net/netfilter/ipvs/ip_vs_conn.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the Netfilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *              Peter Kese <peter.kese@ijs.si>
  10  *              Julian Anastasov <ja@ssi.bg>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  *
  17  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  18  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  19  * and others. Many code here is taken from IP MASQ code of kernel 2.2.
  20  *
  21  * Changes:
  22  *
  23  */
  24
  25 #define KMSG_COMPONENT "IPVS"
  26 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  27
  28 #include <linux/interrupt.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/kernel.h>
  32 #include <linux/module.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/proc_fs.h>              /* for proc_net_* */
  35 #include <linux/slab.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/jhash.h>
  38 #include <linux/random.h>
  39
  40 #include <net/net_namespace.h>
  41 #include <net/ip_vs.h>
  42
  43
  44 #ifndef CONFIG_IP_VS_TAB_BITS
  45 #define CONFIG_IP_VS_TAB_BITS   12
  46 #endif
  47
  48 /*
  49  * Connection hash size. Default is what was selected at compile time.
  50 */
  51 int ip_vs_conn_tab_bits = CONFIG_IP_VS_TAB_BITS;
  52 module_param_named(conn_tab_bits, ip_vs_conn_tab_bits, int, 0444);
  53 MODULE_PARM_DESC(conn_tab_bits, "Set connections' hash size");
  54
  55 /* size and mask values */
  56 int ip_vs_conn_tab_size;
  57 int ip_vs_conn_tab_mask;
  58
  59 /*
  60  *  Connection hash table: for input and output packets lookups of IPVS
  61  */
  62 static struct list_head *ip_vs_conn_tab;
  63
  64 /*  SLAB cache for IPVS connections */
  65 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
  66
  67 /*  counter for current IPVS connections */
  68 static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
  69
  70 /*  counter for no client port connections */
  71 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
  72
  73 /* random value for IPVS connection hash */
  74 static unsigned int ip_vs_conn_rnd;
  75
  76 /*
  77  *  Fine locking granularity for big connection hash table
  78  */
  79 #define CT_LOCKARRAY_BITS  4
  80 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
  81 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
  82
  83 struct ip_vs_aligned_lock
  84 {
  85         rwlock_t        l;
  86 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
  87
  88 /* lock array for conn table */
  89 static struct ip_vs_aligned_lock
  90 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
  91
  92 static inline void ct_read_lock(unsigned key)
  93 {
  94         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  95 }
  96
  97 static inline void ct_read_unlock(unsigned key)
  98 {
  99         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 100 }
 101
 102 static inline void ct_write_lock(unsigned key)
 103 {
 104         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 105 }
 106
 107 static inline void ct_write_unlock(unsigned key)
 108 {
 109         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 110 }
 111
 112 static inline void ct_read_lock_bh(unsigned key)
 113 {
 114         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 115 }
 116
 117 static inline void ct_read_unlock_bh(unsigned key)
 118 {
 119         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 120 }
 121
 122 static inline void ct_write_lock_bh(unsigned key)
 123 {
 124         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 125 }
 126
 127 static inline void ct_write_unlock_bh(unsigned key)
 128 {
 129         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 130 }
 131
 132
 133 /*
 134  *      Returns hash value for IPVS connection entry
 135  */
 136 static unsigned int ip_vs_conn_hashkey(int af, unsigned proto,
 137                                        const union nf_inet_addr *addr,
 138                                        __be16 port)
 139 {
 140 #ifdef CONFIG_IP_VS_IPV6
 141         if (af == AF_INET6)
 142                 return jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
 143                                     (__force u32)port, proto, ip_vs_conn_rnd)
 144                         & ip_vs_conn_tab_mask;
 145 #endif
 146         return jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
 147                             ip_vs_conn_rnd)
 148                 & ip_vs_conn_tab_mask;
 149 }
 150
 151
 152 /*
 153  *      Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
 154  *      returns bool success.
 155  */
 156 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
 157 {
 158         unsigned hash;
 159         int ret;
 160
 161         /* Hash by protocol, client address and port */
 162         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
 163
 164         ct_write_lock(hash);
 165
 166         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
 167                 list_add(&cp->c_list, &ip_vs_conn_tab[hash]);
 168                 cp->flags |= IP_VS_CONN_F_HASHED;
 169                 atomic_inc(&cp->refcnt);
 170                 ret = 1;
 171         } else {
 172                 pr_err("%s(): request for already hashed, called from %pF\n",
 173                        __func__, __builtin_return_address(0));
 174                 ret = 0;
 175         }
 176
 177         ct_write_unlock(hash);
 178
 179         return ret;
 180 }
 181
 182
 183 /*
 184  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
 185  *      returns bool success.
