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         memcpy(key->ipv6_src, nh->saddr.in6_u.u6_addr8, sizeof(key->ipv6_src));
 120         memcpy(key->ipv6_dst, nh->daddr.in6_u.u6_addr8, sizeof(key->ipv6_dst));
 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 (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->dead = false;
 200
 201         return flow;
 202 }
 203
 204 void flow_free_tbl(struct tbl_node *node)
 205 {
 206         struct sw_flow *flow = flow_cast(node);
 207
 208         flow->dead = true;
 209         flow_put(flow);
 210 }
 211
 212 /* RCU callback used by flow_deferred_free. */
 213 static void rcu_free_flow_callback(struct rcu_head *rcu)
 214 {
 215         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 216
 217         flow->dead = true;
 218         flow_put(flow);
 219 }
 220
 221 /* Schedules 'flow' to be freed after the next RCU grace period.
 222  * The caller must hold rcu_read_lock for this to be sensible. */
 223 void flow_deferred_free(struct sw_flow *flow)
 224 {
 225         call_rcu(&flow->rcu, rcu_free_flow_callback);
 226 }
 227
 228 void flow_hold(struct sw_flow *flow)
 229 {
 230         atomic_inc(&flow->refcnt);
 231 }
 232
 233 void flow_put(struct sw_flow *flow)
 234 {
 235         if (unlikely(!flow))
 236                 return;
 237
 238         if (atomic_dec_and_test(&flow->refcnt)) {
 239                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 240                 kmem_cache_free(flow_cache, flow);
 241         }
 242 }
 243
 244 /* RCU callback used by flow_deferred_free_acts. */
 245 static void rcu_free_acts_callback(struct rcu_head *rcu)
 246 {
 247         struct sw_flow_actions *sf_acts = container_of(rcu,
 248                         struct sw_flow_actions, rcu);
 249         kfree(sf_acts);
 250 }
 251
 252 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 253  * The caller must hold rcu_read_lock for this to be sensible. */
 254 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 255 {
 256         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 257 }
 258
 259 static void parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 260 {
 261         struct qtag_prefix {
 262                 __be16 eth_type; /* ETH_P_8021Q */
 263                 __be16 tci;
 264         };
 265         struct qtag_prefix *qp;
 266
 267         if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
 268                 return;
 269
 270         qp = (struct qtag_prefix *) skb->data;
 271         key->dl_tci = qp->tci | htons(VLAN_TAG_PRESENT);
 272         __skb_pull(skb, sizeof(struct qtag_prefix));
 273 }
 274
 275 static __be16 parse_ethertype(struct sk_buff *skb)
 276 {
 277         struct llc_snap_hdr {
 278                 u8  dsap;  /* Always 0xAA */
 279                 u8  ssap;  /* Always 0xAA */
 280                 u8  ctrl;
 281                 u8  oui[3];
 282                 __be16 ethertype;
 283         };
 284         struct llc_snap_hdr *llc;
 285         __be16 proto;
 286
 287         proto = *(__be16 *) skb->data;
 288         __skb_pull(skb, sizeof(__be16));
 289
 290         if (ntohs(proto) >= 1536)
 291                 return proto;
 292
 293         if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
 294                 return htons(ETH_P_802_2);
 295
 296         llc = (struct llc_snap_hdr *) skb->data;
 297         if (llc->dsap != LLC_SAP_SNAP ||
 298             llc->ssap != LLC_SAP_SNAP ||
 299             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 300                 return htons(ETH_P_802_2);
 301
 302         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 303         return llc->ethertype;
 304 }
 305
 306 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 307                 int nh_len)
 308 {
 309         struct icmp6hdr *icmp = icmp6_hdr(skb);
 310
 311         /* The ICMPv6 type and code fields use the 16-bit transport port
 312          * fields, so we need to store them in 16-bit network byte order. */
 313         key->tp_src = htons(icmp->icmp6_type);
 314         key->tp_dst = htons(icmp->icmp6_code);
 315
 316         if (!icmp->icmp6_code
 317                         && ((icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
 318                           || (icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT))) {
 319                 int icmp_len = skb->len - skb_transport_offset(skb);
 320                 struct nd_msg *nd;
 321                 int offset;
 322
 323                 /* In order to process neighbor discovery options, we need the
 324                  * entire packet. */
