net/netfilter/nf_conntrack_core.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Connection state tracking for netfilter.  This is separated from,
   3    but required by, the NAT layer; it can also be used by an iptables
   4    extension. */
   5
   6 /* (C) 1999-2001 Paul `Rusty' Russell
   7  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   8  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   9  * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
  10  */
  11
  12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13
  14 #include <linux/types.h>
  15 #include <linux/netfilter.h>
  16 #include <linux/module.h>
  17 #include <linux/sched.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/proc_fs.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/stddef.h>
  22 #include <linux/slab.h>
  23 #include <linux/random.h>
  24 #include <linux/jhash.h>
  25 #include <linux/siphash.h>
  26 #include <linux/err.h>
  27 #include <linux/percpu.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/notifier.h>
  30 #include <linux/kernel.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/socket.h>
  33 #include <linux/mm.h>
  34 #include <linux/nsproxy.h>
  35 #include <linux/rculist_nulls.h>
  36
  37 #include <net/netfilter/nf_conntrack.h>
  38 #include <net/netfilter/nf_conntrack_l4proto.h>
  39 #include <net/netfilter/nf_conntrack_expect.h>
  40 #include <net/netfilter/nf_conntrack_helper.h>
  41 #include <net/netfilter/nf_conntrack_seqadj.h>
  42 #include <net/netfilter/nf_conntrack_core.h>
  43 #include <net/netfilter/nf_conntrack_extend.h>
  44 #include <net/netfilter/nf_conntrack_acct.h>
  45 #include <net/netfilter/nf_conntrack_ecache.h>
  46 #include <net/netfilter/nf_conntrack_zones.h>
  47 #include <net/netfilter/nf_conntrack_timestamp.h>
  48 #include <net/netfilter/nf_conntrack_timeout.h>
  49 #include <net/netfilter/nf_conntrack_labels.h>
  50 #include <net/netfilter/nf_conntrack_synproxy.h>
  51 #include <net/netfilter/nf_nat.h>
  52 #include <net/netfilter/nf_nat_helper.h>
  53 #include <net/netns/hash.h>
  54 #include <net/ip.h>
  55
  56 #include "nf_internals.h"
  57
  58 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
  59 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
  60
  61 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
  62 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
  63
  64 struct hlist_nulls_head *nf_conntrack_hash __read_mostly;
  65 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
  66
  67 struct conntrack_gc_work {
  68         struct delayed_work     dwork;
  69         u32                     last_bucket;
  70         bool                    exiting;
  71         bool                    early_drop;
  72         long                    next_gc_run;
  73 };
  74
  75 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
  76 static DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
  77 static __read_mostly bool nf_conntrack_locks_all;
  78
  79 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
  80 #define GC_MAX_BUCKETS_DIV      128u
  81 /* upper bound of full table scan */
  82 #define GC_MAX_SCAN_JIFFIES     (16u * HZ)
  83 /* desired ratio of entries found to be expired */
  84 #define GC_EVICT_RATIO  50u
  85
  86 static struct conntrack_gc_work conntrack_gc_work;
  87
  88 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
  89 {
  90         /* 1) Acquire the lock */
  91         spin_lock(lock);
  92
  93         /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
  94          * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
  95          */
  96         if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false))
  97                 return;
  98
  99         /* fast path failed, unlock */
 100         spin_unlock(lock);
 101
 102         /* Slow path 1) get global lock */
 103         spin_lock(&nf_conntrack_locks_all_lock);
 104
 105         /* Slow path 2) get the lock we want */
 106         spin_lock(lock);
 107
 108         /* Slow path 3) release the global lock */
 109         spin_unlock(&nf_conntrack_locks_all_lock);
 110 }
 111 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
 112
 113 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
 114 {
 115         h1 %= CONNTRACK_LOCKS;
 116         h2 %= CONNTRACK_LOCKS;
 117         spin_unlock(&nf_conntrack_locks[h1]);
 118         if (h1 != h2)
 119                 spin_unlock(&nf_conntrack_locks[h2]);
 120 }
 121
 122 /* return true if we need to recompute hashes (in case hash table was resized) */
 123 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
 124                                      unsigned int h2, unsigned int sequence)
 125 {
 126         h1 %= CONNTRACK_LOCKS;
 127         h2 %= CONNTRACK_LOCKS;
 128         if (h1 <= h2) {
 129                 nf_conntrack_lock(&nf_conntrack_locks[h1]);
 130                 if (h1 != h2)
 131                         spin_lock_nested(&nf_conntrack_locks[h2],
 132                                          SINGLE_DEPTH_NESTING);
 133         } else {
 134                 nf_conntrack_lock(&nf_conntrack_locks[h2]);
 135                 spin_lock_nested(&nf_conntrack_locks[h1],
 136                                  SINGLE_DEPTH_NESTING);
 137         }
 138         if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
 139                 nf_conntrack_double_unlock(h1, h2);
 140                 return true;
 141         }
 142         return false;
 143 }
 144
 145 static void nf_conntrack_all_lock(void)
 146 {
 147         int i;
 148
 149         spin_lock(&nf_conntrack_locks_all_lock);
 150
 151         nf_conntrack_locks_all = true;
 152
 153         for (i = 0; i < CONNTRACK_LOCKS; i++) {
 154                 spin_lock(&nf_conntrack_locks[i]);
 155
 156                 /* This spin_unlock provides the "release" to ensure that
 157                  * nf_conntrack_locks_all==true is visible to everyone that
 158                  * acquired spin_lock(&nf_conntrack_locks[]).
 159                  */
 160                 spin_unlock(&nf_conntrack_locks[i]);
 161         }
 162 }
 163
 164 static void nf_conntrack_all_unlock(void)
 165 {
 166         /* All prior stores must be complete before we clear
 167          * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
 168          * might observe the false value but not the entire
 169          * critical section.
 170          * It pairs with the smp_load_acquire() in nf_conntrack_lock()
 171          */
 172         smp_store_release(&nf_conntrack_locks_all, false);
 173         spin_unlock(&nf_conntrack_locks_all_lock);
 174 }
 175
 176 unsigned int nf_conntrack_htable_size __read_mostly;
 177 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
 178
 179 unsigned int nf_conntrack_max __read_mostly;
 180 EXPORT_SYMBOL_GPL(nf_conntrack_max);
 181 seqcount_t nf_conntrack_generation __read_mostly;
 182 static unsigned int nf_conntrack_hash_rnd __read_mostly;
 183
 184 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
 185                               const struct net *net)
 186 {
 187         unsigned int n;
 188         u32 seed;
 189
 190         get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
 191
 192         /* The direction must be ignored, so we hash everything up to the
 193          * destination ports (which is a multiple of 4) and treat the last
 194          * three bytes manually.
 195          */
 196         seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
 197         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
 198         return jhash2((u32 *)tuple, n, seed ^
 199                       (((__force __u16)tuple->dst.u.all << 16) |
 200                       tuple->dst.protonum));
 201 }
 202
 203 static u32 scale_hash(u32 hash)
 204 {
 205         return reciprocal_scale(hash, nf_conntrack_htable_size);
 206 }
 207
 208 static u32 __hash_conntrack(const struct net *net,
 209                             const struct nf_conntrack_tuple *tuple,
 210                             unsigned int size)
 211 {
 212         return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
 213 }
 214
 215 static u32 hash_conntrack(const struct net *net,
 216                           const struct nf_conntrack_tuple *tuple)
 217 {
 218         return scale_hash(hash_conntrack_raw(tuple, net));
 219 }
 220
 221 static bool nf_ct_get_tuple_ports(const struct sk_buff *skb,
 222                                   unsigned int dataoff,
 223                                   struct nf_conntrack_tuple *tuple)
 224 {       struct {
 225                 __be16 sport;
 226                 __be16 dport;
 227         } _inet_hdr, *inet_hdr;
 228
 229         /* Actually only need first 4 bytes to get ports. */
 230         inet_hdr = skb_header_pointer(skb, dataoff, sizeof(_inet_hdr), &_inet_hdr);
 231         if (!inet_hdr)
 232                 return false;
 233
 234         tuple->src.u.udp.port = inet_hdr->sport;
 235         tuple->dst.u.udp.port = inet_hdr->dport;
 236         return true;
 237 }
 238
 239 static bool
 240 nf_ct_get_tuple(const struct sk_buff *skb,
 241                 unsigned int nhoff,
 242                 unsigned int dataoff,
 243                 u_int16_t l3num,
 244                 u_int8_t protonum,
 245                 struct net *net,
 246                 struct nf_conntrack_tuple *tuple)
 247 {
 248         unsigned int size;
 249         const __be32 *ap;
 250         __be32 _addrs[8];
 251
 252         memset(tuple, 0, sizeof(*tuple));
 253
 254         tuple->src.l3num = l3num;
 255         switch (l3num) {
 256         case NFPROTO_IPV4:
 257                 nhoff += offsetof(struct iphdr, saddr);
 258                 size = 2 * sizeof(__be32);
 259                 break;
 260         case NFPROTO_IPV6:
 261                 nhoff += offsetof(struct ipv6hdr, saddr);
 262                 size = sizeof(_addrs);
 263                 break;
 264         default:
 265                 return true;
 266         }
 267
 268         ap = skb_header_pointer(skb, nhoff, size, _addrs);
 269         if (!ap)
 270                 return false;
 271
 272         switch (l3num) {
 273         case NFPROTO_IPV4:
 274                 tuple->src.u3.ip = ap[0];
 275                 tuple->dst.u3.ip = ap[1];
 276                 break;
 277         case NFPROTO_IPV6:
 278                 memcpy(tuple->src.u3.ip6, ap, sizeof(tuple->src.u3.ip6));
 279                 memcpy(tuple->dst.u3.ip6, ap + 4, sizeof(tuple->dst.u3.ip6));
 280                 break;
 281         }
 282
 283         tuple->dst.protonum = protonum;
 284         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 285
 286         switch (protonum) {
 287 #if IS_ENABLED(CONFIG_IPV6)
 288         case IPPROTO_ICMPV6:
 289                 return icmpv6_pkt_to_tuple(skb, dataoff, net, tuple);
 290 #endif
 291         case IPPROTO_ICMP:
 292                 return icmp_pkt_to_tuple(skb, dataoff, net, tuple);
 293 #ifdef CONFIG_NF_CT_PROTO_GRE
 294         case IPPROTO_GRE:
 295                 return gre_pkt_to_tuple(skb, dataoff, net, tuple);
 296 #endif
 297         case IPPROTO_TCP:
 298         case IPPROTO_UDP: /* fallthrough */
 299                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 300 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
 301         case IPPROTO_UDPLITE:
 302                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 303 #endif
 304 #ifdef CONFIG_NF_CT_PROTO_SCTP
 305         case IPPROTO_SCTP:
 306                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 307 #endif
 308 #ifdef CONFIG_NF_CT_PROTO_DCCP
 309         case IPPROTO_DCCP:
 310                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 311 #endif
 312         default:
 313                 break;
 314         }
 315
 316         return true;
 317 }
 318
 319 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
 320                             u_int8_t *protonum)
 321 {
 322         int dataoff = -1;
 323         const struct iphdr *iph;
 324         struct iphdr _iph;
 325
 326         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
 327         if (!iph)
 328                 return -1;
 329
 330         /* Conntrack defragments packets, we might still see fragments
 331          * inside ICMP packets though.
