datapath/flow_netlink.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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include "flow.h"
  22 #include "datapath.h"
  23 #include "mpls.h"
  24 #include <linux/uaccess.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/etherdevice.h>
  27 #include <linux/if_ether.h>
  28 #include <linux/if_vlan.h>
  29 #include <net/llc_pdu.h>
  30 #include <linux/kernel.h>
  31 #include <linux/jhash.h>
  32 #include <linux/jiffies.h>
  33 #include <linux/llc.h>
  34 #include <linux/module.h>
  35 #include <linux/in.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/if_arp.h>
  38 #include <linux/ip.h>
  39 #include <linux/ipv6.h>
  40 #include <linux/sctp.h>
  41 #include <linux/tcp.h>
  42 #include <linux/udp.h>
  43 #include <linux/icmp.h>
  44 #include <linux/icmpv6.h>
  45 #include <linux/rculist.h>
  46 #include <net/geneve.h>
  47 #include <net/ip.h>
  48 #include <net/ip_tunnels.h>
  49 #include <net/ipv6.h>
  50 #include <net/ndisc.h>
  51
  52 #include "flow_netlink.h"
  53
  54 static void update_range__(struct sw_flow_match *match,
  55                            size_t offset, size_t size, bool is_mask)
  56 {
  57         struct sw_flow_key_range *range = NULL;
  58         size_t start = rounddown(offset, sizeof(long));
  59         size_t end = roundup(offset + size, sizeof(long));
  60
  61         if (!is_mask)
  62                 range = &match->range;
  63         else if (match->mask)
  64                 range = &match->mask->range;
  65
  66         if (!range)
  67                 return;
  68
  69         if (range->start == range->end) {
  70                 range->start = start;
  71                 range->end = end;
  72                 return;
  73         }
  74
  75         if (range->start > start)
  76                 range->start = start;
  77
  78         if (range->end < end)
  79                 range->end = end;
  80 }
  81
  82 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  83         do { \
  84                 update_range__(match, offsetof(struct sw_flow_key, field),  \
  85                                      sizeof((match)->key->field), is_mask); \
  86                 if (is_mask) {                                              \
  87                         if ((match)->mask)                                  \
  88                                 (match)->mask->key.field = value;           \
  89                 } else {                                                    \
  90                         (match)->key->field = value;                        \
  91                 }                                                           \
  92         } while (0)
  93
  94 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask) \
  95         do { \
  96                 update_range__(match, offset, len, is_mask);                \
  97                 if (is_mask) {                                              \
  98                         if ((match)->mask)                                  \
  99                                 memcpy((u8 *)&(match)->mask->key + offset, value_p, len);\
 100                 } else {                                                    \
 101                         memcpy((u8 *)(match)->key + offset, value_p, len);         \
 102                 }                                                           \
 103         } while (0)
 104
 105 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
 106         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 107                                   value_p, len, is_mask)
 108
 109 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask) \
 110         do { \
 111                 update_range__(match, offsetof(struct sw_flow_key, field),  \
 112                                      sizeof((match)->key->field), is_mask); \
 113                 if (is_mask) {                                              \
 114                         if ((match)->mask)                                  \
 115                                 memset((u8 *)&(match)->mask->key.field, value,\
 116                                        sizeof((match)->mask->key.field));   \
 117                 } else {                                                    \
 118                         memset((u8 *)&(match)->key->field, value,           \
 119                                sizeof((match)->key->field));                \
 120                 }                                                           \
 121         } while (0)
 122
 123 static bool match_validate(const struct sw_flow_match *match,
 124                            u64 key_attrs, u64 mask_attrs)
 125 {
 126         u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
 127         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 128
 129         /* The following mask attributes allowed only if they
 130          * pass the validation tests. */
 131         mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
 132                         | (1ULL << OVS_KEY_ATTR_IPV6)
 133                         | (1ULL << OVS_KEY_ATTR_TCP)
 134                         | (1ULL << OVS_KEY_ATTR_TCP_FLAGS)
 135                         | (1ULL << OVS_KEY_ATTR_UDP)
 136                         | (1ULL << OVS_KEY_ATTR_SCTP)
 137                         | (1ULL << OVS_KEY_ATTR_ICMP)
 138                         | (1ULL << OVS_KEY_ATTR_ICMPV6)
 139                         | (1ULL << OVS_KEY_ATTR_ARP)
 140                         | (1ULL << OVS_KEY_ATTR_ND)
 141                         | (1ULL << OVS_KEY_ATTR_MPLS));
 142
 143         /* Always allowed mask fields. */
 144         mask_allowed |= ((1ULL << OVS_KEY_ATTR_TUNNEL)
 145                        | (1ULL << OVS_KEY_ATTR_IN_PORT)
 146                        | (1ULL << OVS_KEY_ATTR_ETHERTYPE));
 147
 148         /* Check key attributes. */
 149         if (match->key->eth.type == htons(ETH_P_ARP)
 150                         || match->key->eth.type == htons(ETH_P_RARP)) {
 151                 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
 152                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 153                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
 154         }
 155
 156
 157         if (eth_p_mpls(match->key->eth.type)) {
 158                 key_expected |= 1ULL << OVS_KEY_ATTR_MPLS;
 159                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 160                         mask_allowed |= 1ULL << OVS_KEY_ATTR_MPLS;
 161         }
 162
 163         if (match->key->eth.type == htons(ETH_P_IP)) {
 164                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
 165                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 166                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
 167
