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_OFFSET 13
  64 #define TCP_FLAG_MASK 0x3f
  65
  66 void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
  67 {
  68         struct sw_flow_stats *stats = &flow->stats[smp_processor_id()];
  69         u8 tcp_flags = 0;
  70
  71         if ((flow->key.eth.type == htons(ETH_P_IP) ||
  72              flow->key.eth.type == htons(ETH_P_IPV6)) &&
  73             flow->key.ip.proto == IPPROTO_TCP &&
  74             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
  75                 u8 *tcp = (u8 *)tcp_hdr(skb);
  76                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
  77         }
  78
  79         spin_lock(&stats->lock);
  80         stats->used = jiffies;
  81         stats->packet_count++;
  82         stats->byte_count += skb->len;
  83         stats->tcp_flags |= tcp_flags;
  84         spin_unlock(&stats->lock);
  85 }
  86
  87 void ovs_flow_stats_get(struct sw_flow *flow, struct sw_flow_stats *res)
  88 {
  89         int cpu, cur_cpu;
  90
  91         memset(res, 0, sizeof(*res));
  92
  93         cur_cpu = get_cpu();
  94         for_each_possible_cpu(cpu) {
  95                 struct sw_flow_stats *stats = &flow->stats[cpu];
  96
  97                 if (cpu == cur_cpu)
  98                         local_bh_disable();
  99
 100                 spin_lock(&stats->lock);
 101                 if (time_after(stats->used, res->used))
 102                         res->used = stats->used;
 103                 res->packet_count += stats->packet_count;
 104                 res->byte_count += stats->byte_count;
 105                 res->tcp_flags |= stats->tcp_flags;
 106                 spin_unlock(&stats->lock);
 107
 108                 if (cpu == cur_cpu)
 109                         local_bh_enable();
 110
 111         }
 112         put_cpu();
 113 }
 114
 115 void ovs_flow_stats_clear(struct sw_flow *flow)
 116 {
 117         int cpu, cur_cpu;
 118
 119         cur_cpu = get_cpu();
 120         for_each_possible_cpu(cpu) {
 121                 struct sw_flow_stats *stats = &flow->stats[cpu];
 122
 123                 if (cpu == cur_cpu)
 124                         local_bh_disable();
 125
 126                 spin_lock(&stats->lock);
 127                 stats->used = 0;
 128                 stats->packet_count = 0;
 129                 stats->byte_count = 0;
 130                 stats->tcp_flags = 0;
 131                 spin_unlock(&stats->lock);
 132
 133                 if (cpu == cur_cpu)
 134                         local_bh_enable();
 135         }
 136         put_cpu();
 137 }
 138
 139 static int check_header(struct sk_buff *skb, int len)
 140 {
 141         if (unlikely(skb->len < len))
 142                 return -EINVAL;
 143         if (unlikely(!pskb_may_pull(skb, len)))
 144                 return -ENOMEM;
 145         return 0;
 146 }
 147
 148 static bool arphdr_ok(struct sk_buff *skb)
 149 {
 150         return pskb_may_pull(skb, skb_network_offset(skb) +
 151                                   sizeof(struct arp_eth_header));
 152 }
 153
 154 static int check_iphdr(struct sk_buff *skb)
 155 {
 156         unsigned int nh_ofs = skb_network_offset(skb);
 157         unsigned int ip_len;
 158         int err;
 159
 160         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 161         if (unlikely(err))
 162                 return err;
 163
 164         ip_len = ip_hdrlen(skb);
 165         if (unlikely(ip_len < sizeof(struct iphdr) ||
 166                      skb->len < nh_ofs + ip_len))
 167                 return -EINVAL;
 168
 169         skb_set_transport_header(skb, nh_ofs + ip_len);
 170         return 0;
 171 }
 172
 173 static bool tcphdr_ok(struct sk_buff *skb)
 174 {
 175         int th_ofs = skb_transport_offset(skb);
 176         int tcp_len;
 177
 178         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 179                 return false;
 180
 181         tcp_len = tcp_hdrlen(skb);
 182         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 183                      skb->len < th_ofs + tcp_len))
 184                 return false;
 185
 186         return true;
 187 }
 188
 189 static bool udphdr_ok(struct sk_buff *skb)
 190 {
 191         return pskb_may_pull(skb, skb_transport_offset(skb) +
 192                                   sizeof(struct udphdr));
 193 }
 194