 332          */
 333         if (iph->frag_off & htons(IP_OFFSET))
 334                 return -1;
 335
 336         dataoff = nhoff + (iph->ihl << 2);
 337         *protonum = iph->protocol;
 338
 339         /* Check bogus IP headers */
 340         if (dataoff > skb->len) {
 341                 pr_debug("bogus IPv4 packet: nhoff %u, ihl %u, skblen %u\n",
 342                          nhoff, iph->ihl << 2, skb->len);
 343                 return -1;
 344         }
 345         return dataoff;
 346 }
 347
 348 #if IS_ENABLED(CONFIG_IPV6)
 349 static int ipv6_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
 350                             u8 *protonum)
 351 {
 352         int protoff = -1;
 353         unsigned int extoff = nhoff + sizeof(struct ipv6hdr);
 354         __be16 frag_off;
 355         u8 nexthdr;
 356
 357         if (skb_copy_bits(skb, nhoff + offsetof(struct ipv6hdr, nexthdr),
 358                           &nexthdr, sizeof(nexthdr)) != 0) {
 359                 pr_debug("can't get nexthdr\n");
 360                 return -1;
 361         }
 362         protoff = ipv6_skip_exthdr(skb, extoff, &nexthdr, &frag_off);
 363         /*
 364          * (protoff == skb->len) means the packet has not data, just
 365          * IPv6 and possibly extensions headers, but it is tracked anyway
 366          */
 367         if (protoff < 0 || (frag_off & htons(~0x7)) != 0) {
 368                 pr_debug("can't find proto in pkt\n");
 369                 return -1;
 370         }
 371
 372         *protonum = nexthdr;
 373         return protoff;
 374 }
 375 #endif
 376
 377 static int get_l4proto(const struct sk_buff *skb,
 378                        unsigned int nhoff, u8 pf, u8 *l4num)
 379 {
 380         switch (pf) {
 381         case NFPROTO_IPV4:
 382                 return ipv4_get_l4proto(skb, nhoff, l4num);
 383 #if IS_ENABLED(CONFIG_IPV6)
 384         case NFPROTO_IPV6:
 385                 return ipv6_get_l4proto(skb, nhoff, l4num);
 386 #endif
 387         default:
 388                 *l4num = 0;
 389                 break;
 390         }
 391         return -1;
 392 }
 393
 394 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 395                        u_int16_t l3num,
 396                        struct net *net, struct nf_conntrack_tuple *tuple)
 397 {
 398         u8 protonum;
 399         int protoff;
 400
 401         protoff = get_l4proto(skb, nhoff, l3num, &protonum);
 402         if (protoff <= 0)
 403                 return false;
 404
 405         return nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple);
 406 }
 407 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 408
 409 bool
 410 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 411                    const struct nf_conntrack_tuple *orig)
 412 {
 413         memset(inverse, 0, sizeof(*inverse));
 414
 415         inverse->src.l3num = orig->src.l3num;
 416
 417         switch (orig->src.l3num) {
 418         case NFPROTO_IPV4:
 419                 inverse->src.u3.ip = orig->dst.u3.ip;
 420                 inverse->dst.u3.ip = orig->src.u3.ip;
 421                 break;
 422         case NFPROTO_IPV6:
 423                 inverse->src.u3.in6 = orig->dst.u3.in6;
 424                 inverse->dst.u3.in6 = orig->src.u3.in6;
 425                 break;
 426         default:
 427                 break;
 428         }
 429
 430         inverse->dst.dir = !orig->dst.dir;
 431
 432         inverse->dst.protonum = orig->dst.protonum;
 433
 434         switch (orig->dst.protonum) {
 435         case IPPROTO_ICMP:
 436                 return nf_conntrack_invert_icmp_tuple(inverse, orig);
 437 #if IS_ENABLED(CONFIG_IPV6)
 438         case IPPROTO_ICMPV6:
 439                 return nf_conntrack_invert_icmpv6_tuple(inverse, orig);
 440 #endif
 441         }
 442
 443         inverse->src.u.all = orig->dst.u.all;
 444         inverse->dst.u.all = orig->src.u.all;
 445         return true;
 446 }
 447 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 448
 449 /* Generate a almost-unique pseudo-id for a given conntrack.
 450  *
 451  * intentionally doesn't re-use any of the seeds used for hash
 452  * table location, we assume id gets exposed to userspace.
 453  *
 454  * Following nf_conn items do not change throughout lifetime
 455  * of the nf_conn:
 456  *
 457  * 1. nf_conn address
 458  * 2. nf_conn->master address (normally NULL)
 459  * 3. the associated net namespace
 460  * 4. the original direction tuple
 461  */
 462 u32 nf_ct_get_id(const struct nf_conn *ct)
 463 {
 464         static __read_mostly siphash_key_t ct_id_seed;
 465         unsigned long a, b, c, d;
 466
 467         net_get_random_once(&ct_id_seed, sizeof(ct_id_seed));
 468
 469         a = (unsigned long)ct;
 470         b = (unsigned long)ct->master;
 471         c = (unsigned long)nf_ct_net(ct);
 472         d = (unsigned long)siphash(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 473                                    sizeof(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple),
 474                                    &ct_id_seed);
 475 #ifdef CONFIG_64BIT
 476         return siphash_4u64((u64)a, (u64)b, (u64)c, (u64)d, &ct_id_seed);
 477 #else
 478         return siphash_4u32((u32)a, (u32)b, (u32)c, (u32)d, &ct_id_seed);
 479 #endif
 480 }
 481 EXPORT_SYMBOL_GPL(nf_ct_get_id);
 482
 483 static void
 484 clean_from_lists(struct nf_conn *ct)
 485 {
 486         pr_debug("clean_from_lists(%p)\n", ct);
 487         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 488         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 489
 490         /* Destroy all pending expectations */
 491         nf_ct_remove_expectations(ct);
 492 }
 493
 494 /* must be called with local_bh_disable */
 495 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
 496 {
 497         struct ct_pcpu *pcpu;
 498
 499         /* add this conntrack to the (per cpu) dying list */
 500         ct->cpu = smp_processor_id();
 501         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 502
 503         spin_lock(&pcpu->lock);
 504         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 505                              &pcpu->dying);
 506         spin_unlock(&pcpu->lock);
 507 }
 508
 509 /* must be called with local_bh_disable */
 510 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
 511 {
 512         struct ct_pcpu *pcpu;
 513
 514         /* add this conntrack to the (per cpu) unconfirmed list */
 515         ct->cpu = smp_processor_id();
 516         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 517
 518         spin_lock(&pcpu->lock);
 519         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 520                              &pcpu->unconfirmed);
 521         spin_unlock(&pcpu->lock);
 522 }
 523
 524 /* must be called with local_bh_disable */
 525 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
 526 {
 527         struct ct_pcpu *pcpu;
 528
 529         /* We overload first tuple to link into unconfirmed or dying list.*/
 530         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 531
 532         spin_lock(&pcpu->lock);
 533         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 534         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 535         spin_unlock(&pcpu->lock);
 536 }
 537
 538 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
 539
 540 /* Released via destroy_conntrack() */
 541 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
 542                                  const struct nf_conntrack_zone *zone,
 543                                  gfp_t flags)
 544 {
 545         struct nf_conn *tmpl, *p;
 546
 547         if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) {
 548                 tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags);
 549                 if (!tmpl)
 550                         return NULL;
 551
 552                 p = tmpl;
 553                 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
 554                 if (tmpl != p) {
 555                         tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
 556                         tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p;
 557                 }
 558         } else {
 559                 tmpl = kzalloc(sizeof(*tmpl), flags);
 560                 if (!tmpl)
 561                         return NULL;
 562         }
 563
 564         tmpl->status = IPS_TEMPLATE;
 565         write_pnet(&tmpl->ct_net, net);
 566         nf_ct_zone_add(tmpl, zone);
 567         atomic_set(&tmpl->ct_general.use, 0);
 568
 569         return tmpl;
 570 }
 571 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
 572
 573 void nf_ct_tmpl_free(struct nf_conn *tmpl)
 574 {
 575         nf_ct_ext_destroy(tmpl);
 576
 577         if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK)
 578                 kfree((char *)tmpl - tmpl->proto.tmpl_padto);
 579         else
 580                 kfree(tmpl);
 581 }
 582 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
 583
 584 static void destroy_gre_conntrack(struct nf_conn *ct)
 585 {
 586 #ifdef CONFIG_NF_CT_PROTO_GRE
 587         struct nf_conn *master = ct->master;
 588
 589         if (master)
 590                 nf_ct_gre_keymap_destroy(master);
 591 #endif
 592 }
 593
 594 static void
 595 destroy_conntrack(struct nf_conntrack *nfct)
 596 {
 597         struct nf_conn *ct = (struct nf_conn *)nfct;
 598
 599         pr_debug("destroy_conntrack(%p)\n", ct);
 600         WARN_ON(atomic_read(&nfct->use) != 0);
 601
 602         if (unlikely(nf_ct_is_template(ct))) {
 603                 nf_ct_tmpl_free(ct);
 604                 return;
 605         }
 606
 607         if (unlikely(nf_ct_protonum(ct) == IPPROTO_GRE))
 608                 destroy_gre_conntrack(ct);
 609
 610         local_bh_disable();
 611         /* Expectations will have been removed in clean_from_lists,
 612          * except TFTP can create an expectation on the first packet,
 613          * before connection is in the list, so we need to clean here,
 614          * too.
