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 __read_mostly spinlock_t nf_conntrack_locks_all_lock;
  77 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
  78 static __read_mostly bool nf_conntrack_locks_all;
  79
  80 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
  81 #define GC_MAX_BUCKETS_DIV      128u
  82 /* upper bound of full table scan */
  83 #define GC_MAX_SCAN_JIFFIES     (16u * HZ)
  84 /* desired ratio of entries found to be expired */
  85 #define GC_EVICT_RATIO  50u
  86
  87 static struct conntrack_gc_work conntrack_gc_work;
  88
  89 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
  90 {
  91         /* 1) Acquire the lock */
  92         spin_lock(lock);
  93
  94         /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
  95          * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
  96          */
  97         if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false))
  98                 return;
  99
 100         /* fast path failed, unlock */
 101         spin_unlock(lock);
 102
 103         /* Slow path 1) get global lock */
 104         spin_lock(&nf_conntrack_locks_all_lock);
 105
 106         /* Slow path 2) get the lock we want */
 107         spin_lock(lock);
 108
 109         /* Slow path 3) release the global lock */
 110         spin_unlock(&nf_conntrack_locks_all_lock);
 111 }
 112 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
 113
 114 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
 115 {
 116         h1 %= CONNTRACK_LOCKS;
 117         h2 %= CONNTRACK_LOCKS;
 118         spin_unlock(&nf_conntrack_locks[h1]);
 119         if (h1 != h2)
 120                 spin_unlock(&nf_conntrack_locks[h2]);
 121 }
 122
 123 /* return true if we need to recompute hashes (in case hash table was resized) */
 124 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
 125                                      unsigned int h2, unsigned int sequence)
 126 {
 127         h1 %= CONNTRACK_LOCKS;
 128         h2 %= CONNTRACK_LOCKS;
 129         if (h1 <= h2) {
 130                 nf_conntrack_lock(&nf_conntrack_locks[h1]);
 131                 if (h1 != h2)
 132                         spin_lock_nested(&nf_conntrack_locks[h2],
 133                                          SINGLE_DEPTH_NESTING);
 134         } else {
 135                 nf_conntrack_lock(&nf_conntrack_locks[h2]);
 136                 spin_lock_nested(&nf_conntrack_locks[h1],
 137                                  SINGLE_DEPTH_NESTING);
 138         }
 139         if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
 140                 nf_conntrack_double_unlock(h1, h2);
 141                 return true;
 142         }
 143         return false;
 144 }
 145
 146 static void nf_conntrack_all_lock(void)
 147 {
 148         int i;
 149
 150         spin_lock(&nf_conntrack_locks_all_lock);
 151
 152         nf_conntrack_locks_all = true;
 153
 154         for (i = 0; i < CONNTRACK_LOCKS; i++) {
 155                 spin_lock(&nf_conntrack_locks[i]);
 156
 157                 /* This spin_unlock provides the "release" to ensure that
 158                  * nf_conntrack_locks_all==true is visible to everyone that
 159                  * acquired spin_lock(&nf_conntrack_locks[]).
 160                  */
 161                 spin_unlock(&nf_conntrack_locks[i]);
 162         }
 163 }
 164
 165 static void nf_conntrack_all_unlock(void)
 166 {
 167         /* All prior stores must be complete before we clear
 168          * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
 169          * might observe the false value but not the entire
 170          * critical section.
 171          * It pairs with the smp_load_acquire() in nf_conntrack_lock()
 172          */
 173         smp_store_release(&nf_conntrack_locks_all, false);
 174         spin_unlock(&nf_conntrack_locks_all_lock);
 175 }
 176
 177 unsigned int nf_conntrack_htable_size __read_mostly;
 178 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
 179
 180 unsigned int nf_conntrack_max __read_mostly;
 181 EXPORT_SYMBOL_GPL(nf_conntrack_max);
 182 seqcount_t nf_conntrack_generation __read_mostly;
 183 static unsigned int nf_conntrack_hash_rnd __read_mostly;
 184
 185 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
 186                               const struct net *net)
 187 {
 188         unsigned int n;
 189         u32 seed;
 190
 191         get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
 192
 193         /* The direction must be ignored, so we hash everything up to the
 194          * destination ports (which is a multiple of 4) and treat the last
 195          * three bytes manually.
 196          */
 197         seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
 198         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
 199         return jhash2((u32 *)tuple, n, seed ^
 200                       (((__force __u16)tuple->dst.u.all << 16) |
 201                       tuple->dst.protonum));
 202 }
 203
 204 static u32 scale_hash(u32 hash)
 205 {
 206         return reciprocal_scale(hash, nf_conntrack_htable_size);
 207 }
 208
 209 static u32 __hash_conntrack(const struct net *net,
 210                             const struct nf_conntrack_tuple *tuple,
 211                             unsigned int size)
 212 {
 213         return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
 214 }
 215
 216 static u32 hash_conntrack(const struct net *net,
 217                           const struct nf_conntrack_tuple *tuple)
 218 {
 219         return scale_hash(hash_conntrack_raw(tuple, net));
 220 }
 221
 222 static bool nf_ct_get_tuple_ports(const struct sk_buff *skb,
 223                                   unsigned int dataoff,
 224                                   struct nf_conntrack_tuple *tuple)
 225 {       struct {
 226                 __be16 sport;
 227                 __be16 dport;
 228         } _inet_hdr, *inet_hdr;
 229
 230         /* Actually only need first 4 bytes to get ports. */
 231         inet_hdr = skb_header_pointer(skb, dataoff, sizeof(_inet_hdr), &_inet_hdr);
 232         if (!inet_hdr)
 233                 return false;
 234
 235         tuple->src.u.udp.port = inet_hdr->sport;
 236         tuple->dst.u.udp.port = inet_hdr->dport;
 237         return true;
 238 }
 239
 240 static bool
 241 nf_ct_get_tuple(const struct sk_buff *skb,
 242                 unsigned int nhoff,
 243                 unsigned int dataoff,
 244                 u_int16_t l3num,
 245                 u_int8_t protonum,
 246                 struct net *net,
 247                 struct nf_conntrack_tuple *tuple)
 248 {
 249         unsigned int size;
 250         const __be32 *ap;
 251         __be32 _addrs[8];
 252
 253         memset(tuple, 0, sizeof(*tuple));
 254
 255         tuple->src.l3num = l3num;
 256         switch (l3num) {
 257         case NFPROTO_IPV4:
 258                 nhoff += offsetof(struct iphdr, saddr);
 259                 size = 2 * sizeof(__be32);
 260                 break;
 261         case NFPROTO_IPV6:
 262                 nhoff += offsetof(struct ipv6hdr, saddr);
 263                 size = sizeof(_addrs);
 264                 break;
 265         default:
 266                 return true;
 267         }
 268
 269         ap = skb_header_pointer(skb, nhoff, size, _addrs);
 270         if (!ap)
 271                 return false;
 272
 273         switch (l3num) {
 274         case NFPROTO_IPV4:
 275                 tuple->src.u3.ip = ap[0];
 276                 tuple->dst.u3.ip = ap[1];
 277                 break;
 278         case NFPROTO_IPV6:
 279                 memcpy(tuple->src.u3.ip6, ap, sizeof(tuple->src.u3.ip6));
 280                 memcpy(tuple->dst.u3.ip6, ap + 4, sizeof(tuple->dst.u3.ip6));
 281                 break;
 282         }
 283
 284         tuple->dst.protonum = protonum;
 285         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 286
 287         switch (protonum) {
 288 #if IS_ENABLED(CONFIG_IPV6)
 289         case IPPROTO_ICMPV6:
 290                 return icmpv6_pkt_to_tuple(skb, dataoff, net, tuple);
 291 #endif
 292         case IPPROTO_ICMP:
 293                 return icmp_pkt_to_tuple(skb, dataoff, net, tuple);
 294 #ifdef CONFIG_NF_CT_PROTO_GRE
 295         case IPPROTO_GRE:
 296                 return gre_pkt_to_tuple(skb, dataoff, net, tuple);
 297 #endif
 298         case IPPROTO_TCP:
 299         case IPPROTO_UDP: /* fallthrough */
 300                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 301 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
 302         case IPPROTO_UDPLITE:
 303                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 304 #endif
 305 #ifdef CONFIG_NF_CT_PROTO_SCTP
 306         case IPPROTO_SCTP:
 307                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 308 #endif
 309 #ifdef CONFIG_NF_CT_PROTO_DCCP
 310         case IPPROTO_DCCP:
 311                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 312 #endif
 313         default:
 314                 break;
 315         }
 316
 317         return true;
 318 }
 319
 320 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
 321                             u_int8_t *protonum)
 322 {
 323         int dataoff = -1;
 324         const struct iphdr *iph;
 325         struct iphdr _iph;
 326
 327         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
 328         if (!iph)
 329                 return -1;
 330
 331         /* Conntrack defragments packets, we might still see fragments
 332          * inside ICMP packets though.
