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 int check_header(struct sk_buff *skb, int len)
  43 {
  44         if (unlikely(skb->len < len))
  45                 return -EINVAL;
  46         if (unlikely(!pskb_may_pull(skb, len)))
  47                 return -ENOMEM;
  48         return 0;
  49 }
  50
  51 static inline bool arphdr_ok(struct sk_buff *skb)
  52 {
  53         return pskb_may_pull(skb, skb_network_offset(skb) +
  54                                   sizeof(struct arp_eth_header));
  55 }
  56
  57 static inline int check_iphdr(struct sk_buff *skb)
  58 {
  59         unsigned int nh_ofs = skb_network_offset(skb);
  60         unsigned int ip_len;
  61         int err;
  62
  63         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  64         if (unlikely(err))
  65                 return err;
  66
  67         ip_len = ip_hdrlen(skb);
  68         if (unlikely(ip_len < sizeof(struct iphdr) ||
  69                      skb->len < nh_ofs + ip_len))
  70                 return -EINVAL;
  71
  72         skb_set_transport_header(skb, nh_ofs + ip_len);
  73         return 0;
  74 }
  75
  76 static inline bool tcphdr_ok(struct sk_buff *skb)
  77 {
  78         int th_ofs = skb_transport_offset(skb);
  79         int tcp_len;
  80
  81         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  82                 return false;
  83
  84         tcp_len = tcp_hdrlen(skb);
  85         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  86                      skb->len < th_ofs + tcp_len))
  87                 return false;
  88
  89         return true;
  90 }
  91
  92 static inline bool udphdr_ok(struct sk_buff *skb)
  93 {
  94         return pskb_may_pull(skb, skb_transport_offset(skb) +
  95                                   sizeof(struct udphdr));
  96 }
  97
  98 static inline bool icmphdr_ok(struct sk_buff *skb)
  99 {
 100         return pskb_may_pull(skb, skb_transport_offset(skb) +
 101                                   sizeof(struct icmphdr));
 102 }
 103
 104 u64 flow_used_time(unsigned long flow_jiffies)
 105 {
 106         struct timespec cur_ts;
 107         u64 cur_ms, idle_ms;
 108
 109         ktime_get_ts(&cur_ts);
 110         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 111         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 112                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 113
 114         return cur_ms - idle_ms;
 115 }
 116
 117 #define SW_FLOW_KEY_OFFSET(field)               \
 118         offsetof(struct sw_flow_key, field) +   \
 119         FIELD_SIZEOF(struct sw_flow_key, field)
 120
 121 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 122                          int *key_lenp)
 123 {
 124         unsigned int nh_ofs = skb_network_offset(skb);
 125         unsigned int nh_len;
 126         int payload_ofs;
 127         struct ipv6hdr *nh;
 128         uint8_t nexthdr;
 129         int err;
 130
 131         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.addr);
 132
 133         err = check_header(skb, nh_ofs + sizeof(*nh));
 134         if (unlikely(err))
 135                 return err;
 136
 137         nh = ipv6_hdr(skb);
 138         nexthdr = nh->nexthdr;
 139         payload_ofs = (u8 *)(nh + 1) - skb->data;
 140
 141         key->ip.nw_proto = NEXTHDR_NONE;
 142         key->ip.nw_tos = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
 143         ipv6_addr_copy(&key->ipv6.addr.src, &nh->saddr);
 144         ipv6_addr_copy(&key->ipv6.addr.dst, &nh->daddr);
 145
 146         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr);
 147         if (unlikely(payload_ofs < 0))
 148                 return -EINVAL;
 149
 150         nh_len = payload_ofs - nh_ofs;
 151         skb_set_transport_header(skb, nh_ofs + nh_len);
 152         key->ip.nw_proto = nexthdr;
 153         return nh_len;
 154 }
 155
 156 static bool icmp6hdr_ok(struct sk_buff *skb)
 157 {
 158         return pskb_may_pull(skb, skb_transport_offset(skb) +
 159                                   sizeof(struct icmp6hdr));
 160 }
 161
 162 #define TCP_FLAGS_OFFSET 13
 163 #define TCP_FLAG_MASK 0x3f
 164
 165 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
 166 {
 167         u8 tcp_flags = 0;
 168
 169         if (flow->key.eth.type == htons(ETH_P_IP) &&
 170             flow->key.ip.nw_proto == IPPROTO_TCP) {
 171                 u8 *tcp = (u8 *)tcp_hdr(skb);
 172                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 173         }
 174
 175         spin_lock_bh(&flow->lock);
 176         flow->used = jiffies;
 177         flow->packet_count++;
 178         flow->byte_count += skb->len;
 179         flow->tcp_flags |= tcp_flags;
 180         spin_unlock_bh(&flow->lock);
 181 }
 182
 183 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
 184 {
 185         int actions_len = nla_len(actions);
 186         struct sw_flow_actions *sfa;