 615          */
 616         nf_ct_remove_expectations(ct);
 617
 618         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 619
 620         local_bh_enable();
 621
 622         if (ct->master)
 623                 nf_ct_put(ct->master);
 624
 625         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 626         nf_conntrack_free(ct);
 627 }
 628
 629 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 630 {
 631         struct net *net = nf_ct_net(ct);
 632         unsigned int hash, reply_hash;
 633         unsigned int sequence;
 634
 635         nf_ct_helper_destroy(ct);
 636
 637         local_bh_disable();
 638         do {
 639                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 640                 hash = hash_conntrack(net,
 641                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 642                 reply_hash = hash_conntrack(net,
 643                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 644         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 645
 646         clean_from_lists(ct);
 647         nf_conntrack_double_unlock(hash, reply_hash);
 648
 649         nf_ct_add_to_dying_list(ct);
 650
 651         local_bh_enable();
 652 }
 653
 654 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 655 {
 656         struct nf_conn_tstamp *tstamp;
 657
 658         if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
 659                 return false;
 660
 661         tstamp = nf_conn_tstamp_find(ct);
 662         if (tstamp && tstamp->stop == 0)
 663                 tstamp->stop = ktime_get_real_ns();
 664
 665         if (nf_conntrack_event_report(IPCT_DESTROY, ct,
 666                                     portid, report) < 0) {
 667                 /* destroy event was not delivered. nf_ct_put will
 668                  * be done by event cache worker on redelivery.
 669                  */
 670                 nf_ct_delete_from_lists(ct);
 671                 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
 672                 return false;
 673         }
 674
 675         nf_conntrack_ecache_work(nf_ct_net(ct));
 676         nf_ct_delete_from_lists(ct);
 677         nf_ct_put(ct);
 678         return true;
 679 }
 680 EXPORT_SYMBOL_GPL(nf_ct_delete);
 681
 682 static inline bool
 683 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
 684                 const struct nf_conntrack_tuple *tuple,
 685                 const struct nf_conntrack_zone *zone,
 686                 const struct net *net)
 687 {
 688         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 689
 690         /* A conntrack can be recreated with the equal tuple,
 691          * so we need to check that the conntrack is confirmed
 692          */
 693         return nf_ct_tuple_equal(tuple, &h->tuple) &&
 694                nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
 695                nf_ct_is_confirmed(ct) &&
 696                net_eq(net, nf_ct_net(ct));
 697 }
 698
 699 static inline bool
 700 nf_ct_match(const struct nf_conn *ct1, const struct nf_conn *ct2)
 701 {
 702         return nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 703                                  &ct2->tuplehash[IP_CT_DIR_ORIGINAL].tuple) &&
 704                nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_REPLY].tuple,
 705                                  &ct2->tuplehash[IP_CT_DIR_REPLY].tuple) &&
 706                nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_ORIGINAL) &&
 707                nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_REPLY) &&
 708                net_eq(nf_ct_net(ct1), nf_ct_net(ct2));
 709 }
 710
 711 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
 712 static void nf_ct_gc_expired(struct nf_conn *ct)
 713 {
 714         if (!atomic_inc_not_zero(&ct->ct_general.use))
 715                 return;
 716
 717         if (nf_ct_should_gc(ct))
 718                 nf_ct_kill(ct);
 719
 720         nf_ct_put(ct);
 721 }
 722
 723 /*
 724  * Warning :
 725  * - Caller must take a reference on returned object
 726  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 727  */
 728 static struct nf_conntrack_tuple_hash *
 729 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
 730                       const struct nf_conntrack_tuple *tuple, u32 hash)
 731 {
 732         struct nf_conntrack_tuple_hash *h;
 733         struct hlist_nulls_head *ct_hash;
 734         struct hlist_nulls_node *n;
 735         unsigned int bucket, hsize;
 736
 737 begin:
 738         nf_conntrack_get_ht(&ct_hash, &hsize);
 739         bucket = reciprocal_scale(hash, hsize);
 740
 741         hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
 742                 struct nf_conn *ct;
 743
 744                 ct = nf_ct_tuplehash_to_ctrack(h);
 745                 if (nf_ct_is_expired(ct)) {
 746                         nf_ct_gc_expired(ct);
 747                         continue;
 748                 }
 749
 750                 if (nf_ct_key_equal(h, tuple, zone, net))
 751                         return h;
 752         }
 753         /*
 754          * if the nulls value we got at the end of this lookup is
 755          * not the expected one, we must restart lookup.
 756          * We probably met an item that was moved to another chain.
 757          */
 758         if (get_nulls_value(n) != bucket) {
 759                 NF_CT_STAT_INC_ATOMIC(net, search_restart);
 760                 goto begin;
 761         }
 762
 763         return NULL;
 764 }
 765
 766 /* Find a connection corresponding to a tuple. */
 767 static struct nf_conntrack_tuple_hash *
 768 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 769                         const struct nf_conntrack_tuple *tuple, u32 hash)
 770 {
 771         struct nf_conntrack_tuple_hash *h;
 772         struct nf_conn *ct;
 773
 774         rcu_read_lock();
 775
 776         h = ____nf_conntrack_find(net, zone, tuple, hash);
 777         if (h) {
 778                 /* We have a candidate that matches the tuple we're interested
 779                  * in, try to obtain a reference and re-check tuple
 780                  */
 781                 ct = nf_ct_tuplehash_to_ctrack(h);
 782                 if (likely(atomic_inc_not_zero(&ct->ct_general.use))) {
 783                         if (likely(nf_ct_key_equal(h, tuple, zone, net)))
 784                                 goto found;
 785
 786                         /* TYPESAFE_BY_RCU recycled the candidate */
 787                         nf_ct_put(ct);
 788                 }
 789
 790                 h = NULL;
 791         }
 792 found:
 793         rcu_read_unlock();
 794
 795         return h;
 796 }
 797
 798 struct nf_conntrack_tuple_hash *
 799 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 800                       const struct nf_conntrack_tuple *tuple)
 801 {
 802         return __nf_conntrack_find_get(net, zone, tuple,
 803                                        hash_conntrack_raw(tuple, net));
 804 }
 805 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 806
 807 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 808                                        unsigned int hash,
 809                                        unsigned int reply_hash)
 810 {
 811         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 812                            &nf_conntrack_hash[hash]);
 813         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 814                            &nf_conntrack_hash[reply_hash]);
 815 }
 816
 817 int
 818 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 819 {
 820         const struct nf_conntrack_zone *zone;
 821         struct net *net = nf_ct_net(ct);
 822         unsigned int hash, reply_hash;
 823         struct nf_conntrack_tuple_hash *h;
 824         struct hlist_nulls_node *n;
 825         unsigned int sequence;
 826
 827         zone = nf_ct_zone(ct);
 828
 829         local_bh_disable();
 830         do {
 831                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 832                 hash = hash_conntrack(net,
 833                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 834                 reply_hash = hash_conntrack(net,
 835                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 836         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 837
 838         /* See if there's one in the list already, including reverse */
 839         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
 840                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 841                                     zone, net))
 842                         goto out;
 843
 844         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
 845                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 846                                     zone, net))
 847                         goto out;
 848
 849         smp_wmb();
 850         /* The caller holds a reference to this object */
 851         atomic_set(&ct->ct_general.use, 2);
 852         __nf_conntrack_hash_insert(ct, hash, reply_hash);
 853         nf_conntrack_double_unlock(hash, reply_hash);
 854         NF_CT_STAT_INC(net, insert);
 855         local_bh_enable();
 856         return 0;
 857
 858 out:
 859         nf_conntrack_double_unlock(hash, reply_hash);
 860         NF_CT_STAT_INC(net, insert_failed);
 861         local_bh_enable();
 862         return -EEXIST;
 863 }
 864 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 865
 866 static inline void nf_ct_acct_update(struct nf_conn *ct,
 867                                      enum ip_conntrack_info ctinfo,
 868                                      unsigned int len)
 869 {
 870         struct nf_conn_acct *acct;
 871
 872         acct = nf_conn_acct_find(ct);
 873         if (acct) {
 874                 struct nf_conn_counter *counter = acct->counter;
 875
 876                 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
 877                 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
 878         }
 879 }
 880
 881 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
 882                              const struct nf_conn *loser_ct)
 883 {
 884         struct nf_conn_acct *acct;
 885
 886         acct = nf_conn_acct_find(loser_ct);
 887         if (acct) {
 888                 struct nf_conn_counter *counter = acct->counter;
 889                 unsigned int bytes;
 890
 891                 /* u32 should be fine since we must have seen one packet. */
 892                 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
 893                 nf_ct_acct_update(ct, ctinfo, bytes);
 894         }
 895 }
 896
 897 /* Resolve race on insertion if this protocol allows this. */
 898 static __cold noinline int
 899 nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
 900                     enum ip_conntrack_info ctinfo,
 901                     struct nf_conntrack_tuple_hash *h)
 902 {
 903         /* This is the conntrack entry already in hashes that won race. */
 904         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 905         const struct nf_conntrack_l4proto *l4proto;
 906         enum ip_conntrack_info oldinfo;
 907         struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo);
 908
 909         l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
 910         if (l4proto->allow_clash &&
 911             !nf_ct_is_dying(ct) &&
 912             atomic_inc_not_zero(&ct->ct_general.use)) {
 913                 if (((ct->status & IPS_NAT_DONE_MASK) == 0) ||
 914                     nf_ct_match(ct, loser_ct)) {
 915                         nf_ct_acct_merge(ct, ctinfo, loser_ct);
 916                         nf_conntrack_put(&loser_ct->ct_general);
 917                         nf_ct_set(skb, ct, oldinfo);
 918                         return NF_ACCEPT;
 919                 }
 920                 nf_ct_put(ct);
 921         }
 922         NF_CT_STAT_INC(net, drop);
 923         return NF_DROP;
 924 }
 925
 926 /* Confirm a connection given skb; places it in hash table */
 927 int
 928 __nf_conntrack_confirm(struct sk_buff *skb)
 929 {
 930         const struct nf_conntrack_zone *zone;
 931         unsigned int hash, reply_hash;
 932         struct nf_conntrack_tuple_hash *h;
 933         struct nf_conn *ct;
 934         struct nf_conn_help *help;
 935         struct nf_conn_tstamp *tstamp;
 936         struct hlist_nulls_node *n;
 937         enum ip_conntrack_info ctinfo;
 938         struct net *net;
 939         unsigned int sequence;
 940         int ret = NF_DROP;
 941
 942         ct = nf_ct_get(skb, &ctinfo);
 943         net = nf_ct_net(ct);
 944
 945         /* ipt_REJECT uses nf_conntrack_attach to attach related
 946            ICMP/TCP RST packets in other direction.  Actual packet
 947            which created connection will be IP_CT_NEW or for an
 948            expected connection, IP_CT_RELATED. */
 949         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 950                 return NF_ACCEPT;
 951
 952         zone = nf_ct_zone(ct);
 953         local_bh_disable();
 954
 955         do {
 956                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 957                 /* reuse the hash saved before */
 958                 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 959                 hash = scale_hash(hash);
 960                 reply_hash = hash_conntrack(net,
 961                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 962
 963         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 964
 965         /* We're not in hash table, and we refuse to set up related
 966          * connections for unconfirmed conns.  But packet copies and
 967          * REJECT will give spurious warnings here.