 333          */
 334         if (iph->frag_off & htons(IP_OFFSET))
 335                 return -1;
 336
 337         dataoff = nhoff + (iph->ihl << 2);
 338         *protonum = iph->protocol;
 339
 340         /* Check bogus IP headers */
 341         if (dataoff > skb->len) {
 342                 pr_debug("bogus IPv4 packet: nhoff %u, ihl %u, skblen %u\n",
 343                          nhoff, iph->ihl << 2, skb->len);
 344                 return -1;
 345         }
 346         return dataoff;
 347 }
 348
 349 #if IS_ENABLED(CONFIG_IPV6)
 350 static int ipv6_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
 351                             u8 *protonum)
 352 {
 353         int protoff = -1;
 354         unsigned int extoff = nhoff + sizeof(struct ipv6hdr);
 355         __be16 frag_off;
 356         u8 nexthdr;
 357
 358         if (skb_copy_bits(skb, nhoff + offsetof(struct ipv6hdr, nexthdr),
 359                           &nexthdr, sizeof(nexthdr)) != 0) {
 360                 pr_debug("can't get nexthdr\n");
 361                 return -1;
 362         }
 363         protoff = ipv6_skip_exthdr(skb, extoff, &nexthdr, &frag_off);
 364         /*
 365          * (protoff == skb->len) means the packet has not data, just
 366          * IPv6 and possibly extensions headers, but it is tracked anyway
 367          */
 368         if (protoff < 0 || (frag_off & htons(~0x7)) != 0) {
 369                 pr_debug("can't find proto in pkt\n");
 370                 return -1;
 371         }
 372
 373         *protonum = nexthdr;
 374         return protoff;
 375 }
 376 #endif
 377
 378 static int get_l4proto(const struct sk_buff *skb,
 379                        unsigned int nhoff, u8 pf, u8 *l4num)
 380 {
 381         switch (pf) {
 382         case NFPROTO_IPV4:
 383                 return ipv4_get_l4proto(skb, nhoff, l4num);
 384 #if IS_ENABLED(CONFIG_IPV6)
 385         case NFPROTO_IPV6:
 386                 return ipv6_get_l4proto(skb, nhoff, l4num);
 387 #endif
 388         default:
 389                 *l4num = 0;
 390                 break;
 391         }
 392         return -1;
 393 }
 394
 395 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 396                        u_int16_t l3num,
 397                        struct net *net, struct nf_conntrack_tuple *tuple)
 398 {
 399         u8 protonum;
 400         int protoff;
 401
 402         protoff = get_l4proto(skb, nhoff, l3num, &protonum);
 403         if (protoff <= 0)
 404                 return false;
 405
 406         return nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple);
 407 }
 408 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 409
 410 bool
 411 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 412                    const struct nf_conntrack_tuple *orig)
 413 {
 414         memset(inverse, 0, sizeof(*inverse));
 415
 416         inverse->src.l3num = orig->src.l3num;
 417
 418         switch (orig->src.l3num) {
 419         case NFPROTO_IPV4:
 420                 inverse->src.u3.ip = orig->dst.u3.ip;
 421                 inverse->dst.u3.ip = orig->src.u3.ip;
 422                 break;
 423         case NFPROTO_IPV6:
 424                 inverse->src.u3.in6 = orig->dst.u3.in6;
 425                 inverse->dst.u3.in6 = orig->src.u3.in6;
 426                 break;
 427         default:
 428                 break;
 429         }
 430
 431         inverse->dst.dir = !orig->dst.dir;
 432
 433         inverse->dst.protonum = orig->dst.protonum;
 434
 435         switch (orig->dst.protonum) {
 436         case IPPROTO_ICMP:
 437                 return nf_conntrack_invert_icmp_tuple(inverse, orig);
 438 #if IS_ENABLED(CONFIG_IPV6)
 439         case IPPROTO_ICMPV6:
 440                 return nf_conntrack_invert_icmpv6_tuple(inverse, orig);
 441 #endif
 442         }
 443
 444         inverse->src.u.all = orig->dst.u.all;
 445         inverse->dst.u.all = orig->src.u.all;
 446         return true;
 447 }
 448 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 449
 450 /* Generate a almost-unique pseudo-id for a given conntrack.
 451  *
 452  * intentionally doesn't re-use any of the seeds used for hash
 453  * table location, we assume id gets exposed to userspace.
 454  *
 455  * Following nf_conn items do not change throughout lifetime
 456  * of the nf_conn after it has been committed to main hash table:
 457  *
 458  * 1. nf_conn address
 459  * 2. nf_conn->ext address
 460  * 3. nf_conn->master address (normally NULL)
 461  * 4. tuple
 462  * 5. the associated net namespace
 463  */
 464 u32 nf_ct_get_id(const struct nf_conn *ct)
 465 {
 466         static __read_mostly siphash_key_t ct_id_seed;
 467         unsigned long a, b, c, d;
 468
 469         net_get_random_once(&ct_id_seed, sizeof(ct_id_seed));
 470
 471         a = (unsigned long)ct;
 472         b = (unsigned long)ct->master ^ net_hash_mix(nf_ct_net(ct));
 473         c = (unsigned long)ct->ext;
 474         d = (unsigned long)siphash(&ct->tuplehash, sizeof(ct->tuplehash),
 475                                    &ct_id_seed);
 476 #ifdef CONFIG_64BIT
 477         return siphash_4u64((u64)a, (u64)b, (u64)c, (u64)d, &ct_id_seed);
 478 #else
 479         return siphash_4u32((u32)a, (u32)b, (u32)c, (u32)d, &ct_id_seed);
 480 #endif
 481 }
 482 EXPORT_SYMBOL_GPL(nf_ct_get_id);
 483
 484 static void
 485 clean_from_lists(struct nf_conn *ct)
 486 {
 487         pr_debug("clean_from_lists(%p)\n", ct);
 488         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 489         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 490
 491         /* Destroy all pending expectations */
 492         nf_ct_remove_expectations(ct);
 493 }
 494
 495 /* must be called with local_bh_disable */
 496 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
 497 {
 498         struct ct_pcpu *pcpu;
 499
 500         /* add this conntrack to the (per cpu) dying list */
 501         ct->cpu = smp_processor_id();
 502         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 503
 504         spin_lock(&pcpu->lock);
 505         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 506                              &pcpu->dying);
 507         spin_unlock(&pcpu->lock);
 508 }
 509
 510 /* must be called with local_bh_disable */
 511 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
 512 {
 513         struct ct_pcpu *pcpu;
 514
 515         /* add this conntrack to the (per cpu) unconfirmed list */
 516         ct->cpu = smp_processor_id();
 517         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 518
 519         spin_lock(&pcpu->lock);
 520         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 521                              &pcpu->unconfirmed);
 522         spin_unlock(&pcpu->lock);
 523 }
 524
 525 /* must be called with local_bh_disable */
 526 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
 527 {
 528         struct ct_pcpu *pcpu;
 529
 530         /* We overload first tuple to link into unconfirmed or dying list.*/
 531         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 532
 533         spin_lock(&pcpu->lock);
 534         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 535         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 536         spin_unlock(&pcpu->lock);
 537 }
 538
 539 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
 540
 541 /* Released via destroy_conntrack() */
 542 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
 543                                  const struct nf_conntrack_zone *zone,
 544                                  gfp_t flags)
 545 {
 546         struct nf_conn *tmpl, *p;
 547
 548         if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) {
 549                 tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags);
 550                 if (!tmpl)
 551                         return NULL;
 552
 553                 p = tmpl;
 554                 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
 555                 if (tmpl != p) {
 556                         tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
 557                         tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p;
 558                 }
 559         } else {
 560                 tmpl = kzalloc(sizeof(*tmpl), flags);
 561                 if (!tmpl)
 562                         return NULL;
 563         }
 564
 565         tmpl->status = IPS_TEMPLATE;
 566         write_pnet(&tmpl->ct_net, net);
 567         nf_ct_zone_add(tmpl, zone);
 568         atomic_set(&tmpl->ct_general.use, 0);
 569
 570         return tmpl;
 571 }
 572 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
 573
 574 void nf_ct_tmpl_free(struct nf_conn *tmpl)
 575 {
 576         nf_ct_ext_destroy(tmpl);
 577         nf_ct_ext_free(tmpl);
 578
 579         if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK)
 580                 kfree((char *)tmpl - tmpl->proto.tmpl_padto);
 581         else
 582                 kfree(tmpl);
 583 }
 584 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
 585
 586 static void destroy_gre_conntrack(struct nf_conn *ct)
 587 {
 588 #ifdef CONFIG_NF_CT_PROTO_GRE
 589         struct nf_conn *master = ct->master;
 590
 591         if (master)
 592                 nf_ct_gre_keymap_destroy(master);
 593 #endif
 594 }
 595
 596 static void
 597 destroy_conntrack(struct nf_conntrack *nfct)
 598 {
 599         struct nf_conn *ct = (struct nf_conn *)nfct;
 600
 601         pr_debug("destroy_conntrack(%p)\n", ct);
 602         WARN_ON(atomic_read(&nfct->use) != 0);
 603
 604         if (unlikely(nf_ct_is_template(ct))) {
 605                 nf_ct_tmpl_free(ct);
 606                 return;
 607         }
 608
 609         if (unlikely(nf_ct_protonum(ct) == IPPROTO_GRE))
 610                 destroy_gre_conntrack(ct);
 611
 612         local_bh_disable();
 613         /* Expectations will have been removed in clean_from_lists,
 614          * except TFTP can create an expectation on the first packet,
 615          * before connection is in the list, so we need to clean here,
 616          * too.
 617          */
 618         nf_ct_remove_expectations(ct);
 619
 620         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 621
 622         local_bh_enable();
 623
 624         if (ct->master)
 625                 nf_ct_put(ct->master);
 626
 627         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 628         nf_conntrack_free(ct);
 629 }
 630
 631 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 632 {
 633         struct net *net = nf_ct_net(ct);
 634         unsigned int hash, reply_hash;
 635         unsigned int sequence;
 636
 637         nf_ct_helper_destroy(ct);
 638
 639         local_bh_disable();
 640         do {
 641                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 642                 hash = hash_conntrack(net,
 643                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 644                 reply_hash = hash_conntrack(net,
 645                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 646         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 647
 648         clean_from_lists(ct);
 649         nf_conntrack_double_unlock(hash, reply_hash);
 650
 651         nf_ct_add_to_dying_list(ct);
 652
 653         local_bh_enable();
 654 }
 655
 656 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 657 {
 658         struct nf_conn_tstamp *tstamp;
 659
 660         if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
 661                 return false;
 662
 663         tstamp = nf_conn_tstamp_find(ct);
 664         if (tstamp && tstamp->stop == 0)
 665                 tstamp->stop = ktime_get_real_ns();
 666
 667         if (nf_conntrack_event_report(IPCT_DESTROY, ct,
 668                                     portid, report) < 0) {
 669                 /* destroy event was not delivered. nf_ct_put will
 670                  * be done by event cache worker on redelivery.
