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                     next_bucket;
  70         bool                    exiting;
  71         bool                    early_drop;
  72 };
  73
  74 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
  75 static DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
  76 static __read_mostly bool nf_conntrack_locks_all;
  77
  78 #define GC_SCAN_INTERVAL        (120u * HZ)
  79 #define GC_SCAN_MAX_DURATION    msecs_to_jiffies(10)
  80
  81 static struct conntrack_gc_work conntrack_gc_work;
  82
  83 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
  84 {
  85         /* 1) Acquire the lock */
  86         spin_lock(lock);
  87
  88         /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
  89          * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
  90          */
  91         if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false))
  92                 return;
  93
  94         /* fast path failed, unlock */
  95         spin_unlock(lock);
  96
  97         /* Slow path 1) get global lock */
  98         spin_lock(&nf_conntrack_locks_all_lock);
  99
 100         /* Slow path 2) get the lock we want */
 101         spin_lock(lock);
 102
 103         /* Slow path 3) release the global lock */
 104         spin_unlock(&nf_conntrack_locks_all_lock);
 105 }
 106 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
 107
 108 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
 109 {
 110         h1 %= CONNTRACK_LOCKS;
 111         h2 %= CONNTRACK_LOCKS;
 112         spin_unlock(&nf_conntrack_locks[h1]);
 113         if (h1 != h2)
 114                 spin_unlock(&nf_conntrack_locks[h2]);
 115 }
 116
 117 /* return true if we need to recompute hashes (in case hash table was resized) */
 118 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
 119                                      unsigned int h2, unsigned int sequence)
 120 {
 121         h1 %= CONNTRACK_LOCKS;
 122         h2 %= CONNTRACK_LOCKS;
 123         if (h1 <= h2) {
 124                 nf_conntrack_lock(&nf_conntrack_locks[h1]);
 125                 if (h1 != h2)
 126                         spin_lock_nested(&nf_conntrack_locks[h2],
 127                                          SINGLE_DEPTH_NESTING);
 128         } else {
 129                 nf_conntrack_lock(&nf_conntrack_locks[h2]);
 130                 spin_lock_nested(&nf_conntrack_locks[h1],
 131                                  SINGLE_DEPTH_NESTING);
 132         }
 133         if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
 134                 nf_conntrack_double_unlock(h1, h2);
 135                 return true;
 136         }
 137         return false;
 138 }
 139
 140 static void nf_conntrack_all_lock(void)
 141         __acquires(&nf_conntrack_locks_all_lock)
 142 {
 143         int i;
 144
 145         spin_lock(&nf_conntrack_locks_all_lock);
 146
 147         /* For nf_contrack_locks_all, only the latest time when another
 148          * CPU will see an update is controlled, by the "release" of the
 149          * spin_lock below.
 150          * The earliest time is not controlled, an thus KCSAN could detect
 151          * a race when nf_conntract_lock() reads the variable.
 152          * WRITE_ONCE() is used to ensure the compiler will not
 153          * optimize the write.
 154          */
 155         WRITE_ONCE(nf_conntrack_locks_all, true);
 156
 157         for (i = 0; i < CONNTRACK_LOCKS; i++) {
 158                 spin_lock(&nf_conntrack_locks[i]);
 159
 160                 /* This spin_unlock provides the "release" to ensure that
 161                  * nf_conntrack_locks_all==true is visible to everyone that
 162                  * acquired spin_lock(&nf_conntrack_locks[]).
 163                  */
 164                 spin_unlock(&nf_conntrack_locks[i]);
 165         }
 166 }
 167
 168 static void nf_conntrack_all_unlock(void)
 169         __releases(&nf_conntrack_locks_all_lock)
 170 {
 171         /* All prior stores must be complete before we clear
 172          * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
 173          * might observe the false value but not the entire
 174          * critical section.
 175          * It pairs with the smp_load_acquire() in nf_conntrack_lock()
 176          */
 177         smp_store_release(&nf_conntrack_locks_all, false);
 178         spin_unlock(&nf_conntrack_locks_all_lock);
 179 }
 180
 181 unsigned int nf_conntrack_htable_size __read_mostly;
 182 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
 183
 184 unsigned int nf_conntrack_max __read_mostly;
 185 EXPORT_SYMBOL_GPL(nf_conntrack_max);
 186 seqcount_spinlock_t nf_conntrack_generation __read_mostly;
 187 static unsigned int nf_conntrack_hash_rnd __read_mostly;
 188
 189 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
 190                               const struct net *net)
 191 {
 192         unsigned int n;
 193         u32 seed;
 194
 195         get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
 196
 197         /* The direction must be ignored, so we hash everything up to the
 198          * destination ports (which is a multiple of 4) and treat the last
 199          * three bytes manually.
 200          */
 201         seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
 202         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
 203         return jhash2((u32 *)tuple, n, seed ^
 204                       (((__force __u16)tuple->dst.u.all << 16) |
 205                       tuple->dst.protonum));
 206 }
 207
 208 static u32 scale_hash(u32 hash)
 209 {
 210         return reciprocal_scale(hash, nf_conntrack_htable_size);
 211 }
 212
 213 static u32 __hash_conntrack(const struct net *net,
 214                             const struct nf_conntrack_tuple *tuple,
 215                             unsigned int size)
 216 {
 217         return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
 218 }
 219
 220 static u32 hash_conntrack(const struct net *net,
 221                           const struct nf_conntrack_tuple *tuple)
 222 {
 223         return scale_hash(hash_conntrack_raw(tuple, net));
 224 }
 225
 226 static bool nf_ct_get_tuple_ports(const struct sk_buff *skb,
 227                                   unsigned int dataoff,
 228                                   struct nf_conntrack_tuple *tuple)
 229 {       struct {
 230                 __be16 sport;
 231                 __be16 dport;
 232         } _inet_hdr, *inet_hdr;
 233
 234         /* Actually only need first 4 bytes to get ports. */
 235         inet_hdr = skb_header_pointer(skb, dataoff, sizeof(_inet_hdr), &_inet_hdr);
 236         if (!inet_hdr)
 237                 return false;
 238
 239         tuple->src.u.udp.port = inet_hdr->sport;
 240         tuple->dst.u.udp.port = inet_hdr->dport;
 241         return true;
 242 }
 243
 244 static bool
 245 nf_ct_get_tuple(const struct sk_buff *skb,
 246                 unsigned int nhoff,
 247                 unsigned int dataoff,
 248                 u_int16_t l3num,
 249                 u_int8_t protonum,
 250                 struct net *net,
 251                 struct nf_conntrack_tuple *tuple)
 252 {
 253         unsigned int size;
 254         const __be32 *ap;
 255         __be32 _addrs[8];
 256
 257         memset(tuple, 0, sizeof(*tuple));
 258
 259         tuple->src.l3num = l3num;
 260         switch (l3num) {
 261         case NFPROTO_IPV4:
 262                 nhoff += offsetof(struct iphdr, saddr);
 263                 size = 2 * sizeof(__be32);
 264                 break;
 265         case NFPROTO_IPV6:
 266                 nhoff += offsetof(struct ipv6hdr, saddr);
 267                 size = sizeof(_addrs);
 268                 break;
 269         default:
 270                 return true;
 271         }
 272
 273         ap = skb_header_pointer(skb, nhoff, size, _addrs);
 274         if (!ap)
 275                 return false;
 276
 277         switch (l3num) {
 278         case NFPROTO_IPV4:
 279                 tuple->src.u3.ip = ap[0];
 280                 tuple->dst.u3.ip = ap[1];
 281                 break;
 282         case NFPROTO_IPV6:
 283                 memcpy(tuple->src.u3.ip6, ap, sizeof(tuple->src.u3.ip6));
 284                 memcpy(tuple->dst.u3.ip6, ap + 4, sizeof(tuple->dst.u3.ip6));
 285                 break;
 286         }
 287
 288         tuple->dst.protonum = protonum;
 289         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 290
 291         switch (protonum) {
 292 #if IS_ENABLED(CONFIG_IPV6)
 293         case IPPROTO_ICMPV6:
 294                 return icmpv6_pkt_to_tuple(skb, dataoff, net, tuple);
 295 #endif
 296         case IPPROTO_ICMP:
 297                 return icmp_pkt_to_tuple(skb, dataoff, net, tuple);
 298 #ifdef CONFIG_NF_CT_PROTO_GRE
 299         case IPPROTO_GRE:
 300                 return gre_pkt_to_tuple(skb, dataoff, net, tuple);
 301 #endif
 302         case IPPROTO_TCP:
 303         case IPPROTO_UDP: /* fallthrough */
 304                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 305 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
 306         case IPPROTO_UDPLITE:
 307                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 308 #endif
 309 #ifdef CONFIG_NF_CT_PROTO_SCTP
 310         case IPPROTO_SCTP:
 311                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 312 #endif
 313 #ifdef CONFIG_NF_CT_PROTO_DCCP
 314         case IPPROTO_DCCP:
 315                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 316 #endif
 317         default:
 318                 break;
 319         }
 320
 321         return true;
 322 }
 323
 324 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
 325                             u_int8_t *protonum)
 326 {
 327         int dataoff = -1;
 328         const struct iphdr *iph;
 329         struct iphdr _iph;
 330
 331         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
 332         if (!iph)
 333                 return -1;
 334
 335         /* Conntrack defragments packets, we might still see fragments
 336          * inside ICMP packets though.
 337          */
 338         if (iph->frag_off & htons(IP_OFFSET))
 339                 return -1;
 340
 341         dataoff = nhoff + (iph->ihl << 2);
 342         *protonum = iph->protocol;
 343
 344         /* Check bogus IP headers */
 345         if (dataoff > skb->len) {
 346                 pr_debug("bogus IPv4 packet: nhoff %u, ihl %u, skblen %u\n",
 347                          nhoff, iph->ihl << 2, skb->len);
 348                 return -1;
 349         }
 350         return dataoff;
 351 }
 352
 353 #if IS_ENABLED(CONFIG_IPV6)
 354 static int ipv6_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
 355                             u8 *protonum)
 356 {
 357         int protoff = -1;
 358         unsigned int extoff = nhoff + sizeof(struct ipv6hdr);
 359         __be16 frag_off;
 360         u8 nexthdr;
 361
 362         if (skb_copy_bits(skb, nhoff + offsetof(struct ipv6hdr, nexthdr),
 363                           &nexthdr, sizeof(nexthdr)) != 0) {
 364                 pr_debug("can't get nexthdr\n");
 365                 return -1;
 366         }
 367         protoff = ipv6_skip_exthdr(skb, extoff, &nexthdr, &frag_off);
 368         /*
 369          * (protoff == skb->len) means the packet has not data, just
 370          * IPv6 and possibly extensions headers, but it is tracked anyway
 371          */
 372         if (protoff < 0 || (frag_off & htons(~0x7)) != 0) {
 373                 pr_debug("can't find proto in pkt\n");
 374                 return -1;
 375         }
 376
 377         *protonum = nexthdr;
 378         return protoff;
 379 }
 380 #endif
 381
 382 static int get_l4proto(const struct sk_buff *skb,
 383                        unsigned int nhoff, u8 pf, u8 *l4num)
 384 {
 385         switch (pf) {
 386         case NFPROTO_IPV4:
 387                 return ipv4_get_l4proto(skb, nhoff, l4num);
 388 #if IS_ENABLED(CONFIG_IPV6)
 389         case NFPROTO_IPV6:
 390                 return ipv6_get_l4proto(skb, nhoff, l4num);
 391 #endif
 392         default:
 393                 *l4num = 0;
 394                 break;
 395         }
 396         return -1;
 397 }
 398
 399 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 400                        u_int16_t l3num,
 401                        struct net *net, struct nf_conntrack_tuple *tuple)
 402 {
 403         u8 protonum;
 404         int protoff;
 405
 406         protoff = get_l4proto(skb, nhoff, l3num, &protonum);
 407         if (protoff <= 0)
 408                 return false;
 409
 410         return nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple);
 411 }
 412 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 413
 414 bool
 415 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 416                    const struct nf_conntrack_tuple *orig)
 417 {
 418         memset(inverse, 0, sizeof(*inverse));
 419
 420         inverse->src.l3num = orig->src.l3num;
 421
 422         switch (orig->src.l3num) {
 423         case NFPROTO_IPV4:
 424                 inverse->src.u3.ip = orig->dst.u3.ip;
 425                 inverse->dst.u3.ip = orig->src.u3.ip;
 426                 break;
 427         case NFPROTO_IPV6:
 428                 inverse->src.u3.in6 = orig->dst.u3.in6;
 429                 inverse->dst.u3.in6 = orig->src.u3.in6;
 430                 break;
 431         default:
 432                 break;
 433         }
 434
 435         inverse->dst.dir = !orig->dst.dir;
 436
 437         inverse->dst.protonum = orig->dst.protonum;
 438
 439         switch (orig->dst.protonum) {
 440         case IPPROTO_ICMP:
 441                 return nf_conntrack_invert_icmp_tuple(inverse, orig);
 442 #if IS_ENABLED(CONFIG_IPV6)
 443         case IPPROTO_ICMPV6:
 444                 return nf_conntrack_invert_icmpv6_tuple(inverse, orig);
 445 #endif
 446         }
 447
 448         inverse->src.u.all = orig->dst.u.all;
 449         inverse->dst.u.all = orig->src.u.all;
 450         return true;
 451 }
 452 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 453
 454 /* Generate a almost-unique pseudo-id for a given conntrack.