 187
 188         /* At least DP_MAX_PORTS actions are required to be able to flood a
 189          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 190          * etc. */
 191         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 192                 return ERR_PTR(-EINVAL);
 193
 194         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 195         if (!sfa)
 196                 return ERR_PTR(-ENOMEM);
 197
 198         sfa->actions_len = actions_len;
 199         memcpy(sfa->actions, nla_data(actions), actions_len);
 200         return sfa;
 201 }
 202
 203 struct sw_flow *flow_alloc(void)
 204 {
 205         struct sw_flow *flow;
 206
 207         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 208         if (!flow)
 209                 return ERR_PTR(-ENOMEM);
 210
 211         spin_lock_init(&flow->lock);
 212         atomic_set(&flow->refcnt, 1);
 213         flow->sf_acts = NULL;
 214         flow->dead = false;
 215
 216         return flow;
 217 }
 218
 219 void flow_free_tbl(struct tbl_node *node)
 220 {
 221         struct sw_flow *flow = flow_cast(node);
 222
 223         flow->dead = true;
 224         flow_put(flow);
 225 }
 226
 227 /* RCU callback used by flow_deferred_free. */
 228 static void rcu_free_flow_callback(struct rcu_head *rcu)
 229 {
 230         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 231
 232         flow->dead = true;
 233         flow_put(flow);
 234 }
 235
 236 /* Schedules 'flow' to be freed after the next RCU grace period.
 237  * The caller must hold rcu_read_lock for this to be sensible. */
 238 void flow_deferred_free(struct sw_flow *flow)
 239 {
 240         call_rcu(&flow->rcu, rcu_free_flow_callback);
 241 }
 242
 243 void flow_hold(struct sw_flow *flow)
 244 {
 245         atomic_inc(&flow->refcnt);
 246 }
 247
 248 void flow_put(struct sw_flow *flow)
 249 {
 250         if (unlikely(!flow))
 251                 return;
 252
 253         if (atomic_dec_and_test(&flow->refcnt)) {
 254                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 255                 kmem_cache_free(flow_cache, flow);
 256         }
 257 }
 258
 259 /* RCU callback used by flow_deferred_free_acts. */
 260 static void rcu_free_acts_callback(struct rcu_head *rcu)
 261 {
 262         struct sw_flow_actions *sf_acts = container_of(rcu,
 263                         struct sw_flow_actions, rcu);
 264         kfree(sf_acts);
 265 }
 266
 267 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 268  * The caller must hold rcu_read_lock for this to be sensible. */
 269 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 270 {
 271         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 272 }
 273
 274 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 275 {
 276         struct qtag_prefix {
 277                 __be16 eth_type; /* ETH_P_8021Q */
 278                 __be16 tci;
 279         };
 280         struct qtag_prefix *qp;
 281
 282         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 283                                          sizeof(__be16))))
 284                 return -ENOMEM;
 285
 286         qp = (struct qtag_prefix *) skb->data;
 287         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 288         __skb_pull(skb, sizeof(struct qtag_prefix));
 289
 290         return 0;
 291 }
 292
 293 static __be16 parse_ethertype(struct sk_buff *skb)
 294 {
 295         struct llc_snap_hdr {
 296                 u8  dsap;  /* Always 0xAA */
 297                 u8  ssap;  /* Always 0xAA */
 298                 u8  ctrl;
 299                 u8  oui[3];
 300                 __be16 ethertype;
 301         };
 302         struct llc_snap_hdr *llc;
 303         __be16 proto;
 304
 305         proto = *(__be16 *) skb->data;
 306         __skb_pull(skb, sizeof(__be16));
 307
 308         if (ntohs(proto) >= 1536)
 309                 return proto;
 310
 311         if (skb->len < sizeof(struct llc_snap_hdr))
 312                 return htons(ETH_P_802_2);
 313
 314         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 315                 return htons(0);
 316
 317         llc = (struct llc_snap_hdr *) skb->data;
 318         if (llc->dsap != LLC_SAP_SNAP ||
 319             llc->ssap != LLC_SAP_SNAP ||
 320             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 321                 return htons(ETH_P_802_2);
 322
 323         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 324         return llc->ethertype;
 325 }
 326
 327 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 328                         int *key_lenp, int nh_len)
 329 {
 330         struct icmp6hdr *icmp = icmp6_hdr(skb);
 331         int error = 0;
 332         int key_len;
 333
 334         /* The ICMPv6 type and code fields use the 16-bit transport port
 335          * fields, so we need to store them in 16-bit network byte order.
