {
 538         switch (IP_VS_FWD_METHOD(cp)) {
 539         case IP_VS_CONN_F_MASQ:
 540                 cp->packet_xmit = ip_vs_nat_xmit_v6;
 541                 break;
 542
 543         case IP_VS_CONN_F_TUNNEL:
 544                 if (cp->daf == AF_INET6)
 545                         cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 546                 else
 547                         cp->packet_xmit = ip_vs_tunnel_xmit;
 548                 break;
 549
 550         case IP_VS_CONN_F_DROUTE:
 551                 cp->packet_xmit = ip_vs_dr_xmit_v6;
 552                 break;
 553
 554         case IP_VS_CONN_F_LOCALNODE:
 555                 cp->packet_xmit = ip_vs_null_xmit;
 556                 break;
 557
 558         case IP_VS_CONN_F_BYPASS:
 559                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
 560                 break;
 561         }
 562 }
 563 #endif
 564
 565
 566 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
 567 {
 568         return atomic_read(&dest->activeconns)
 569                 + atomic_read(&dest->inactconns);
 570 }
 571
 572 /*
 573  *      Bind a connection entry with a virtual service destination
 574  *      Called just after a new connection entry is created.
 575  */
 576 static inline void
 577 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
 578 {
 579         unsigned int conn_flags;
 580         __u32 flags;
 581
 582         /* if dest is NULL, then return directly */
 583         if (!dest)
 584                 return;
 585
 586         /* Increase the refcnt counter of the dest */
 587         ip_vs_dest_hold(dest);
 588
 589         conn_flags = atomic_read(&dest->conn_flags);
 590         if (cp->protocol != IPPROTO_UDP)
 591                 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
 592         flags = cp->flags;
 593         /* Bind with the destination and its corresponding transmitter */
 594         if (flags & IP_VS_CONN_F_SYNC) {
 595                 /* if the connection is not template and is created
 596                  * by sync, preserve the activity flag.
 597                  */
 598                 if (!(flags & IP_VS_CONN_F_TEMPLATE))
 599                         conn_flags &= ~IP_VS_CONN_F_INACTIVE;
 600                 /* connections inherit forwarding method from dest */
 601                 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
 602         }
 603         flags |= conn_flags;
 604         cp->flags = flags;
 605         cp->dest = dest;
 606
 607         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
 608                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 609                       "dest->refcnt:%d\n",
 610                       ip_vs_proto_name(cp->protocol),
 611                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 612                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 613                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 614                       ip_vs_fwd_tag(cp), cp->state,
 615                       cp->flags, refcount_read(&cp->refcnt),
 616                       refcount_read(&dest->refcnt));
 617
 618         /* Update the connection counters */
 619         if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
 620                 /* It is a normal connection, so modify the counters
 621                  * according to the flags, later the protocol can
 622                  * update them on state change
 623                  */
 624                 if (!(flags & IP_VS_CONN_F_INACTIVE))
 625                         atomic_inc(&dest->activeconns);
 626                 else
 627                         atomic_inc(&dest->inactconns);
 628         } else {
 629                 /* It is a persistent connection/template, so increase
 630                    the persistent connection counter */
 631                 atomic_inc(&dest->persistconns);
 632         }
 633
 634         if (dest->u_threshold != 0 &&
 635             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
 636                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
 637 }
 638
 639
 640 /*
 641  * Check if there is a destination for the connection, if so
 642  * bind the connection to the destination.