 455  *
 456  * intentionally doesn't re-use any of the seeds used for hash
 457  * table location, we assume id gets exposed to userspace.
 458  *
 459  * Following nf_conn items do not change throughout lifetime
 460  * of the nf_conn:
 461  *
 462  * 1. nf_conn address
 463  * 2. nf_conn->master address (normally NULL)
 464  * 3. the associated net namespace
 465  * 4. the original direction tuple
 466  */
 467 u32 nf_ct_get_id(const struct nf_conn *ct)
 468 {
 469         static __read_mostly siphash_key_t ct_id_seed;
 470         unsigned long a, b, c, d;
 471
 472         net_get_random_once(&ct_id_seed, sizeof(ct_id_seed));
 473
 474         a = (unsigned long)ct;
 475         b = (unsigned long)ct->master;
 476         c = (unsigned long)nf_ct_net(ct);
 477         d = (unsigned long)siphash(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 478                                    sizeof(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple),
 479                                    &ct_id_seed);
 480 #ifdef CONFIG_64BIT
 481         return siphash_4u64((u64)a, (u64)b, (u64)c, (u64)d, &ct_id_seed);
 482 #else
 483         return siphash_4u32((u32)a, (u32)b, (u32)c, (u32)d, &ct_id_seed);
 484 #endif
 485 }
 486 EXPORT_SYMBOL_GPL(nf_ct_get_id);
 487
 488 static void
 489 clean_from_lists(struct nf_conn *ct)
 490 {
 491         pr_debug("clean_from_lists(%p)\n", ct);
 492         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 493         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 494
 495         /* Destroy all pending expectations */
 496         nf_ct_remove_expectations(ct);
 497 }
 498
 499 /* must be called with local_bh_disable */
 500 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
 501 {
 502         struct ct_pcpu *pcpu;
 503
 504         /* add this conntrack to the (per cpu) dying list */
 505         ct->cpu = smp_processor_id();
 506         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 507
 508         spin_lock(&pcpu->lock);
 509         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 510                              &pcpu->dying);
 511         spin_unlock(&pcpu->lock);
 512 }
 513
 514 /* must be called with local_bh_disable */
 515 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
 516 {
 517         struct ct_pcpu *pcpu;
 518
 519         /* add this conntrack to the (per cpu) unconfirmed list */
 520         ct->cpu = smp_processor_id();
 521         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 522
 523         spin_lock(&pcpu->lock);
 524         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 525                              &pcpu->unconfirmed);
 526         spin_unlock(&pcpu->lock);
 527 }
 528
 529 /* must be called with local_bh_disable */
 530 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
 531 {
 532         struct ct_pcpu *pcpu;
 533
 534         /* We overload first tuple to link into unconfirmed or dying list.*/
 535         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 536
 537         spin_lock(&pcpu->lock);
 538         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 539         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 540         spin_unlock(&pcpu->lock);
 541 }
 542
 543 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
 544
 545 /* Released via destroy_conntrack() */
 546 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
 547                                  const struct nf_conntrack_zone *zone,
 548                                  gfp_t flags)
 549 {
 550         struct nf_conn *tmpl, *p;
 551
 552         if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) {
 553                 tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags);
 554                 if (!tmpl)
 555                         return NULL;
 556
 557                 p = tmpl;
 558                 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
 559                 if (tmpl != p) {
 560                         tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
 561                         tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p;
 562                 }
 563         } else {
 564                 tmpl = kzalloc(sizeof(*tmpl), flags);
 565                 if (!tmpl)
 566                         return NULL;
 567         }
 568
 569         tmpl->status = IPS_TEMPLATE;
 570         write_pnet(&tmpl->ct_net, net);
 571         nf_ct_zone_add(tmpl, zone);
 572         atomic_set(&tmpl->ct_general.use, 0);
 573
 574         return tmpl;
 575 }
 576 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
 577
 578 void nf_ct_tmpl_free(struct nf_conn *tmpl)
 579 {
 580         nf_ct_ext_destroy(tmpl);
 581
 582         if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK)
 583                 kfree((char *)tmpl - tmpl->proto.tmpl_padto);
 584         else
 585                 kfree(tmpl);
 586 }
 587 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
 588
 589 static void destroy_gre_conntrack(struct nf_conn *ct)
 590 {
 591 #ifdef CONFIG_NF_CT_PROTO_GRE
 592         struct nf_conn *master = ct->master;
 593
 594         if (master)
 595                 nf_ct_gre_keymap_destroy(master);
 596 #endif
 597 }
 598
 599 static void
 600 destroy_conntrack(struct nf_conntrack *nfct)
 601 {
 602         struct nf_conn *ct = (struct nf_conn *)nfct;
 603
 604         pr_debug("destroy_conntrack(%p)\n", ct);
 605         WARN_ON(atomic_read(&nfct->use) != 0);
 606
 607         if (unlikely(nf_ct_is_template(ct))) {
 608                 nf_ct_tmpl_free(ct);
 609                 return;
 610         }
 611
 612         if (unlikely(nf_ct_protonum(ct) == IPPROTO_GRE))
 613                 destroy_gre_conntrack(ct);
 614
 615         local_bh_disable();
 616         /* Expectations will have been removed in clean_from_lists,
 617          * except TFTP can create an expectation on the first packet,
 618          * before connection is in the list, so we need to clean here,
 619          * too.
 620          */
 621         nf_ct_remove_expectations(ct);
 622
 623         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 624
 625         local_bh_enable();
 626
 627         if (ct->master)
 628                 nf_ct_put(ct->master);
 629
 630         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 631         nf_conntrack_free(ct);
 632 }
 633
 634 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 635 {
 636         struct net *net = nf_ct_net(ct);
 637         unsigned int hash, reply_hash;
 638         unsigned int sequence;
 639
 640         nf_ct_helper_destroy(ct);
 641
 642         local_bh_disable();
 643         do {
 644                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 645                 hash = hash_conntrack(net,
 646                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 647                 reply_hash = hash_conntrack(net,
 648                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 649         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 650
 651         clean_from_lists(ct);
 652         nf_conntrack_double_unlock(hash, reply_hash);
 653
 654         nf_ct_add_to_dying_list(ct);
 655
 656         local_bh_enable();
 657 }
 658
 659 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 660 {
 661         struct nf_conn_tstamp *tstamp;
 662         struct net *net;
 663
 664         if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
 665                 return false;
 666
 667         tstamp = nf_conn_tstamp_find(ct);
 668         if (tstamp) {
 669                 s32 timeout = ct->timeout - nfct_time_stamp;
 670
 671                 tstamp->stop = ktime_get_real_ns();
 672                 if (timeout < 0)
 673                         tstamp->stop -= jiffies_to_nsecs(-timeout);
 674         }
 675
 676         if (nf_conntrack_event_report(IPCT_DESTROY, ct,
 677                                     portid, report) < 0) {
 678                 /* destroy event was not delivered. nf_ct_put will
 679                  * be done by event cache worker on redelivery.
 680                  */
 681                 nf_ct_delete_from_lists(ct);
 682                 nf_conntrack_ecache_work(nf_ct_net(ct), NFCT_ECACHE_DESTROY_FAIL);
 683                 return false;
 684         }
 685
 686         net = nf_ct_net(ct);
 687         if (nf_conntrack_ecache_dwork_pending(net))
 688                 nf_conntrack_ecache_work(net, NFCT_ECACHE_DESTROY_SENT);
 689         nf_ct_delete_from_lists(ct);
 690         nf_ct_put(ct);
 691         return true;
 692 }
 693 EXPORT_SYMBOL_GPL(nf_ct_delete);
 694
 695 static inline bool
 696 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
 697                 const struct nf_conntrack_tuple *tuple,
 698                 const struct nf_conntrack_zone *zone,
 699                 const struct net *net)
 700 {
 701         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 702
 703         /* A conntrack can be recreated with the equal tuple,
 704          * so we need to check that the conntrack is confirmed
 705          */
 706         return nf_ct_tuple_equal(tuple, &h->tuple) &&
 707                nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
 708                nf_ct_is_confirmed(ct) &&
 709                net_eq(net, nf_ct_net(ct));
 710 }
 711
 712 static inline bool
 713 nf_ct_match(const struct nf_conn *ct1, const struct nf_conn *ct2)
 714 {
 715         return nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 716                                  &ct2->tuplehash[IP_CT_DIR_ORIGINAL].tuple) &&
 717                nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_REPLY].tuple,
 718                                  &ct2->tuplehash[IP_CT_DIR_REPLY].tuple) &&
 719                nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_ORIGINAL) &&
 720                nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_REPLY) &&
 721                net_eq(nf_ct_net(ct1), nf_ct_net(ct2));
 722 }
 723
 724 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
 725 static void nf_ct_gc_expired(struct nf_conn *ct)
 726 {
 727         if (!atomic_inc_not_zero(&ct->ct_general.use))
 728                 return;
 729
 730         if (nf_ct_should_gc(ct))
 731                 nf_ct_kill(ct);
 732
 733         nf_ct_put(ct);
 734 }
 735
 736 /*
 737  * Warning :
 738  * - Caller must take a reference on returned object
 739  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 740  */
 741 static struct nf_conntrack_tuple_hash *
 742 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
 743                       const struct nf_conntrack_tuple *tuple, u32 hash)
 744 {
 745         struct nf_conntrack_tuple_hash *h;
 746         struct hlist_nulls_head *ct_hash;
 747         struct hlist_nulls_node *n;
 748         unsigned int bucket, hsize;
 749
 750 begin:
 751         nf_conntrack_get_ht(&ct_hash, &hsize);
 752         bucket = reciprocal_scale(hash, hsize);
 753
 754         hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
 755                 struct nf_conn *ct;
 756
 757                 ct = nf_ct_tuplehash_to_ctrack(h);
 758                 if (nf_ct_is_expired(ct)) {
 759                         nf_ct_gc_expired(ct);
 760                         continue;
 761                 }
 762
 763                 if (nf_ct_key_equal(h, tuple, zone, net))
 764                         return h;
 765         }
 766         /*
 767          * if the nulls value we got at the end of this lookup is
 768          * not the expected one, we must restart lookup.