 325                 if (icmp_len < sizeof(*nd))
 326                         goto invalid;
 327                 if (unlikely(!pskb_may_pull(skb, skb->len)))
 328                         return -ENOMEM;
 329
 330                 nd = (struct nd_msg *)skb_transport_header(skb);
 331                 memcpy(key->nd_target, &nd->target, sizeof(key->nd_target));
 332
 333                 icmp_len -= sizeof(*nd);
 334                 offset = 0;
 335                 while (icmp_len >= 8) {
 336                         struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd->opt + offset);
 337                         int opt_len = nd_opt->nd_opt_len * 8;
 338
 339                         if (!opt_len || (opt_len > icmp_len))
 340                                 goto invalid;
 341
 342                         /* Store the link layer address if the appropriate option is
 343                          * provided.  It is considered an error if the same link
 344                          * layer option is specified twice. */
 345                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 346                                         && opt_len == 8) {
 347                                 if (!is_zero_ether_addr(key->arp_sha))
 348                                         goto invalid;
 349                                 memcpy(key->arp_sha,
 350                                                 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 351                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 352                                         && opt_len == 8) {
 353                                 if (!is_zero_ether_addr(key->arp_tha))
 354                                         goto invalid;
 355                                 memcpy(key->arp_tha,
 356                                                 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 357                         }
 358
 359                         icmp_len -= opt_len;
 360                         offset += opt_len;
 361                 }
 362         }
 363
 364         return 0;
 365
 366 invalid:
 367         memset(key->nd_target, 0, sizeof(key->nd_target));
 368         memset(key->arp_sha, 0, sizeof(key->arp_sha));
 369         memset(key->arp_tha, 0, sizeof(key->arp_tha));
 370
 371         return 0;
 372 }
 373
 374 /**
 375  * flow_extract - extracts a flow key from an Ethernet frame.
 376  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 377  * Ethernet header
 378  * @in_port: port number on which @skb was received.
 379  * @key: output flow key
 380  * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
 381  * not contain an IPv4 packet or if it is not a fragment.
 382  *
 383  * The caller must ensure that skb->len >= ETH_HLEN.
 384  *
 385  * Returns 0 if successful, otherwise a negative errno value.
 386  *
 387  * Initializes @skb header pointers as follows:
 388  *
 389  *    - skb->mac_header: the Ethernet header.
 390  *
 391  *    - skb->network_header: just past the Ethernet header, or just past the
 392  *      VLAN header, to the first byte of the Ethernet payload.
 393  *
 394  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 395  *      on output, then just past the IP header, if one is present and
 396  *      of a correct length, otherwise the same as skb->network_header.
 397  *      For other key->dl_type values it is left untouched.
 398  */
 399 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 400                  bool *is_frag)
 401 {
 402         struct ethhdr *eth;
 403
 404         memset(key, 0, sizeof(*key));
 405         key->tun_id = OVS_CB(skb)->tun_id;
 406         key->in_port = in_port;
 407         *is_frag = false;
 408
 409         /*
 410          * We would really like to pull as many bytes as we could possibly
 411          * want to parse into the linear data area.  Currently, for IPv4,
 412          * that is:
 413          *
 414          *    14     Ethernet header
 415          *     4     VLAN header
 416          *    60     max IP header with options
 417          *    20     max TCP/UDP/ICMP header (don't care about options)
 418          *    --
 419          *    98
 420          *
 421          * But Xen only allocates 64 or 72 bytes for the linear data area in
 422          * netback, which means that we would reallocate and copy the skb's
 423          * linear data on every packet if we did that.  So instead just pull 64
 424          * bytes, which is always sufficient without IP options, and then check
 425          * whether we need to pull more later when we look at the IP header.