 643  */
 644 void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
 645 {
 646         struct ip_vs_dest *dest;
 647
 648         rcu_read_lock();
 649
 650         /* This function is only invoked by the synchronization code. We do
 651          * not currently support heterogeneous pools with synchronization,
 652          * so we can make the assumption that the svc_af is the same as the
 653          * dest_af
 654          */
 655         dest = ip_vs_find_dest(cp->ipvs, cp->af, cp->af, &cp->daddr,
 656                                cp->dport, &cp->vaddr, cp->vport,
 657                                cp->protocol, cp->fwmark, cp->flags);
 658         if (dest) {
 659                 struct ip_vs_proto_data *pd;
 660
 661                 spin_lock_bh(&cp->lock);
 662                 if (cp->dest) {
 663                         spin_unlock_bh(&cp->lock);
 664                         rcu_read_unlock();
 665                         return;
 666                 }
 667
 668                 /* Applications work depending on the forwarding method
 669                  * but better to reassign them always when binding dest */
 670                 if (cp->app)
 671                         ip_vs_unbind_app(cp);
 672
 673                 ip_vs_bind_dest(cp, dest);
 674                 spin_unlock_bh(&cp->lock);
 675
 676                 /* Update its packet transmitter */
 677                 cp->packet_xmit = NULL;
 678 #ifdef CONFIG_IP_VS_IPV6
 679                 if (cp->af == AF_INET6)
 680                         ip_vs_bind_xmit_v6(cp);
 681                 else
 682 #endif
 683                         ip_vs_bind_xmit(cp);
 684
 685                 pd = ip_vs_proto_data_get(cp->ipvs, cp->protocol);
 686                 if (pd && atomic_read(&pd->appcnt))
 687                         ip_vs_bind_app(cp, pd->pp);
 688         }
 689         rcu_read_unlock();
 690 }
 691
 692
 693 /*
 694  *      Unbind a connection entry with its VS destination
 695  *      Called by the ip_vs_conn_expire function.
 696  */
 697 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
 698 {
 699         struct ip_vs_dest *dest = cp->dest;
 700
 701         if (!dest)
 702                 return;
 703
 704         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
 705                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 706                       "dest->refcnt:%d\n",
 707                       ip_vs_proto_name(cp->protocol),
 708                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 709                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 710                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 711                       ip_vs_fwd_tag(cp), cp->state,
 712                       cp->flags, refcount_read(&cp->refcnt),
 713                       refcount_read(&dest->refcnt));
 714
 715         /* Update the connection counters */
 716         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 717                 /* It is a normal connection, so decrease the inactconns
 718                    or activeconns counter */
 719                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
 720                         atomic_dec(&dest->inactconns);
 721                 } else {
 722                         atomic_dec(&dest->activeconns);
 723                 }
 724         } else {
 725                 /* It is a persistent connection/template, so decrease
 726                    the persistent connection counter */
 727                 atomic_dec(&dest->persistconns);
 728         }
 729
 730         if (dest->l_threshold != 0) {
 731                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
 732                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 733         } else if (dest->u_threshold != 0) {
 734                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
 735                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 736         } else {
 737                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 738                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 739         }
 740
 741         ip_vs_dest_put(dest);
 742 }
 743
 744 static int expire_quiescent_template(struct netns_ipvs *ipvs,
 745                                      struct ip_vs_dest *dest)
 746 {
 747 #ifdef CONFIG_SYSCTL
 748         return ipvs->sysctl_expire_quiescent_template &&
 749                 (atomic_read(&dest->weight) == 0);
 750 #else
 751         return 0;
 752 #endif
 753 }
 754
 755 /*
 756  *      Checking if the destination of a connection template is available.
 757  *      If available, return 1, otherwise invalidate this connection
 758  *      template and return 0.
 759  */
 760 int ip_vs_check_template(struct ip_vs_conn *ct, struct ip_vs_dest *cdest)
 761 {
 762         struct ip_vs_dest *dest = ct->dest;
 763         struct netns_ipvs *ipvs = ct->ipvs;
 764
 765         /*
 766          * Checking the dest server status.