 336          */
 337         key->ipv6.tp.src = htons(icmp->icmp6_type);
 338         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 339         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 340
 341         if (icmp->icmp6_code == 0 &&
 342             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 343              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 344                 int icmp_len = skb->len - skb_transport_offset(skb);
 345                 struct nd_msg *nd;
 346                 int offset;
 347
 348                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 349
 350                 /* In order to process neighbor discovery options, we need the
 351                  * entire packet.
 352                  */
 353                 if (unlikely(icmp_len < sizeof(*nd)))
 354                         goto out;
 355                 if (unlikely(skb_linearize(skb))) {
 356                         error = -ENOMEM;
 357                         goto out;
 358                 }
 359
 360                 nd = (struct nd_msg *)skb_transport_header(skb);
 361                 ipv6_addr_copy(&key->ipv6.nd.target, &nd->target);
 362                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 363
 364                 icmp_len -= sizeof(*nd);
 365                 offset = 0;
 366                 while (icmp_len >= 8) {
 367                         struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd->opt + offset);
 368                         int opt_len = nd_opt->nd_opt_len * 8;
 369
 370                         if (unlikely(!opt_len || opt_len > icmp_len))
 371                                 goto invalid;
 372
 373                         /* Store the link layer address if the appropriate
 374                          * option is provided.  It is considered an error if
 375                          * the same link layer option is specified twice.
 376                          */
 377                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 378                             && opt_len == 8) {
 379                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 380                                         goto invalid;
 381                                 memcpy(key->ipv6.nd.sll,
 382                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 383                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 384                                    && opt_len == 8) {
 385                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 386                                         goto invalid;
 387                                 memcpy(key->ipv6.nd.tll,
 388                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 389                         }
 390
 391                         icmp_len -= opt_len;
 392                         offset += opt_len;
 393                 }
 394         }
 395
 396         goto out;
 397
 398 invalid:
 399         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 400         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 401         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 402
 403 out:
 404         *key_lenp = key_len;
 405         return error;
 406 }
 407
 408 /**
 409  * flow_extract - extracts a flow key from an Ethernet frame.
 410  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 411  * Ethernet header
 412  * @in_port: port number on which @skb was received.
 413  * @key: output flow key
 414  * @key_lenp: length of output flow key
 415  * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
 416  * not contain an IPv4 packet or if it is not a fragment.
 417  *
 418  * The caller must ensure that skb->len >= ETH_HLEN.
 419  *
 420  * Returns 0 if successful, otherwise a negative errno value.
 421  *
 422  * Initializes @skb header pointers as follows:
 423  *
 424  *    - skb->mac_header: the Ethernet header.
 425  *
 426  *    - skb->network_header: just past the Ethernet header, or just past the
 427  *      VLAN header, to the first byte of the Ethernet payload.
 428  *
 429  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 430  *      on output, then just past the IP header, if one is present and
 431  *      of a correct length, otherwise the same as skb->network_header.
 432  *      For other key->dl_type values it is left untouched.
 433  */
 434 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 435                  int *key_lenp, bool *is_frag)
 436 {
 437         int error = 0;
 438         int key_len = SW_FLOW_KEY_OFFSET(eth);
 439         struct ethhdr *eth;
 440
 441         memset(key, 0, sizeof(*key));
 442         key->eth.tun_id = OVS_CB(skb)->tun_id;
 443         key->eth.in_port = in_port;
 444         *is_frag = false;
 445
 446         skb_reset_mac_header(skb);
 447
 448         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 449          * header in the linear data area.
