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
  36 static struct kmem_cache *flow_cache;
  37 static unsigned int hash_seed __read_mostly;
  38
  39 static inline bool arphdr_ok(struct sk_buff *skb)
  40 {
  41         return skb->len >= skb_network_offset(skb) + sizeof(struct arp_eth_header);
  42 }
  43
  44 static inline int check_iphdr(struct sk_buff *skb)
  45 {
  46         unsigned int nh_ofs = skb_network_offset(skb);
  47         unsigned int ip_len;
  48
  49         if (skb->len < nh_ofs + sizeof(struct iphdr))
  50                 return -EINVAL;
  51
  52         ip_len = ip_hdrlen(skb);
  53         if (ip_len < sizeof(struct iphdr) || skb->len < nh_ofs + ip_len)
  54                 return -EINVAL;
  55
  56         /*
  57          * Pull enough header bytes to account for the IP header plus the
  58          * longest transport header that we parse, currently 20 bytes for TCP.
  59          */
  60         if (!pskb_may_pull(skb, min(nh_ofs + ip_len + 20, skb->len)))
  61                 return -ENOMEM;
  62
  63         skb_set_transport_header(skb, nh_ofs + ip_len);
  64         return 0;
  65 }
  66
  67 static inline bool tcphdr_ok(struct sk_buff *skb)
  68 {
  69         int th_ofs = skb_transport_offset(skb);
  70         if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
  71                 int tcp_len = tcp_hdrlen(skb);
  72                 return (tcp_len >= sizeof(struct tcphdr)
  73                         && skb->len >= th_ofs + tcp_len);
  74         }
  75         return false;
  76 }
  77
  78 static inline bool udphdr_ok(struct sk_buff *skb)
  79 {
  80         return skb->len >= skb_transport_offset(skb) + sizeof(struct udphdr);
  81 }
  82
  83 static inline bool icmphdr_ok(struct sk_buff *skb)
  84 {
  85         return skb->len >= skb_transport_offset(skb) + sizeof(struct icmphdr);
  86 }
  87
  88 u64 flow_used_time(unsigned long flow_jiffies)
  89 {
  90         struct timespec cur_ts;
  91         u64 cur_ms, idle_ms;
  92
  93         ktime_get_ts(&cur_ts);
  94         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
  95         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
  96                  cur_ts.tv_nsec / NSEC_PER_MSEC;
  97
  98         return cur_ms - idle_ms;
  99 }
 100
 101 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 102 {
 103         unsigned int nh_ofs = skb_network_offset(skb);
 104         unsigned int nh_len;
 105         int payload_ofs;
 106         int payload_len;
 107         struct ipv6hdr *nh;
 108         uint8_t nexthdr;
 109
 110         if (unlikely(skb->len < nh_ofs + sizeof(*nh)))
 111                 return -EINVAL;
 112
 113         nh = ipv6_hdr(skb);
 114         nexthdr = nh->nexthdr;
 115         payload_ofs = (u8 *)(nh + 1) - skb->data;
 116         payload_len = ntohs(nh->payload_len);
 117
 118         memcpy(key->ipv6_src, nh->saddr.in6_u.u6_addr8, sizeof(key->ipv6_src));
 119         memcpy(key->ipv6_dst, nh->daddr.in6_u.u6_addr8, sizeof(key->ipv6_dst));
 120         key->nw_tos = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
 121         key->nw_proto = NEXTHDR_NONE;
 122
 123         /* We don't process jumbograms. */
 124         if (!payload_len)
 125                 return -EINVAL;
 126
 127         if (unlikely(skb->len < nh_ofs + sizeof(*nh) + payload_len))
 128                 return -EINVAL;
 129
 130         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr);
 131         if (payload_ofs < 0) {
 132                 return -EINVAL;
 133         }
 134         nh_len = payload_ofs - nh_ofs;
 135
 136         /* Ensure that the payload length claimed is at least large enough
 137          * for the headers we've already processed. */
 138         if (payload_len < nh_len - sizeof(*nh))
 139                 return -EINVAL;
 140
 141         /* Pull enough header bytes to account for the IP header plus the
 142          * longest transport header that we parse, currently 20 bytes for TCP.