 769          * We probably met an item that was moved to another chain.
 770          */
 771         if (get_nulls_value(n) != bucket) {
 772                 NF_CT_STAT_INC_ATOMIC(net, search_restart);
 773                 goto begin;
 774         }
 775
 776         return NULL;
 777 }
 778
 779 /* Find a connection corresponding to a tuple. */
 780 static struct nf_conntrack_tuple_hash *
 781 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 782                         const struct nf_conntrack_tuple *tuple, u32 hash)
 783 {
 784         struct nf_conntrack_tuple_hash *h;
 785         struct nf_conn *ct;
 786
 787         rcu_read_lock();
 788
 789         h = ____nf_conntrack_find(net, zone, tuple, hash);
 790         if (h) {
 791                 /* We have a candidate that matches the tuple we're interested
 792                  * in, try to obtain a reference and re-check tuple
 793                  */
 794                 ct = nf_ct_tuplehash_to_ctrack(h);
 795                 if (likely(atomic_inc_not_zero(&ct->ct_general.use))) {
 796                         if (likely(nf_ct_key_equal(h, tuple, zone, net)))
 797                                 goto found;
 798
 799                         /* TYPESAFE_BY_RCU recycled the candidate */
 800                         nf_ct_put(ct);
 801                 }
 802
 803                 h = NULL;
 804         }
 805 found:
 806         rcu_read_unlock();
 807
 808         return h;
 809 }
 810
 811 struct nf_conntrack_tuple_hash *
 812 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 813                       const struct nf_conntrack_tuple *tuple)
 814 {
 815         return __nf_conntrack_find_get(net, zone, tuple,
 816                                        hash_conntrack_raw(tuple, net));
 817 }
 818 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 819
 820 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 821                                        unsigned int hash,
 822                                        unsigned int reply_hash)
 823 {
 824         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 825                            &nf_conntrack_hash[hash]);
 826         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 827                            &nf_conntrack_hash[reply_hash]);
 828 }
 829
 830 int
 831 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 832 {
 833         const struct nf_conntrack_zone *zone;
 834         struct net *net = nf_ct_net(ct);
 835         unsigned int hash, reply_hash;
 836         struct nf_conntrack_tuple_hash *h;
 837         struct hlist_nulls_node *n;
 838         unsigned int sequence;
 839
 840         zone = nf_ct_zone(ct);
 841
 842         local_bh_disable();
 843         do {
 844                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 845                 hash = hash_conntrack(net,
 846                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 847                 reply_hash = hash_conntrack(net,
 848                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 849         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 850
 851         /* See if there's one in the list already, including reverse */
 852         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
 853                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 854                                     zone, net))
 855                         goto out;
 856
 857         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
 858                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 859                                     zone, net))
 860                         goto out;
 861
 862         smp_wmb();
 863         /* The caller holds a reference to this object */
 864         atomic_set(&ct->ct_general.use, 2);
 865         __nf_conntrack_hash_insert(ct, hash, reply_hash);
 866         nf_conntrack_double_unlock(hash, reply_hash);
 867         NF_CT_STAT_INC(net, insert);
 868         local_bh_enable();
 869         return 0;
 870
 871 out:
 872         nf_conntrack_double_unlock(hash, reply_hash);
 873         local_bh_enable();
 874         return -EEXIST;
 875 }
 876 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 877
 878 void nf_ct_acct_add(struct nf_conn *ct, u32 dir, unsigned int packets,
 879                     unsigned int bytes)
 880 {
 881         struct nf_conn_acct *acct;
 882
 883         acct = nf_conn_acct_find(ct);
 884         if (acct) {
 885                 struct nf_conn_counter *counter = acct->counter;
 886
 887                 atomic64_add(packets, &counter[dir].packets);
 888                 atomic64_add(bytes, &counter[dir].bytes);
 889         }
 890 }
 891 EXPORT_SYMBOL_GPL(nf_ct_acct_add);
 892
 893 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
 894                              const struct nf_conn *loser_ct)
 895 {
 896         struct nf_conn_acct *acct;
 897
 898         acct = nf_conn_acct_find(loser_ct);
 899         if (acct) {
 900                 struct nf_conn_counter *counter = acct->counter;
 901                 unsigned int bytes;
 902
 903                 /* u32 should be fine since we must have seen one packet. */
 904                 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
 905                 nf_ct_acct_update(ct, CTINFO2DIR(ctinfo), bytes);
 906         }
 907 }
 908
 909 static void __nf_conntrack_insert_prepare(struct nf_conn *ct)
 910 {
 911         struct nf_conn_tstamp *tstamp;
 912
 913         atomic_inc(&ct->ct_general.use);
 914         ct->status |= IPS_CONFIRMED;
 915
 916         /* set conntrack timestamp, if enabled. */
 917         tstamp = nf_conn_tstamp_find(ct);
 918         if (tstamp)
 919                 tstamp->start = ktime_get_real_ns();
 920 }
 921
 922 /* caller must hold locks to prevent concurrent changes */
 923 static int __nf_ct_resolve_clash(struct sk_buff *skb,
 924                                  struct nf_conntrack_tuple_hash *h)
 925 {
 926         /* This is the conntrack entry already in hashes that won race. */
 927         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 928         enum ip_conntrack_info ctinfo;
 929         struct nf_conn *loser_ct;
 930
 931         loser_ct = nf_ct_get(skb, &ctinfo);
 932
 933         if (nf_ct_is_dying(ct))
 934                 return NF_DROP;
 935
 936         if (((ct->status & IPS_NAT_DONE_MASK) == 0) ||
 937             nf_ct_match(ct, loser_ct)) {
 938                 struct net *net = nf_ct_net(ct);
 939
 940                 nf_conntrack_get(&ct->ct_general);
 941
 942                 nf_ct_acct_merge(ct, ctinfo, loser_ct);
 943                 nf_ct_add_to_dying_list(loser_ct);
 944                 nf_conntrack_put(&loser_ct->ct_general);
 945                 nf_ct_set(skb, ct, ctinfo);
 946
 947                 NF_CT_STAT_INC(net, clash_resolve);
 948                 return NF_ACCEPT;
 949         }
 950
 951         return NF_DROP;
 952 }
 953
 954 /**
 955  * nf_ct_resolve_clash_harder - attempt to insert clashing conntrack entry
 956  *
 957  * @skb: skb that causes the collision
 958  * @repl_idx: hash slot for reply direction
 959  *
 960  * Called when origin or reply direction had a clash.
 961  * The skb can be handled without packet drop provided the reply direction
 962  * is unique or there the existing entry has the identical tuple in both
 963  * directions.
 964  *
 965  * Caller must hold conntrack table locks to prevent concurrent updates.
 966  *
 967  * Returns NF_DROP if the clash could not be handled.
 968  */
 969 static int nf_ct_resolve_clash_harder(struct sk_buff *skb, u32 repl_idx)
 970 {
 971         struct nf_conn *loser_ct = (struct nf_conn *)skb_nfct(skb);
 972         const struct nf_conntrack_zone *zone;
 973         struct nf_conntrack_tuple_hash *h;
 974         struct hlist_nulls_node *n;
 975         struct net *net;
 976
 977         zone = nf_ct_zone(loser_ct);
 978         net = nf_ct_net(loser_ct);
 979
 980         /* Reply direction must never result in a clash, unless both origin
 981          * and reply tuples are identical.
 982          */
 983         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[repl_idx], hnnode) {
 984                 if (nf_ct_key_equal(h,
 985                                     &loser_ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 986                                     zone, net))
 987                         return __nf_ct_resolve_clash(skb, h);
 988         }
 989
 990         /* We want the clashing entry to go away real soon: 1 second timeout. */
 991         loser_ct->timeout = nfct_time_stamp + HZ;
 992
 993         /* IPS_NAT_CLASH removes the entry automatically on the first
 994          * reply.  Also prevents UDP tracker from moving the entry to
 995          * ASSURED state, i.e. the entry can always be evicted under
 996          * pressure.
 997          */
 998         loser_ct->status |= IPS_FIXED_TIMEOUT | IPS_NAT_CLASH;
 999
1000         __nf_conntrack_insert_prepare(loser_ct);
1001
1002         /* fake add for ORIGINAL dir: we want lookups to only find the entry
1003          * already in the table.  This also hides the clashing entry from
1004          * ctnetlink iteration, i.e. conntrack -L won't show them.
1005          */
1006         hlist_nulls_add_fake(&loser_ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
1007
1008         hlist_nulls_add_head_rcu(&loser_ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
1009                                  &nf_conntrack_hash[repl_idx]);
1010
1011         NF_CT_STAT_INC(net, clash_resolve);
1012         return NF_ACCEPT;
1013 }
1014
1015 /**
1016  * nf_ct_resolve_clash - attempt to handle clash without packet drop
1017  *
1018  * @skb: skb that causes the clash
1019  * @h: tuplehash of the clashing entry already in table
1020  * @reply_hash: hash slot for reply direction
1021  *
1022  * A conntrack entry can be inserted to the connection tracking table
1023  * if there is no existing entry with an identical tuple.
1024  *
1025  * If there is one, @skb (and the assocated, unconfirmed conntrack) has
1026  * to be dropped.  In case @skb is retransmitted, next conntrack lookup
1027  * will find the already-existing entry.
1028  *
1029  * The major problem with such packet drop is the extra delay added by
1030  * the packet loss -- it will take some time for a retransmit to occur
1031  * (or the sender to time out when waiting for a reply).
1032  *
1033  * This function attempts to handle the situation without packet drop.
1034  *
1035  * If @skb has no NAT transformation or if the colliding entries are
1036  * exactly the same, only the to-be-confirmed conntrack entry is discarded
1037  * and @skb is associated with the conntrack entry already in the table.
1038  *
1039  * Failing that, the new, unconfirmed conntrack is still added to the table
1040  * provided that the collision only occurs in the ORIGINAL direction.
1041  * The new entry will be added only in the non-clashing REPLY direction,
1042  * so packets in the ORIGINAL direction will continue to match the existing
1043  * entry.  The new entry will also have a fixed timeout so it expires --
1044  * due to the collision, it will only see reply traffic.
1045  *
1046  * Returns NF_DROP if the clash could not be resolved.
