net/core/flow_dissector.c

   1 #include <linux/kernel.h>
   2 #include <linux/skbuff.h>
   3 #include <linux/export.h>
   4 #include <linux/ip.h>
   5 #include <linux/ipv6.h>
   6 #include <linux/if_vlan.h>
   7 #include <net/ip.h>
   8 #include <net/ipv6.h>
   9 #include <linux/igmp.h>
  10 #include <linux/icmp.h>
  11 #include <linux/sctp.h>
  12 #include <linux/dccp.h>
  13 #include <linux/if_tunnel.h>
  14 #include <linux/if_pppox.h>
  15 #include <linux/ppp_defs.h>
  16 #include <linux/stddef.h>
  17 #include <linux/if_ether.h>
  18 #include <linux/mpls.h>
  19 #include <net/flow_dissector.h>
  20 #include <scsi/fc/fc_fcoe.h>
  21
  22 static bool dissector_uses_key(const struct flow_dissector *flow_dissector,
  23                                enum flow_dissector_key_id key_id)
  24 {
  25         return flow_dissector->used_keys & (1 << key_id);
  26 }
  27
  28 static void dissector_set_key(struct flow_dissector *flow_dissector,
  29                               enum flow_dissector_key_id key_id)
  30 {
  31         flow_dissector->used_keys |= (1 << key_id);
  32 }
  33
  34 static void *skb_flow_dissector_target(struct flow_dissector *flow_dissector,
  35                                        enum flow_dissector_key_id key_id,
  36                                        void *target_container)
  37 {
  38         return ((char *) target_container) + flow_dissector->offset[key_id];
  39 }
  40
  41 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
  42                              const struct flow_dissector_key *key,
  43                              unsigned int key_count)
  44 {
  45         unsigned int i;
  46
  47         memset(flow_dissector, 0, sizeof(*flow_dissector));
  48
  49         for (i = 0; i < key_count; i++, key++) {
  50                 /* User should make sure that every key target offset is withing
  51                  * boundaries of unsigned short.
  52                  */
  53                 BUG_ON(key->offset > USHRT_MAX);
  54                 BUG_ON(dissector_uses_key(flow_dissector,
  55                                           key->key_id));
  56
  57                 dissector_set_key(flow_dissector, key->key_id);
  58                 flow_dissector->offset[key->key_id] = key->offset;
  59         }
  60
  61         /* Ensure that the dissector always includes control and basic key.
  62          * That way we are able to avoid handling lack of these in fast path.
  63          */
  64         BUG_ON(!dissector_uses_key(flow_dissector,
  65                                    FLOW_DISSECTOR_KEY_CONTROL));
  66         BUG_ON(!dissector_uses_key(flow_dissector,
  67                                    FLOW_DISSECTOR_KEY_BASIC));
  68 }
  69 EXPORT_SYMBOL(skb_flow_dissector_init);
  70
  71 /**
  72  * __skb_flow_get_ports - extract the upper layer ports and return them
  73  * @skb: sk_buff to extract the ports from
  74  * @thoff: transport header offset
  75  * @ip_proto: protocol for which to get port offset
  76  * @data: raw buffer pointer to the packet, if NULL use skb->data
  77  * @hlen: packet header length, if @data is NULL use skb_headlen(skb)
  78  *
  79  * The function will try to retrieve the ports at offset thoff + poff where poff
  80  * is the protocol port offset returned from proto_ports_offset
  81  */
  82 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
  83                             void *data, int hlen)
  84 {
  85         int poff = proto_ports_offset(ip_proto);
  86
  87         if (!data) {
  88                 data = skb->data;
  89                 hlen = skb_headlen(skb);
  90         }
  91
  92         if (poff >= 0) {
  93                 __be32 *ports, _ports;
  94
  95                 ports = __skb_header_pointer(skb, thoff + poff,
  96                                              sizeof(_ports), data, hlen, &_ports);
  97                 if (ports)
  98                         return *ports;
  99         }
 100
 101         return 0;
 102 }
 103 EXPORT_SYMBOL(__skb_flow_get_ports);
 104
 105 /**
 106  * __skb_flow_dissect - extract the flow_keys struct and return it
 107  * @skb: sk_buff to extract the flow from, can be NULL if the rest are specified
 108  * @flow_dissector: list of keys to dissect
 109  * @target_container: target structure to put dissected values into
 110  * @data: raw buffer pointer to the packet, if NULL use skb->data
 111  * @proto: protocol for which to get the flow, if @data is NULL use skb->protocol
 112  * @nhoff: network header offset, if @data is NULL use skb_network_offset(skb)
 113  * @hlen: packet header length, if @data is NULL use skb_headlen(skb)
 114  *
 115  * The function will try to retrieve individual keys into target specified
 116  * by flow_dissector from either the skbuff or a raw buffer specified by the
 117  * rest parameters.