 968          */
 969
 970         /* Another skb with the same unconfirmed conntrack may
 971          * win the race. This may happen for bridge(br_flood)
 972          * or broadcast/multicast packets do skb_clone with
 973          * unconfirmed conntrack.
 974          */
 975         if (unlikely(nf_ct_is_confirmed(ct))) {
 976                 WARN_ON_ONCE(1);
 977                 nf_conntrack_double_unlock(hash, reply_hash);
 978                 local_bh_enable();
 979                 return NF_DROP;
 980         }
 981
 982         pr_debug("Confirming conntrack %p\n", ct);
 983         /* We have to check the DYING flag after unlink to prevent
 984          * a race against nf_ct_get_next_corpse() possibly called from
 985          * user context, else we insert an already 'dead' hash, blocking
 986          * further use of that particular connection -JM.
 987          */
 988         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 989
 990         if (unlikely(nf_ct_is_dying(ct))) {
 991                 nf_ct_add_to_dying_list(ct);
 992                 goto dying;
 993         }
 994
 995         /* See if there's one in the list already, including reverse:
 996            NAT could have grabbed it without realizing, since we're
 997            not in the hash.  If there is, we lost race. */
 998         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
 999                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1000                                     zone, net))
1001                         goto out;
1002
1003         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
1004                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
1005                                     zone, net))
1006                         goto out;
1007
1008         /* Timer relative to confirmation time, not original
1009            setting time, otherwise we'd get timer wrap in
1010            weird delay cases. */
1011         ct->timeout += nfct_time_stamp;
1012         atomic_inc(&ct->ct_general.use);
1013         ct->status |= IPS_CONFIRMED;
1014
1015         /* set conntrack timestamp, if enabled. */
1016         tstamp = nf_conn_tstamp_find(ct);
1017         if (tstamp)
1018                 tstamp->start = ktime_get_real_ns();
1019
1020         /* Since the lookup is lockless, hash insertion must be done after
1021          * starting the timer and setting the CONFIRMED bit. The RCU barriers
1022          * guarantee that no other CPU can find the conntrack before the above
1023          * stores are visible.
1024          */
1025         __nf_conntrack_hash_insert(ct, hash, reply_hash);
1026         nf_conntrack_double_unlock(hash, reply_hash);
1027         local_bh_enable();
1028
1029         help = nfct_help(ct);
1030         if (help && help->helper)
1031                 nf_conntrack_event_cache(IPCT_HELPER, ct);
1032
1033         nf_conntrack_event_cache(master_ct(ct) ?
1034                                  IPCT_RELATED : IPCT_NEW, ct);
1035         return NF_ACCEPT;
1036
1037 out:
1038         nf_ct_add_to_dying_list(ct);
1039         ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
1040 dying:
1041         nf_conntrack_double_unlock(hash, reply_hash);
1042         NF_CT_STAT_INC(net, insert_failed);
1043         local_bh_enable();
1044         return ret;
1045 }
1046 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
1047
1048 /* Returns true if a connection correspondings to the tuple (required
1049    for NAT). */
1050 int
1051 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
1052                          const struct nf_conn *ignored_conntrack)
1053 {
1054         struct net *net = nf_ct_net(ignored_conntrack);
1055         const struct nf_conntrack_zone *zone;
1056         struct nf_conntrack_tuple_hash *h;
1057         struct hlist_nulls_head *ct_hash;
1058         unsigned int hash, hsize;
1059         struct hlist_nulls_node *n;
1060         struct nf_conn *ct;
1061
1062         zone = nf_ct_zone(ignored_conntrack);
1063
1064         rcu_read_lock();
1065  begin:
1066         nf_conntrack_get_ht(&ct_hash, &hsize);
1067         hash = __hash_conntrack(net, tuple, hsize);
1068
1069         hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
1070                 ct = nf_ct_tuplehash_to_ctrack(h);
1071
1072                 if (ct == ignored_conntrack)
1073                         continue;
1074
1075                 if (nf_ct_is_expired(ct)) {
1076                         nf_ct_gc_expired(ct);
1077                         continue;
1078                 }
1079
1080                 if (nf_ct_key_equal(h, tuple, zone, net)) {
1081                         /* Tuple is taken already, so caller will need to find
1082                          * a new source port to use.
1083                          *
1084                          * Only exception:
1085                          * If the *original tuples* are identical, then both
1086                          * conntracks refer to the same flow.
1087                          * This is a rare situation, it can occur e.g. when
1088                          * more than one UDP packet is sent from same socket
1089                          * in different threads.
1090                          *
1091                          * Let nf_ct_resolve_clash() deal with this later.
1092                          */
1093                         if (nf_ct_tuple_equal(&ignored_conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1094                                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple))
1095                                 continue;
1096
1097                         NF_CT_STAT_INC_ATOMIC(net, found);
1098                         rcu_read_unlock();
1099                         return 1;
1100                 }
1101         }
1102
1103         if (get_nulls_value(n) != hash) {
1104                 NF_CT_STAT_INC_ATOMIC(net, search_restart);
1105                 goto begin;
1106         }
1107
1108         rcu_read_unlock();
1109
1110         return 0;
1111 }
1112 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
1113
1114 #define NF_CT_EVICTION_RANGE    8
1115
1116 /* There's a small race here where we may free a just-assured
1117    connection.  Too bad: we're in trouble anyway. */
1118 static unsigned int early_drop_list(struct net *net,
1119                                     struct hlist_nulls_head *head)
1120 {
1121         struct nf_conntrack_tuple_hash *h;
1122         struct hlist_nulls_node *n;
1123         unsigned int drops = 0;
1124         struct nf_conn *tmp;
1125
1126         hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
1127                 tmp = nf_ct_tuplehash_to_ctrack(h);
1128
1129                 if (test_bit(IPS_OFFLOAD_BIT, &tmp->status))
1130                         continue;
1131
1132                 if (nf_ct_is_expired(tmp)) {
1133                         nf_ct_gc_expired(tmp);
1134                         continue;
1135                 }
1136
1137                 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
1138                     !net_eq(nf_ct_net(tmp), net) ||
1139                     nf_ct_is_dying(tmp))
1140                         continue;
1141
1142                 if (!atomic_inc_not_zero(&tmp->ct_general.use))
1143                         continue;
1144
1145                 /* kill only if still in same netns -- might have moved due to
1146                  * SLAB_TYPESAFE_BY_RCU rules.
1147                  *
1148                  * We steal the timer reference.  If that fails timer has
1149                  * already fired or someone else deleted it. Just drop ref
1150                  * and move to next entry.
1151                  */
1152                 if (net_eq(nf_ct_net(tmp), net) &&
1153                     nf_ct_is_confirmed(tmp) &&
1154                     nf_ct_delete(tmp, 0, 0))
1155                         drops++;
1156
1157                 nf_ct_put(tmp);
1158         }
1159
1160         return drops;
1161 }
1162
1163 static noinline int early_drop(struct net *net, unsigned int hash)
1164 {
1165         unsigned int i, bucket;
1166
1167         for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
1168                 struct hlist_nulls_head *ct_hash;
1169                 unsigned int hsize, drops;
1170
1171                 rcu_read_lock();
1172                 nf_conntrack_get_ht(&ct_hash, &hsize);
1173                 if (!i)
1174                         bucket = reciprocal_scale(hash, hsize);
1175                 else
1176                         bucket = (bucket + 1) % hsize;
1177
1178                 drops = early_drop_list(net, &ct_hash[bucket]);
1179                 rcu_read_unlock();
1180
1181                 if (drops) {
1182                         NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
1183                         return true;
1184                 }
1185         }
1186
1187         return false;
1188 }
1189
1190 static bool gc_worker_skip_ct(const struct nf_conn *ct)
1191 {
1192         return !nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct);
1193 }
1194
1195 static bool gc_worker_can_early_drop(const struct nf_conn *ct)
1196 {
1197         const struct nf_conntrack_l4proto *l4proto;
1198
1199         if (!test_bit(IPS_ASSURED_BIT, &ct->status))
1200                 return true;
1201
1202         l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
1203         if (l4proto->can_early_drop && l4proto->can_early_drop(ct))
1204                 return true;
1205
1206         return false;
1207 }
1208
1209 #define DAY     (86400 * HZ)
1210
1211 /* Set an arbitrary timeout large enough not to ever expire, this save
1212  * us a check for the IPS_OFFLOAD_BIT from the packet path via
1213  * nf_ct_is_expired().
1214  */
1215 static void nf_ct_offload_timeout(struct nf_conn *ct)
1216 {
1217         if (nf_ct_expires(ct) < DAY / 2)
1218                 ct->timeout = nfct_time_stamp + DAY;
1219 }
1220
1221 static void gc_worker(struct work_struct *work)
1222 {
1223         unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u);
1224         unsigned int i, goal, buckets = 0, expired_count = 0;
1225         unsigned int nf_conntrack_max95 = 0;
1226         struct conntrack_gc_work *gc_work;
1227         unsigned int ratio, scanned = 0;
1228         unsigned long next_run;
1229
1230         gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
1231
1232         goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV;
1233         i = gc_work->last_bucket;
1234         if (gc_work->early_drop)
1235                 nf_conntrack_max95 = nf_conntrack_max / 100u * 95u;
1236
1237         do {
1238                 struct nf_conntrack_tuple_hash *h;
1239                 struct hlist_nulls_head *ct_hash;
1240                 struct hlist_nulls_node *n;
1241                 unsigned int hashsz;
1242                 struct nf_conn *tmp;
1243
1244                 i++;
1245                 rcu_read_lock();
1246
1247                 nf_conntrack_get_ht(&ct_hash, &hashsz);
1248                 if (i >= hashsz)
1249                         i = 0;
1250
1251                 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
1252                         struct net *net;
1253
1254                         tmp = nf_ct_tuplehash_to_ctrack(h);
1255
1256                         scanned++;
1257                         if (test_bit(IPS_OFFLOAD_BIT, &tmp->status)) {
1258                                 nf_ct_offload_timeout(tmp);
1259                                 continue;
1260                         }
1261
1262                         if (nf_ct_is_expired(tmp)) {
1263                                 nf_ct_gc_expired(tmp);
1264                                 expired_count++;
1265                                 continue;
1266                         }
1267
1268                         if (nf_conntrack_max95 == 0 || gc_worker_skip_ct(tmp))
1269                                 continue;
1270
1271                         net = nf_ct_net(tmp);
1272                         if (atomic_read(&net->ct.count) < nf_conntrack_max95)
1273                                 continue;
1274
1275                         /* need to take reference to avoid possible races */
1276                         if (!atomic_inc_not_zero(&tmp->ct_general.use))
1277                                 continue;
1278
1279                         if (gc_worker_skip_ct(tmp)) {
1280                                 nf_ct_put(tmp);
1281                                 continue;
1282                         }
1283
1284                         if (gc_worker_can_early_drop(tmp))
1285                                 nf_ct_kill(tmp);
1286
1287                         nf_ct_put(tmp);
1288                 }
1289
1290                 /* could check get_nulls_value() here and restart if ct
1291                  * was moved to another chain.  But given gc is best-effort
1292                  * we will just continue with next hash slot.