 143          * To dig deeper than the transport header, transport parsers may need
 144          * to pull more header bytes.
 145          */
 146         if (unlikely(!pskb_may_pull(skb, min(nh_ofs + nh_len + 20, skb->len))))
 147                 return -ENOMEM;
 148
 149         skb_set_transport_header(skb, nh_ofs + nh_len);
 150         key->nw_proto = nexthdr;
 151         return nh_len;
 152 }
 153
 154 static bool icmp6hdr_ok(struct sk_buff *skb)
 155 {
 156         return skb->len >= skb_transport_offset(skb) + sizeof(struct icmp6hdr);
 157 }
 158
 159 #define TCP_FLAGS_OFFSET 13
 160 #define TCP_FLAG_MASK 0x3f
 161
 162 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
 163 {
 164         u8 tcp_flags = 0;
 165
 166         if (flow->key.dl_type == htons(ETH_P_IP) &&
 167             flow->key.nw_proto == IPPROTO_TCP) {
 168                 u8 *tcp = (u8 *)tcp_hdr(skb);
 169                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 170         }
 171
 172         spin_lock_bh(&flow->lock);
 173         flow->used = jiffies;
 174         flow->packet_count++;
 175         flow->byte_count += skb->len;
 176         flow->tcp_flags |= tcp_flags;
 177         spin_unlock_bh(&flow->lock);
 178 }
 179
 180 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
 181 {
 182         int actions_len = nla_len(actions);
 183         struct sw_flow_actions *sfa;
 184
 185         /* At least DP_MAX_PORTS actions are required to be able to flood a
 186          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 187          * etc. */
 188         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 189                 return ERR_PTR(-EINVAL);
 190
 191         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 192         if (!sfa)
 193                 return ERR_PTR(-ENOMEM);
 194
 195         sfa->actions_len = actions_len;
 196         memcpy(sfa->actions, nla_data(actions), actions_len);
 197         return sfa;
 198 }
 199
 200 struct sw_flow *flow_alloc(void)
 201 {
 202         struct sw_flow *flow;
 203
 204         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 205         if (!flow)
 206                 return ERR_PTR(-ENOMEM);
 207
 208         spin_lock_init(&flow->lock);
 209         atomic_set(&flow->refcnt, 1);
 210         flow->dead = false;
 211
 212         return flow;
 213 }
 214
 215 void flow_free_tbl(struct tbl_node *node)
 216 {
 217         struct sw_flow *flow = flow_cast(node);
 218
 219         flow->dead = true;
 220         flow_put(flow);
 221 }
 222
 223 /* RCU callback used by flow_deferred_free. */
 224 static void rcu_free_flow_callback(struct rcu_head *rcu)
 225 {
 226         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 227
 228         flow->dead = true;
 229         flow_put(flow);
 230 }
 231
 232 /* Schedules 'flow' to be freed after the next RCU grace period.
