datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2013 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #include "flow.h"
  20 #include "datapath.h"
  21 #include <linux/uaccess.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <net/llc_pdu.h>
  27 #include <linux/kernel.h>
  28 #include <linux/jhash.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/llc.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/rcupdate.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/sctp.h>
  38 #include <linux/smp.h>
  39 #include <linux/tcp.h>
  40 #include <linux/udp.h>
  41 #include <linux/icmp.h>
  42 #include <linux/icmpv6.h>
  43 #include <linux/rculist.h>
  44 #include <net/ip.h>
  45 #include <net/ipv6.h>
  46 #include <net/ndisc.h>
  47
  48 #include "vlan.h"
  49
  50 u64 ovs_flow_used_time(unsigned long flow_jiffies)
  51 {
  52         struct timespec cur_ts;
  53         u64 cur_ms, idle_ms;
  54
  55         ktime_get_ts(&cur_ts);
  56         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
  57         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
  58                  cur_ts.tv_nsec / NSEC_PER_MSEC;
  59
  60         return cur_ms - idle_ms;
  61 }
  62
  63 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
  64
  65 void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
  66 {
  67         struct sw_flow_stats *stats = &flow->stats[smp_processor_id()];
  68         __be16 tcp_flags = 0;
  69
  70         if ((flow->key.eth.type == htons(ETH_P_IP) ||
  71              flow->key.eth.type == htons(ETH_P_IPV6)) &&
  72             flow->key.ip.proto == IPPROTO_TCP &&
  73             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
  74                 tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
  75         }
  76
  77         spin_lock(&stats->lock);
  78         stats->used = jiffies;
  79         stats->packet_count++;
  80         stats->byte_count += skb->len;
  81         stats->tcp_flags |= tcp_flags;
  82         spin_unlock(&stats->lock);
  83 }
  84
  85 void ovs_flow_stats_get(struct sw_flow *flow, struct sw_flow_stats *res)
  86 {
  87         int cpu, cur_cpu;
  88
  89         memset(res, 0, sizeof(*res));
  90
  91         cur_cpu = get_cpu();
  92         for_each_possible_cpu(cpu) {
  93                 struct sw_flow_stats *stats = &flow->stats[cpu];
  94
  95                 if (cpu == cur_cpu)
  96                         local_bh_disable();
  97
  98                 spin_lock(&stats->lock);
  99                 if (time_after(stats->used, res->used))
 100                         res->used = stats->used;
 101                 res->packet_count += stats->packet_count;
 102                 res->byte_count += stats->byte_count;
 103                 res->tcp_flags |= stats->tcp_flags;
 104                 spin_unlock(&stats->lock);
 105
 106                 if (cpu == cur_cpu)
 107                         local_bh_enable();
 108
 109         }
 110         put_cpu();
 111 }
 112
 113 void ovs_flow_stats_clear(struct sw_flow *flow)
 114 {
 115         int cpu, cur_cpu;
 116
 117         cur_cpu = get_cpu();
 118         for_each_possible_cpu(cpu) {
 119                 struct sw_flow_stats *stats = &flow->stats[cpu];
 120
 121                 if (cpu == cur_cpu)
 122                         local_bh_disable();
 123
 124                 spin_lock(&stats->lock);
 125                 stats->used = 0;
 126                 stats->packet_count = 0;
 127                 stats->byte_count = 0;
 128                 stats->tcp_flags = 0;
 129                 spin_unlock(&stats->lock);
 130
 131                 if (cpu == cur_cpu)
 132                         local_bh_enable();
 133         }
 134         put_cpu();
 135 }
 136
 137 static int check_header(struct sk_buff *skb, int len)
 138 {
 139         if (unlikely(skb->len < len))
 140                 return -EINVAL;
 141         if (unlikely(!pskb_may_pull(skb, len)))
 142                 return -ENOMEM;
 143         return 0;
 144 }
 145
 146 static bool arphdr_ok(struct sk_buff *skb)
 147 {
 148         return pskb_may_pull(skb, skb_network_offset(skb) +
 149                                   sizeof(struct arp_eth_header));
 150 }
 151
 152 static int check_iphdr(struct sk_buff *skb)
 153 {
 154         unsigned int nh_ofs = skb_network_offset(skb);
 155         unsigned int ip_len;
 156         int err;
 157
 158         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 159         if (unlikely(err))
 160                 return err;
 161
 162         ip_len = ip_hdrlen(skb);
 163         if (unlikely(ip_len < sizeof(struct iphdr) ||
 164                      skb->len < nh_ofs + ip_len))
 165                 return -EINVAL;
 166
 167         skb_set_transport_header(skb, nh_ofs + ip_len);
 168         return 0;
 169 }
 170
 171 static bool tcphdr_ok(struct sk_buff *skb)
 172 {
 173         int th_ofs = skb_transport_offset(skb);
 174         int tcp_len;
 175