 186  */
 187 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
 188 {
 189         unsigned hash;
 190         int ret;
 191
 192         /* unhash it and decrease its reference counter */
 193         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
 194
 195         ct_write_lock(hash);
 196
 197         if (cp->flags & IP_VS_CONN_F_HASHED) {
 198                 list_del(&cp->c_list);
 199                 cp->flags &= ~IP_VS_CONN_F_HASHED;
 200                 atomic_dec(&cp->refcnt);
 201                 ret = 1;
 202         } else
 203                 ret = 0;
 204
 205         ct_write_unlock(hash);
 206
 207         return ret;
 208 }
 209
 210
 211 /*
 212  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 213  *  Called for pkts coming from OUTside-to-INside.
 214  *      s_addr, s_port: pkt source address (foreign host)
 215  *      d_addr, d_port: pkt dest address (load balancer)
 216  */
 217 static inline struct ip_vs_conn *__ip_vs_conn_in_get
 218 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 219  const union nf_inet_addr *d_addr, __be16 d_port)
 220 {
 221         unsigned hash;
 222         struct ip_vs_conn *cp;
 223
 224         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
 225
 226         ct_read_lock(hash);
 227
 228         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 229                 if (cp->af == af &&
 230                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
 231                     ip_vs_addr_equal(af, d_addr, &cp->vaddr) &&
 232                     s_port == cp->cport && d_port == cp->vport &&
 233                     ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
 234                     protocol == cp->protocol) {
 235                         /* HIT */
 236                         atomic_inc(&cp->refcnt);
 237                         ct_read_unlock(hash);
 238                         return cp;
 239                 }
 240         }
 241
 242         ct_read_unlock(hash);
 243
 244         return NULL;
 245 }
 246
 247 struct ip_vs_conn *ip_vs_conn_in_get
 248 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 249  const union nf_inet_addr *d_addr, __be16 d_port)
 250 {
 251         struct ip_vs_conn *cp;
 252
 253         cp = __ip_vs_conn_in_get(af, protocol, s_addr, s_port, d_addr, d_port);
 254         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt))
 255                 cp = __ip_vs_conn_in_get(af, protocol, s_addr, 0, d_addr,
 256                                          d_port);
 257
 258         IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
 259                       ip_vs_proto_name(protocol),
 260                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
 261                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
 262                       cp ? "hit" : "not hit");
 263
 264         return cp;
 265 }
 266
 267 /* Get reference to connection template */
 268 struct ip_vs_conn *ip_vs_ct_in_get
 269 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 270  const union nf_inet_addr *d_addr, __be16 d_port)
 271 {
 272         unsigned hash;
 273         struct ip_vs_conn *cp;
 274
 275         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
 276
 277         ct_read_lock(hash);
 278
 279         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 280                 if (cp->af == af &&
 281                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
 282                     /* protocol should only be IPPROTO_IP if
 283                      * d_addr is a fwmark */
 284                     ip_vs_addr_equal(protocol == IPPROTO_IP ? AF_UNSPEC : af,
 285                                      d_addr, &cp->vaddr) &&
 286                     s_port == cp->cport && d_port == cp->vport &&
 287                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
 288                     protocol == cp->protocol) {
 289                         /* HIT */
 290                         atomic_inc(&cp->refcnt);
 291                         goto out;
 292                 }
 293         }
 294         cp = NULL;
 295
 296   out:
 297         ct_read_unlock(hash);
 298
 299         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
 300                       ip_vs_proto_name(protocol),
 301                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
 302                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
 303                       cp ? "hit" : "not hit");
 304
 305         return cp;
 306 }
 307
 308 /*
 309  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 310  *  Called for pkts coming from inside-to-OUTside.