 426          */
 427         if (!pskb_may_pull(skb, min(skb->len, 64u)))
 428                 return -ENOMEM;
 429
 430         skb_reset_mac_header(skb);
 431
 432         /* Link layer. */
 433         eth = eth_hdr(skb);
 434         memcpy(key->dl_src, eth->h_source, ETH_ALEN);
 435         memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
 436
 437         /* dl_type, dl_vlan, dl_vlan_pcp. */
 438         __skb_pull(skb, 2 * ETH_ALEN);
 439
 440         if (vlan_tx_tag_present(skb))
 441                 key->dl_tci = htons(vlan_get_tci(skb));
 442         else if (eth->h_proto == htons(ETH_P_8021Q))
 443                 parse_vlan(skb, key);
 444
 445         key->dl_type = parse_ethertype(skb);
 446         skb_reset_network_header(skb);
 447         __skb_push(skb, skb->data - (unsigned char *)eth);
 448
 449         /* Network layer. */
 450         if (key->dl_type == htons(ETH_P_IP)) {
 451                 struct iphdr *nh;
 452                 int error;
 453
 454                 error = check_iphdr(skb);
 455                 if (unlikely(error)) {
 456                         if (error == -EINVAL) {
 457                                 skb->transport_header = skb->network_header;
 458                                 return 0;
 459                         }
 460                         return error;
 461                 }
 462
 463                 nh = ip_hdr(skb);
 464                 key->ipv4_src = nh->saddr;
 465                 key->ipv4_dst = nh->daddr;
 466                 key->nw_tos = nh->tos & ~INET_ECN_MASK;
 467                 key->nw_proto = nh->protocol;
 468
 469                 /* Transport layer. */
 470                 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET)) &&
 471                     !(skb_shinfo(skb)->gso_type & SKB_GSO_UDP)) {
 472                         if (key->nw_proto == IPPROTO_TCP) {
 473                                 if (tcphdr_ok(skb)) {
 474                                         struct tcphdr *tcp = tcp_hdr(skb);
 475                                         key->tp_src = tcp->source;
 476                                         key->tp_dst = tcp->dest;
 477                                 }
 478                         } else if (key->nw_proto == IPPROTO_UDP) {
 479                                 if (udphdr_ok(skb)) {
 480                                         struct udphdr *udp = udp_hdr(skb);
 481                                         key->tp_src = udp->source;
 482                                         key->tp_dst = udp->dest;
 483                                 }
 484                         } else if (key->nw_proto == IPPROTO_ICMP) {
 485                                 if (icmphdr_ok(skb)) {
 486                                         struct icmphdr *icmp = icmp_hdr(skb);
 487                                         /* The ICMP type and code fields use the 16-bit
 488                                          * transport port fields, so we need to store them
 489                                          * in 16-bit network byte order. */
 490                                         key->tp_src = htons(icmp->type);
 491                                         key->tp_dst = htons(icmp->code);
 492                                 }
 493                         }
 494                 } else
 495                         *is_frag = true;
 496
 497         } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 498                 struct arp_eth_header *arp;
 499
 500                 arp = (struct arp_eth_header *)skb_network_header(skb);
 501
 502                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 503                                 && arp->ar_pro == htons(ETH_P_IP)
 504                                 && arp->ar_hln == ETH_ALEN
 505                                 && arp->ar_pln == 4) {
 506
 507                         /* We only match on the lower 8 bits of the opcode. */
 508                         if (ntohs(arp->ar_op) <= 0xff)
 509                                 key->nw_proto = ntohs(arp->ar_op);
 510
 511                         if (key->nw_proto == ARPOP_REQUEST
 512                                         || key->nw_proto == ARPOP_REPLY) {
 513                                 memcpy(&key->ipv4_src, arp->ar_sip, sizeof(key->ipv4_src));