1047  */
1048 static __cold noinline int
1049 nf_ct_resolve_clash(struct sk_buff *skb, struct nf_conntrack_tuple_hash *h,
1050                     u32 reply_hash)
1051 {
1052         /* This is the conntrack entry already in hashes that won race. */
1053         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
1054         const struct nf_conntrack_l4proto *l4proto;
1055         enum ip_conntrack_info ctinfo;
1056         struct nf_conn *loser_ct;
1057         struct net *net;
1058         int ret;
1059
1060         loser_ct = nf_ct_get(skb, &ctinfo);
1061         net = nf_ct_net(loser_ct);
1062
1063         l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
1064         if (!l4proto->allow_clash)
1065                 goto drop;
1066
1067         ret = __nf_ct_resolve_clash(skb, h);
1068         if (ret == NF_ACCEPT)
1069                 return ret;
1070
1071         ret = nf_ct_resolve_clash_harder(skb, reply_hash);
1072         if (ret == NF_ACCEPT)
1073                 return ret;
1074
1075 drop:
1076         nf_ct_add_to_dying_list(loser_ct);
1077         NF_CT_STAT_INC(net, drop);
1078         NF_CT_STAT_INC(net, insert_failed);
1079         return NF_DROP;
1080 }
1081
1082 /* Confirm a connection given skb; places it in hash table */
1083 int
1084 __nf_conntrack_confirm(struct sk_buff *skb)
1085 {
1086         const struct nf_conntrack_zone *zone;
1087         unsigned int hash, reply_hash;
1088         struct nf_conntrack_tuple_hash *h;
1089         struct nf_conn *ct;
1090         struct nf_conn_help *help;
1091         struct hlist_nulls_node *n;
1092         enum ip_conntrack_info ctinfo;
1093         struct net *net;
1094         unsigned int sequence;
1095         int ret = NF_DROP;
1096
1097         ct = nf_ct_get(skb, &ctinfo);
1098         net = nf_ct_net(ct);
1099
1100         /* ipt_REJECT uses nf_conntrack_attach to attach related
1101            ICMP/TCP RST packets in other direction.  Actual packet
1102            which created connection will be IP_CT_NEW or for an
1103            expected connection, IP_CT_RELATED. */
1104         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
1105                 return NF_ACCEPT;
1106
1107         zone = nf_ct_zone(ct);
1108         local_bh_disable();
1109
1110         do {
1111                 sequence = read_seqcount_begin(&nf_conntrack_generation);
1112                 /* reuse the hash saved before */
1113                 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
1114                 hash = scale_hash(hash);
1115                 reply_hash = hash_conntrack(net,
1116                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
1117
1118         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
1119
1120         /* We're not in hash table, and we refuse to set up related
1121          * connections for unconfirmed conns.  But packet copies and
1122          * REJECT will give spurious warnings here.
1123          */
1124
1125         /* Another skb with the same unconfirmed conntrack may
1126          * win the race. This may happen for bridge(br_flood)
1127          * or broadcast/multicast packets do skb_clone with
1128          * unconfirmed conntrack.
1129          */
1130         if (unlikely(nf_ct_is_confirmed(ct))) {
1131                 WARN_ON_ONCE(1);
1132                 nf_conntrack_double_unlock(hash, reply_hash);
1133                 local_bh_enable();
1134                 return NF_DROP;
1135         }
1136
1137         pr_debug("Confirming conntrack %p\n", ct);
1138         /* We have to check the DYING flag after unlink to prevent
1139          * a race against nf_ct_get_next_corpse() possibly called from
1140          * user context, else we insert an already 'dead' hash, blocking
1141          * further use of that particular connection -JM.
1142          */
1143         nf_ct_del_from_dying_or_unconfirmed_list(ct);
1144
1145         if (unlikely(nf_ct_is_dying(ct))) {
1146                 nf_ct_add_to_dying_list(ct);
1147                 NF_CT_STAT_INC(net, insert_failed);
1148                 goto dying;
1149         }
1150
1151         /* See if there's one in the list already, including reverse:
1152            NAT could have grabbed it without realizing, since we're
1153            not in the hash.  If there is, we lost race. */
1154         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
1155                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1156                                     zone, net))
1157                         goto out;
1158
1159         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
1160                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
1161                                     zone, net))
1162                         goto out;
1163
1164         /* Timer relative to confirmation time, not original
1165            setting time, otherwise we'd get timer wrap in
1166            weird delay cases. */
1167         ct->timeout += nfct_time_stamp;
1168
1169         __nf_conntrack_insert_prepare(ct);
1170
1171         /* Since the lookup is lockless, hash insertion must be done after
1172          * starting the timer and setting the CONFIRMED bit. The RCU barriers
1173          * guarantee that no other CPU can find the conntrack before the above
1174          * stores are visible.
1175          */
1176         __nf_conntrack_hash_insert(ct, hash, reply_hash);
1177         nf_conntrack_double_unlock(hash, reply_hash);
1178         local_bh_enable();
1179
1180         help = nfct_help(ct);
1181         if (help && help->helper)
1182                 nf_conntrack_event_cache(IPCT_HELPER, ct);
1183
1184         nf_conntrack_event_cache(master_ct(ct) ?
1185                                  IPCT_RELATED : IPCT_NEW, ct);
1186         return NF_ACCEPT;
1187
1188 out:
1189         ret = nf_ct_resolve_clash(skb, h, reply_hash);
1190 dying:
1191         nf_conntrack_double_unlock(hash, reply_hash);
1192         local_bh_enable();
1193         return ret;
1194 }
1195 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
1196
1197 /* Returns true if a connection correspondings to the tuple (required
1198    for NAT). */
1199 int
1200 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
1201                          const struct nf_conn *ignored_conntrack)
1202 {
1203         struct net *net = nf_ct_net(ignored_conntrack);
1204         const struct nf_conntrack_zone *zone;
1205         struct nf_conntrack_tuple_hash *h;
1206         struct hlist_nulls_head *ct_hash;
1207         unsigned int hash, hsize;
1208         struct hlist_nulls_node *n;
1209         struct nf_conn *ct;
1210
1211         zone = nf_ct_zone(ignored_conntrack);
1212
1213         rcu_read_lock();
1214  begin:
1215         nf_conntrack_get_ht(&ct_hash, &hsize);
1216         hash = __hash_conntrack(net, tuple, hsize);
1217
1218         hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
1219                 ct = nf_ct_tuplehash_to_ctrack(h);
1220
1221                 if (ct == ignored_conntrack)
1222                         continue;
1223
1224                 if (nf_ct_is_expired(ct)) {
1225                         nf_ct_gc_expired(ct);
1226                         continue;
1227                 }
1228
1229                 if (nf_ct_key_equal(h, tuple, zone, net)) {
1230                         /* Tuple is taken already, so caller will need to find
1231                          * a new source port to use.
1232                          *
1233                          * Only exception:
1234                          * If the *original tuples* are identical, then both
1235                          * conntracks refer to the same flow.
1236                          * This is a rare situation, it can occur e.g. when
1237                          * more than one UDP packet is sent from same socket
1238                          * in different threads.
1239                          *
1240                          * Let nf_ct_resolve_clash() deal with this later.
1241                          */
1242                         if (nf_ct_tuple_equal(&ignored_conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1243                                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple) &&
1244                                               nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL))
1245                                 continue;
1246
1247                         NF_CT_STAT_INC_ATOMIC(net, found);
1248                         rcu_read_unlock();
1249                         return 1;
1250                 }
1251         }
1252
1253         if (get_nulls_value(n) != hash) {
1254                 NF_CT_STAT_INC_ATOMIC(net, search_restart);
1255                 goto begin;
1256         }
1257
1258         rcu_read_unlock();
1259
1260         return 0;
1261 }
1262 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
1263
1264 #define NF_CT_EVICTION_RANGE    8
1265
1266 /* There's a small race here where we may free a just-assured
1267    connection.  Too bad: we're in trouble anyway. */
1268 static unsigned int early_drop_list(struct net *net,
1269                                     struct hlist_nulls_head *head)
1270 {
1271         struct nf_conntrack_tuple_hash *h;
1272         struct hlist_nulls_node *n;
1273         unsigned int drops = 0;
1274         struct nf_conn *tmp;
1275
1276         hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
1277                 tmp = nf_ct_tuplehash_to_ctrack(h);
1278
1279                 if (test_bit(IPS_OFFLOAD_BIT, &tmp->status))
1280                         continue;
1281
1282                 if (nf_ct_is_expired(tmp)) {
1283                         nf_ct_gc_expired(tmp);
1284                         continue;
1285                 }
1286
1287                 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
1288                     !net_eq(nf_ct_net(tmp), net) ||
1289                     nf_ct_is_dying(tmp))
1290                         continue;
1291
1292                 if (!atomic_inc_not_zero(&tmp->ct_general.use))
1293                         continue;
1294
1295                 /* kill only if still in same netns -- might have moved due to
1296                  * SLAB_TYPESAFE_BY_RCU rules.
1297                  *
1298                  * We steal the timer reference.  If that fails timer has
1299                  * already fired or someone else deleted it. Just drop ref
1300                  * and move to next entry.
1301                  */
1302                 if (net_eq(nf_ct_net(tmp), net) &&
1303                     nf_ct_is_confirmed(tmp) &&
1304                     nf_ct_delete(tmp, 0, 0))
1305                         drops++;
1306
1307                 nf_ct_put(tmp);
1308         }
1309
1310         return drops;
1311 }
1312
1313 static noinline int early_drop(struct net *net, unsigned int hash)
1314 {
1315         unsigned int i, bucket;
1316
1317         for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
1318                 struct hlist_nulls_head *ct_hash;
1319                 unsigned int hsize, drops;
1320
1321                 rcu_read_lock();
1322                 nf_conntrack_get_ht(&ct_hash, &hsize);
1323                 if (!i)
1324                         bucket = reciprocal_scale(hash, hsize);
1325                 else
1326                         bucket = (bucket + 1) % hsize;
1327
1328                 drops = early_drop_list(net, &ct_hash[bucket]);
1329                 rcu_read_unlock();
1330
1331                 if (drops) {
1332                         NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
1333                         return true;
1334                 }
1335         }
1336
1337         return false;
1338 }
1339
1340 static bool gc_worker_skip_ct(const struct nf_conn *ct)
1341 {
1342         return !nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct);
1343 }
1344
1345 static bool gc_worker_can_early_drop(const struct nf_conn *ct)
1346 {
1347         const struct nf_conntrack_l4proto *l4proto;
1348
1349         if (!test_bit(IPS_ASSURED_BIT, &ct->status))
1350                 return true;
1351
1352         l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
1353         if (l4proto->can_early_drop && l4proto->can_early_drop(ct))
1354                 return true;
1355
1356         return false;
1357 }
1358
1359 static void gc_worker(struct work_struct *work)
1360 {
1361         unsigned long end_time = jiffies + GC_SCAN_MAX_DURATION;
1362         unsigned int i, hashsz, nf_conntrack_max95 = 0;
1363         unsigned long next_run = GC_SCAN_INTERVAL;
1364         struct conntrack_gc_work *gc_work;
1365         gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
1366
1367         i = gc_work->next_bucket;
1368         if (gc_work->early_drop)
1369                 nf_conntrack_max95 = nf_conntrack_max / 100u * 95u;
1370
1371         do {
1372                 struct nf_conntrack_tuple_hash *h;
1373                 struct hlist_nulls_head *ct_hash;
1374                 struct hlist_nulls_node *n;
1375                 struct nf_conn *tmp;
1376
1377                 rcu_read_lock();
1378
1379                 nf_conntrack_get_ht(&ct_hash, &hashsz);
1380                 if (i >= hashsz) {
1381                         rcu_read_unlock();
1382                         break;
1383                 }
1384
1385                 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
1386                         struct nf_conntrack_net *cnet;
1387                         struct net *net;
1388
1389                         tmp = nf_ct_tuplehash_to_ctrack(h);
1390
1391                         if (test_bit(IPS_OFFLOAD_BIT, &tmp->status)) {
1392                                 nf_ct_offload_timeout(tmp);
1393                                 continue;
1394                         }
1395
1396                         if (nf_ct_is_expired(tmp)) {
1397                                 nf_ct_gc_expired(tmp);
1398                                 continue;
1399                         }
1400
1401                         if (nf_conntrack_max95 == 0 || gc_worker_skip_ct(tmp))
1402                                 continue;
1403
1404                         net = nf_ct_net(tmp);
1405                         cnet = nf_ct_pernet(net);
1406                         if (atomic_read(&cnet->count) < nf_conntrack_max95)
1407                                 continue;
1408
1409                         /* need to take reference to avoid possible races */
1410                         if (!atomic_inc_not_zero(&tmp->ct_general.use))
1411                                 continue;
1412
1413                         if (gc_worker_skip_ct(tmp)) {
1414                                 nf_ct_put(tmp);
1415                                 continue;
1416                         }
1417
1418                         if (gc_worker_can_early_drop(tmp))
1419                                 nf_ct_kill(tmp);
1420
1421                         nf_ct_put(tmp);
1422                 }
1423
1424                 /* could check get_nulls_value() here and restart if ct
1425                  * was moved to another chain.  But given gc is best-effort
1426                  * we will just continue with next hash slot.
