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 extern unsigned int nf_conntrack_net_id;
  59
  60 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
  61 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
  62
  63 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
  64 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
  65
  66 struct hlist_nulls_head *nf_conntrack_hash __read_mostly;
  67 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
  68
  69 struct conntrack_gc_work {
  70         struct delayed_work     dwork;
  71         u32                     last_bucket;
  72         bool                    exiting;
  73         bool                    early_drop;
  74         long                    next_gc_run;
  75 };
  76
  77 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
  78 static DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
  79 static __read_mostly bool nf_conntrack_locks_all;
  80
  81 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
  82 #define GC_MAX_BUCKETS_DIV      128u
  83 /* upper bound of full table scan */
  84 #define GC_MAX_SCAN_JIFFIES     (16u * HZ)
  85 /* desired ratio of entries found to be expired */
  86 #define GC_EVICT_RATIO  50u
  87
  88 static struct conntrack_gc_work conntrack_gc_work;
  89
  90 extern unsigned int nf_conntrack_net_id;
  91
  92 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
  93 {
  94         /* 1) Acquire the lock */
  95         spin_lock(lock);
  96
  97         /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
  98          * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
  99          */
 100         if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false))
 101                 return;
 102
 103         /* fast path failed, unlock */
 104         spin_unlock(lock);
 105
 106         /* Slow path 1) get global lock */
 107         spin_lock(&nf_conntrack_locks_all_lock);
 108
 109         /* Slow path 2) get the lock we want */
 110         spin_lock(lock);
 111
 112         /* Slow path 3) release the global lock */
 113         spin_unlock(&nf_conntrack_locks_all_lock);
 114 }
 115 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
 116
 117 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
 118 {
 119         h1 %= CONNTRACK_LOCKS;
 120         h2 %= CONNTRACK_LOCKS;
 121         spin_unlock(&nf_conntrack_locks[h1]);
 122         if (h1 != h2)
 123                 spin_unlock(&nf_conntrack_locks[h2]);
 124 }
 125
 126 /* return true if we need to recompute hashes (in case hash table was resized) */
 127 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
 128                                      unsigned int h2, unsigned int sequence)
 129 {
 130         h1 %= CONNTRACK_LOCKS;
 131         h2 %= CONNTRACK_LOCKS;
 132         if (h1 <= h2) {
 133                 nf_conntrack_lock(&nf_conntrack_locks[h1]);
 134                 if (h1 != h2)
 135                         spin_lock_nested(&nf_conntrack_locks[h2],
 136                                          SINGLE_DEPTH_NESTING);
 137         } else {
 138                 nf_conntrack_lock(&nf_conntrack_locks[h2]);
 139                 spin_lock_nested(&nf_conntrack_locks[h1],
 140                                  SINGLE_DEPTH_NESTING);
 141         }
 142         if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
 143                 nf_conntrack_double_unlock(h1, h2);
 144                 return true;
 145         }
 146         return false;
 147 }
 148
 149 static void nf_conntrack_all_lock(void)
 150         __acquires(&nf_conntrack_locks_all_lock)
 151 {
 152         int i;
 153
 154         spin_lock(&nf_conntrack_locks_all_lock);
 155
 156         nf_conntrack_locks_all = true;
 157
 158         for (i = 0; i < CONNTRACK_LOCKS; i++) {
 159                 spin_lock(&nf_conntrack_locks[i]);
 160
 161                 /* This spin_unlock provides the "release" to ensure that
 162                  * nf_conntrack_locks_all==true is visible to everyone that
 163                  * acquired spin_lock(&nf_conntrack_locks[]).
 164                  */
 165                 spin_unlock(&nf_conntrack_locks[i]);
 166         }
 167 }
 168
 169 static void nf_conntrack_all_unlock(void)
 170         __releases(&nf_conntrack_locks_all_lock)
 171 {
 172         /* All prior stores must be complete before we clear
 173          * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
 174          * might observe the false value but not the entire
 175          * critical section.
 176          * It pairs with the smp_load_acquire() in nf_conntrack_lock()
 177          */
 178         smp_store_release(&nf_conntrack_locks_all, false);
 179         spin_unlock(&nf_conntrack_locks_all_lock);
 180 }
 181
 182 unsigned int nf_conntrack_htable_size __read_mostly;
 183 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
 184
 185 unsigned int nf_conntrack_max __read_mostly;
 186 EXPORT_SYMBOL_GPL(nf_conntrack_max);
 187 seqcount_spinlock_t nf_conntrack_generation __read_mostly;
 188 static unsigned int nf_conntrack_hash_rnd __read_mostly;
 189
 190 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
 191                               const struct net *net)
 192 {
 193         unsigned int n;
 194         u32 seed;
 195
 196         get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
 197
 198         /* The direction must be ignored, so we hash everything up to the
 199          * destination ports (which is a multiple of 4) and treat the last
 200          * three bytes manually.
 201          */
 202         seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
 203         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
 204         return jhash2((u32 *)tuple, n, seed ^
 205                       (((__force __u16)tuple->dst.u.all << 16) |
 206                       tuple->dst.protonum));
 207 }
 208
 209 static u32 scale_hash(u32 hash)
 210 {
 211         return reciprocal_scale(hash, nf_conntrack_htable_size);
 212 }
 213
 214 static u32 __hash_conntrack(const struct net *net,
 215                             const struct nf_conntrack_tuple *tuple,
 216                             unsigned int size)
 217 {
 218         return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
 219 }
 220
 221 static u32 hash_conntrack(const struct net *net,
 222                           const struct nf_conntrack_tuple *tuple)
 223 {
 224         return scale_hash(hash_conntrack_raw(tuple, net));
 225 }
 226
 227 static bool nf_ct_get_tuple_ports(const struct sk_buff *skb,
 228                                   unsigned int dataoff,
 229                                   struct nf_conntrack_tuple *tuple)
 230 {       struct {
 231                 __be16 sport;
 232                 __be16 dport;
 233         } _inet_hdr, *inet_hdr;
 234
 235         /* Actually only need first 4 bytes to get ports. */
 236         inet_hdr = skb_header_pointer(skb, dataoff, sizeof(_inet_hdr), &_inet_hdr);
 237         if (!inet_hdr)
 238                 return false;
 239
 240         tuple->src.u.udp.port = inet_hdr->sport;
 241         tuple->dst.u.udp.port = inet_hdr->dport;
 242         return true;
 243 }
 244
 245 static bool
 246 nf_ct_get_tuple(const struct sk_buff *skb,
 247                 unsigned int nhoff,
 248                 unsigned int dataoff,
 249                 u_int16_t l3num,
 250                 u_int8_t protonum,
 251                 struct net *net,
 252                 struct nf_conntrack_tuple *tuple)
 253 {
 254         unsigned int size;
 255         const __be32 *ap;
 256         __be32 _addrs[8];
 257
 258         memset(tuple, 0, sizeof(*tuple));
 259
 260         tuple->src.l3num = l3num;
 261         switch (l3num) {
 262         case NFPROTO_IPV4:
 263                 nhoff += offsetof(struct iphdr, saddr);
 264                 size = 2 * sizeof(__be32);
 265                 break;
 266         case NFPROTO_IPV6:
 267                 nhoff += offsetof(struct ipv6hdr, saddr);
 268                 size = sizeof(_addrs);
 269                 break;
 270         default:
 271                 return true;
 272         }
 273
 274         ap = skb_header_pointer(skb, nhoff, size, _addrs);
 275         if (!ap)
 276                 return false;
 277
 278         switch (l3num) {
 279         case NFPROTO_IPV4:
 280                 tuple->src.u3.ip = ap[0];
 281                 tuple->dst.u3.ip = ap[1];
 282                 break;
 283         case NFPROTO_IPV6:
 284                 memcpy(tuple->src.u3.ip6, ap, sizeof(tuple->src.u3.ip6));
 285                 memcpy(tuple->dst.u3.ip6, ap + 4, sizeof(tuple->dst.u3.ip6));
 286                 break;
 287         }
 288
 289         tuple->dst.protonum = protonum;
 290         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 291
 292         switch (protonum) {
 293 #if IS_ENABLED(CONFIG_IPV6)
 294         case IPPROTO_ICMPV6:
 295                 return icmpv6_pkt_to_tuple(skb, dataoff, net, tuple);
 296 #endif
 297         case IPPROTO_ICMP:
 298                 return icmp_pkt_to_tuple(skb, dataoff, net, tuple);
 299 #ifdef CONFIG_NF_CT_PROTO_GRE
 300         case IPPROTO_GRE:
 301                 return gre_pkt_to_tuple(skb, dataoff, net, tuple);
 302 #endif
 303         case IPPROTO_TCP:
 304         case IPPROTO_UDP: /* fallthrough */
 305                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 306 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
 307         case IPPROTO_UDPLITE:
 308                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 309 #endif
 310 #ifdef CONFIG_NF_CT_PROTO_SCTP
 311         case IPPROTO_SCTP:
 312                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 313 #endif
 314 #ifdef CONFIG_NF_CT_PROTO_DCCP
 315         case IPPROTO_DCCP:
 316                 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
 317 #endif
 318         default:
 319                 break;
 320         }
 321
 322         return true;
 323 }
 324
 325 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
 326                             u_int8_t *protonum)
 327 {
 328         int dataoff = -1;
 329         const struct iphdr *iph;
 330         struct iphdr _iph;
 331
 332         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
 333         if (!iph)
 334                 return -1;
 335
 336         /* Conntrack defragments packets, we might still see fragments
 337          * inside ICMP packets though.
 338          */
 339         if (iph->frag_off & htons(IP_OFFSET))
 340                 return -1;
 341
 342         dataoff = nhoff + (iph->ihl << 2);
 343         *protonum = iph->protocol;
 344
 345         /* Check bogus IP headers */
 346         if (dataoff > skb->len) {
 347                 pr_debug("bogus IPv4 packet: nhoff %u, ihl %u, skblen %u\n",
 348                          nhoff, iph->ihl << 2, skb->len);
 349                 return -1;
 350         }
 351         return dataoff;
 352 }
 353
 354 #if IS_ENABLED(CONFIG_IPV6)
 355 static int ipv6_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
 356                             u8 *protonum)
 357 {
 358         int protoff = -1;
 359         unsigned int extoff = nhoff + sizeof(struct ipv6hdr);
 360         __be16 frag_off;
 361         u8 nexthdr;
 362
 363         if (skb_copy_bits(skb, nhoff + offsetof(struct ipv6hdr, nexthdr),
 364                           &nexthdr, sizeof(nexthdr)) != 0) {
 365                 pr_debug("can't get nexthdr\n");
 366                 return -1;
 367         }
 368         protoff = ipv6_skip_exthdr(skb, extoff, &nexthdr, &frag_off);
 369         /*
 370          * (protoff == skb->len) means the packet has not data, just
 371          * IPv6 and possibly extensions headers, but it is tracked anyway
 372          */
 373         if (protoff < 0 || (frag_off & htons(~0x7)) != 0) {
 374                 pr_debug("can't find proto in pkt\n");
 375                 return -1;
 376         }
 377
 378         *protonum = nexthdr;
 379         return protoff;
 380 }
 381 #endif
 382
 383 static int get_l4proto(const struct sk_buff *skb,
 384                        unsigned int nhoff, u8 pf, u8 *l4num)
 385 {
 386         switch (pf) {
 387         case NFPROTO_IPV4:
 388                 return ipv4_get_l4proto(skb, nhoff, l4num);
 389 #if IS_ENABLED(CONFIG_IPV6)
 390         case NFPROTO_IPV6:
 391                 return ipv6_get_l4proto(skb, nhoff, l4num);
 392 #endif
 393         default:
 394                 *l4num = 0;
 395                 break;
 396         }
 397         return -1;
 398 }
 399
 400 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 401                        u_int16_t l3num,
 402                        struct net *net, struct nf_conntrack_tuple *tuple)
 403 {
 404         u8 protonum;
 405         int protoff;
 406
 407         protoff = get_l4proto(skb, nhoff, l3num, &protonum);
 408         if (protoff <= 0)
 409                 return false;
 410
 411         return nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple);
 412 }
 413 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 414
 415 bool
 416 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 417                    const struct nf_conntrack_tuple *orig)
 418 {
 419         memset(inverse, 0, sizeof(*inverse));
 420
 421         inverse->src.l3num = orig->src.l3num;
 422
 423         switch (orig->src.l3num) {
 424         case NFPROTO_IPV4:
 425                 inverse->src.u3.ip = orig->dst.u3.ip;
 426                 inverse->dst.u3.ip = orig->src.u3.ip;
 427                 break;
 428         case NFPROTO_IPV6:
 429                 inverse->src.u3.in6 = orig->dst.u3.in6;
 430                 inverse->dst.u3.in6 = orig->src.u3.in6;
 431                 break;
 432         default:
 433                 break;
 434         }
 435
 436         inverse->dst.dir = !orig->dst.dir;
 437
 438         inverse->dst.protonum = orig->dst.protonum;
 439
 440         switch (orig->dst.protonum) {
 441         case IPPROTO_ICMP:
 442                 return nf_conntrack_invert_icmp_tuple(inverse, orig);
 443 #if IS_ENABLED(CONFIG_IPV6)
 444         case IPPROTO_ICMPV6:
 445                 return nf_conntrack_invert_icmpv6_tuple(inverse, orig);
 446 #endif
 447         }
 448
 449         inverse->src.u.all = orig->dst.u.all;
 450         inverse->dst.u.all = orig->src.u.all;
 451         return true;
 452 }
 453 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 454
 455 /* Generate a almost-unique pseudo-id for a given conntrack.