 311  *      s_addr, s_port: pkt source address (inside host)
 312  *      d_addr, d_port: pkt dest address (foreign host)
 313  */
 314 struct ip_vs_conn *ip_vs_conn_out_get
 315 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 316  const union nf_inet_addr *d_addr, __be16 d_port)
 317 {
 318         unsigned hash;
 319         struct ip_vs_conn *cp, *ret=NULL;
 320
 321         /*
 322          *      Check for "full" addressed entries
 323          */
 324         hash = ip_vs_conn_hashkey(af, protocol, d_addr, d_port);
 325
 326         ct_read_lock(hash);
 327
 328         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 329                 if (cp->af == af &&
 330                     ip_vs_addr_equal(af, d_addr, &cp->caddr) &&
 331                     ip_vs_addr_equal(af, s_addr, &cp->daddr) &&
 332                     d_port == cp->cport && s_port == cp->dport &&
 333                     protocol == cp->protocol) {
 334                         /* HIT */
 335                         atomic_inc(&cp->refcnt);
 336                         ret = cp;
 337                         break;
 338                 }
 339         }
 340
 341         ct_read_unlock(hash);
 342
 343         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
 344                       ip_vs_proto_name(protocol),
 345                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
 346                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
 347                       ret ? "hit" : "not hit");
 348
 349         return ret;
 350 }
 351
 352
 353 /*
 354  *      Put back the conn and restart its timer with its timeout
 355  */
 356 void ip_vs_conn_put(struct ip_vs_conn *cp)
 357 {
 358         /* reset it expire in its timeout */
 359         mod_timer(&cp->timer, jiffies+cp->timeout);
 360
 361         __ip_vs_conn_put(cp);
 362 }
 363
 364
 365 /*
 366  *      Fill a no_client_port connection with a client port number
 367  */
 368 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
 369 {
 370         if (ip_vs_conn_unhash(cp)) {
 371                 spin_lock(&cp->lock);
 372                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
 373                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 374                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
 375                         cp->cport = cport;
 376                 }
 377                 spin_unlock(&cp->lock);
 378
 379                 /* hash on new dport */
 380                 ip_vs_conn_hash(cp);
 381         }
 382 }
 383
 384
 385 /*
 386  *      Bind a connection entry with the corresponding packet_xmit.
 387  *      Called by ip_vs_conn_new.
 388  */
 389 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
 390 {
 391         switch (IP_VS_FWD_METHOD(cp)) {
 392         case IP_VS_CONN_F_MASQ:
 393                 cp->packet_xmit = ip_vs_nat_xmit;
 394                 break;
 395
 396         case IP_VS_CONN_F_TUNNEL:
 397                 cp->packet_xmit = ip_vs_tunnel_xmit;
 398                 break;
 399
 400         case IP_VS_CONN_F_DROUTE:
 401                 cp->packet_xmit = ip_vs_dr_xmit;
 402                 break;
 403
 404         case IP_VS_CONN_F_LOCALNODE:
 405                 cp->packet_xmit = ip_vs_null_xmit;
 406                 break;
 407
 408         case IP_VS_CONN_F_BYPASS:
 409                 cp->packet_xmit = ip_vs_bypass_xmit;
 410                 break;
 411         }
 412 }
 413
 414 #ifdef CONFIG_IP_VS_IPV6
 415 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
 416 {
 417         switch (IP_VS_FWD_METHOD(cp)) {
 418         case IP_VS_CONN_F_MASQ:
 419                 cp->packet_xmit = ip_vs_nat_xmit_v6;
 420                 break;
 421
 422         case IP_VS_CONN_F_TUNNEL:
 423                 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 424                 break;
 425
 426         case IP_VS_CONN_F_DROUTE:
 427                 cp->packet_xmit = ip_vs_dr_xmit_v6;
 428                 break;
 429
 430         case IP_VS_CONN_F_LOCALNODE:
 431                 cp->packet_xmit = ip_vs_null_xmit;
 432                 break;
 433
 434         case IP_VS_CONN_F_BYPASS:
 435                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
 436                 break;
 437         }
 438 }
 439 #endif
 440
 441
 442 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
 443 {
 444         return atomic_read(&dest->activeconns)
 445                 + atomic_read(&dest->inactconns);
 446 }
 447
 448 /*
 449  *      Bind a connection entry with a virtual service destination
 450  *      Called just after a new connection entry is created.