 233  * The caller must hold rcu_read_lock for this to be sensible. */
 234 void flow_deferred_free(struct sw_flow *flow)
 235 {
 236         call_rcu(&flow->rcu, rcu_free_flow_callback);
 237 }
 238
 239 void flow_hold(struct sw_flow *flow)
 240 {
 241         atomic_inc(&flow->refcnt);
 242 }
 243
 244 void flow_put(struct sw_flow *flow)
 245 {
 246         if (unlikely(!flow))
 247                 return;
 248
 249         if (atomic_dec_and_test(&flow->refcnt)) {
 250                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 251                 kmem_cache_free(flow_cache, flow);
 252         }
 253 }
 254
 255 /* RCU callback used by flow_deferred_free_acts. */
 256 static void rcu_free_acts_callback(struct rcu_head *rcu)
 257 {
 258         struct sw_flow_actions *sf_acts = container_of(rcu,
 259                         struct sw_flow_actions, rcu);
 260         kfree(sf_acts);
 261 }
 262
 263 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 264  * The caller must hold rcu_read_lock for this to be sensible. */
 265 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 266 {
 267         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 268 }
 269
 270 static void parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 271 {
 272         struct qtag_prefix {
 273                 __be16 eth_type; /* ETH_P_8021Q */
 274                 __be16 tci;
 275         };
 276         struct qtag_prefix *qp;
 277
 278         if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
 279                 return;
 280
 281         qp = (struct qtag_prefix *) skb->data;
 282         key->dl_tci = qp->tci | htons(VLAN_TAG_PRESENT);
 283         __skb_pull(skb, sizeof(struct qtag_prefix));
 284 }
 285
 286 static __be16 parse_ethertype(struct sk_buff *skb)
 287 {
 288         struct llc_snap_hdr {
 289                 u8  dsap;  /* Always 0xAA */
 290                 u8  ssap;  /* Always 0xAA */
 291                 u8  ctrl;
 292                 u8  oui[3];
 293                 __be16 ethertype;
 294         };
 295         struct llc_snap_hdr *llc;
 296         __be16 proto;
 297
 298         proto = *(__be16 *) skb->data;
 299         __skb_pull(skb, sizeof(__be16));
 300
 301         if (ntohs(proto) >= 1536)
 302                 return proto;
 303
 304         if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
 305                 return htons(ETH_P_802_2);
 306
 307         llc = (struct llc_snap_hdr *) skb->data;
 308         if (llc->dsap != LLC_SAP_SNAP ||
 309             llc->ssap != LLC_SAP_SNAP ||
 310             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 311                 return htons(ETH_P_802_2);
 312
 313         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 314         return llc->ethertype;
 315 }
 316
 317 /**
 318  * flow_extract - extracts a flow key from an Ethernet frame.
 319  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 320  * Ethernet header
 321  * @in_port: port number on which @skb was received.
 322  * @key: output flow key
 323  * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
 324  * not contain an IPv4 packet or if it is not a fragment.
 325  *
 326  * The caller must ensure that skb->len >= ETH_HLEN.
 327  *
 328  * Returns 0 if successful, otherwise a negative errno value.
 329  *
 330  * Initializes @skb header pointers as follows:
 331  *
 332  *    - skb->mac_header: the Ethernet header.
 333  *
 334  *    - skb->network_header: just past the Ethernet header, or just past the
 335  *      VLAN header, to the first byte of the Ethernet payload.
 336  *
 337  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 338  *      on output, then just past the IP header, if one is present and
 339  *      of a correct length, otherwise the same as skb->network_header.
 340  *      For other key->dl_type values it is left untouched.
 341  */
 342 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 343                  bool *is_frag)
 344 {
 345         struct ethhdr *eth;
 346
 347         memset(key, 0, sizeof(*key));
 348         key->tun_id = OVS_CB(skb)->tun_id;
 349         key->in_port = in_port;
 350         *is_frag = false;
 351
 352         /*
 353          * We would really like to pull as many bytes as we could possibly
 354          * want to parse into the linear data area.  Currently, for IPv4,
 355          * that is:
 356          *
 357          *    14     Ethernet header
 358          *     4     VLAN header
 359          *    60     max IP header with options
 360          *    20     max TCP/UDP/ICMP header (don't care about options)
 361          *    --
 362          *    98
 363          *
 364          * But Xen only allocates 64 or 72 bytes for the linear data area in
 365          * netback, which means that we would reallocate and copy the skb's
 366          * linear data on every packet if we did that.  So instead just pull 64
 367          * bytes, which is always sufficient without IP options, and then check
 368          * whether we need to pull more later when we look at the IP header.
