datapath/flow.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 #include "flow.h"
  10 #include "datapath.h"
  11 #include <asm/uaccess.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/if_ether.h>
  15 #include <linux/if_vlan.h>
  16 #include <net/llc_pdu.h>
  17 #include <linux/kernel.h>
  18 #include <linux/jhash.h>
  19 #include <linux/jiffies.h>
  20 #include <linux/llc.h>
  21 #include <linux/module.h>
  22 #include <linux/in.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/if_arp.h>
  25 #include <linux/if_ether.h>
  26 #include <linux/ip.h>
  27 #include <linux/ipv6.h>
  28 #include <linux/tcp.h>
  29 #include <linux/udp.h>
  30 #include <linux/icmp.h>
  31 #include <linux/icmpv6.h>
  32 #include <net/inet_ecn.h>
  33 #include <net/ip.h>
  34 #include <net/ipv6.h>
  35 #include <net/ndisc.h>
  36
  37 #include "vlan.h"
  38
  39 static struct kmem_cache *flow_cache;
  40 static unsigned int hash_seed __read_mostly;
  41
  42 static inline bool arphdr_ok(struct sk_buff *skb)
  43 {
  44         return skb->len >= skb_network_offset(skb) + sizeof(struct arp_eth_header);
  45 }
  46
  47 static inline int check_iphdr(struct sk_buff *skb)
  48 {
  49         unsigned int nh_ofs = skb_network_offset(skb);
  50         unsigned int ip_len;
  51
  52         if (skb->len < nh_ofs + sizeof(struct iphdr))
  53                 return -EINVAL;
  54
  55         ip_len = ip_hdrlen(skb);
  56         if (ip_len < sizeof(struct iphdr) || skb->len < nh_ofs + ip_len)
  57                 return -EINVAL;
  58
  59         /*
  60          * Pull enough header bytes to account for the IP header plus the
  61          * longest transport header that we parse, currently 20 bytes for TCP.
  62          */
  63         if (!pskb_may_pull(skb, min(nh_ofs + ip_len + 20, skb->len)))
  64                 return -ENOMEM;
  65
  66         skb_set_transport_header(skb, nh_ofs + ip_len);
  67         return 0;
  68 }
  69
  70 static inline bool tcphdr_ok(struct sk_buff *skb)
  71 {
  72         int th_ofs = skb_transport_offset(skb);
  73         if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
  74                 int tcp_len = tcp_hdrlen(skb);
  75                 return (tcp_len >= sizeof(struct tcphdr)
  76                         && skb->len >= th_ofs + tcp_len);
  77         }
  78         return false;
  79 }
  80
  81 static inline bool udphdr_ok(struct sk_buff *skb)
  82 {
  83         return skb->len >= skb_transport_offset(skb) + sizeof(struct udphdr);
  84 }
  85
  86 static inline bool icmphdr_ok(struct sk_buff *skb)
  87 {
  88         return skb->len >= skb_transport_offset(skb) + sizeof(struct icmphdr);
  89 }
  90
  91 u64 flow_used_time(unsigned long flow_jiffies)
  92 {
  93         struct timespec cur_ts;
  94         u64 cur_ms, idle_ms;
  95
  96         ktime_get_ts(&cur_ts);
  97         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
  98         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
  99                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 100
 101         return cur_ms - idle_ms;
 102 }
 103
 104 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 105 {
 106         unsigned int nh_ofs = skb_network_offset(skb);
 107         unsigned int nh_len;
 108         int payload_ofs;
 109         struct ipv6hdr *nh;
 110         uint8_t nexthdr;
 111
 112         if (unlikely(skb->len < nh_ofs + sizeof(*nh)))
 113                 return -EINVAL;
 114
 115         nh = ipv6_hdr(skb);
 116         nexthdr = nh->nexthdr;
 117         payload_ofs = (u8 *)(nh + 1) - skb->data;
 118
 119         ipv6_addr_copy(&key->ipv6_src, &nh->saddr);
 120         ipv6_addr_copy(&key->ipv6_dst, &nh->daddr);
 121         key->nw_tos = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
 122         key->nw_proto = NEXTHDR_NONE;
 123
 124         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr);
 125         if (unlikely(payload_ofs < 0))
 126                 return -EINVAL;
 127
 128         nh_len = payload_ofs - nh_ofs;
 129
 130         /* Pull enough header bytes to account for the IP header plus the
 131          * longest transport header that we parse, currently 20 bytes for TCP.
 132          * To dig deeper than the transport header, transport parsers may need
 133          * to pull more header bytes.