 514                                 memcpy(&key->ipv4_dst, arp->ar_tip, sizeof(key->ipv4_dst));
 515                                 memcpy(key->arp_sha, arp->ar_sha, ETH_ALEN);
 516                                 memcpy(key->arp_tha, arp->ar_tha, ETH_ALEN);
 517                         }
 518                 }
 519         } else if (key->dl_type == htons(ETH_P_IPV6)) {
 520                 int nh_len;             /* IPv6 Header + Extensions */
 521
 522                 nh_len = parse_ipv6hdr(skb, key);
 523                 if (unlikely(nh_len < 0)) {
 524                         if (nh_len == -EINVAL) {
 525                                 skb->transport_header = skb->network_header;
 526                                 return 0;
 527                         }
 528                         return nh_len;
 529                 }
 530
 531                 /* Transport layer. */
 532                 if (key->nw_proto == NEXTHDR_TCP) {
 533                         if (tcphdr_ok(skb)) {
 534                                 struct tcphdr *tcp = tcp_hdr(skb);
 535                                 key->tp_src = tcp->source;
 536                                 key->tp_dst = tcp->dest;
 537                         }
 538                 } else if (key->nw_proto == NEXTHDR_UDP) {
 539                         if (udphdr_ok(skb)) {
 540                                 struct udphdr *udp = udp_hdr(skb);
 541                                 key->tp_src = udp->source;
 542                                 key->tp_dst = udp->dest;
 543                         }
 544                 } else if (key->nw_proto == NEXTHDR_ICMP) {
 545                         if (icmp6hdr_ok(skb)) {
 546                                 int error = parse_icmpv6(skb, key, nh_len);
 547                                 if (error < 0)
 548                                         return error;
 549                         }
 550                 }
 551         }
 552         return 0;
 553 }
 554
 555 u32 flow_hash(const struct sw_flow_key *key)
 556 {
 557         return jhash2((u32*)key, sizeof(*key) / sizeof(u32), hash_seed);
 558 }
 559
 560 int flow_cmp(const struct tbl_node *node, void *key2_)
 561 {
 562         const struct sw_flow_key *key1 = &flow_cast(node)->key;
 563         const struct sw_flow_key *key2 = key2_;
 564
 565         return !memcmp(key1, key2, sizeof(struct sw_flow_key));
 566 }
 567
 568 /**
 569  * flow_from_nlattrs - parses Netlink attributes into a flow key.
 570  * @swkey: receives the extracted flow key.
 571  * @key: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
 572  * sequence.
 573  *
 574  * This state machine accepts the following forms, with [] for optional
 575  * elements and | for alternatives:
 576  *
 577  * [tun_id] in_port ethernet [8021q] [ethertype \
 578  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
 579  */
 580 int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
 581 {
 582         const struct nlattr *nla;
 583         u16 prev_type;
 584         int rem;
 585
 586         memset(swkey, 0, sizeof(*swkey));
 587         swkey->dl_type = htons(ETH_P_802_2);
 588
 589         prev_type = ODP_KEY_ATTR_UNSPEC;
 590         nla_for_each_nested(nla, attr, rem) {
 591                 static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = {
 592                         [ODP_KEY_ATTR_TUN_ID] = 8,
 593                         [ODP_KEY_ATTR_IN_PORT] = 4,
 594                         [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet),
 595                         [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q),
 596                         [ODP_KEY_ATTR_ETHERTYPE] = 2,
 597                         [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4),
 598                         [ODP_KEY_ATTR_IPV6] = sizeof(struct odp_key_ipv6),
 599                         [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp),
 600                         [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp),
 601                         [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
 602                         [ODP_KEY_ATTR_ICMPV6] = sizeof(struct odp_key_icmpv6),
 603                         [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
 604                         [ODP_KEY_ATTR_ND] = sizeof(struct odp_key_nd),
