datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2015 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 <linux/uaccess.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/etherdevice.h>
  22 #include <linux/if_ether.h>
  23 #include <linux/if_vlan.h>
  24 #include <net/llc_pdu.h>
  25 #include <linux/kernel.h>
  26 #include <linux/jhash.h>
  27 #include <linux/jiffies.h>
  28 #include <linux/llc.h>
  29 #include <linux/module.h>
  30 #include <linux/in.h>
  31 #include <linux/rcupdate.h>
  32 #include <linux/cpumask.h>
  33 #include <linux/if_arp.h>
  34 #include <linux/ip.h>
  35 #include <linux/ipv6.h>
  36 #include <linux/mpls.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 <linux/timekeeping.h>
  45 #include <net/ip.h>
  46 #include <net/ipv6.h>
  47 #include <net/mpls.h>
  48 #include <net/ndisc.h>
  49 #include <net/nsh.h>
  50
  51 #include "datapath.h"
  52 #include "conntrack.h"
  53 #include "flow.h"
  54 #include "flow_netlink.h"
  55 #include "vport.h"
  56
  57 u64 ovs_flow_used_time(unsigned long flow_jiffies)
  58 {
  59         struct timespec64 cur_ts;
  60         u64 cur_ms, idle_ms;
  61
  62         ktime_get_ts64(&cur_ts);
  63         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
  64         cur_ms = (u64)(u32)cur_ts.tv_sec * MSEC_PER_SEC +
  65                  cur_ts.tv_nsec / NSEC_PER_MSEC;
  66
  67         return cur_ms - idle_ms;
  68 }
  69
  70 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
  71
  72 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
  73                            const struct sk_buff *skb)
  74 {
  75         struct sw_flow_stats *stats;
  76         unsigned int cpu = smp_processor_id();
  77         int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
  78
  79         stats = rcu_dereference(flow->stats[cpu]);
  80
  81         /* Check if already have CPU-specific stats. */
  82         if (likely(stats)) {
  83                 spin_lock(&stats->lock);
  84                 /* Mark if we write on the pre-allocated stats. */
  85                 if (cpu == 0 && unlikely(flow->stats_last_writer != cpu))
  86                         flow->stats_last_writer = cpu;
  87         } else {
  88                 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
  89                 spin_lock(&stats->lock);
  90
  91                 /* If the current CPU is the only writer on the
  92                  * pre-allocated stats keep using them.
  93                  */
  94                 if (unlikely(flow->stats_last_writer != cpu)) {
  95                         /* A previous locker may have already allocated the
  96                          * stats, so we need to check again.  If CPU-specific
  97                          * stats were already allocated, we update the pre-
  98                          * allocated stats as we have already locked them.
  99                          */
 100                         if (likely(flow->stats_last_writer != -1) &&
 101                             likely(!rcu_access_pointer(flow->stats[cpu]))) {
 102                                 /* Try to allocate CPU-specific stats. */
 103                                 struct sw_flow_stats *new_stats;
 104
 105                                 new_stats =
 106                                         kmem_cache_alloc_node(flow_stats_cache,
 107                                                               GFP_NOWAIT |
 108                                                               __GFP_THISNODE |
 109                                                               __GFP_NOWARN |
 110                                                               __GFP_NOMEMALLOC,
 111                                                               numa_node_id());
 112                                 if (likely(new_stats)) {
 113                                         new_stats->used = jiffies;
 114                                         new_stats->packet_count = 1;
 115                                         new_stats->byte_count = len;
 116                                         new_stats->tcp_flags = tcp_flags;
 117                                         spin_lock_init(&new_stats->lock);
 118
 119                                         rcu_assign_pointer(flow->stats[cpu],
 120                                                            new_stats);
 121                                         cpumask_set_cpu(cpu, &flow->cpu_used_mask);
 122                                         goto unlock;
 123                                 }
 124                         }
 125                         flow->stats_last_writer = cpu;
 126                 }
 127         }
 128
 129         stats->used = jiffies;
 130         stats->packet_count++;
 131         stats->byte_count += len;
 132         stats->tcp_flags |= tcp_flags;
 133 unlock:
 134         spin_unlock(&stats->lock);
 135 }
 136
 137 /* Must be called with rcu_read_lock or ovs_mutex. */
 138 void ovs_flow_stats_get(const struct sw_flow *flow,
 139                         struct ovs_flow_stats *ovs_stats,
 140                         unsigned long *used, __be16 *tcp_flags)
 141 {
 142         int cpu;
 143
 144         *used = 0;
 145         *tcp_flags = 0;
 146         memset(ovs_stats, 0, sizeof(*ovs_stats));
 147
 148         /* We open code this to make sure cpu 0 is always considered */
 149         for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
 150                 struct sw_flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
 151
 152                 if (stats) {
 153                         /* Local CPU may write on non-local stats, so we must
 154                          * block bottom-halves here.