 451  */
 452 static inline void
 453 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
 454 {
 455         /* if dest is NULL, then return directly */
 456         if (!dest)
 457                 return;
 458
 459         /* Increase the refcnt counter of the dest */
 460         atomic_inc(&dest->refcnt);
 461
 462         /* Bind with the destination and its corresponding transmitter */
 463         if ((cp->flags & IP_VS_CONN_F_SYNC) &&
 464             (!(cp->flags & IP_VS_CONN_F_TEMPLATE)))
 465                 /* if the connection is not template and is created
 466                  * by sync, preserve the activity flag.
 467                  */
 468                 cp->flags |= atomic_read(&dest->conn_flags) &
 469                              (~IP_VS_CONN_F_INACTIVE);
 470         else
 471                 cp->flags |= atomic_read(&dest->conn_flags);
 472         cp->dest = dest;
 473
 474         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
 475                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 476                       "dest->refcnt:%d\n",
 477                       ip_vs_proto_name(cp->protocol),
 478                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 479                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 480                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
 481                       ip_vs_fwd_tag(cp), cp->state,
 482                       cp->flags, atomic_read(&cp->refcnt),
 483                       atomic_read(&dest->refcnt));
 484
 485         /* Update the connection counters */
 486         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 487                 /* It is a normal connection, so increase the inactive
 488                    connection counter because it is in TCP SYNRECV
 489                    state (inactive) or other protocol inacive state */
 490                 if ((cp->flags & IP_VS_CONN_F_SYNC) &&
 491                     (!(cp->flags & IP_VS_CONN_F_INACTIVE)))
 492                         atomic_inc(&dest->activeconns);
 493                 else
 494                         atomic_inc(&dest->inactconns);
 495         } else {
 496                 /* It is a persistent connection/template, so increase
 497                    the peristent connection counter */
 498                 atomic_inc(&dest->persistconns);
 499         }
 500
 501         if (dest->u_threshold != 0 &&
 502             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
 503                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
 504 }
 505
 506
 507 /*
 508  * Check if there is a destination for the connection, if so
 509  * bind the connection to the destination.
 510  */
 511 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
 512 {
 513         struct ip_vs_dest *dest;
 514
 515         if ((cp) && (!cp->dest)) {
 516                 dest = ip_vs_find_dest(cp->af, &cp->daddr, cp->dport,
 517                                        &cp->vaddr, cp->vport,
 518                                        cp->protocol);
 519                 ip_vs_bind_dest(cp, dest);
 520                 return dest;
 521         } else
 522                 return NULL;
 523 }
 524
 525
 526 /*
 527  *      Unbind a connection entry with its VS destination
 528  *      Called by the ip_vs_conn_expire function.
 529  */
 530 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
 531 {
 532         struct ip_vs_dest *dest = cp->dest;
 533
 534         if (!dest)
 535                 return;
 536
 537         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
 538                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 539                       "dest->refcnt:%d\n",
 540                       ip_vs_proto_name(cp->protocol),
 541                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 542                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 543                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
 544                       ip_vs_fwd_tag(cp), cp->state,
 545                       cp->flags, atomic_read(&cp->refcnt),
 546                       atomic_read(&dest->refcnt));
 547
 548         /* Update the connection counters */
 549         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 550                 /* It is a normal connection, so decrease the inactconns
 551                    or activeconns counter */
 552                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
 553                         atomic_dec(&dest->inactconns);
 554                 } else {
 555                         atomic_dec(&dest->activeconns);
 556                 }
 557         } else {
 558                 /* It is a persistent connection/template, so decrease
 559                    the peristent connection counter */
 560                 atomic_dec(&dest->persistconns);
 561         }
 562
 563         if (dest->l_threshold != 0) {
 564                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
 565                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 566         } else if (dest->u_threshold != 0) {
 567                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
 568                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 569         } else {
 570                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 571                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 572         }
 573
 574         /*
 575          * Simply decrease the refcnt of the dest, because the
 576          * dest will be either in service's destination list
 577          * or in the trash.
 578          */
 579         atomic_dec(&dest->refcnt);
 580 }
 581
 582
 583 /*
 584  *      Checking if the destination of a connection template is available.
 585  *      If available, return 1, otherwise invalidate this connection
 586  *      template and return 0.
 587  */
 588 int ip_vs_check_template(struct ip_vs_conn *ct)
 589 {
 590         struct ip_vs_dest *dest = ct->dest;
 591
 592         /*
 593          * Checking the dest server status.