 134          */
 135         if (unlikely(!pskb_may_pull(skb, min(nh_ofs + nh_len + 20, skb->len))))
 136                 return -ENOMEM;
 137
 138         skb_set_transport_header(skb, nh_ofs + nh_len);
 139         key->nw_proto = nexthdr;
 140         return nh_len;
 141 }
 142
 143 static bool icmp6hdr_ok(struct sk_buff *skb)
 144 {
 145         return skb->len >= skb_transport_offset(skb) + sizeof(struct icmp6hdr);
 146 }
 147
 148 #define TCP_FLAGS_OFFSET 13
 149 #define TCP_FLAG_MASK 0x3f
 150
 151 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
 152 {
 153         u8 tcp_flags = 0;
 154
 155         if (flow->key.dl_type == htons(ETH_P_IP) &&
 156             flow->key.nw_proto == IPPROTO_TCP) {
 157                 u8 *tcp = (u8 *)tcp_hdr(skb);
 158                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 159         }
 160
 161         spin_lock_bh(&flow->lock);
 162         flow->used = jiffies;
 163         flow->packet_count++;
 164         flow->byte_count += skb->len;
 165         flow->tcp_flags |= tcp_flags;
 166         spin_unlock_bh(&flow->lock);
 167 }
 168
 169 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
 170 {
 171         int actions_len = nla_len(actions);
 172         struct sw_flow_actions *sfa;
 173
 174         /* At least DP_MAX_PORTS actions are required to be able to flood a
 175          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 176          * etc. */
 177         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 178                 return ERR_PTR(-EINVAL);
 179
 180         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 181         if (!sfa)
 182                 return ERR_PTR(-ENOMEM);
 183
 184         sfa->actions_len = actions_len;
 185         memcpy(sfa->actions, nla_data(actions), actions_len);
 186         return sfa;
 187 }
 188
 189 struct sw_flow *flow_alloc(void)
 190 {
 191         struct sw_flow *flow;
 192
 193         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 194         if (!flow)
 195                 return ERR_PTR(-ENOMEM);
 196
 197         spin_lock_init(&flow->lock);
 198         atomic_set(&flow->refcnt, 1);
 199         flow->sf_acts = NULL;
 200         flow->dead = false;
 201
 202         return flow;
 203 }
 204
 205 void flow_free_tbl(struct tbl_node *node)
 206 {
 207         struct sw_flow *flow = flow_cast(node);
 208
 209         flow->dead = true;
 210         flow_put(flow);
 211 }
 212
 213 /* RCU callback used by flow_deferred_free. */
 214 static void rcu_free_flow_callback(struct rcu_head *rcu)
 215 {
 216         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 217
 218         flow->dead = true;
 219         flow_put(flow);
 220 }
 221
 222 /* Schedules 'flow' to be freed after the next RCU grace period.
 223  * The caller must hold rcu_read_lock for this to be sensible. */
 224 void flow_deferred_free(struct sw_flow *flow)
 225 {
 226         call_rcu(&flow->rcu, rcu_free_flow_callback);
 227 }
 228
 229 void flow_hold(struct sw_flow *flow)
 230 {
 231         atomic_inc(&flow->refcnt);
 232 }
 233
 234 void flow_put(struct sw_flow *flow)
 235 {
 236         if (unlikely(!flow))
 237                 return;
 238
 239         if (atomic_dec_and_test(&flow->refcnt)) {
 240                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 241                 kmem_cache_free(flow_cache, flow);
 242         }
 243 }
 244
 245 /* RCU callback used by flow_deferred_free_acts. */
 246 static void rcu_free_acts_callback(struct rcu_head *rcu)
 247 {
 248         struct sw_flow_actions *sf_acts = container_of(rcu,
 249                         struct sw_flow_actions, rcu);
 250         kfree(sf_acts);
 251 }
 252
 253 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 254  * The caller must hold rcu_read_lock for this to be sensible. */
 255 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 256 {
 257         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 258 }
 259
 260 static void parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 261 {
 262         struct qtag_prefix {
 263                 __be16 eth_type; /* ETH_P_8021Q */
 264                 __be16 tci;
 265         };
 266         struct qtag_prefix *qp;
 267
 268         if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
 269                 return;
 270
 271         qp = (struct qtag_prefix *) skb->data;
 272         key->dl_tci = qp->tci | htons(VLAN_TAG_PRESENT);
 273         __skb_pull(skb, sizeof(struct qtag_prefix));
 274 }
 275
 276 static __be16 parse_ethertype(struct sk_buff *skb)
 277 {
 278         struct llc_snap_hdr {
 279                 u8  dsap;  /* Always 0xAA */
 280                 u8  ssap;  /* Always 0xAA */
 281                 u8  ctrl;
 282                 u8  oui[3];
 283                 __be16 ethertype;
 284         };
 285         struct llc_snap_hdr *llc;
 286         __be16 proto;
 287
 288         proto = *(__be16 *) skb->data;
 289         __skb_pull(skb, sizeof(__be16));
 290
 291         if (ntohs(proto) >= 1536)
 292                 return proto;
 293
 294         if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
 295                 return htons(ETH_P_802_2);
 296
 297         llc = (struct llc_snap_hdr *) skb->data;
 298         if (llc->dsap != LLC_SAP_SNAP ||
 299             llc->ssap != LLC_SAP_SNAP ||
 300             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 301                 return htons(ETH_P_802_2);
 302
 303         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 304         return llc->ethertype;
 305 }
 306
 307 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 308                         int nh_len)
 309 {
 310         struct icmp6hdr *icmp = icmp6_hdr(skb);
 311
 312         /* The ICMPv6 type and code fields use the 16-bit transport port
 313          * fields, so we need to store them in 16-bit network byte order.