 155                          */
 156                         spin_lock_bh(&stats->lock);
 157                         if (!*used || time_after(stats->used, *used))
 158                                 *used = stats->used;
 159                         *tcp_flags |= stats->tcp_flags;
 160                         ovs_stats->n_packets += stats->packet_count;
 161                         ovs_stats->n_bytes += stats->byte_count;
 162                         spin_unlock_bh(&stats->lock);
 163                 }
 164         }
 165 }
 166
 167 /* Called with ovs_mutex. */
 168 void ovs_flow_stats_clear(struct sw_flow *flow)
 169 {
 170         int cpu;
 171
 172         /* We open code this to make sure cpu 0 is always considered */
 173         for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
 174                 struct sw_flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
 175
 176                 if (stats) {
 177                         spin_lock_bh(&stats->lock);
 178                         stats->used = 0;
 179                         stats->packet_count = 0;
 180                         stats->byte_count = 0;
 181                         stats->tcp_flags = 0;
 182                         spin_unlock_bh(&stats->lock);
 183                 }
 184         }
 185 }
 186
 187 static int check_header(struct sk_buff *skb, int len)
 188 {
 189         if (unlikely(skb->len < len))
 190                 return -EINVAL;
 191         if (unlikely(!pskb_may_pull(skb, len)))
 192                 return -ENOMEM;
 193         return 0;
 194 }
 195
 196 static bool arphdr_ok(struct sk_buff *skb)
 197 {
 198         return pskb_may_pull(skb, skb_network_offset(skb) +
 199                                   sizeof(struct arp_eth_header));
 200 }
 201
 202 static int check_iphdr(struct sk_buff *skb)
 203 {
 204         unsigned int nh_ofs = skb_network_offset(skb);
 205         unsigned int ip_len;
 206         int err;
 207
 208         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 209         if (unlikely(err))
 210                 return err;
 211
 212         ip_len = ip_hdrlen(skb);
 213         if (unlikely(ip_len < sizeof(struct iphdr) ||
 214                      skb->len < nh_ofs + ip_len))
 215                 return -EINVAL;
 216
 217         skb_set_transport_header(skb, nh_ofs + ip_len);
 218         return 0;
 219 }
 220
 221 static bool tcphdr_ok(struct sk_buff *skb)
 222 {
 223         int th_ofs = skb_transport_offset(skb);
 224         int tcp_len;
 225
 226         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 227                 return false;
 228
 229         tcp_len = tcp_hdrlen(skb);
 230         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 231                      skb->len < th_ofs + tcp_len))
 232                 return false;
 233
 234         return true;
 235 }
 236
 237 static bool udphdr_ok(struct sk_buff *skb)
 238 {
 239         return pskb_may_pull(skb, skb_transport_offset(skb) +
 240                                   sizeof(struct udphdr));
 241 }
 242
 243 static bool sctphdr_ok(struct sk_buff *skb)
 244 {
 245         return pskb_may_pull(skb, skb_transport_offset(skb) +
 246                                   sizeof(struct sctphdr));
 247 }
 248
 249 static bool icmphdr_ok(struct sk_buff *skb)
 250 {
 251         return pskb_may_pull(skb, skb_transport_offset(skb) +
 252                                   sizeof(struct icmphdr));
 253 }
 254
 255 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 256 {
 257         unsigned short frag_off;
 258         unsigned int payload_ofs = 0;
 259         unsigned int nh_ofs = skb_network_offset(skb);
 260         unsigned int nh_len;
 261         struct ipv6hdr *nh;
 262         int err, nexthdr, flags = 0;
 263