 594          */
 595         if ((dest == NULL) ||
 596             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
 597             (sysctl_ip_vs_expire_quiescent_template &&
 598              (atomic_read(&dest->weight) == 0))) {
 599                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
 600                               "protocol %s s:%s:%d v:%s:%d "
 601                               "-> d:%s:%d\n",
 602                               ip_vs_proto_name(ct->protocol),
 603                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
 604                               ntohs(ct->cport),
 605                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
 606                               ntohs(ct->vport),
 607                               IP_VS_DBG_ADDR(ct->af, &ct->daddr),
 608                               ntohs(ct->dport));
 609
 610                 /*
 611                  * Invalidate the connection template
 612                  */
 613                 if (ct->vport != htons(0xffff)) {
 614                         if (ip_vs_conn_unhash(ct)) {
 615                                 ct->dport = htons(0xffff);
 616                                 ct->vport = htons(0xffff);
 617                                 ct->cport = 0;
 618                                 ip_vs_conn_hash(ct);
 619                         }
 620                 }
 621
 622                 /*
 623                  * Simply decrease the refcnt of the template,
 624                  * don't restart its timer.
 625                  */
 626                 atomic_dec(&ct->refcnt);
 627                 return 0;
 628         }
 629         return 1;
 630 }
 631
 632 static void ip_vs_conn_expire(unsigned long data)
 633 {
 634         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
 635
 636         cp->timeout = 60*HZ;
 637
 638         /*
 639          *      hey, I'm using it
 640          */
 641         atomic_inc(&cp->refcnt);
 642
 643         /*
 644          *      do I control anybody?
 645          */
 646         if (atomic_read(&cp->n_control))
 647                 goto expire_later;
 648
 649         /*
 650          *      unhash it if it is hashed in the conn table
 651          */
 652         if (!ip_vs_conn_unhash(cp))
 653                 goto expire_later;
 654
 655         /*
 656          *      refcnt==1 implies I'm the only one referrer
 657          */
 658         if (likely(atomic_read(&cp->refcnt) == 1)) {
 659                 /* delete the timer if it is activated by other users */
 660                 if (timer_pending(&cp->timer))
 661                         del_timer(&cp->timer);
 662
 663                 /* does anybody control me? */
 664                 if (cp->control)
 665                         ip_vs_control_del(cp);
 666
 667                 if (unlikely(cp->app != NULL))
 668                         ip_vs_unbind_app(cp);
 669                 ip_vs_unbind_dest(cp);
 670                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
 671                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 672                 atomic_dec(&ip_vs_conn_count);
 673
 674                 kmem_cache_free(ip_vs_conn_cachep, cp);
 675                 return;
 676         }
 677
 678         /* hash it back to the table */
 679         ip_vs_conn_hash(cp);
 680
 681   expire_later:
 682         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
 683                   atomic_read(&cp->refcnt)-1,
 684                   atomic_read(&cp->n_control));
 685
 686         ip_vs_conn_put(cp);
 687 }
 688
 689
 690 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
 691 {
 692         if (del_timer(&cp->timer))
 693                 mod_timer(&cp->timer, jiffies);
 694 }
 695
 696
 697 /*
 698  *      Create a new connection entry and hash it into the ip_vs_conn_tab
 699  */
 700 struct ip_vs_conn *
 701 ip_vs_conn_new(int af, int proto, const union nf_inet_addr *caddr, __be16 cport,
 702                const union nf_inet_addr *vaddr, __be16 vport,
 703                const union nf_inet_addr *daddr, __be16 dport, unsigned flags,
 704                struct ip_vs_dest *dest)
 705 {
 706         struct ip_vs_conn *cp;
 707         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
 708
 709         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
 710         if (cp == NULL) {
 711                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
 712                 return NULL;
 713         }
 714
 715         INIT_LIST_HEAD(&cp->c_list);
 716         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
 717         cp->af             = af;
 718         cp->protocol       = proto;
 719         ip_vs_addr_copy(af, &cp->caddr, caddr);
 720         cp->cport          = cport;
 721         ip_vs_addr_copy(af, &cp->vaddr, vaddr);
 722         cp->vport          = vport;
 723         /* proto should only be IPPROTO_IP if d_addr is a fwmark */
 724         ip_vs_addr_copy(proto == IPPROTO_IP ? AF_UNSPEC : af,
 725                         &cp->daddr, daddr);
 726         cp->dport          = dport;
 727         cp->flags          = flags;
 728         spin_lock_init(&cp->lock);
 729
 730         /*
 731          * Set the entry is referenced by the current thread before hashing
 732          * it in the table, so that other thread run ip_vs_random_dropentry
 733          * but cannot drop this entry.