 118  *
 119  * Caller must take care of zeroing target container memory.
 120  */
 121 bool __skb_flow_dissect(const struct sk_buff *skb,
 122                         struct flow_dissector *flow_dissector,
 123                         void *target_container,
 124                         void *data, __be16 proto, int nhoff, int hlen,
 125                         unsigned int flags)
 126 {
 127         struct flow_dissector_key_control *key_control;
 128         struct flow_dissector_key_basic *key_basic;
 129         struct flow_dissector_key_addrs *key_addrs;
 130         struct flow_dissector_key_ports *key_ports;
 131         struct flow_dissector_key_tags *key_tags;
 132         struct flow_dissector_key_keyid *key_keyid;
 133         u8 ip_proto = 0;
 134         bool ret;
 135
 136         if (!data) {
 137                 data = skb->data;
 138                 proto = skb->protocol;
 139                 nhoff = skb_network_offset(skb);
 140                 hlen = skb_headlen(skb);
 141         }
 142
 143         /* It is ensured by skb_flow_dissector_init() that control key will
 144          * be always present.
 145          */
 146         key_control = skb_flow_dissector_target(flow_dissector,
 147                                                 FLOW_DISSECTOR_KEY_CONTROL,
 148                                                 target_container);
 149
 150         /* It is ensured by skb_flow_dissector_init() that basic key will
 151          * be always present.
 152          */
 153         key_basic = skb_flow_dissector_target(flow_dissector,
 154                                               FLOW_DISSECTOR_KEY_BASIC,
 155                                               target_container);
 156
 157         if (dissector_uses_key(flow_dissector,
 158                                FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
 159                 struct ethhdr *eth = eth_hdr(skb);
 160                 struct flow_dissector_key_eth_addrs *key_eth_addrs;
 161
 162                 key_eth_addrs = skb_flow_dissector_target(flow_dissector,
 163                                                           FLOW_DISSECTOR_KEY_ETH_ADDRS,
 164                                                           target_container);
 165                 memcpy(key_eth_addrs, &eth->h_dest, sizeof(*key_eth_addrs));
 166         }
 167
 168 again:
 169         switch (proto) {
 170         case htons(ETH_P_IP): {
 171                 const struct iphdr *iph;
 172                 struct iphdr _iph;
 173 ip:
 174                 iph = __skb_header_pointer(skb, nhoff, sizeof(_iph), data, hlen, &_iph);
 175                 if (!iph || iph->ihl < 5)
 176                         goto out_bad;
 177                 nhoff += iph->ihl * 4;
 178
 179                 ip_proto = iph->protocol;
 180
 181                 if (!dissector_uses_key(flow_dissector,
 182                                         FLOW_DISSECTOR_KEY_IPV4_ADDRS))
 183                         break;
 184
 185                 key_addrs = skb_flow_dissector_target(flow_dissector,
 186                               FLOW_DISSECTOR_KEY_IPV4_ADDRS, target_container);
 187                 memcpy(&key_addrs->v4addrs, &iph->saddr,
 188                        sizeof(key_addrs->v4addrs));
 189                 key_control->addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
 190
 191                 if (ip_is_fragment(iph)) {
 192                         key_control->flags |= FLOW_DIS_IS_FRAGMENT;
 193
 194                         if (iph->frag_off & htons(IP_OFFSET)) {
 195                                 goto out_good;
 196                         } else {
 197                                 key_control->flags |= FLOW_DIS_FIRST_FRAG;
 198                                 if (!(flags & FLOW_DISSECTOR_F_PARSE_1ST_FRAG))
 199                                         goto out_good;
 200                         }
 201                 }
 202
 203                 if (flags & FLOW_DISSECTOR_F_STOP_AT_L3)
 204                         goto out_good;
 205
 206                 break;
 207         }
 208         case htons(ETH_P_IPV6): {
 209                 const struct ipv6hdr *iph;
 210                 struct ipv6hdr _iph;
 211
 212 ipv6:
 213                 iph = __skb_header_pointer(skb, nhoff, sizeof(_iph), data, hlen, &_iph);
 214                 if (!iph)
 215                         goto out_bad;
 216
 217                 ip_proto = iph->nexthdr;
 218                 nhoff += sizeof(struct ipv6hdr);
 219
 220                 if (dissector_uses_key(flow_dissector,
 221                                        FLOW_DISSECTOR_KEY_IPV6_ADDRS)) {
 222                         struct flow_dissector_key_ipv6_addrs *key_ipv6_addrs;
 223
 224                         key_ipv6_addrs = skb_flow_dissector_target(flow_dissector,