 314          */
 315         key->tp_src = htons(icmp->icmp6_type);
 316         key->tp_dst = htons(icmp->icmp6_code);
 317
 318         if (icmp->icmp6_code == 0 &&
 319             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 320              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 321                 int icmp_len = skb->len - skb_transport_offset(skb);
 322                 struct nd_msg *nd;
 323                 int offset;
 324
 325                 /* In order to process neighbor discovery options, we need the
 326                  * entire packet.
 327                  */
 328                 if (unlikely(icmp_len < sizeof(*nd)))
 329                         return 0;
 330                 if (unlikely(skb_linearize(skb)))
 331                         return -ENOMEM;
 332
 333                 nd = (struct nd_msg *)skb_transport_header(skb);
 334                 ipv6_addr_copy(&key->nd_target, &nd->target);
 335
 336                 icmp_len -= sizeof(*nd);
 337                 offset = 0;
 338                 while (icmp_len >= 8) {
 339                         struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd->opt + offset);
 340                         int opt_len = nd_opt->nd_opt_len * 8;
 341
 342                         if (unlikely(!opt_len || opt_len > icmp_len))
 343                                 goto invalid;
 344
 345                         /* Store the link layer address if the appropriate
 346                          * option is provided.  It is considered an error if
 347                          * the same link layer option is specified twice.
 348                          */
 349                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 350                             && opt_len == 8) {
 351                                 if (unlikely(!is_zero_ether_addr(key->arp_sha)))
 352                                         goto invalid;
 353                                 memcpy(key->arp_sha,
 354                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 355                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 356                                    && opt_len == 8) {
 357                                 if (unlikely(!is_zero_ether_addr(key->arp_tha)))
 358                                         goto invalid;
 359                                 memcpy(key->arp_tha,
 360                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 361                         }
 362
 363                         icmp_len -= opt_len;
 364                         offset += opt_len;
 365                 }
 366         }
 367
 368         return 0;
 369
 370 invalid:
 371         memset(&key->nd_target, 0, sizeof(key->nd_target));
 372         memset(key->arp_sha, 0, sizeof(key->arp_sha));
 373         memset(key->arp_tha, 0, sizeof(key->arp_tha));
 374
 375         return 0;
 376 }
 377
 378 /**
 379  * flow_extract - extracts a flow key from an Ethernet frame.
 380  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 381  * Ethernet header
 382  * @in_port: port number on which @skb was received.
 383  * @key: output flow key
 384  * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
 385  * not contain an IPv4 packet or if it is not a fragment.
 386  *
 387  * The caller must ensure that skb->len >= ETH_HLEN.
 388  *
 389  * Returns 0 if successful, otherwise a negative errno value.
 390  *
 391  * Initializes @skb header pointers as follows:
 392  *
 393  *    - skb->mac_header: the Ethernet header.
 394  *
 395  *    - skb->network_header: just past the Ethernet header, or just past the
 396  *      VLAN header, to the first byte of the Ethernet payload.
 397  *
 398  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 399  *      on output, then just past the IP header, if one is present and
 400  *      of a correct length, otherwise the same as skb->network_header.
 401  *      For other key->dl_type values it is left untouched.
 402  */
 403 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 404                  bool *is_frag)
 405 {
 406         struct ethhdr *eth;
 407
 408         memset(key, 0, sizeof(*key));
 409         key->tun_id = OVS_CB(skb)->tun_id;
 410         key->in_port = in_port;
 411         *is_frag = false;
 412
 413         /*
 414          * We would really like to pull as many bytes as we could possibly
 415          * want to parse into the linear data area.  Currently, for IPv4,
 416          * that is:
 417          *
 418          *    14     Ethernet header
 419          *     4     VLAN header
 420          *    60     max IP header with options
 421          *    20     max TCP/UDP/ICMP header (don't care about options)
 422          *    --
 423          *    98
 424          *
 425          * But Xen only allocates 64 or 72 bytes for the linear data area in
 426          * netback, which means that we would reallocate and copy the skb's
 427          * linear data on every packet if we did that.  So instead just pull 64
 428          * bytes, which is always sufficient without IP options, and then check
 429          * whether we need to pull more later when we look at the IP header.