 264         err = check_header(skb, nh_ofs + sizeof(*nh));
 265         if (unlikely(err))
 266                 return err;
 267
 268         nh = ipv6_hdr(skb);
 269
 270         key->ip.proto = NEXTHDR_NONE;
 271         key->ip.tos = ipv6_get_dsfield(nh);
 272         key->ip.ttl = nh->hop_limit;
 273         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 274         key->ipv6.addr.src = nh->saddr;
 275         key->ipv6.addr.dst = nh->daddr;
 276
 277         nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
 278         if (flags & IP6_FH_F_FRAG) {
 279                 if (frag_off) {
 280                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 281                         key->ip.proto = nexthdr;
 282                         return 0;
 283                 }
 284                 key->ip.frag = OVS_FRAG_TYPE_FIRST;
 285         } else {
 286                 key->ip.frag = OVS_FRAG_TYPE_NONE;
 287         }
 288
 289         /* Delayed handling of error in ipv6_find_hdr() as it
 290          * always sets flags and frag_off to a valid value which may be
 291          * used to set key->ip.frag above.
 292          */
 293         if (unlikely(nexthdr < 0))
 294                 return -EPROTO;
 295
 296         nh_len = payload_ofs - nh_ofs;
 297         skb_set_transport_header(skb, nh_ofs + nh_len);
 298         key->ip.proto = nexthdr;
 299         return nh_len;
 300 }
 301
 302 static bool icmp6hdr_ok(struct sk_buff *skb)
 303 {
 304         return pskb_may_pull(skb, skb_transport_offset(skb) +
 305                                   sizeof(struct icmp6hdr));
 306 }
 307
 308 /**
 309  * Parse vlan tag from vlan header.
 310  * Returns ERROR on memory error.
 311  * Returns 0 if it encounters a non-vlan or incomplete packet.
 312  * Returns 1 after successfully parsing vlan tag.
 313  */
 314 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
 315                           bool untag_vlan)
 316 {
 317         struct vlan_head *vh = (struct vlan_head *)skb->data;
 318
 319         if (likely(!eth_type_vlan(vh->tpid)))
 320                 return 0;
 321
 322         if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
 323                 return 0;
 324
 325         if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
 326                                  sizeof(__be16))))
 327                 return -ENOMEM;
 328
 329         vh = (struct vlan_head *)skb->data;
 330         key_vh->tci = vh->tci | htons(VLAN_CFI_MASK);
 331         key_vh->tpid = vh->tpid;
 332
 333         if (unlikely(untag_vlan)) {
 334                 int offset = skb->data - skb_mac_header(skb);
 335                 u16 tci;
 336                 int err;
 337
 338                 __skb_push(skb, offset);
 339                 err = __skb_vlan_pop(skb, &tci);
 340                 __skb_pull(skb, offset);
 341                 if (err)
 342                         return err;
 343                 __vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
 344         } else {
 345                 __skb_pull(skb, sizeof(struct vlan_head));
 346         }
 347         return 1;
 348 }
 349
 350 static void clear_vlan(struct sw_flow_key *key)
 351 {
 352         key->eth.vlan.tci = 0;
 353         key->eth.vlan.tpid = 0;
 354         key->eth.cvlan.tci = 0;
 355         key->eth.cvlan.tpid = 0;
 356 }
 357
 358 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 359 {
 360         int res;
 361
 362         key->eth.vlan.tci = 0;
 363         key->eth.vlan.tpid = 0;
 364         key->eth.cvlan.tci = 0;
 365         key->eth.cvlan.tpid = 0;
 366
 367         if (skb_vlan_tag_present(skb)) {
 368                 key->eth.vlan.tci = htons(skb->vlan_tci) | htons(VLAN_CFI_MASK);
 369                 key->eth.vlan.tpid = skb->vlan_proto;