 225                                                                    FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 226                                                                    target_container);
 227
 228                         memcpy(key_ipv6_addrs, &iph->saddr, sizeof(*key_ipv6_addrs));
 229                         key_control->addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
 230                 }
 231
 232                 if ((dissector_uses_key(flow_dissector,
 233                                         FLOW_DISSECTOR_KEY_FLOW_LABEL) ||
 234                      (flags & FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL)) &&
 235                     ip6_flowlabel(iph)) {
 236                         __be32 flow_label = ip6_flowlabel(iph);
 237
 238                         if (dissector_uses_key(flow_dissector,
 239                                                FLOW_DISSECTOR_KEY_FLOW_LABEL)) {
 240                                 key_tags = skb_flow_dissector_target(flow_dissector,
 241                                                                      FLOW_DISSECTOR_KEY_FLOW_LABEL,
 242                                                                      target_container);
 243                                 key_tags->flow_label = ntohl(flow_label);
 244                         }
 245                         if (flags & FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL)
 246                                 goto out_good;
 247                 }
 248
 249                 if (flags & FLOW_DISSECTOR_F_STOP_AT_L3)
 250                         goto out_good;
 251
 252                 break;
 253         }
 254         case htons(ETH_P_8021AD):
 255         case htons(ETH_P_8021Q): {
 256                 const struct vlan_hdr *vlan;
 257                 struct vlan_hdr _vlan;
 258
 259                 vlan = __skb_header_pointer(skb, nhoff, sizeof(_vlan), data, hlen, &_vlan);
 260                 if (!vlan)
 261                         goto out_bad;
 262
 263                 if (dissector_uses_key(flow_dissector,
 264                                        FLOW_DISSECTOR_KEY_VLANID)) {
 265                         key_tags = skb_flow_dissector_target(flow_dissector,
 266                                                              FLOW_DISSECTOR_KEY_VLANID,
 267                                                              target_container);
 268
 269                         key_tags->vlan_id = skb_vlan_tag_get_id(skb);
 270                 }
 271
 272                 proto = vlan->h_vlan_encapsulated_proto;
 273                 nhoff += sizeof(*vlan);
 274                 goto again;
 275         }
 276         case htons(ETH_P_PPP_SES): {
 277                 struct {
 278                         struct pppoe_hdr hdr;
 279                         __be16 proto;
 280                 } *hdr, _hdr;
 281                 hdr = __skb_header_pointer(skb, nhoff, sizeof(_hdr), data, hlen, &_hdr);
 282                 if (!hdr)
 283                         goto out_bad;
 284                 proto = hdr->proto;
 285                 nhoff += PPPOE_SES_HLEN;
 286                 switch (proto) {
 287                 case htons(PPP_IP):
 288                         goto ip;
 289                 case htons(PPP_IPV6):
 290                         goto ipv6;
 291                 default:
 292                         goto out_bad;
 293                 }
 294         }
 295         case htons(ETH_P_TIPC): {
 296                 struct {
 297                         __be32 pre[3];
 298                         __be32 srcnode;
 299                 } *hdr, _hdr;
 300                 hdr = __skb_header_pointer(skb, nhoff, sizeof(_hdr), data, hlen, &_hdr);
 301                 if (!hdr)
 302                         goto out_bad;
 303
 304                 if (dissector_uses_key(flow_dissector,
 305                                        FLOW_DISSECTOR_KEY_TIPC_ADDRS)) {
 306                         key_addrs = skb_flow_dissector_target(flow_dissector,
 307                                                               FLOW_DISSECTOR_KEY_TIPC_ADDRS,
 308                                                               target_container);
 309                         key_addrs->tipcaddrs.srcnode = hdr->srcnode;
 310                         key_control->addr_type = FLOW_DISSECTOR_KEY_TIPC_ADDRS;
 311                 }
 312                 goto out_good;
 313         }
 314
 315         case htons(ETH_P_MPLS_UC):
 316         case htons(ETH_P_MPLS_MC): {
 317                 struct mpls_label *hdr, _hdr[2];
 318 mpls:
 319                 hdr = __skb_header_pointer(skb, nhoff, sizeof(_hdr), data,
 320                                            hlen, &_hdr);
 321                 if (!hdr)
 322                         goto out_bad;
 323
 324                 if ((ntohl(hdr[0].entry) & MPLS_LS_LABEL_MASK) >>
 325                      MPLS_LS_LABEL_SHIFT == MPLS_LABEL_ENTROPY) {
