datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2014 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 flow_stats *stats;
  68         __be16 tcp_flags = 0;
  69
  70         if (!flow->stats.is_percpu)
  71                 stats = flow->stats.stat;
  72         else
  73                 stats = this_cpu_ptr(flow->stats.cpu_stats);
  74
  75         if ((flow->key.eth.type == htons(ETH_P_IP) ||
  76              flow->key.eth.type == htons(ETH_P_IPV6)) &&
  77             flow->key.ip.proto == IPPROTO_TCP &&
  78             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
  79                 tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
  80         }
  81
  82         spin_lock(&stats->lock);
  83         stats->used = jiffies;
  84         stats->packet_count++;
  85         stats->byte_count += skb->len;
  86         stats->tcp_flags |= tcp_flags;
  87         spin_unlock(&stats->lock);
  88 }
  89
  90 static void stats_read(struct flow_stats *stats,
  91                        struct ovs_flow_stats *ovs_stats,
  92                        unsigned long *used, __be16 *tcp_flags)
  93 {
  94         spin_lock(&stats->lock);
  95         if (!*used || time_after(stats->used, *used))
  96                 *used = stats->used;
  97         *tcp_flags |= stats->tcp_flags;
  98         ovs_stats->n_packets += stats->packet_count;
  99         ovs_stats->n_bytes += stats->byte_count;
 100         spin_unlock(&stats->lock);
 101 }
 102
 103 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
 104                         unsigned long *used, __be16 *tcp_flags)
 105 {
 106         int cpu;
 107
 108         *used = 0;
 109         *tcp_flags = 0;
 110         memset(ovs_stats, 0, sizeof(*ovs_stats));
 111
 112         local_bh_disable();
 113         if (!flow->stats.is_percpu) {
 114                 stats_read(flow->stats.stat, true, ovs_stats, used, tcp_flags);
 115         } else {
 116                 for_each_possible_cpu(cpu) {
 117                         struct flow_stats *stats;
 118
 119                         stats = per_cpu_ptr(flow->stats.cpu_stats, cpu);
 120                         stats_read(stats, ovs_stats, used, tcp_flags);
 121                 }
 122         }
 123         local_bh_enable();
 124 }
 125
 126 static void stats_reset(struct flow_stats *stats)
 127 {
 128         spin_lock(&stats->lock);
 129         stats->used = 0;
 130         stats->packet_count = 0;
 131         stats->byte_count = 0;
 132         stats->tcp_flags = 0;
 133         spin_unlock(&stats->lock);
 134 }
 135
 136 void ovs_flow_stats_clear(struct sw_flow *flow)
 137 {
 138         int cpu;
 139
 140         local_bh_disable();
 141         if (!flow->stats.is_percpu) {
 142                 stats_reset(flow->stats.stat, true);
 143         } else {
 144                 for_each_possible_cpu(cpu)
 145                         stats_reset(per_cpu_ptr(flow->stats.cpu_stats, cpu));
 146         }
 147         local_bh_enable();
 148 }
 149
 150 static int check_header(struct sk_buff *skb, int len)
 151 {
 152         if (unlikely(skb->len < len))
 153                 return -EINVAL;
 154         if (unlikely(!pskb_may_pull(skb, len)))
 155                 return -ENOMEM;
 156         return 0;
 157 }
 158
 159 static bool arphdr_ok(struct sk_buff *skb)
 160 {
 161         return pskb_may_pull(skb, skb_network_offset(skb) +
 162                                   sizeof(struct arp_eth_header));
 163 }
 164
 165 static int check_iphdr(struct sk_buff *skb)
 166 {
 167         unsigned int nh_ofs = skb_network_offset(skb);
 168         unsigned int ip_len;
 169         int err;
 170
 171         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 172         if (unlikely(err))
 173                 return err;
 174
 175         ip_len = ip_hdrlen(skb);
 176         if (unlikely(ip_len < sizeof(struct iphdr) ||
 177                      skb->len < nh_ofs + ip_len))
 178                 return -EINVAL;
 179
 180         skb_set_transport_header(skb, nh_ofs + ip_len);
 181         return 0;
 182 }
 183
 184 static bool tcphdr_ok(struct sk_buff *skb)
 185 {
 186         int th_ofs = skb_transport_offset(skb);
 187         int tcp_len;
 188
 189         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 190                 return false;
 191
 192         tcp_len = tcp_hdrlen(skb);
 193         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 194                      skb->len < th_ofs + tcp_len))
 195                 return false;
 196
 197         return true;
 198 }
 199
 200 static bool udphdr_ok(struct sk_buff *skb)