 734          */
 735         atomic_set(&cp->refcnt, 1);
 736
 737         atomic_set(&cp->n_control, 0);
 738         atomic_set(&cp->in_pkts, 0);
 739
 740         atomic_inc(&ip_vs_conn_count);
 741         if (flags & IP_VS_CONN_F_NO_CPORT)
 742                 atomic_inc(&ip_vs_conn_no_cport_cnt);
 743
 744         /* Bind the connection with a destination server */
 745         ip_vs_bind_dest(cp, dest);
 746
 747         /* Set its state and timeout */
 748         cp->state = 0;
 749         cp->timeout = 3*HZ;
 750
 751         /* Bind its packet transmitter */
 752 #ifdef CONFIG_IP_VS_IPV6
 753         if (af == AF_INET6)
 754                 ip_vs_bind_xmit_v6(cp);
 755         else
 756 #endif
 757                 ip_vs_bind_xmit(cp);
 758
 759         if (unlikely(pp && atomic_read(&pp->appcnt)))
 760                 ip_vs_bind_app(cp, pp);
 761
 762         /* Hash it in the ip_vs_conn_tab finally */
 763         ip_vs_conn_hash(cp);
 764
 765         return cp;
 766 }
 767
 768
 769 /*
 770  *      /proc/net/ip_vs_conn entries
 771  */
 772 #ifdef CONFIG_PROC_FS
 773
 774 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
 775 {
 776         int idx;
 777         struct ip_vs_conn *cp;
 778
 779         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
 780                 ct_read_lock_bh(idx);
 781                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
 782                         if (pos-- == 0) {
 783                                 seq->private = &ip_vs_conn_tab[idx];
 784                                 return cp;
 785                         }
 786                 }
 787                 ct_read_unlock_bh(idx);
 788         }
 789
 790         return NULL;
 791 }
 792
 793 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
 794 {
 795         seq->private = NULL;
 796         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
 797 }
 798
 799 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 800 {
 801         struct ip_vs_conn *cp = v;
 802         struct list_head *e, *l = seq->private;
 803         int idx;
 804
 805         ++*pos;
 806         if (v == SEQ_START_TOKEN)
 807                 return ip_vs_conn_array(seq, 0);
 808
 809         /* more on same hash chain? */
 810         if ((e = cp->c_list.next) != l)
 811                 return list_entry(e, struct ip_vs_conn, c_list);
 812
 813         idx = l - ip_vs_conn_tab;
 814         ct_read_unlock_bh(idx);
 815
 816         while (++idx < ip_vs_conn_tab_size) {
 817                 ct_read_lock_bh(idx);
 818                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
 819                         seq->private = &ip_vs_conn_tab[idx];
 820                         return cp;
 821                 }
 822                 ct_read_unlock_bh(idx);
 823         }
 824         seq->private = NULL;
 825         return NULL;
 826 }
 827
 828 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
 829 {
 830         struct list_head *l = seq->private;
 831
 832         if (l)
 833                 ct_read_unlock_bh(l - ip_vs_conn_tab);
 834 }
 835
 836 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
 837 {
 838
 839         if (v == SEQ_START_TOKEN)
 840                 seq_puts(seq,
 841    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
 842         else {
 843                 const struct ip_vs_conn *cp = v;
 844
 845 #ifdef CONFIG_IP_VS_IPV6
 846                 if (cp->af == AF_INET6)
 847                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %7lu\n",
 848                                 ip_vs_proto_name(cp->protocol),
 849                                 &cp->caddr.in6, ntohs(cp->cport),
 850                                 &cp->vaddr.in6, ntohs(cp->vport),
 851                                 &cp->daddr.in6, ntohs(cp->dport),
 852                                 ip_vs_state_name(cp->protocol, cp->state),
 853                                 (cp->timer.expires-jiffies)/HZ);
 854                 else
 855 #endif
 856                         seq_printf(seq,
 857                                 "%-3s %08X %04X %08X %04X"
 858                                 " %08X %04X %-11s %7lu\n",
 859                                 ip_vs_proto_name(cp->protocol),
 860                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
 861                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
 862                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
 863                                 ip_vs_state_name(cp->protocol, cp->state),
 864                                 (cp->timer.expires-jiffies)/HZ);
 865         }
 866         return 0;
 867 }
 868
 869 static const struct seq_operations ip_vs_conn_seq_ops = {
 870         .start = ip_vs_conn_seq_start,