 767          */
 768         if ((dest == NULL) ||
 769             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
 770             expire_quiescent_template(ipvs, dest) ||
 771             (cdest && (dest != cdest))) {
 772                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
 773                               "protocol %s s:%s:%d v:%s:%d "
 774                               "-> d:%s:%d\n",
 775                               ip_vs_proto_name(ct->protocol),
 776                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
 777                               ntohs(ct->cport),
 778                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
 779                               ntohs(ct->vport),
 780                               IP_VS_DBG_ADDR(ct->daf, &ct->daddr),
 781                               ntohs(ct->dport));
 782
 783                 /*
 784                  * Invalidate the connection template
 785                  */
 786                 if (ct->vport != htons(0xffff)) {
 787                         if (ip_vs_conn_unhash(ct)) {
 788                                 ct->dport = htons(0xffff);
 789                                 ct->vport = htons(0xffff);
 790                                 ct->cport = 0;
 791                                 ip_vs_conn_hash(ct);
 792                         }
 793                 }
 794
 795                 /*
 796                  * Simply decrease the refcnt of the template,
 797                  * don't restart its timer.
 798                  */
 799                 __ip_vs_conn_put(ct);
 800                 return 0;
 801         }
 802         return 1;
 803 }
 804
 805 static void ip_vs_conn_rcu_free(struct rcu_head *head)
 806 {
 807         struct ip_vs_conn *cp = container_of(head, struct ip_vs_conn,
 808                                              rcu_head);
 809
 810         ip_vs_pe_put(cp->pe);
 811         kfree(cp->pe_data);
 812         kmem_cache_free(ip_vs_conn_cachep, cp);
 813 }
 814
 815 static void ip_vs_conn_expire(struct timer_list *t)
 816 {
 817         struct ip_vs_conn *cp = from_timer(cp, t, timer);
 818         struct netns_ipvs *ipvs = cp->ipvs;
 819
 820         /*
 821          *      do I control anybody?
 822          */
 823         if (atomic_read(&cp->n_control))
 824                 goto expire_later;
 825
 826         /* Unlink conn if not referenced anymore */
 827         if (likely(ip_vs_conn_unlink(cp))) {
 828                 /* delete the timer if it is activated by other users */
 829                 del_timer(&cp->timer);
 830
 831                 /* does anybody control me? */
 832                 if (cp->control)
 833                         ip_vs_control_del(cp);
 834
 835                 if ((cp->flags & IP_VS_CONN_F_NFCT) &&
 836                     !(cp->flags & IP_VS_CONN_F_ONE_PACKET)) {
 837                         /* Do not access conntracks during subsys cleanup
 838                          * because nf_conntrack_find_get can not be used after
 839                          * conntrack cleanup for the net.
 840                          */
 841                         smp_rmb();
 842                         if (ipvs->enable)
 843                                 ip_vs_conn_drop_conntrack(cp);
 844                 }
 845
 846                 if (unlikely(cp->app != NULL))
 847                         ip_vs_unbind_app(cp);
 848                 ip_vs_unbind_dest(cp);
 849                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
 850                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 851                 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
 852                         ip_vs_conn_rcu_free(&cp->rcu_head);
 853                 else
 854                         call_rcu(&cp->rcu_head, ip_vs_conn_rcu_free);
 855                 atomic_dec(&ipvs->conn_count);
 856                 return;
 857         }
 858
 859   expire_later:
 860         IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
 861                   refcount_read(&cp->refcnt),
 862                   atomic_read(&cp->n_control));
 863
 864         refcount_inc(&cp->refcnt);
 865         cp->timeout = 60*HZ;
 866
 867         if (ipvs->sync_state & IP_VS_STATE_MASTER)
 868                 ip_vs_sync_conn(ipvs, cp, sysctl_sync_threshold(ipvs));
 869
 870         __ip_vs_conn_put_timer(cp);
 871 }
 872
 873 /* Modify timer, so that it expires as soon as possible.
 874  * Can be called without reference only if under RCU lock.
 875  */
 876 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
 877 {
 878         /* Using mod_timer_pending will ensure the timer is not
 879          * modified after the final del_timer in ip_vs_conn_expire.