 450          */
 451         eth = eth_hdr(skb);
 452         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 453         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 454
 455         __skb_pull(skb, 2 * ETH_ALEN);
 456
 457         if (vlan_tx_tag_present(skb))
 458                 key->eth.tci = htons(vlan_get_tci(skb));
 459         else if (eth->h_proto == htons(ETH_P_8021Q))
 460                 if (unlikely(parse_vlan(skb, key)))
 461                         return -ENOMEM;
 462
 463         key->eth.type = parse_ethertype(skb);
 464         if (unlikely(key->eth.type == htons(0)))
 465                 return -ENOMEM;
 466
 467         skb_reset_network_header(skb);
 468         __skb_push(skb, skb->data - skb_mac_header(skb));
 469
 470         /* Network layer. */
 471         if (key->eth.type == htons(ETH_P_IP)) {
 472                 struct iphdr *nh;
 473
 474                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 475
 476                 error = check_iphdr(skb);
 477                 if (unlikely(error)) {
 478                         if (error == -EINVAL) {
 479                                 skb->transport_header = skb->network_header;
 480                                 error = 0;
 481                         }
 482                         goto out;
 483                 }
 484
 485                 nh = ip_hdr(skb);
 486                 key->ipv4.addr.src = nh->saddr;
 487                 key->ipv4.addr.dst = nh->daddr;
 488                 key->ip.nw_tos = nh->tos & ~INET_ECN_MASK;
 489                 key->ip.nw_proto = nh->protocol;
 490
 491                 /* Transport layer. */
 492                 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET)) &&
 493                     !(skb_shinfo(skb)->gso_type & SKB_GSO_UDP)) {
 494                         if (key->ip.nw_proto == IPPROTO_TCP) {
 495                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 496                                 if (tcphdr_ok(skb)) {
 497                                         struct tcphdr *tcp = tcp_hdr(skb);
 498                                         key->ipv4.tp.src = tcp->source;
 499                                         key->ipv4.tp.dst = tcp->dest;
 500                                 }
 501                         } else if (key->ip.nw_proto == IPPROTO_UDP) {
 502                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 503                                 if (udphdr_ok(skb)) {
 504                                         struct udphdr *udp = udp_hdr(skb);
 505                                         key->ipv4.tp.src = udp->source;
 506                                         key->ipv4.tp.dst = udp->dest;
 507                                 }
 508                         } else if (key->ip.nw_proto == IPPROTO_ICMP) {
 509                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 510                                 if (icmphdr_ok(skb)) {
 511                                         struct icmphdr *icmp = icmp_hdr(skb);
 512                                         /* The ICMP type and code fields use the 16-bit
 513                                          * transport port fields, so we need to store them
 514                                          * in 16-bit network byte order. */
 515                                         key->ipv4.tp.src = htons(icmp->type);
 516                                         key->ipv4.tp.dst = htons(icmp->code);
 517                                 }
 518                         }
 519                 } else
 520                         *is_frag = true;
 521
 522         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 523                 struct arp_eth_header *arp;
 524
 525                 arp = (struct arp_eth_header *)skb_network_header(skb);
 526
 527                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 528                                 && arp->ar_pro == htons(ETH_P_IP)
 529                                 && arp->ar_hln == ETH_ALEN
 530                                 && arp->ar_pln == 4) {
 531
 532                         /* We only match on the lower 8 bits of the opcode. */
 533                         if (ntohs(arp->ar_op) <= 0xff)
 534                                 key->ip.nw_proto = ntohs(arp->ar_op);
 535
 536                         if (key->ip.nw_proto == ARPOP_REQUEST
 537                                         || key->ip.nw_proto == ARPOP_REPLY) {
 538                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 539                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 540                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 541                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 542                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 543                         }
 544                 }
 545         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 546                 int nh_len;             /* IPv6 Header + Extensions */
 547
 548                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 549                 if (unlikely(nh_len < 0)) {
 550                         if (nh_len == -EINVAL)
 551                                 skb->transport_header = skb->network_header;
 552                         else
 553                                 error = nh_len;
 554                         goto out;
 555                 }
 556
 557                 /* Transport layer. */
 558                 if (key->ip.nw_proto == NEXTHDR_TCP) {
 559                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 560                         if (tcphdr_ok(skb)) {
 561                                 struct tcphdr *tcp = tcp_hdr(skb);
 562                                 key->ipv6.tp.src = tcp->source;