 456  *
 457  * intentionally doesn't re-use any of the seeds used for hash
 458  * table location, we assume id gets exposed to userspace.
 459  *
 460  * Following nf_conn items do not change throughout lifetime
 461  * of the nf_conn:
 462  *
 463  * 1. nf_conn address
 464  * 2. nf_conn->master address (normally NULL)
 465  * 3. the associated net namespace
 466  * 4. the original direction tuple
 467  */
 468 u32 nf_ct_get_id(const struct nf_conn *ct)
 469 {
 470         static __read_mostly siphash_key_t ct_id_seed;
 471         unsigned long a, b, c, d;
 472
 473         net_get_random_once(&ct_id_seed, sizeof(ct_id_seed));
 474
 475         a = (unsigned long)ct;
 476         b = (unsigned long)ct->master;
 477         c = (unsigned long)nf_ct_net(ct);
 478         d = (unsigned long)siphash(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 479                                    sizeof(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple),
 480                                    &ct_id_seed);
 481 #ifdef CONFIG_64BIT
 482         return siphash_4u64((u64)a, (u64)b, (u64)c, (u64)d, &ct_id_seed);
 483 #else
 484         return siphash_4u32((u32)a, (u32)b, (u32)c, (u32)d, &ct_id_seed);
 485 #endif
 486 }
 487 EXPORT_SYMBOL_GPL(nf_ct_get_id);
 488
 489 static void
 490 clean_from_lists(struct nf_conn *ct)
 491 {
 492         pr_debug("clean_from_lists(%p)\n", ct);
 493         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 494         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 495
 496         /* Destroy all pending expectations */
 497         nf_ct_remove_expectations(ct);
 498 }
 499
 500 /* must be called with local_bh_disable */
 501 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
 502 {
 503         struct ct_pcpu *pcpu;
 504
 505         /* add this conntrack to the (per cpu) dying list */
 506         ct->cpu = smp_processor_id();
 507         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 508
 509         spin_lock(&pcpu->lock);
 510         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 511                              &pcpu->dying);
 512         spin_unlock(&pcpu->lock);
 513 }
 514
 515 /* must be called with local_bh_disable */
 516 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
 517 {
 518         struct ct_pcpu *pcpu;
 519
 520         /* add this conntrack to the (per cpu) unconfirmed list */
 521         ct->cpu = smp_processor_id();
 522         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 523
 524         spin_lock(&pcpu->lock);
 525         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 526                              &pcpu->unconfirmed);
 527         spin_unlock(&pcpu->lock);
 528 }
 529
 530 /* must be called with local_bh_disable */
 531 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
 532 {
 533         struct ct_pcpu *pcpu;
 534
 535         /* We overload first tuple to link into unconfirmed or dying list.*/
 536         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 537
 538         spin_lock(&pcpu->lock);
 539         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 540         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 541         spin_unlock(&pcpu->lock);
 542 }
 543
 544 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
 545
 546 /* Released via destroy_conntrack() */
 547 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
 548                                  const struct nf_conntrack_zone *zone,
 549                                  gfp_t flags)
 550 {
 551         struct nf_conn *tmpl, *p;
 552
 553         if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) {
 554                 tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags);
 555                 if (!tmpl)
 556                         return NULL;
 557
 558                 p = tmpl;
 559                 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
 560                 if (tmpl != p) {
 561                         tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
 562                         tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p;
 563                 }
 564         } else {
 565                 tmpl = kzalloc(sizeof(*tmpl), flags);
 566                 if (!tmpl)
 567                         return NULL;
 568         }
 569
 570         tmpl->status = IPS_TEMPLATE;
 571         write_pnet(&tmpl->ct_net, net);
 572         nf_ct_zone_add(tmpl, zone);
 573         atomic_set(&tmpl->ct_general.use, 0);
 574
 575         return tmpl;
 576 }
 577 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
 578
 579 void nf_ct_tmpl_free(struct nf_conn *tmpl)
 580 {
 581         nf_ct_ext_destroy(tmpl);
 582
 583         if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK)
 584                 kfree((char *)tmpl - tmpl->proto.tmpl_padto);
 585         else
 586                 kfree(tmpl);
 587 }
 588 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
 589
 590 static void destroy_gre_conntrack(struct nf_conn *ct)
 591 {
 592 #ifdef CONFIG_NF_CT_PROTO_GRE
 593         struct nf_conn *master = ct->master;
 594
 595         if (master)
 596                 nf_ct_gre_keymap_destroy(master);
 597 #endif
 598 }
 599
 600 static void
 601 destroy_conntrack(struct nf_conntrack *nfct)
 602 {
 603         struct nf_conn *ct = (struct nf_conn *)nfct;
 604
 605         pr_debug("destroy_conntrack(%p)\n", ct);
 606         WARN_ON(atomic_read(&nfct->use) != 0);
 607
 608         if (unlikely(nf_ct_is_template(ct))) {
 609                 nf_ct_tmpl_free(ct);
 610                 return;
 611         }
 612
 613         if (unlikely(nf_ct_protonum(ct) == IPPROTO_GRE))
 614                 destroy_gre_conntrack(ct);
 615
 616         local_bh_disable();
 617         /* Expectations will have been removed in clean_from_lists,
 618          * except TFTP can create an expectation on the first packet,
 619          * before connection is in the list, so we need to clean here,
 620          * too.
 621          */
 622         nf_ct_remove_expectations(ct);
 623
 624         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 625
 626         local_bh_enable();
 627
 628         if (ct->master)
 629                 nf_ct_put(ct->master);
 630
 631         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 632         nf_conntrack_free(ct);
 633 }
 634
 635 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 636 {
 637         struct net *net = nf_ct_net(ct);
 638         unsigned int hash, reply_hash;
 639         unsigned int sequence;
 640
 641         nf_ct_helper_destroy(ct);
 642
 643         local_bh_disable();
 644         do {
 645                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 646                 hash = hash_conntrack(net,
 647                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 648                 reply_hash = hash_conntrack(net,
 649                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 650         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 651
 652         clean_from_lists(ct);
 653         nf_conntrack_double_unlock(hash, reply_hash);
 654
 655         nf_ct_add_to_dying_list(ct);
 656
 657         local_bh_enable();
 658 }
 659
 660 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 661 {
 662         struct nf_conn_tstamp *tstamp;
 663         struct net *net;
 664
 665         if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
 666                 return false;
 667
 668         tstamp = nf_conn_tstamp_find(ct);
 669         if (tstamp && tstamp->stop == 0)
 670                 tstamp->stop = ktime_get_real_ns();
 671
 672         if (nf_conntrack_event_report(IPCT_DESTROY, ct,
 673                                     portid, report) < 0) {
 674                 /* destroy event was not delivered. nf_ct_put will
 675                  * be done by event cache worker on redelivery.
 676                  */
 677                 nf_ct_delete_from_lists(ct);
 678                 nf_conntrack_ecache_work(nf_ct_net(ct), NFCT_ECACHE_DESTROY_FAIL);
 679                 return false;
 680         }
 681
 682         net = nf_ct_net(ct);
 683         if (nf_conntrack_ecache_dwork_pending(net))
 684                 nf_conntrack_ecache_work(net, NFCT_ECACHE_DESTROY_SENT);
 685         nf_ct_delete_from_lists(ct);
 686         nf_ct_put(ct);
 687         return true;
 688 }
 689 EXPORT_SYMBOL_GPL(nf_ct_delete);
 690
 691 static inline bool
 692 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
 693                 const struct nf_conntrack_tuple *tuple,
 694                 const struct nf_conntrack_zone *zone,
 695                 const struct net *net)
 696 {
 697         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 698
 699         /* A conntrack can be recreated with the equal tuple,
 700          * so we need to check that the conntrack is confirmed
 701          */
 702         return nf_ct_tuple_equal(tuple, &h->tuple) &&
 703                nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
 704                nf_ct_is_confirmed(ct) &&
 705                net_eq(net, nf_ct_net(ct));
 706 }
 707
 708 static inline bool
 709 nf_ct_match(const struct nf_conn *ct1, const struct nf_conn *ct2)
 710 {
 711         return nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 712                                  &ct2->tuplehash[IP_CT_DIR_ORIGINAL].tuple) &&
 713                nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_REPLY].tuple,
 714                                  &ct2->tuplehash[IP_CT_DIR_REPLY].tuple) &&
 715                nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_ORIGINAL) &&
 716                nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_REPLY) &&
 717                net_eq(nf_ct_net(ct1), nf_ct_net(ct2));
 718 }
 719
 720 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
 721 static void nf_ct_gc_expired(struct nf_conn *ct)
 722 {
 723         if (!atomic_inc_not_zero(&ct->ct_general.use))
 724                 return;
 725
 726         if (nf_ct_should_gc(ct))
 727                 nf_ct_kill(ct);
 728
 729         nf_ct_put(ct);
 730 }
 731
 732 /*
 733  * Warning :
 734  * - Caller must take a reference on returned object
 735  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 736  */
 737 static struct nf_conntrack_tuple_hash *
 738 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
 739                       const struct nf_conntrack_tuple *tuple, u32 hash)
 740 {
 741         struct nf_conntrack_tuple_hash *h;
 742         struct hlist_nulls_head *ct_hash;
 743         struct hlist_nulls_node *n;
 744         unsigned int bucket, hsize;
 745
 746 begin:
 747         nf_conntrack_get_ht(&ct_hash, &hsize);
 748         bucket = reciprocal_scale(hash, hsize);
 749
 750         hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
 751                 struct nf_conn *ct;
 752
 753                 ct = nf_ct_tuplehash_to_ctrack(h);
 754                 if (nf_ct_is_expired(ct)) {
 755                         nf_ct_gc_expired(ct);
 756                         continue;
 757                 }
 758
 759                 if (nf_ct_key_equal(h, tuple, zone, net))
 760                         return h;
 761         }
 762         /*
 763          * if the nulls value we got at the end of this lookup is
 764          * not the expected one, we must restart lookup.