 176         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 177                 return false;
 178
 179         tcp_len = tcp_hdrlen(skb);
 180         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 181                      skb->len < th_ofs + tcp_len))
 182                 return false;
 183
 184         return true;
 185 }
 186
 187 static bool udphdr_ok(struct sk_buff *skb)
 188 {
 189         return pskb_may_pull(skb, skb_transport_offset(skb) +
 190                                   sizeof(struct udphdr));
 191 }
 192
 193 static bool sctphdr_ok(struct sk_buff *skb)
 194 {
 195         return pskb_may_pull(skb, skb_transport_offset(skb) +
 196                                   sizeof(struct sctphdr));
 197 }
 198
 199 static bool icmphdr_ok(struct sk_buff *skb)
 200 {
 201         return pskb_may_pull(skb, skb_transport_offset(skb) +
 202                                   sizeof(struct icmphdr));
 203 }
 204
 205 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 206 {
 207         unsigned int nh_ofs = skb_network_offset(skb);
 208         unsigned int nh_len;
 209         int payload_ofs;
 210         struct ipv6hdr *nh;
 211         uint8_t nexthdr;
 212         __be16 frag_off;
 213         int err;
 214
 215         err = check_header(skb, nh_ofs + sizeof(*nh));
 216         if (unlikely(err))
 217                 return err;
 218
 219         nh = ipv6_hdr(skb);
 220         nexthdr = nh->nexthdr;
 221         payload_ofs = (u8 *)(nh + 1) - skb->data;
 222
 223         key->ip.proto = NEXTHDR_NONE;
 224         key->ip.tos = ipv6_get_dsfield(nh);
 225         key->ip.ttl = nh->hop_limit;
 226         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 227         key->ipv6.addr.src = nh->saddr;
 228         key->ipv6.addr.dst = nh->daddr;
 229
 230         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 231         if (unlikely(payload_ofs < 0))
 232                 return -EINVAL;
 233
 234         if (frag_off) {
 235                 if (frag_off & htons(~0x7))
 236                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 237                 else
 238                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 239         }
 240
 241         nh_len = payload_ofs - nh_ofs;
 242         skb_set_transport_header(skb, nh_ofs + nh_len);
 243         key->ip.proto = nexthdr;
 244         return nh_len;
 245 }
 246
 247 static bool icmp6hdr_ok(struct sk_buff *skb)
 248 {
 249         return pskb_may_pull(skb, skb_transport_offset(skb) +
 250                                   sizeof(struct icmp6hdr));
 251 }
 252
 253 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 254 {
 255         struct qtag_prefix {
 256                 __be16 eth_type; /* ETH_P_8021Q */
 257                 __be16 tci;
 258         };
 259         struct qtag_prefix *qp;
 260
 261         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 262                 return 0;
 263
 264         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 265                                          sizeof(__be16))))
 266                 return -ENOMEM;
 267
 268         qp = (struct qtag_prefix *) skb->data;
 269         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 270         __skb_pull(skb, sizeof(struct qtag_prefix));
 271
 272         return 0;
 273 }
 274
 275 static __be16 parse_ethertype(struct sk_buff *skb)
 276 {
 277         struct llc_snap_hdr {
 278                 u8  dsap;  /* Always 0xAA */
 279                 u8  ssap;  /* Always 0xAA */
 280                 u8  ctrl;
 281                 u8  oui[3];
 282                 __be16 ethertype;
 283         };
 284         struct llc_snap_hdr *llc;
 285         __be16 proto;
 286
 287         proto = *(__be16 *) skb->data;
 288         __skb_pull(skb, sizeof(__be16));
 289
 290         if (ntohs(proto) >= ETH_P_802_3_MIN)
 291                 return proto;
 292
 293         if (skb->len < sizeof(struct llc_snap_hdr))
 294                 return htons(ETH_P_802_2);
 295
 296         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 297                 return htons(0);
 298
 299         llc = (struct llc_snap_hdr *) skb->data;
 300         if (llc->dsap != LLC_SAP_SNAP ||
 301             llc->ssap != LLC_SAP_SNAP ||
 302             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 303                 return htons(ETH_P_802_2);
 304
 305         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 306
 307         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
 308                 return llc->ethertype;
 309
 310         return htons(ETH_P_802_2);
 311 }
 312
 313 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 314                         int nh_len)
 315 {
 316         struct icmp6hdr *icmp = icmp6_hdr(skb);
 317
 318         /* The ICMPv6 type and code fields use the 16-bit transport port
 319          * fields, so we need to store them in 16-bit network byte order.