 563                                 key->ipv6.tp.dst = tcp->dest;
 564                         }
 565                 } else if (key->ip.nw_proto == NEXTHDR_UDP) {
 566                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 567                         if (udphdr_ok(skb)) {
 568                                 struct udphdr *udp = udp_hdr(skb);
 569                                 key->ipv6.tp.src = udp->source;
 570                                 key->ipv6.tp.dst = udp->dest;
 571                         }
 572                 } else if (key->ip.nw_proto == NEXTHDR_ICMP) {
 573                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 574                         if (icmp6hdr_ok(skb)) {
 575                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 576                                 if (error < 0)
 577                                         goto out;
 578                         }
 579                 }
 580         }
 581
 582 out:
 583         *key_lenp = key_len;
 584         return error;
 585 }
 586
 587 u32 flow_hash(const struct sw_flow_key *key, int key_len)
 588 {
 589         return jhash2((u32*)key, DIV_ROUND_UP(key_len, sizeof(u32)), hash_seed);
 590 }
 591
 592 int flow_cmp(const struct tbl_node *node, void *key2_, int len)
 593 {
 594         const struct sw_flow_key *key1 = &flow_cast(node)->key;
 595         const struct sw_flow_key *key2 = key2_;
 596
 597         return !memcmp(key1, key2, len);
 598 }
 599
 600 /**
 601  * flow_from_nlattrs - parses Netlink attributes into a flow key.
 602  * @swkey: receives the extracted flow key.
 603  * @key_lenp: number of bytes used in @swkey.
 604  * @attr: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
 605  * sequence.
 606  *
 607  * This state machine accepts the following forms, with [] for optional
 608  * elements and | for alternatives:
 609  *
 610  * [tun_id] in_port ethernet [8021q] [ethertype \
 611  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
 612  */
 613 int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 614                       const struct nlattr *attr)
 615 {
 616         int error = 0;
 617         const struct nlattr *nla;
 618         u16 prev_type;
 619         int rem;
 620         int key_len;
 621
 622         memset(swkey, 0, sizeof(*swkey));
 623         swkey->eth.type = htons(ETH_P_802_2);
 624         key_len = SW_FLOW_KEY_OFFSET(eth);
 625
 626         prev_type = ODP_KEY_ATTR_UNSPEC;
 627         nla_for_each_nested(nla, attr, rem) {
 628                 static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = {
 629                         [ODP_KEY_ATTR_TUN_ID] = 8,
 630                         [ODP_KEY_ATTR_IN_PORT] = 4,
 631                         [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet),
 632                         [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q),
 633                         [ODP_KEY_ATTR_ETHERTYPE] = 2,
 634                         [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4),
 635                         [ODP_KEY_ATTR_IPV6] = sizeof(struct odp_key_ipv6),
 636                         [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp),
 637                         [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp),
 638                         [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
 639                         [ODP_KEY_ATTR_ICMPV6] = sizeof(struct odp_key_icmpv6),
 640                         [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
 641                         [ODP_KEY_ATTR_ND] = sizeof(struct odp_key_nd),
 642                 };
 643
 644                 const struct odp_key_ethernet *eth_key;
 645                 const struct odp_key_8021q *q_key;
 646                 const struct odp_key_ipv4 *ipv4_key;
 647                 const struct odp_key_ipv6 *ipv6_key;
 648                 const struct odp_key_tcp *tcp_key;
 649                 const struct odp_key_udp *udp_key;
 650                 const struct odp_key_icmp *icmp_key;
 651                 const struct odp_key_icmpv6 *icmpv6_key;
 652                 const struct odp_key_arp *arp_key;
 653                 const struct odp_key_nd *nd_key;
 654
 655                 int type = nla_type(nla);
 656
 657                 if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
 658                         goto invalid;
 659
 660 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
 661                 switch (TRANSITION(prev_type, type)) {
 662                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
 663                         swkey->eth.tun_id = nla_get_be64(nla);
 664                         break;
 665
 666                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
 667                 case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
 668                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
 669                                 goto invalid;
 670                         swkey->eth.in_port = nla_get_u32(nla);
 671                         break;
 672
 673                 case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET):
 674                         eth_key = nla_data(nla);
 675                         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
 676                         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
 677                         break;
 678
 679                 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q):
 680                         q_key = nla_data(nla);
 681                         /* Only standard 0x8100 VLANs currently supported. */
 682                         if (q_key->q_tpid != htons(ETH_P_8021Q))
 683                                 goto invalid;
 684                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 685                                 goto invalid;