 370         } else {
 371                 /* Parse outer vlan tag in the non-accelerated case. */
 372                 res = parse_vlan_tag(skb, &key->eth.vlan, true);
 373                 if (res <= 0)
 374                         return res;
 375         }
 376
 377         /* Parse inner vlan tag. */
 378         res = parse_vlan_tag(skb, &key->eth.cvlan, false);
 379         if (res <= 0)
 380                 return res;
 381
 382         return 0;
 383 }
 384
 385 static __be16 parse_ethertype(struct sk_buff *skb)
 386 {
 387         struct llc_snap_hdr {
 388                 u8  dsap;  /* Always 0xAA */
 389                 u8  ssap;  /* Always 0xAA */
 390                 u8  ctrl;
 391                 u8  oui[3];
 392                 __be16 ethertype;
 393         };
 394         struct llc_snap_hdr *llc;
 395         __be16 proto;
 396
 397         proto = *(__be16 *) skb->data;
 398         __skb_pull(skb, sizeof(__be16));
 399
 400         if (eth_proto_is_802_3(proto))
 401                 return proto;
 402
 403         if (skb->len < sizeof(struct llc_snap_hdr))
 404                 return htons(ETH_P_802_2);
 405
 406         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 407                 return htons(0);
 408
 409         llc = (struct llc_snap_hdr *) skb->data;
 410         if (llc->dsap != LLC_SAP_SNAP ||
 411             llc->ssap != LLC_SAP_SNAP ||
 412             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 413                 return htons(ETH_P_802_2);
 414
 415         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 416
 417         if (eth_proto_is_802_3(llc->ethertype))
 418                 return llc->ethertype;
 419
 420         return htons(ETH_P_802_2);
 421 }
 422
 423 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 424                         int nh_len)
 425 {
 426         struct icmp6hdr *icmp = icmp6_hdr(skb);
 427
 428         /* The ICMPv6 type and code fields use the 16-bit transport port
 429          * fields, so we need to store them in 16-bit network byte order.
 430          */
 431         key->tp.src = htons(icmp->icmp6_type);
 432         key->tp.dst = htons(icmp->icmp6_code);
 433         memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
 434
 435         if (icmp->icmp6_code == 0 &&
 436             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 437              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 438                 int icmp_len = skb->len - skb_transport_offset(skb);
 439                 struct nd_msg *nd;
 440                 int offset;
 441
 442                 /* In order to process neighbor discovery options, we need the
 443                  * entire packet.
 444                  */
 445                 if (unlikely(icmp_len < sizeof(*nd)))
 446                         return 0;
 447
 448                 if (unlikely(skb_linearize(skb)))
 449                         return -ENOMEM;
 450
 451                 nd = (struct nd_msg *)skb_transport_header(skb);
 452                 key->ipv6.nd.target = nd->target;
 453
 454                 icmp_len -= sizeof(*nd);
 455                 offset = 0;
 456                 while (icmp_len >= 8) {
 457                         struct nd_opt_hdr *nd_opt =
 458                                  (struct nd_opt_hdr *)(nd->opt + offset);
 459                         int opt_len = nd_opt->nd_opt_len * 8;
 460
 461                         if (unlikely(!opt_len || opt_len > icmp_len))
 462                                 return 0;
 463
 464                         /* Store the link layer address if the appropriate
 465                          * option is provided.  It is considered an error if
 466                          * the same link layer option is specified twice.