 765          * We probably met an item that was moved to another chain.
 766          */
 767         if (get_nulls_value(n) != bucket) {
 768                 NF_CT_STAT_INC_ATOMIC(net, search_restart);
 769                 goto begin;
 770         }
 771
 772         return NULL;
 773 }
 774
 775 /* Find a connection corresponding to a tuple. */
 776 static struct nf_conntrack_tuple_hash *
 777 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 778                         const struct nf_conntrack_tuple *tuple, u32 hash)
 779 {
 780         struct nf_conntrack_tuple_hash *h;
 781         struct nf_conn *ct;
 782
 783         rcu_read_lock();
 784
 785         h = ____nf_conntrack_find(net, zone, tuple, hash);
 786         if (h) {
 787                 /* We have a candidate that matches the tuple we're interested
 788                  * in, try to obtain a reference and re-check tuple
 789                  */
 790                 ct = nf_ct_tuplehash_to_ctrack(h);
 791                 if (likely(atomic_inc_not_zero(&ct->ct_general.use))) {
 792                         if (likely(nf_ct_key_equal(h, tuple, zone, net)))
 793                                 goto found;
 794
 795                         /* TYPESAFE_BY_RCU recycled the candidate */
 796                         nf_ct_put(ct);
 797                 }
 798
 799                 h = NULL;
 800         }
 801 found:
 802         rcu_read_unlock();
 803
 804         return h;
 805 }
 806
 807 struct nf_conntrack_tuple_hash *
 808 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 809                       const struct nf_conntrack_tuple *tuple)
 810 {
 811         return __nf_conntrack_find_get(net, zone, tuple,
 812                                        hash_conntrack_raw(tuple, net));
 813 }
 814 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 815
 816 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 817                                        unsigned int hash,
 818                                        unsigned int reply_hash)
 819 {
 820         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 821                            &nf_conntrack_hash[hash]);
 822         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 823                            &nf_conntrack_hash[reply_hash]);
 824 }
 825
 826 int
 827 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 828 {
 829         const struct nf_conntrack_zone *zone;
 830         struct net *net = nf_ct_net(ct);
 831         unsigned int hash, reply_hash;
 832         struct nf_conntrack_tuple_hash *h;
 833         struct hlist_nulls_node *n;
 834         unsigned int sequence;
 835
 836         zone = nf_ct_zone(ct);
 837
 838         local_bh_disable();
 839         do {
 840                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 841                 hash = hash_conntrack(net,
 842                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 843                 reply_hash = hash_conntrack(net,
 844                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 845         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 846
 847         /* See if there's one in the list already, including reverse */
 848         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
 849                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 850                                     zone, net))
 851                         goto out;
 852
 853         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
 854                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 855                                     zone, net))
 856                         goto out;
 857
 858         smp_wmb();
 859         /* The caller holds a reference to this object */
 860         atomic_set(&ct->ct_general.use, 2);
 861         __nf_conntrack_hash_insert(ct, hash, reply_hash);
 862         nf_conntrack_double_unlock(hash, reply_hash);
 863         NF_CT_STAT_INC(net, insert);
 864         local_bh_enable();
 865         return 0;
 866
 867 out:
 868         nf_conntrack_double_unlock(hash, reply_hash);
 869         local_bh_enable();
 870         return -EEXIST;
 871 }
 872 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 873
 874 void nf_ct_acct_add(struct nf_conn *ct, u32 dir, unsigned int packets,
 875                     unsigned int bytes)
 876 {
 877         struct nf_conn_acct *acct;
 878
 879         acct = nf_conn_acct_find(ct);
 880         if (acct) {
 881                 struct nf_conn_counter *counter = acct->counter;
 882
 883                 atomic64_add(packets, &counter[dir].packets);
 884                 atomic64_add(bytes, &counter[dir].bytes);
 885         }
 886 }
 887 EXPORT_SYMBOL_GPL(nf_ct_acct_add);
 888
 889 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
 890                              const struct nf_conn *loser_ct)
 891 {
 892         struct nf_conn_acct *acct;
 893
 894         acct = nf_conn_acct_find(loser_ct);
 895         if (acct) {
 896                 struct nf_conn_counter *counter = acct->counter;
 897                 unsigned int bytes;
 898
 899                 /* u32 should be fine since we must have seen one packet. */
 900                 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
 901                 nf_ct_acct_update(ct, CTINFO2DIR(ctinfo), bytes);
 902         }
 903 }
 904
 905 static void __nf_conntrack_insert_prepare(struct nf_conn *ct)
 906 {
 907         struct nf_conn_tstamp *tstamp;
 908
 909         atomic_inc(&ct->ct_general.use);
 910         ct->status |= IPS_CONFIRMED;
 911
 912         /* set conntrack timestamp, if enabled. */
 913         tstamp = nf_conn_tstamp_find(ct);
 914         if (tstamp)
 915                 tstamp->start = ktime_get_real_ns();
 916 }
 917
 918 /* caller must hold locks to prevent concurrent changes */
 919 static int __nf_ct_resolve_clash(struct sk_buff *skb,
 920                                  struct nf_conntrack_tuple_hash *h)
 921 {
 922         /* This is the conntrack entry already in hashes that won race. */
 923         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 924         enum ip_conntrack_info ctinfo;
 925         struct nf_conn *loser_ct;
 926
 927         loser_ct = nf_ct_get(skb, &ctinfo);
 928
 929         if (nf_ct_is_dying(ct))
 930                 return NF_DROP;
 931
 932         if (((ct->status & IPS_NAT_DONE_MASK) == 0) ||
 933             nf_ct_match(ct, loser_ct)) {
 934                 struct net *net = nf_ct_net(ct);
 935
 936                 nf_conntrack_get(&ct->ct_general);
 937
 938                 nf_ct_acct_merge(ct, ctinfo, loser_ct);
 939                 nf_ct_add_to_dying_list(loser_ct);
 940                 nf_conntrack_put(&loser_ct->ct_general);
 941                 nf_ct_set(skb, ct, ctinfo);
 942
 943                 NF_CT_STAT_INC(net, clash_resolve);
 944                 return NF_ACCEPT;
 945         }
 946
 947         return NF_DROP;
 948 }
 949
 950 /**
 951  * nf_ct_resolve_clash_harder - attempt to insert clashing conntrack entry
 952  *
 953  * @skb: skb that causes the collision
 954  * @repl_idx: hash slot for reply direction
 955  *
 956  * Called when origin or reply direction had a clash.
 957  * The skb can be handled without packet drop provided the reply direction
 958  * is unique or there the existing entry has the identical tuple in both
 959  * directions.
 960  *
 961  * Caller must hold conntrack table locks to prevent concurrent updates.
 962  *
 963  * Returns NF_DROP if the clash could not be handled.
 964  */
 965 static int nf_ct_resolve_clash_harder(struct sk_buff *skb, u32 repl_idx)
 966 {
 967         struct nf_conn *loser_ct = (struct nf_conn *)skb_nfct(skb);
 968         const struct nf_conntrack_zone *zone;
 969         struct nf_conntrack_tuple_hash *h;
 970         struct hlist_nulls_node *n;
 971         struct net *net;
 972
 973         zone = nf_ct_zone(loser_ct);
 974         net = nf_ct_net(loser_ct);
 975
 976         /* Reply direction must never result in a clash, unless both origin
 977          * and reply tuples are identical.
 978          */
 979         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[repl_idx], hnnode) {
 980                 if (nf_ct_key_equal(h,
 981                                     &loser_ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 982                                     zone, net))
 983                         return __nf_ct_resolve_clash(skb, h);
 984         }
 985
 986         /* We want the clashing entry to go away real soon: 1 second timeout. */
 987         loser_ct->timeout = nfct_time_stamp + HZ;
 988
 989         /* IPS_NAT_CLASH removes the entry automatically on the first
 990          * reply.  Also prevents UDP tracker from moving the entry to
 991          * ASSURED state, i.e. the entry can always be evicted under
 992          * pressure.
 993          */
 994         loser_ct->status |= IPS_FIXED_TIMEOUT | IPS_NAT_CLASH;
 995
 996         __nf_conntrack_insert_prepare(loser_ct);
 997
 998         /* fake add for ORIGINAL dir: we want lookups to only find the entry
 999          * already in the table.  This also hides the clashing entry from
1000          * ctnetlink iteration, i.e. conntrack -L won't show them.
1001          */
1002         hlist_nulls_add_fake(&loser_ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
1003
1004         hlist_nulls_add_head_rcu(&loser_ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
1005                                  &nf_conntrack_hash[repl_idx]);
1006
1007         NF_CT_STAT_INC(net, clash_resolve);
1008         return NF_ACCEPT;
1009 }
1010
1011 /**
1012  * nf_ct_resolve_clash - attempt to handle clash without packet drop
1013  *
1014  * @skb: skb that causes the clash
1015  * @h: tuplehash of the clashing entry already in table
1016  * @reply_hash: hash slot for reply direction
1017  *
1018  * A conntrack entry can be inserted to the connection tracking table
1019  * if there is no existing entry with an identical tuple.
1020  *
1021  * If there is one, @skb (and the assocated, unconfirmed conntrack) has
1022  * to be dropped.  In case @skb is retransmitted, next conntrack lookup
1023  * will find the already-existing entry.
1024  *
1025  * The major problem with such packet drop is the extra delay added by
1026  * the packet loss -- it will take some time for a retransmit to occur
1027  * (or the sender to time out when waiting for a reply).
1028  *
1029  * This function attempts to handle the situation without packet drop.
1030  *
1031  * If @skb has no NAT transformation or if the colliding entries are
1032  * exactly the same, only the to-be-confirmed conntrack entry is discarded
1033  * and @skb is associated with the conntrack entry already in the table.
1034  *
1035  * Failing that, the new, unconfirmed conntrack is still added to the table
1036  * provided that the collision only occurs in the ORIGINAL direction.
1037  * The new entry will be added only in the non-clashing REPLY direction,
1038  * so packets in the ORIGINAL direction will continue to match the existing
1039  * entry.  The new entry will also have a fixed timeout so it expires --
1040  * due to the collision, it will only see reply traffic.
1041  *
1042  * Returns NF_DROP if the clash could not be resolved.