 201 {
 202         return pskb_may_pull(skb, skb_transport_offset(skb) +
 203                                   sizeof(struct udphdr));
 204 }
 205
 206 static bool sctphdr_ok(struct sk_buff *skb)
 207 {
 208         return pskb_may_pull(skb, skb_transport_offset(skb) +
 209                                   sizeof(struct sctphdr));
 210 }
 211
 212 static bool icmphdr_ok(struct sk_buff *skb)
 213 {
 214         return pskb_may_pull(skb, skb_transport_offset(skb) +
 215                                   sizeof(struct icmphdr));
 216 }
 217
 218 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 219 {
 220         unsigned int nh_ofs = skb_network_offset(skb);
 221         unsigned int nh_len;
 222         int payload_ofs;
 223         struct ipv6hdr *nh;
 224         uint8_t nexthdr;
 225         __be16 frag_off;
 226         int err;
 227
 228         err = check_header(skb, nh_ofs + sizeof(*nh));
 229         if (unlikely(err))
 230                 return err;
 231
 232         nh = ipv6_hdr(skb);
 233         nexthdr = nh->nexthdr;
 234         payload_ofs = (u8 *)(nh + 1) - skb->data;
 235
 236         key->ip.proto = NEXTHDR_NONE;
 237         key->ip.tos = ipv6_get_dsfield(nh);
 238         key->ip.ttl = nh->hop_limit;
 239         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 240         key->ipv6.addr.src = nh->saddr;
 241         key->ipv6.addr.dst = nh->daddr;
 242
 243         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 244         if (unlikely(payload_ofs < 0))
 245                 return -EINVAL;
 246
 247         if (frag_off) {
 248                 if (frag_off & htons(~0x7))
 249                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 250                 else
 251                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 252         }
 253
 254         nh_len = payload_ofs - nh_ofs;
 255         skb_set_transport_header(skb, nh_ofs + nh_len);
 256         key->ip.proto = nexthdr;
 257         return nh_len;
 258 }
 259
 260 static bool icmp6hdr_ok(struct sk_buff *skb)
 261 {
 262         return pskb_may_pull(skb, skb_transport_offset(skb) +
 263                                   sizeof(struct icmp6hdr));
 264 }
 265
 266 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 267 {
 268         struct qtag_prefix {
 269                 __be16 eth_type; /* ETH_P_8021Q */
 270                 __be16 tci;
 271         };
 272         struct qtag_prefix *qp;
 273
 274         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 275                 return 0;
 276
 277         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 278                                          sizeof(__be16))))
 279                 return -ENOMEM;
 280
 281         qp = (struct qtag_prefix *) skb->data;
 282         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 283         __skb_pull(skb, sizeof(struct qtag_prefix));
 284
 285         return 0;
 286 }
 287
 288 static __be16 parse_ethertype(struct sk_buff *skb)
 289 {
 290         struct llc_snap_hdr {
 291                 u8  dsap;  /* Always 0xAA */
 292                 u8  ssap;  /* Always 0xAA */
 293                 u8  ctrl;
 294                 u8  oui[3];
 295                 __be16 ethertype;
 296         };
 297         struct llc_snap_hdr *llc;
 298         __be16 proto;
 299
 300         proto = *(__be16 *) skb->data;
 301         __skb_pull(skb, sizeof(__be16));
 302
 303         if (ntohs(proto) >= ETH_P_802_3_MIN)
 304                 return proto;
 305
 306         if (skb->len < sizeof(struct llc_snap_hdr))
 307                 return htons(ETH_P_802_2);
 308
 309         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 310                 return htons(0);
 311
 312         llc = (struct llc_snap_hdr *) skb->data;
 313         if (llc->dsap != LLC_SAP_SNAP ||
 314             llc->ssap != LLC_SAP_SNAP ||
 315             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 316                 return htons(ETH_P_802_2);
 317
 318         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 319
 320         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
 321                 return llc->ethertype;
 322
 323         return htons(ETH_P_802_2);
 324 }
 325
 326 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 327                         int nh_len)
 328 {
 329         struct icmp6hdr *icmp = icmp6_hdr(skb);
 330
 331         /* The ICMPv6 type and code fields use the 16-bit transport port
 332          * fields, so we need to store them in 16-bit network byte order.