 871         .next  = ip_vs_conn_seq_next,
 872         .stop  = ip_vs_conn_seq_stop,
 873         .show  = ip_vs_conn_seq_show,
 874 };
 875
 876 static int ip_vs_conn_open(struct inode *inode, struct file *file)
 877 {
 878         return seq_open(file, &ip_vs_conn_seq_ops);
 879 }
 880
 881 static const struct file_operations ip_vs_conn_fops = {
 882         .owner   = THIS_MODULE,
 883         .open    = ip_vs_conn_open,
 884         .read    = seq_read,
 885         .llseek  = seq_lseek,
 886         .release = seq_release,
 887 };
 888
 889 static const char *ip_vs_origin_name(unsigned flags)
 890 {
 891         if (flags & IP_VS_CONN_F_SYNC)
 892                 return "SYNC";
 893         else
 894                 return "LOCAL";
 895 }
 896
 897 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
 898 {
 899
 900         if (v == SEQ_START_TOKEN)
 901                 seq_puts(seq,
 902    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
 903         else {
 904                 const struct ip_vs_conn *cp = v;
 905
 906 #ifdef CONFIG_IP_VS_IPV6
 907                 if (cp->af == AF_INET6)
 908                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
 909                                 ip_vs_proto_name(cp->protocol),
 910                                 &cp->caddr.in6, ntohs(cp->cport),
 911                                 &cp->vaddr.in6, ntohs(cp->vport),
 912                                 &cp->daddr.in6, ntohs(cp->dport),
 913                                 ip_vs_state_name(cp->protocol, cp->state),
 914                                 ip_vs_origin_name(cp->flags),
 915                                 (cp->timer.expires-jiffies)/HZ);
 916                 else
 917 #endif
 918                         seq_printf(seq,
 919                                 "%-3s %08X %04X %08X %04X "
 920                                 "%08X %04X %-11s %-6s %7lu\n",
 921                                 ip_vs_proto_name(cp->protocol),
 922                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
 923                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
 924                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
 925                                 ip_vs_state_name(cp->protocol, cp->state),
 926                                 ip_vs_origin_name(cp->flags),
 927                                 (cp->timer.expires-jiffies)/HZ);
 928         }
 929         return 0;
 930 }
 931
 932 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
 933         .start = ip_vs_conn_seq_start,
 934         .next  = ip_vs_conn_seq_next,
 935         .stop  = ip_vs_conn_seq_stop,
 936         .show  = ip_vs_conn_sync_seq_show,
 937 };
 938
 939 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
 940 {
 941         return seq_open(file, &ip_vs_conn_sync_seq_ops);
 942 }
 943
 944 static const struct file_operations ip_vs_conn_sync_fops = {
 945         .owner   = THIS_MODULE,
 946         .open    = ip_vs_conn_sync_open,
 947         .read    = seq_read,
 948         .llseek  = seq_lseek,
 949         .release = seq_release,
 950 };
 951
 952 #endif
 953
 954
 955 /*
 956  *      Randomly drop connection entries before running out of memory
 957  */
 958 static inline int todrop_entry(struct ip_vs_conn *cp)
 959 {
 960         /*
 961          * The drop rate array needs tuning for real environments.
 962          * Called from timer bh only => no locking
 963          */
 964         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
 965         static char todrop_counter[9] = {0};
 966         int i;
 967
 968         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
 969            This will leave enough time for normal connection to get
 970            through. */
 971         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
 972                 return 0;
 973
 974         /* Don't drop the entry if its number of incoming packets is not
 975            located in [0, 8] */
 976         i = atomic_read(&cp->in_pkts);
 977         if (i > 8 || i < 0) return 0;
 978
 979         if (!todrop_rate[i]) return 0;
 980         if (--todrop_counter[i] > 0) return 0;
 981
 982         todrop_counter[i] = todrop_rate[i];
 983         return 1;
 984 }
 985
 986 /* Called from keventd and must protect itself from softirqs */
 987 void ip_vs_random_dropentry(void)
 988 {
 989         int idx;
 990         struct ip_vs_conn *cp;
 991
 992         /*
 993          * Randomly scan 1/32 of the whole table every second
 994          */
 995         for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
 996                 unsigned hash = net_random() & ip_vs_conn_tab_mask;
 997
 998                 /*
 999                  *  Lock is actually needed in this loop.