 430          */
 431         if (!pskb_may_pull(skb, min(skb->len, 64u)))
 432                 return -ENOMEM;
 433
 434         skb_reset_mac_header(skb);
 435
 436         /* Link layer. */
 437         eth = eth_hdr(skb);
 438         memcpy(key->dl_src, eth->h_source, ETH_ALEN);
 439         memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
 440
 441         /* dl_type, dl_vlan, dl_vlan_pcp. */
 442         __skb_pull(skb, 2 * ETH_ALEN);
 443
 444         if (vlan_tx_tag_present(skb))
 445                 key->dl_tci = htons(vlan_get_tci(skb));
 446         else if (eth->h_proto == htons(ETH_P_8021Q))
 447                 parse_vlan(skb, key);
 448
 449         key->dl_type = parse_ethertype(skb);
 450         skb_reset_network_header(skb);
 451         __skb_push(skb, skb->data - (unsigned char *)eth);
 452
 453         /* Network layer. */
 454         if (key->dl_type == htons(ETH_P_IP)) {
 455                 struct iphdr *nh;
 456                 int error;
 457
 458                 error = check_iphdr(skb);
 459                 if (unlikely(error)) {
 460                         if (error == -EINVAL) {
 461                                 skb->transport_header = skb->network_header;
 462                                 return 0;
 463                         }
 464                         return error;
 465                 }
 466
 467                 nh = ip_hdr(skb);
 468                 key->ipv4_src = nh->saddr;
 469                 key->ipv4_dst = nh->daddr;
 470                 key->nw_tos = nh->tos & ~INET_ECN_MASK;
 471                 key->nw_proto = nh->protocol;
 472
 473                 /* Transport layer. */
 474                 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET)) &&
 475                     !(skb_shinfo(skb)->gso_type & SKB_GSO_UDP)) {
 476                         if (key->nw_proto == IPPROTO_TCP) {
 477                                 if (tcphdr_ok(skb)) {
 478                                         struct tcphdr *tcp = tcp_hdr(skb);
 479                                         key->tp_src = tcp->source;
 480                                         key->tp_dst = tcp->dest;
 481                                 }
 482                         } else if (key->nw_proto == IPPROTO_UDP) {
 483                                 if (udphdr_ok(skb)) {
 484                                         struct udphdr *udp = udp_hdr(skb);
 485                                         key->tp_src = udp->source;
 486                                         key->tp_dst = udp->dest;
 487                                 }
 488                         } else if (key->nw_proto == IPPROTO_ICMP) {
 489                                 if (icmphdr_ok(skb)) {
 490                                         struct icmphdr *icmp = icmp_hdr(skb);
 491                                         /* The ICMP type and code fields use the 16-bit
 492                                          * transport port fields, so we need to store them
 493                                          * in 16-bit network byte order. */
 494                                         key->tp_src = htons(icmp->type);
 495                                         key->tp_dst = htons(icmp->code);
 496                                 }
 497                         }
 498                 } else
 499                         *is_frag = true;
 500
 501         } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 502                 struct arp_eth_header *arp;
 503
 504                 arp = (struct arp_eth_header *)skb_network_header(skb);
 505
 506                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 507                                 && arp->ar_pro == htons(ETH_P_IP)
 508                                 && arp->ar_hln == ETH_ALEN
 509                                 && arp->ar_pln == 4) {
 510
 511                         /* We only match on the lower 8 bits of the opcode. */
 512                         if (ntohs(arp->ar_op) <= 0xff)
 513                                 key->nw_proto = ntohs(arp->ar_op);
 514
 515                         if (key->nw_proto == ARPOP_REQUEST
 516                                         || key->nw_proto == ARPOP_REPLY) {
 517                                 memcpy(&key->ipv4_src, arp->ar_sip, sizeof(key->ipv4_src));
 518                                 memcpy(&key->ipv4_dst, arp->ar_tip, sizeof(key->ipv4_dst));
 519                                 memcpy(key->arp_sha, arp->ar_sha, ETH_ALEN);
 520                                 memcpy(key->arp_tha, arp->ar_tha, ETH_ALEN);
 521                         }
 522                 }
 523         } else if (key->dl_type == htons(ETH_P_IPV6)) {
 524                 int nh_len;             /* IPv6 Header + Extensions */
 525
 526                 nh_len = parse_ipv6hdr(skb, key);
 527                 if (unlikely(nh_len < 0)) {
 528                         if (nh_len == -EINVAL) {
 529                                 skb->transport_header = skb->network_header;
 530                                 return 0;
 531                         }
 532                         return nh_len;
 533                 }
 534