 333          */
 334         key->ipv6.tp.src = htons(icmp->icmp6_type);
 335         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 336
 337         if (icmp->icmp6_code == 0 &&
 338             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 339              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 340                 int icmp_len = skb->len - skb_transport_offset(skb);
 341                 struct nd_msg *nd;
 342                 int offset;
 343
 344                 /* In order to process neighbor discovery options, we need the
 345                  * entire packet.
 346                  */
 347                 if (unlikely(icmp_len < sizeof(*nd)))
 348                         return 0;
 349
 350                 if (unlikely(skb_linearize(skb)))
 351                         return -ENOMEM;
 352
 353                 nd = (struct nd_msg *)skb_transport_header(skb);
 354                 key->ipv6.nd.target = nd->target;
 355
 356                 icmp_len -= sizeof(*nd);
 357                 offset = 0;
 358                 while (icmp_len >= 8) {
 359                         struct nd_opt_hdr *nd_opt =
 360                                  (struct nd_opt_hdr *)(nd->opt + offset);
 361                         int opt_len = nd_opt->nd_opt_len * 8;
 362
 363                         if (unlikely(!opt_len || opt_len > icmp_len))
 364                                 return 0;
 365
 366                         /* Store the link layer address if the appropriate
 367                          * option is provided.  It is considered an error if
 368                          * the same link layer option is specified twice.
 369                          */
 370                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 371                             && opt_len == 8) {
 372                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 373                                         goto invalid;
 374                                 memcpy(key->ipv6.nd.sll,
 375                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 376                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 377                                    && opt_len == 8) {
 378                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 379                                         goto invalid;
 380                                 memcpy(key->ipv6.nd.tll,
 381                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 382                         }
 383
 384                         icmp_len -= opt_len;
 385                         offset += opt_len;
 386                 }
 387         }
 388
 389         return 0;
 390
 391 invalid:
 392         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 393         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 394         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 395
 396         return 0;
 397 }
 398
 399 /**
 400  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 401  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 402  * Ethernet header
 403  * @in_port: port number on which @skb was received.
 404  * @key: output flow key
 405  *
 406  * The caller must ensure that skb->len >= ETH_HLEN.
 407  *
 408  * Returns 0 if successful, otherwise a negative errno value.
 409  *
 410  * Initializes @skb header pointers as follows:
 411  *
 412  *    - skb->mac_header: the Ethernet header.
 413  *
 414  *    - skb->network_header: just past the Ethernet header, or just past the
 415  *      VLAN header, to the first byte of the Ethernet payload.
 416  *
 417  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 418  *      on output, then just past the IP header, if one is present and
 419  *      of a correct length, otherwise the same as skb->network_header.
 420  *      For other key->eth.type values it is left untouched.
 421  */
 422 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
 423 {
 424         int error;
 425         struct ethhdr *eth;
 426
 427         memset(key, 0, sizeof(*key));
 428
 429         key->phy.priority = skb->priority;
 430         if (OVS_CB(skb)->tun_key)
 431                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 432         key->phy.in_port = in_port;
 433         key->phy.skb_mark = skb->mark;
 434
 435         skb_reset_mac_header(skb);
 436
 437         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 438          * header in the linear data area.
 439          */
 440         eth = eth_hdr(skb);
 441         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 442         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 443
 444         __skb_pull(skb, 2 * ETH_ALEN);
 445         /* We are going to push all headers that we pull, so no need to
 446          * update skb->csum here. */
 447
 448         if (vlan_tx_tag_present(skb))
 449                 key->eth.tci = htons(vlan_get_tci(skb));
 450         else if (eth->h_proto == htons(ETH_P_8021Q))
 451                 if (unlikely(parse_vlan(skb, key)))
 452                         return -ENOMEM;
 453
 454         key->eth.type = parse_ethertype(skb);
 455         if (unlikely(key->eth.type == htons(0)))
 456                 return -ENOMEM;
 457
 458         skb_reset_network_header(skb);
 459         __skb_push(skb, skb->data - skb_mac_header(skb));
 460
 461         /* Network layer. */
 462         if (key->eth.type == htons(ETH_P_IP)) {
 463                 struct iphdr *nh;
 464                 __be16 offset;
 465
 466                 error = check_iphdr(skb);
 467                 if (unlikely(error)) {
 468                         if (error == -EINVAL) {
 469                                 skb->transport_header = skb->network_header;
 470                                 error = 0;
 471                         }
 472                         return error;
 473                 }
 474
 475                 nh = ip_hdr(skb);
 476                 key->ipv4.addr.src = nh->saddr;