1043  */
1044 static __cold noinline int
1045 nf_ct_resolve_clash(struct sk_buff *skb, struct nf_conntrack_tuple_hash *h,
1046                     u32 reply_hash)
1047 {
1048         /* This is the conntrack entry already in hashes that won race. */
1049         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
1050         const struct nf_conntrack_l4proto *l4proto;
1051         enum ip_conntrack_info ctinfo;
1052         struct nf_conn *loser_ct;
1053         struct net *net;
1054         int ret;
1055
1056         loser_ct = nf_ct_get(skb, &ctinfo);
1057         net = nf_ct_net(loser_ct);
1058
1059         l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
1060         if (!l4proto->allow_clash)
1061                 goto drop;
1062
1063         ret = __nf_ct_resolve_clash(skb, h);
1064         if (ret == NF_ACCEPT)
1065                 return ret;
1066
1067         ret = nf_ct_resolve_clash_harder(skb, reply_hash);
1068         if (ret == NF_ACCEPT)
1069                 return ret;
1070
1071 drop:
1072         nf_ct_add_to_dying_list(loser_ct);
1073         NF_CT_STAT_INC(net, drop);
1074         NF_CT_STAT_INC(net, insert_failed);
1075         return NF_DROP;
1076 }
1077
1078 /* Confirm a connection given skb; places it in hash table */
1079 int
1080 __nf_conntrack_confirm(struct sk_buff *skb)
1081 {
1082         const struct nf_conntrack_zone *zone;
1083         unsigned int hash, reply_hash;
1084         struct nf_conntrack_tuple_hash *h;
1085         struct nf_conn *ct;
1086         struct nf_conn_help *help;
1087         struct hlist_nulls_node *n;
1088         enum ip_conntrack_info ctinfo;
1089         struct net *net;
1090         unsigned int sequence;
1091         int ret = NF_DROP;
1092
1093         ct = nf_ct_get(skb, &ctinfo);
1094         net = nf_ct_net(ct);
1095
1096         /* ipt_REJECT uses nf_conntrack_attach to attach related
1097            ICMP/TCP RST packets in other direction.  Actual packet
1098            which created connection will be IP_CT_NEW or for an
1099            expected connection, IP_CT_RELATED. */
1100         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
1101                 return NF_ACCEPT;
1102
1103         zone = nf_ct_zone(ct);
1104         local_bh_disable();
1105
1106         do {
1107                 sequence = read_seqcount_begin(&nf_conntrack_generation);
1108                 /* reuse the hash saved before */
1109                 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
1110                 hash = scale_hash(hash);
1111                 reply_hash = hash_conntrack(net,
1112                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
1113
1114         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
1115
1116         /* We're not in hash table, and we refuse to set up related
1117          * connections for unconfirmed conns.  But packet copies and
1118          * REJECT will give spurious warnings here.
1119          */
1120
1121         /* Another skb with the same unconfirmed conntrack may
1122          * win the race. This may happen for bridge(br_flood)
1123          * or broadcast/multicast packets do skb_clone with
1124          * unconfirmed conntrack.
1125          */
1126         if (unlikely(nf_ct_is_confirmed(ct))) {
1127                 WARN_ON_ONCE(1);
1128                 nf_conntrack_double_unlock(hash, reply_hash);
1129                 local_bh_enable();
1130                 return NF_DROP;
1131         }
1132
1133         pr_debug("Confirming conntrack %p\n", ct);
1134         /* We have to check the DYING flag after unlink to prevent
1135          * a race against nf_ct_get_next_corpse() possibly called from
1136          * user context, else we insert an already 'dead' hash, blocking
1137          * further use of that particular connection -JM.
1138          */
1139         nf_ct_del_from_dying_or_unconfirmed_list(ct);
1140
1141         if (unlikely(nf_ct_is_dying(ct))) {
1142                 nf_ct_add_to_dying_list(ct);
1143                 NF_CT_STAT_INC(net, insert_failed);
1144                 goto dying;
1145         }
1146
1147         /* See if there's one in the list already, including reverse:
1148            NAT could have grabbed it without realizing, since we're
1149            not in the hash.  If there is, we lost race. */
1150         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
1151                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1152                                     zone, net))
1153                         goto out;
1154
1155         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
1156                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
1157                                     zone, net))
1158                         goto out;
1159
1160         /* Timer relative to confirmation time, not original
1161            setting time, otherwise we'd get timer wrap in
1162            weird delay cases. */
1163         ct->timeout += nfct_time_stamp;
1164
1165         __nf_conntrack_insert_prepare(ct);
1166
1167         /* Since the lookup is lockless, hash insertion must be done after
1168          * starting the timer and setting the CONFIRMED bit. The RCU barriers
1169          * guarantee that no other CPU can find the conntrack before the above
1170          * stores are visible.
1171          */
1172         __nf_conntrack_hash_insert(ct, hash, reply_hash);
1173         nf_conntrack_double_unlock(hash, reply_hash);
1174         local_bh_enable();
1175
1176         help = nfct_help(ct);
1177         if (help && help->helper)
1178                 nf_conntrack_event_cache(IPCT_HELPER, ct);
1179
1180         nf_conntrack_event_cache(master_ct(ct) ?
1181                                  IPCT_RELATED : IPCT_NEW, ct);
1182         return NF_ACCEPT;
1183
1184 out:
1185         ret = nf_ct_resolve_clash(skb, h, reply_hash);
1186 dying:
1187         nf_conntrack_double_unlock(hash, reply_hash);
1188         local_bh_enable();
1189         return ret;
1190 }
1191 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
1192
1193 /* Returns true if a connection correspondings to the tuple (required
1194    for NAT). */
1195 int
1196 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
1197                          const struct nf_conn *ignored_conntrack)
1198 {
1199         struct net *net = nf_ct_net(ignored_conntrack);
1200         const struct nf_conntrack_zone *zone;
1201         struct nf_conntrack_tuple_hash *h;
1202         struct hlist_nulls_head *ct_hash;
1203         unsigned int hash, hsize;
1204         struct hlist_nulls_node *n;
1205         struct nf_conn *ct;
1206
1207         zone = nf_ct_zone(ignored_conntrack);
1208
1209         rcu_read_lock();
1210  begin:
1211         nf_conntrack_get_ht(&ct_hash, &hsize);
1212         hash = __hash_conntrack(net, tuple, hsize);
1213
1214         hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
1215                 ct = nf_ct_tuplehash_to_ctrack(h);
1216
1217                 if (ct == ignored_conntrack)
1218                         continue;
1219
1220                 if (nf_ct_is_expired(ct)) {
1221                         nf_ct_gc_expired(ct);
1222                         continue;
1223                 }
1224
1225                 if (nf_ct_key_equal(h, tuple, zone, net)) {
1226                         /* Tuple is taken already, so caller will need to find
1227                          * a new source port to use.
1228                          *
1229                          * Only exception:
1230                          * If the *original tuples* are identical, then both
1231                          * conntracks refer to the same flow.
1232                          * This is a rare situation, it can occur e.g. when
1233                          * more than one UDP packet is sent from same socket
1234                          * in different threads.
1235                          *
1236                          * Let nf_ct_resolve_clash() deal with this later.
1237                          */
1238                         if (nf_ct_tuple_equal(&ignored_conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1239                                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple) &&
1240                                               nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL))
1241                                 continue;
1242
1243                         NF_CT_STAT_INC_ATOMIC(net, found);
1244                         rcu_read_unlock();
1245                         return 1;
1246                 }
1247         }
1248
1249         if (get_nulls_value(n) != hash) {
1250                 NF_CT_STAT_INC_ATOMIC(net, search_restart);
1251                 goto begin;
1252         }
1253
1254         rcu_read_unlock();
1255
1256         return 0;
1257 }
1258 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
1259
1260 #define NF_CT_EVICTION_RANGE    8
1261
1262 /* There's a small race here where we may free a just-assured
1263    connection.  Too bad: we're in trouble anyway. */
1264 static unsigned int early_drop_list(struct net *net,
1265                                     struct hlist_nulls_head *head)
1266 {
1267         struct nf_conntrack_tuple_hash *h;
1268         struct hlist_nulls_node *n;
1269         unsigned int drops = 0;
1270         struct nf_conn *tmp;
1271
1272         hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
1273                 tmp = nf_ct_tuplehash_to_ctrack(h);
1274
1275                 if (test_bit(IPS_OFFLOAD_BIT, &tmp->status))
1276                         continue;
1277
1278                 if (nf_ct_is_expired(tmp)) {
1279                         nf_ct_gc_expired(tmp);
1280                         continue;
1281                 }
1282
1283                 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
1284                     !net_eq(nf_ct_net(tmp), net) ||
1285                     nf_ct_is_dying(tmp))
1286                         continue;
1287
1288                 if (!atomic_inc_not_zero(&tmp->ct_general.use))
1289                         continue;
1290
1291                 /* kill only if still in same netns -- might have moved due to
1292                  * SLAB_TYPESAFE_BY_RCU rules.
1293                  *
1294                  * We steal the timer reference.  If that fails timer has
1295                  * already fired or someone else deleted it. Just drop ref
1296                  * and move to next entry.
1297                  */
1298                 if (net_eq(nf_ct_net(tmp), net) &&
1299                     nf_ct_is_confirmed(tmp) &&
1300                     nf_ct_delete(tmp, 0, 0))
1301                         drops++;
1302
1303                 nf_ct_put(tmp);
1304         }
1305
1306         return drops;
1307 }
1308
1309 static noinline int early_drop(struct net *net, unsigned int hash)
1310 {
1311         unsigned int i, bucket;
1312
1313         for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
1314                 struct hlist_nulls_head *ct_hash;
1315                 unsigned int hsize, drops;
1316
1317                 rcu_read_lock();
1318                 nf_conntrack_get_ht(&ct_hash, &hsize);
1319                 if (!i)
1320                         bucket = reciprocal_scale(hash, hsize);
1321                 else
1322                         bucket = (bucket + 1) % hsize;
1323
1324                 drops = early_drop_list(net, &ct_hash[bucket]);
1325                 rcu_read_unlock();
1326
1327                 if (drops) {
1328                         NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
1329                         return true;
1330                 }
1331         }
1332
1333         return false;
1334 }
1335
1336 static bool gc_worker_skip_ct(const struct nf_conn *ct)
1337 {
1338         return !nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct);
1339 }
1340
1341 static bool gc_worker_can_early_drop(const struct nf_conn *ct)
1342 {
1343         const struct nf_conntrack_l4proto *l4proto;
1344
1345         if (!test_bit(IPS_ASSURED_BIT, &ct->status))
1346                 return true;
1347
1348         l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
1349         if (l4proto->can_early_drop && l4proto->can_early_drop(ct))
1350                 return true;
1351
1352         return false;
1353 }
1354
1355 static void gc_worker(struct work_struct *work)
1356 {
1357         unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u);
1358         unsigned int i, goal, buckets = 0, expired_count = 0;
1359         unsigned int nf_conntrack_max95 = 0;
1360         struct conntrack_gc_work *gc_work;
1361         unsigned int ratio, scanned = 0;
1362         unsigned long next_run;
1363
1364         gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
1365
1366         goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV;
1367         i = gc_work->last_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                 unsigned int hashsz;
1376                 struct nf_conn *tmp;
1377
1378                 i++;
1379                 rcu_read_lock();
1380
1381                 nf_conntrack_get_ht(&ct_hash, &hashsz);
1382                 if (i >= hashsz)
1383                         i = 0;
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                         scanned++;
1392                         if (test_bit(IPS_OFFLOAD_BIT, &tmp->status)) {
1393                                 nf_ct_offload_timeout(tmp);
1394                                 continue;
1395                         }
1396
1397                         if (nf_ct_is_expired(tmp)) {
1398                                 nf_ct_gc_expired(tmp);
1399                                 expired_count++;
1400                                 continue;
1401                         }
1402
1403                         if (nf_conntrack_max95 == 0 || gc_worker_skip_ct(tmp))
1404                                 continue;
1405
1406                         net = nf_ct_net(tmp);
1407                         cnet = net_generic(net, nf_conntrack_net_id);
1408                         if (atomic_read(&cnet->count) < nf_conntrack_max95)
1409                                 continue;
1410
1411                         /* need to take reference to avoid possible races */
1412                         if (!atomic_inc_not_zero(&tmp->ct_general.use))
1413                                 continue;
1414
1415                         if (gc_worker_skip_ct(tmp)) {
1416                                 nf_ct_put(tmp);
1417                                 continue;
1418                         }
1419
1420                         if (gc_worker_can_early_drop(tmp))
1421                                 nf_ct_kill(tmp);
1422
1423                         nf_ct_put(tmp);
1424                 }
1425
1426                 /* could check get_nulls_value() here and restart if ct
1427                  * was moved to another chain.  But given gc is best-effort
1428                  * we will just continue with next hash slot.