 671                  */
 672                 nf_ct_delete_from_lists(ct);
 673                 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
 674                 return false;
 675         }
 676
 677         nf_conntrack_ecache_work(nf_ct_net(ct));
 678         nf_ct_delete_from_lists(ct);
 679         nf_ct_put(ct);
 680         return true;
 681 }
 682 EXPORT_SYMBOL_GPL(nf_ct_delete);
 683
 684 static inline bool
 685 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
 686                 const struct nf_conntrack_tuple *tuple,
 687                 const struct nf_conntrack_zone *zone,
 688                 const struct net *net)
 689 {
 690         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 691
 692         /* A conntrack can be recreated with the equal tuple,
 693          * so we need to check that the conntrack is confirmed
 694          */
 695         return nf_ct_tuple_equal(tuple, &h->tuple) &&
 696                nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
 697                nf_ct_is_confirmed(ct) &&
 698                net_eq(net, nf_ct_net(ct));
 699 }
 700
 701 static inline bool
 702 nf_ct_match(const struct nf_conn *ct1, const struct nf_conn *ct2)
 703 {
 704         return nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 705                                  &ct2->tuplehash[IP_CT_DIR_ORIGINAL].tuple) &&
 706                nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_REPLY].tuple,
 707                                  &ct2->tuplehash[IP_CT_DIR_REPLY].tuple) &&
 708                nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_ORIGINAL) &&
 709                nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_REPLY) &&
 710                net_eq(nf_ct_net(ct1), nf_ct_net(ct2));
 711 }
 712
 713 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
 714 static void nf_ct_gc_expired(struct nf_conn *ct)
 715 {
 716         if (!atomic_inc_not_zero(&ct->ct_general.use))
 717                 return;
 718
 719         if (nf_ct_should_gc(ct))
 720                 nf_ct_kill(ct);
 721
 722         nf_ct_put(ct);
 723 }
 724
 725 /*
 726  * Warning :
 727  * - Caller must take a reference on returned object
 728  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 729  */
 730 static struct nf_conntrack_tuple_hash *
 731 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
 732                       const struct nf_conntrack_tuple *tuple, u32 hash)
 733 {
 734         struct nf_conntrack_tuple_hash *h;
 735         struct hlist_nulls_head *ct_hash;
 736         struct hlist_nulls_node *n;
 737         unsigned int bucket, hsize;
 738
 739 begin:
 740         nf_conntrack_get_ht(&ct_hash, &hsize);
 741         bucket = reciprocal_scale(hash, hsize);
 742
 743         hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
 744                 struct nf_conn *ct;
 745
 746                 ct = nf_ct_tuplehash_to_ctrack(h);
 747                 if (nf_ct_is_expired(ct)) {
 748                         nf_ct_gc_expired(ct);
 749                         continue;
 750                 }
 751
 752                 if (nf_ct_key_equal(h, tuple, zone, net))
 753                         return h;
 754         }
 755         /*
 756          * if the nulls value we got at the end of this lookup is
 757          * not the expected one, we must restart lookup.
 758          * We probably met an item that was moved to another chain.
 759          */
 760         if (get_nulls_value(n) != bucket) {
 761                 NF_CT_STAT_INC_ATOMIC(net, search_restart);
 762                 goto begin;
 763         }
 764
 765         return NULL;
 766 }
 767
 768 /* Find a connection corresponding to a tuple. */
 769 static struct nf_conntrack_tuple_hash *
 770 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 771                         const struct nf_conntrack_tuple *tuple, u32 hash)
 772 {
 773         struct nf_conntrack_tuple_hash *h;
 774         struct nf_conn *ct;
 775
 776         rcu_read_lock();
 777
 778         h = ____nf_conntrack_find(net, zone, tuple, hash);
 779         if (h) {
 780                 /* We have a candidate that matches the tuple we're interested
 781                  * in, try to obtain a reference and re-check tuple
 782                  */
 783                 ct = nf_ct_tuplehash_to_ctrack(h);
 784                 if (likely(atomic_inc_not_zero(&ct->ct_general.use))) {
 785                         if (likely(nf_ct_key_equal(h, tuple, zone, net)))
 786                                 goto found;
 787
 788                         /* TYPESAFE_BY_RCU recycled the candidate */
 789                         nf_ct_put(ct);
 790                 }
 791
 792                 h = NULL;
 793         }
 794 found:
 795         rcu_read_unlock();
 796
 797         return h;
 798 }
 799
 800 struct nf_conntrack_tuple_hash *
 801 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 802                       const struct nf_conntrack_tuple *tuple)
 803 {
 804         return __nf_conntrack_find_get(net, zone, tuple,
 805                                        hash_conntrack_raw(tuple, net));
 806 }
 807 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 808
 809 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 810                                        unsigned int hash,
 811                                        unsigned int reply_hash)
 812 {
 813         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 814                            &nf_conntrack_hash[hash]);
 815         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 816                            &nf_conntrack_hash[reply_hash]);
 817 }
 818
 819 int
 820 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 821 {
 822         const struct nf_conntrack_zone *zone;
 823         struct net *net = nf_ct_net(ct);
 824         unsigned int hash, reply_hash;
 825         struct nf_conntrack_tuple_hash *h;
 826         struct hlist_nulls_node *n;
 827         unsigned int sequence;
 828
 829         zone = nf_ct_zone(ct);
 830
 831         local_bh_disable();
 832         do {
 833                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 834                 hash = hash_conntrack(net,
 835                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 836                 reply_hash = hash_conntrack(net,
 837                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 838         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 839
 840         /* See if there's one in the list already, including reverse */
 841         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
 842                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 843                                     zone, net))
 844                         goto out;
 845
 846         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
 847                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 848                                     zone, net))
 849                         goto out;
 850
 851         smp_wmb();
 852         /* The caller holds a reference to this object */
 853         atomic_set(&ct->ct_general.use, 2);
 854         __nf_conntrack_hash_insert(ct, hash, reply_hash);
 855         nf_conntrack_double_unlock(hash, reply_hash);
 856         NF_CT_STAT_INC(net, insert);
 857         local_bh_enable();
 858         return 0;
 859
 860 out:
 861         nf_conntrack_double_unlock(hash, reply_hash);
 862         NF_CT_STAT_INC(net, insert_failed);
 863         local_bh_enable();
 864         return -EEXIST;
 865 }
 866 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 867
 868 static inline void nf_ct_acct_update(struct nf_conn *ct,
 869                                      enum ip_conntrack_info ctinfo,
 870                                      unsigned int len)
 871 {
 872         struct nf_conn_acct *acct;
 873
 874         acct = nf_conn_acct_find(ct);
 875         if (acct) {
 876                 struct nf_conn_counter *counter = acct->counter;
 877
 878                 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
 879                 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
 880         }
 881 }
 882
 883 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
 884                              const struct nf_conn *loser_ct)
 885 {
 886         struct nf_conn_acct *acct;
 887
 888         acct = nf_conn_acct_find(loser_ct);
 889         if (acct) {
 890                 struct nf_conn_counter *counter = acct->counter;
 891                 unsigned int bytes;
 892
 893                 /* u32 should be fine since we must have seen one packet. */
 894                 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
 895                 nf_ct_acct_update(ct, ctinfo, bytes);
 896         }
 897 }
 898
 899 /* Resolve race on insertion if this protocol allows this. */
 900 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
 901                                enum ip_conntrack_info ctinfo,
 902                                struct nf_conntrack_tuple_hash *h)
 903 {
 904         /* This is the conntrack entry already in hashes that won race. */
 905         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 906         const struct nf_conntrack_l4proto *l4proto;
 907         enum ip_conntrack_info oldinfo;
 908         struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo);
 909
 910         l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
 911         if (l4proto->allow_clash &&
 912             !nf_ct_is_dying(ct) &&
 913             atomic_inc_not_zero(&ct->ct_general.use)) {
 914                 if (((ct->status & IPS_NAT_DONE_MASK) == 0) ||
 915                     nf_ct_match(ct, loser_ct)) {
 916                         nf_ct_acct_merge(ct, ctinfo, loser_ct);
 917                         nf_conntrack_put(&loser_ct->ct_general);
 918                         nf_ct_set(skb, ct, oldinfo);
 919                         return NF_ACCEPT;
 920                 }
 921                 nf_ct_put(ct);
 922         }
 923         NF_CT_STAT_INC(net, drop);
 924         return NF_DROP;
 925 }
 926
 927 /* Confirm a connection given skb; places it in hash table */
 928 int
 929 __nf_conntrack_confirm(struct sk_buff *skb)
 930 {
 931         const struct nf_conntrack_zone *zone;
 932         unsigned int hash, reply_hash;
 933         struct nf_conntrack_tuple_hash *h;
 934         struct nf_conn *ct;
 935         struct nf_conn_help *help;
 936         struct nf_conn_tstamp *tstamp;
 937         struct hlist_nulls_node *n;
 938         enum ip_conntrack_info ctinfo;
 939         struct net *net;
 940         unsigned int sequence;
 941         int ret = NF_DROP;
 942
 943         ct = nf_ct_get(skb, &ctinfo);
 944         net = nf_ct_net(ct);
 945
 946         /* ipt_REJECT uses nf_conntrack_attach to attach related
 947            ICMP/TCP RST packets in other direction.  Actual packet
 948            which created connection will be IP_CT_NEW or for an
 949            expected connection, IP_CT_RELATED. */
 950         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 951                 return NF_ACCEPT;
 952
 953         zone = nf_ct_zone(ct);
 954         local_bh_disable();
 955
 956         do {
 957                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 958                 /* reuse the hash saved before */
 959                 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 960                 hash = scale_hash(hash);
 961                 reply_hash = hash_conntrack(net,
 962                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 963
 964         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 965
 966         /* We're not in hash table, and we refuse to set up related
 967          * connections for unconfirmed conns.  But packet copies and
 968          * REJECT will give spurious warnings here.
 969          */
 970
 971         /* Another skb with the same unconfirmed conntrack may
 972          * win the race. This may happen for bridge(br_flood)
 973          * or broadcast/multicast packets do skb_clone with
 974          * unconfirmed conntrack.
 975          */
 976         if (unlikely(nf_ct_is_confirmed(ct))) {
 977                 WARN_ON_ONCE(1);
 978                 nf_conntrack_double_unlock(hash, reply_hash);
 979                 local_bh_enable();
 980                 return NF_DROP;
 981         }
 982
 983         pr_debug("Confirming conntrack %p\n", ct);
 984         /* We have to check the DYING flag after unlink to prevent
 985          * a race against nf_ct_get_next_corpse() possibly called from
 986          * user context, else we insert an already 'dead' hash, blocking
 987          * further use of that particular connection -JM.
 988          */
 989         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 990
 991         if (unlikely(nf_ct_is_dying(ct))) {
 992                 nf_ct_add_to_dying_list(ct);
 993                 goto dying;
 994         }
 995
 996         /* See if there's one in the list already, including reverse:
 997            NAT could have grabbed it without realizing, since we're
 998            not in the hash.  If there is, we lost race. */
 999         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
1000                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1001                                     zone, net))
1002                         goto out;
1003
1004         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
1005                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
1006                                     zone, net))
1007                         goto out;
1008
1009         /* Timer relative to confirmation time, not original
1010            setting time, otherwise we'd get timer wrap in
1011            weird delay cases. */
1012         ct->timeout += nfct_time_stamp;
1013         atomic_inc(&ct->ct_general.use);
1014         ct->status |= IPS_CONFIRMED;
1015
1016         /* set conntrack timestamp, if enabled. */
1017         tstamp = nf_conn_tstamp_find(ct);
1018         if (tstamp)
1019                 tstamp->start = ktime_get_real_ns();
1020
1021         /* Since the lookup is lockless, hash insertion must be done after
1022          * starting the timer and setting the CONFIRMED bit. The RCU barriers
1023          * guarantee that no other CPU can find the conntrack before the above
1024          * stores are visible.