 467                          */
 468                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 469                             && opt_len == 8) {
 470                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 471                                         goto invalid;
 472                                 ether_addr_copy(key->ipv6.nd.sll,
 473                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 474                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 475                                    && opt_len == 8) {
 476                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 477                                         goto invalid;
 478                                 ether_addr_copy(key->ipv6.nd.tll,
 479                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 480                         }
 481
 482                         icmp_len -= opt_len;
 483                         offset += opt_len;
 484                 }
 485         }
 486
 487         return 0;
 488
 489 invalid:
 490         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 491         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 492         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 493
 494         return 0;
 495 }
 496
 497 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 498 {
 499         struct nshhdr *nh;
 500         unsigned int nh_ofs = skb_network_offset(skb);
 501         u8 version, length;
 502         int err;
 503
 504         err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
 505         if (unlikely(err))
 506                 return err;
 507
 508         nh = nsh_hdr(skb);
 509         version = nsh_get_ver(nh);
 510         length = nsh_hdr_len(nh);
 511
 512         if (version != 0)
 513                 return -EINVAL;
 514
 515         err = check_header(skb, nh_ofs + length);
 516         if (unlikely(err))
 517                 return err;
 518
 519         nh = nsh_hdr(skb);
 520         key->nsh.base.flags = nsh_get_flags(nh);
 521         key->nsh.base.ttl = nsh_get_ttl(nh);
 522         key->nsh.base.mdtype = nh->mdtype;
 523         key->nsh.base.np = nh->np;
 524         key->nsh.base.path_hdr = nh->path_hdr;
 525         switch (key->nsh.base.mdtype) {
 526         case NSH_M_TYPE1:
 527                 if (length != NSH_M_TYPE1_LEN)
 528                         return -EINVAL;
 529                 memcpy(key->nsh.context, nh->md1.context,
 530                        sizeof(nh->md1));
 531                 break;
 532         case NSH_M_TYPE2:
 533                 memset(key->nsh.context, 0,
 534                        sizeof(nh->md1));
 535                 break;
 536         default:
 537                 return -EINVAL;
 538         }
 539
 540         return 0;
 541 }
 542
 543 /**
 544  * key_extract_l3l4 - extracts L3/L4 header information.
 545  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 546  *       L3 header
 547  * @key: output flow key
 548  */
 549 static int key_extract_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
 550 {
 551         int error;
 552
 553         /* Network layer. */
 554         if (key->eth.type == htons(ETH_P_IP)) {
 555                 struct iphdr *nh;
 556                 __be16 offset;
 557
 558                 error = check_iphdr(skb);
 559                 if (unlikely(error)) {
 560                         memset(&key->ip, 0, sizeof(key->ip));
 561                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 562                         if (error == -EINVAL) {
 563                                 skb->transport_header = skb->network_header;
 564                                 error = 0;
 565                         }
 566                         return error;
 567                 }
 568
 569                 nh = ip_hdr(skb);
 570                 key->ipv4.addr.src = nh->saddr;
 571                 key->ipv4.addr.dst = nh->daddr;
 572
 573                 key->ip.proto = nh->protocol;
 574                 key->ip.tos = nh->tos;
 575                 key->ip.ttl = nh->ttl;
 576
 577                 offset = nh->frag_off & htons(IP_OFFSET);
 578                 if (offset) {
 579                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 580                         memset(&key->tp, 0, sizeof(key->tp));
 581                         return 0;
 582                 }
 583 #ifdef HAVE_SKB_GSO_UDP
 584                 if (nh->frag_off & htons(IP_MF) ||
 585                         skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 586 #else
 587                 if (nh->frag_off & htons(IP_MF))
 588 #endif
 589                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 590                 else
 591                         key->ip.frag = OVS_FRAG_TYPE_NONE;
 592
 593                 /* Transport layer. */
 594                 if (key->ip.proto == IPPROTO_TCP) {
 595                         if (tcphdr_ok(skb)) {
 596                                 struct tcphdr *tcp = tcp_hdr(skb);
 597                                 key->tp.src = tcp->source;
 598                                 key->tp.dst = tcp->dest;
 599                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 600                         } else {
 601                                 memset(&key->tp, 0, sizeof(key->tp));
 602                         }
 603
 604                 } else if (key->ip.proto == IPPROTO_UDP) {
 605                         if (udphdr_ok(skb)) {
 606                                 struct udphdr *udp = udp_hdr(skb);
 607                                 key->tp.src = udp->source;
 608                                 key->tp.dst = udp->dest;
 609                         } else {
 610                                 memset(&key->tp, 0, sizeof(key->tp));
 611                         }
 612                 } else if (key->ip.proto == IPPROTO_SCTP) {
 613                         if (sctphdr_ok(skb)) {
 614                                 struct sctphdr *sctp = sctp_hdr(skb);
 615                                 key->tp.src = sctp->source;
 616                                 key->tp.dst = sctp->dest;
 617                         } else {
 618                                 memset(&key->tp, 0, sizeof(key->tp));
 619                         }
 620                 } else if (key->ip.proto == IPPROTO_ICMP) {
 621                         if (icmphdr_ok(skb)) {
 622                                 struct icmphdr *icmp = icmp_hdr(skb);
 623                                 /* The ICMP type and code fields use the 16-bit
 624                                  * transport port fields, so we need to store
 625                                  * them in 16-bit network byte order.