1429                  */
1430                 rcu_read_unlock();
1431                 cond_resched();
1432         } while (++buckets < goal);
1433
1434         if (gc_work->exiting)
1435                 return;
1436
1437         /*
1438          * Eviction will normally happen from the packet path, and not
1439          * from this gc worker.
1440          *
1441          * This worker is only here to reap expired entries when system went
1442          * idle after a busy period.
1443          *
1444          * The heuristics below are supposed to balance conflicting goals:
1445          *
1446          * 1. Minimize time until we notice a stale entry
1447          * 2. Maximize scan intervals to not waste cycles
1448          *
1449          * Normally, expire ratio will be close to 0.
1450          *
1451          * As soon as a sizeable fraction of the entries have expired
1452          * increase scan frequency.
1453          */
1454         ratio = scanned ? expired_count * 100 / scanned : 0;
1455         if (ratio > GC_EVICT_RATIO) {
1456                 gc_work->next_gc_run = min_interval;
1457         } else {
1458                 unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV;
1459
1460                 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0);
1461
1462                 gc_work->next_gc_run += min_interval;
1463                 if (gc_work->next_gc_run > max)
1464                         gc_work->next_gc_run = max;
1465         }
1466
1467         next_run = gc_work->next_gc_run;
1468         gc_work->last_bucket = i;
1469         gc_work->early_drop = false;
1470         queue_delayed_work(system_power_efficient_wq, &gc_work->dwork, next_run);
1471 }
1472
1473 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1474 {
1475         INIT_DEFERRABLE_WORK(&gc_work->dwork, gc_worker);
1476         gc_work->next_gc_run = HZ;
1477         gc_work->exiting = false;
1478 }
1479
1480 static struct nf_conn *
1481 __nf_conntrack_alloc(struct net *net,
1482                      const struct nf_conntrack_zone *zone,
1483                      const struct nf_conntrack_tuple *orig,
1484                      const struct nf_conntrack_tuple *repl,
1485                      gfp_t gfp, u32 hash)
1486 {
1487         struct nf_conntrack_net *cnet = net_generic(net, nf_conntrack_net_id);
1488         unsigned int ct_count;
1489         struct nf_conn *ct;
1490
1491         /* We don't want any race condition at early drop stage */
1492         ct_count = atomic_inc_return(&cnet->count);
1493
1494         if (nf_conntrack_max && unlikely(ct_count > nf_conntrack_max)) {
1495                 if (!early_drop(net, hash)) {
1496                         if (!conntrack_gc_work.early_drop)
1497                                 conntrack_gc_work.early_drop = true;
1498                         atomic_dec(&cnet->count);
1499                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1500                         return ERR_PTR(-ENOMEM);
1501                 }
1502         }
1503
1504         /*
1505          * Do not use kmem_cache_zalloc(), as this cache uses
1506          * SLAB_TYPESAFE_BY_RCU.
1507          */
1508         ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1509         if (ct == NULL)
1510                 goto out;
1511
1512         spin_lock_init(&ct->lock);
1513         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1514         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1515         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1516         /* save hash for reusing when confirming */
1517         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1518         ct->status = 0;
1519         ct->timeout = 0;
1520         write_pnet(&ct->ct_net, net);
1521         memset(&ct->__nfct_init_offset, 0,
1522                offsetof(struct nf_conn, proto) -
1523                offsetof(struct nf_conn, __nfct_init_offset));
1524
1525         nf_ct_zone_add(ct, zone);
1526
1527         /* Because we use RCU lookups, we set ct_general.use to zero before
1528          * this is inserted in any list.
1529          */
1530         atomic_set(&ct->ct_general.use, 0);
1531         return ct;
1532 out:
1533         atomic_dec(&cnet->count);
1534         return ERR_PTR(-ENOMEM);
1535 }
1536
1537 struct nf_conn *nf_conntrack_alloc(struct net *net,
1538                                    const struct nf_conntrack_zone *zone,
1539                                    const struct nf_conntrack_tuple *orig,
1540                                    const struct nf_conntrack_tuple *repl,
1541                                    gfp_t gfp)
1542 {
1543         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1544 }
1545 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1546
1547 void nf_conntrack_free(struct nf_conn *ct)
1548 {
1549         struct net *net = nf_ct_net(ct);
1550         struct nf_conntrack_net *cnet;
1551
1552         /* A freed object has refcnt == 0, that's
1553          * the golden rule for SLAB_TYPESAFE_BY_RCU
1554          */
1555         WARN_ON(atomic_read(&ct->ct_general.use) != 0);
1556
1557         nf_ct_ext_destroy(ct);
1558         kmem_cache_free(nf_conntrack_cachep, ct);
1559         cnet = net_generic(net, nf_conntrack_net_id);
1560
1561         smp_mb__before_atomic();
1562         atomic_dec(&cnet->count);
1563 }
1564 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1565
1566
1567 /* Allocate a new conntrack: we return -ENOMEM if classification
1568    failed due to stress.  Otherwise it really is unclassifiable. */
1569 static noinline struct nf_conntrack_tuple_hash *
1570 init_conntrack(struct net *net, struct nf_conn *tmpl,
1571                const struct nf_conntrack_tuple *tuple,
1572                struct sk_buff *skb,
1573                unsigned int dataoff, u32 hash)
1574 {
1575         struct nf_conn *ct;
1576         struct nf_conn_help *help;
1577         struct nf_conntrack_tuple repl_tuple;
1578         struct nf_conntrack_ecache *ecache;
1579         struct nf_conntrack_expect *exp = NULL;
1580         const struct nf_conntrack_zone *zone;
1581         struct nf_conn_timeout *timeout_ext;
1582         struct nf_conntrack_zone tmp;
1583         struct nf_conntrack_net *cnet;
1584
1585         if (!nf_ct_invert_tuple(&repl_tuple, tuple)) {
1586                 pr_debug("Can't invert tuple.\n");
1587                 return NULL;
1588         }
1589
1590         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1591         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1592                                   hash);
1593         if (IS_ERR(ct))
1594                 return (struct nf_conntrack_tuple_hash *)ct;
1595
1596         if (!nf_ct_add_synproxy(ct, tmpl)) {
1597                 nf_conntrack_free(ct);
1598                 return ERR_PTR(-ENOMEM);
1599         }
1600
1601         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1602
1603         if (timeout_ext)
1604                 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1605                                       GFP_ATOMIC);
1606
1607         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1608         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1609         nf_ct_labels_ext_add(ct);
1610
1611         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1612         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1613                                  ecache ? ecache->expmask : 0,
1614                              GFP_ATOMIC);
1615
1616         local_bh_disable();
1617         cnet = net_generic(net, nf_conntrack_net_id);
1618         if (cnet->expect_count) {
1619                 spin_lock(&nf_conntrack_expect_lock);
1620                 exp = nf_ct_find_expectation(net, zone, tuple);
1621                 if (exp) {
1622                         pr_debug("expectation arrives ct=%p exp=%p\n",
1623                                  ct, exp);
1624                         /* Welcome, Mr. Bond.  We've been expecting you... */
1625                         __set_bit(IPS_EXPECTED_BIT, &ct->status);
1626                         /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1627                         ct->master = exp->master;
1628                         if (exp->helper) {
1629                                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
1630                                 if (help)
1631                                         rcu_assign_pointer(help->helper, exp->helper);
1632                         }
1633
1634 #ifdef CONFIG_NF_CONNTRACK_MARK
1635                         ct->mark = exp->master->mark;
1636 #endif
1637 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1638                         ct->secmark = exp->master->secmark;
1639 #endif
1640                         NF_CT_STAT_INC(net, expect_new);
1641                 }
1642                 spin_unlock(&nf_conntrack_expect_lock);
1643         }
1644         if (!exp)
1645                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1646
1647         /* Now it is inserted into the unconfirmed list, bump refcount */
1648         nf_conntrack_get(&ct->ct_general);
1649         nf_ct_add_to_unconfirmed_list(ct);
1650
1651         local_bh_enable();
1652
1653         if (exp) {
1654                 if (exp->expectfn)
1655                         exp->expectfn(ct, exp);
1656                 nf_ct_expect_put(exp);
1657         }
1658
1659         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1660 }
1661
1662 /* On success, returns 0, sets skb->_nfct | ctinfo */
1663 static int
1664 resolve_normal_ct(struct nf_conn *tmpl,
1665                   struct sk_buff *skb,
1666                   unsigned int dataoff,
1667                   u_int8_t protonum,
1668                   const struct nf_hook_state *state)
1669 {
1670         const struct nf_conntrack_zone *zone;
1671         struct nf_conntrack_tuple tuple;
1672         struct nf_conntrack_tuple_hash *h;
1673         enum ip_conntrack_info ctinfo;
1674         struct nf_conntrack_zone tmp;
1675         struct nf_conn *ct;
1676         u32 hash;
1677
1678         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1679                              dataoff, state->pf, protonum, state->net,
1680                              &tuple)) {
1681                 pr_debug("Can't get tuple\n");
1682                 return 0;
1683         }
1684
1685         /* look for tuple match */
1686         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1687         hash = hash_conntrack_raw(&tuple, state->net);
1688         h = __nf_conntrack_find_get(state->net, zone, &tuple, hash);
1689         if (!h) {
1690                 h = init_conntrack(state->net, tmpl, &tuple,
1691                                    skb, dataoff, hash);
1692                 if (!h)
1693                         return 0;
1694                 if (IS_ERR(h))
1695                         return PTR_ERR(h);
1696         }
1697         ct = nf_ct_tuplehash_to_ctrack(h);
1698
1699         /* It exists; we have (non-exclusive) reference. */
1700         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1701                 ctinfo = IP_CT_ESTABLISHED_REPLY;
1702         } else {
1703                 /* Once we've had two way comms, always ESTABLISHED. */
1704                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1705                         pr_debug("normal packet for %p\n", ct);
1706                         ctinfo = IP_CT_ESTABLISHED;
1707                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1708                         pr_debug("related packet for %p\n", ct);
1709                         ctinfo = IP_CT_RELATED;
1710                 } else {
1711                         pr_debug("new packet for %p\n", ct);
1712                         ctinfo = IP_CT_NEW;
1713                 }
1714         }
1715         nf_ct_set(skb, ct, ctinfo);
1716         return 0;
1717 }
1718
1719 /*
1720  * icmp packets need special treatment to handle error messages that are
1721  * related to a connection.
1722  *
1723  * Callers need to check if skb has a conntrack assigned when this
1724  * helper returns; in such case skb belongs to an already known connection.