 326                         if (dissector_uses_key(flow_dissector,
 327                                                FLOW_DISSECTOR_KEY_MPLS_ENTROPY)) {
 328                                 key_keyid = skb_flow_dissector_target(flow_dissector,
 329                                                                       FLOW_DISSECTOR_KEY_MPLS_ENTROPY,
 330                                                                       target_container);
 331                                 key_keyid->keyid = hdr[1].entry &
 332                                         htonl(MPLS_LS_LABEL_MASK);
 333                         }
 334
 335                         goto out_good;
 336                 }
 337
 338                 goto out_good;
 339         }
 340
 341         case htons(ETH_P_FCOE):
 342                 key_control->thoff = (u16)(nhoff + FCOE_HEADER_LEN);
 343                 /* fall through */
 344         default:
 345                 goto out_bad;
 346         }
 347
 348 ip_proto_again:
 349         switch (ip_proto) {
 350         case IPPROTO_GRE: {
 351                 struct gre_hdr {
 352                         __be16 flags;
 353                         __be16 proto;
 354                 } *hdr, _hdr;
 355
 356                 hdr = __skb_header_pointer(skb, nhoff, sizeof(_hdr), data, hlen, &_hdr);
 357                 if (!hdr)
 358                         goto out_bad;
 359                 /*
 360                  * Only look inside GRE if version zero and no
 361                  * routing
 362                  */
 363                 if (hdr->flags & (GRE_VERSION | GRE_ROUTING))
 364                         break;
 365
 366                 proto = hdr->proto;
 367                 nhoff += 4;
 368                 if (hdr->flags & GRE_CSUM)
 369                         nhoff += 4;
 370                 if (hdr->flags & GRE_KEY) {
 371                         const __be32 *keyid;
 372                         __be32 _keyid;
 373
 374                         keyid = __skb_header_pointer(skb, nhoff, sizeof(_keyid),
 375                                                      data, hlen, &_keyid);
 376
 377                         if (!keyid)
 378                                 goto out_bad;
 379
 380                         if (dissector_uses_key(flow_dissector,
 381                                                FLOW_DISSECTOR_KEY_GRE_KEYID)) {
 382                                 key_keyid = skb_flow_dissector_target(flow_dissector,
 383                                                                       FLOW_DISSECTOR_KEY_GRE_KEYID,
 384                                                                       target_container);
 385                                 key_keyid->keyid = *keyid;
 386                         }
 387                         nhoff += 4;
 388                 }
 389                 if (hdr->flags & GRE_SEQ)
 390                         nhoff += 4;
 391                 if (proto == htons(ETH_P_TEB)) {
 392                         const struct ethhdr *eth;
 393                         struct ethhdr _eth;
 394
 395                         eth = __skb_header_pointer(skb, nhoff,
 396                                                    sizeof(_eth),
 397                                                    data, hlen, &_eth);
 398                         if (!eth)
 399                                 goto out_bad;
 400                         proto = eth->h_proto;
 401                         nhoff += sizeof(*eth);
 402
 403                         /* Cap headers that we access via pointers at the
 404                          * end of the Ethernet header as our maximum alignment
 405                          * at that point is only 2 bytes.
 406                          */
 407                         if (NET_IP_ALIGN)
 408                                 hlen = nhoff;
 409                 }
 410
 411                 key_control->flags |= FLOW_DIS_ENCAPSULATION;
 412                 if (flags & FLOW_DISSECTOR_F_STOP_AT_ENCAP)
 413                         goto out_good;
 414
 415                 goto again;
 416         }
 417         case NEXTHDR_HOP:
 418         case NEXTHDR_ROUTING:
 419         case NEXTHDR_DEST: {
 420                 u8 _opthdr[2], *opthdr;
 421
 422                 if (proto != htons(ETH_P_IPV6))
 423                         break;
 424
 425                 opthdr = __skb_header_pointer(skb, nhoff, sizeof(_opthdr),
 426                                               data, hlen, &_opthdr);
 427                 if (!opthdr)
 428                         goto out_bad;
 429
 430                 ip_proto = opthdr[0];
 431                 nhoff += (opthdr[1] + 1) << 3;
 432
 433                 goto ip_proto_again;
 434         }
 435         case NEXTHDR_FRAGMENT: {
 436                 struct frag_hdr _fh, *fh;
 437
 438                 if (proto != htons(ETH_P_IPV6))
 439                         break;
 440