 686                         swkey->eth.tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
 687                         break;
 688
 689                 case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE):
 690                 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE):
 691                         swkey->eth.type = nla_get_be16(nla);
 692                         if (ntohs(swkey->eth.type) < 1536)
 693                                 goto invalid;
 694                         break;
 695
 696                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4):
 697                         key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 698                         if (swkey->eth.type != htons(ETH_P_IP))
 699                                 goto invalid;
 700                         ipv4_key = nla_data(nla);
 701                         swkey->ip.nw_proto = ipv4_key->ipv4_proto;
 702                         swkey->ip.nw_tos = ipv4_key->ipv4_tos;
 703                         swkey->ipv4.addr.src = ipv4_key->ipv4_src;
 704                         swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
 705                         if (swkey->ip.nw_tos & INET_ECN_MASK)
 706                                 goto invalid;
 707                         break;
 708
 709                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV6):
 710                         key_len = SW_FLOW_KEY_OFFSET(ipv6.addr);
 711                         if (swkey->eth.type != htons(ETH_P_IPV6))
 712                                 goto invalid;
 713                         ipv6_key = nla_data(nla);
 714                         swkey->ip.nw_proto = ipv6_key->ipv6_proto;
 715                         swkey->ip.nw_tos = ipv6_key->ipv6_tos;
 716                         memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
 717                                         sizeof(swkey->ipv6.addr.src));
 718                         memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
 719                                         sizeof(swkey->ipv6.addr.dst));
 720                         if (swkey->ip.nw_tos & INET_ECN_MASK)
 721                                 goto invalid;
 722                         break;
 723
 724                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP):
 725                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 726                         if (swkey->ip.nw_proto != IPPROTO_TCP)
 727                                 goto invalid;
 728                         tcp_key = nla_data(nla);
 729                         swkey->ipv4.tp.src = tcp_key->tcp_src;
 730                         swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 731                         break;
 732
 733                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_TCP):
 734                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 735                         if (swkey->ip.nw_proto != IPPROTO_TCP)
 736                                 goto invalid;
 737                         tcp_key = nla_data(nla);
 738                         swkey->ipv6.tp.src = tcp_key->tcp_src;
 739                         swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 740                         break;
 741
 742                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP):
 743                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 744                         if (swkey->ip.nw_proto != IPPROTO_UDP)
 745                                 goto invalid;
 746                         udp_key = nla_data(nla);
 747                         swkey->ipv4.tp.src = udp_key->udp_src;
 748                         swkey->ipv4.tp.dst = udp_key->udp_dst;
 749                         break;
 750
 751                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_UDP):
 752                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 753                         if (swkey->ip.nw_proto != IPPROTO_UDP)
 754                                 goto invalid;
 755                         udp_key = nla_data(nla);
 756                         swkey->ipv6.tp.src = udp_key->udp_src;
 757                         swkey->ipv6.tp.dst = udp_key->udp_dst;
 758                         break;
 759
 760                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP):
 761                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 762                         if (swkey->ip.nw_proto != IPPROTO_ICMP)
 763                                 goto invalid;
 764                         icmp_key = nla_data(nla);
 765                         swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 766                         swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 767                         break;
 768
 769                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_ICMPV6):
 770                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 771                         if (swkey->ip.nw_proto != IPPROTO_ICMPV6)
 772                                 goto invalid;
 773                         icmpv6_key = nla_data(nla);
 774                         swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 775                         swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 776                         break;
 777
 778                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP):
 779                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 780                         if (swkey->eth.type != htons(ETH_P_ARP))
 781                                 goto invalid;
 782                         arp_key = nla_data(nla);
 783                         swkey->ipv4.addr.src = arp_key->arp_sip;
 784                         swkey->ipv4.addr.dst = arp_key->arp_tip;