1725  */
1726 static unsigned int __cold
1727 nf_conntrack_handle_icmp(struct nf_conn *tmpl,
1728                          struct sk_buff *skb,
1729                          unsigned int dataoff,
1730                          u8 protonum,
1731                          const struct nf_hook_state *state)
1732 {
1733         int ret;
1734
1735         if (state->pf == NFPROTO_IPV4 && protonum == IPPROTO_ICMP)
1736                 ret = nf_conntrack_icmpv4_error(tmpl, skb, dataoff, state);
1737 #if IS_ENABLED(CONFIG_IPV6)
1738         else if (state->pf == NFPROTO_IPV6 && protonum == IPPROTO_ICMPV6)
1739                 ret = nf_conntrack_icmpv6_error(tmpl, skb, dataoff, state);
1740 #endif
1741         else
1742                 return NF_ACCEPT;
1743
1744         if (ret <= 0)
1745                 NF_CT_STAT_INC_ATOMIC(state->net, error);
1746
1747         return ret;
1748 }
1749
1750 static int generic_packet(struct nf_conn *ct, struct sk_buff *skb,
1751                           enum ip_conntrack_info ctinfo)
1752 {
1753         const unsigned int *timeout = nf_ct_timeout_lookup(ct);
1754
1755         if (!timeout)
1756                 timeout = &nf_generic_pernet(nf_ct_net(ct))->timeout;
1757
1758         nf_ct_refresh_acct(ct, ctinfo, skb, *timeout);
1759         return NF_ACCEPT;
1760 }
1761
1762 /* Returns verdict for packet, or -1 for invalid. */
1763 static int nf_conntrack_handle_packet(struct nf_conn *ct,
1764                                       struct sk_buff *skb,
1765                                       unsigned int dataoff,
1766                                       enum ip_conntrack_info ctinfo,
1767                                       const struct nf_hook_state *state)
1768 {
1769         switch (nf_ct_protonum(ct)) {
1770         case IPPROTO_TCP:
1771                 return nf_conntrack_tcp_packet(ct, skb, dataoff,
1772                                                ctinfo, state);
1773         case IPPROTO_UDP:
1774                 return nf_conntrack_udp_packet(ct, skb, dataoff,
1775                                                ctinfo, state);
1776         case IPPROTO_ICMP:
1777                 return nf_conntrack_icmp_packet(ct, skb, ctinfo, state);
1778 #if IS_ENABLED(CONFIG_IPV6)
1779         case IPPROTO_ICMPV6:
1780                 return nf_conntrack_icmpv6_packet(ct, skb, ctinfo, state);
1781 #endif
1782 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
1783         case IPPROTO_UDPLITE:
1784                 return nf_conntrack_udplite_packet(ct, skb, dataoff,
1785                                                    ctinfo, state);
1786 #endif
1787 #ifdef CONFIG_NF_CT_PROTO_SCTP
1788         case IPPROTO_SCTP:
1789                 return nf_conntrack_sctp_packet(ct, skb, dataoff,
1790                                                 ctinfo, state);
1791 #endif
1792 #ifdef CONFIG_NF_CT_PROTO_DCCP
1793         case IPPROTO_DCCP:
1794                 return nf_conntrack_dccp_packet(ct, skb, dataoff,
1795                                                 ctinfo, state);
1796 #endif
1797 #ifdef CONFIG_NF_CT_PROTO_GRE
1798         case IPPROTO_GRE:
1799                 return nf_conntrack_gre_packet(ct, skb, dataoff,
1800                                                ctinfo, state);
1801 #endif
1802         }
1803
1804         return generic_packet(ct, skb, ctinfo);
1805 }
1806
1807 unsigned int
1808 nf_conntrack_in(struct sk_buff *skb, const struct nf_hook_state *state)
1809 {
1810         enum ip_conntrack_info ctinfo;
1811         struct nf_conn *ct, *tmpl;
1812         u_int8_t protonum;
1813         int dataoff, ret;
1814
1815         tmpl = nf_ct_get(skb, &ctinfo);
1816         if (tmpl || ctinfo == IP_CT_UNTRACKED) {
1817                 /* Previously seen (loopback or untracked)?  Ignore. */
1818                 if ((tmpl && !nf_ct_is_template(tmpl)) ||
1819                      ctinfo == IP_CT_UNTRACKED)
1820                         return NF_ACCEPT;
1821                 skb->_nfct = 0;
1822         }
1823
1824         /* rcu_read_lock()ed by nf_hook_thresh */
1825         dataoff = get_l4proto(skb, skb_network_offset(skb), state->pf, &protonum);
1826         if (dataoff <= 0) {
1827                 pr_debug("not prepared to track yet or error occurred\n");
1828                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1829                 ret = NF_ACCEPT;
1830                 goto out;
1831         }
1832
1833         if (protonum == IPPROTO_ICMP || protonum == IPPROTO_ICMPV6) {
1834                 ret = nf_conntrack_handle_icmp(tmpl, skb, dataoff,
1835                                                protonum, state);
1836                 if (ret <= 0) {
1837                         ret = -ret;
1838                         goto out;
1839                 }
1840                 /* ICMP[v6] protocol trackers may assign one conntrack. */
1841                 if (skb->_nfct)
1842                         goto out;
1843         }
1844 repeat:
1845         ret = resolve_normal_ct(tmpl, skb, dataoff,
1846                                 protonum, state);
1847         if (ret < 0) {
1848                 /* Too stressed to deal. */
1849                 NF_CT_STAT_INC_ATOMIC(state->net, drop);
1850                 ret = NF_DROP;
1851                 goto out;
1852         }
1853
1854         ct = nf_ct_get(skb, &ctinfo);
1855         if (!ct) {
1856                 /* Not valid part of a connection */
1857                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1858                 ret = NF_ACCEPT;
1859                 goto out;
1860         }
1861
1862         ret = nf_conntrack_handle_packet(ct, skb, dataoff, ctinfo, state);
1863         if (ret <= 0) {
1864                 /* Invalid: inverse of the return code tells
1865                  * the netfilter core what to do */
1866                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1867                 nf_conntrack_put(&ct->ct_general);
1868                 skb->_nfct = 0;
1869                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1870                 if (ret == -NF_DROP)
1871                         NF_CT_STAT_INC_ATOMIC(state->net, drop);
1872                 /* Special case: TCP tracker reports an attempt to reopen a
1873                  * closed/aborted connection. We have to go back and create a
1874                  * fresh conntrack.
1875                  */
1876                 if (ret == -NF_REPEAT)
1877                         goto repeat;
1878                 ret = -ret;
1879                 goto out;
1880         }
1881
1882         if (ctinfo == IP_CT_ESTABLISHED_REPLY &&
1883             !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1884                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1885 out:
1886         if (tmpl)
1887                 nf_ct_put(tmpl);
1888
1889         return ret;
1890 }
1891 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1892
1893 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1894    implicitly racy: see __nf_conntrack_confirm */
1895 void nf_conntrack_alter_reply(struct nf_conn *ct,
1896                               const struct nf_conntrack_tuple *newreply)
1897 {
1898         struct nf_conn_help *help = nfct_help(ct);
1899
1900         /* Should be unconfirmed, so not in hash table yet */
1901         WARN_ON(nf_ct_is_confirmed(ct));
1902
1903         pr_debug("Altering reply tuple of %p to ", ct);
1904         nf_ct_dump_tuple(newreply);
1905
1906         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1907         if (ct->master || (help && !hlist_empty(&help->expectations)))
1908                 return;
1909
1910         rcu_read_lock();
1911         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1912         rcu_read_unlock();
1913 }
1914 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1915
1916 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1917 void __nf_ct_refresh_acct(struct nf_conn *ct,
1918                           enum ip_conntrack_info ctinfo,
1919                           const struct sk_buff *skb,
1920                           u32 extra_jiffies,
1921                           bool do_acct)
1922 {
1923         /* Only update if this is not a fixed timeout */
1924         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1925                 goto acct;
1926
1927         /* If not in hash table, timer will not be active yet */
1928         if (nf_ct_is_confirmed(ct))
1929                 extra_jiffies += nfct_time_stamp;
1930
1931         if (READ_ONCE(ct->timeout) != extra_jiffies)
1932                 WRITE_ONCE(ct->timeout, extra_jiffies);
1933 acct:
1934         if (do_acct)
1935                 nf_ct_acct_update(ct, CTINFO2DIR(ctinfo), skb->len);
1936 }
1937 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1938
1939 bool nf_ct_kill_acct(struct nf_conn *ct,
1940                      enum ip_conntrack_info ctinfo,
1941                      const struct sk_buff *skb)
1942 {
1943         nf_ct_acct_update(ct, CTINFO2DIR(ctinfo), skb->len);
1944
1945         return nf_ct_delete(ct, 0, 0);
1946 }
1947 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1948
1949 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1950
1951 #include <linux/netfilter/nfnetlink.h>
1952 #include <linux/netfilter/nfnetlink_conntrack.h>
1953 #include <linux/mutex.h>
1954
1955 /* Generic function for tcp/udp/sctp/dccp and alike. */
1956 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1957                                const struct nf_conntrack_tuple *tuple)
1958 {
1959         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1960             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1961                 goto nla_put_failure;
1962         return 0;
1963
1964 nla_put_failure:
1965         return -1;
1966 }
1967 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1968
1969 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1970         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1971         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1972 };
1973 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1974
1975 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1976                                struct nf_conntrack_tuple *t,
1977                                u_int32_t flags)
1978 {
1979         if (flags & CTA_FILTER_FLAG(CTA_PROTO_SRC_PORT)) {
1980                 if (!tb[CTA_PROTO_SRC_PORT])
1981                         return -EINVAL;
1982
1983                 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1984         }
1985
1986         if (flags & CTA_FILTER_FLAG(CTA_PROTO_DST_PORT)) {
1987                 if (!tb[CTA_PROTO_DST_PORT])
1988                         return -EINVAL;
1989
1990                 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1991         }
1992
1993         return 0;
1994 }
1995 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1996
1997 unsigned int nf_ct_port_nlattr_tuple_size(void)
1998 {
1999         static unsigned int size __read_mostly;
2000
2001         if (!size)
2002                 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
2003
2004         return size;
2005 }
2006 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
2007 #endif
2008
2009 /* Used by ipt_REJECT and ip6t_REJECT. */
2010 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
2011 {
2012         struct nf_conn *ct;
2013         enum ip_conntrack_info ctinfo;
2014
2015         /* This ICMP is in reverse direction to the packet which caused it */
2016         ct = nf_ct_get(skb, &ctinfo);
2017         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
2018                 ctinfo = IP_CT_RELATED_REPLY;
2019         else
2020                 ctinfo = IP_CT_RELATED;
2021
2022         /* Attach to new skbuff, and increment count */
2023         nf_ct_set(nskb, ct, ctinfo);
2024         nf_conntrack_get(skb_nfct(nskb));
2025 }
2026
2027 static int __nf_conntrack_update(struct net *net, struct sk_buff *skb,
2028                                  struct nf_conn *ct,
2029                                  enum ip_conntrack_info ctinfo)
2030 {
2031         struct nf_conntrack_tuple_hash *h;
2032         struct nf_conntrack_tuple tuple;
2033         struct nf_nat_hook *nat_hook;
2034         unsigned int status;
2035         int dataoff;
2036         u16 l3num;
2037         u8 l4num;
2038
2039         l3num = nf_ct_l3num(ct);
2040
2041         dataoff = get_l4proto(skb, skb_network_offset(skb), l3num, &l4num);
2042         if (dataoff <= 0)
2043                 return -1;
2044
2045         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
2046                              l4num, net, &tuple))
2047                 return -1;
2048
2049         if (ct->status & IPS_SRC_NAT) {
2050                 memcpy(tuple.src.u3.all,
2051                        ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.all,
2052                        sizeof(tuple.src.u3.all));
2053                 tuple.src.u.all =
2054                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.all;
2055         }
2056
2057         if (ct->status & IPS_DST_NAT) {
2058                 memcpy(tuple.dst.u3.all,
2059                        ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.all,
2060                        sizeof(tuple.dst.u3.all));
2061                 tuple.dst.u.all =
2062                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u.all;
2063         }
2064
2065         h = nf_conntrack_find_get(net, nf_ct_zone(ct), &tuple);
2066         if (!h)
2067                 return 0;
2068
2069         /* Store status bits of the conntrack that is clashing to re-do NAT
2070          * mangling according to what it has been done already to this packet.