 441                 fh = __skb_header_pointer(skb, nhoff, sizeof(_fh),
 442                                           data, hlen, &_fh);
 443
 444                 if (!fh)
 445                         goto out_bad;
 446
 447                 key_control->flags |= FLOW_DIS_IS_FRAGMENT;
 448
 449                 nhoff += sizeof(_fh);
 450
 451                 if (!(fh->frag_off & htons(IP6_OFFSET))) {
 452                         key_control->flags |= FLOW_DIS_FIRST_FRAG;
 453                         if (flags & FLOW_DISSECTOR_F_PARSE_1ST_FRAG) {
 454                                 ip_proto = fh->nexthdr;
 455                                 goto ip_proto_again;
 456                         }
 457                 }
 458                 goto out_good;
 459         }
 460         case IPPROTO_IPIP:
 461                 proto = htons(ETH_P_IP);
 462
 463                 key_control->flags |= FLOW_DIS_ENCAPSULATION;
 464                 if (flags & FLOW_DISSECTOR_F_STOP_AT_ENCAP)
 465                         goto out_good;
 466
 467                 goto ip;
 468         case IPPROTO_IPV6:
 469                 proto = htons(ETH_P_IPV6);
 470
 471                 key_control->flags |= FLOW_DIS_ENCAPSULATION;
 472                 if (flags & FLOW_DISSECTOR_F_STOP_AT_ENCAP)
 473                         goto out_good;
 474
 475                 goto ipv6;
 476         case IPPROTO_MPLS:
 477                 proto = htons(ETH_P_MPLS_UC);
 478                 goto mpls;
 479         default:
 480                 break;
 481         }
 482
 483         if (dissector_uses_key(flow_dissector,
 484                                FLOW_DISSECTOR_KEY_PORTS)) {
 485                 key_ports = skb_flow_dissector_target(flow_dissector,
 486                                                       FLOW_DISSECTOR_KEY_PORTS,
 487                                                       target_container);
 488                 key_ports->ports = __skb_flow_get_ports(skb, nhoff, ip_proto,
 489                                                         data, hlen);
 490         }
 491
 492 out_good:
 493         ret = true;
 494
 495         key_control->thoff = (u16)nhoff;
 496 out:
 497         key_basic->n_proto = proto;
 498         key_basic->ip_proto = ip_proto;
 499
 500         return ret;
 501
 502 out_bad:
 503         ret = false;
 504         key_control->thoff = min_t(u16, nhoff, skb ? skb->len : hlen);
 505         goto out;
 506 }
 507 EXPORT_SYMBOL(__skb_flow_dissect);
 508
 509 static u32 hashrnd __read_mostly;
 510 static __always_inline void __flow_hash_secret_init(void)
 511 {
 512         net_get_random_once(&hashrnd, sizeof(hashrnd));
 513 }
 514
 515 static __always_inline u32 __flow_hash_words(const u32 *words, u32 length,
 516                                              u32 keyval)
 517 {
 518         return jhash2(words, length, keyval);
 519 }
 520
 521 static inline const u32 *flow_keys_hash_start(const struct flow_keys *flow)
 522 {
 523         const void *p = flow;
 524
 525         BUILD_BUG_ON(FLOW_KEYS_HASH_OFFSET % sizeof(u32));
 526         return (const u32 *)(p + FLOW_KEYS_HASH_OFFSET);
 527 }
 528
 529 static inline size_t flow_keys_hash_length(const struct flow_keys *flow)
 530 {
 531         size_t diff = FLOW_KEYS_HASH_OFFSET + sizeof(flow->addrs);
 532         BUILD_BUG_ON((sizeof(*flow) - FLOW_KEYS_HASH_OFFSET) % sizeof(u32));
 533         BUILD_BUG_ON(offsetof(typeof(*flow), addrs) !=
 534                      sizeof(*flow) - sizeof(flow->addrs));
 535
 536         switch (flow->control.addr_type) {
 537         case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
 538                 diff -= sizeof(flow->addrs.v4addrs);
 539                 break;
 540         case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
 541                 diff -= sizeof(flow->addrs.v6addrs);
 542                 break;
 543         case FLOW_DISSECTOR_KEY_TIPC_ADDRS:
 544                 diff -= sizeof(flow->addrs.tipcaddrs);
 545                 break;
 546         }
 547         return (sizeof(*flow) - diff) / sizeof(u32);
 548 }
 549
 550 __be32 flow_get_u32_src(const struct flow_keys *flow)
 551 {
 552         switch (flow->control.addr_type) {
 553         case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
 554                 return flow->addrs.v4addrs.src;
 555         case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
 556                 return (__force __be32)ipv6_addr_hash(
 557                         &flow->addrs.v6addrs.src);
 558         case FLOW_DISSECTOR_KEY_TIPC_ADDRS:
 559                 return flow->addrs.tipcaddrs.srcnode;
 560         default:
 561                 return 0;
 562         }
 563 }
 564 EXPORT_SYMBOL(flow_get_u32_src);
 565
 566 __be32 flow_get_u32_dst(const struct flow_keys *flow)
 567 {
 568         switch (flow->control.addr_type) {
 569         case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