 626                                  */
 627                                 key->tp.src = htons(icmp->type);
 628                                 key->tp.dst = htons(icmp->code);
 629                         } else {
 630                                 memset(&key->tp, 0, sizeof(key->tp));
 631                         }
 632                 }
 633
 634         } else if (key->eth.type == htons(ETH_P_ARP) ||
 635                    key->eth.type == htons(ETH_P_RARP)) {
 636                 struct arp_eth_header *arp;
 637                 bool arp_available = arphdr_ok(skb);
 638
 639                 arp = (struct arp_eth_header *)skb_network_header(skb);
 640
 641                 if (arp_available &&
 642                     arp->ar_hrd == htons(ARPHRD_ETHER) &&
 643                     arp->ar_pro == htons(ETH_P_IP) &&
 644                     arp->ar_hln == ETH_ALEN &&
 645                     arp->ar_pln == 4) {
 646
 647                         /* We only match on the lower 8 bits of the opcode. */
 648                         if (ntohs(arp->ar_op) <= 0xff)
 649                                 key->ip.proto = ntohs(arp->ar_op);
 650                         else
 651                                 key->ip.proto = 0;
 652
 653                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 654                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 655                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
 656                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
 657                 } else {
 658                         memset(&key->ip, 0, sizeof(key->ip));
 659                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 660                 }
 661         } else if (eth_p_mpls(key->eth.type)) {
 662                 u8 label_count = 1;
 663
 664                 memset(&key->mpls, 0, sizeof(key->mpls));
 665                 skb_set_inner_network_header(skb, skb->mac_len);
 666                 while (1) {
 667                         __be32 lse;
 668
 669                         error = check_header(skb, skb->mac_len +
 670                                              label_count * MPLS_HLEN);
 671                         if (unlikely(error))
 672                                 return 0;
 673
 674                         memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
 675
 676                         if (label_count <= MPLS_LABEL_DEPTH)
 677                                 memcpy(&key->mpls.lse[label_count - 1], &lse,
 678                                        MPLS_HLEN);
 679
 680                         skb_set_inner_network_header(skb, skb->mac_len +
 681                                                      label_count * MPLS_HLEN);
 682                         if (lse & htonl(MPLS_LS_S_MASK))
 683                                 break;
 684
 685                         label_count++;
 686                 }
 687                 if (label_count > MPLS_LABEL_DEPTH)
 688                         label_count = MPLS_LABEL_DEPTH;
 689
 690                 key->mpls.num_labels_mask = GENMASK(label_count - 1, 0);
 691         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 692                 int nh_len;             /* IPv6 Header + Extensions */
 693
 694                 nh_len = parse_ipv6hdr(skb, key);
 695                 if (unlikely(nh_len < 0)) {
 696                         switch (nh_len) {
 697                         case -EINVAL:
 698                                 memset(&key->ip, 0, sizeof(key->ip));
 699                                 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
 700                                 /* fall-through */
 701                         case -EPROTO:
 702                                 skb->transport_header = skb->network_header;
 703                                 error = 0;
 704                                 break;
 705                         default:
 706                                 error = nh_len;
 707                         }
 708                         return error;
 709                 }
 710
 711                 if (key->ip.frag == OVS_FRAG_TYPE_LATER) {
 712                         memset(&key->tp, 0, sizeof(key->tp));
 713                         return 0;
 714                 }
 715 #ifdef HAVE_SKB_GSO_UDP
 716                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 717                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 718
 719 #endif
 720                 /* Transport layer. */
 721                 if (key->ip.proto == NEXTHDR_TCP) {