1293                  */
1294                 rcu_read_unlock();
1295                 cond_resched();
1296         } while (++buckets < goal);
1297
1298         if (gc_work->exiting)
1299                 return;
1300
1301         /*
1302          * Eviction will normally happen from the packet path, and not
1303          * from this gc worker.
1304          *
1305          * This worker is only here to reap expired entries when system went
1306          * idle after a busy period.
1307          *
1308          * The heuristics below are supposed to balance conflicting goals:
1309          *
1310          * 1. Minimize time until we notice a stale entry
1311          * 2. Maximize scan intervals to not waste cycles
1312          *
1313          * Normally, expire ratio will be close to 0.
1314          *
1315          * As soon as a sizeable fraction of the entries have expired
1316          * increase scan frequency.
1317          */
1318         ratio = scanned ? expired_count * 100 / scanned : 0;
1319         if (ratio > GC_EVICT_RATIO) {
1320                 gc_work->next_gc_run = min_interval;
1321         } else {
1322                 unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV;
1323
1324                 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0);
1325
1326                 gc_work->next_gc_run += min_interval;
1327                 if (gc_work->next_gc_run > max)
1328                         gc_work->next_gc_run = max;
1329         }
1330
1331         next_run = gc_work->next_gc_run;
1332         gc_work->last_bucket = i;
1333         gc_work->early_drop = false;
1334         queue_delayed_work(system_power_efficient_wq, &gc_work->dwork, next_run);
1335 }
1336
1337 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1338 {
1339         INIT_DEFERRABLE_WORK(&gc_work->dwork, gc_worker);
1340         gc_work->next_gc_run = HZ;
1341         gc_work->exiting = false;
1342 }
1343
1344 static struct nf_conn *
1345 __nf_conntrack_alloc(struct net *net,
1346                      const struct nf_conntrack_zone *zone,
1347                      const struct nf_conntrack_tuple *orig,
1348                      const struct nf_conntrack_tuple *repl,
1349                      gfp_t gfp, u32 hash)
1350 {
1351         struct nf_conn *ct;
1352
1353         /* We don't want any race condition at early drop stage */
1354         atomic_inc(&net->ct.count);
1355
1356         if (nf_conntrack_max &&
1357             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
1358                 if (!early_drop(net, hash)) {
1359                         if (!conntrack_gc_work.early_drop)
1360                                 conntrack_gc_work.early_drop = true;
1361                         atomic_dec(&net->ct.count);
1362                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1363                         return ERR_PTR(-ENOMEM);
1364                 }
1365         }
1366
1367         /*
1368          * Do not use kmem_cache_zalloc(), as this cache uses
1369          * SLAB_TYPESAFE_BY_RCU.
1370          */
1371         ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1372         if (ct == NULL)
1373                 goto out;
1374
1375         spin_lock_init(&ct->lock);
1376         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1377         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1378         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1379         /* save hash for reusing when confirming */
1380         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1381         ct->status = 0;
1382         ct->timeout = 0;
1383         write_pnet(&ct->ct_net, net);
1384         memset(&ct->__nfct_init_offset[0], 0,
1385                offsetof(struct nf_conn, proto) -
1386                offsetof(struct nf_conn, __nfct_init_offset[0]));
1387
1388         nf_ct_zone_add(ct, zone);
1389
1390         /* Because we use RCU lookups, we set ct_general.use to zero before
1391          * this is inserted in any list.
1392          */
1393         atomic_set(&ct->ct_general.use, 0);
1394         return ct;
1395 out:
1396         atomic_dec(&net->ct.count);
1397         return ERR_PTR(-ENOMEM);
1398 }
1399
1400 struct nf_conn *nf_conntrack_alloc(struct net *net,
1401                                    const struct nf_conntrack_zone *zone,
1402                                    const struct nf_conntrack_tuple *orig,
1403                                    const struct nf_conntrack_tuple *repl,
1404                                    gfp_t gfp)
1405 {
1406         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1407 }
1408 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1409
1410 void nf_conntrack_free(struct nf_conn *ct)
1411 {
1412         struct net *net = nf_ct_net(ct);
1413
1414         /* A freed object has refcnt == 0, that's
1415          * the golden rule for SLAB_TYPESAFE_BY_RCU
1416          */
1417         WARN_ON(atomic_read(&ct->ct_general.use) != 0);
1418
1419         nf_ct_ext_destroy(ct);
1420         kmem_cache_free(nf_conntrack_cachep, ct);
1421         smp_mb__before_atomic();
1422         atomic_dec(&net->ct.count);
1423 }
1424 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1425
1426
1427 /* Allocate a new conntrack: we return -ENOMEM if classification
1428    failed due to stress.  Otherwise it really is unclassifiable. */
1429 static noinline struct nf_conntrack_tuple_hash *
1430 init_conntrack(struct net *net, struct nf_conn *tmpl,
1431                const struct nf_conntrack_tuple *tuple,
1432                struct sk_buff *skb,
1433                unsigned int dataoff, u32 hash)
1434 {
1435         struct nf_conn *ct;
1436         struct nf_conn_help *help;
1437         struct nf_conntrack_tuple repl_tuple;
1438         struct nf_conntrack_ecache *ecache;
1439         struct nf_conntrack_expect *exp = NULL;
1440         const struct nf_conntrack_zone *zone;
1441         struct nf_conn_timeout *timeout_ext;
1442         struct nf_conntrack_zone tmp;
1443
1444         if (!nf_ct_invert_tuple(&repl_tuple, tuple)) {
1445                 pr_debug("Can't invert tuple.\n");
1446                 return NULL;
1447         }
1448
1449         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1450         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1451                                   hash);
1452         if (IS_ERR(ct))
1453                 return (struct nf_conntrack_tuple_hash *)ct;
1454
1455         if (!nf_ct_add_synproxy(ct, tmpl)) {
1456                 nf_conntrack_free(ct);
1457                 return ERR_PTR(-ENOMEM);
1458         }
1459
1460         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1461
1462         if (timeout_ext)
1463                 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1464                                       GFP_ATOMIC);
1465
1466         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1467         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1468         nf_ct_labels_ext_add(ct);
1469
1470         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1471         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1472                                  ecache ? ecache->expmask : 0,
1473                              GFP_ATOMIC);
1474
1475         local_bh_disable();
1476         if (net->ct.expect_count) {
1477                 spin_lock(&nf_conntrack_expect_lock);
1478                 exp = nf_ct_find_expectation(net, zone, tuple);
1479                 if (exp) {
1480                         pr_debug("expectation arrives ct=%p exp=%p\n",
1481                                  ct, exp);
1482                         /* Welcome, Mr. Bond.  We've been expecting you... */
1483                         __set_bit(IPS_EXPECTED_BIT, &ct->status);
1484                         /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1485                         ct->master = exp->master;
1486                         if (exp->helper) {
1487                                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
1488                                 if (help)
1489                                         rcu_assign_pointer(help->helper, exp->helper);
1490                         }
1491
1492 #ifdef CONFIG_NF_CONNTRACK_MARK
1493                         ct->mark = exp->master->mark;
1494 #endif
1495 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1496                         ct->secmark = exp->master->secmark;
1497 #endif
1498                         NF_CT_STAT_INC(net, expect_new);
1499                 }
1500                 spin_unlock(&nf_conntrack_expect_lock);
1501         }
1502         if (!exp)
1503                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1504
1505         /* Now it is inserted into the unconfirmed list, bump refcount */
1506         nf_conntrack_get(&ct->ct_general);
1507         nf_ct_add_to_unconfirmed_list(ct);
1508
1509         local_bh_enable();
1510
1511         if (exp) {
1512                 if (exp->expectfn)
1513                         exp->expectfn(ct, exp);
1514                 nf_ct_expect_put(exp);
1515         }
1516
1517         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1518 }
1519
1520 /* On success, returns 0, sets skb->_nfct | ctinfo */
1521 static int
1522 resolve_normal_ct(struct nf_conn *tmpl,
1523                   struct sk_buff *skb,
1524                   unsigned int dataoff,
1525                   u_int8_t protonum,
1526                   const struct nf_hook_state *state)
1527 {
1528         const struct nf_conntrack_zone *zone;
1529         struct nf_conntrack_tuple tuple;
1530         struct nf_conntrack_tuple_hash *h;
1531         enum ip_conntrack_info ctinfo;
1532         struct nf_conntrack_zone tmp;
1533         struct nf_conn *ct;
1534         u32 hash;
1535
1536         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1537                              dataoff, state->pf, protonum, state->net,
1538                              &tuple)) {
1539                 pr_debug("Can't get tuple\n");
1540                 return 0;
1541         }
1542
1543         /* look for tuple match */
1544         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1545         hash = hash_conntrack_raw(&tuple, state->net);
1546         h = __nf_conntrack_find_get(state->net, zone, &tuple, hash);
1547         if (!h) {
1548                 h = init_conntrack(state->net, tmpl, &tuple,
1549                                    skb, dataoff, hash);
1550                 if (!h)
1551                         return 0;
1552                 if (IS_ERR(h))
1553                         return PTR_ERR(h);
1554         }
1555         ct = nf_ct_tuplehash_to_ctrack(h);
1556
1557         /* It exists; we have (non-exclusive) reference. */
1558         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1559                 ctinfo = IP_CT_ESTABLISHED_REPLY;
1560         } else {
1561                 /* Once we've had two way comms, always ESTABLISHED. */
1562                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1563                         pr_debug("normal packet for %p\n", ct);
1564                         ctinfo = IP_CT_ESTABLISHED;
1565                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1566                         pr_debug("related packet for %p\n", ct);
1567                         ctinfo = IP_CT_RELATED;
1568                 } else {
1569                         pr_debug("new packet for %p\n", ct);
1570                         ctinfo = IP_CT_NEW;
1571                 }
1572         }
1573         nf_ct_set(skb, ct, ctinfo);
1574         return 0;
1575 }
1576
1577 /*
1578  * icmp packets need special treatment to handle error messages that are
1579  * related to a connection.
1580  *
1581  * Callers need to check if skb has a conntrack assigned when this
1582  * helper returns; in such case skb belongs to an already known connection.