 369          */
 370         if (!pskb_may_pull(skb, min(skb->len, 64u)))
 371                 return -ENOMEM;
 372
 373         skb_reset_mac_header(skb);
 374
 375         /* Link layer. */
 376         eth = eth_hdr(skb);
 377         memcpy(key->dl_src, eth->h_source, ETH_ALEN);
 378         memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
 379
 380         /* dl_type, dl_vlan, dl_vlan_pcp. */
 381         __skb_pull(skb, 2 * ETH_ALEN);
 382         if (eth->h_proto == htons(ETH_P_8021Q))
 383                 parse_vlan(skb, key);
 384         key->dl_type = parse_ethertype(skb);
 385         skb_reset_network_header(skb);
 386         __skb_push(skb, skb->data - (unsigned char *)eth);
 387
 388         /* Network layer. */
 389         if (key->dl_type == htons(ETH_P_IP)) {
 390                 struct iphdr *nh;
 391                 int error;
 392
 393                 error = check_iphdr(skb);
 394                 if (unlikely(error)) {
 395                         if (error == -EINVAL) {
 396                                 skb->transport_header = skb->network_header;
 397                                 return 0;
 398                         }
 399                         return error;
 400                 }
 401
 402                 nh = ip_hdr(skb);
 403                 key->ipv4_src = nh->saddr;
 404                 key->ipv4_dst = nh->daddr;
 405                 key->nw_tos = nh->tos & ~INET_ECN_MASK;
 406                 key->nw_proto = nh->protocol;
 407
 408                 /* Transport layer. */
 409                 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET)) &&
 410                     !(skb_shinfo(skb)->gso_type & SKB_GSO_UDP)) {
 411                         if (key->nw_proto == IPPROTO_TCP) {
 412                                 if (tcphdr_ok(skb)) {
 413                                         struct tcphdr *tcp = tcp_hdr(skb);
 414                                         key->tp_src = tcp->source;
 415                                         key->tp_dst = tcp->dest;
 416                                 }
 417                         } else if (key->nw_proto == IPPROTO_UDP) {
 418                                 if (udphdr_ok(skb)) {
 419                                         struct udphdr *udp = udp_hdr(skb);
 420                                         key->tp_src = udp->source;
 421                                         key->tp_dst = udp->dest;
 422                                 }
 423                         } else if (key->nw_proto == IPPROTO_ICMP) {
 424                                 if (icmphdr_ok(skb)) {
 425                                         struct icmphdr *icmp = icmp_hdr(skb);
 426                                         /* The ICMP type and code fields use the 16-bit
 427                                          * transport port fields, so we need to store them
 428                                          * in 16-bit network byte order. */
 429                                         key->tp_src = htons(icmp->type);
 430                                         key->tp_dst = htons(icmp->code);
 431                                 }
 432                         }
 433                 } else
 434                         *is_frag = true;
 435
 436         } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 437                 struct arp_eth_header *arp;
 438
 439                 arp = (struct arp_eth_header *)skb_network_header(skb);
 440
 441                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 442                                 && arp->ar_pro == htons(ETH_P_IP)
 443                                 && arp->ar_hln == ETH_ALEN
 444                                 && arp->ar_pln == 4) {
 445
 446                         /* We only match on the lower 8 bits of the opcode. */
 447                         if (ntohs(arp->ar_op) <= 0xff)
 448                                 key->nw_proto = ntohs(arp->ar_op);
 449
 450                         if (key->nw_proto == ARPOP_REQUEST
 451                                         || key->nw_proto == ARPOP_REPLY) {
 452                                 memcpy(&key->ipv4_src, arp->ar_sip, sizeof(key->ipv4_src));
 453                                 memcpy(&key->ipv4_dst, arp->ar_tip, sizeof(key->ipv4_dst));
 454                                 memcpy(key->arp_sha, arp->ar_sha, ETH_ALEN);
 455                                 memcpy(key->arp_tha, arp->ar_tha, ETH_ALEN);
 456                         }
 457                 }
 458         } else if (key->dl_type == htons(ETH_P_IPV6)) {
 459                 int nh_len;             /* IPv6 Header + Extensions */
 460
 461                 nh_len = parse_ipv6hdr(skb, key);