 785                         if (arp_key->arp_op & htons(0xff00))
 786                                 goto invalid;
 787                         swkey->ip.nw_proto = ntohs(arp_key->arp_op);
 788                         memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
 789                         memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
 790                         break;
 791
 792                 case TRANSITION(ODP_KEY_ATTR_ICMPV6, ODP_KEY_ATTR_ND):
 793                         key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 794                         if (swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_SOLICITATION)
 795                             && swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
 796                                 goto invalid;
 797                         nd_key = nla_data(nla);
 798                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 799                                         sizeof(swkey->ipv6.nd.target));
 800                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 801                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 802                         break;
 803
 804                 default:
 805                         goto invalid;
 806                 }
 807
 808                 prev_type = type;
 809         }
 810         if (rem)
 811                 goto invalid;
 812
 813         switch (prev_type) {
 814         case ODP_KEY_ATTR_UNSPEC:
 815                 goto invalid;
 816
 817         case ODP_KEY_ATTR_TUN_ID:
 818         case ODP_KEY_ATTR_IN_PORT:
 819                 goto invalid;
 820
 821         case ODP_KEY_ATTR_ETHERNET:
 822         case ODP_KEY_ATTR_8021Q:
 823                 goto ok;
 824
 825         case ODP_KEY_ATTR_ETHERTYPE:
 826                 if (swkey->eth.type == htons(ETH_P_IP) ||
 827                     swkey->eth.type == htons(ETH_P_ARP))
 828                         goto invalid;
 829                 goto ok;
 830
 831         case ODP_KEY_ATTR_IPV4:
 832                 if (swkey->ip.nw_proto == IPPROTO_TCP ||
 833                     swkey->ip.nw_proto == IPPROTO_UDP ||
 834                     swkey->ip.nw_proto == IPPROTO_ICMP)
 835                         goto invalid;
 836                 goto ok;
 837
 838         case ODP_KEY_ATTR_IPV6:
 839                 if (swkey->ip.nw_proto == IPPROTO_TCP ||
 840                     swkey->ip.nw_proto == IPPROTO_UDP ||
 841                     swkey->ip.nw_proto == IPPROTO_ICMPV6)
 842                         goto invalid;
 843                 goto ok;
 844
 845         case ODP_KEY_ATTR_ICMPV6:
 846                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 847                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
 848                         goto invalid;
 849                 goto ok;
 850
 851         case ODP_KEY_ATTR_TCP:
 852         case ODP_KEY_ATTR_UDP:
 853         case ODP_KEY_ATTR_ICMP:
 854         case ODP_KEY_ATTR_ARP:
 855         case ODP_KEY_ATTR_ND:
 856                 goto ok;
 857
 858         default:
 859                 WARN_ON_ONCE(1);
 860         }
 861
 862 invalid:
 863         error = -EINVAL;
 864
 865 ok:
 866         WARN_ON_ONCE(!key_len && !error);
 867         *key_lenp = key_len;
 868         return error;
 869 }
 870
 871 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
 872 {
 873         struct odp_key_ethernet *eth_key;
 874         struct nlattr *nla;
 875
 876         /* This is an imperfect sanity-check that FLOW_BUFSIZE doesn't need
 877          * to be updated, but will at least raise awareness when new ODP key
 878          * types are added. */
 879         BUILD_BUG_ON(__ODP_KEY_ATTR_MAX != 14);
 880
 881         if (swkey->eth.tun_id != cpu_to_be64(0))
 882                 NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->eth.tun_id);
 883
 884         NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->eth.in_port);
 885
 886         nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key));
 887         if (!nla)
 888                 goto nla_put_failure;
 889         eth_key = nla_data(nla);
 890         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
 891         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
 892
 893         if (swkey->eth.tci != htons(0)) {
 894                 struct odp_key_8021q q_key;
 895
 896                 q_key.q_tpid = htons(ETH_P_8021Q);
 897                 q_key.q_tci = swkey->eth.tci & ~htons(VLAN_TAG_PRESENT);
 898                 NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
 899         }
 900
 901         if (swkey->eth.type == htons(ETH_P_802_2))
 902                 return 0;
 903
 904         NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->eth.type);
 905
 906         if (swkey->eth.type == htons(ETH_P_IP)) {
 907                 struct odp_key_ipv4 *ipv4_key;
 908
 909                 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key));
 910                 if (!nla)
 911                         goto nla_put_failure;
 912                 ipv4_key = nla_data(nla);
 913                 memset(ipv4_key, 0, sizeof(struct odp_key_ipv4));
 914                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
 915                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
 916                 ipv4_key->ipv4_proto = swkey->ip.nw_proto;
 917                 ipv4_key->ipv4_tos = swkey->ip.nw_tos;
 918         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 919                 struct odp_key_ipv6 *ipv6_key;
 920
 921                 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV6, sizeof(*ipv6_key));
 922                 if (!nla)
 923                         goto nla_put_failure;
 924                 ipv6_key = nla_data(nla);