 168                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 169                         if (match->key->ip.proto == IPPROTO_UDP) {
 170                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 171                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 172                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 173                         }
 174
 175                         if (match->key->ip.proto == IPPROTO_SCTP) {
 176                                 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
 177                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 178                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
 179                         }
 180
 181                         if (match->key->ip.proto == IPPROTO_TCP) {
 182                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 183                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 184                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 185                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 186                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 187                                 }
 188                         }
 189
 190                         if (match->key->ip.proto == IPPROTO_ICMP) {
 191                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
 192                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 193                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
 194                         }
 195                 }
 196         }
 197
 198         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 199                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
 200                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 201                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
 202
 203                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 204                         if (match->key->ip.proto == IPPROTO_UDP) {
 205                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 206                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 207                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 208                         }
 209
 210                         if (match->key->ip.proto == IPPROTO_SCTP) {
 211                                 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
 212                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 213                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
 214                         }
 215
 216                         if (match->key->ip.proto == IPPROTO_TCP) {
 217                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 218                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 219                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 220                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 221                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 222                                 }
 223                         }
 224
 225                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 226                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 227                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 228                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 229
 230                                 if (match->key->tp.src ==
 231                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 232                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 233                                         key_expected |= 1ULL << OVS_KEY_ATTR_ND;
 234                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 235                                                 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
 236                                 }
 237                         }
 238                 }
 239         }
 240
 241         if ((key_attrs & key_expected) != key_expected) {
 242                 /* Key attributes check failed. */
 243                 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
 244                                 (unsigned long long)key_attrs, (unsigned long long)key_expected);
 245                 return false;
 246         }
 247
 248         if ((mask_attrs & mask_allowed) != mask_attrs) {
 249                 /* Mask attributes check failed. */
 250                 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
 251                                 (unsigned long long)mask_attrs, (unsigned long long)mask_allowed);
 252                 return false;
 253         }
 254
 255         return true;
 256 }
 257
 258 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 259 static const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 260         [OVS_KEY_ATTR_ENCAP] = -1,
 261         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 262         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 263         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
 264         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 265         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 266         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 267         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 268         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 269         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 270         [OVS_KEY_ATTR_TCP_FLAGS] = sizeof(__be16),
 271         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 272         [OVS_KEY_ATTR_SCTP] = sizeof(struct ovs_key_sctp),
 273         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 274         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 275         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 276         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 277         [OVS_KEY_ATTR_DP_HASH] = sizeof(u32),
 278         [OVS_KEY_ATTR_RECIRC_ID] = sizeof(u32),
 279         [OVS_KEY_ATTR_TUNNEL] = -1,
 280         [OVS_KEY_ATTR_MPLS] = sizeof(struct ovs_key_mpls),
 281 };
 282
 283 static bool is_all_zero(const u8 *fp, size_t size)
 284 {
 285         int i;
 286
 287         if (!fp)
 288                 return false;
 289
 290         for (i = 0; i < size; i++)
 291                 if (fp[i])
 292                         return false;
 293
 294         return true;
 295 }
 296
 297 static int __parse_flow_nlattrs(const struct nlattr *attr,
 298                                 const struct nlattr *a[],
 299                                 u64 *attrsp, bool nz)
 300 {
 301         const struct nlattr *nla;
 302         u64 attrs;
 303         int rem;
 304
 305         attrs = *attrsp;