1025          */
1026         __nf_conntrack_hash_insert(ct, hash, reply_hash);
1027         nf_conntrack_double_unlock(hash, reply_hash);
1028         local_bh_enable();
1029
1030         help = nfct_help(ct);
1031         if (help && help->helper)
1032                 nf_conntrack_event_cache(IPCT_HELPER, ct);
1033
1034         nf_conntrack_event_cache(master_ct(ct) ?
1035                                  IPCT_RELATED : IPCT_NEW, ct);
1036         return NF_ACCEPT;
1037
1038 out:
1039         nf_ct_add_to_dying_list(ct);
1040         ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
1041 dying:
1042         nf_conntrack_double_unlock(hash, reply_hash);
1043         NF_CT_STAT_INC(net, insert_failed);
1044         local_bh_enable();
1045         return ret;
1046 }
1047 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
1048
1049 /* Returns true if a connection correspondings to the tuple (required
1050    for NAT). */
1051 int
1052 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
1053                          const struct nf_conn *ignored_conntrack)
1054 {
1055         struct net *net = nf_ct_net(ignored_conntrack);
1056         const struct nf_conntrack_zone *zone;
1057         struct nf_conntrack_tuple_hash *h;
1058         struct hlist_nulls_head *ct_hash;
1059         unsigned int hash, hsize;
1060         struct hlist_nulls_node *n;
1061         struct nf_conn *ct;
1062
1063         zone = nf_ct_zone(ignored_conntrack);
1064
1065         rcu_read_lock();
1066  begin:
1067         nf_conntrack_get_ht(&ct_hash, &hsize);
1068         hash = __hash_conntrack(net, tuple, hsize);
1069
1070         hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
1071                 ct = nf_ct_tuplehash_to_ctrack(h);
1072
1073                 if (ct == ignored_conntrack)
1074                         continue;
1075
1076                 if (nf_ct_is_expired(ct)) {
1077                         nf_ct_gc_expired(ct);
1078                         continue;
1079                 }
1080
1081                 if (nf_ct_key_equal(h, tuple, zone, net)) {
1082                         /* Tuple is taken already, so caller will need to find
1083                          * a new source port to use.
1084                          *
1085                          * Only exception:
1086                          * If the *original tuples* are identical, then both
1087                          * conntracks refer to the same flow.
1088                          * This is a rare situation, it can occur e.g. when
1089                          * more than one UDP packet is sent from same socket
1090                          * in different threads.
1091                          *
1092                          * Let nf_ct_resolve_clash() deal with this later.
1093                          */
1094                         if (nf_ct_tuple_equal(&ignored_conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1095                                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple))
1096                                 continue;
1097
1098                         NF_CT_STAT_INC_ATOMIC(net, found);
1099                         rcu_read_unlock();
1100                         return 1;
1101                 }
1102         }
1103
1104         if (get_nulls_value(n) != hash) {
1105                 NF_CT_STAT_INC_ATOMIC(net, search_restart);
1106                 goto begin;
1107         }
1108
1109         rcu_read_unlock();
1110
1111         return 0;
1112 }
1113 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
1114
1115 #define NF_CT_EVICTION_RANGE    8
1116
1117 /* There's a small race here where we may free a just-assured
1118    connection.  Too bad: we're in trouble anyway. */
1119 static unsigned int early_drop_list(struct net *net,
1120                                     struct hlist_nulls_head *head)
1121 {
1122         struct nf_conntrack_tuple_hash *h;
1123         struct hlist_nulls_node *n;
1124         unsigned int drops = 0;
1125         struct nf_conn *tmp;
1126
1127         hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
1128                 tmp = nf_ct_tuplehash_to_ctrack(h);
1129
1130                 if (test_bit(IPS_OFFLOAD_BIT, &tmp->status))
1131                         continue;
1132
1133                 if (nf_ct_is_expired(tmp)) {
1134                         nf_ct_gc_expired(tmp);
1135                         continue;
1136                 }
1137
1138                 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
1139                     !net_eq(nf_ct_net(tmp), net) ||
1140                     nf_ct_is_dying(tmp))
1141                         continue;
1142
1143                 if (!atomic_inc_not_zero(&tmp->ct_general.use))
1144                         continue;
1145
1146                 /* kill only if still in same netns -- might have moved due to
1147                  * SLAB_TYPESAFE_BY_RCU rules.
1148                  *
1149                  * We steal the timer reference.  If that fails timer has
1150                  * already fired or someone else deleted it. Just drop ref
1151                  * and move to next entry.
1152                  */
1153                 if (net_eq(nf_ct_net(tmp), net) &&
1154                     nf_ct_is_confirmed(tmp) &&
1155                     nf_ct_delete(tmp, 0, 0))
1156                         drops++;
1157
1158                 nf_ct_put(tmp);
1159         }
1160
1161         return drops;
1162 }
1163
1164 static noinline int early_drop(struct net *net, unsigned int hash)
1165 {
1166         unsigned int i, bucket;
1167
1168         for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
1169                 struct hlist_nulls_head *ct_hash;
1170                 unsigned int hsize, drops;
1171
1172                 rcu_read_lock();
1173                 nf_conntrack_get_ht(&ct_hash, &hsize);
1174                 if (!i)
1175                         bucket = reciprocal_scale(hash, hsize);
1176                 else
1177                         bucket = (bucket + 1) % hsize;
1178
1179                 drops = early_drop_list(net, &ct_hash[bucket]);
1180                 rcu_read_unlock();
1181
1182                 if (drops) {
1183                         NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
1184                         return true;
1185                 }
1186         }
1187
1188         return false;
1189 }
1190
1191 static bool gc_worker_skip_ct(const struct nf_conn *ct)
1192 {
1193         return !nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct);
1194 }
1195
1196 static bool gc_worker_can_early_drop(const struct nf_conn *ct)
1197 {
1198         const struct nf_conntrack_l4proto *l4proto;
1199
1200         if (!test_bit(IPS_ASSURED_BIT, &ct->status))
1201                 return true;
1202
1203         l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
1204         if (l4proto->can_early_drop && l4proto->can_early_drop(ct))
1205                 return true;
1206
1207         return false;
1208 }
1209
1210 #define DAY     (86400 * HZ)
1211
1212 /* Set an arbitrary timeout large enough not to ever expire, this save
1213  * us a check for the IPS_OFFLOAD_BIT from the packet path via
1214  * nf_ct_is_expired().
1215  */
1216 static void nf_ct_offload_timeout(struct nf_conn *ct)
1217 {
1218         if (nf_ct_expires(ct) < DAY / 2)
1219                 ct->timeout = nfct_time_stamp + DAY;
1220 }
1221
1222 static void gc_worker(struct work_struct *work)
1223 {
1224         unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u);
1225         unsigned int i, goal, buckets = 0, expired_count = 0;
1226         unsigned int nf_conntrack_max95 = 0;
1227         struct conntrack_gc_work *gc_work;
1228         unsigned int ratio, scanned = 0;
1229         unsigned long next_run;
1230
1231         gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
1232
1233         goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV;
1234         i = gc_work->last_bucket;
1235         if (gc_work->early_drop)
1236                 nf_conntrack_max95 = nf_conntrack_max / 100u * 95u;
1237
1238         do {
1239                 struct nf_conntrack_tuple_hash *h;
1240                 struct hlist_nulls_head *ct_hash;
1241                 struct hlist_nulls_node *n;
1242                 unsigned int hashsz;
1243                 struct nf_conn *tmp;
1244
1245                 i++;
1246                 rcu_read_lock();
1247
1248                 nf_conntrack_get_ht(&ct_hash, &hashsz);
1249                 if (i >= hashsz)
1250                         i = 0;
1251
1252                 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
1253                         struct net *net;
1254
1255                         tmp = nf_ct_tuplehash_to_ctrack(h);
1256
1257                         scanned++;
1258                         if (test_bit(IPS_OFFLOAD_BIT, &tmp->status)) {
1259                                 nf_ct_offload_timeout(tmp);
1260                                 continue;
1261                         }
1262
1263                         if (nf_ct_is_expired(tmp)) {
1264                                 nf_ct_gc_expired(tmp);
1265                                 expired_count++;
1266                                 continue;
1267                         }
1268
1269                         if (nf_conntrack_max95 == 0 || gc_worker_skip_ct(tmp))
1270                                 continue;
1271
1272                         net = nf_ct_net(tmp);
1273                         if (atomic_read(&net->ct.count) < nf_conntrack_max95)
1274                                 continue;
1275
1276                         /* need to take reference to avoid possible races */
1277                         if (!atomic_inc_not_zero(&tmp->ct_general.use))
1278                                 continue;
1279
1280                         if (gc_worker_skip_ct(tmp)) {
1281                                 nf_ct_put(tmp);
1282                                 continue;
1283                         }
1284
1285                         if (gc_worker_can_early_drop(tmp))
1286                                 nf_ct_kill(tmp);
1287
1288                         nf_ct_put(tmp);
1289                 }
1290
1291                 /* could check get_nulls_value() here and restart if ct
1292                  * was moved to another chain.  But given gc is best-effort
1293                  * we will just continue with next hash slot.
1294                  */
1295                 rcu_read_unlock();
1296                 cond_resched();
1297         } while (++buckets < goal);
1298
1299         if (gc_work->exiting)
1300                 return;
1301
1302         /*
1303          * Eviction will normally happen from the packet path, and not
1304          * from this gc worker.
1305          *
1306          * This worker is only here to reap expired entries when system went
1307          * idle after a busy period.
1308          *
1309          * The heuristics below are supposed to balance conflicting goals:
1310          *
1311          * 1. Minimize time until we notice a stale entry
1312          * 2. Maximize scan intervals to not waste cycles
1313          *
1314          * Normally, expire ratio will be close to 0.
1315          *
1316          * As soon as a sizeable fraction of the entries have expired
1317          * increase scan frequency.