 880          */
 881         if (timer_pending(&cp->timer) &&
 882             time_after(cp->timer.expires, jiffies))
 883                 mod_timer_pending(&cp->timer, jiffies);
 884 }
 885
 886
 887 /*
 888  *      Create a new connection entry and hash it into the ip_vs_conn_tab
 889  */
 890 struct ip_vs_conn *
 891 ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af,
 892                const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
 893                struct ip_vs_dest *dest, __u32 fwmark)
 894 {
 895         struct ip_vs_conn *cp;
 896         struct netns_ipvs *ipvs = p->ipvs;
 897         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->ipvs,
 898                                                            p->protocol);
 899
 900         cp = kmem_cache_alloc(ip_vs_conn_cachep, GFP_ATOMIC);
 901         if (cp == NULL) {
 902                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
 903                 return NULL;
 904         }
 905
 906         INIT_HLIST_NODE(&cp->c_list);
 907         timer_setup(&cp->timer, ip_vs_conn_expire, 0);
 908         cp->ipvs           = ipvs;
 909         cp->af             = p->af;
 910         cp->daf            = dest_af;
 911         cp->protocol       = p->protocol;
 912         ip_vs_addr_set(p->af, &cp->caddr, p->caddr);
 913         cp->cport          = p->cport;
 914         /* proto should only be IPPROTO_IP if p->vaddr is a fwmark */
 915         ip_vs_addr_set(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
 916                        &cp->vaddr, p->vaddr);
 917         cp->vport          = p->vport;
 918         ip_vs_addr_set(cp->daf, &cp->daddr, daddr);
 919         cp->dport          = dport;
 920         cp->flags          = flags;
 921         cp->fwmark         = fwmark;
 922         if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
 923                 ip_vs_pe_get(p->pe);
 924                 cp->pe = p->pe;
 925                 cp->pe_data = p->pe_data;
 926                 cp->pe_data_len = p->pe_data_len;
 927         } else {
 928                 cp->pe = NULL;
 929                 cp->pe_data = NULL;
 930                 cp->pe_data_len = 0;
 931         }
 932         spin_lock_init(&cp->lock);
 933
 934         /*
 935          * Set the entry is referenced by the current thread before hashing
 936          * it in the table, so that other thread run ip_vs_random_dropentry
 937          * but cannot drop this entry.
 938          */
 939         refcount_set(&cp->refcnt, 1);
 940
 941         cp->control = NULL;
 942         atomic_set(&cp->n_control, 0);
 943         atomic_set(&cp->in_pkts, 0);
 944
 945         cp->packet_xmit = NULL;
 946         cp->app = NULL;
 947         cp->app_data = NULL;
 948         /* reset struct ip_vs_seq */
 949         cp->in_seq.delta = 0;
 950         cp->out_seq.delta = 0;
 951
 952         atomic_inc(&ipvs->conn_count);
 953         if (flags & IP_VS_CONN_F_NO_CPORT)
 954                 atomic_inc(&ip_vs_conn_no_cport_cnt);
 955
 956         /* Bind the connection with a destination server */
 957         cp->dest = NULL;
 958         ip_vs_bind_dest(cp, dest);
 959
 960         /* Set its state and timeout */
 961         cp->state = 0;
 962         cp->old_state = 0;
 963         cp->timeout = 3*HZ;
 964         cp->sync_endtime = jiffies & ~3UL;
 965
 966         /* Bind its packet transmitter */
 967 #ifdef CONFIG_IP_VS_IPV6
 968         if (p->af == AF_INET6)
 969                 ip_vs_bind_xmit_v6(cp);
 970         else
 971 #endif
 972                 ip_vs_bind_xmit(cp);
 973
 974         if (unlikely(pd && atomic_read(&pd->appcnt)))
 975                 ip_vs_bind_app(cp, pd->pp);
 976
 977         /*
 978          * Allow conntrack to be preserved. By default, conntrack
 979          * is created and destroyed for every packet.
 980          * Sometimes keeping conntrack can be useful for
 981          * IP_VS_CONN_F_ONE_PACKET too.