 605                 };
 606
 607                 const struct odp_key_ethernet *eth_key;
 608                 const struct odp_key_8021q *q_key;
 609                 const struct odp_key_ipv4 *ipv4_key;
 610                 const struct odp_key_ipv6 *ipv6_key;
 611                 const struct odp_key_tcp *tcp_key;
 612                 const struct odp_key_udp *udp_key;
 613                 const struct odp_key_icmp *icmp_key;
 614                 const struct odp_key_icmpv6 *icmpv6_key;
 615                 const struct odp_key_arp *arp_key;
 616                 const struct odp_key_nd *nd_key;
 617
 618                 int type = nla_type(nla);
 619
 620                 if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
 621                         return -EINVAL;
 622
 623 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
 624                 switch (TRANSITION(prev_type, type)) {
 625                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
 626                         swkey->tun_id = nla_get_be64(nla);
 627                         break;
 628
 629                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
 630                 case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
 631                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
 632                                 return -EINVAL;
 633                         swkey->in_port = nla_get_u32(nla);
 634                         break;
 635
 636                 case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET):
 637                         eth_key = nla_data(nla);
 638                         memcpy(swkey->dl_src, eth_key->eth_src, ETH_ALEN);
 639                         memcpy(swkey->dl_dst, eth_key->eth_dst, ETH_ALEN);
 640                         break;
 641
 642                 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q):
 643                         q_key = nla_data(nla);
 644                         /* Only standard 0x8100 VLANs currently supported. */
 645                         if (q_key->q_tpid != htons(ETH_P_8021Q))
 646                                 return -EINVAL;
 647                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 648                                 return -EINVAL;
 649                         swkey->dl_tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
 650                         break;
 651
 652                 case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE):
 653                 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE):
 654                         swkey->dl_type = nla_get_be16(nla);
 655                         if (ntohs(swkey->dl_type) < 1536)
 656                                 return -EINVAL;
 657                         break;
 658
 659                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4):
 660                         if (swkey->dl_type != htons(ETH_P_IP))
 661                                 return -EINVAL;
 662                         ipv4_key = nla_data(nla);
 663                         swkey->ipv4_src = ipv4_key->ipv4_src;
 664                         swkey->ipv4_dst = ipv4_key->ipv4_dst;
 665                         swkey->nw_proto = ipv4_key->ipv4_proto;
 666                         swkey->nw_tos = ipv4_key->ipv4_tos;
 667                         if (swkey->nw_tos & INET_ECN_MASK)
 668                                 return -EINVAL;
 669                         break;
 670
 671                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV6):
 672                         if (swkey->dl_type != htons(ETH_P_IPV6))
 673                                 return -EINVAL;
 674                         ipv6_key = nla_data(nla);
 675                         memcpy(swkey->ipv6_src, ipv6_key->ipv6_src,
 676                                         sizeof(swkey->ipv6_src));
 677                         memcpy(swkey->ipv6_dst, ipv6_key->ipv6_dst,
 678                                         sizeof(swkey->ipv6_dst));
 679                         swkey->nw_proto = ipv6_key->ipv6_proto;
 680                         swkey->nw_tos = ipv6_key->ipv6_tos;
 681                         if (swkey->nw_tos & INET_ECN_MASK)
 682                                 return -EINVAL;
 683                         break;