 195 static bool sctphdr_ok(struct sk_buff *skb)
 196 {
 197         return pskb_may_pull(skb, skb_transport_offset(skb) +
 198                                   sizeof(struct sctphdr));
 199 }
 200
 201 static bool icmphdr_ok(struct sk_buff *skb)
 202 {
 203         return pskb_may_pull(skb, skb_transport_offset(skb) +
 204                                   sizeof(struct icmphdr));
 205 }
 206
 207 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 208 {
 209         unsigned int nh_ofs = skb_network_offset(skb);
 210         unsigned int nh_len;
 211         int payload_ofs;
 212         struct ipv6hdr *nh;
 213         uint8_t nexthdr;
 214         __be16 frag_off;
 215         int err;
 216
 217         err = check_header(skb, nh_ofs + sizeof(*nh));
 218         if (unlikely(err))
 219                 return err;
 220
 221         nh = ipv6_hdr(skb);
 222         nexthdr = nh->nexthdr;
 223         payload_ofs = (u8 *)(nh + 1) - skb->data;
 224
 225         key->ip.proto = NEXTHDR_NONE;
 226         key->ip.tos = ipv6_get_dsfield(nh);
 227         key->ip.ttl = nh->hop_limit;
 228         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 229         key->ipv6.addr.src = nh->saddr;
 230         key->ipv6.addr.dst = nh->daddr;
 231
 232         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 233         if (unlikely(payload_ofs < 0))
 234                 return -EINVAL;
 235
 236         if (frag_off) {
 237                 if (frag_off & htons(~0x7))
 238                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 239                 else
 240                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 241         }
 242
 243         nh_len = payload_ofs - nh_ofs;
 244         skb_set_transport_header(skb, nh_ofs + nh_len);
 245         key->ip.proto = nexthdr;
 246         return nh_len;
 247 }
 248
 249 static bool icmp6hdr_ok(struct sk_buff *skb)
 250 {
 251         return pskb_may_pull(skb, skb_transport_offset(skb) +
 252                                   sizeof(struct icmp6hdr));
 253 }
 254
 255 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 256 {
 257         struct qtag_prefix {
 258                 __be16 eth_type; /* ETH_P_8021Q */
 259                 __be16 tci;
 260         };
 261         struct qtag_prefix *qp;
 262
 263         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 264                 return 0;
 265
 266         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 267                                          sizeof(__be16))))
 268                 return -ENOMEM;
 269
 270         qp = (struct qtag_prefix *) skb->data;
 271         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 272         __skb_pull(skb, sizeof(struct qtag_prefix));
 273
 274         return 0;
 275 }
 276
 277 static __be16 parse_ethertype(struct sk_buff *skb)
 278 {
 279         struct llc_snap_hdr {
 280                 u8  dsap;  /* Always 0xAA */
 281                 u8  ssap;  /* Always 0xAA */
 282                 u8  ctrl;
 283                 u8  oui[3];
 284                 __be16 ethertype;
 285         };
 286         struct llc_snap_hdr *llc;
 287         __be16 proto;
 288
 289         proto = *(__be16 *) skb->data;
 290         __skb_pull(skb, sizeof(__be16));
 291
 292         if (ntohs(proto) >= ETH_P_802_3_MIN)
 293                 return proto;
 294
 295         if (skb->len < sizeof(struct llc_snap_hdr))
 296                 return htons(ETH_P_802_2);
 297
 298         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 299                 return htons(0);
 300
 301         llc = (struct llc_snap_hdr *) skb->data;
 302         if (llc->dsap != LLC_SAP_SNAP ||
 303             llc->ssap != LLC_SAP_SNAP ||
 304             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 305                 return htons(ETH_P_802_2);
 306
 307         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 308
 309         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
 310                 return llc->ethertype;
 311
 312         return htons(ETH_P_802_2);
 313 }
 314
 315 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 316                         int nh_len)
 317 {
 318         struct icmp6hdr *icmp = icmp6_hdr(skb);
 319
 320         /* The ICMPv6 type and code fields use the 16-bit transport port
 321          * fields, so we need to store them in 16-bit network byte order.