 982          */
 983
 984         if (ip_vs_conntrack_enabled(ipvs))
 985                 cp->flags |= IP_VS_CONN_F_NFCT;
 986
 987         /* Hash it in the ip_vs_conn_tab finally */
 988         ip_vs_conn_hash(cp);
 989
 990         return cp;
 991 }
 992
 993 /*
 994  *      /proc/net/ip_vs_conn entries
 995  */
 996 #ifdef CONFIG_PROC_FS
 997 struct ip_vs_iter_state {
 998         struct seq_net_private  p;
 999         struct hlist_head       *l;
1000 };
1001
1002 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
1003 {
1004         int idx;
1005         struct ip_vs_conn *cp;
1006         struct ip_vs_iter_state *iter = seq->private;
1007
1008         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1009                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1010                         /* __ip_vs_conn_get() is not needed by
1011                          * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
1012                          */
1013                         if (pos-- == 0) {
1014                                 iter->l = &ip_vs_conn_tab[idx];
1015                                 return cp;
1016                         }
1017                 }
1018                 cond_resched_rcu();
1019         }
1020
1021         return NULL;
1022 }
1023
1024 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
1025         __acquires(RCU)
1026 {
1027         struct ip_vs_iter_state *iter = seq->private;
1028
1029         iter->l = NULL;
1030         rcu_read_lock();
1031         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
1032 }
1033
1034 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1035 {
1036         struct ip_vs_conn *cp = v;
1037         struct ip_vs_iter_state *iter = seq->private;
1038         struct hlist_node *e;
1039         struct hlist_head *l = iter->l;
1040         int idx;
1041
1042         ++*pos;
1043         if (v == SEQ_START_TOKEN)
1044                 return ip_vs_conn_array(seq, 0);
1045
1046         /* more on same hash chain? */
1047         e = rcu_dereference(hlist_next_rcu(&cp->c_list));
1048         if (e)
1049                 return hlist_entry(e, struct ip_vs_conn, c_list);
1050
1051         idx = l - ip_vs_conn_tab;
1052         while (++idx < ip_vs_conn_tab_size) {
1053                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1054                         iter->l = &ip_vs_conn_tab[idx];
1055                         return cp;
1056                 }
1057                 cond_resched_rcu();
1058         }
1059         iter->l = NULL;
1060         return NULL;
1061 }
1062
1063 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1064         __releases(RCU)
1065 {
1066         rcu_read_unlock();
1067 }
1068
1069 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1070 {
1071
1072         if (v == SEQ_START_TOKEN)
1073                 seq_puts(seq,
1074    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires PEName PEData\n");
1075         else {
1076                 const struct ip_vs_conn *cp = v;
1077                 struct net *net = seq_file_net(seq);
1078                 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1079                 size_t len = 0;
1080                 char dbuf[IP_VS_ADDRSTRLEN];
1081
1082                 if (!net_eq(cp->ipvs->net, net))
1083                         return 0;
1084                 if (cp->pe_data) {
1085                         pe_data[0] = ' ';
1086                         len = strlen(cp->pe->name);
1087                         memcpy(pe_data + 1, cp->pe->name, len);
1088                         pe_data[len + 1] = ' ';
1089                         len += 2;
1090                         len += cp->pe->show_pe_data(cp, pe_data + len);
1091                 }
1092                 pe_data[len] = '\0';
1093
1094 #ifdef CONFIG_IP_VS_IPV6
1095                 if (cp->daf == AF_INET6)
1096                         snprintf(dbuf, sizeof(dbuf), "%pI6", &cp->daddr.in6);
1097                 else
1098 #endif
1099                         snprintf(dbuf, sizeof(dbuf), "%08X",
1100                                  ntohl(cp->daddr.ip));
1101
1102 #ifdef CONFIG_IP_VS_IPV6
1103                 if (cp->af == AF_INET6)
1104                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1105                                 "%s %04X %-11s %7lu%s\n",