 535                 /* Transport layer. */
 536                 if (key->nw_proto == NEXTHDR_TCP) {
 537                         if (tcphdr_ok(skb)) {
 538                                 struct tcphdr *tcp = tcp_hdr(skb);
 539                                 key->tp_src = tcp->source;
 540                                 key->tp_dst = tcp->dest;
 541                         }
 542                 } else if (key->nw_proto == NEXTHDR_UDP) {
 543                         if (udphdr_ok(skb)) {
 544                                 struct udphdr *udp = udp_hdr(skb);
 545                                 key->tp_src = udp->source;
 546                                 key->tp_dst = udp->dest;
 547                         }
 548                 } else if (key->nw_proto == NEXTHDR_ICMP) {
 549                         if (icmp6hdr_ok(skb)) {
 550                                 int error = parse_icmpv6(skb, key, nh_len);
 551                                 if (error < 0)
 552                                         return error;
 553                         }
 554                 }
 555         }
 556         return 0;
 557 }
 558
 559 u32 flow_hash(const struct sw_flow_key *key)
 560 {
 561         return jhash2((u32*)key, sizeof(*key) / sizeof(u32), hash_seed);
 562 }
 563
 564 int flow_cmp(const struct tbl_node *node, void *key2_)
 565 {
 566         const struct sw_flow_key *key1 = &flow_cast(node)->key;
 567         const struct sw_flow_key *key2 = key2_;
 568
 569         return !memcmp(key1, key2, sizeof(struct sw_flow_key));
 570 }
 571
 572 /**
 573  * flow_from_nlattrs - parses Netlink attributes into a flow key.
 574  * @swkey: receives the extracted flow key.
 575  * @key: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
 576  * sequence.
 577  *
 578  * This state machine accepts the following forms, with [] for optional
 579  * elements and | for alternatives:
 580  *
 581  * [tun_id] in_port ethernet [8021q] [ethertype \
 582  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
 583  */
 584 int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
 585 {
 586         const struct nlattr *nla;
 587         u16 prev_type;
 588         int rem;
 589
 590         memset(swkey, 0, sizeof(*swkey));
 591         swkey->dl_type = htons(ETH_P_802_2);
 592
 593         prev_type = ODP_KEY_ATTR_UNSPEC;
 594         nla_for_each_nested(nla, attr, rem) {
 595                 static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = {
 596                         [ODP_KEY_ATTR_TUN_ID] = 8,
 597                         [ODP_KEY_ATTR_IN_PORT] = 4,
 598                         [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet),
 599                         [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q),
 600                         [ODP_KEY_ATTR_ETHERTYPE] = 2,
 601                         [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4),
 602                         [ODP_KEY_ATTR_IPV6] = sizeof(struct odp_key_ipv6),
 603                         [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp),
 604                         [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp),
 605                         [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
 606                         [ODP_KEY_ATTR_ICMPV6] = sizeof(struct odp_key_icmpv6),
 607                         [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
 608                         [ODP_KEY_ATTR_ND] = sizeof(struct odp_key_nd),
 609                 };
 610
 611                 const struct odp_key_ethernet *eth_key;
 612                 const struct odp_key_8021q *q_key;
 613                 const struct odp_key_ipv4 *ipv4_key;
 614                 const struct odp_key_ipv6 *ipv6_key;
 615                 const struct odp_key_tcp *tcp_key;
 616                 const struct odp_key_udp *udp_key;
 617                 const struct odp_key_icmp *icmp_key;
 618                 const struct odp_key_icmpv6 *icmpv6_key;
 619                 const struct odp_key_arp *arp_key;
 620                 const struct odp_key_nd *nd_key;
 621
 622                 int type = nla_type(nla);
 623
 624                 if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
 625                         return -EINVAL;
 626
 627 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
 628                 switch (TRANSITION(prev_type, type)) {
 629                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
 630                         swkey->tun_id = nla_get_be64(nla);
 631                         break;
 632
 633                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
 634                 case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
 635                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
 636                                 return -EINVAL;
 637                         swkey->in_port = nla_get_u32(nla);
 638                         break;
 639
 640                 case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET):
 641                         eth_key = nla_data(nla);
 642                         memcpy(swkey->dl_src, eth_key->eth_src, ETH_ALEN);