 320          */
 321         key->ipv6.tp.src = htons(icmp->icmp6_type);
 322         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 323
 324         if (icmp->icmp6_code == 0 &&
 325             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 326              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 327                 int icmp_len = skb->len - skb_transport_offset(skb);
 328                 struct nd_msg *nd;
 329                 int offset;
 330
 331                 /* In order to process neighbor discovery options, we need the
 332                  * entire packet.
 333                  */
 334                 if (unlikely(icmp_len < sizeof(*nd)))
 335                         return 0;
 336
 337                 if (unlikely(skb_linearize(skb)))
 338                         return -ENOMEM;
 339
 340                 nd = (struct nd_msg *)skb_transport_header(skb);
 341                 key->ipv6.nd.target = nd->target;
 342
 343                 icmp_len -= sizeof(*nd);
 344                 offset = 0;
 345                 while (icmp_len >= 8) {
 346                         struct nd_opt_hdr *nd_opt =
 347                                  (struct nd_opt_hdr *)(nd->opt + offset);
 348                         int opt_len = nd_opt->nd_opt_len * 8;
 349
 350                         if (unlikely(!opt_len || opt_len > icmp_len))
 351                                 return 0;
 352
 353                         /* Store the link layer address if the appropriate
 354                          * option is provided.  It is considered an error if
 355                          * the same link layer option is specified twice.
 356                          */
 357                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 358                             && opt_len == 8) {
 359                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 360                                         goto invalid;
 361                                 memcpy(key->ipv6.nd.sll,
 362                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 363                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 364                                    && opt_len == 8) {
 365                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 366                                         goto invalid;
 367                                 memcpy(key->ipv6.nd.tll,
 368                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 369                         }
 370
 371                         icmp_len -= opt_len;
 372                         offset += opt_len;
 373                 }
 374         }
 375
 376         return 0;
 377
 378 invalid:
 379         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 380         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 381         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 382
 383         return 0;
 384 }
 385
 386 /**
 387  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 388  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 389  * Ethernet header
 390  * @in_port: port number on which @skb was received.
 391  * @key: output flow key
 392  *
 393  * The caller must ensure that skb->len >= ETH_HLEN.
 394  *
 395  * Returns 0 if successful, otherwise a negative errno value.
 396  *
 397  * Initializes @skb header pointers as follows:
 398  *
 399  *    - skb->mac_header: the Ethernet header.
 400  *
 401  *    - skb->network_header: just past the Ethernet header, or just past the
 402  *      VLAN header, to the first byte of the Ethernet payload.
 403  *
 404  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 405  *      on output, then just past the IP header, if one is present and
 406  *      of a correct length, otherwise the same as skb->network_header.
 407  *      For other key->eth.type values it is left untouched.
 408  */
 409 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
 410 {
 411         int error;
 412         struct ethhdr *eth;
 413
 414         memset(key, 0, sizeof(*key));
 415
 416         key->phy.priority = skb->priority;
 417         if (OVS_CB(skb)->tun_key)
 418                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 419         key->phy.in_port = in_port;
 420         key->phy.skb_mark = skb->mark;
 421
 422         skb_reset_mac_header(skb);
 423
 424         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 425          * header in the linear data area.