1318          */
1319         ratio = scanned ? expired_count * 100 / scanned : 0;
1320         if (ratio > GC_EVICT_RATIO) {
1321                 gc_work->next_gc_run = min_interval;
1322         } else {
1323                 unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV;
1324
1325                 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0);
1326
1327                 gc_work->next_gc_run += min_interval;
1328                 if (gc_work->next_gc_run > max)
1329                         gc_work->next_gc_run = max;
1330         }
1331
1332         next_run = gc_work->next_gc_run;
1333         gc_work->last_bucket = i;
1334         gc_work->early_drop = false;
1335         queue_delayed_work(system_power_efficient_wq, &gc_work->dwork, next_run);
1336 }
1337
1338 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1339 {
1340         INIT_DEFERRABLE_WORK(&gc_work->dwork, gc_worker);
1341         gc_work->next_gc_run = HZ;
1342         gc_work->exiting = false;
1343 }
1344
1345 static struct nf_conn *
1346 __nf_conntrack_alloc(struct net *net,
1347                      const struct nf_conntrack_zone *zone,
1348                      const struct nf_conntrack_tuple *orig,
1349                      const struct nf_conntrack_tuple *repl,
1350                      gfp_t gfp, u32 hash)
1351 {
1352         struct nf_conn *ct;
1353
1354         /* We don't want any race condition at early drop stage */
1355         atomic_inc(&net->ct.count);
1356
1357         if (nf_conntrack_max &&
1358             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
1359                 if (!early_drop(net, hash)) {
1360                         if (!conntrack_gc_work.early_drop)
1361                                 conntrack_gc_work.early_drop = true;
1362                         atomic_dec(&net->ct.count);
1363                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1364                         return ERR_PTR(-ENOMEM);
1365                 }
1366         }
1367
1368         /*
1369          * Do not use kmem_cache_zalloc(), as this cache uses
1370          * SLAB_TYPESAFE_BY_RCU.
1371          */
1372         ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1373         if (ct == NULL)
1374                 goto out;
1375
1376         spin_lock_init(&ct->lock);
1377         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1378         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1379         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1380         /* save hash for reusing when confirming */
1381         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1382         ct->status = 0;
1383         ct->timeout = 0;
1384         write_pnet(&ct->ct_net, net);
1385         memset(&ct->__nfct_init_offset[0], 0,
1386                offsetof(struct nf_conn, proto) -
1387                offsetof(struct nf_conn, __nfct_init_offset[0]));
1388
1389         nf_ct_zone_add(ct, zone);
1390
1391         /* Because we use RCU lookups, we set ct_general.use to zero before
1392          * this is inserted in any list.
1393          */
1394         atomic_set(&ct->ct_general.use, 0);
1395         return ct;
1396 out:
1397         atomic_dec(&net->ct.count);
1398         return ERR_PTR(-ENOMEM);
1399 }
1400
1401 struct nf_conn *nf_conntrack_alloc(struct net *net,
1402                                    const struct nf_conntrack_zone *zone,
1403                                    const struct nf_conntrack_tuple *orig,
1404                                    const struct nf_conntrack_tuple *repl,
1405                                    gfp_t gfp)
1406 {
1407         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1408 }
1409 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1410
1411 void nf_conntrack_free(struct nf_conn *ct)
1412 {
1413         struct net *net = nf_ct_net(ct);
1414
1415         /* A freed object has refcnt == 0, that's
1416          * the golden rule for SLAB_TYPESAFE_BY_RCU
1417          */
1418         WARN_ON(atomic_read(&ct->ct_general.use) != 0);
1419
1420         nf_ct_ext_destroy(ct);
1421         nf_ct_ext_free(ct);
1422         kmem_cache_free(nf_conntrack_cachep, ct);
1423         smp_mb__before_atomic();
1424         atomic_dec(&net->ct.count);
1425 }
1426 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1427
1428
1429 /* Allocate a new conntrack: we return -ENOMEM if classification
1430    failed due to stress.  Otherwise it really is unclassifiable. */
1431 static noinline struct nf_conntrack_tuple_hash *
1432 init_conntrack(struct net *net, struct nf_conn *tmpl,
1433                const struct nf_conntrack_tuple *tuple,
1434                struct sk_buff *skb,
1435                unsigned int dataoff, u32 hash)
1436 {
1437         struct nf_conn *ct;
1438         struct nf_conn_help *help;
1439         struct nf_conntrack_tuple repl_tuple;
1440         struct nf_conntrack_ecache *ecache;
1441         struct nf_conntrack_expect *exp = NULL;
1442         const struct nf_conntrack_zone *zone;
1443         struct nf_conn_timeout *timeout_ext;
1444         struct nf_conntrack_zone tmp;
1445
1446         if (!nf_ct_invert_tuple(&repl_tuple, tuple)) {
1447                 pr_debug("Can't invert tuple.\n");
1448                 return NULL;
1449         }
1450
1451         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1452         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1453                                   hash);
1454         if (IS_ERR(ct))
1455                 return (struct nf_conntrack_tuple_hash *)ct;
1456
1457         if (!nf_ct_add_synproxy(ct, tmpl)) {
1458                 nf_conntrack_free(ct);
1459                 return ERR_PTR(-ENOMEM);
1460         }
1461
1462         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1463
1464         if (timeout_ext)
1465                 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1466                                       GFP_ATOMIC);
1467
1468         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1469         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1470         nf_ct_labels_ext_add(ct);
1471
1472         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1473         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1474                                  ecache ? ecache->expmask : 0,
1475                              GFP_ATOMIC);
1476
1477         local_bh_disable();
1478         if (net->ct.expect_count) {
1479                 spin_lock(&nf_conntrack_expect_lock);
1480                 exp = nf_ct_find_expectation(net, zone, tuple);
1481                 if (exp) {
1482                         pr_debug("expectation arrives ct=%p exp=%p\n",
1483                                  ct, exp);
1484                         /* Welcome, Mr. Bond.  We've been expecting you... */
1485                         __set_bit(IPS_EXPECTED_BIT, &ct->status);
1486                         /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1487                         ct->master = exp->master;
1488                         if (exp->helper) {
1489                                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
1490                                 if (help)
1491                                         rcu_assign_pointer(help->helper, exp->helper);
1492                         }
1493
1494 #ifdef CONFIG_NF_CONNTRACK_MARK
1495                         ct->mark = exp->master->mark;
1496 #endif
1497 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1498                         ct->secmark = exp->master->secmark;
1499 #endif
1500                         NF_CT_STAT_INC(net, expect_new);
1501                 }
1502                 spin_unlock(&nf_conntrack_expect_lock);
1503         }
1504         if (!exp)
1505                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1506
1507         /* Now it is inserted into the unconfirmed list, bump refcount */
1508         nf_conntrack_get(&ct->ct_general);
1509         nf_ct_add_to_unconfirmed_list(ct);
1510
1511         local_bh_enable();
1512
1513         if (exp) {
1514                 if (exp->expectfn)
1515                         exp->expectfn(ct, exp);
1516                 nf_ct_expect_put(exp);
1517         }
1518
1519         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1520 }
1521
1522 /* On success, returns 0, sets skb->_nfct | ctinfo */
1523 static int
1524 resolve_normal_ct(struct nf_conn *tmpl,
1525                   struct sk_buff *skb,
1526                   unsigned int dataoff,
1527                   u_int8_t protonum,
1528                   const struct nf_hook_state *state)
1529 {
1530         const struct nf_conntrack_zone *zone;
1531         struct nf_conntrack_tuple tuple;
1532         struct nf_conntrack_tuple_hash *h;
1533         enum ip_conntrack_info ctinfo;
1534         struct nf_conntrack_zone tmp;
1535         struct nf_conn *ct;
1536         u32 hash;
1537
1538         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1539                              dataoff, state->pf, protonum, state->net,
1540                              &tuple)) {
1541                 pr_debug("Can't get tuple\n");
1542                 return 0;
1543         }
1544
1545         /* look for tuple match */
1546         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1547         hash = hash_conntrack_raw(&tuple, state->net);
1548         h = __nf_conntrack_find_get(state->net, zone, &tuple, hash);
1549         if (!h) {
1550                 h = init_conntrack(state->net, tmpl, &tuple,
1551                                    skb, dataoff, hash);
1552                 if (!h)
1553                         return 0;
1554                 if (IS_ERR(h))
1555                         return PTR_ERR(h);
1556         }
1557         ct = nf_ct_tuplehash_to_ctrack(h);
1558
1559         /* It exists; we have (non-exclusive) reference. */
1560         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1561                 ctinfo = IP_CT_ESTABLISHED_REPLY;
1562         } else {
1563                 /* Once we've had two way comms, always ESTABLISHED. */
1564                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1565                         pr_debug("normal packet for %p\n", ct);
1566                         ctinfo = IP_CT_ESTABLISHED;
1567                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1568                         pr_debug("related packet for %p\n", ct);
1569                         ctinfo = IP_CT_RELATED;
1570                 } else {
1571                         pr_debug("new packet for %p\n", ct);
1572                         ctinfo = IP_CT_NEW;
1573                 }
1574         }
1575         nf_ct_set(skb, ct, ctinfo);
1576         return 0;
1577 }
1578
1579 /*
1580  * icmp packets need special treatment to handle error messages that are
1581  * related to a connection.
1582  *
1583  * Callers need to check if skb has a conntrack assigned when this
1584  * helper returns; in such case skb belongs to an already known connection.