1427                  */
1428                 rcu_read_unlock();
1429                 cond_resched();
1430                 i++;
1431
1432                 if (time_after(jiffies, end_time) && i < hashsz) {
1433                         gc_work->next_bucket = i;
1434                         next_run = 0;
1435                         break;
1436                 }
1437         } while (i < hashsz);
1438
1439         if (gc_work->exiting)
1440                 return;
1441
1442         /*
1443          * Eviction will normally happen from the packet path, and not
1444          * from this gc worker.
1445          *
1446          * This worker is only here to reap expired entries when system went
1447          * idle after a busy period.
1448          */
1449         if (next_run) {
1450                 gc_work->early_drop = false;
1451                 gc_work->next_bucket = 0;
1452         }
1453         queue_delayed_work(system_power_efficient_wq, &gc_work->dwork, next_run);
1454 }
1455
1456 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1457 {
1458         INIT_DEFERRABLE_WORK(&gc_work->dwork, gc_worker);
1459         gc_work->exiting = false;
1460 }
1461
1462 static struct nf_conn *
1463 __nf_conntrack_alloc(struct net *net,
1464                      const struct nf_conntrack_zone *zone,
1465                      const struct nf_conntrack_tuple *orig,
1466                      const struct nf_conntrack_tuple *repl,
1467                      gfp_t gfp, u32 hash)
1468 {
1469         struct nf_conntrack_net *cnet = nf_ct_pernet(net);
1470         unsigned int ct_count;
1471         struct nf_conn *ct;
1472
1473         /* We don't want any race condition at early drop stage */
1474         ct_count = atomic_inc_return(&cnet->count);
1475
1476         if (nf_conntrack_max && unlikely(ct_count > nf_conntrack_max)) {
1477                 if (!early_drop(net, hash)) {
1478                         if (!conntrack_gc_work.early_drop)
1479                                 conntrack_gc_work.early_drop = true;
1480                         atomic_dec(&cnet->count);
1481                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1482                         return ERR_PTR(-ENOMEM);
1483                 }
1484         }
1485
1486         /*
1487          * Do not use kmem_cache_zalloc(), as this cache uses
1488          * SLAB_TYPESAFE_BY_RCU.
1489          */
1490         ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1491         if (ct == NULL)
1492                 goto out;
1493
1494         spin_lock_init(&ct->lock);
1495         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1496         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1497         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1498         /* save hash for reusing when confirming */
1499         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1500         ct->status = 0;
1501         ct->timeout = 0;
1502         write_pnet(&ct->ct_net, net);
1503         memset(&ct->__nfct_init_offset, 0,
1504                offsetof(struct nf_conn, proto) -
1505                offsetof(struct nf_conn, __nfct_init_offset));
1506
1507         nf_ct_zone_add(ct, zone);
1508
1509         /* Because we use RCU lookups, we set ct_general.use to zero before
1510          * this is inserted in any list.
1511          */
1512         atomic_set(&ct->ct_general.use, 0);
1513         return ct;
1514 out:
1515         atomic_dec(&cnet->count);
1516         return ERR_PTR(-ENOMEM);
1517 }
1518
1519 struct nf_conn *nf_conntrack_alloc(struct net *net,
1520                                    const struct nf_conntrack_zone *zone,
1521                                    const struct nf_conntrack_tuple *orig,
1522                                    const struct nf_conntrack_tuple *repl,
1523                                    gfp_t gfp)
1524 {
1525         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1526 }
1527 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1528
1529 void nf_conntrack_free(struct nf_conn *ct)
1530 {
1531         struct net *net = nf_ct_net(ct);
1532         struct nf_conntrack_net *cnet;
1533
1534         /* A freed object has refcnt == 0, that's
1535          * the golden rule for SLAB_TYPESAFE_BY_RCU
1536          */
1537         WARN_ON(atomic_read(&ct->ct_general.use) != 0);
1538
1539         nf_ct_ext_destroy(ct);
1540         kmem_cache_free(nf_conntrack_cachep, ct);
1541         cnet = nf_ct_pernet(net);
1542
1543         smp_mb__before_atomic();
1544         atomic_dec(&cnet->count);
1545 }
1546 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1547
1548
1549 /* Allocate a new conntrack: we return -ENOMEM if classification
1550    failed due to stress.  Otherwise it really is unclassifiable. */
1551 static noinline struct nf_conntrack_tuple_hash *
1552 init_conntrack(struct net *net, struct nf_conn *tmpl,
1553                const struct nf_conntrack_tuple *tuple,
1554                struct sk_buff *skb,
1555                unsigned int dataoff, u32 hash)
1556 {
1557         struct nf_conn *ct;
1558         struct nf_conn_help *help;
1559         struct nf_conntrack_tuple repl_tuple;
1560         struct nf_conntrack_ecache *ecache;
1561         struct nf_conntrack_expect *exp = NULL;
1562         const struct nf_conntrack_zone *zone;
1563         struct nf_conn_timeout *timeout_ext;
1564         struct nf_conntrack_zone tmp;
1565         struct nf_conntrack_net *cnet;
1566
1567         if (!nf_ct_invert_tuple(&repl_tuple, tuple)) {
1568                 pr_debug("Can't invert tuple.\n");
1569                 return NULL;
1570         }
1571
1572         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1573         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1574                                   hash);
1575         if (IS_ERR(ct))
1576                 return (struct nf_conntrack_tuple_hash *)ct;
1577
1578         if (!nf_ct_add_synproxy(ct, tmpl)) {
1579                 nf_conntrack_free(ct);
1580                 return ERR_PTR(-ENOMEM);
1581         }
1582
1583         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1584
1585         if (timeout_ext)
1586                 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1587                                       GFP_ATOMIC);
1588
1589         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1590         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1591         nf_ct_labels_ext_add(ct);
1592
1593         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1594         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1595                                  ecache ? ecache->expmask : 0,
1596                              GFP_ATOMIC);
1597
1598         local_bh_disable();
1599         cnet = nf_ct_pernet(net);
1600         if (cnet->expect_count) {
1601                 spin_lock(&nf_conntrack_expect_lock);
1602                 exp = nf_ct_find_expectation(net, zone, tuple);
1603                 if (exp) {
1604                         pr_debug("expectation arrives ct=%p exp=%p\n",
1605                                  ct, exp);
1606                         /* Welcome, Mr. Bond.  We've been expecting you... */
1607                         __set_bit(IPS_EXPECTED_BIT, &ct->status);
1608                         /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1609                         ct->master = exp->master;
1610                         if (exp->helper) {
1611                                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
1612                                 if (help)
1613                                         rcu_assign_pointer(help->helper, exp->helper);
1614                         }
1615
1616 #ifdef CONFIG_NF_CONNTRACK_MARK
1617                         ct->mark = exp->master->mark;
1618 #endif
1619 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1620                         ct->secmark = exp->master->secmark;
1621 #endif
1622                         NF_CT_STAT_INC(net, expect_new);
1623                 }
1624                 spin_unlock(&nf_conntrack_expect_lock);
1625         }
1626         if (!exp)
1627                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1628
1629         /* Now it is inserted into the unconfirmed list, bump refcount */
1630         nf_conntrack_get(&ct->ct_general);
1631         nf_ct_add_to_unconfirmed_list(ct);
1632
1633         local_bh_enable();
1634
1635         if (exp) {
1636                 if (exp->expectfn)
1637                         exp->expectfn(ct, exp);
1638                 nf_ct_expect_put(exp);
1639         }
1640
1641         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1642 }
1643
1644 /* On success, returns 0, sets skb->_nfct | ctinfo */
1645 static int
1646 resolve_normal_ct(struct nf_conn *tmpl,
1647                   struct sk_buff *skb,
1648                   unsigned int dataoff,
1649                   u_int8_t protonum,
1650                   const struct nf_hook_state *state)
1651 {
1652         const struct nf_conntrack_zone *zone;
1653         struct nf_conntrack_tuple tuple;
1654         struct nf_conntrack_tuple_hash *h;
1655         enum ip_conntrack_info ctinfo;
1656         struct nf_conntrack_zone tmp;
1657         struct nf_conn *ct;
1658         u32 hash;
1659
1660         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1661                              dataoff, state->pf, protonum, state->net,
1662                              &tuple)) {
1663                 pr_debug("Can't get tuple\n");
1664                 return 0;
1665         }
1666
1667         /* look for tuple match */
1668         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1669         hash = hash_conntrack_raw(&tuple, state->net);
1670         h = __nf_conntrack_find_get(state->net, zone, &tuple, hash);
1671         if (!h) {
1672                 h = init_conntrack(state->net, tmpl, &tuple,
1673                                    skb, dataoff, hash);
1674                 if (!h)
1675                         return 0;
1676                 if (IS_ERR(h))
1677                         return PTR_ERR(h);
1678         }
1679         ct = nf_ct_tuplehash_to_ctrack(h);
1680
1681         /* It exists; we have (non-exclusive) reference. */
1682         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1683                 ctinfo = IP_CT_ESTABLISHED_REPLY;
1684         } else {
1685                 /* Once we've had two way comms, always ESTABLISHED. */
1686                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1687                         pr_debug("normal packet for %p\n", ct);
1688                         ctinfo = IP_CT_ESTABLISHED;
1689                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1690                         pr_debug("related packet for %p\n", ct);
1691                         ctinfo = IP_CT_RELATED;
1692                 } else {
1693                         pr_debug("new packet for %p\n", ct);
1694                         ctinfo = IP_CT_NEW;
1695                 }
1696         }
1697         nf_ct_set(skb, ct, ctinfo);
1698         return 0;
1699 }
1700
1701 /*
1702  * icmp packets need special treatment to handle error messages that are
1703  * related to a connection.
1704  *
1705  * Callers need to check if skb has a conntrack assigned when this
1706  * helper returns; in such case skb belongs to an already known connection.