 684
 685                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP):
 686                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_TCP):
 687                         if (swkey->nw_proto != IPPROTO_TCP)
 688                                 return -EINVAL;
 689                         tcp_key = nla_data(nla);
 690                         swkey->tp_src = tcp_key->tcp_src;
 691                         swkey->tp_dst = tcp_key->tcp_dst;
 692                         break;
 693
 694                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP):
 695                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_UDP):
 696                         if (swkey->nw_proto != IPPROTO_UDP)
 697                                 return -EINVAL;
 698                         udp_key = nla_data(nla);
 699                         swkey->tp_src = udp_key->udp_src;
 700                         swkey->tp_dst = udp_key->udp_dst;
 701                         break;
 702
 703                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP):
 704                         if (swkey->nw_proto != IPPROTO_ICMP)
 705                                 return -EINVAL;
 706                         icmp_key = nla_data(nla);
 707                         swkey->tp_src = htons(icmp_key->icmp_type);
 708                         swkey->tp_dst = htons(icmp_key->icmp_code);
 709                         break;
 710
 711                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_ICMPV6):
 712                         if (swkey->nw_proto != IPPROTO_ICMPV6)
 713                                 return -EINVAL;
 714                         icmpv6_key = nla_data(nla);
 715                         swkey->tp_src = htons(icmpv6_key->icmpv6_type);
 716                         swkey->tp_dst = htons(icmpv6_key->icmpv6_code);
 717                         break;
 718
 719                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP):
 720                         if (swkey->dl_type != htons(ETH_P_ARP))
 721                                 return -EINVAL;
 722                         arp_key = nla_data(nla);
 723                         swkey->ipv4_src = arp_key->arp_sip;
 724                         swkey->ipv4_dst = arp_key->arp_tip;
 725                         if (arp_key->arp_op & htons(0xff00))
 726                                 return -EINVAL;
 727                         swkey->nw_proto = ntohs(arp_key->arp_op);
 728                         memcpy(swkey->arp_sha, arp_key->arp_sha, ETH_ALEN);
 729                         memcpy(swkey->arp_tha, arp_key->arp_tha, ETH_ALEN);
 730                         break;
 731
 732                 case TRANSITION(ODP_KEY_ATTR_ICMPV6, ODP_KEY_ATTR_ND):
 733                         if (swkey->tp_src != htons(NDISC_NEIGHBOUR_SOLICITATION)
 734                                         && swkey->tp_src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
 735                                 return -EINVAL;
 736                         nd_key = nla_data(nla);
 737                         memcpy(swkey->nd_target, nd_key->nd_target,
 738                                         sizeof(swkey->nd_target));
 739                         memcpy(swkey->arp_sha, nd_key->nd_sll, ETH_ALEN);
 740                         memcpy(swkey->arp_tha, nd_key->nd_tll, ETH_ALEN);
 741                         break;
 742
 743                 default:
 744                         return -EINVAL;
 745                 }
 746
 747                 prev_type = type;
 748         }
 749         if (rem)
 750                 return -EINVAL;
 751
 752         switch (prev_type) {
 753         case ODP_KEY_ATTR_UNSPEC:
 754                 return -EINVAL;
 755
 756         case ODP_KEY_ATTR_TUN_ID:
 757         case ODP_KEY_ATTR_IN_PORT:
 758                 return -EINVAL;
 759
 760         case ODP_KEY_ATTR_ETHERNET:
 761         case ODP_KEY_ATTR_8021Q:
 762                 return 0;
 763
 764         case ODP_KEY_ATTR_ETHERTYPE:
 765                 if (swkey->dl_type == htons(ETH_P_IP) ||
 766                     swkey->dl_type == htons(ETH_P_ARP))
 767                         return -EINVAL;
 768                 return 0;
 769
 770         case ODP_KEY_ATTR_IPV4:
 771                 if (swkey->nw_proto == IPPROTO_TCP ||