 306         nla_for_each_nested(nla, attr, rem) {
 307                 u16 type = nla_type(nla);
 308                 int expected_len;
 309
 310                 if (type > OVS_KEY_ATTR_MAX) {
 311                         OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
 312                                   type, OVS_KEY_ATTR_MAX);
 313                         return -EINVAL;
 314                 }
 315
 316                 if (attrs & (1ULL << type)) {
 317                         OVS_NLERR("Duplicate key attribute (type %d).\n", type);
 318                         return -EINVAL;
 319                 }
 320
 321                 expected_len = ovs_key_lens[type];
 322                 if (nla_len(nla) != expected_len && expected_len != -1) {
 323                         OVS_NLERR("Key attribute has unexpected length (type=%d"
 324                                   ", length=%d, expected=%d).\n", type,
 325                                   nla_len(nla), expected_len);
 326                         return -EINVAL;
 327                 }
 328
 329                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 330                         attrs |= 1ULL << type;
 331                         a[type] = nla;
 332                 }
 333         }
 334         if (rem) {
 335                 OVS_NLERR("Message has %d unknown bytes.\n", rem);
 336                 return -EINVAL;
 337         }
 338
 339         *attrsp = attrs;
 340         return 0;
 341 }
 342
 343 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 344                                    const struct nlattr *a[], u64 *attrsp)
 345 {
 346         return __parse_flow_nlattrs(attr, a, attrsp, true);
 347 }
 348
 349 static int parse_flow_nlattrs(const struct nlattr *attr,
 350                               const struct nlattr *a[], u64 *attrsp)
 351 {
 352         return __parse_flow_nlattrs(attr, a, attrsp, false);
 353 }
 354
 355 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
 356                                 struct sw_flow_match *match, bool is_mask)
 357 {
 358         struct nlattr *a;
 359         int rem;
 360         bool ttl = false;
 361         __be16 tun_flags = 0;
 362
 363         nla_for_each_nested(a, attr, rem) {
 364                 int type = nla_type(a);
 365                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 366                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
 367                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
 368                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
 369                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
 370                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
 371                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
 372                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
 373                         [OVS_TUNNEL_KEY_ATTR_OAM] = 0,
 374                         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS] = -1,
 375                 };
 376
 377                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 378                         OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
 379                         type, OVS_TUNNEL_KEY_ATTR_MAX);
 380                         return -EINVAL;
 381                 }
 382
 383                 if (ovs_tunnel_key_lens[type] != nla_len(a) &&
 384                     ovs_tunnel_key_lens[type] != -1) {
 385                         OVS_NLERR("IPv4 tunnel attribute type has unexpected "
 386                                   " length (type=%d, length=%d, expected=%d).\n",
 387                                   type, nla_len(a), ovs_tunnel_key_lens[type]);
 388                         return -EINVAL;
 389                 }
 390
 391                 switch (type) {
 392                 case OVS_TUNNEL_KEY_ATTR_ID:
 393                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 394                                         nla_get_be64(a), is_mask);
 395                         tun_flags |= TUNNEL_KEY;
 396                         break;
 397                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 398                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
 399                                         nla_get_be32(a), is_mask);
 400                         break;
 401                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 402                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
 403                                         nla_get_be32(a), is_mask);
 404                         break;
 405                 case OVS_TUNNEL_KEY_ATTR_TOS:
 406                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
 407                                         nla_get_u8(a), is_mask);
 408                         break;
 409                 case OVS_TUNNEL_KEY_ATTR_TTL:
 410                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
 411                                         nla_get_u8(a), is_mask);
 412                         ttl = true;
 413                         break;
 414                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 415                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 416                         break;
 417                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 418                         tun_flags |= TUNNEL_CSUM;
 419                         break;
 420                 case OVS_TUNNEL_KEY_ATTR_OAM:
 421                         tun_flags |= TUNNEL_OAM;
 422                         break;
 423                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 424                         tun_flags |= TUNNEL_OPTIONS_PRESENT;
 425                         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 426                                 OVS_NLERR("Geneve option length exceeds "
 427                                           "maximum size (len %d, max %zu).\n",
 428                                           nla_len(a),
 429                                           sizeof(match->key->tun_opts));
 430                                 return -EINVAL;
 431                         }
 432
 433                         if (nla_len(a) % 4 != 0) {
 434                                 OVS_NLERR("Geneve option length is not "
 435                                           "a multiple of 4 (len %d).\n",
 436                                           nla_len(a));
 437                                 return -EINVAL;
 438                         }
 439
 440                         /* We need to record the length of the options passed
 441                          * down, otherwise packets with the same format but
 442                          * additional options will be silently matched.