 643                         memcpy(swkey->dl_dst, eth_key->eth_dst, ETH_ALEN);
 644                         break;
 645
 646                 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q):
 647                         q_key = nla_data(nla);
 648                         /* Only standard 0x8100 VLANs currently supported. */
 649                         if (q_key->q_tpid != htons(ETH_P_8021Q))
 650                                 return -EINVAL;
 651                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 652                                 return -EINVAL;
 653                         swkey->dl_tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
 654                         break;
 655
 656                 case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE):
 657                 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE):
 658                         swkey->dl_type = nla_get_be16(nla);
 659                         if (ntohs(swkey->dl_type) < 1536)
 660                                 return -EINVAL;
 661                         break;
 662
 663                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4):
 664                         if (swkey->dl_type != htons(ETH_P_IP))
 665                                 return -EINVAL;
 666                         ipv4_key = nla_data(nla);
 667                         swkey->ipv4_src = ipv4_key->ipv4_src;
 668                         swkey->ipv4_dst = ipv4_key->ipv4_dst;
 669                         swkey->nw_proto = ipv4_key->ipv4_proto;
 670                         swkey->nw_tos = ipv4_key->ipv4_tos;
 671                         if (swkey->nw_tos & INET_ECN_MASK)
 672                                 return -EINVAL;
 673                         break;
 674
 675                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV6):
 676                         if (swkey->dl_type != htons(ETH_P_IPV6))
 677                                 return -EINVAL;
 678                         ipv6_key = nla_data(nla);
 679                         memcpy(&swkey->ipv6_src, ipv6_key->ipv6_src,
 680                                         sizeof(swkey->ipv6_src));
 681                         memcpy(&swkey->ipv6_dst, ipv6_key->ipv6_dst,
 682                                         sizeof(swkey->ipv6_dst));
 683                         swkey->nw_proto = ipv6_key->ipv6_proto;
 684                         swkey->nw_tos = ipv6_key->ipv6_tos;
 685                         if (swkey->nw_tos & INET_ECN_MASK)
 686                                 return -EINVAL;
 687                         break;
 688
 689                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP):
 690                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_TCP):
 691                         if (swkey->nw_proto != IPPROTO_TCP)
 692                                 return -EINVAL;
 693                         tcp_key = nla_data(nla);
 694                         swkey->tp_src = tcp_key->tcp_src;
 695                         swkey->tp_dst = tcp_key->tcp_dst;
 696                         break;
 697
 698                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP):
 699                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_UDP):
 700                         if (swkey->nw_proto != IPPROTO_UDP)
 701                                 return -EINVAL;
 702                         udp_key = nla_data(nla);
 703                         swkey->tp_src = udp_key->udp_src;
 704                         swkey->tp_dst = udp_key->udp_dst;
 705                         break;
 706
 707                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP):
 708                         if (swkey->nw_proto != IPPROTO_ICMP)
 709                                 return -EINVAL;
 710                         icmp_key = nla_data(nla);
 711                         swkey->tp_src = htons(icmp_key->icmp_type);
 712                         swkey->tp_dst = htons(icmp_key->icmp_code);
 713                         break;
 714
 715                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_ICMPV6):
 716                         if (swkey->nw_proto != IPPROTO_ICMPV6)
 717                                 return -EINVAL;
 718                         icmpv6_key = nla_data(nla);
 719                         swkey->tp_src = htons(icmpv6_key->icmpv6_type);
 720                         swkey->tp_dst = htons(icmpv6_key->icmpv6_code);
 721                         break;
 722
 723                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP):
 724                         if (swkey->dl_type != htons(ETH_P_ARP))
 725                                 return -EINVAL;
 726                         arp_key = nla_data(nla);
 727                         swkey->ipv4_src = arp_key->arp_sip;
 728                         swkey->ipv4_dst = arp_key->arp_tip;
 729                         if (arp_key->arp_op & htons(0xff00))
 730                                 return -EINVAL;
 731                         swkey->nw_proto = ntohs(arp_key->arp_op);
 732                         memcpy(swkey->arp_sha, arp_key->arp_sha, ETH_ALEN);
 733                         memcpy(swkey->arp_tha, arp_key->arp_tha, ETH_ALEN);