1583  */
1584 static unsigned int __cold
1585 nf_conntrack_handle_icmp(struct nf_conn *tmpl,
1586                          struct sk_buff *skb,
1587                          unsigned int dataoff,
1588                          u8 protonum,
1589                          const struct nf_hook_state *state)
1590 {
1591         int ret;
1592
1593         if (state->pf == NFPROTO_IPV4 && protonum == IPPROTO_ICMP)
1594                 ret = nf_conntrack_icmpv4_error(tmpl, skb, dataoff, state);
1595 #if IS_ENABLED(CONFIG_IPV6)
1596         else if (state->pf == NFPROTO_IPV6 && protonum == IPPROTO_ICMPV6)
1597                 ret = nf_conntrack_icmpv6_error(tmpl, skb, dataoff, state);
1598 #endif
1599         else
1600                 return NF_ACCEPT;
1601
1602         if (ret <= 0) {
1603                 NF_CT_STAT_INC_ATOMIC(state->net, error);
1604                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1605         }
1606
1607         return ret;
1608 }
1609
1610 static int generic_packet(struct nf_conn *ct, struct sk_buff *skb,
1611                           enum ip_conntrack_info ctinfo)
1612 {
1613         const unsigned int *timeout = nf_ct_timeout_lookup(ct);
1614
1615         if (!timeout)
1616                 timeout = &nf_generic_pernet(nf_ct_net(ct))->timeout;
1617
1618         nf_ct_refresh_acct(ct, ctinfo, skb, *timeout);
1619         return NF_ACCEPT;
1620 }
1621
1622 /* Returns verdict for packet, or -1 for invalid. */
1623 static int nf_conntrack_handle_packet(struct nf_conn *ct,
1624                                       struct sk_buff *skb,
1625                                       unsigned int dataoff,
1626                                       enum ip_conntrack_info ctinfo,
1627                                       const struct nf_hook_state *state)
1628 {
1629         switch (nf_ct_protonum(ct)) {
1630         case IPPROTO_TCP:
1631                 return nf_conntrack_tcp_packet(ct, skb, dataoff,
1632                                                ctinfo, state);
1633         case IPPROTO_UDP:
1634                 return nf_conntrack_udp_packet(ct, skb, dataoff,
1635                                                ctinfo, state);
1636         case IPPROTO_ICMP:
1637                 return nf_conntrack_icmp_packet(ct, skb, ctinfo, state);
1638 #if IS_ENABLED(CONFIG_IPV6)
1639         case IPPROTO_ICMPV6:
1640                 return nf_conntrack_icmpv6_packet(ct, skb, ctinfo, state);
1641 #endif
1642 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
1643         case IPPROTO_UDPLITE:
1644                 return nf_conntrack_udplite_packet(ct, skb, dataoff,
1645                                                    ctinfo, state);
1646 #endif
1647 #ifdef CONFIG_NF_CT_PROTO_SCTP
1648         case IPPROTO_SCTP:
1649                 return nf_conntrack_sctp_packet(ct, skb, dataoff,
1650                                                 ctinfo, state);
1651 #endif
1652 #ifdef CONFIG_NF_CT_PROTO_DCCP
1653         case IPPROTO_DCCP:
1654                 return nf_conntrack_dccp_packet(ct, skb, dataoff,
1655                                                 ctinfo, state);
1656 #endif
1657 #ifdef CONFIG_NF_CT_PROTO_GRE
1658         case IPPROTO_GRE:
1659                 return nf_conntrack_gre_packet(ct, skb, dataoff,
1660                                                ctinfo, state);
1661 #endif
1662         }
1663
1664         return generic_packet(ct, skb, ctinfo);
1665 }
1666
1667 unsigned int
1668 nf_conntrack_in(struct sk_buff *skb, const struct nf_hook_state *state)
1669 {
1670         enum ip_conntrack_info ctinfo;
1671         struct nf_conn *ct, *tmpl;
1672         u_int8_t protonum;
1673         int dataoff, ret;
1674
1675         tmpl = nf_ct_get(skb, &ctinfo);
1676         if (tmpl || ctinfo == IP_CT_UNTRACKED) {
1677                 /* Previously seen (loopback or untracked)?  Ignore. */
1678                 if ((tmpl && !nf_ct_is_template(tmpl)) ||
1679                      ctinfo == IP_CT_UNTRACKED) {
1680                         NF_CT_STAT_INC_ATOMIC(state->net, ignore);
1681                         return NF_ACCEPT;
1682                 }
1683                 skb->_nfct = 0;
1684         }
1685
1686         /* rcu_read_lock()ed by nf_hook_thresh */
1687         dataoff = get_l4proto(skb, skb_network_offset(skb), state->pf, &protonum);
1688         if (dataoff <= 0) {
1689                 pr_debug("not prepared to track yet or error occurred\n");
1690                 NF_CT_STAT_INC_ATOMIC(state->net, error);
1691                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1692                 ret = NF_ACCEPT;
1693                 goto out;
1694         }
1695
1696         if (protonum == IPPROTO_ICMP || protonum == IPPROTO_ICMPV6) {
1697                 ret = nf_conntrack_handle_icmp(tmpl, skb, dataoff,
1698                                                protonum, state);
1699                 if (ret <= 0) {
1700                         ret = -ret;
1701                         goto out;
1702                 }
1703                 /* ICMP[v6] protocol trackers may assign one conntrack. */
1704                 if (skb->_nfct)
1705                         goto out;
1706         }
1707 repeat:
1708         ret = resolve_normal_ct(tmpl, skb, dataoff,
1709                                 protonum, state);
1710         if (ret < 0) {
1711                 /* Too stressed to deal. */
1712                 NF_CT_STAT_INC_ATOMIC(state->net, drop);
1713                 ret = NF_DROP;
1714                 goto out;
1715         }
1716
1717         ct = nf_ct_get(skb, &ctinfo);
1718         if (!ct) {
1719                 /* Not valid part of a connection */
1720                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1721                 ret = NF_ACCEPT;
1722                 goto out;
1723         }
1724
1725         ret = nf_conntrack_handle_packet(ct, skb, dataoff, ctinfo, state);
1726         if (ret <= 0) {
1727                 /* Invalid: inverse of the return code tells
1728                  * the netfilter core what to do */
1729                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1730                 nf_conntrack_put(&ct->ct_general);
1731                 skb->_nfct = 0;
1732                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1733                 if (ret == -NF_DROP)
1734                         NF_CT_STAT_INC_ATOMIC(state->net, drop);
1735                 /* Special case: TCP tracker reports an attempt to reopen a
1736                  * closed/aborted connection. We have to go back and create a
1737                  * fresh conntrack.
1738                  */
1739                 if (ret == -NF_REPEAT)
1740                         goto repeat;
1741                 ret = -ret;
1742                 goto out;
1743         }
1744
1745         if (ctinfo == IP_CT_ESTABLISHED_REPLY &&
1746             !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1747                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1748 out:
1749         if (tmpl)
1750                 nf_ct_put(tmpl);
1751
1752         return ret;
1753 }
1754 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1755
1756 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1757    implicitly racy: see __nf_conntrack_confirm */
1758 void nf_conntrack_alter_reply(struct nf_conn *ct,
1759                               const struct nf_conntrack_tuple *newreply)
1760 {
1761         struct nf_conn_help *help = nfct_help(ct);
1762
1763         /* Should be unconfirmed, so not in hash table yet */
1764         WARN_ON(nf_ct_is_confirmed(ct));
1765
1766         pr_debug("Altering reply tuple of %p to ", ct);
1767         nf_ct_dump_tuple(newreply);
1768
1769         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1770         if (ct->master || (help && !hlist_empty(&help->expectations)))
1771                 return;
1772
1773         rcu_read_lock();
1774         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1775         rcu_read_unlock();
1776 }
1777 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1778
1779 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1780 void __nf_ct_refresh_acct(struct nf_conn *ct,
1781                           enum ip_conntrack_info ctinfo,
1782                           const struct sk_buff *skb,
1783                           u32 extra_jiffies,
1784                           bool do_acct)
1785 {
1786         /* Only update if this is not a fixed timeout */
1787         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1788                 goto acct;
1789
1790         /* If not in hash table, timer will not be active yet */
1791         if (nf_ct_is_confirmed(ct))
1792                 extra_jiffies += nfct_time_stamp;
1793
1794         if (READ_ONCE(ct->timeout) != extra_jiffies)
1795                 WRITE_ONCE(ct->timeout, extra_jiffies);
1796 acct:
1797         if (do_acct)
1798                 nf_ct_acct_update(ct, ctinfo, skb->len);
1799 }
1800 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1801
1802 bool nf_ct_kill_acct(struct nf_conn *ct,
1803                      enum ip_conntrack_info ctinfo,
1804                      const struct sk_buff *skb)
1805 {
1806         nf_ct_acct_update(ct, ctinfo, skb->len);
1807
1808         return nf_ct_delete(ct, 0, 0);
1809 }
1810 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1811
1812 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1813
1814 #include <linux/netfilter/nfnetlink.h>
1815 #include <linux/netfilter/nfnetlink_conntrack.h>
1816 #include <linux/mutex.h>
1817
1818 /* Generic function for tcp/udp/sctp/dccp and alike. */
1819 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1820                                const struct nf_conntrack_tuple *tuple)
1821 {
1822         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1823             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1824                 goto nla_put_failure;
1825         return 0;
1826
1827 nla_put_failure:
1828         return -1;
1829 }
1830 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1831
1832 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1833         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1834         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1835 };
1836 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1837
1838 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1839                                struct nf_conntrack_tuple *t)
1840 {
1841         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1842                 return -EINVAL;
1843
1844         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1845         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1846
1847         return 0;
1848 }
1849 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1850
1851 unsigned int nf_ct_port_nlattr_tuple_size(void)
1852 {
1853         static unsigned int size __read_mostly;
1854
1855         if (!size)
1856                 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1857
1858         return size;
1859 }
1860 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1861 #endif
1862
1863 /* Used by ipt_REJECT and ip6t_REJECT. */
1864 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1865 {
1866         struct nf_conn *ct;
1867         enum ip_conntrack_info ctinfo;
1868
1869         /* This ICMP is in reverse direction to the packet which caused it */
1870         ct = nf_ct_get(skb, &ctinfo);
1871         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1872                 ctinfo = IP_CT_RELATED_REPLY;
1873         else
1874                 ctinfo = IP_CT_RELATED;
1875
1876         /* Attach to new skbuff, and increment count */
1877         nf_ct_set(nskb, ct, ctinfo);
1878         nf_conntrack_get(skb_nfct(nskb));
1879 }
1880
1881 static int nf_conntrack_update(struct net *net, struct sk_buff *skb)
1882 {
1883         struct nf_conntrack_tuple_hash *h;
1884         struct nf_conntrack_tuple tuple;
1885         enum ip_conntrack_info ctinfo;
1886         struct nf_nat_hook *nat_hook;
1887         unsigned int status;
1888         struct nf_conn *ct;
1889         int dataoff;
1890         u16 l3num;
1891         u8 l4num;
1892
1893         ct = nf_ct_get(skb, &ctinfo);
1894         if (!ct || nf_ct_is_confirmed(ct))
1895                 return 0;
1896
1897         l3num = nf_ct_l3num(ct);
1898
1899         dataoff = get_l4proto(skb, skb_network_offset(skb), l3num, &l4num);
1900         if (dataoff <= 0)
1901                 return -1;
1902
1903         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
1904                              l4num, net, &tuple))
1905                 return -1;
1906
1907         if (ct->status & IPS_SRC_NAT) {
1908                 memcpy(tuple.src.u3.all,
1909                        ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.all,
1910                        sizeof(tuple.src.u3.all));
1911                 tuple.src.u.all =
1912                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.all;
1913         }
1914
1915         if (ct->status & IPS_DST_NAT) {
1916                 memcpy(tuple.dst.u3.all,
1917                        ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.all,
1918                        sizeof(tuple.dst.u3.all));
1919                 tuple.dst.u.all =
1920                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u.all;
1921         }
1922
1923         h = nf_conntrack_find_get(net, nf_ct_zone(ct), &tuple);
1924         if (!h)
1925                 return 0;
1926
1927         /* Store status bits of the conntrack that is clashing to re-do NAT
1928          * mangling according to what it has been done already to this packet.