1585  */
1586 static unsigned int __cold
1587 nf_conntrack_handle_icmp(struct nf_conn *tmpl,
1588                          struct sk_buff *skb,
1589                          unsigned int dataoff,
1590                          u8 protonum,
1591                          const struct nf_hook_state *state)
1592 {
1593         int ret;
1594
1595         if (state->pf == NFPROTO_IPV4 && protonum == IPPROTO_ICMP)
1596                 ret = nf_conntrack_icmpv4_error(tmpl, skb, dataoff, state);
1597 #if IS_ENABLED(CONFIG_IPV6)
1598         else if (state->pf == NFPROTO_IPV6 && protonum == IPPROTO_ICMPV6)
1599                 ret = nf_conntrack_icmpv6_error(tmpl, skb, dataoff, state);
1600 #endif
1601         else
1602                 return NF_ACCEPT;
1603
1604         if (ret <= 0) {
1605                 NF_CT_STAT_INC_ATOMIC(state->net, error);
1606                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1607         }
1608
1609         return ret;
1610 }
1611
1612 static int generic_packet(struct nf_conn *ct, struct sk_buff *skb,
1613                           enum ip_conntrack_info ctinfo)
1614 {
1615         const unsigned int *timeout = nf_ct_timeout_lookup(ct);
1616
1617         if (!timeout)
1618                 timeout = &nf_generic_pernet(nf_ct_net(ct))->timeout;
1619
1620         nf_ct_refresh_acct(ct, ctinfo, skb, *timeout);
1621         return NF_ACCEPT;
1622 }
1623
1624 /* Returns verdict for packet, or -1 for invalid. */
1625 static int nf_conntrack_handle_packet(struct nf_conn *ct,
1626                                       struct sk_buff *skb,
1627                                       unsigned int dataoff,
1628                                       enum ip_conntrack_info ctinfo,
1629                                       const struct nf_hook_state *state)
1630 {
1631         switch (nf_ct_protonum(ct)) {
1632         case IPPROTO_TCP:
1633                 return nf_conntrack_tcp_packet(ct, skb, dataoff,
1634                                                ctinfo, state);
1635         case IPPROTO_UDP:
1636                 return nf_conntrack_udp_packet(ct, skb, dataoff,
1637                                                ctinfo, state);
1638         case IPPROTO_ICMP:
1639                 return nf_conntrack_icmp_packet(ct, skb, ctinfo, state);
1640 #if IS_ENABLED(CONFIG_IPV6)
1641         case IPPROTO_ICMPV6:
1642                 return nf_conntrack_icmpv6_packet(ct, skb, ctinfo, state);
1643 #endif
1644 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
1645         case IPPROTO_UDPLITE:
1646                 return nf_conntrack_udplite_packet(ct, skb, dataoff,
1647                                                    ctinfo, state);
1648 #endif
1649 #ifdef CONFIG_NF_CT_PROTO_SCTP
1650         case IPPROTO_SCTP:
1651                 return nf_conntrack_sctp_packet(ct, skb, dataoff,
1652                                                 ctinfo, state);
1653 #endif
1654 #ifdef CONFIG_NF_CT_PROTO_DCCP
1655         case IPPROTO_DCCP:
1656                 return nf_conntrack_dccp_packet(ct, skb, dataoff,
1657                                                 ctinfo, state);
1658 #endif
1659 #ifdef CONFIG_NF_CT_PROTO_GRE
1660         case IPPROTO_GRE:
1661                 return nf_conntrack_gre_packet(ct, skb, dataoff,
1662                                                ctinfo, state);
1663 #endif
1664         }
1665
1666         return generic_packet(ct, skb, ctinfo);
1667 }
1668
1669 unsigned int
1670 nf_conntrack_in(struct sk_buff *skb, const struct nf_hook_state *state)
1671 {
1672         enum ip_conntrack_info ctinfo;
1673         struct nf_conn *ct, *tmpl;
1674         u_int8_t protonum;
1675         int dataoff, ret;
1676
1677         tmpl = nf_ct_get(skb, &ctinfo);
1678         if (tmpl || ctinfo == IP_CT_UNTRACKED) {
1679                 /* Previously seen (loopback or untracked)?  Ignore. */
1680                 if ((tmpl && !nf_ct_is_template(tmpl)) ||
1681                      ctinfo == IP_CT_UNTRACKED) {
1682                         NF_CT_STAT_INC_ATOMIC(state->net, ignore);
1683                         return NF_ACCEPT;
1684                 }
1685                 skb->_nfct = 0;
1686         }
1687
1688         /* rcu_read_lock()ed by nf_hook_thresh */
1689         dataoff = get_l4proto(skb, skb_network_offset(skb), state->pf, &protonum);
1690         if (dataoff <= 0) {
1691                 pr_debug("not prepared to track yet or error occurred\n");
1692                 NF_CT_STAT_INC_ATOMIC(state->net, error);
1693                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1694                 ret = NF_ACCEPT;
1695                 goto out;
1696         }
1697
1698         if (protonum == IPPROTO_ICMP || protonum == IPPROTO_ICMPV6) {
1699                 ret = nf_conntrack_handle_icmp(tmpl, skb, dataoff,
1700                                                protonum, state);
1701                 if (ret <= 0) {
1702                         ret = -ret;
1703                         goto out;
1704                 }
1705                 /* ICMP[v6] protocol trackers may assign one conntrack. */
1706                 if (skb->_nfct)
1707                         goto out;
1708         }
1709 repeat:
1710         ret = resolve_normal_ct(tmpl, skb, dataoff,
1711                                 protonum, state);
1712         if (ret < 0) {
1713                 /* Too stressed to deal. */
1714                 NF_CT_STAT_INC_ATOMIC(state->net, drop);
1715                 ret = NF_DROP;
1716                 goto out;
1717         }
1718
1719         ct = nf_ct_get(skb, &ctinfo);
1720         if (!ct) {
1721                 /* Not valid part of a connection */
1722                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1723                 ret = NF_ACCEPT;
1724                 goto out;
1725         }
1726
1727         ret = nf_conntrack_handle_packet(ct, skb, dataoff, ctinfo, state);
1728         if (ret <= 0) {
1729                 /* Invalid: inverse of the return code tells
1730                  * the netfilter core what to do */
1731                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1732                 nf_conntrack_put(&ct->ct_general);
1733                 skb->_nfct = 0;
1734                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1735                 if (ret == -NF_DROP)
1736                         NF_CT_STAT_INC_ATOMIC(state->net, drop);
1737                 /* Special case: TCP tracker reports an attempt to reopen a
1738                  * closed/aborted connection. We have to go back and create a
1739                  * fresh conntrack.
1740                  */
1741                 if (ret == -NF_REPEAT)
1742                         goto repeat;
1743                 ret = -ret;
1744                 goto out;
1745         }
1746
1747         if (ctinfo == IP_CT_ESTABLISHED_REPLY &&
1748             !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1749                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1750 out:
1751         if (tmpl)
1752                 nf_ct_put(tmpl);
1753
1754         return ret;
1755 }
1756 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1757
1758 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1759    implicitly racy: see __nf_conntrack_confirm */
1760 void nf_conntrack_alter_reply(struct nf_conn *ct,
1761                               const struct nf_conntrack_tuple *newreply)
1762 {
1763         struct nf_conn_help *help = nfct_help(ct);
1764
1765         /* Should be unconfirmed, so not in hash table yet */
1766         WARN_ON(nf_ct_is_confirmed(ct));
1767
1768         pr_debug("Altering reply tuple of %p to ", ct);
1769         nf_ct_dump_tuple(newreply);
1770
1771         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1772         if (ct->master || (help && !hlist_empty(&help->expectations)))
1773                 return;
1774
1775         rcu_read_lock();
1776         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1777         rcu_read_unlock();
1778 }
1779 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1780
1781 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1782 void __nf_ct_refresh_acct(struct nf_conn *ct,
1783                           enum ip_conntrack_info ctinfo,
1784                           const struct sk_buff *skb,
1785                           u32 extra_jiffies,
1786                           bool do_acct)
1787 {
1788         /* Only update if this is not a fixed timeout */
1789         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1790                 goto acct;
1791
1792         /* If not in hash table, timer will not be active yet */
1793         if (nf_ct_is_confirmed(ct))
1794                 extra_jiffies += nfct_time_stamp;
1795
1796         if (ct->timeout != extra_jiffies)
1797                 ct->timeout = extra_jiffies;
1798 acct:
1799         if (do_acct)
1800                 nf_ct_acct_update(ct, ctinfo, skb->len);
1801 }
1802 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1803
1804 bool nf_ct_kill_acct(struct nf_conn *ct,
1805                      enum ip_conntrack_info ctinfo,
1806                      const struct sk_buff *skb)
1807 {
1808         nf_ct_acct_update(ct, ctinfo, skb->len);
1809
1810         return nf_ct_delete(ct, 0, 0);
1811 }
1812 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1813
1814 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1815
1816 #include <linux/netfilter/nfnetlink.h>
1817 #include <linux/netfilter/nfnetlink_conntrack.h>
1818 #include <linux/mutex.h>
1819
1820 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1821  * in ip_conntrack_core, since we don't want the protocols to autoload
1822  * or depend on ctnetlink */
1823 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1824                                const struct nf_conntrack_tuple *tuple)
1825 {
1826         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1827             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1828                 goto nla_put_failure;
1829         return 0;
1830
1831 nla_put_failure:
1832         return -1;
1833 }
1834 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1835
1836 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1837         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1838         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1839 };
1840 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1841
1842 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1843                                struct nf_conntrack_tuple *t)
1844 {
1845         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1846                 return -EINVAL;
1847
1848         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1849         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1850
1851         return 0;
1852 }
1853 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1854
1855 unsigned int nf_ct_port_nlattr_tuple_size(void)
1856 {
1857         static unsigned int size __read_mostly;
1858
1859         if (!size)
1860                 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1861
1862         return size;
1863 }
1864 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1865 #endif
1866
1867 /* Used by ipt_REJECT and ip6t_REJECT. */
1868 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1869 {
1870         struct nf_conn *ct;
1871         enum ip_conntrack_info ctinfo;
1872
1873         /* This ICMP is in reverse direction to the packet which caused it */
1874         ct = nf_ct_get(skb, &ctinfo);
1875         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1876                 ctinfo = IP_CT_RELATED_REPLY;
1877         else
1878                 ctinfo = IP_CT_RELATED;
1879
1880         /* Attach to new skbuff, and increment count */
1881         nf_ct_set(nskb, ct, ctinfo);
1882         nf_conntrack_get(skb_nfct(nskb));
1883 }
1884
1885 static int nf_conntrack_update(struct net *net, struct sk_buff *skb)
1886 {
1887         struct nf_conntrack_tuple_hash *h;
1888         struct nf_conntrack_tuple tuple;
1889         enum ip_conntrack_info ctinfo;
1890         struct nf_nat_hook *nat_hook;
1891         unsigned int status;
1892         struct nf_conn *ct;
1893         int dataoff;
1894         u16 l3num;
1895         u8 l4num;
1896
1897         ct = nf_ct_get(skb, &ctinfo);
1898         if (!ct || nf_ct_is_confirmed(ct))
1899                 return 0;
1900
1901         l3num = nf_ct_l3num(ct);
1902
1903         dataoff = get_l4proto(skb, skb_network_offset(skb), l3num, &l4num);
1904         if (dataoff <= 0)
1905                 return -1;
1906
1907         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
1908                              l4num, net, &tuple))
1909                 return -1;
1910
1911         if (ct->status & IPS_SRC_NAT) {
1912                 memcpy(tuple.src.u3.all,
1913                        ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.all,
1914                        sizeof(tuple.src.u3.all));
1915                 tuple.src.u.all =
1916                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.all;
1917         }
1918
1919         if (ct->status & IPS_DST_NAT) {
1920                 memcpy(tuple.dst.u3.all,
1921                        ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.all,
1922                        sizeof(tuple.dst.u3.all));
1923                 tuple.dst.u.all =
1924                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u.all;
1925         }
1926
1927         h = nf_conntrack_find_get(net, nf_ct_zone(ct), &tuple);
1928         if (!h)
1929                 return 0;
1930
1931         /* Store status bits of the conntrack that is clashing to re-do NAT
1932          * mangling according to what it has been done already to this packet.