 322          */
 323         key->ipv6.tp.src = htons(icmp->icmp6_type);
 324         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 325
 326         if (icmp->icmp6_code == 0 &&
 327             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 328              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 329                 int icmp_len = skb->len - skb_transport_offset(skb);
 330                 struct nd_msg *nd;
 331                 int offset;
 332
 333                 /* In order to process neighbor discovery options, we need the
 334                  * entire packet.
 335                  */
 336                 if (unlikely(icmp_len < sizeof(*nd)))
 337                         return 0;
 338
 339                 if (unlikely(skb_linearize(skb)))
 340                         return -ENOMEM;
 341
 342                 nd = (struct nd_msg *)skb_transport_header(skb);
 343                 key->ipv6.nd.target = nd->target;
 344
 345                 icmp_len -= sizeof(*nd);
 346                 offset = 0;
 347                 while (icmp_len >= 8) {
 348                         struct nd_opt_hdr *nd_opt =
 349                                  (struct nd_opt_hdr *)(nd->opt + offset);
 350                         int opt_len = nd_opt->nd_opt_len * 8;
 351
 352                         if (unlikely(!opt_len || opt_len > icmp_len))
 353                                 return 0;
 354
 355                         /* Store the link layer address if the appropriate
 356                          * option is provided.  It is considered an error if
 357                          * the same link layer option is specified twice.
 358                          */
 359                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 360                             && opt_len == 8) {
 361                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 362                                         goto invalid;
 363                                 memcpy(key->ipv6.nd.sll,
 364                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 365                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 366                                    && opt_len == 8) {
 367                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 368                                         goto invalid;
 369                                 memcpy(key->ipv6.nd.tll,
 370                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 371                         }
 372
 373                         icmp_len -= opt_len;
 374                         offset += opt_len;
 375                 }
 376         }
 377
 378         return 0;
 379
 380 invalid:
 381         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 382         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 383         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 384
 385         return 0;
 386 }
 387
 388 /**
 389  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 390  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 391  * Ethernet header
 392  * @in_port: port number on which @skb was received.
 393  * @key: output flow key
 394  *
 395  * The caller must ensure that skb->len >= ETH_HLEN.
 396  *
 397  * Returns 0 if successful, otherwise a negative errno value.
 398  *
 399  * Initializes @skb header pointers as follows:
 400  *
 401  *    - skb->mac_header: the Ethernet header.
 402  *
 403  *    - skb->network_header: just past the Ethernet header, or just past the
 404  *      VLAN header, to the first byte of the Ethernet payload.
 405  *
 406  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 407  *      on output, then just past the IP header, if one is present and
 408  *      of a correct length, otherwise the same as skb->network_header.
 409  *      For other key->eth.type values it is left untouched.
 410  */
 411 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
 412 {
 413         int error;
 414         struct ethhdr *eth;
 415
 416         memset(key, 0, sizeof(*key));
 417
 418         key->phy.priority = skb->priority;
 419         if (OVS_CB(skb)->tun_key)
 420                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 421         key->phy.in_port = in_port;
 422         key->phy.skb_mark = skb->mark;
 423
 424         skb_reset_mac_header(skb);
 425
 426         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 427          * header in the linear data area.