 570                 return flow->addrs.v4addrs.dst;
 571         case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
 572                 return (__force __be32)ipv6_addr_hash(
 573                         &flow->addrs.v6addrs.dst);
 574         default:
 575                 return 0;
 576         }
 577 }
 578 EXPORT_SYMBOL(flow_get_u32_dst);
 579
 580 static inline void __flow_hash_consistentify(struct flow_keys *keys)
 581 {
 582         int addr_diff, i;
 583
 584         switch (keys->control.addr_type) {
 585         case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
 586                 addr_diff = (__force u32)keys->addrs.v4addrs.dst -
 587                             (__force u32)keys->addrs.v4addrs.src;
 588                 if ((addr_diff < 0) ||
 589                     (addr_diff == 0 &&
 590                      ((__force u16)keys->ports.dst <
 591                       (__force u16)keys->ports.src))) {
 592                         swap(keys->addrs.v4addrs.src, keys->addrs.v4addrs.dst);
 593                         swap(keys->ports.src, keys->ports.dst);
 594                 }
 595                 break;
 596         case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
 597                 addr_diff = memcmp(&keys->addrs.v6addrs.dst,
 598                                    &keys->addrs.v6addrs.src,
 599                                    sizeof(keys->addrs.v6addrs.dst));
 600                 if ((addr_diff < 0) ||
 601                     (addr_diff == 0 &&
 602                      ((__force u16)keys->ports.dst <
 603                       (__force u16)keys->ports.src))) {
 604                         for (i = 0; i < 4; i++)
 605                                 swap(keys->addrs.v6addrs.src.s6_addr32[i],
 606                                      keys->addrs.v6addrs.dst.s6_addr32[i]);
 607                         swap(keys->ports.src, keys->ports.dst);
 608                 }
 609                 break;
 610         }
 611 }
 612
 613 static inline u32 __flow_hash_from_keys(struct flow_keys *keys, u32 keyval)
 614 {
 615         u32 hash;
 616
 617         __flow_hash_consistentify(keys);
 618
 619         hash = __flow_hash_words(flow_keys_hash_start(keys),
 620                                  flow_keys_hash_length(keys), keyval);
 621         if (!hash)
 622                 hash = 1;
 623
 624         return hash;
 625 }
 626
 627 u32 flow_hash_from_keys(struct flow_keys *keys)
 628 {
 629         __flow_hash_secret_init();
 630         return __flow_hash_from_keys(keys, hashrnd);
 631 }
 632 EXPORT_SYMBOL(flow_hash_from_keys);
 633
 634 static inline u32 ___skb_get_hash(const struct sk_buff *skb,
 635                                   struct flow_keys *keys, u32 keyval)
 636 {
 637         skb_flow_dissect_flow_keys(skb, keys,
 638                                    FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL);
 639
 640         return __flow_hash_from_keys(keys, keyval);
 641 }
 642
 643 struct _flow_keys_digest_data {
 644         __be16  n_proto;
 645         u8      ip_proto;
 646         u8      padding;
 647         __be32  ports;
 648         __be32  src;
 649         __be32  dst;
 650 };
 651
 652 void make_flow_keys_digest(struct flow_keys_digest *digest,
 653                            const struct flow_keys *flow)
 654 {
 655         struct _flow_keys_digest_data *data =
 656             (struct _flow_keys_digest_data *)digest;
 657
 658         BUILD_BUG_ON(sizeof(*data) > sizeof(*digest));
 659
 660         memset(digest, 0, sizeof(*digest));
 661
 662         data->n_proto = flow->basic.n_proto;
 663         data->ip_proto = flow->basic.ip_proto;
 664         data->ports = flow->ports.ports;
 665         data->src = flow->addrs.v4addrs.src;
 666         data->dst = flow->addrs.v4addrs.dst;
 667 }
 668 EXPORT_SYMBOL(make_flow_keys_digest);
 669
 670 static struct flow_dissector flow_keys_dissector_symmetric __read_mostly;
 671
 672 u32 __skb_get_hash_symmetric(struct sk_buff *skb)
 673 {
 674         struct flow_keys keys;
 675
 676         __flow_hash_secret_init();
 677
 678         memset(&keys, 0, sizeof(keys));
 679         __skb_flow_dissect(skb, &flow_keys_dissector_symmetric, &keys,
 680                            NULL, 0, 0, 0,
 681                            FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL);
 682
 683         return __flow_hash_from_keys(&keys, hashrnd);
 684 }
 685 EXPORT_SYMBOL_GPL(__skb_get_hash_symmetric);
 686
 687 /**
 688  * __skb_get_hash: calculate a flow hash
 689  * @skb: sk_buff to calculate flow hash from
 690  *
 691  * This function calculates a flow hash based on src/dst addresses
 692  * and src/dst port numbers.  Sets hash in skb to non-zero hash value
 693  * on success, zero indicates no valid hash.  Also, sets l4_hash in skb
 694  * if hash is a canonical 4-tuple hash over transport ports.