 426          */
 427         eth = eth_hdr(skb);
 428         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 429         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 430
 431         __skb_pull(skb, 2 * ETH_ALEN);
 432         /* We are going to push all headers that we pull, so no need to
 433          * update skb->csum here. */
 434
 435         if (vlan_tx_tag_present(skb))
 436                 key->eth.tci = htons(vlan_get_tci(skb));
 437         else if (eth->h_proto == htons(ETH_P_8021Q))
 438                 if (unlikely(parse_vlan(skb, key)))
 439                         return -ENOMEM;
 440
 441         key->eth.type = parse_ethertype(skb);
 442         if (unlikely(key->eth.type == htons(0)))
 443                 return -ENOMEM;
 444
 445         skb_reset_network_header(skb);
 446         __skb_push(skb, skb->data - skb_mac_header(skb));
 447
 448         /* Network layer. */
 449         if (key->eth.type == htons(ETH_P_IP)) {
 450                 struct iphdr *nh;
 451                 __be16 offset;
 452
 453                 error = check_iphdr(skb);
 454                 if (unlikely(error)) {
 455                         if (error == -EINVAL) {
 456                                 skb->transport_header = skb->network_header;
 457                                 error = 0;
 458                         }
 459                         return error;
 460                 }
 461
 462                 nh = ip_hdr(skb);
 463                 key->ipv4.addr.src = nh->saddr;
 464                 key->ipv4.addr.dst = nh->daddr;
 465
 466                 key->ip.proto = nh->protocol;
 467                 key->ip.tos = nh->tos;
 468                 key->ip.ttl = nh->ttl;
 469
 470                 offset = nh->frag_off & htons(IP_OFFSET);
 471                 if (offset) {
 472                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 473                         return 0;
 474                 }
 475                 if (nh->frag_off & htons(IP_MF) ||
 476                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 477                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 478
 479                 /* Transport layer. */
 480                 if (key->ip.proto == IPPROTO_TCP) {
 481                         if (tcphdr_ok(skb)) {
 482                                 struct tcphdr *tcp = tcp_hdr(skb);
 483                                 key->ipv4.tp.src = tcp->source;
 484                                 key->ipv4.tp.dst = tcp->dest;
 485                                 key->ipv4.tp.flags = TCP_FLAGS_BE16(tcp);
 486                         }
 487                 } else if (key->ip.proto == IPPROTO_UDP) {
 488                         if (udphdr_ok(skb)) {
 489                                 struct udphdr *udp = udp_hdr(skb);
 490                                 key->ipv4.tp.src = udp->source;
 491                                 key->ipv4.tp.dst = udp->dest;
 492                         }
 493                 } else if (key->ip.proto == IPPROTO_SCTP) {
 494                         if (sctphdr_ok(skb)) {
 495                                 struct sctphdr *sctp = sctp_hdr(skb);
 496                                 key->ipv4.tp.src = sctp->source;
 497                                 key->ipv4.tp.dst = sctp->dest;
 498                         }
 499                 } else if (key->ip.proto == IPPROTO_ICMP) {
 500                         if (icmphdr_ok(skb)) {
 501                                 struct icmphdr *icmp = icmp_hdr(skb);
 502                                 /* The ICMP type and code fields use the 16-bit
 503                                  * transport port fields, so we need to store
 504                                  * them in 16-bit network byte order. */
 505                                 key->ipv4.tp.src = htons(icmp->type);
 506                                 key->ipv4.tp.dst = htons(icmp->code);
 507                         }
 508                 }
 509
 510         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 511                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 512                 struct arp_eth_header *arp;
 513
 514                 arp = (struct arp_eth_header *)skb_network_header(skb);
 515
 516                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 517                                 && arp->ar_pro == htons(ETH_P_IP)
 518                                 && arp->ar_hln == ETH_ALEN
 519                                 && arp->ar_pln == 4) {
 520
 521                         /* We only match on the lower 8 bits of the opcode. */
 522                         if (ntohs(arp->ar_op) <= 0xff)
 523                                 key->ip.proto = ntohs(arp->ar_op);
 524                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 525                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 526                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 527                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 528                 }
 529         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 530                 int nh_len;             /* IPv6 Header + Extensions */
 531
 532                 nh_len = parse_ipv6hdr(skb, key);
 533                 if (unlikely(nh_len < 0)) {
 534                         if (nh_len == -EINVAL) {
 535                                 skb->transport_header = skb->network_header;
 536                                 error = 0;
 537                         } else {
 538                                 error = nh_len;
 539                         }
 540                         return error;
 541                 }
 542
 543                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 544                         return 0;
 545                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 546                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 547
 548                 /* Transport layer. */
 549                 if (key->ip.proto == NEXTHDR_TCP) {
 550                         if (tcphdr_ok(skb)) {
 551                                 struct tcphdr *tcp = tcp_hdr(skb);
 552                                 key->ipv6.tp.src = tcp->source;
 553                                 key->ipv6.tp.dst = tcp->dest;
 554                                 key->ipv6.tp.flags = TCP_FLAGS_BE16(tcp);
 555                         }
 556                 } else if (key->ip.proto == NEXTHDR_UDP) {
 557                         if (udphdr_ok(skb)) {
 558                                 struct udphdr *udp = udp_hdr(skb);
 559                                 key->ipv6.tp.src = udp->source;
 560                                 key->ipv6.tp.dst = udp->dest;
 561                         }
 562                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 563                         if (sctphdr_ok(skb)) {
 564                                 struct sctphdr *sctp = sctp_hdr(skb);
 565                                 key->ipv6.tp.src = sctp->source;
 566                                 key->ipv6.tp.dst = sctp->dest;
 567                         }
 568                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 569                         if (icmp6hdr_ok(skb)) {
 570                                 error = parse_icmpv6(skb, key, nh_len);
 571                                 if (error)
 572                                         return error;
 573                         }
 574                 }
 575         }
 576
 577         return 0;
 578 }