1707  */
1708 static unsigned int __cold
1709 nf_conntrack_handle_icmp(struct nf_conn *tmpl,
1710                          struct sk_buff *skb,
1711                          unsigned int dataoff,
1712                          u8 protonum,
1713                          const struct nf_hook_state *state)
1714 {
1715         int ret;
1716
1717         if (state->pf == NFPROTO_IPV4 && protonum == IPPROTO_ICMP)
1718                 ret = nf_conntrack_icmpv4_error(tmpl, skb, dataoff, state);
1719 #if IS_ENABLED(CONFIG_IPV6)
1720         else if (state->pf == NFPROTO_IPV6 && protonum == IPPROTO_ICMPV6)
1721                 ret = nf_conntrack_icmpv6_error(tmpl, skb, dataoff, state);
1722 #endif
1723         else
1724                 return NF_ACCEPT;
1725
1726         if (ret <= 0)
1727                 NF_CT_STAT_INC_ATOMIC(state->net, error);
1728
1729         return ret;
1730 }
1731
1732 static int generic_packet(struct nf_conn *ct, struct sk_buff *skb,
1733                           enum ip_conntrack_info ctinfo)
1734 {
1735         const unsigned int *timeout = nf_ct_timeout_lookup(ct);
1736
1737         if (!timeout)
1738                 timeout = &nf_generic_pernet(nf_ct_net(ct))->timeout;
1739
1740         nf_ct_refresh_acct(ct, ctinfo, skb, *timeout);
1741         return NF_ACCEPT;
1742 }
1743
1744 /* Returns verdict for packet, or -1 for invalid. */
1745 static int nf_conntrack_handle_packet(struct nf_conn *ct,
1746                                       struct sk_buff *skb,
1747                                       unsigned int dataoff,
1748                                       enum ip_conntrack_info ctinfo,
1749                                       const struct nf_hook_state *state)
1750 {
1751         switch (nf_ct_protonum(ct)) {
1752         case IPPROTO_TCP:
1753                 return nf_conntrack_tcp_packet(ct, skb, dataoff,
1754                                                ctinfo, state);
1755         case IPPROTO_UDP:
1756                 return nf_conntrack_udp_packet(ct, skb, dataoff,
1757                                                ctinfo, state);
1758         case IPPROTO_ICMP:
1759                 return nf_conntrack_icmp_packet(ct, skb, ctinfo, state);
1760 #if IS_ENABLED(CONFIG_IPV6)
1761         case IPPROTO_ICMPV6:
1762                 return nf_conntrack_icmpv6_packet(ct, skb, ctinfo, state);
1763 #endif
1764 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
1765         case IPPROTO_UDPLITE:
1766                 return nf_conntrack_udplite_packet(ct, skb, dataoff,
1767                                                    ctinfo, state);
1768 #endif
1769 #ifdef CONFIG_NF_CT_PROTO_SCTP
1770         case IPPROTO_SCTP:
1771                 return nf_conntrack_sctp_packet(ct, skb, dataoff,
1772                                                 ctinfo, state);
1773 #endif
1774 #ifdef CONFIG_NF_CT_PROTO_DCCP
1775         case IPPROTO_DCCP:
1776                 return nf_conntrack_dccp_packet(ct, skb, dataoff,
1777                                                 ctinfo, state);
1778 #endif
1779 #ifdef CONFIG_NF_CT_PROTO_GRE
1780         case IPPROTO_GRE:
1781                 return nf_conntrack_gre_packet(ct, skb, dataoff,
1782                                                ctinfo, state);
1783 #endif
1784         }
1785
1786         return generic_packet(ct, skb, ctinfo);
1787 }
1788
1789 unsigned int
1790 nf_conntrack_in(struct sk_buff *skb, const struct nf_hook_state *state)
1791 {
1792         enum ip_conntrack_info ctinfo;
1793         struct nf_conn *ct, *tmpl;
1794         u_int8_t protonum;
1795         int dataoff, ret;
1796
1797         tmpl = nf_ct_get(skb, &ctinfo);
1798         if (tmpl || ctinfo == IP_CT_UNTRACKED) {
1799                 /* Previously seen (loopback or untracked)?  Ignore. */
1800                 if ((tmpl && !nf_ct_is_template(tmpl)) ||
1801                      ctinfo == IP_CT_UNTRACKED)
1802                         return NF_ACCEPT;
1803                 skb->_nfct = 0;
1804         }
1805
1806         /* rcu_read_lock()ed by nf_hook_thresh */
1807         dataoff = get_l4proto(skb, skb_network_offset(skb), state->pf, &protonum);
1808         if (dataoff <= 0) {
1809                 pr_debug("not prepared to track yet or error occurred\n");
1810                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1811                 ret = NF_ACCEPT;
1812                 goto out;
1813         }
1814
1815         if (protonum == IPPROTO_ICMP || protonum == IPPROTO_ICMPV6) {
1816                 ret = nf_conntrack_handle_icmp(tmpl, skb, dataoff,
1817                                                protonum, state);
1818                 if (ret <= 0) {
1819                         ret = -ret;
1820                         goto out;
1821                 }
1822                 /* ICMP[v6] protocol trackers may assign one conntrack. */
1823                 if (skb->_nfct)
1824                         goto out;
1825         }
1826 repeat:
1827         ret = resolve_normal_ct(tmpl, skb, dataoff,
1828                                 protonum, state);
1829         if (ret < 0) {
1830                 /* Too stressed to deal. */
1831                 NF_CT_STAT_INC_ATOMIC(state->net, drop);
1832                 ret = NF_DROP;
1833                 goto out;
1834         }
1835
1836         ct = nf_ct_get(skb, &ctinfo);
1837         if (!ct) {
1838                 /* Not valid part of a connection */
1839                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1840                 ret = NF_ACCEPT;
1841                 goto out;
1842         }
1843
1844         ret = nf_conntrack_handle_packet(ct, skb, dataoff, ctinfo, state);
1845         if (ret <= 0) {
1846                 /* Invalid: inverse of the return code tells
1847                  * the netfilter core what to do */
1848                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1849                 nf_conntrack_put(&ct->ct_general);
1850                 skb->_nfct = 0;
1851                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1852                 if (ret == -NF_DROP)
1853                         NF_CT_STAT_INC_ATOMIC(state->net, drop);
1854                 /* Special case: TCP tracker reports an attempt to reopen a
1855                  * closed/aborted connection. We have to go back and create a
1856                  * fresh conntrack.
1857                  */
1858                 if (ret == -NF_REPEAT)
1859                         goto repeat;
1860                 ret = -ret;
1861                 goto out;
1862         }
1863
1864         if (ctinfo == IP_CT_ESTABLISHED_REPLY &&
1865             !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1866                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1867 out:
1868         if (tmpl)
1869                 nf_ct_put(tmpl);
1870
1871         return ret;
1872 }
1873 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1874
1875 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1876    implicitly racy: see __nf_conntrack_confirm */
1877 void nf_conntrack_alter_reply(struct nf_conn *ct,
1878                               const struct nf_conntrack_tuple *newreply)
1879 {
1880         struct nf_conn_help *help = nfct_help(ct);
1881
1882         /* Should be unconfirmed, so not in hash table yet */
1883         WARN_ON(nf_ct_is_confirmed(ct));
1884
1885         pr_debug("Altering reply tuple of %p to ", ct);
1886         nf_ct_dump_tuple(newreply);
1887
1888         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1889         if (ct->master || (help && !hlist_empty(&help->expectations)))
1890                 return;
1891
1892         rcu_read_lock();
1893         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1894         rcu_read_unlock();
1895 }
1896 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1897
1898 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1899 void __nf_ct_refresh_acct(struct nf_conn *ct,
1900                           enum ip_conntrack_info ctinfo,
1901                           const struct sk_buff *skb,
1902                           u32 extra_jiffies,
1903                           bool do_acct)
1904 {
1905         /* Only update if this is not a fixed timeout */
1906         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1907                 goto acct;
1908
1909         /* If not in hash table, timer will not be active yet */
1910         if (nf_ct_is_confirmed(ct))
1911                 extra_jiffies += nfct_time_stamp;
1912
1913         if (READ_ONCE(ct->timeout) != extra_jiffies)
1914                 WRITE_ONCE(ct->timeout, extra_jiffies);
1915 acct:
1916         if (do_acct)
1917                 nf_ct_acct_update(ct, CTINFO2DIR(ctinfo), skb->len);
1918 }
1919 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1920
1921 bool nf_ct_kill_acct(struct nf_conn *ct,
1922                      enum ip_conntrack_info ctinfo,
1923                      const struct sk_buff *skb)
1924 {
1925         nf_ct_acct_update(ct, CTINFO2DIR(ctinfo), skb->len);
1926
1927         return nf_ct_delete(ct, 0, 0);
1928 }
1929 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1930
1931 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1932
1933 #include <linux/netfilter/nfnetlink.h>
1934 #include <linux/netfilter/nfnetlink_conntrack.h>
1935 #include <linux/mutex.h>
1936
1937 /* Generic function for tcp/udp/sctp/dccp and alike. */
1938 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1939                                const struct nf_conntrack_tuple *tuple)
1940 {
1941         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1942             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1943                 goto nla_put_failure;
1944         return 0;
1945
1946 nla_put_failure:
1947         return -1;
1948 }
1949 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1950
1951 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1952         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1953         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1954 };
1955 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1956
1957 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1958                                struct nf_conntrack_tuple *t,
1959                                u_int32_t flags)
1960 {
1961         if (flags & CTA_FILTER_FLAG(CTA_PROTO_SRC_PORT)) {
1962                 if (!tb[CTA_PROTO_SRC_PORT])
1963                         return -EINVAL;
1964
1965                 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1966         }
1967
1968         if (flags & CTA_FILTER_FLAG(CTA_PROTO_DST_PORT)) {
1969                 if (!tb[CTA_PROTO_DST_PORT])
1970                         return -EINVAL;
1971
1972                 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1973         }
1974
1975         return 0;
1976 }
1977 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1978
1979 unsigned int nf_ct_port_nlattr_tuple_size(void)
1980 {
1981         static unsigned int size __read_mostly;
1982
1983         if (!size)
1984                 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1985
1986         return size;
1987 }
1988 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1989 #endif
1990
1991 /* Used by ipt_REJECT and ip6t_REJECT. */
1992 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1993 {
1994         struct nf_conn *ct;
1995         enum ip_conntrack_info ctinfo;
1996
1997         /* This ICMP is in reverse direction to the packet which caused it */
1998         ct = nf_ct_get(skb, &ctinfo);
1999         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
2000                 ctinfo = IP_CT_RELATED_REPLY;
2001         else
2002                 ctinfo = IP_CT_RELATED;
2003
2004         /* Attach to new skbuff, and increment count */
2005         nf_ct_set(nskb, ct, ctinfo);
2006         nf_conntrack_get(skb_nfct(nskb));
2007 }
2008
2009 static int __nf_conntrack_update(struct net *net, struct sk_buff *skb,
2010                                  struct nf_conn *ct,
2011                                  enum ip_conntrack_info ctinfo)
2012 {
2013         struct nf_conntrack_tuple_hash *h;
2014         struct nf_conntrack_tuple tuple;
2015         struct nf_nat_hook *nat_hook;
2016         unsigned int status;
2017         int dataoff;
2018         u16 l3num;
2019         u8 l4num;
2020
2021         l3num = nf_ct_l3num(ct);
2022
2023         dataoff = get_l4proto(skb, skb_network_offset(skb), l3num, &l4num);
2024         if (dataoff <= 0)
2025                 return -1;
2026
2027         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
2028                              l4num, net, &tuple))
2029                 return -1;
2030
2031         if (ct->status & IPS_SRC_NAT) {
2032                 memcpy(tuple.src.u3.all,
2033                        ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.all,
2034                        sizeof(tuple.src.u3.all));
2035                 tuple.src.u.all =
2036                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.all;
2037         }
2038
2039         if (ct->status & IPS_DST_NAT) {
2040                 memcpy(tuple.dst.u3.all,
2041                        ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.all,
2042                        sizeof(tuple.dst.u3.all));
2043                 tuple.dst.u.all =
2044                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u.all;
2045         }
2046
2047         h = nf_conntrack_find_get(net, nf_ct_zone(ct), &tuple);
2048         if (!h)
2049                 return 0;
2050
2051         /* Store status bits of the conntrack that is clashing to re-do NAT
2052          * mangling according to what it has been done already to this packet.