 695  */
 696 void __skb_get_hash(struct sk_buff *skb)
 697 {
 698         struct flow_keys keys;
 699
 700         __flow_hash_secret_init();
 701
 702         __skb_set_sw_hash(skb, ___skb_get_hash(skb, &keys, hashrnd),
 703                           flow_keys_have_l4(&keys));
 704 }
 705 EXPORT_SYMBOL(__skb_get_hash);
 706
 707 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb)
 708 {
 709         struct flow_keys keys;
 710
 711         return ___skb_get_hash(skb, &keys, perturb);
 712 }
 713 EXPORT_SYMBOL(skb_get_hash_perturb);
 714
 715 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
 716 {
 717         struct flow_keys keys;
 718
 719         memset(&keys, 0, sizeof(keys));
 720
 721         memcpy(&keys.addrs.v6addrs.src, &fl6->saddr,
 722                sizeof(keys.addrs.v6addrs.src));
 723         memcpy(&keys.addrs.v6addrs.dst, &fl6->daddr,
 724                sizeof(keys.addrs.v6addrs.dst));
 725         keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
 726         keys.ports.src = fl6->fl6_sport;
 727         keys.ports.dst = fl6->fl6_dport;
 728         keys.keyid.keyid = fl6->fl6_gre_key;
 729         keys.tags.flow_label = (__force u32)fl6->flowlabel;
 730         keys.basic.ip_proto = fl6->flowi6_proto;
 731
 732         __skb_set_sw_hash(skb, flow_hash_from_keys(&keys),
 733                           flow_keys_have_l4(&keys));
 734
 735         return skb->hash;
 736 }
 737 EXPORT_SYMBOL(__skb_get_hash_flowi6);
 738
 739 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
 740 {
 741         struct flow_keys keys;
 742
 743         memset(&keys, 0, sizeof(keys));
 744
 745         keys.addrs.v4addrs.src = fl4->saddr;
 746         keys.addrs.v4addrs.dst = fl4->daddr;
 747         keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
 748         keys.ports.src = fl4->fl4_sport;
 749         keys.ports.dst = fl4->fl4_dport;
 750         keys.keyid.keyid = fl4->fl4_gre_key;
 751         keys.basic.ip_proto = fl4->flowi4_proto;
 752
 753         __skb_set_sw_hash(skb, flow_hash_from_keys(&keys),
 754                           flow_keys_have_l4(&keys));
 755
 756         return skb->hash;
 757 }
 758 EXPORT_SYMBOL(__skb_get_hash_flowi4);
 759
 760 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
 761                    const struct flow_keys *keys, int hlen)
 762 {
 763         u32 poff = keys->control.thoff;
 764
 765         switch (keys->basic.ip_proto) {
 766         case IPPROTO_TCP: {
 767                 /* access doff as u8 to avoid unaligned access */
 768                 const u8 *doff;
 769                 u8 _doff;
 770
 771                 doff = __skb_header_pointer(skb, poff + 12, sizeof(_doff),
 772                                             data, hlen, &_doff);
 773                 if (!doff)
 774                         return poff;
 775
 776                 poff += max_t(u32, sizeof(struct tcphdr), (*doff & 0xF0) >> 2);
 777                 break;
 778         }
 779         case IPPROTO_UDP:
 780         case IPPROTO_UDPLITE:
 781                 poff += sizeof(struct udphdr);
 782                 break;
 783         /* For the rest, we do not really care about header
 784          * extensions at this point for now.
 785          */
 786         case IPPROTO_ICMP:
 787                 poff += sizeof(struct icmphdr);
 788                 break;
 789         case IPPROTO_ICMPV6:
 790                 poff += sizeof(struct icmp6hdr);
 791                 break;
 792         case IPPROTO_IGMP:
 793                 poff += sizeof(struct igmphdr);
 794                 break;
 795         case IPPROTO_DCCP:
 796                 poff += sizeof(struct dccp_hdr);
 797                 break;
 798         case IPPROTO_SCTP:
 799                 poff += sizeof(struct sctphdr);
 800                 break;
 801         }
 802
 803         return poff;
 804 }
 805
 806 /**
 807  * skb_get_poff - get the offset to the payload
 808  * @skb: sk_buff to get the payload offset from
 809  *
 810  * The function will get the offset to the payload as far as it could
 811  * be dissected.  The main user is currently BPF, so that we can dynamically
 812  * truncate packets without needing to push actual payload to the user
 813  * space and can analyze headers only, instead.