 462                 if (unlikely(nh_len < 0)) {
 463                         if (nh_len == -EINVAL) {
 464                                 skb->transport_header = skb->network_header;
 465                                 return 0;
 466                         }
 467                         return nh_len;
 468                 }
 469
 470                 /* Transport layer. */
 471                 if (key->nw_proto == NEXTHDR_TCP) {
 472                         if (tcphdr_ok(skb)) {
 473                                 struct tcphdr *tcp = tcp_hdr(skb);
 474                                 key->tp_src = tcp->source;
 475                                 key->tp_dst = tcp->dest;
 476                         }
 477                 } else if (key->nw_proto == NEXTHDR_UDP) {
 478                         if (udphdr_ok(skb)) {
 479                                 struct udphdr *udp = udp_hdr(skb);
 480                                 key->tp_src = udp->source;
 481                                 key->tp_dst = udp->dest;
 482                         }
 483                 } else if (key->nw_proto == NEXTHDR_ICMP) {
 484                         if (icmp6hdr_ok(skb)) {
 485                                 struct icmp6hdr *icmp = icmp6_hdr(skb);
 486                                 /* The ICMPv6 type and code fields use the 16-bit
 487                                  * transport port fields, so we need to store them
 488                                  * in 16-bit network byte order. */
 489                                 key->tp_src = htons(icmp->icmp6_type);
 490                                 key->tp_dst = htons(icmp->icmp6_code);
 491                         }
 492                 }
 493         }
 494         return 0;
 495 }
 496
 497 u32 flow_hash(const struct sw_flow_key *key)
 498 {
 499         return jhash2((u32*)key, sizeof(*key) / sizeof(u32), hash_seed);
 500 }
 501
 502 int flow_cmp(const struct tbl_node *node, void *key2_)
 503 {
 504         const struct sw_flow_key *key1 = &flow_cast(node)->key;
 505         const struct sw_flow_key *key2 = key2_;
 506
 507         return !memcmp(key1, key2, sizeof(struct sw_flow_key));
 508 }
 509
 510 /**
 511  * flow_from_nlattrs - parses Netlink attributes into a flow key.
 512  * @swkey: receives the extracted flow key.
 513  * @key: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
 514  * sequence.
 515  *
 516  * This state machine accepts the following forms, with [] for optional
 517  * elements and | for alternatives:
 518  *
 519  * [tun_id] in_port ethernet [8021q] [ethertype \
 520  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6] | ARP]]
 521  */
 522 int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
 523 {
 524         const struct nlattr *nla;
 525         u16 prev_type;
 526         int rem;
 527
 528         memset(swkey, 0, sizeof(*swkey));
 529         swkey->dl_type = htons(ETH_P_802_2);
 530
 531         prev_type = ODP_KEY_ATTR_UNSPEC;
 532         nla_for_each_nested(nla, attr, rem) {
 533                 static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = {
 534                         [ODP_KEY_ATTR_TUN_ID] = 8,
 535                         [ODP_KEY_ATTR_IN_PORT] = 4,
 536                         [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet),
 537                         [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q),
 538                         [ODP_KEY_ATTR_ETHERTYPE] = 2,
 539                         [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4),
 540                         [ODP_KEY_ATTR_IPV6] = sizeof(struct odp_key_ipv6),
 541                         [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp),
 542                         [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp),
 543                         [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
 544                         [ODP_KEY_ATTR_ICMPV6] = sizeof(struct odp_key_icmpv6),
 545                         [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
 546                 };
 547
 548                 const struct odp_key_ethernet *eth_key;
 549                 const struct odp_key_8021q *q_key;
 550                 const struct odp_key_ipv4 *ipv4_key;
 551                 const struct odp_key_ipv6 *ipv6_key;
 552                 const struct odp_key_tcp *tcp_key;
 553                 const struct odp_key_udp *udp_key;
 554                 const struct odp_key_icmp *icmp_key;
 555                 const struct odp_key_icmpv6 *icmpv6_key;
 556                 const struct odp_key_arp *arp_key;
 557
 558                 int type = nla_type(nla);
 559
 560                 if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