1106                                 ip_vs_proto_name(cp->protocol),
1107                                 &cp->caddr.in6, ntohs(cp->cport),
1108                                 &cp->vaddr.in6, ntohs(cp->vport),
1109                                 dbuf, ntohs(cp->dport),
1110                                 ip_vs_state_name(cp->protocol, cp->state),
1111                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1112                 else
1113 #endif
1114                         seq_printf(seq,
1115                                 "%-3s %08X %04X %08X %04X"
1116                                 " %s %04X %-11s %7lu%s\n",
1117                                 ip_vs_proto_name(cp->protocol),
1118                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1119                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1120                                 dbuf, ntohs(cp->dport),
1121                                 ip_vs_state_name(cp->protocol, cp->state),
1122                                 (cp->timer.expires-jiffies)/HZ, pe_data);
1123         }
1124         return 0;
1125 }
1126
1127 static const struct seq_operations ip_vs_conn_seq_ops = {
1128         .start = ip_vs_conn_seq_start,
1129         .next  = ip_vs_conn_seq_next,
1130         .stop  = ip_vs_conn_seq_stop,
1131         .show  = ip_vs_conn_seq_show,
1132 };
1133
1134 static const char *ip_vs_origin_name(unsigned int flags)
1135 {
1136         if (flags & IP_VS_CONN_F_SYNC)
1137                 return "SYNC";
1138         else
1139                 return "LOCAL";
1140 }
1141
1142 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1143 {
1144         char dbuf[IP_VS_ADDRSTRLEN];
1145
1146         if (v == SEQ_START_TOKEN)
1147                 seq_puts(seq,
1148    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
1149         else {
1150                 const struct ip_vs_conn *cp = v;
1151                 struct net *net = seq_file_net(seq);
1152
1153                 if (!net_eq(cp->ipvs->net, net))
1154                         return 0;
1155
1156 #ifdef CONFIG_IP_VS_IPV6
1157                 if (cp->daf == AF_INET6)
1158                         snprintf(dbuf, sizeof(dbuf), "%pI6", &cp->daddr.in6);
1159                 else
1160 #endif
1161                         snprintf(dbuf, sizeof(dbuf), "%08X",
1162                                  ntohl(cp->daddr.ip));
1163
1164 #ifdef CONFIG_IP_VS_IPV6
1165                 if (cp->af == AF_INET6)
1166                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1167                                 "%s %04X %-11s %-6s %7lu\n",
1168                                 ip_vs_proto_name(cp->protocol),
1169                                 &cp->caddr.in6, ntohs(cp->cport),
1170                                 &cp->vaddr.in6, ntohs(cp->vport),
1171                                 dbuf, ntohs(cp->dport),
1172                                 ip_vs_state_name(cp->protocol, cp->state),
1173                                 ip_vs_origin_name(cp->flags),
1174                                 (cp->timer.expires-jiffies)/HZ);
1175                 else
1176 #endif
1177                         seq_printf(seq,
1178                                 "%-3s %08X %04X %08X %04X "
1179                                 "%s %04X %-11s %-6s %7lu\n",
1180                                 ip_vs_proto_name(cp->protocol),
1181                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
1182                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1183                                 dbuf, ntohs(cp->dport),
1184                                 ip_vs_state_name(cp->protocol, cp->state),
1185                                 ip_vs_origin_name(cp->flags),
1186                                 (cp->timer.expires-jiffies)/HZ);
1187         }
1188         return 0;
1189 }
1190
1191 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1192         .start = ip_vs_conn_seq_start,
1193         .next  = ip_vs_conn_seq_next,
1194         .stop  = ip_vs_conn_seq_stop,
1195         .show  = ip_vs_conn_sync_seq_show,
1196 };
1197 #endif
1198
1199
1200 /*
1201  *      Randomly drop connection entries before running out of memory
1202  */
1203 static inline int todrop_entry(struct ip_vs_conn *cp)
1204 {
1205         /*
1206          * The drop rate array needs tuning for real environments.