 722                         if (tcphdr_ok(skb)) {
 723                                 struct tcphdr *tcp = tcp_hdr(skb);
 724                                 key->tp.src = tcp->source;
 725                                 key->tp.dst = tcp->dest;
 726                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 727                         } else {
 728                                 memset(&key->tp, 0, sizeof(key->tp));
 729                         }
 730                 } else if (key->ip.proto == NEXTHDR_UDP) {
 731                         if (udphdr_ok(skb)) {
 732                                 struct udphdr *udp = udp_hdr(skb);
 733                                 key->tp.src = udp->source;
 734                                 key->tp.dst = udp->dest;
 735                         } else {
 736                                 memset(&key->tp, 0, sizeof(key->tp));
 737                         }
 738                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 739                         if (sctphdr_ok(skb)) {
 740                                 struct sctphdr *sctp = sctp_hdr(skb);
 741                                 key->tp.src = sctp->source;
 742                                 key->tp.dst = sctp->dest;
 743                         } else {
 744                                 memset(&key->tp, 0, sizeof(key->tp));
 745                         }
 746                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 747                         if (icmp6hdr_ok(skb)) {
 748                                 error = parse_icmpv6(skb, key, nh_len);
 749                                 if (error)
 750                                         return error;
 751                         } else {
 752                                 memset(&key->tp, 0, sizeof(key->tp));
 753                         }
 754                 }
 755         } else if (key->eth.type == htons(ETH_P_NSH)) {
 756                 error = parse_nsh(skb, key);
 757                 if (error)
 758                         return error;
 759         }
 760         return 0;
 761 }
 762
 763 /**
 764  * key_extract - extracts a flow key from an Ethernet frame.
 765  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 766  * Ethernet header
 767  * @key: output flow key
 768  *
 769  * The caller must ensure that skb->len >= ETH_HLEN.
 770  *
 771  * Returns 0 if successful, otherwise a negative errno value.
 772  *
 773  * Initializes @skb header fields as follows:
 774  *
 775  *    - skb->mac_header: the L2 header.
 776  *
 777  *    - skb->network_header: just past the L2 header, or just past the
 778  *      VLAN header, to the first byte of the L2 payload.
 779  *
 780  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 781  *      on output, then just past the IP header, if one is present and
 782  *      of a correct length, otherwise the same as skb->network_header.
 783  *      For other key->eth.type values it is left untouched.
 784  *
 785  *    - skb->protocol: the type of the data starting at skb->network_header.
 786  *      Equals to key->eth.type.
 787  */
 788 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
 789 {
 790         struct ethhdr *eth;
 791
 792         /* Flags are always used as part of stats */
 793         key->tp.flags = 0;
 794
 795         skb_reset_mac_header(skb);
 796
 797         /* Link layer. */
 798         clear_vlan(key);
 799         if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
 800                 if (unlikely(eth_type_vlan(skb->protocol)))
 801                         return -EINVAL;
 802
 803                 skb_reset_network_header(skb);
 804                 key->eth.type = skb->protocol;
 805         } else {
 806                 eth = eth_hdr(skb);
 807                 ether_addr_copy(key->eth.src, eth->h_source);
 808                 ether_addr_copy(key->eth.dst, eth->h_dest);
 809
 810                 __skb_pull(skb, 2 * ETH_ALEN);
 811                 /* We are going to push all headers that we pull, so no need to
 812                  * update skb->csum here.
 813                  */
 814
 815                 if (unlikely(parse_vlan(skb, key)))
 816                         return -ENOMEM;
 817
 818                 key->eth.type = parse_ethertype(skb);
 819                 if (unlikely(key->eth.type == htons(0)))
 820                         return -ENOMEM;
 821
 822                 /* Multiple tagged packets need to retain TPID to satisfy
 823                  * skb_vlan_pop(), which will later shift the ethertype into
 824                  * skb->protocol.