2071          */
2072         status = ct->status;
2073
2074         nf_ct_put(ct);
2075         ct = nf_ct_tuplehash_to_ctrack(h);
2076         nf_ct_set(skb, ct, ctinfo);
2077
2078         nat_hook = rcu_dereference(nf_nat_hook);
2079         if (!nat_hook)
2080                 return 0;
2081
2082         if (status & IPS_SRC_NAT &&
2083             nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_SRC,
2084                                 IP_CT_DIR_ORIGINAL) == NF_DROP)
2085                 return -1;
2086
2087         if (status & IPS_DST_NAT &&
2088             nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_DST,
2089                                 IP_CT_DIR_ORIGINAL) == NF_DROP)
2090                 return -1;
2091
2092         return 0;
2093 }
2094
2095 /* This packet is coming from userspace via nf_queue, complete the packet
2096  * processing after the helper invocation in nf_confirm().
2097  */
2098 static int nf_confirm_cthelper(struct sk_buff *skb, struct nf_conn *ct,
2099                                enum ip_conntrack_info ctinfo)
2100 {
2101         const struct nf_conntrack_helper *helper;
2102         const struct nf_conn_help *help;
2103         int protoff;
2104
2105         help = nfct_help(ct);
2106         if (!help)
2107                 return 0;
2108
2109         helper = rcu_dereference(help->helper);
2110         if (!(helper->flags & NF_CT_HELPER_F_USERSPACE))
2111                 return 0;
2112
2113         switch (nf_ct_l3num(ct)) {
2114         case NFPROTO_IPV4:
2115                 protoff = skb_network_offset(skb) + ip_hdrlen(skb);
2116                 break;
2117 #if IS_ENABLED(CONFIG_IPV6)
2118         case NFPROTO_IPV6: {
2119                 __be16 frag_off;
2120                 u8 pnum;
2121
2122                 pnum = ipv6_hdr(skb)->nexthdr;
2123                 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &pnum,
2124                                            &frag_off);
2125                 if (protoff < 0 || (frag_off & htons(~0x7)) != 0)
2126                         return 0;
2127                 break;
2128         }
2129 #endif
2130         default:
2131                 return 0;
2132         }
2133
2134         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
2135             !nf_is_loopback_packet(skb)) {
2136                 if (!nf_ct_seq_adjust(skb, ct, ctinfo, protoff)) {
2137                         NF_CT_STAT_INC_ATOMIC(nf_ct_net(ct), drop);
2138                         return -1;
2139                 }
2140         }
2141
2142         /* We've seen it coming out the other side: confirm it */
2143         return nf_conntrack_confirm(skb) == NF_DROP ? - 1 : 0;
2144 }
2145
2146 static int nf_conntrack_update(struct net *net, struct sk_buff *skb)
2147 {
2148         enum ip_conntrack_info ctinfo;
2149         struct nf_conn *ct;
2150         int err;
2151
2152         ct = nf_ct_get(skb, &ctinfo);
2153         if (!ct)
2154                 return 0;
2155
2156         if (!nf_ct_is_confirmed(ct)) {
2157                 err = __nf_conntrack_update(net, skb, ct, ctinfo);
2158                 if (err < 0)
2159                         return err;
2160
2161                 ct = nf_ct_get(skb, &ctinfo);
2162         }
2163
2164         return nf_confirm_cthelper(skb, ct, ctinfo);
2165 }
2166
2167 static bool nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
2168                                        const struct sk_buff *skb)
2169 {
2170         const struct nf_conntrack_tuple *src_tuple;
2171         const struct nf_conntrack_tuple_hash *hash;
2172         struct nf_conntrack_tuple srctuple;
2173         enum ip_conntrack_info ctinfo;
2174         struct nf_conn *ct;
2175
2176         ct = nf_ct_get(skb, &ctinfo);
2177         if (ct) {
2178                 src_tuple = nf_ct_tuple(ct, CTINFO2DIR(ctinfo));
2179                 memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
2180                 return true;
2181         }
2182
2183         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
2184                                NFPROTO_IPV4, dev_net(skb->dev),
2185                                &srctuple))
2186                 return false;
2187
2188         hash = nf_conntrack_find_get(dev_net(skb->dev),
2189                                      &nf_ct_zone_dflt,
2190                                      &srctuple);
2191         if (!hash)
2192                 return false;
2193
2194         ct = nf_ct_tuplehash_to_ctrack(hash);
2195         src_tuple = nf_ct_tuple(ct, !hash->tuple.dst.dir);
2196         memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
2197         nf_ct_put(ct);
2198
2199         return true;
2200 }
2201
2202 /* Bring out ya dead! */
2203 static struct nf_conn *
2204 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
2205                 void *data, unsigned int *bucket)
2206 {
2207         struct nf_conntrack_tuple_hash *h;
2208         struct nf_conn *ct;
2209         struct hlist_nulls_node *n;
2210         spinlock_t *lockp;
2211
2212         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
2213                 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
2214                 local_bh_disable();
2215                 nf_conntrack_lock(lockp);
2216                 if (*bucket < nf_conntrack_htable_size) {
2217                         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
2218                                 if (NF_CT_DIRECTION(h) != IP_CT_DIR_REPLY)
2219                                         continue;
2220                                 /* All nf_conn objects are added to hash table twice, one
2221                                  * for original direction tuple, once for the reply tuple.
2222                                  *
2223                                  * Exception: In the IPS_NAT_CLASH case, only the reply
2224                                  * tuple is added (the original tuple already existed for
2225                                  * a different object).
2226                                  *
2227                                  * We only need to call the iterator once for each
2228                                  * conntrack, so we just use the 'reply' direction
2229                                  * tuple while iterating.
2230                                  */
2231                                 ct = nf_ct_tuplehash_to_ctrack(h);
2232                                 if (iter(ct, data))
2233                                         goto found;
2234                         }
2235                 }
2236                 spin_unlock(lockp);
2237                 local_bh_enable();
2238                 cond_resched();
2239         }
2240
2241         return NULL;
2242 found:
2243         atomic_inc(&ct->ct_general.use);
2244         spin_unlock(lockp);
2245         local_bh_enable();
2246         return ct;
2247 }
2248
2249 static void nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data),
2250                                   void *data, u32 portid, int report)
2251 {
2252         unsigned int bucket = 0, sequence;
2253         struct nf_conn *ct;
2254
2255         might_sleep();
2256
2257         for (;;) {
2258                 sequence = read_seqcount_begin(&nf_conntrack_generation);
2259
2260                 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
2261                         /* Time to push up daises... */
2262
2263                         nf_ct_delete(ct, portid, report);
2264                         nf_ct_put(ct);
2265                         cond_resched();
2266                 }
2267
2268                 if (!read_seqcount_retry(&nf_conntrack_generation, sequence))
2269                         break;
2270                 bucket = 0;
2271         }
2272 }
2273
2274 struct iter_data {
2275         int (*iter)(struct nf_conn *i, void *data);
2276         void *data;
2277         struct net *net;
2278 };
2279
2280 static int iter_net_only(struct nf_conn *i, void *data)
2281 {
2282         struct iter_data *d = data;
2283
2284         if (!net_eq(d->net, nf_ct_net(i)))
2285                 return 0;
2286
2287         return d->iter(i, d->data);
2288 }
2289
2290 static void
2291 __nf_ct_unconfirmed_destroy(struct net *net)
2292 {
2293         int cpu;
2294
2295         for_each_possible_cpu(cpu) {
2296                 struct nf_conntrack_tuple_hash *h;
2297                 struct hlist_nulls_node *n;
2298                 struct ct_pcpu *pcpu;
2299
2300                 pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2301
2302                 spin_lock_bh(&pcpu->lock);
2303                 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
2304                         struct nf_conn *ct;
2305
2306                         ct = nf_ct_tuplehash_to_ctrack(h);
2307
2308                         /* we cannot call iter() on unconfirmed list, the
2309                          * owning cpu can reallocate ct->ext at any time.
2310                          */
2311                         set_bit(IPS_DYING_BIT, &ct->status);
2312                 }
2313                 spin_unlock_bh(&pcpu->lock);
2314                 cond_resched();
2315         }
2316 }
2317
2318 void nf_ct_unconfirmed_destroy(struct net *net)
2319 {
2320         struct nf_conntrack_net *cnet = net_generic(net, nf_conntrack_net_id);
2321
2322         might_sleep();
2323
2324         if (atomic_read(&cnet->count) > 0) {
2325                 __nf_ct_unconfirmed_destroy(net);
2326                 nf_queue_nf_hook_drop(net);
2327                 synchronize_net();
2328         }
2329 }
2330 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy);
2331
2332 void nf_ct_iterate_cleanup_net(struct net *net,
2333                                int (*iter)(struct nf_conn *i, void *data),
2334                                void *data, u32 portid, int report)
2335 {
2336         struct nf_conntrack_net *cnet = net_generic(net, nf_conntrack_net_id);
2337         struct iter_data d;
2338
2339         might_sleep();
2340
2341         if (atomic_read(&cnet->count) == 0)
2342                 return;
2343
2344         d.iter = iter;
2345         d.data = data;
2346         d.net = net;
2347
2348         nf_ct_iterate_cleanup(iter_net_only, &d, portid, report);
2349 }
2350 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net);
2351
2352 /**
2353  * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
2354  * @iter: callback to invoke for each conntrack
2355  * @data: data to pass to @iter
2356  *
2357  * Like nf_ct_iterate_cleanup, but first marks conntracks on the
2358  * unconfirmed list as dying (so they will not be inserted into
2359  * main table).
2360  *
2361  * Can only be called in module exit path.
2362  */
2363 void
2364 nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data)
2365 {
2366         struct net *net;
2367
2368         down_read(&net_rwsem);
2369         for_each_net(net) {
2370                 struct nf_conntrack_net *cnet = net_generic(net, nf_conntrack_net_id);
2371
2372                 if (atomic_read(&cnet->count) == 0)
2373                         continue;
2374                 __nf_ct_unconfirmed_destroy(net);
2375                 nf_queue_nf_hook_drop(net);
2376         }
2377         up_read(&net_rwsem);
2378
2379         /* Need to wait for netns cleanup worker to finish, if its
2380          * running -- it might have deleted a net namespace from
2381          * the global list, so our __nf_ct_unconfirmed_destroy() might
2382          * not have affected all namespaces.
2383          */
2384         net_ns_barrier();
2385
2386         /* a conntrack could have been unlinked from unconfirmed list
2387          * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
2388          * This makes sure its inserted into conntrack table.
2389          */
2390         synchronize_net();
2391
2392         nf_ct_iterate_cleanup(iter, data, 0, 0);
2393 }
2394 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy);
2395
2396 static int kill_all(struct nf_conn *i, void *data)
2397 {
2398         return net_eq(nf_ct_net(i), data);
2399 }
2400
2401 void nf_conntrack_cleanup_start(void)
2402 {
2403         conntrack_gc_work.exiting = true;
2404         RCU_INIT_POINTER(ip_ct_attach, NULL);
2405 }
2406
2407 void nf_conntrack_cleanup_end(void)
2408 {
2409         RCU_INIT_POINTER(nf_ct_hook, NULL);
2410         cancel_delayed_work_sync(&conntrack_gc_work.dwork);
2411         kvfree(nf_conntrack_hash);
2412
2413         nf_conntrack_proto_fini();
2414         nf_conntrack_seqadj_fini();
2415         nf_conntrack_labels_fini();
2416         nf_conntrack_helper_fini();
2417         nf_conntrack_timeout_fini();
2418         nf_conntrack_ecache_fini();
2419         nf_conntrack_tstamp_fini();
2420         nf_conntrack_acct_fini();
2421         nf_conntrack_expect_fini();
2422
2423         kmem_cache_destroy(nf_conntrack_cachep);
2424 }
2425
2426 /*
2427  * Mishearing the voices in his head, our hero wonders how he's
2428  * supposed to kill the mall.