1207          * Called from timer bh only => no locking
1208          */
1209         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1210         static char todrop_counter[9] = {0};
1211         int i;
1212
1213         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1214            This will leave enough time for normal connection to get
1215            through. */
1216         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1217                 return 0;
1218
1219         /* Don't drop the entry if its number of incoming packets is not
1220            located in [0, 8] */
1221         i = atomic_read(&cp->in_pkts);
1222         if (i > 8 || i < 0) return 0;
1223
1224         if (!todrop_rate[i]) return 0;
1225         if (--todrop_counter[i] > 0) return 0;
1226
1227         todrop_counter[i] = todrop_rate[i];
1228         return 1;
1229 }
1230
1231 static inline bool ip_vs_conn_ops_mode(struct ip_vs_conn *cp)
1232 {
1233         struct ip_vs_service *svc;
1234
1235         if (!cp->dest)
1236                 return false;
1237         svc = rcu_dereference(cp->dest->svc);
1238         return svc && (svc->flags & IP_VS_SVC_F_ONEPACKET);
1239 }
1240
1241 /* Called from keventd and must protect itself from softirqs */
1242 void ip_vs_random_dropentry(struct netns_ipvs *ipvs)
1243 {
1244         int idx;
1245         struct ip_vs_conn *cp, *cp_c;
1246
1247         rcu_read_lock();
1248         /*
1249          * Randomly scan 1/32 of the whole table every second
1250          */
1251         for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1252                 unsigned int hash = prandom_u32() & ip_vs_conn_tab_mask;
1253
1254                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
1255                         if (cp->ipvs != ipvs)
1256                                 continue;
1257                         if (cp->flags & IP_VS_CONN_F_TEMPLATE) {
1258                                 if (atomic_read(&cp->n_control) ||
1259                                     !ip_vs_conn_ops_mode(cp))
1260                                         continue;
1261                                 else
1262                                         /* connection template of OPS */
1263                                         goto try_drop;
1264                         }
1265                         if (cp->protocol == IPPROTO_TCP) {
1266                                 switch(cp->state) {
1267                                 case IP_VS_TCP_S_SYN_RECV:
1268                                 case IP_VS_TCP_S_SYNACK:
1269                                         break;
1270
1271                                 case IP_VS_TCP_S_ESTABLISHED:
1272                                         if (todrop_entry(cp))
1273                                                 break;
1274                                         continue;
1275
1276                                 default:
1277                                         continue;
1278                                 }
1279                         } else if (cp->protocol == IPPROTO_SCTP) {
1280                                 switch (cp->state) {
1281                                 case IP_VS_SCTP_S_INIT1:
1282                                 case IP_VS_SCTP_S_INIT:
1283                                         break;
1284                                 case IP_VS_SCTP_S_ESTABLISHED:
1285                                         if (todrop_entry(cp))
1286                                                 break;
1287                                         continue;
1288                                 default:
1289                                         continue;
1290                                 }
1291                         } else {
1292 try_drop:
1293                                 if (!todrop_entry(cp))
1294                                         continue;
1295                         }
1296
1297                         IP_VS_DBG(4, "del connection\n");
1298                         ip_vs_conn_expire_now(cp);
1299                         cp_c = cp->control;
1300                         /* cp->control is valid only with reference to cp */
1301                         if (cp_c && __ip_vs_conn_get(cp)) {
1302                                 IP_VS_DBG(4, "del conn template\n");
1303                                 ip_vs_conn_expire_now(cp_c);
1304                                 __ip_vs_conn_put(cp);
1305                         }
1306                 }
1307                 cond_resched_rcu();
