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