1933          */
1934         status = ct->status;
1935
1936         nf_ct_put(ct);
1937         ct = nf_ct_tuplehash_to_ctrack(h);
1938         nf_ct_set(skb, ct, ctinfo);
1939
1940         nat_hook = rcu_dereference(nf_nat_hook);
1941         if (!nat_hook)
1942                 return 0;
1943
1944         if (status & IPS_SRC_NAT &&
1945             nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_SRC,
1946                                 IP_CT_DIR_ORIGINAL) == NF_DROP)
1947                 return -1;
1948
1949         if (status & IPS_DST_NAT &&
1950             nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_DST,
1951                                 IP_CT_DIR_ORIGINAL) == NF_DROP)
1952                 return -1;
1953
1954         return 0;
1955 }
1956
1957 static bool nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
1958                                        const struct sk_buff *skb)
1959 {
1960         const struct nf_conntrack_tuple *src_tuple;
1961         const struct nf_conntrack_tuple_hash *hash;
1962         struct nf_conntrack_tuple srctuple;
1963         enum ip_conntrack_info ctinfo;
1964         struct nf_conn *ct;
1965
1966         ct = nf_ct_get(skb, &ctinfo);
1967         if (ct) {
1968                 src_tuple = nf_ct_tuple(ct, CTINFO2DIR(ctinfo));
1969                 memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
1970                 return true;
1971         }
1972
1973         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
1974                                NFPROTO_IPV4, dev_net(skb->dev),
1975                                &srctuple))
1976                 return false;
1977
1978         hash = nf_conntrack_find_get(dev_net(skb->dev),
1979                                      &nf_ct_zone_dflt,
1980                                      &srctuple);
1981         if (!hash)
1982                 return false;
1983
1984         ct = nf_ct_tuplehash_to_ctrack(hash);
1985         src_tuple = nf_ct_tuple(ct, !hash->tuple.dst.dir);
1986         memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
1987         nf_ct_put(ct);
1988
1989         return true;
1990 }
1991
1992 /* Bring out ya dead! */
1993 static struct nf_conn *
1994 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
1995                 void *data, unsigned int *bucket)
1996 {
1997         struct nf_conntrack_tuple_hash *h;
1998         struct nf_conn *ct;
1999         struct hlist_nulls_node *n;
2000         spinlock_t *lockp;
2001
2002         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
2003                 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
2004                 local_bh_disable();
2005                 nf_conntrack_lock(lockp);
2006                 if (*bucket < nf_conntrack_htable_size) {
2007                         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
2008                                 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
2009                                         continue;
2010                                 ct = nf_ct_tuplehash_to_ctrack(h);
2011                                 if (iter(ct, data))
2012                                         goto found;
2013                         }
2014                 }
2015                 spin_unlock(lockp);
2016                 local_bh_enable();
2017                 cond_resched();
2018         }
2019
2020         return NULL;
2021 found:
2022         atomic_inc(&ct->ct_general.use);
2023         spin_unlock(lockp);
2024         local_bh_enable();
2025         return ct;
2026 }
2027
2028 static void nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data),
2029                                   void *data, u32 portid, int report)
2030 {
2031         unsigned int bucket = 0, sequence;
2032         struct nf_conn *ct;
2033
2034         might_sleep();
2035
2036         for (;;) {
2037                 sequence = read_seqcount_begin(&nf_conntrack_generation);
2038
2039                 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
2040                         /* Time to push up daises... */
2041
2042                         nf_ct_delete(ct, portid, report);
2043                         nf_ct_put(ct);
2044                         cond_resched();
2045                 }
2046
2047                 if (!read_seqcount_retry(&nf_conntrack_generation, sequence))
2048                         break;
2049                 bucket = 0;
2050         }
2051 }
2052
2053 struct iter_data {
2054         int (*iter)(struct nf_conn *i, void *data);
2055         void *data;
2056         struct net *net;
2057 };
2058
2059 static int iter_net_only(struct nf_conn *i, void *data)
2060 {
2061         struct iter_data *d = data;
2062
2063         if (!net_eq(d->net, nf_ct_net(i)))
2064                 return 0;
2065
2066         return d->iter(i, d->data);
2067 }
2068
2069 static void
2070 __nf_ct_unconfirmed_destroy(struct net *net)
2071 {
2072         int cpu;
2073
2074         for_each_possible_cpu(cpu) {
2075                 struct nf_conntrack_tuple_hash *h;
2076                 struct hlist_nulls_node *n;
2077                 struct ct_pcpu *pcpu;
2078
2079                 pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2080
2081                 spin_lock_bh(&pcpu->lock);
2082                 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
2083                         struct nf_conn *ct;
2084
2085                         ct = nf_ct_tuplehash_to_ctrack(h);
2086
2087                         /* we cannot call iter() on unconfirmed list, the
2088                          * owning cpu can reallocate ct->ext at any time.
2089                          */
2090                         set_bit(IPS_DYING_BIT, &ct->status);
2091                 }
2092                 spin_unlock_bh(&pcpu->lock);
2093                 cond_resched();
2094         }
2095 }
2096
2097 void nf_ct_unconfirmed_destroy(struct net *net)
2098 {
2099         might_sleep();
2100
2101         if (atomic_read(&net->ct.count) > 0) {
2102                 __nf_ct_unconfirmed_destroy(net);
2103                 nf_queue_nf_hook_drop(net);
2104                 synchronize_net();
2105         }
2106 }
2107 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy);
2108
2109 void nf_ct_iterate_cleanup_net(struct net *net,
2110                                int (*iter)(struct nf_conn *i, void *data),
2111                                void *data, u32 portid, int report)
2112 {
2113         struct iter_data d;
2114
2115         might_sleep();
2116
2117         if (atomic_read(&net->ct.count) == 0)
2118                 return;
2119
2120         d.iter = iter;
2121         d.data = data;
2122         d.net = net;
2123
2124         nf_ct_iterate_cleanup(iter_net_only, &d, portid, report);
2125 }
2126 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net);
2127
2128 /**
2129  * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
2130  * @iter: callback to invoke for each conntrack
2131  * @data: data to pass to @iter
2132  *
2133  * Like nf_ct_iterate_cleanup, but first marks conntracks on the
2134  * unconfirmed list as dying (so they will not be inserted into
2135  * main table).
2136  *
2137  * Can only be called in module exit path.
2138  */
2139 void
2140 nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data)
2141 {
2142         struct net *net;
2143
2144         down_read(&net_rwsem);
2145         for_each_net(net) {
2146                 if (atomic_read(&net->ct.count) == 0)
2147                         continue;
2148                 __nf_ct_unconfirmed_destroy(net);
2149                 nf_queue_nf_hook_drop(net);
2150         }
2151         up_read(&net_rwsem);
2152
2153         /* Need to wait for netns cleanup worker to finish, if its
2154          * running -- it might have deleted a net namespace from
2155          * the global list, so our __nf_ct_unconfirmed_destroy() might
2156          * not have affected all namespaces.
2157          */
2158         net_ns_barrier();
2159
2160         /* a conntrack could have been unlinked from unconfirmed list
2161          * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
2162          * This makes sure its inserted into conntrack table.
2163          */
2164         synchronize_net();
2165
2166         nf_ct_iterate_cleanup(iter, data, 0, 0);
2167 }
2168 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy);
2169
2170 static int kill_all(struct nf_conn *i, void *data)
2171 {
2172         return net_eq(nf_ct_net(i), data);
2173 }
2174
2175 void nf_conntrack_cleanup_start(void)
2176 {
2177         conntrack_gc_work.exiting = true;
2178         RCU_INIT_POINTER(ip_ct_attach, NULL);
2179 }
2180
2181 void nf_conntrack_cleanup_end(void)
2182 {
2183         RCU_INIT_POINTER(nf_ct_hook, NULL);
2184         cancel_delayed_work_sync(&conntrack_gc_work.dwork);
2185         kvfree(nf_conntrack_hash);
2186
2187         nf_conntrack_proto_fini();
2188         nf_conntrack_seqadj_fini();
2189         nf_conntrack_labels_fini();
2190         nf_conntrack_helper_fini();
2191         nf_conntrack_timeout_fini();
2192         nf_conntrack_ecache_fini();
2193         nf_conntrack_tstamp_fini();
2194         nf_conntrack_acct_fini();
2195         nf_conntrack_expect_fini();
2196
2197         kmem_cache_destroy(nf_conntrack_cachep);
2198 }
2199
2200 /*
2201  * Mishearing the voices in his head, our hero wonders how he's
2202  * supposed to kill the mall.
2203  */
2204 void nf_conntrack_cleanup_net(struct net *net)
2205 {
2206         LIST_HEAD(single);
2207
2208         list_add(&net->exit_list, &single);
2209         nf_conntrack_cleanup_net_list(&single);
2210 }
2211
2212 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
2213 {
2214         int busy;
2215         struct net *net;
2216
2217         /*
2218          * This makes sure all current packets have passed through
2219          *  netfilter framework.  Roll on, two-stage module
2220          *  delete...