1308         }
1309         rcu_read_unlock();
1310 }
1311
1312
1313 /*
1314  *      Flush all the connection entries in the ip_vs_conn_tab
1315  */
1316 static void ip_vs_conn_flush(struct netns_ipvs *ipvs)
1317 {
1318         int idx;
1319         struct ip_vs_conn *cp, *cp_c;
1320
1321 flush_again:
1322         rcu_read_lock();
1323         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1324
1325                 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1326                         if (cp->ipvs != ipvs)
1327                                 continue;
1328                         IP_VS_DBG(4, "del connection\n");
1329                         ip_vs_conn_expire_now(cp);
1330                         cp_c = cp->control;
1331                         /* cp->control is valid only with reference to cp */
1332                         if (cp_c && __ip_vs_conn_get(cp)) {
1333                                 IP_VS_DBG(4, "del conn template\n");
1334                                 ip_vs_conn_expire_now(cp_c);
1335                                 __ip_vs_conn_put(cp);
1336                         }
1337                 }
1338                 cond_resched_rcu();
1339         }
1340         rcu_read_unlock();
1341
1342         /* the counter may be not NULL, because maybe some conn entries
1343            are run by slow timer handler or unhashed but still referred */
1344         if (atomic_read(&ipvs->conn_count) != 0) {
1345                 schedule();
1346                 goto flush_again;
1347         }
1348 }
1349 /*
1350  * per netns init and exit
1351  */
1352 int __net_init ip_vs_conn_net_init(struct netns_ipvs *ipvs)
1353 {
1354         atomic_set(&ipvs->conn_count, 0);
1355
1356         proc_create_net("ip_vs_conn", 0, ipvs->net->proc_net,
1357                         &ip_vs_conn_seq_ops, sizeof(struct ip_vs_iter_state));
1358         proc_create_net("ip_vs_conn_sync", 0, ipvs->net->proc_net,
1359                         &ip_vs_conn_sync_seq_ops,
1360                         sizeof(struct ip_vs_iter_state));
1361         return 0;
1362 }
1363
1364 void __net_exit ip_vs_conn_net_cleanup(struct netns_ipvs *ipvs)
1365 {
1366         /* flush all the connection entries first */
1367         ip_vs_conn_flush(ipvs);
1368         remove_proc_entry("ip_vs_conn", ipvs->net->proc_net);
1369         remove_proc_entry("ip_vs_conn_sync", ipvs->net->proc_net);
1370 }
1371
1372 int __init ip_vs_conn_init(void)
1373 {
1374         int idx;
1375
1376         /* Compute size and mask */
1377         ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1378         ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1379
1380         /*
1381          * Allocate the connection hash table and initialize its list heads
1382          */
1383         ip_vs_conn_tab = vmalloc(array_size(ip_vs_conn_tab_size,
1384                                             sizeof(*ip_vs_conn_tab)));
1385         if (!ip_vs_conn_tab)
1386                 return -ENOMEM;
1387
1388         /* Allocate ip_vs_conn slab cache */
1389         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1390                                               sizeof(struct ip_vs_conn), 0,
1391                                               SLAB_HWCACHE_ALIGN, NULL);
1392         if (!ip_vs_conn_cachep) {
1393                 vfree(ip_vs_conn_tab);
1394                 return -ENOMEM;
1395         }
1396
1397         pr_info("Connection hash table configured "
1398                 "(size=%d, memory=%ldKbytes)\n",
1399                 ip_vs_conn_tab_size,
1400                 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1401         IP_VS_DBG(0, "Each connection entry needs %zd bytes at least\n",
1402                   sizeof(struct ip_vs_conn));
1403
1404         for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1405                 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1406
1407         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1408                 spin_lock_init(&__ip_vs_conntbl_lock_array[idx].l);
1409         }
1410
1411         /* calculate the random value for connection hash */
1412         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1413
1414         return 0;
1415 }
1416
1417 void ip_vs_conn_cleanup(void)
1418 {
1419         /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1420         rcu_barrier();
1421         /* Release the empty cache */
1422         kmem_cache_destroy(ip_vs_conn_cachep);
1423         vfree(ip_vs_conn_tab);
1424 }