 734                         break;
 735
 736                 case TRANSITION(ODP_KEY_ATTR_ICMPV6, ODP_KEY_ATTR_ND):
 737                         if (swkey->tp_src != htons(NDISC_NEIGHBOUR_SOLICITATION)
 738                             && swkey->tp_src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
 739                                 return -EINVAL;
 740                         nd_key = nla_data(nla);
 741                         memcpy(&swkey->nd_target, nd_key->nd_target,
 742                                         sizeof(swkey->nd_target));
 743                         memcpy(swkey->arp_sha, nd_key->nd_sll, ETH_ALEN);
 744                         memcpy(swkey->arp_tha, nd_key->nd_tll, ETH_ALEN);
 745                         break;
 746
 747                 default:
 748                         return -EINVAL;
 749                 }
 750
 751                 prev_type = type;
 752         }
 753         if (rem)
 754                 return -EINVAL;
 755
 756         switch (prev_type) {
 757         case ODP_KEY_ATTR_UNSPEC:
 758                 return -EINVAL;
 759
 760         case ODP_KEY_ATTR_TUN_ID:
 761         case ODP_KEY_ATTR_IN_PORT:
 762                 return -EINVAL;
 763
 764         case ODP_KEY_ATTR_ETHERNET:
 765         case ODP_KEY_ATTR_8021Q:
 766                 return 0;
 767
 768         case ODP_KEY_ATTR_ETHERTYPE:
 769                 if (swkey->dl_type == htons(ETH_P_IP) ||
 770                     swkey->dl_type == htons(ETH_P_ARP))
 771                         return -EINVAL;
 772                 return 0;
 773
 774         case ODP_KEY_ATTR_IPV4:
 775                 if (swkey->nw_proto == IPPROTO_TCP ||
 776                     swkey->nw_proto == IPPROTO_UDP ||
 777                     swkey->nw_proto == IPPROTO_ICMP)
 778                         return -EINVAL;
 779                 return 0;
 780
 781         case ODP_KEY_ATTR_IPV6:
 782                 if (swkey->nw_proto == IPPROTO_TCP ||
 783                     swkey->nw_proto == IPPROTO_UDP ||
 784                     swkey->nw_proto == IPPROTO_ICMPV6)
 785                         return -EINVAL;
 786                 return 0;
 787
 788         case ODP_KEY_ATTR_ICMPV6:
 789                 if (swkey->tp_src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 790                     swkey->tp_src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
 791                         return -EINVAL;
 792                 return 0;
 793
 794         case ODP_KEY_ATTR_TCP:
 795         case ODP_KEY_ATTR_UDP:
 796         case ODP_KEY_ATTR_ICMP:
 797         case ODP_KEY_ATTR_ARP:
 798         case ODP_KEY_ATTR_ND:
 799                 return 0;
 800         }
 801
 802         WARN_ON_ONCE(1);
 803         return -EINVAL;
 804 }
 805
 806 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
 807 {
 808         struct odp_key_ethernet *eth_key;
 809         struct nlattr *nla;
 810
 811         /* This is an imperfect sanity-check that FLOW_BUFSIZE doesn't need
 812          * to be updated, but will at least raise awareness when new ODP key
 813          * types are added. */
 814         BUILD_BUG_ON(__ODP_KEY_ATTR_MAX != 14);
 815
 816         if (swkey->tun_id != cpu_to_be64(0))
 817                 NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id);
 818
 819         NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->in_port);
 820
 821         nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key));
 822         if (!nla)
 823                 goto nla_put_failure;
 824         eth_key = nla_data(nla);
 825         memcpy(eth_key->eth_src, swkey->dl_src, ETH_ALEN);
 826         memcpy(eth_key->eth_dst, swkey->dl_dst, ETH_ALEN);
 827
 828         if (swkey->dl_tci != htons(0)) {
 829                 struct odp_key_8021q q_key;
 830
 831                 q_key.q_tpid = htons(ETH_P_8021Q);
 832                 q_key.q_tci = swkey->dl_tci & ~htons(VLAN_TAG_PRESENT);
 833                 NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
 834         }
 835
 836         if (swkey->dl_type == htons(ETH_P_802_2))
 837                 return 0;
 838
 839         NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->dl_type);
 840
 841         if (swkey->dl_type == htons(ETH_P_IP)) {
 842                 struct odp_key_ipv4 *ipv4_key;
 843
 844                 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key));
 845                 if (!nla)
 846                         goto nla_put_failure;
 847                 ipv4_key = nla_data(nla);
 848                 memset(ipv4_key, 0, sizeof(struct odp_key_ipv4));
 849                 ipv4_key->ipv4_src = swkey->ipv4_src;
 850                 ipv4_key->ipv4_dst = swkey->ipv4_dst;
 851                 ipv4_key->ipv4_proto = swkey->nw_proto;
 852                 ipv4_key->ipv4_tos = swkey->nw_tos;
 853         } else if (swkey->dl_type == htons(ETH_P_IPV6)) {
 854                 struct odp_key_ipv6 *ipv6_key;
 855
 856                 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV6, sizeof(*ipv6_key));
 857                 if (!nla)
 858                         goto nla_put_failure;