 561                         return -EINVAL;
 562
 563 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
 564                 switch (TRANSITION(prev_type, type)) {
 565                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
 566                         swkey->tun_id = nla_get_be64(nla);
 567                         break;
 568
 569                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
 570                 case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
 571                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
 572                                 return -EINVAL;
 573                         swkey->in_port = nla_get_u32(nla);
 574                         break;
 575
 576                 case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET):
 577                         eth_key = nla_data(nla);
 578                         memcpy(swkey->dl_src, eth_key->eth_src, ETH_ALEN);
 579                         memcpy(swkey->dl_dst, eth_key->eth_dst, ETH_ALEN);
 580                         break;
 581
 582                 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q):
 583                         q_key = nla_data(nla);
 584                         /* Only standard 0x8100 VLANs currently supported. */
 585                         if (q_key->q_tpid != htons(ETH_P_8021Q))
 586                                 return -EINVAL;
 587                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 588                                 return -EINVAL;
 589                         swkey->dl_tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
 590                         break;
 591
 592                 case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE):
 593                 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE):
 594                         swkey->dl_type = nla_get_be16(nla);
 595                         if (ntohs(swkey->dl_type) < 1536)
 596                                 return -EINVAL;
 597                         break;
 598
 599                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4):
 600                         if (swkey->dl_type != htons(ETH_P_IP))
 601                                 return -EINVAL;
 602                         ipv4_key = nla_data(nla);
 603                         swkey->ipv4_src = ipv4_key->ipv4_src;
 604                         swkey->ipv4_dst = ipv4_key->ipv4_dst;
 605                         swkey->nw_proto = ipv4_key->ipv4_proto;
 606                         swkey->nw_tos = ipv4_key->ipv4_tos;
 607                         if (swkey->nw_tos & INET_ECN_MASK)
 608                                 return -EINVAL;
 609                         break;
 610
 611                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV6):
 612                         if (swkey->dl_type != htons(ETH_P_IPV6))
 613                                 return -EINVAL;
 614                         ipv6_key = nla_data(nla);
 615                         memcpy(swkey->ipv6_src, ipv6_key->ipv6_src,
 616                                         sizeof(swkey->ipv6_src));
 617                         memcpy(swkey->ipv6_dst, ipv6_key->ipv6_dst,
 618                                         sizeof(swkey->ipv6_dst));
 619                         swkey->nw_proto = ipv6_key->ipv6_proto;
 620                         swkey->nw_tos = ipv6_key->ipv6_tos;
 621                         if (swkey->nw_tos & INET_ECN_MASK)
 622                                 return -EINVAL;
 623                         break;
 624
 625                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP):
 626                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_TCP):
 627                         if (swkey->nw_proto != IPPROTO_TCP)
 628                                 return -EINVAL;
 629                         tcp_key = nla_data(nla);
 630                         swkey->tp_src = tcp_key->tcp_src;
 631                         swkey->tp_dst = tcp_key->tcp_dst;
 632                         break;
 633
 634                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP):
 635                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_UDP):
 636                         if (swkey->nw_proto != IPPROTO_UDP)
 637                                 return -EINVAL;
 638                         udp_key = nla_data(nla);
 639                         swkey->tp_src = udp_key->udp_src;
 640                         swkey->tp_dst = udp_key->udp_dst;
 641                         break;
 642
 643                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP):
 644                         if (swkey->nw_proto != IPPROTO_ICMP)