2053          */
2054         status = ct->status;
2055
2056         nf_ct_put(ct);
2057         ct = nf_ct_tuplehash_to_ctrack(h);
2058         nf_ct_set(skb, ct, ctinfo);
2059
2060         nat_hook = rcu_dereference(nf_nat_hook);
2061         if (!nat_hook)
2062                 return 0;
2063
2064         if (status & IPS_SRC_NAT &&
2065             nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_SRC,
2066                                 IP_CT_DIR_ORIGINAL) == NF_DROP)
2067                 return -1;
2068
2069         if (status & IPS_DST_NAT &&
2070             nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_DST,
2071                                 IP_CT_DIR_ORIGINAL) == NF_DROP)
2072                 return -1;
2073
2074         return 0;
2075 }
2076
2077 /* This packet is coming from userspace via nf_queue, complete the packet
2078  * processing after the helper invocation in nf_confirm().
2079  */
2080 static int nf_confirm_cthelper(struct sk_buff *skb, struct nf_conn *ct,
2081                                enum ip_conntrack_info ctinfo)
2082 {
2083         const struct nf_conntrack_helper *helper;
2084         const struct nf_conn_help *help;
2085         int protoff;
2086
2087         help = nfct_help(ct);
2088         if (!help)
2089                 return 0;
2090
2091         helper = rcu_dereference(help->helper);
2092         if (!(helper->flags & NF_CT_HELPER_F_USERSPACE))
2093                 return 0;
2094
2095         switch (nf_ct_l3num(ct)) {
2096         case NFPROTO_IPV4:
2097                 protoff = skb_network_offset(skb) + ip_hdrlen(skb);
2098                 break;
2099 #if IS_ENABLED(CONFIG_IPV6)
2100         case NFPROTO_IPV6: {
2101                 __be16 frag_off;
2102                 u8 pnum;
2103
2104                 pnum = ipv6_hdr(skb)->nexthdr;
2105                 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &pnum,
2106                                            &frag_off);
2107                 if (protoff < 0 || (frag_off & htons(~0x7)) != 0)
2108                         return 0;
2109                 break;
2110         }
2111 #endif
2112         default:
2113                 return 0;
2114         }
2115
2116         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
2117             !nf_is_loopback_packet(skb)) {
2118                 if (!nf_ct_seq_adjust(skb, ct, ctinfo, protoff)) {
2119                         NF_CT_STAT_INC_ATOMIC(nf_ct_net(ct), drop);
2120                         return -1;
2121                 }
2122         }
2123
2124         /* We've seen it coming out the other side: confirm it */
2125         return nf_conntrack_confirm(skb) == NF_DROP ? - 1 : 0;
2126 }
2127
2128 static int nf_conntrack_update(struct net *net, struct sk_buff *skb)
2129 {
2130         enum ip_conntrack_info ctinfo;
2131         struct nf_conn *ct;
2132         int err;
2133
2134         ct = nf_ct_get(skb, &ctinfo);
2135         if (!ct)
2136                 return 0;
2137
2138         if (!nf_ct_is_confirmed(ct)) {
2139                 err = __nf_conntrack_update(net, skb, ct, ctinfo);
2140                 if (err < 0)
2141                         return err;
2142
2143                 ct = nf_ct_get(skb, &ctinfo);
2144         }
2145
2146         return nf_confirm_cthelper(skb, ct, ctinfo);
2147 }
2148
2149 static bool nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
2150                                        const struct sk_buff *skb)
2151 {
2152         const struct nf_conntrack_tuple *src_tuple;
2153         const struct nf_conntrack_tuple_hash *hash;
2154         struct nf_conntrack_tuple srctuple;
2155         enum ip_conntrack_info ctinfo;
2156         struct nf_conn *ct;
2157
2158         ct = nf_ct_get(skb, &ctinfo);
2159         if (ct) {
2160                 src_tuple = nf_ct_tuple(ct, CTINFO2DIR(ctinfo));
2161                 memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
2162                 return true;
2163         }
2164
2165         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
2166                                NFPROTO_IPV4, dev_net(skb->dev),
2167                                &srctuple))
2168                 return false;
2169
2170         hash = nf_conntrack_find_get(dev_net(skb->dev),
2171                                      &nf_ct_zone_dflt,
2172                                      &srctuple);
2173         if (!hash)
2174                 return false;
2175
2176         ct = nf_ct_tuplehash_to_ctrack(hash);
2177         src_tuple = nf_ct_tuple(ct, !hash->tuple.dst.dir);
2178         memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
2179         nf_ct_put(ct);
2180
2181         return true;
2182 }
2183
2184 /* Bring out ya dead! */
2185 static struct nf_conn *
2186 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
2187                 void *data, unsigned int *bucket)
2188 {
2189         struct nf_conntrack_tuple_hash *h;
2190         struct nf_conn *ct;
2191         struct hlist_nulls_node *n;
2192         spinlock_t *lockp;
2193
2194         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
2195                 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
2196                 local_bh_disable();
2197                 nf_conntrack_lock(lockp);
2198                 if (*bucket < nf_conntrack_htable_size) {
2199                         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
2200                                 if (NF_CT_DIRECTION(h) != IP_CT_DIR_REPLY)
2201                                         continue;
2202                                 /* All nf_conn objects are added to hash table twice, one
2203                                  * for original direction tuple, once for the reply tuple.
2204                                  *
2205                                  * Exception: In the IPS_NAT_CLASH case, only the reply
2206                                  * tuple is added (the original tuple already existed for
2207                                  * a different object).
2208                                  *
2209                                  * We only need to call the iterator once for each
2210                                  * conntrack, so we just use the 'reply' direction
2211                                  * tuple while iterating.
2212                                  */
2213                                 ct = nf_ct_tuplehash_to_ctrack(h);
2214                                 if (iter(ct, data))
2215                                         goto found;
2216                         }
2217                 }
2218                 spin_unlock(lockp);
2219                 local_bh_enable();
2220                 cond_resched();
2221         }
2222
2223         return NULL;
2224 found:
2225         atomic_inc(&ct->ct_general.use);
2226         spin_unlock(lockp);
2227         local_bh_enable();
2228         return ct;
2229 }
2230
2231 static void nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data),
2232                                   void *data, u32 portid, int report)
2233 {
2234         unsigned int bucket = 0, sequence;
2235         struct nf_conn *ct;
2236
2237         might_sleep();
2238
2239         for (;;) {
2240                 sequence = read_seqcount_begin(&nf_conntrack_generation);
2241
2242                 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
2243                         /* Time to push up daises... */
2244
2245                         nf_ct_delete(ct, portid, report);
2246                         nf_ct_put(ct);
2247                         cond_resched();
2248                 }
2249
2250                 if (!read_seqcount_retry(&nf_conntrack_generation, sequence))
2251                         break;
2252                 bucket = 0;
2253         }
2254 }
2255
2256 struct iter_data {
2257         int (*iter)(struct nf_conn *i, void *data);
2258         void *data;
2259         struct net *net;
2260 };
2261
2262 static int iter_net_only(struct nf_conn *i, void *data)
2263 {
2264         struct iter_data *d = data;
2265
2266         if (!net_eq(d->net, nf_ct_net(i)))
2267                 return 0;
2268
2269         return d->iter(i, d->data);
2270 }
2271
2272 static void
2273 __nf_ct_unconfirmed_destroy(struct net *net)
2274 {
2275         int cpu;
2276
2277         for_each_possible_cpu(cpu) {
2278                 struct nf_conntrack_tuple_hash *h;
2279                 struct hlist_nulls_node *n;
2280                 struct ct_pcpu *pcpu;
2281
2282                 pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2283
2284                 spin_lock_bh(&pcpu->lock);
2285                 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
2286                         struct nf_conn *ct;
2287
2288                         ct = nf_ct_tuplehash_to_ctrack(h);
2289
2290                         /* we cannot call iter() on unconfirmed list, the
2291                          * owning cpu can reallocate ct->ext at any time.
2292                          */
2293                         set_bit(IPS_DYING_BIT, &ct->status);
2294                 }
2295                 spin_unlock_bh(&pcpu->lock);
2296                 cond_resched();
2297         }
2298 }
2299
2300 void nf_ct_unconfirmed_destroy(struct net *net)
2301 {
2302         struct nf_conntrack_net *cnet = nf_ct_pernet(net);
2303
2304         might_sleep();
2305
2306         if (atomic_read(&cnet->count) > 0) {
2307                 __nf_ct_unconfirmed_destroy(net);
2308                 nf_queue_nf_hook_drop(net);
2309                 synchronize_net();
2310         }
2311 }
2312 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy);
2313
2314 void nf_ct_iterate_cleanup_net(struct net *net,
2315                                int (*iter)(struct nf_conn *i, void *data),
2316                                void *data, u32 portid, int report)
2317 {
2318         struct nf_conntrack_net *cnet = nf_ct_pernet(net);
2319         struct iter_data d;
2320
2321         might_sleep();
2322
2323         if (atomic_read(&cnet->count) == 0)
2324                 return;
2325
2326         d.iter = iter;
2327         d.data = data;
2328         d.net = net;
2329
2330         nf_ct_iterate_cleanup(iter_net_only, &d, portid, report);
2331 }
2332 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net);
2333
2334 /**
2335  * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
2336  * @iter: callback to invoke for each conntrack
2337  * @data: data to pass to @iter
2338  *
2339  * Like nf_ct_iterate_cleanup, but first marks conntracks on the
2340  * unconfirmed list as dying (so they will not be inserted into
2341  * main table).
2342  *
2343  * Can only be called in module exit path.
2344  */
2345 void
2346 nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data)
2347 {
2348         struct net *net;
2349
2350         down_read(&net_rwsem);
2351         for_each_net(net) {
2352                 struct nf_conntrack_net *cnet = nf_ct_pernet(net);
2353
2354                 if (atomic_read(&cnet->count) == 0)
2355                         continue;
2356                 __nf_ct_unconfirmed_destroy(net);
2357                 nf_queue_nf_hook_drop(net);
2358         }
2359         up_read(&net_rwsem);
2360
2361         /* Need to wait for netns cleanup worker to finish, if its
2362          * running -- it might have deleted a net namespace from
2363          * the global list, so our __nf_ct_unconfirmed_destroy() might
2364          * not have affected all namespaces.
2365          */
2366         net_ns_barrier();
2367
2368         /* a conntrack could have been unlinked from unconfirmed list
2369          * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
2370          * This makes sure its inserted into conntrack table.
2371          */
2372         synchronize_net();
2373
2374         nf_ct_iterate_cleanup(iter, data, 0, 0);
2375 }
2376 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy);
2377
2378 static int kill_all(struct nf_conn *i, void *data)
2379 {
2380         return net_eq(nf_ct_net(i), data);
2381 }
2382
2383 void nf_conntrack_cleanup_start(void)
2384 {
2385         conntrack_gc_work.exiting = true;
2386         RCU_INIT_POINTER(ip_ct_attach, NULL);
2387 }
2388
2389 void nf_conntrack_cleanup_end(void)
2390 {
2391         RCU_INIT_POINTER(nf_ct_hook, NULL);
2392         cancel_delayed_work_sync(&conntrack_gc_work.dwork);
2393         kvfree(nf_conntrack_hash);
2394
2395         nf_conntrack_proto_fini();
2396         nf_conntrack_seqadj_fini();
2397         nf_conntrack_labels_fini();
2398         nf_conntrack_helper_fini();
2399         nf_conntrack_timeout_fini();
2400         nf_conntrack_ecache_fini();
2401         nf_conntrack_tstamp_fini();
2402         nf_conntrack_acct_fini();
2403         nf_conntrack_expect_fini();
2404
2405         kmem_cache_destroy(nf_conntrack_cachep);
2406 }
2407
2408 /*
2409  * Mishearing the voices in his head, our hero wonders how he's
2410  * supposed to kill the mall.