1000                  */
1001                 ct_write_lock_bh(hash);
1002
1003                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
1004                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1005                                 /* connection template */
1006                                 continue;
1007
1008                         if (cp->protocol == IPPROTO_TCP) {
1009                                 switch(cp->state) {
1010                                 case IP_VS_TCP_S_SYN_RECV:
1011                                 case IP_VS_TCP_S_SYNACK:
1012                                         break;
1013
1014                                 case IP_VS_TCP_S_ESTABLISHED:
1015                                         if (todrop_entry(cp))
1016                                                 break;
1017                                         continue;
1018
1019                                 default:
1020                                         continue;
1021                                 }
1022                         } else {
1023                                 if (!todrop_entry(cp))
1024                                         continue;
1025                         }
1026
1027                         IP_VS_DBG(4, "del connection\n");
1028                         ip_vs_conn_expire_now(cp);
1029                         if (cp->control) {
1030                                 IP_VS_DBG(4, "del conn template\n");
1031                                 ip_vs_conn_expire_now(cp->control);
1032                         }
1033                 }
1034                 ct_write_unlock_bh(hash);
1035         }
1036 }
1037
1038
1039 /*
1040  *      Flush all the connection entries in the ip_vs_conn_tab
1041  */
1042 static void ip_vs_conn_flush(void)
1043 {
1044         int idx;
1045         struct ip_vs_conn *cp;
1046
1047   flush_again:
1048         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1049                 /*
1050                  *  Lock is actually needed in this loop.
1051                  */
1052                 ct_write_lock_bh(idx);
1053
1054                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
1055
1056                         IP_VS_DBG(4, "del connection\n");
1057                         ip_vs_conn_expire_now(cp);
1058                         if (cp->control) {
1059                                 IP_VS_DBG(4, "del conn template\n");
1060                                 ip_vs_conn_expire_now(cp->control);
1061                         }
1062                 }
1063                 ct_write_unlock_bh(idx);
1064         }
1065
1066         /* the counter may be not NULL, because maybe some conn entries
1067            are run by slow timer handler or unhashed but still referred */
1068         if (atomic_read(&ip_vs_conn_count) != 0) {
1069                 schedule();
1070                 goto flush_again;
1071         }
1072 }
1073
1074
1075 int __init ip_vs_conn_init(void)
1076 {
1077         int idx;
1078
1079         /* Compute size and mask */
1080         ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1081         ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1082
1083         /*
1084          * Allocate the connection hash table and initialize its list heads
1085          */
1086         ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size *
1087                                  sizeof(struct list_head));
1088         if (!ip_vs_conn_tab)
1089                 return -ENOMEM;
1090
1091         /* Allocate ip_vs_conn slab cache */
1092         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1093                                               sizeof(struct ip_vs_conn), 0,
1094                                               SLAB_HWCACHE_ALIGN, NULL);
1095         if (!ip_vs_conn_cachep) {
1096                 vfree(ip_vs_conn_tab);
1097                 return -ENOMEM;
1098         }
1099
1100         pr_info("Connection hash table configured "
1101                 "(size=%d, memory=%ldKbytes)\n",
1102                 ip_vs_conn_tab_size,
1103                 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1104         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1105                   sizeof(struct ip_vs_conn));
1106
1107         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1108                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
1109         }
1110
1111         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1112                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1113         }
1114
1115         proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops);
1116         proc_net_fops_create(&init_net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops);
1117
1118         /* calculate the random value for connection hash */
1119         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1120
1121         return 0;
1122 }
1123
1124
1125 void ip_vs_conn_cleanup(void)
1126 {
1127         /* flush all the connection entries first */
1128         ip_vs_conn_flush();
1129
1130         /* Release the empty cache */
1131         kmem_cache_destroy(ip_vs_conn_cachep);
1132         proc_net_remove(&init_net, "ip_vs_conn");
1133         proc_net_remove(&init_net, "ip_vs_conn_sync");
1134         vfree(ip_vs_conn_tab);
1135 }