1929          */
1930         status = ct->status;
1931
1932         nf_ct_put(ct);
1933         ct = nf_ct_tuplehash_to_ctrack(h);
1934         nf_ct_set(skb, ct, ctinfo);
1935
1936         nat_hook = rcu_dereference(nf_nat_hook);
1937         if (!nat_hook)
1938                 return 0;
1939
1940         if (status & IPS_SRC_NAT &&
1941             nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_SRC,
1942                                 IP_CT_DIR_ORIGINAL) == NF_DROP)
1943                 return -1;
1944
1945         if (status & IPS_DST_NAT &&
1946             nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_DST,
1947                                 IP_CT_DIR_ORIGINAL) == NF_DROP)
1948                 return -1;
1949
1950         return 0;
1951 }
1952
1953 static bool nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
1954                                        const struct sk_buff *skb)
1955 {
1956         const struct nf_conntrack_tuple *src_tuple;
1957         const struct nf_conntrack_tuple_hash *hash;
1958         struct nf_conntrack_tuple srctuple;
1959         enum ip_conntrack_info ctinfo;
1960         struct nf_conn *ct;
1961
1962         ct = nf_ct_get(skb, &ctinfo);
1963         if (ct) {
1964                 src_tuple = nf_ct_tuple(ct, CTINFO2DIR(ctinfo));
1965                 memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
1966                 return true;
1967         }
1968
1969         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
1970                                NFPROTO_IPV4, dev_net(skb->dev),
1971                                &srctuple))
1972                 return false;
1973
1974         hash = nf_conntrack_find_get(dev_net(skb->dev),
1975                                      &nf_ct_zone_dflt,
1976                                      &srctuple);
1977         if (!hash)
1978                 return false;
1979
1980         ct = nf_ct_tuplehash_to_ctrack(hash);
1981         src_tuple = nf_ct_tuple(ct, !hash->tuple.dst.dir);
1982         memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
1983         nf_ct_put(ct);
1984
1985         return true;
1986 }
1987
1988 /* Bring out ya dead! */
1989 static struct nf_conn *
1990 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
1991                 void *data, unsigned int *bucket)
1992 {
1993         struct nf_conntrack_tuple_hash *h;
1994         struct nf_conn *ct;
1995         struct hlist_nulls_node *n;
1996         spinlock_t *lockp;
1997
1998         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1999                 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
2000                 local_bh_disable();
2001                 nf_conntrack_lock(lockp);
2002                 if (*bucket < nf_conntrack_htable_size) {
2003                         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
2004                                 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
2005                                         continue;
2006                                 ct = nf_ct_tuplehash_to_ctrack(h);
2007                                 if (iter(ct, data))
2008                                         goto found;
2009                         }
2010                 }
2011                 spin_unlock(lockp);
2012                 local_bh_enable();
2013                 cond_resched();
2014         }
2015
2016         return NULL;
2017 found:
2018         atomic_inc(&ct->ct_general.use);
2019         spin_unlock(lockp);
2020         local_bh_enable();
2021         return ct;
2022 }
2023
2024 static void nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data),
2025                                   void *data, u32 portid, int report)
2026 {
2027         unsigned int bucket = 0, sequence;
2028         struct nf_conn *ct;
2029
2030         might_sleep();
2031
2032         for (;;) {
2033                 sequence = read_seqcount_begin(&nf_conntrack_generation);
2034
2035                 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
2036                         /* Time to push up daises... */
2037
2038                         nf_ct_delete(ct, portid, report);
2039                         nf_ct_put(ct);
2040                         cond_resched();
2041                 }
2042
2043                 if (!read_seqcount_retry(&nf_conntrack_generation, sequence))
2044                         break;
2045                 bucket = 0;
2046         }
2047 }
2048
2049 struct iter_data {
2050         int (*iter)(struct nf_conn *i, void *data);
2051         void *data;
2052         struct net *net;
2053 };
2054
2055 static int iter_net_only(struct nf_conn *i, void *data)
2056 {
2057         struct iter_data *d = data;
2058
2059         if (!net_eq(d->net, nf_ct_net(i)))
2060                 return 0;
2061
2062         return d->iter(i, d->data);
2063 }
2064
2065 static void
2066 __nf_ct_unconfirmed_destroy(struct net *net)
2067 {
2068         int cpu;
2069
2070         for_each_possible_cpu(cpu) {
2071                 struct nf_conntrack_tuple_hash *h;
2072                 struct hlist_nulls_node *n;
2073                 struct ct_pcpu *pcpu;
2074
2075                 pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2076
2077                 spin_lock_bh(&pcpu->lock);
2078                 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
2079                         struct nf_conn *ct;
2080
2081                         ct = nf_ct_tuplehash_to_ctrack(h);
2082
2083                         /* we cannot call iter() on unconfirmed list, the
2084                          * owning cpu can reallocate ct->ext at any time.
2085                          */
2086                         set_bit(IPS_DYING_BIT, &ct->status);
2087                 }
2088                 spin_unlock_bh(&pcpu->lock);
2089                 cond_resched();
2090         }
2091 }
2092
2093 void nf_ct_unconfirmed_destroy(struct net *net)
2094 {
2095         might_sleep();
2096
2097         if (atomic_read(&net->ct.count) > 0) {
2098                 __nf_ct_unconfirmed_destroy(net);
2099                 nf_queue_nf_hook_drop(net);
2100                 synchronize_net();
2101         }
2102 }
2103 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy);
2104
2105 void nf_ct_iterate_cleanup_net(struct net *net,
2106                                int (*iter)(struct nf_conn *i, void *data),
2107                                void *data, u32 portid, int report)
2108 {
2109         struct iter_data d;
2110
2111         might_sleep();
2112
2113         if (atomic_read(&net->ct.count) == 0)
2114                 return;
2115
2116         d.iter = iter;
2117         d.data = data;
2118         d.net = net;
2119
2120         nf_ct_iterate_cleanup(iter_net_only, &d, portid, report);
2121 }
2122 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net);
2123
2124 /**
2125  * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
2126  * @iter: callback to invoke for each conntrack
2127  * @data: data to pass to @iter
2128  *
2129  * Like nf_ct_iterate_cleanup, but first marks conntracks on the
2130  * unconfirmed list as dying (so they will not be inserted into
2131  * main table).
2132  *
2133  * Can only be called in module exit path.
2134  */
2135 void
2136 nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data)
2137 {
2138         struct net *net;
2139
2140         down_read(&net_rwsem);
2141         for_each_net(net) {
2142                 if (atomic_read(&net->ct.count) == 0)
2143                         continue;
2144                 __nf_ct_unconfirmed_destroy(net);
2145                 nf_queue_nf_hook_drop(net);
2146         }
2147         up_read(&net_rwsem);
2148
2149         /* Need to wait for netns cleanup worker to finish, if its
2150          * running -- it might have deleted a net namespace from
2151          * the global list, so our __nf_ct_unconfirmed_destroy() might
2152          * not have affected all namespaces.
2153          */
2154         net_ns_barrier();
2155
2156         /* a conntrack could have been unlinked from unconfirmed list
2157          * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
2158          * This makes sure its inserted into conntrack table.
2159          */
2160         synchronize_net();
2161
2162         nf_ct_iterate_cleanup(iter, data, 0, 0);
2163 }
2164 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy);
2165
2166 static int kill_all(struct nf_conn *i, void *data)
2167 {
2168         return net_eq(nf_ct_net(i), data);
2169 }
2170
2171 void nf_conntrack_cleanup_start(void)
2172 {
2173         conntrack_gc_work.exiting = true;
2174         RCU_INIT_POINTER(ip_ct_attach, NULL);
2175 }
2176
2177 void nf_conntrack_cleanup_end(void)
2178 {
2179         RCU_INIT_POINTER(nf_ct_hook, NULL);
2180         cancel_delayed_work_sync(&conntrack_gc_work.dwork);
2181         kvfree(nf_conntrack_hash);
2182
2183         nf_conntrack_proto_fini();
2184         nf_conntrack_seqadj_fini();
2185         nf_conntrack_labels_fini();
2186         nf_conntrack_helper_fini();
2187         nf_conntrack_timeout_fini();
2188         nf_conntrack_ecache_fini();
2189         nf_conntrack_tstamp_fini();
2190         nf_conntrack_acct_fini();
2191         nf_conntrack_expect_fini();
2192
2193         kmem_cache_destroy(nf_conntrack_cachep);
2194 }
2195
2196 /*
2197  * Mishearing the voices in his head, our hero wonders how he's
2198  * supposed to kill the mall.
2199  */
2200 void nf_conntrack_cleanup_net(struct net *net)
2201 {
2202         LIST_HEAD(single);
2203
2204         list_add(&net->exit_list, &single);
2205         nf_conntrack_cleanup_net_list(&single);
2206 }
2207
2208 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
2209 {
2210         int busy;
2211         struct net *net;
2212
2213         /*
2214          * This makes sure all current packets have passed through
2215          *  netfilter framework.  Roll on, two-stage module
2216          *  delete...