 925                 memset(ipv6_key, 0, sizeof(struct odp_key_ipv6));
 926                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
 927                                 sizeof(ipv6_key->ipv6_src));
 928                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
 929                                 sizeof(ipv6_key->ipv6_dst));
 930                 ipv6_key->ipv6_proto = swkey->ip.nw_proto;
 931                 ipv6_key->ipv6_tos = swkey->ip.nw_tos;
 932         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
 933                 struct odp_key_arp *arp_key;
 934
 935                 nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key));
 936                 if (!nla)
 937                         goto nla_put_failure;
 938                 arp_key = nla_data(nla);
 939                 memset(arp_key, 0, sizeof(struct odp_key_arp));
 940                 arp_key->arp_sip = swkey->ipv4.addr.src;
 941                 arp_key->arp_tip = swkey->ipv4.addr.dst;
 942                 arp_key->arp_op = htons(swkey->ip.nw_proto);
 943                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
 944                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
 945         }
 946
 947         if (swkey->eth.type == htons(ETH_P_IP) ||
 948             swkey->eth.type == htons(ETH_P_IPV6)) {
 949
 950                 if (swkey->ip.nw_proto == IPPROTO_TCP) {
 951                         struct odp_key_tcp *tcp_key;
 952
 953                         nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key));
 954                         if (!nla)
 955                                 goto nla_put_failure;
 956                         tcp_key = nla_data(nla);
 957                         if (swkey->eth.type == htons(ETH_P_IP)) {
 958                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
 959                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
 960                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 961                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
 962                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
 963                         }
 964                 } else if (swkey->ip.nw_proto == IPPROTO_UDP) {
 965                         struct odp_key_udp *udp_key;
 966
 967                         nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key));
 968                         if (!nla)
 969                                 goto nla_put_failure;
 970                         udp_key = nla_data(nla);
 971                         if (swkey->eth.type == htons(ETH_P_IP)) {
 972                                 udp_key->udp_src = swkey->ipv4.tp.src;
 973                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
 974                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 975                                 udp_key->udp_src = swkey->ipv6.tp.src;
 976                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
 977                         }
 978                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
 979                            swkey->ip.nw_proto == IPPROTO_ICMP) {
 980                         struct odp_key_icmp *icmp_key;
 981
 982                         nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key));
 983                         if (!nla)
 984                                 goto nla_put_failure;
 985                         icmp_key = nla_data(nla);
 986                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
 987                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
 988                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
 989                            swkey->ip.nw_proto == IPPROTO_ICMPV6) {
 990                         struct odp_key_icmpv6 *icmpv6_key;
 991
 992                         nla = nla_reserve(skb, ODP_KEY_ATTR_ICMPV6,
 993                                                 sizeof(*icmpv6_key));
 994                         if (!nla)
 995                                 goto nla_put_failure;
 996                         icmpv6_key = nla_data(nla);
 997                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
 998                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
 999
1000                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1001                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1002                                 struct odp_key_nd *nd_key;
1003
1004                                 nla = nla_reserve(skb, ODP_KEY_ATTR_ND, sizeof(*nd_key));
1005                                 if (!nla)
1006                                         goto nla_put_failure;
1007                                 nd_key = nla_data(nla);
1008                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1009                                                         sizeof(nd_key->nd_target));
1010                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1011                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1012                         }
1013                 }
1014         }
1015
1016         return 0;
1017
1018 nla_put_failure:
1019         return -EMSGSIZE;
1020 }
1021
1022 /* Initializes the flow module.
1023  * Returns zero if successful or a negative error code. */
1024 int flow_init(void)
1025 {
1026         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1027                                         0, NULL);
1028         if (flow_cache == NULL)
1029                 return -ENOMEM;
1030
1031         get_random_bytes(&hash_seed, sizeof(hash_seed));
1032
1033         return 0;
1034 }
1035
1036 /* Uninitializes the flow module. */
1037 void flow_exit(void)
1038 {
1039         kmem_cache_destroy(flow_cache);
1040 }