2221          */
2222         synchronize_net();
2223 i_see_dead_people:
2224         busy = 0;
2225         list_for_each_entry(net, net_exit_list, exit_list) {
2226                 nf_ct_iterate_cleanup(kill_all, net, 0, 0);
2227                 if (atomic_read(&net->ct.count) != 0)
2228                         busy = 1;
2229         }
2230         if (busy) {
2231                 schedule();
2232                 goto i_see_dead_people;
2233         }
2234
2235         list_for_each_entry(net, net_exit_list, exit_list) {
2236                 nf_conntrack_proto_pernet_fini(net);
2237                 nf_conntrack_ecache_pernet_fini(net);
2238                 nf_conntrack_expect_pernet_fini(net);
2239                 free_percpu(net->ct.stat);
2240                 free_percpu(net->ct.pcpu_lists);
2241         }
2242 }
2243
2244 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
2245 {
2246         struct hlist_nulls_head *hash;
2247         unsigned int nr_slots, i;
2248
2249         if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
2250                 return NULL;
2251
2252         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
2253         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
2254
2255         hash = kvmalloc_array(nr_slots, sizeof(struct hlist_nulls_head),
2256                               GFP_KERNEL | __GFP_ZERO);
2257
2258         if (hash && nulls)
2259                 for (i = 0; i < nr_slots; i++)
2260                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
2261
2262         return hash;
2263 }
2264 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
2265
2266 int nf_conntrack_hash_resize(unsigned int hashsize)
2267 {
2268         int i, bucket;
2269         unsigned int old_size;
2270         struct hlist_nulls_head *hash, *old_hash;
2271         struct nf_conntrack_tuple_hash *h;
2272         struct nf_conn *ct;
2273
2274         if (!hashsize)
2275                 return -EINVAL;
2276
2277         hash = nf_ct_alloc_hashtable(&hashsize, 1);
2278         if (!hash)
2279                 return -ENOMEM;
2280
2281         old_size = nf_conntrack_htable_size;
2282         if (old_size == hashsize) {
2283                 kvfree(hash);
2284                 return 0;
2285         }
2286
2287         local_bh_disable();
2288         nf_conntrack_all_lock();
2289         write_seqcount_begin(&nf_conntrack_generation);
2290
2291         /* Lookups in the old hash might happen in parallel, which means we
2292          * might get false negatives during connection lookup. New connections
2293          * created because of a false negative won't make it into the hash
2294          * though since that required taking the locks.
2295          */
2296
2297         for (i = 0; i < nf_conntrack_htable_size; i++) {
2298                 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
2299                         h = hlist_nulls_entry(nf_conntrack_hash[i].first,
2300                                               struct nf_conntrack_tuple_hash, hnnode);
2301                         ct = nf_ct_tuplehash_to_ctrack(h);
2302                         hlist_nulls_del_rcu(&h->hnnode);
2303                         bucket = __hash_conntrack(nf_ct_net(ct),
2304                                                   &h->tuple, hashsize);
2305                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
2306                 }
2307         }
2308         old_size = nf_conntrack_htable_size;
2309         old_hash = nf_conntrack_hash;
2310
2311         nf_conntrack_hash = hash;
2312         nf_conntrack_htable_size = hashsize;
2313
2314         write_seqcount_end(&nf_conntrack_generation);
2315         nf_conntrack_all_unlock();
2316         local_bh_enable();
2317
2318         synchronize_net();
2319         kvfree(old_hash);
2320         return 0;
2321 }
2322
2323 int nf_conntrack_set_hashsize(const char *val, const struct kernel_param *kp)
2324 {
2325         unsigned int hashsize;
2326         int rc;
2327
2328         if (current->nsproxy->net_ns != &init_net)
2329                 return -EOPNOTSUPP;
2330
2331         /* On boot, we can set this without any fancy locking. */
2332         if (!nf_conntrack_hash)
2333                 return param_set_uint(val, kp);
2334
2335         rc = kstrtouint(val, 0, &hashsize);
2336         if (rc)
2337                 return rc;
2338
2339         return nf_conntrack_hash_resize(hashsize);
2340 }
2341 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
2342
2343 static __always_inline unsigned int total_extension_size(void)
2344 {
2345         /* remember to add new extensions below */
2346         BUILD_BUG_ON(NF_CT_EXT_NUM > 9);
2347
2348         return sizeof(struct nf_ct_ext) +
2349                sizeof(struct nf_conn_help)
2350 #if IS_ENABLED(CONFIG_NF_NAT)
2351                 + sizeof(struct nf_conn_nat)
2352 #endif
2353                 + sizeof(struct nf_conn_seqadj)
2354                 + sizeof(struct nf_conn_acct)
2355 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2356                 + sizeof(struct nf_conntrack_ecache)
2357 #endif
2358 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2359                 + sizeof(struct nf_conn_tstamp)
2360 #endif
2361 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2362                 + sizeof(struct nf_conn_timeout)
2363 #endif
2364 #ifdef CONFIG_NF_CONNTRACK_LABELS
2365                 + sizeof(struct nf_conn_labels)
2366 #endif
2367 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2368                 + sizeof(struct nf_conn_synproxy)
2369 #endif
2370         ;
2371 };
2372
2373 int nf_conntrack_init_start(void)
2374 {
2375         unsigned long nr_pages = totalram_pages();
2376         int max_factor = 8;
2377         int ret = -ENOMEM;
2378         int i;
2379
2380         /* struct nf_ct_ext uses u8 to store offsets/size */
2381         BUILD_BUG_ON(total_extension_size() > 255u);
2382
2383         seqcount_init(&nf_conntrack_generation);
2384
2385         for (i = 0; i < CONNTRACK_LOCKS; i++)
2386                 spin_lock_init(&nf_conntrack_locks[i]);
2387
2388         if (!nf_conntrack_htable_size) {
2389                 /* Idea from tcp.c: use 1/16384 of memory.
2390                  * On i386: 32MB machine has 512 buckets.
2391                  * >= 1GB machines have 16384 buckets.
2392                  * >= 4GB machines have 65536 buckets.
2393                  */
2394                 nf_conntrack_htable_size
2395                         = (((nr_pages << PAGE_SHIFT) / 16384)
2396                            / sizeof(struct hlist_head));
2397                 if (nr_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
2398                         nf_conntrack_htable_size = 65536;
2399                 else if (nr_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
2400                         nf_conntrack_htable_size = 16384;
2401                 if (nf_conntrack_htable_size < 32)
2402                         nf_conntrack_htable_size = 32;
2403
2404                 /* Use a max. factor of four by default to get the same max as
2405                  * with the old struct list_heads. When a table size is given
2406                  * we use the old value of 8 to avoid reducing the max.
2407                  * entries. */
2408                 max_factor = 4;
2409         }
2410
2411         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
2412         if (!nf_conntrack_hash)
2413                 return -ENOMEM;
2414
2415         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
2416
2417         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
2418                                                 sizeof(struct nf_conn),
2419                                                 NFCT_INFOMASK + 1,
2420                                                 SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
2421         if (!nf_conntrack_cachep)
2422                 goto err_cachep;
2423
2424         ret = nf_conntrack_expect_init();
2425         if (ret < 0)
2426                 goto err_expect;
2427
2428         ret = nf_conntrack_acct_init();
2429         if (ret < 0)
2430                 goto err_acct;
2431
2432         ret = nf_conntrack_tstamp_init();
2433         if (ret < 0)
2434                 goto err_tstamp;
2435
2436         ret = nf_conntrack_ecache_init();
2437         if (ret < 0)
2438                 goto err_ecache;
2439
2440         ret = nf_conntrack_timeout_init();
2441         if (ret < 0)
2442                 goto err_timeout;
2443
2444         ret = nf_conntrack_helper_init();
2445         if (ret < 0)
2446                 goto err_helper;
2447
2448         ret = nf_conntrack_labels_init();
2449         if (ret < 0)
2450                 goto err_labels;
2451
2452         ret = nf_conntrack_seqadj_init();
2453         if (ret < 0)
2454                 goto err_seqadj;
2455
2456         ret = nf_conntrack_proto_init();
2457         if (ret < 0)
2458                 goto err_proto;
2459
2460         conntrack_gc_work_init(&conntrack_gc_work);
2461         queue_delayed_work(system_power_efficient_wq, &conntrack_gc_work.dwork, HZ);
2462
2463         return 0;
2464
2465 err_proto:
2466         nf_conntrack_seqadj_fini();
2467 err_seqadj:
2468         nf_conntrack_labels_fini();
2469 err_labels:
2470         nf_conntrack_helper_fini();
2471 err_helper:
2472         nf_conntrack_timeout_fini();
2473 err_timeout:
2474         nf_conntrack_ecache_fini();
2475 err_ecache:
2476         nf_conntrack_tstamp_fini();
2477 err_tstamp:
2478         nf_conntrack_acct_fini();
2479 err_acct:
2480         nf_conntrack_expect_fini();
2481 err_expect:
2482         kmem_cache_destroy(nf_conntrack_cachep);
2483 err_cachep:
2484         kvfree(nf_conntrack_hash);
2485         return ret;
2486 }
2487
2488 static struct nf_ct_hook nf_conntrack_hook = {
2489         .update         = nf_conntrack_update,
2490         .destroy        = destroy_conntrack,
2491         .get_tuple_skb  = nf_conntrack_get_tuple_skb,
2492 };
2493
2494 void nf_conntrack_init_end(void)
2495 {
2496         /* For use by REJECT target */
2497         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
2498         RCU_INIT_POINTER(nf_ct_hook, &nf_conntrack_hook);
2499 }
2500
2501 /*
2502  * We need to use special "null" values, not used in hash table
2503  */
2504 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
2505 #define DYING_NULLS_VAL         ((1<<30)+1)
2506 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
2507
2508 int nf_conntrack_init_net(struct net *net)
2509 {
2510         int ret = -ENOMEM;
2511         int cpu;
2512
2513         BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER);
2514         BUILD_BUG_ON_NOT_POWER_OF_2(CONNTRACK_LOCKS);
2515         atomic_set(&net->ct.count, 0);
2516
2517         net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
2518         if (!net->ct.pcpu_lists)
2519                 goto err_stat;
2520
2521         for_each_possible_cpu(cpu) {
2522                 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2523
2524                 spin_lock_init(&pcpu->lock);
2525                 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
2526                 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
2527         }
2528
2529         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
2530         if (!net->ct.stat)
2531                 goto err_pcpu_lists;
2532
2533         ret = nf_conntrack_expect_pernet_init(net);
2534         if (ret < 0)
2535                 goto err_expect;
2536
2537         nf_conntrack_acct_pernet_init(net);
2538         nf_conntrack_tstamp_pernet_init(net);
2539         nf_conntrack_ecache_pernet_init(net);
2540         nf_conntrack_helper_pernet_init(net);
2541         nf_conntrack_proto_pernet_init(net);
2542
2543         return 0;
2544
2545 err_expect:
2546         free_percpu(net->ct.stat);
2547 err_pcpu_lists:
2548         free_percpu(net->ct.pcpu_lists);
2549 err_stat:
2550         return ret;
2551 }