2217          */
2218         synchronize_net();
2219 i_see_dead_people:
2220         busy = 0;
2221         list_for_each_entry(net, net_exit_list, exit_list) {
2222                 nf_ct_iterate_cleanup(kill_all, net, 0, 0);
2223                 if (atomic_read(&net->ct.count) != 0)
2224                         busy = 1;
2225         }
2226         if (busy) {
2227                 schedule();
2228                 goto i_see_dead_people;
2229         }
2230
2231         list_for_each_entry(net, net_exit_list, exit_list) {
2232                 nf_conntrack_proto_pernet_fini(net);
2233                 nf_conntrack_ecache_pernet_fini(net);
2234                 nf_conntrack_expect_pernet_fini(net);
2235                 free_percpu(net->ct.stat);
2236                 free_percpu(net->ct.pcpu_lists);
2237         }
2238 }
2239
2240 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
2241 {
2242         struct hlist_nulls_head *hash;
2243         unsigned int nr_slots, i;
2244
2245         if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
2246                 return NULL;
2247
2248         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
2249         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
2250
2251         hash = kvmalloc_array(nr_slots, sizeof(struct hlist_nulls_head),
2252                               GFP_KERNEL | __GFP_ZERO);
2253
2254         if (hash && nulls)
2255                 for (i = 0; i < nr_slots; i++)
2256                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
2257
2258         return hash;
2259 }
2260 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
2261
2262 int nf_conntrack_hash_resize(unsigned int hashsize)
2263 {
2264         int i, bucket;
2265         unsigned int old_size;
2266         struct hlist_nulls_head *hash, *old_hash;
2267         struct nf_conntrack_tuple_hash *h;
2268         struct nf_conn *ct;
2269
2270         if (!hashsize)
2271                 return -EINVAL;
2272
2273         hash = nf_ct_alloc_hashtable(&hashsize, 1);
2274         if (!hash)
2275                 return -ENOMEM;
2276
2277         old_size = nf_conntrack_htable_size;
2278         if (old_size == hashsize) {
2279                 kvfree(hash);
2280                 return 0;
2281         }
2282
2283         local_bh_disable();
2284         nf_conntrack_all_lock();
2285         write_seqcount_begin(&nf_conntrack_generation);
2286
2287         /* Lookups in the old hash might happen in parallel, which means we
2288          * might get false negatives during connection lookup. New connections
2289          * created because of a false negative won't make it into the hash
2290          * though since that required taking the locks.
2291          */
2292
2293         for (i = 0; i < nf_conntrack_htable_size; i++) {
2294                 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
2295                         h = hlist_nulls_entry(nf_conntrack_hash[i].first,
2296                                               struct nf_conntrack_tuple_hash, hnnode);
2297                         ct = nf_ct_tuplehash_to_ctrack(h);
2298                         hlist_nulls_del_rcu(&h->hnnode);
2299                         bucket = __hash_conntrack(nf_ct_net(ct),
2300                                                   &h->tuple, hashsize);
2301                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
2302                 }
2303         }
2304         old_size = nf_conntrack_htable_size;
2305         old_hash = nf_conntrack_hash;
2306
2307         nf_conntrack_hash = hash;
2308         nf_conntrack_htable_size = hashsize;
2309
2310         write_seqcount_end(&nf_conntrack_generation);
2311         nf_conntrack_all_unlock();
2312         local_bh_enable();
2313
2314         synchronize_net();
2315         kvfree(old_hash);
2316         return 0;
2317 }
2318
2319 int nf_conntrack_set_hashsize(const char *val, const struct kernel_param *kp)
2320 {
2321         unsigned int hashsize;
2322         int rc;
2323
2324         if (current->nsproxy->net_ns != &init_net)
2325                 return -EOPNOTSUPP;
2326
2327         /* On boot, we can set this without any fancy locking. */
2328         if (!nf_conntrack_hash)
2329                 return param_set_uint(val, kp);
2330
2331         rc = kstrtouint(val, 0, &hashsize);
2332         if (rc)
2333                 return rc;
2334
2335         return nf_conntrack_hash_resize(hashsize);
2336 }
2337
2338 static __always_inline unsigned int total_extension_size(void)
2339 {
2340         /* remember to add new extensions below */
2341         BUILD_BUG_ON(NF_CT_EXT_NUM > 9);
2342
2343         return sizeof(struct nf_ct_ext) +
2344                sizeof(struct nf_conn_help)
2345 #if IS_ENABLED(CONFIG_NF_NAT)
2346                 + sizeof(struct nf_conn_nat)
2347 #endif
2348                 + sizeof(struct nf_conn_seqadj)
2349                 + sizeof(struct nf_conn_acct)
2350 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2351                 + sizeof(struct nf_conntrack_ecache)
2352 #endif
2353 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2354                 + sizeof(struct nf_conn_tstamp)
2355 #endif
2356 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2357                 + sizeof(struct nf_conn_timeout)
2358 #endif
2359 #ifdef CONFIG_NF_CONNTRACK_LABELS
2360                 + sizeof(struct nf_conn_labels)
2361 #endif
2362 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2363                 + sizeof(struct nf_conn_synproxy)
2364 #endif
2365         ;
2366 };
2367
2368 int nf_conntrack_init_start(void)
2369 {
2370         unsigned long nr_pages = totalram_pages();
2371         int max_factor = 8;
2372         int ret = -ENOMEM;
2373         int i;
2374
2375         /* struct nf_ct_ext uses u8 to store offsets/size */
2376         BUILD_BUG_ON(total_extension_size() > 255u);
2377
2378         seqcount_init(&nf_conntrack_generation);
2379
2380         for (i = 0; i < CONNTRACK_LOCKS; i++)
2381                 spin_lock_init(&nf_conntrack_locks[i]);
2382
2383         if (!nf_conntrack_htable_size) {
2384                 /* Idea from tcp.c: use 1/16384 of memory.
2385                  * On i386: 32MB machine has 512 buckets.
2386                  * >= 1GB machines have 16384 buckets.
2387                  * >= 4GB machines have 65536 buckets.
2388                  */
2389                 nf_conntrack_htable_size
2390                         = (((nr_pages << PAGE_SHIFT) / 16384)
2391                            / sizeof(struct hlist_head));
2392                 if (nr_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
2393                         nf_conntrack_htable_size = 65536;
2394                 else if (nr_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
2395                         nf_conntrack_htable_size = 16384;
2396                 if (nf_conntrack_htable_size < 32)
2397                         nf_conntrack_htable_size = 32;
2398
2399                 /* Use a max. factor of four by default to get the same max as
2400                  * with the old struct list_heads. When a table size is given
2401                  * we use the old value of 8 to avoid reducing the max.
2402                  * entries. */
2403                 max_factor = 4;
2404         }
2405
2406         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
2407         if (!nf_conntrack_hash)
2408                 return -ENOMEM;
2409
2410         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
2411
2412         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
2413                                                 sizeof(struct nf_conn),
2414                                                 NFCT_INFOMASK + 1,
2415                                                 SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
2416         if (!nf_conntrack_cachep)
2417                 goto err_cachep;
2418
2419         ret = nf_conntrack_expect_init();
2420         if (ret < 0)
2421                 goto err_expect;
2422
2423         ret = nf_conntrack_acct_init();
2424         if (ret < 0)
2425                 goto err_acct;
2426
2427         ret = nf_conntrack_tstamp_init();
2428         if (ret < 0)
2429                 goto err_tstamp;
2430
2431         ret = nf_conntrack_ecache_init();
2432         if (ret < 0)
2433                 goto err_ecache;
2434
2435         ret = nf_conntrack_timeout_init();
2436         if (ret < 0)
2437                 goto err_timeout;
2438
2439         ret = nf_conntrack_helper_init();
2440         if (ret < 0)
2441                 goto err_helper;
2442
2443         ret = nf_conntrack_labels_init();
2444         if (ret < 0)
2445                 goto err_labels;
2446
2447         ret = nf_conntrack_seqadj_init();
2448         if (ret < 0)
2449                 goto err_seqadj;
2450
2451         ret = nf_conntrack_proto_init();
2452         if (ret < 0)
2453                 goto err_proto;
2454
2455         conntrack_gc_work_init(&conntrack_gc_work);
2456         queue_delayed_work(system_power_efficient_wq, &conntrack_gc_work.dwork, HZ);
2457
2458         return 0;
2459
2460 err_proto:
2461         nf_conntrack_seqadj_fini();
2462 err_seqadj:
2463         nf_conntrack_labels_fini();
2464 err_labels:
2465         nf_conntrack_helper_fini();
2466 err_helper:
2467         nf_conntrack_timeout_fini();
2468 err_timeout:
2469         nf_conntrack_ecache_fini();
2470 err_ecache:
2471         nf_conntrack_tstamp_fini();
2472 err_tstamp:
2473         nf_conntrack_acct_fini();
2474 err_acct:
2475         nf_conntrack_expect_fini();
2476 err_expect:
2477         kmem_cache_destroy(nf_conntrack_cachep);
2478 err_cachep:
2479         kvfree(nf_conntrack_hash);
2480         return ret;
2481 }
2482
2483 static struct nf_ct_hook nf_conntrack_hook = {
2484         .update         = nf_conntrack_update,
2485         .destroy        = destroy_conntrack,
2486         .get_tuple_skb  = nf_conntrack_get_tuple_skb,
2487 };
2488
2489 void nf_conntrack_init_end(void)
2490 {
2491         /* For use by REJECT target */
2492         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
2493         RCU_INIT_POINTER(nf_ct_hook, &nf_conntrack_hook);
2494 }
2495
2496 /*
2497  * We need to use special "null" values, not used in hash table
2498  */
2499 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
2500 #define DYING_NULLS_VAL         ((1<<30)+1)
2501 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
2502
2503 int nf_conntrack_init_net(struct net *net)
2504 {
2505         int ret = -ENOMEM;
2506         int cpu;
2507
2508         BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER);
2509         BUILD_BUG_ON_NOT_POWER_OF_2(CONNTRACK_LOCKS);
2510         atomic_set(&net->ct.count, 0);
2511
2512         net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
2513         if (!net->ct.pcpu_lists)
2514                 goto err_stat;
2515
2516         for_each_possible_cpu(cpu) {
2517                 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2518
2519                 spin_lock_init(&pcpu->lock);
2520                 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
2521                 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
2522         }
2523
2524         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
2525         if (!net->ct.stat)
2526                 goto err_pcpu_lists;
2527
2528         ret = nf_conntrack_expect_pernet_init(net);
2529         if (ret < 0)
2530                 goto err_expect;
2531
2532         nf_conntrack_acct_pernet_init(net);
2533         nf_conntrack_tstamp_pernet_init(net);
2534         nf_conntrack_ecache_pernet_init(net);
2535         nf_conntrack_helper_pernet_init(net);
2536         nf_conntrack_proto_pernet_init(net);
2537
2538         return 0;
2539
2540 err_expect:
2541         free_percpu(net->ct.stat);
2542 err_pcpu_lists:
2543         free_percpu(net->ct.pcpu_lists);
2544 err_stat:
2545         return ret;
2546 }