 825                  */
 826                 if (key->eth.cvlan.tci & htons(VLAN_CFI_MASK))
 827                         skb->protocol = key->eth.cvlan.tpid;
 828                 else
 829                         skb->protocol = key->eth.type;
 830
 831                 skb_reset_network_header(skb);
 832                 __skb_push(skb, skb->data - skb_mac_header(skb));
 833         }
 834
 835         skb_reset_mac_len(skb);
 836
 837         /* Fill out L3/L4 key info, if any */
 838         return key_extract_l3l4(skb, key);
 839 }
 840
 841 /* In the case of conntrack fragment handling it expects L3 headers,
 842  * add a helper.
 843  */
 844 int ovs_flow_key_update_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
 845 {
 846         return key_extract_l3l4(skb, key);
 847 }
 848
 849 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
 850 {
 851         int res;
 852
 853         res = key_extract(skb, key);
 854         if (!res)
 855                 key->mac_proto &= ~SW_FLOW_KEY_INVALID;
 856
 857         return res;
 858 }
 859
 860 static int key_extract_mac_proto(struct sk_buff *skb)
 861 {
 862         switch (skb->dev->type) {
 863         case ARPHRD_ETHER:
 864                 return MAC_PROTO_ETHERNET;
 865         case ARPHRD_NONE:
 866                 if (skb->protocol == htons(ETH_P_TEB))
 867                         return MAC_PROTO_ETHERNET;
 868                 return MAC_PROTO_NONE;
 869         }
 870         WARN_ON_ONCE(1);
 871         return -EINVAL;
 872 }
 873
 874 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
 875                          struct sk_buff *skb, struct sw_flow_key *key)
 876 {
 877         int res, err;
 878
 879         /* Extract metadata from packet. */
 880         if (tun_info) {
 881                 key->tun_proto = ip_tunnel_info_af(tun_info);
 882                 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
 883                 BUILD_BUG_ON(((1 << (sizeof(tun_info->options_len) * 8)) - 1) >
 884                              sizeof(key->tun_opts));
 885
 886                 if (tun_info->options_len) {
 887                         ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
 888                                                 tun_info);
 889                         key->tun_opts_len = tun_info->options_len;
 890                 } else {
 891                         key->tun_opts_len = 0;
 892                 }
 893         } else  {
 894                 key->tun_proto = 0;
 895                 key->tun_opts_len = 0;
 896                 memset(&key->tun_key, 0, sizeof(key->tun_key));
 897         }
 898
 899         key->phy.priority = skb->priority;
 900         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
 901         key->phy.skb_mark = skb->mark;
 902         key->ovs_flow_hash = 0;
 903         res = key_extract_mac_proto(skb);
 904         if (res < 0)
 905                 return res;
 906         key->mac_proto = res;
 907         key->recirc_id = 0;
 908
 909         err = key_extract(skb, key);
 910         if (!err)
 911                 ovs_ct_fill_key(skb, key);   /* Must be after key_extract(). */
 912         return err;
 913 }
 914
 915 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
 916                                    struct sk_buff *skb,
 917                                    struct sw_flow_key *key, bool log)
 918 {
 919         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 920         u64 attrs = 0;
 921         int err;
 922
 923         err = parse_flow_nlattrs(attr, a, &attrs, log);
 924         if (err)
 925                 return -EINVAL;
 926
 927         /* Extract metadata from netlink attributes. */
 928         err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
 929         if (err)
 930                 return err;
 931
 932         /* key_extract assumes that skb->protocol is set-up for
 933          * layer 3 packets which is the case for other callers,
 934          * in particular packets received from the network stack.
 935          * Here the correct value can be set from the metadata
 936          * extracted above.
 937          * For L2 packet key eth type would be zero. skb protocol
 938          * would be set to correct value later during key-extact.
 939          */
 940
 941         skb->protocol = key->eth.type;
 942         err = key_extract(skb, key);
 943         if (err)
 944                 return err;
 945
 946         /* Check that we have conntrack original direction tuple metadata only
 947          * for packets for which it makes sense.  Otherwise the key may be
 948          * corrupted due to overlapping key fields.
 949          */
 950         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
 951             key->eth.type != htons(ETH_P_IP))
 952                 return -EINVAL;
 953         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
 954             (key->eth.type != htons(ETH_P_IPV6) ||
 955              sw_flow_key_is_nd(key)))
 956                 return -EINVAL;
 957
 958         return 0;
 959 }