 645                                 return -EINVAL;
 646                         icmp_key = nla_data(nla);
 647                         swkey->tp_src = htons(icmp_key->icmp_type);
 648                         swkey->tp_dst = htons(icmp_key->icmp_code);
 649                         break;
 650
 651                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_ICMPV6):
 652                         if (swkey->nw_proto != IPPROTO_ICMPV6)
 653                                 return -EINVAL;
 654                         icmpv6_key = nla_data(nla);
 655                         swkey->tp_src = htons(icmpv6_key->icmpv6_type);
 656                         swkey->tp_dst = htons(icmpv6_key->icmpv6_code);
 657                         break;
 658
 659                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP):
 660                         if (swkey->dl_type != htons(ETH_P_ARP))
 661                                 return -EINVAL;
 662                         arp_key = nla_data(nla);
 663                         swkey->ipv4_src = arp_key->arp_sip;
 664                         swkey->ipv4_dst = arp_key->arp_tip;
 665                         if (arp_key->arp_op & htons(0xff00))
 666                                 return -EINVAL;
 667                         swkey->nw_proto = ntohs(arp_key->arp_op);
 668                         memcpy(swkey->arp_sha, arp_key->arp_sha, ETH_ALEN);
 669                         memcpy(swkey->arp_tha, arp_key->arp_tha, ETH_ALEN);
 670                         break;
 671
 672                 default:
 673                         return -EINVAL;
 674                 }
 675
 676                 prev_type = type;
 677         }
 678         if (rem)
 679                 return -EINVAL;
 680
 681         switch (prev_type) {
 682         case ODP_KEY_ATTR_UNSPEC:
 683                 return -EINVAL;
 684
 685         case ODP_KEY_ATTR_TUN_ID:
 686         case ODP_KEY_ATTR_IN_PORT:
 687                 return -EINVAL;
 688
 689         case ODP_KEY_ATTR_ETHERNET:
 690         case ODP_KEY_ATTR_8021Q:
 691                 return 0;
 692
 693         case ODP_KEY_ATTR_ETHERTYPE:
 694                 if (swkey->dl_type == htons(ETH_P_IP) ||
 695                     swkey->dl_type == htons(ETH_P_ARP))
 696                         return -EINVAL;
 697                 return 0;
 698
 699         case ODP_KEY_ATTR_IPV4:
 700                 if (swkey->nw_proto == IPPROTO_TCP ||
 701                     swkey->nw_proto == IPPROTO_UDP ||
 702                     swkey->nw_proto == IPPROTO_ICMP)
 703                         return -EINVAL;
 704                 return 0;
 705
 706         case ODP_KEY_ATTR_IPV6:
 707                 if (swkey->nw_proto == IPPROTO_TCP ||
 708                     swkey->nw_proto == IPPROTO_UDP ||
 709                     swkey->nw_proto == IPPROTO_ICMPV6)
 710                         return -EINVAL;
 711                 return 0;
 712
 713         case ODP_KEY_ATTR_TCP:
 714         case ODP_KEY_ATTR_UDP:
 715         case ODP_KEY_ATTR_ICMP:
 716         case ODP_KEY_ATTR_ICMPV6:
 717         case ODP_KEY_ATTR_ARP:
 718                 return 0;
 719         }
 720
 721         WARN_ON_ONCE(1);
 722         return -EINVAL;
 723 }
 724
 725 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
 726 {
 727         struct odp_key_ethernet *eth_key;
 728         struct nlattr *nla;
 729
 730         if (swkey->tun_id != cpu_to_be64(0))
 731                 NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id);
 732
 733         NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->in_port);
 734
 735         nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key));
 736         if (!nla)
 737                 goto nla_put_failure;
 738         eth_key = nla_data(nla);
 739         memcpy(eth_key->eth_src, swkey->dl_src, ETH_ALEN);
 740         memcpy(eth_key->eth_dst, swkey->dl_dst, ETH_ALEN);
 741
 742         if (swkey->dl_tci != htons(0)) {
 743                 struct odp_key_8021q q_key;
 744
 745                 q_key.q_tpid = htons(ETH_P_8021Q);
 746                 q_key.q_tci = swkey->dl_tci & ~htons(VLAN_TAG_PRESENT);
 747                 NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
 748         }
 749
 750         if (swkey->dl_type == htons(ETH_P_802_2))
 751                 return 0;
 752
 753         NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->dl_type);
 754
 755         if (swkey->dl_type == htons(ETH_P_IP)) {
 756                 struct odp_key_ipv4 *ipv4_key;
 757
 758                 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key));
 759                 if (!nla)
 760                         goto nla_put_failure;