 859                 ipv6_key = nla_data(nla);
 860                 memset(ipv6_key, 0, sizeof(struct odp_key_ipv6));
 861                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6_src,
 862                                 sizeof(ipv6_key->ipv6_src));
 863                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6_dst,
 864                                 sizeof(ipv6_key->ipv6_dst));
 865                 ipv6_key->ipv6_proto = swkey->nw_proto;
 866                 ipv6_key->ipv6_tos = swkey->nw_tos;
 867         } else if (swkey->dl_type == htons(ETH_P_ARP)) {
 868                 struct odp_key_arp *arp_key;
 869
 870                 nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key));
 871                 if (!nla)
 872                         goto nla_put_failure;
 873                 arp_key = nla_data(nla);
 874                 memset(arp_key, 0, sizeof(struct odp_key_arp));
 875                 arp_key->arp_sip = swkey->ipv4_src;
 876                 arp_key->arp_tip = swkey->ipv4_dst;
 877                 arp_key->arp_op = htons(swkey->nw_proto);
 878                 memcpy(arp_key->arp_sha, swkey->arp_sha, ETH_ALEN);
 879                 memcpy(arp_key->arp_tha, swkey->arp_tha, ETH_ALEN);
 880         }
 881
 882         if (swkey->dl_type == htons(ETH_P_IP) ||
 883             swkey->dl_type == htons(ETH_P_IPV6)) {
 884
 885                 if (swkey->nw_proto == IPPROTO_TCP) {
 886                         struct odp_key_tcp *tcp_key;
 887
 888                         nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key));
 889                         if (!nla)
 890                                 goto nla_put_failure;
 891                         tcp_key = nla_data(nla);
 892                         tcp_key->tcp_src = swkey->tp_src;
 893                         tcp_key->tcp_dst = swkey->tp_dst;
 894                 } else if (swkey->nw_proto == IPPROTO_UDP) {
 895                         struct odp_key_udp *udp_key;
 896
 897                         nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key));
 898                         if (!nla)
 899                                 goto nla_put_failure;
 900                         udp_key = nla_data(nla);
 901                         udp_key->udp_src = swkey->tp_src;
 902                         udp_key->udp_dst = swkey->tp_dst;
 903                 } else if (swkey->dl_type == htons(ETH_P_IP) &&
 904                            swkey->nw_proto == IPPROTO_ICMP) {
 905                         struct odp_key_icmp *icmp_key;
 906
 907                         nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key));
 908                         if (!nla)
 909                                 goto nla_put_failure;
 910                         icmp_key = nla_data(nla);
 911                         icmp_key->icmp_type = ntohs(swkey->tp_src);
 912                         icmp_key->icmp_code = ntohs(swkey->tp_dst);
 913                 } else if (swkey->dl_type == htons(ETH_P_IPV6) &&
 914                            swkey->nw_proto == IPPROTO_ICMPV6) {
 915                         struct odp_key_icmpv6 *icmpv6_key;
 916
 917                         nla = nla_reserve(skb, ODP_KEY_ATTR_ICMPV6,
 918                                                 sizeof(*icmpv6_key));
 919                         if (!nla)
 920                                 goto nla_put_failure;
 921                         icmpv6_key = nla_data(nla);
 922                         icmpv6_key->icmpv6_type = ntohs(swkey->tp_src);
 923                         icmpv6_key->icmpv6_code = ntohs(swkey->tp_dst);
 924
 925                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 926                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
 927                                 struct odp_key_nd *nd_key;
 928
 929                                 nla = nla_reserve(skb, ODP_KEY_ATTR_ND, sizeof(*nd_key));
 930                                 if (!nla)
 931                                         goto nla_put_failure;
 932                                 nd_key = nla_data(nla);
 933                                 memcpy(nd_key->nd_target, &swkey->nd_target,
 934                                                         sizeof(nd_key->nd_target));
 935                                 memcpy(nd_key->nd_sll, swkey->arp_sha, ETH_ALEN);
 936                                 memcpy(nd_key->nd_tll, swkey->arp_tha, ETH_ALEN);
 937                         }
 938                 }
 939         }
 940
 941         return 0;
 942
 943 nla_put_failure:
 944         return -EMSGSIZE;
 945 }
 946
 947 /* Initializes the flow module.
 948  * Returns zero if successful or a negative error code. */
 949 int flow_init(void)
 950 {
 951         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
 952                                         0, NULL);
 953         if (flow_cache == NULL)
 954                 return -ENOMEM;
 955
 956         get_random_bytes(&hash_seed, sizeof(hash_seed));
 957
 958         return 0;
 959 }
 960
 961 /* Uninitializes the flow module. */
 962 void flow_exit(void)
 963 {
 964         kmem_cache_destroy(flow_cache);
 965 }