 428          */
 429         eth = eth_hdr(skb);
 430         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 431         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 432
 433         __skb_pull(skb, 2 * ETH_ALEN);
 434         /* We are going to push all headers that we pull, so no need to
 435          * update skb->csum here. */
 436
 437         if (vlan_tx_tag_present(skb))
 438                 key->eth.tci = htons(vlan_get_tci(skb));
 439         else if (eth->h_proto == htons(ETH_P_8021Q))
 440                 if (unlikely(parse_vlan(skb, key)))
 441                         return -ENOMEM;
 442
 443         key->eth.type = parse_ethertype(skb);
 444         if (unlikely(key->eth.type == htons(0)))
 445                 return -ENOMEM;
 446
 447         skb_reset_network_header(skb);
 448         __skb_push(skb, skb->data - skb_mac_header(skb));
 449
 450         /* Network layer. */
 451         if (key->eth.type == htons(ETH_P_IP)) {
 452                 struct iphdr *nh;
 453                 __be16 offset;
 454
 455                 error = check_iphdr(skb);
 456                 if (unlikely(error)) {
 457                         if (error == -EINVAL) {
 458                                 skb->transport_header = skb->network_header;
 459                                 error = 0;
 460                         }
 461                         return error;
 462                 }
 463
 464                 nh = ip_hdr(skb);
 465                 key->ipv4.addr.src = nh->saddr;
 466                 key->ipv4.addr.dst = nh->daddr;
 467
 468                 key->ip.proto = nh->protocol;
 469                 key->ip.tos = nh->tos;
 470                 key->ip.ttl = nh->ttl;
 471
 472                 offset = nh->frag_off & htons(IP_OFFSET);
 473                 if (offset) {
 474                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 475                         return 0;
 476                 }
 477                 if (nh->frag_off & htons(IP_MF) ||
 478                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 479                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 480
 481                 /* Transport layer. */
 482                 if (key->ip.proto == IPPROTO_TCP) {
 483                         if (tcphdr_ok(skb)) {
 484                                 struct tcphdr *tcp = tcp_hdr(skb);
 485                                 key->ipv4.tp.src = tcp->source;
 486                                 key->ipv4.tp.dst = tcp->dest;
 487                         }
 488                 } else if (key->ip.proto == IPPROTO_UDP) {
 489                         if (udphdr_ok(skb)) {
 490                                 struct udphdr *udp = udp_hdr(skb);
 491                                 key->ipv4.tp.src = udp->source;
 492                                 key->ipv4.tp.dst = udp->dest;
 493                         }
 494                 } else if (key->ip.proto == IPPROTO_SCTP) {
 495                         if (sctphdr_ok(skb)) {
 496                                 struct sctphdr *sctp = sctp_hdr(skb);
 497                                 key->ipv4.tp.src = sctp->source;
 498                                 key->ipv4.tp.dst = sctp->dest;
 499                         }
 500                 } else if (key->ip.proto == IPPROTO_ICMP) {
 501                         if (icmphdr_ok(skb)) {
 502                                 struct icmphdr *icmp = icmp_hdr(skb);
 503                                 /* The ICMP type and code fields use the 16-bit
 504                                  * transport port fields, so we need to store
 505                                  * them in 16-bit network byte order. */
 506                                 key->ipv4.tp.src = htons(icmp->type);
 507                                 key->ipv4.tp.dst = htons(icmp->code);
 508                         }
 509                 }
 510
 511         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 512                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 513                 struct arp_eth_header *arp;
 514
 515                 arp = (struct arp_eth_header *)skb_network_header(skb);
 516
 517                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 518                                 && arp->ar_pro == htons(ETH_P_IP)
 519                                 && arp->ar_hln == ETH_ALEN
 520                                 && arp->ar_pln == 4) {
 521
 522                         /* We only match on the lower 8 bits of the opcode. */
 523                         if (ntohs(arp->ar_op) <= 0xff)
 524                                 key->ip.proto = ntohs(arp->ar_op);
 525                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 526                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 527                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 528                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 529                 }
 530         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 531                 int nh_len;             /* IPv6 Header + Extensions */
 532
 533                 nh_len = parse_ipv6hdr(skb, key);
 534                 if (unlikely(nh_len < 0)) {
 535                         if (nh_len == -EINVAL) {
 536                                 skb->transport_header = skb->network_header;
 537                                 error = 0;
 538                         } else {
 539                                 error = nh_len;
 540                         }
 541                         return error;
 542                 }
 543
 544                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 545                         return 0;
 546                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 547                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 548
 549                 /* Transport layer. */
 550                 if (key->ip.proto == NEXTHDR_TCP) {
 551                         if (tcphdr_ok(skb)) {
 552                                 struct tcphdr *tcp = tcp_hdr(skb);
 553                                 key->ipv6.tp.src = tcp->source;
 554                                 key->ipv6.tp.dst = tcp->dest;
 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 }