 814  */
 815 u32 skb_get_poff(const struct sk_buff *skb)
 816 {
 817         struct flow_keys keys;
 818
 819         if (!skb_flow_dissect_flow_keys(skb, &keys, 0))
 820                 return 0;
 821
 822         return __skb_get_poff(skb, skb->data, &keys, skb_headlen(skb));
 823 }
 824
 825 __u32 __get_hash_from_flowi6(const struct flowi6 *fl6, struct flow_keys *keys)
 826 {
 827         memset(keys, 0, sizeof(*keys));
 828
 829         memcpy(&keys->addrs.v6addrs.src, &fl6->saddr,
 830             sizeof(keys->addrs.v6addrs.src));
 831         memcpy(&keys->addrs.v6addrs.dst, &fl6->daddr,
 832             sizeof(keys->addrs.v6addrs.dst));
 833         keys->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
 834         keys->ports.src = fl6->fl6_sport;
 835         keys->ports.dst = fl6->fl6_dport;
 836         keys->keyid.keyid = fl6->fl6_gre_key;
 837         keys->tags.flow_label = (__force u32)fl6->flowlabel;
 838         keys->basic.ip_proto = fl6->flowi6_proto;
 839
 840         return flow_hash_from_keys(keys);
 841 }
 842 EXPORT_SYMBOL(__get_hash_from_flowi6);
 843
 844 __u32 __get_hash_from_flowi4(const struct flowi4 *fl4, struct flow_keys *keys)
 845 {
 846         memset(keys, 0, sizeof(*keys));
 847
 848         keys->addrs.v4addrs.src = fl4->saddr;
 849         keys->addrs.v4addrs.dst = fl4->daddr;
 850         keys->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
 851         keys->ports.src = fl4->fl4_sport;
 852         keys->ports.dst = fl4->fl4_dport;
 853         keys->keyid.keyid = fl4->fl4_gre_key;
 854         keys->basic.ip_proto = fl4->flowi4_proto;
 855
 856         return flow_hash_from_keys(keys);
 857 }
 858 EXPORT_SYMBOL(__get_hash_from_flowi4);
 859
 860 static const struct flow_dissector_key flow_keys_dissector_keys[] = {
 861         {
 862                 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
 863                 .offset = offsetof(struct flow_keys, control),
 864         },
 865         {
 866                 .key_id = FLOW_DISSECTOR_KEY_BASIC,
 867                 .offset = offsetof(struct flow_keys, basic),
 868         },
 869         {
 870                 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 871                 .offset = offsetof(struct flow_keys, addrs.v4addrs),
 872         },
 873         {
 874                 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 875                 .offset = offsetof(struct flow_keys, addrs.v6addrs),
 876         },
 877         {
 878                 .key_id = FLOW_DISSECTOR_KEY_TIPC_ADDRS,
 879                 .offset = offsetof(struct flow_keys, addrs.tipcaddrs),
 880         },
 881         {
 882                 .key_id = FLOW_DISSECTOR_KEY_PORTS,
 883                 .offset = offsetof(struct flow_keys, ports),
 884         },
 885         {
 886                 .key_id = FLOW_DISSECTOR_KEY_VLANID,
 887                 .offset = offsetof(struct flow_keys, tags),
 888         },
 889         {
 890                 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
 891                 .offset = offsetof(struct flow_keys, tags),
 892         },
 893         {
 894                 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
 895                 .offset = offsetof(struct flow_keys, keyid),
 896         },
 897 };
 898
 899 static const struct flow_dissector_key flow_keys_dissector_symmetric_keys[] = {
 900         {
 901                 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
 902                 .offset = offsetof(struct flow_keys, control),
 903         },
 904         {
 905                 .key_id = FLOW_DISSECTOR_KEY_BASIC,
 906                 .offset = offsetof(struct flow_keys, basic),
 907         },
 908         {
 909                 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 910                 .offset = offsetof(struct flow_keys, addrs.v4addrs),
 911         },
 912         {
 913                 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 914                 .offset = offsetof(struct flow_keys, addrs.v6addrs),
 915         },
 916         {
 917                 .key_id = FLOW_DISSECTOR_KEY_PORTS,
 918                 .offset = offsetof(struct flow_keys, ports),
 919         },
 920 };
 921
 922 static const struct flow_dissector_key flow_keys_buf_dissector_keys[] = {
 923         {
 924                 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
 925                 .offset = offsetof(struct flow_keys, control),
 926         },
 927         {
 928                 .key_id = FLOW_DISSECTOR_KEY_BASIC,
 929                 .offset = offsetof(struct flow_keys, basic),
 930         },
 931 };
 932
 933 struct flow_dissector flow_keys_dissector __read_mostly;
 934 EXPORT_SYMBOL(flow_keys_dissector);
 935
 936 struct flow_dissector flow_keys_buf_dissector __read_mostly;
 937
 938 static int __init init_default_flow_dissectors(void)
 939 {
 940         skb_flow_dissector_init(&flow_keys_dissector,
 941                                 flow_keys_dissector_keys,
 942                                 ARRAY_SIZE(flow_keys_dissector_keys));
 943         skb_flow_dissector_init(&flow_keys_dissector_symmetric,
 944                                 flow_keys_dissector_symmetric_keys,
 945                                 ARRAY_SIZE(flow_keys_dissector_symmetric_keys));
 946         skb_flow_dissector_init(&flow_keys_buf_dissector,
 947                                 flow_keys_buf_dissector_keys,
 948                                 ARRAY_SIZE(flow_keys_buf_dissector_keys));
 949         return 0;
 950 }
 951
 952 core_initcall(init_default_flow_dissectors);