 772                     swkey->nw_proto == IPPROTO_UDP ||
 773                     swkey->nw_proto == IPPROTO_ICMP)
 774                         return -EINVAL;
 775                 return 0;
 776
 777         case ODP_KEY_ATTR_IPV6:
 778                 if (swkey->nw_proto == IPPROTO_TCP ||
 779                     swkey->nw_proto == IPPROTO_UDP ||
 780                     swkey->nw_proto == IPPROTO_ICMPV6)
 781                         return -EINVAL;
 782                 return 0;
 783
 784         case ODP_KEY_ATTR_ICMPV6:
 785                 if (swkey->tp_src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 786                     swkey->tp_src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
 787                         return -EINVAL;
 788                 return 0;
 789
 790         case ODP_KEY_ATTR_TCP:
 791         case ODP_KEY_ATTR_UDP:
 792         case ODP_KEY_ATTR_ICMP:
 793         case ODP_KEY_ATTR_ARP:
 794         case ODP_KEY_ATTR_ND:
 795                 return 0;
 796         }
 797
 798         WARN_ON_ONCE(1);
 799         return -EINVAL;
 800 }
 801
 802 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
 803 {
 804         struct odp_key_ethernet *eth_key;
 805         struct nlattr *nla;
 806
 807         /* This is an imperfect sanity-check that FLOW_BUFSIZE doesn't need
 808          * to be updated, but will at least raise awareness when new ODP key
 809          * types are added. */
 810         BUILD_BUG_ON(__ODP_KEY_ATTR_MAX != 14);
 811
 812         if (swkey->tun_id != cpu_to_be64(0))
 813                 NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id);
 814
 815         NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->in_port);
 816
 817         nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key));
 818         if (!nla)
 819                 goto nla_put_failure;
 820         eth_key = nla_data(nla);
 821         memcpy(eth_key->eth_src, swkey->dl_src, ETH_ALEN);
 822         memcpy(eth_key->eth_dst, swkey->dl_dst, ETH_ALEN);
 823
 824         if (swkey->dl_tci != htons(0)) {
 825                 struct odp_key_8021q q_key;
 826
 827                 q_key.q_tpid = htons(ETH_P_8021Q);
 828                 q_key.q_tci = swkey->dl_tci & ~htons(VLAN_TAG_PRESENT);
 829                 NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
 830         }
 831
 832         if (swkey->dl_type == htons(ETH_P_802_2))
 833                 return 0;
 834
 835         NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->dl_type);
 836
 837         if (swkey->dl_type == htons(ETH_P_IP)) {
 838                 struct odp_key_ipv4 *ipv4_key;
 839
 840                 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key));
 841                 if (!nla)
 842                         goto nla_put_failure;
 843                 ipv4_key = nla_data(nla);
 844                 memset(ipv4_key, 0, sizeof(struct odp_key_ipv4));
 845                 ipv4_key->ipv4_src = swkey->ipv4_src;
 846                 ipv4_key->ipv4_dst = swkey->ipv4_dst;
 847                 ipv4_key->ipv4_proto = swkey->nw_proto;
 848                 ipv4_key->ipv4_tos = swkey->nw_tos;
 849         } else if (swkey->dl_type == htons(ETH_P_IPV6)) {
 850                 struct odp_key_ipv6 *ipv6_key;
 851
 852                 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV6, sizeof(*ipv6_key));
 853                 if (!nla)
 854                         goto nla_put_failure;
 855                 ipv6_key = nla_data(nla);
 856                 memset(ipv6_key, 0, sizeof(struct odp_key_ipv6));
 857                 memcpy(ipv6_key->ipv6_src, swkey->ipv6_src,
 858                                 sizeof(ipv6_key->ipv6_src));
 859                 memcpy(ipv6_key->ipv6_dst, swkey->ipv6_dst,
 860                                 sizeof(ipv6_key->ipv6_dst));
 861                 ipv6_key->ipv6_proto = swkey->nw_proto;
 862                 ipv6_key->ipv6_tos = swkey->nw_tos;
 863         } else if (swkey->dl_type == htons(ETH_P_ARP)) {
 864                 struct odp_key_arp *arp_key;
 865
 866                 nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key));
 867                 if (!nla)
 868                         goto nla_put_failure;
 869                 arp_key = nla_data(nla);