2411  */
2412 void nf_conntrack_cleanup_net(struct net *net)
2413 {
2414         LIST_HEAD(single);
2415
2416         list_add(&net->exit_list, &single);
2417         nf_conntrack_cleanup_net_list(&single);
2418 }
2419
2420 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
2421 {
2422         int busy;
2423         struct net *net;
2424
2425         /*
2426          * This makes sure all current packets have passed through
2427          *  netfilter framework.  Roll on, two-stage module
2428          *  delete...
2429          */
2430         synchronize_net();
2431 i_see_dead_people:
2432         busy = 0;
2433         list_for_each_entry(net, net_exit_list, exit_list) {
2434                 struct nf_conntrack_net *cnet = nf_ct_pernet(net);
2435
2436                 nf_ct_iterate_cleanup(kill_all, net, 0, 0);
2437                 if (atomic_read(&cnet->count) != 0)
2438                         busy = 1;
2439         }
2440         if (busy) {
2441                 schedule();
2442                 goto i_see_dead_people;
2443         }
2444
2445         list_for_each_entry(net, net_exit_list, exit_list) {
2446                 nf_conntrack_ecache_pernet_fini(net);
2447                 nf_conntrack_expect_pernet_fini(net);
2448                 free_percpu(net->ct.stat);
2449                 free_percpu(net->ct.pcpu_lists);
2450         }
2451 }
2452
2453 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
2454 {
2455         struct hlist_nulls_head *hash;
2456         unsigned int nr_slots, i;
2457
2458         if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
2459                 return NULL;
2460
2461         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
2462         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
2463
2464         hash = kvcalloc(nr_slots, sizeof(struct hlist_nulls_head), GFP_KERNEL);
2465
2466         if (hash && nulls)
2467                 for (i = 0; i < nr_slots; i++)
2468                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
2469
2470         return hash;
2471 }
2472 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
2473
2474 int nf_conntrack_hash_resize(unsigned int hashsize)
2475 {
2476         int i, bucket;
2477         unsigned int old_size;
2478         struct hlist_nulls_head *hash, *old_hash;
2479         struct nf_conntrack_tuple_hash *h;
2480         struct nf_conn *ct;
2481
2482         if (!hashsize)
2483                 return -EINVAL;
2484
2485         hash = nf_ct_alloc_hashtable(&hashsize, 1);
2486         if (!hash)
2487                 return -ENOMEM;
2488
2489         old_size = nf_conntrack_htable_size;
2490         if (old_size == hashsize) {
2491                 kvfree(hash);
2492                 return 0;
2493         }
2494
2495         local_bh_disable();
2496         nf_conntrack_all_lock();
2497         write_seqcount_begin(&nf_conntrack_generation);
2498
2499         /* Lookups in the old hash might happen in parallel, which means we
2500          * might get false negatives during connection lookup. New connections
2501          * created because of a false negative won't make it into the hash
2502          * though since that required taking the locks.
2503          */
2504
2505         for (i = 0; i < nf_conntrack_htable_size; i++) {
2506                 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
2507                         h = hlist_nulls_entry(nf_conntrack_hash[i].first,
2508                                               struct nf_conntrack_tuple_hash, hnnode);
2509                         ct = nf_ct_tuplehash_to_ctrack(h);
2510                         hlist_nulls_del_rcu(&h->hnnode);
2511                         bucket = __hash_conntrack(nf_ct_net(ct),
2512                                                   &h->tuple, hashsize);
2513                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
2514                 }
2515         }
2516         old_size = nf_conntrack_htable_size;
2517         old_hash = nf_conntrack_hash;
2518
2519         nf_conntrack_hash = hash;
2520         nf_conntrack_htable_size = hashsize;
2521
2522         write_seqcount_end(&nf_conntrack_generation);
2523         nf_conntrack_all_unlock();
2524         local_bh_enable();
2525
2526         synchronize_net();
2527         kvfree(old_hash);
2528         return 0;
2529 }
2530
2531 int nf_conntrack_set_hashsize(const char *val, const struct kernel_param *kp)
2532 {
2533         unsigned int hashsize;
2534         int rc;
2535
2536         if (current->nsproxy->net_ns != &init_net)
2537                 return -EOPNOTSUPP;
2538
2539         /* On boot, we can set this without any fancy locking. */
2540         if (!nf_conntrack_hash)
2541                 return param_set_uint(val, kp);
2542
2543         rc = kstrtouint(val, 0, &hashsize);
2544         if (rc)
2545                 return rc;
2546
2547         return nf_conntrack_hash_resize(hashsize);
2548 }
2549
2550 static __always_inline unsigned int total_extension_size(void)
2551 {
2552         /* remember to add new extensions below */
2553         BUILD_BUG_ON(NF_CT_EXT_NUM > 9);
2554
2555         return sizeof(struct nf_ct_ext) +
2556                sizeof(struct nf_conn_help)
2557 #if IS_ENABLED(CONFIG_NF_NAT)
2558                 + sizeof(struct nf_conn_nat)
2559 #endif
2560                 + sizeof(struct nf_conn_seqadj)
2561                 + sizeof(struct nf_conn_acct)
2562 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2563                 + sizeof(struct nf_conntrack_ecache)
2564 #endif
2565 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2566                 + sizeof(struct nf_conn_tstamp)
2567 #endif
2568 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2569                 + sizeof(struct nf_conn_timeout)
2570 #endif
2571 #ifdef CONFIG_NF_CONNTRACK_LABELS
2572                 + sizeof(struct nf_conn_labels)
2573 #endif
2574 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2575                 + sizeof(struct nf_conn_synproxy)
2576 #endif
2577         ;
2578 };
2579
2580 int nf_conntrack_init_start(void)
2581 {
2582         unsigned long nr_pages = totalram_pages();
2583         int max_factor = 8;
2584         int ret = -ENOMEM;
2585         int i;
2586
2587         /* struct nf_ct_ext uses u8 to store offsets/size */
2588         BUILD_BUG_ON(total_extension_size() > 255u);
2589
2590         seqcount_spinlock_init(&nf_conntrack_generation,
2591                                &nf_conntrack_locks_all_lock);
2592
2593         for (i = 0; i < CONNTRACK_LOCKS; i++)
2594                 spin_lock_init(&nf_conntrack_locks[i]);
2595
2596         if (!nf_conntrack_htable_size) {
2597                 /* Idea from tcp.c: use 1/16384 of memory.
2598                  * On i386: 32MB machine has 512 buckets.
2599                  * >= 1GB machines have 16384 buckets.
2600                  * >= 4GB machines have 65536 buckets.
2601                  */
2602                 nf_conntrack_htable_size
2603                         = (((nr_pages << PAGE_SHIFT) / 16384)
2604                            / sizeof(struct hlist_head));
2605                 if (nr_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
2606                         nf_conntrack_htable_size = 65536;
2607                 else if (nr_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
2608                         nf_conntrack_htable_size = 16384;
2609                 if (nf_conntrack_htable_size < 32)
2610                         nf_conntrack_htable_size = 32;
2611
2612                 /* Use a max. factor of four by default to get the same max as
2613                  * with the old struct list_heads. When a table size is given
2614                  * we use the old value of 8 to avoid reducing the max.
2615                  * entries. */
2616                 max_factor = 4;
2617         }
2618
2619         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
2620         if (!nf_conntrack_hash)
2621                 return -ENOMEM;
2622
2623         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
2624
2625         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
2626                                                 sizeof(struct nf_conn),
2627                                                 NFCT_INFOMASK + 1,
2628                                                 SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
2629         if (!nf_conntrack_cachep)
2630                 goto err_cachep;
2631
2632         ret = nf_conntrack_expect_init();
2633         if (ret < 0)
2634                 goto err_expect;
2635
2636         ret = nf_conntrack_acct_init();
2637         if (ret < 0)
2638                 goto err_acct;
2639
2640         ret = nf_conntrack_tstamp_init();
2641         if (ret < 0)
2642                 goto err_tstamp;
2643
2644         ret = nf_conntrack_ecache_init();
2645         if (ret < 0)
2646                 goto err_ecache;
2647
2648         ret = nf_conntrack_timeout_init();
2649         if (ret < 0)
2650                 goto err_timeout;
2651
2652         ret = nf_conntrack_helper_init();
2653         if (ret < 0)
2654                 goto err_helper;
2655
2656         ret = nf_conntrack_labels_init();
2657         if (ret < 0)
2658                 goto err_labels;
2659
2660         ret = nf_conntrack_seqadj_init();
2661         if (ret < 0)
2662                 goto err_seqadj;
2663
2664         ret = nf_conntrack_proto_init();
2665         if (ret < 0)
2666                 goto err_proto;
2667
2668         conntrack_gc_work_init(&conntrack_gc_work);
2669         queue_delayed_work(system_power_efficient_wq, &conntrack_gc_work.dwork, HZ);
2670
2671         return 0;
2672
2673 err_proto:
2674         nf_conntrack_seqadj_fini();
2675 err_seqadj:
2676         nf_conntrack_labels_fini();
2677 err_labels:
2678         nf_conntrack_helper_fini();
2679 err_helper:
2680         nf_conntrack_timeout_fini();
2681 err_timeout:
2682         nf_conntrack_ecache_fini();
2683 err_ecache:
2684         nf_conntrack_tstamp_fini();
2685 err_tstamp:
2686         nf_conntrack_acct_fini();
2687 err_acct:
2688         nf_conntrack_expect_fini();
2689 err_expect:
2690         kmem_cache_destroy(nf_conntrack_cachep);
2691 err_cachep:
2692         kvfree(nf_conntrack_hash);
2693         return ret;
2694 }
2695
2696 static struct nf_ct_hook nf_conntrack_hook = {
2697         .update         = nf_conntrack_update,
2698         .destroy        = destroy_conntrack,
2699         .get_tuple_skb  = nf_conntrack_get_tuple_skb,
2700 };
2701
2702 void nf_conntrack_init_end(void)
2703 {
2704         /* For use by REJECT target */
2705         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
2706         RCU_INIT_POINTER(nf_ct_hook, &nf_conntrack_hook);
2707 }
2708
2709 /*
2710  * We need to use special "null" values, not used in hash table
2711  */
2712 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
2713 #define DYING_NULLS_VAL         ((1<<30)+1)
2714
2715 int nf_conntrack_init_net(struct net *net)
2716 {
2717         struct nf_conntrack_net *cnet = nf_ct_pernet(net);
2718         int ret = -ENOMEM;
2719         int cpu;
2720
2721         BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER);
2722         BUILD_BUG_ON_NOT_POWER_OF_2(CONNTRACK_LOCKS);
2723         atomic_set(&cnet->count, 0);
2724
2725         net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
2726         if (!net->ct.pcpu_lists)
2727                 goto err_stat;
2728
2729         for_each_possible_cpu(cpu) {
2730                 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2731
2732                 spin_lock_init(&pcpu->lock);
2733                 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
2734                 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
2735         }
2736
2737         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
2738         if (!net->ct.stat)
2739                 goto err_pcpu_lists;
2740
2741         ret = nf_conntrack_expect_pernet_init(net);
2742         if (ret < 0)
2743                 goto err_expect;
2744
2745         nf_conntrack_acct_pernet_init(net);
2746         nf_conntrack_tstamp_pernet_init(net);
2747         nf_conntrack_ecache_pernet_init(net);
2748         nf_conntrack_helper_pernet_init(net);
2749         nf_conntrack_proto_pernet_init(net);
2750
2751         return 0;
2752
2753 err_expect:
2754         free_percpu(net->ct.stat);
2755 err_pcpu_lists:
2756         free_percpu(net->ct.pcpu_lists);
2757 err_stat:
2758         return ret;
2759 }