2429  */
2430 void nf_conntrack_cleanup_net(struct net *net)
2431 {
2432         LIST_HEAD(single);
2433
2434         list_add(&net->exit_list, &single);
2435         nf_conntrack_cleanup_net_list(&single);
2436 }
2437
2438 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
2439 {
2440         int busy;
2441         struct net *net;
2442
2443         /*
2444          * This makes sure all current packets have passed through
2445          *  netfilter framework.  Roll on, two-stage module
2446          *  delete...
2447          */
2448         synchronize_net();
2449 i_see_dead_people:
2450         busy = 0;
2451         list_for_each_entry(net, net_exit_list, exit_list) {
2452                 struct nf_conntrack_net *cnet = net_generic(net, nf_conntrack_net_id);
2453
2454                 nf_ct_iterate_cleanup(kill_all, net, 0, 0);
2455                 if (atomic_read(&cnet->count) != 0)
2456                         busy = 1;
2457         }
2458         if (busy) {
2459                 schedule();
2460                 goto i_see_dead_people;
2461         }
2462
2463         list_for_each_entry(net, net_exit_list, exit_list) {
2464                 nf_conntrack_proto_pernet_fini(net);
2465                 nf_conntrack_ecache_pernet_fini(net);
2466                 nf_conntrack_expect_pernet_fini(net);
2467                 free_percpu(net->ct.stat);
2468                 free_percpu(net->ct.pcpu_lists);
2469         }
2470 }
2471
2472 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
2473 {
2474         struct hlist_nulls_head *hash;
2475         unsigned int nr_slots, i;
2476
2477         if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
2478                 return NULL;
2479
2480         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
2481         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
2482
2483         hash = kvcalloc(nr_slots, sizeof(struct hlist_nulls_head), GFP_KERNEL);
2484
2485         if (hash && nulls)
2486                 for (i = 0; i < nr_slots; i++)
2487                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
2488
2489         return hash;
2490 }
2491 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
2492
2493 int nf_conntrack_hash_resize(unsigned int hashsize)
2494 {
2495         int i, bucket;
2496         unsigned int old_size;
2497         struct hlist_nulls_head *hash, *old_hash;
2498         struct nf_conntrack_tuple_hash *h;
2499         struct nf_conn *ct;
2500
2501         if (!hashsize)
2502                 return -EINVAL;
2503
2504         hash = nf_ct_alloc_hashtable(&hashsize, 1);
2505         if (!hash)
2506                 return -ENOMEM;
2507
2508         old_size = nf_conntrack_htable_size;
2509         if (old_size == hashsize) {
2510                 kvfree(hash);
2511                 return 0;
2512         }
2513
2514         local_bh_disable();
2515         nf_conntrack_all_lock();
2516         write_seqcount_begin(&nf_conntrack_generation);
2517
2518         /* Lookups in the old hash might happen in parallel, which means we
2519          * might get false negatives during connection lookup. New connections
2520          * created because of a false negative won't make it into the hash
2521          * though since that required taking the locks.
2522          */
2523
2524         for (i = 0; i < nf_conntrack_htable_size; i++) {
2525                 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
2526                         h = hlist_nulls_entry(nf_conntrack_hash[i].first,
2527                                               struct nf_conntrack_tuple_hash, hnnode);
2528                         ct = nf_ct_tuplehash_to_ctrack(h);
2529                         hlist_nulls_del_rcu(&h->hnnode);
2530                         bucket = __hash_conntrack(nf_ct_net(ct),
2531                                                   &h->tuple, hashsize);
2532                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
2533                 }
2534         }
2535         old_size = nf_conntrack_htable_size;
2536         old_hash = nf_conntrack_hash;
2537
2538         nf_conntrack_hash = hash;
2539         nf_conntrack_htable_size = hashsize;
2540
2541         write_seqcount_end(&nf_conntrack_generation);
2542         nf_conntrack_all_unlock();
2543         local_bh_enable();
2544
2545         synchronize_net();
2546         kvfree(old_hash);
2547         return 0;
2548 }
2549
2550 int nf_conntrack_set_hashsize(const char *val, const struct kernel_param *kp)
2551 {
2552         unsigned int hashsize;
2553         int rc;
2554
2555         if (current->nsproxy->net_ns != &init_net)
2556                 return -EOPNOTSUPP;
2557
2558         /* On boot, we can set this without any fancy locking. */
2559         if (!nf_conntrack_hash)
2560                 return param_set_uint(val, kp);
2561
2562         rc = kstrtouint(val, 0, &hashsize);
2563         if (rc)
2564                 return rc;
2565
2566         return nf_conntrack_hash_resize(hashsize);
2567 }
2568
2569 static __always_inline unsigned int total_extension_size(void)
2570 {
2571         /* remember to add new extensions below */
2572         BUILD_BUG_ON(NF_CT_EXT_NUM > 9);
2573
2574         return sizeof(struct nf_ct_ext) +
2575                sizeof(struct nf_conn_help)
2576 #if IS_ENABLED(CONFIG_NF_NAT)
2577                 + sizeof(struct nf_conn_nat)
2578 #endif
2579                 + sizeof(struct nf_conn_seqadj)
2580                 + sizeof(struct nf_conn_acct)
2581 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2582                 + sizeof(struct nf_conntrack_ecache)
2583 #endif
2584 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2585                 + sizeof(struct nf_conn_tstamp)
2586 #endif
2587 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2588                 + sizeof(struct nf_conn_timeout)
2589 #endif
2590 #ifdef CONFIG_NF_CONNTRACK_LABELS
2591                 + sizeof(struct nf_conn_labels)
2592 #endif
2593 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2594                 + sizeof(struct nf_conn_synproxy)
2595 #endif
2596         ;
2597 };
2598
2599 int nf_conntrack_init_start(void)
2600 {
2601         unsigned long nr_pages = totalram_pages();
2602         int max_factor = 8;
2603         int ret = -ENOMEM;
2604         int i;
2605
2606         /* struct nf_ct_ext uses u8 to store offsets/size */
2607         BUILD_BUG_ON(total_extension_size() > 255u);
2608
2609         seqcount_spinlock_init(&nf_conntrack_generation,
2610                                &nf_conntrack_locks_all_lock);
2611
2612         for (i = 0; i < CONNTRACK_LOCKS; i++)
2613                 spin_lock_init(&nf_conntrack_locks[i]);
2614
2615         if (!nf_conntrack_htable_size) {
2616                 /* Idea from tcp.c: use 1/16384 of memory.
2617                  * On i386: 32MB machine has 512 buckets.
2618                  * >= 1GB machines have 16384 buckets.
2619                  * >= 4GB machines have 65536 buckets.
2620                  */
2621                 nf_conntrack_htable_size
2622                         = (((nr_pages << PAGE_SHIFT) / 16384)
2623                            / sizeof(struct hlist_head));
2624                 if (nr_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
2625                         nf_conntrack_htable_size = 65536;
2626                 else if (nr_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
2627                         nf_conntrack_htable_size = 16384;
2628                 if (nf_conntrack_htable_size < 32)
2629                         nf_conntrack_htable_size = 32;
2630
2631                 /* Use a max. factor of four by default to get the same max as
2632                  * with the old struct list_heads. When a table size is given
2633                  * we use the old value of 8 to avoid reducing the max.
2634                  * entries. */
2635                 max_factor = 4;
2636         }
2637
2638         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
2639         if (!nf_conntrack_hash)
2640                 return -ENOMEM;
2641
2642         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
2643
2644         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
2645                                                 sizeof(struct nf_conn),
2646                                                 NFCT_INFOMASK + 1,
2647                                                 SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
2648         if (!nf_conntrack_cachep)
2649                 goto err_cachep;
2650
2651         ret = nf_conntrack_expect_init();
2652         if (ret < 0)
2653                 goto err_expect;
2654
2655         ret = nf_conntrack_acct_init();
2656         if (ret < 0)
2657                 goto err_acct;
2658
2659         ret = nf_conntrack_tstamp_init();
2660         if (ret < 0)
2661                 goto err_tstamp;
2662
2663         ret = nf_conntrack_ecache_init();
2664         if (ret < 0)
2665                 goto err_ecache;
2666
2667         ret = nf_conntrack_timeout_init();
2668         if (ret < 0)
2669                 goto err_timeout;
2670
2671         ret = nf_conntrack_helper_init();
2672         if (ret < 0)
2673                 goto err_helper;
2674
2675         ret = nf_conntrack_labels_init();
2676         if (ret < 0)
2677                 goto err_labels;
2678
2679         ret = nf_conntrack_seqadj_init();
2680         if (ret < 0)
2681                 goto err_seqadj;
2682
2683         ret = nf_conntrack_proto_init();
2684         if (ret < 0)
2685                 goto err_proto;
2686
2687         conntrack_gc_work_init(&conntrack_gc_work);
2688         queue_delayed_work(system_power_efficient_wq, &conntrack_gc_work.dwork, HZ);
2689
2690         return 0;
2691
2692 err_proto:
2693         nf_conntrack_seqadj_fini();
2694 err_seqadj:
2695         nf_conntrack_labels_fini();
2696 err_labels:
2697         nf_conntrack_helper_fini();
2698 err_helper:
2699         nf_conntrack_timeout_fini();
2700 err_timeout:
2701         nf_conntrack_ecache_fini();
2702 err_ecache:
2703         nf_conntrack_tstamp_fini();
2704 err_tstamp:
2705         nf_conntrack_acct_fini();
2706 err_acct:
2707         nf_conntrack_expect_fini();
2708 err_expect:
2709         kmem_cache_destroy(nf_conntrack_cachep);
2710 err_cachep:
2711         kvfree(nf_conntrack_hash);
2712         return ret;
2713 }
2714
2715 static struct nf_ct_hook nf_conntrack_hook = {
2716         .update         = nf_conntrack_update,
2717         .destroy        = destroy_conntrack,
2718         .get_tuple_skb  = nf_conntrack_get_tuple_skb,
2719 };
2720
2721 void nf_conntrack_init_end(void)
2722 {
2723         /* For use by REJECT target */
2724         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
2725         RCU_INIT_POINTER(nf_ct_hook, &nf_conntrack_hook);
2726 }
2727
2728 /*
2729  * We need to use special "null" values, not used in hash table
2730  */
2731 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
2732 #define DYING_NULLS_VAL         ((1<<30)+1)
2733
2734 int nf_conntrack_init_net(struct net *net)
2735 {
2736         struct nf_conntrack_net *cnet = net_generic(net, nf_conntrack_net_id);
2737         int ret = -ENOMEM;
2738         int cpu;
2739
2740         BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER);
2741         BUILD_BUG_ON_NOT_POWER_OF_2(CONNTRACK_LOCKS);
2742         atomic_set(&cnet->count, 0);
2743
2744         net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
2745         if (!net->ct.pcpu_lists)
2746                 goto err_stat;
2747
2748         for_each_possible_cpu(cpu) {
2749                 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2750
2751                 spin_lock_init(&pcpu->lock);
2752                 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
2753                 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
2754         }
2755
2756         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
2757         if (!net->ct.stat)
2758                 goto err_pcpu_lists;
2759
2760         ret = nf_conntrack_expect_pernet_init(net);
2761         if (ret < 0)
2762                 goto err_expect;
2763
2764         nf_conntrack_acct_pernet_init(net);
2765         nf_conntrack_tstamp_pernet_init(net);
2766         nf_conntrack_ecache_pernet_init(net);
2767         nf_conntrack_helper_pernet_init(net);
2768         nf_conntrack_proto_pernet_init(net);
2769
2770         return 0;
2771
2772 err_expect:
2773         free_percpu(net->ct.stat);
2774 err_pcpu_lists:
2775         free_percpu(net->ct.pcpu_lists);
2776 err_stat:
2777         return ret;
2778 }