 870                 memset(arp_key, 0, sizeof(struct odp_key_arp));
 871                 arp_key->arp_sip = swkey->ipv4_src;
 872                 arp_key->arp_tip = swkey->ipv4_dst;
 873                 arp_key->arp_op = htons(swkey->nw_proto);
 874                 memcpy(arp_key->arp_sha, swkey->arp_sha, ETH_ALEN);
 875                 memcpy(arp_key->arp_tha, swkey->arp_tha, ETH_ALEN);
 876         }
 877
 878         if (swkey->dl_type == htons(ETH_P_IP)
 879                         || swkey->dl_type == htons(ETH_P_IPV6)) {
 880
 881                 if (swkey->nw_proto == IPPROTO_TCP) {
 882                         struct odp_key_tcp *tcp_key;
 883
 884                         nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key));
 885                         if (!nla)
 886                                 goto nla_put_failure;
 887                         tcp_key = nla_data(nla);
 888                         tcp_key->tcp_src = swkey->tp_src;
 889                         tcp_key->tcp_dst = swkey->tp_dst;
 890                 } else if (swkey->nw_proto == IPPROTO_UDP) {
 891                         struct odp_key_udp *udp_key;
 892
 893                         nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key));
 894                         if (!nla)
 895                                 goto nla_put_failure;
 896                         udp_key = nla_data(nla);
 897                         udp_key->udp_src = swkey->tp_src;
 898                         udp_key->udp_dst = swkey->tp_dst;
 899                 } else if (swkey->dl_type == htons(ETH_P_IP)
 900                                 && swkey->nw_proto == IPPROTO_ICMP) {
 901                         struct odp_key_icmp *icmp_key;
 902
 903                         nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key));
 904                         if (!nla)
 905                                 goto nla_put_failure;
 906                         icmp_key = nla_data(nla);
 907                         icmp_key->icmp_type = ntohs(swkey->tp_src);
 908                         icmp_key->icmp_code = ntohs(swkey->tp_dst);
 909                 } else if (swkey->dl_type == htons(ETH_P_IPV6)
 910                                 && swkey->nw_proto == IPPROTO_ICMPV6) {
 911                         struct odp_key_icmpv6 *icmpv6_key;
 912
 913                         nla = nla_reserve(skb, ODP_KEY_ATTR_ICMPV6, sizeof(*icmpv6_key));
 914                         if (!nla)
 915                                 goto nla_put_failure;
 916                         icmpv6_key = nla_data(nla);
 917                         icmpv6_key->icmpv6_type = ntohs(swkey->tp_src);
 918                         icmpv6_key->icmpv6_code = ntohs(swkey->tp_dst);
 919
 920                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION
 921                                         || icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
 922                                 struct odp_key_nd *nd_key;
 923
 924                                 nla = nla_reserve(skb, ODP_KEY_ATTR_ND, sizeof(*nd_key));
 925                                 if (!nla)
 926                                         goto nla_put_failure;
 927                                 nd_key = nla_data(nla);
 928                                 memcpy(nd_key->nd_target, swkey->nd_target,
 929                                                         sizeof(nd_key->nd_target));
 930                                 memcpy(nd_key->nd_sll, swkey->arp_sha, ETH_ALEN);
 931                                 memcpy(nd_key->nd_tll, swkey->arp_tha, ETH_ALEN);
 932                         }
 933                 }
 934         }
 935
 936         return 0;
 937
 938 nla_put_failure:
 939         return -EMSGSIZE;
 940 }
 941
 942 /* Initializes the flow module.
 943  * Returns zero if successful or a negative error code. */
 944 int flow_init(void)
 945 {
 946         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
 947                                         0, NULL);
 948         if (flow_cache == NULL)
 949                 return -ENOMEM;
 950
 951         get_random_bytes(&hash_seed, sizeof(hash_seed));
 952
 953         return 0;
 954 }
 955
 956 /* Uninitializes the flow module. */
 957 void flow_exit(void)
 958 {
 959         kmem_cache_destroy(flow_cache);
 960 }