 477                 key->ipv4.addr.dst = nh->daddr;
 478
 479                 key->ip.proto = nh->protocol;
 480                 key->ip.tos = nh->tos;
 481                 key->ip.ttl = nh->ttl;
 482
 483                 offset = nh->frag_off & htons(IP_OFFSET);
 484                 if (offset) {
 485                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 486                         return 0;
 487                 }
 488                 if (nh->frag_off & htons(IP_MF) ||
 489                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 490                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 491
 492                 /* Transport layer. */
 493                 if (key->ip.proto == IPPROTO_TCP) {
 494                         if (tcphdr_ok(skb)) {
 495                                 struct tcphdr *tcp = tcp_hdr(skb);
 496                                 key->ipv4.tp.src = tcp->source;
 497                                 key->ipv4.tp.dst = tcp->dest;
 498                                 key->ipv4.tp.flags = TCP_FLAGS_BE16(tcp);
 499                         }
 500                 } else if (key->ip.proto == IPPROTO_UDP) {
 501                         if (udphdr_ok(skb)) {
 502                                 struct udphdr *udp = udp_hdr(skb);
 503                                 key->ipv4.tp.src = udp->source;
 504                                 key->ipv4.tp.dst = udp->dest;
 505                         }
 506                 } else if (key->ip.proto == IPPROTO_SCTP) {
 507                         if (sctphdr_ok(skb)) {
 508                                 struct sctphdr *sctp = sctp_hdr(skb);
 509                                 key->ipv4.tp.src = sctp->source;
 510                                 key->ipv4.tp.dst = sctp->dest;
 511                         }
 512                 } else if (key->ip.proto == IPPROTO_ICMP) {
 513                         if (icmphdr_ok(skb)) {
 514                                 struct icmphdr *icmp = icmp_hdr(skb);
 515                                 /* The ICMP type and code fields use the 16-bit
 516                                  * transport port fields, so we need to store
 517                                  * them in 16-bit network byte order. */
 518                                 key->ipv4.tp.src = htons(icmp->type);
 519                                 key->ipv4.tp.dst = htons(icmp->code);
 520                         }
 521                 }
 522
 523         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 524                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 525                 struct arp_eth_header *arp;
 526
 527                 arp = (struct arp_eth_header *)skb_network_header(skb);
 528
 529                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 530                                 && arp->ar_pro == htons(ETH_P_IP)
 531                                 && arp->ar_hln == ETH_ALEN
 532                                 && arp->ar_pln == 4) {
 533
 534                         /* We only match on the lower 8 bits of the opcode. */
 535                         if (ntohs(arp->ar_op) <= 0xff)
 536                                 key->ip.proto = ntohs(arp->ar_op);
 537                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 538                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 539                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 540                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 541                 }
 542         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 543                 int nh_len;             /* IPv6 Header + Extensions */
 544
 545                 nh_len = parse_ipv6hdr(skb, key);
 546                 if (unlikely(nh_len < 0)) {
 547                         if (nh_len == -EINVAL) {
 548                                 skb->transport_header = skb->network_header;
 549                                 error = 0;
 550                         } else {
 551                                 error = nh_len;
 552                         }
 553                         return error;
 554                 }
 555
 556                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 557                         return 0;
 558                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 559                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 560
 561                 /* Transport layer. */
 562                 if (key->ip.proto == NEXTHDR_TCP) {
 563                         if (tcphdr_ok(skb)) {
 564                                 struct tcphdr *tcp = tcp_hdr(skb);
 565                                 key->ipv6.tp.src = tcp->source;
 566                                 key->ipv6.tp.dst = tcp->dest;
 567                                 key->ipv6.tp.flags = TCP_FLAGS_BE16(tcp);
 568                         }
 569                 } else if (key->ip.proto == NEXTHDR_UDP) {
 570                         if (udphdr_ok(skb)) {
 571                                 struct udphdr *udp = udp_hdr(skb);
 572                                 key->ipv6.tp.src = udp->source;
 573                                 key->ipv6.tp.dst = udp->dest;
 574                         }
 575                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 576                         if (sctphdr_ok(skb)) {
 577                                 struct sctphdr *sctp = sctp_hdr(skb);
 578                                 key->ipv6.tp.src = sctp->source;
 579                                 key->ipv6.tp.dst = sctp->dest;
 580                         }
 581                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 582                         if (icmp6hdr_ok(skb)) {
 583                                 error = parse_icmpv6(skb, key, nh_len);
 584                                 if (error)
 585                                         return error;
 586                         }
 587                 }
 588         }
 589
 590         return 0;
 591 }