 761                 ipv4_key = nla_data(nla);
 762                 ipv4_key->ipv4_src = swkey->ipv4_src;
 763                 ipv4_key->ipv4_dst = swkey->ipv4_dst;
 764                 ipv4_key->ipv4_proto = swkey->nw_proto;
 765                 ipv4_key->ipv4_tos = swkey->nw_tos;
 766         } else if (swkey->dl_type == htons(ETH_P_IPV6)) {
 767                 struct odp_key_ipv6 *ipv6_key;
 768
 769                 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV6, sizeof(*ipv6_key));
 770                 if (!nla)
 771                         goto nla_put_failure;
 772                 ipv6_key = nla_data(nla);
 773                 memcpy(ipv6_key->ipv6_src, swkey->ipv6_src,
 774                                 sizeof(ipv6_key->ipv6_src));
 775                 memcpy(ipv6_key->ipv6_dst, swkey->ipv6_dst,
 776                                 sizeof(ipv6_key->ipv6_dst));
 777                 ipv6_key->ipv6_proto = swkey->nw_proto;
 778                 ipv6_key->ipv6_tos = swkey->nw_tos;
 779         } else if (swkey->dl_type == htons(ETH_P_ARP)) {
 780                 struct odp_key_arp *arp_key;
 781
 782                 nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key));
 783                 if (!nla)
 784                         goto nla_put_failure;
 785                 arp_key = nla_data(nla);
 786                 arp_key->arp_sip = swkey->ipv4_src;
 787                 arp_key->arp_tip = swkey->ipv4_dst;
 788                 arp_key->arp_op = htons(swkey->nw_proto);
 789                 memcpy(arp_key->arp_sha, swkey->arp_sha, ETH_ALEN);
 790                 memcpy(arp_key->arp_tha, swkey->arp_tha, ETH_ALEN);
 791         }
 792
 793         if (swkey->dl_type == htons(ETH_P_IP)
 794                         || swkey->dl_type == htons(ETH_P_IPV6)) {
 795
 796                 if (swkey->nw_proto == IPPROTO_TCP) {
 797                         struct odp_key_tcp *tcp_key;
 798
 799                         nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key));
 800                         if (!nla)
 801                                 goto nla_put_failure;
 802                         tcp_key = nla_data(nla);
 803                         tcp_key->tcp_src = swkey->tp_src;
 804                         tcp_key->tcp_dst = swkey->tp_dst;
 805                 } else if (swkey->nw_proto == IPPROTO_UDP) {
 806                         struct odp_key_udp *udp_key;
 807
 808                         nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key));
 809                         if (!nla)
 810                                 goto nla_put_failure;
 811                         udp_key = nla_data(nla);
 812                         udp_key->udp_src = swkey->tp_src;
 813                         udp_key->udp_dst = swkey->tp_dst;
 814                 } else if (swkey->dl_type == htons(ETH_P_IP)
 815                                 && swkey->nw_proto == IPPROTO_ICMP) {
 816                         struct odp_key_icmp *icmp_key;
 817
 818                         nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key));
 819                         if (!nla)
 820                                 goto nla_put_failure;
 821                         icmp_key = nla_data(nla);
 822                         icmp_key->icmp_type = ntohs(swkey->tp_src);
 823                         icmp_key->icmp_code = ntohs(swkey->tp_dst);
 824                 } else if (swkey->dl_type == htons(ETH_P_IPV6)
 825                                 && swkey->nw_proto == IPPROTO_ICMPV6) {
 826                         struct odp_key_icmpv6 *icmpv6_key;
 827
 828                         nla = nla_reserve(skb, ODP_KEY_ATTR_ICMPV6, sizeof(*icmpv6_key));
 829                         if (!nla)
 830                                 goto nla_put_failure;
 831                         icmpv6_key = nla_data(nla);
 832                         icmpv6_key->icmpv6_type = ntohs(swkey->tp_src);
 833                         icmpv6_key->icmpv6_code = ntohs(swkey->tp_dst);
 834                 }
 835         }
 836
 837         return 0;
 838
 839 nla_put_failure:
 840         return -EMSGSIZE;
 841 }
 842
 843 /* Initializes the flow module.
 844  * Returns zero if successful or a negative error code. */
 845 int flow_init(void)
 846 {
 847         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
 848                                         0, NULL);
 849         if (flow_cache == NULL)
 850                 return -ENOMEM;
 851
 852         get_random_bytes(&hash_seed, sizeof(hash_seed));
 853
 854         return 0;
 855 }
 856
 857 /* Uninitializes the flow module. */
 858 void flow_exit(void)
 859 {
 860         kmem_cache_destroy(flow_cache);
 861 }