 443                          */
 444                         if (!is_mask) {
 445                                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 446                                                 false);
 447                         } else {
 448                                 /* This is somewhat unusual because it looks at
 449                                  * both the key and mask while parsing the
 450                                  * attributes (and by extension assumes the key
 451                                  * is parsed first). Normally, we would verify
 452                                  * that each is the correct length and that the
 453                                  * attributes line up in the validate function.
 454                                  * However, that is difficult because this is
 455                                  * variable length and we won't have the
 456                                  * information later.
 457                                  */
 458                                 if (match->key->tun_opts_len != nla_len(a)) {
 459                                         OVS_NLERR("Geneve option key length (%d)"
 460                                            " is different from mask length (%d).",
 461                                            match->key->tun_opts_len, nla_len(a));
 462                                         return -EINVAL;
 463                                 }
 464
 465                                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff,
 466                                                 true);
 467                         }
 468
 469                         SW_FLOW_KEY_MEMCPY_OFFSET(match,
 470                                 (unsigned long)GENEVE_OPTS((struct sw_flow_key *)0,
 471                                                            nla_len(a)),
 472                                 nla_data(a), nla_len(a), is_mask);
 473                         break;
 474                 default:
 475                         OVS_NLERR("Unknown IPv4 tunnel attribute (%d).\n", type);
 476                         return -EINVAL;
 477                 }
 478         }
 479
 480         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 481
 482         if (rem > 0) {
 483                 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
 484                 return -EINVAL;
 485         }
 486
 487         if (!is_mask) {
 488                 if (!match->key->tun_key.ipv4_dst) {
 489                         OVS_NLERR("IPv4 tunnel destination address is zero.\n");
 490                         return -EINVAL;
 491                 }
 492
 493                 if (!ttl) {
 494                         OVS_NLERR("IPv4 tunnel TTL not specified.\n");
 495                         return -EINVAL;
 496                 }
 497         }
 498
 499         return 0;
 500 }
 501
 502 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
 503                               const struct ovs_key_ipv4_tunnel *output,
 504                               const struct geneve_opt *tun_opts,
 505                               int swkey_tun_opts_len)
 506 {
 507         struct nlattr *nla;
 508
 509         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 510         if (!nla)
 511                 return -EMSGSIZE;
 512
 513         if (output->tun_flags & TUNNEL_KEY &&
 514             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 515                 return -EMSGSIZE;
 516         if (output->ipv4_src &&
 517                 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
 518                 return -EMSGSIZE;
 519         if (output->ipv4_dst &&
 520                 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
 521                 return -EMSGSIZE;
 522         if (output->ipv4_tos &&
 523                 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
 524                 return -EMSGSIZE;
 525         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
 526                 return -EMSGSIZE;
 527         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 528                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 529                 return -EMSGSIZE;
 530         if ((output->tun_flags & TUNNEL_CSUM) &&
 531                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 532                 return -EMSGSIZE;
 533         if ((output->tun_flags & TUNNEL_OAM) &&
 534                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 535                 return -EMSGSIZE;
 536         if (tun_opts &&
 537             nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 538                     swkey_tun_opts_len, tun_opts))
 539                 return -EMSGSIZE;
 540
 541         nla_nest_end(skb, nla);
 542         return 0;
 543 }
 544
 545
 546 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
 547                                  const struct nlattr **a, bool is_mask)
 548 {
 549         if (*attrs & (1ULL << OVS_KEY_ATTR_DP_HASH)) {
 550                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 551
 552                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 553                 *attrs &= ~(1ULL << OVS_KEY_ATTR_DP_HASH);
 554         }
 555
 556         if (*attrs & (1ULL << OVS_KEY_ATTR_RECIRC_ID)) {
 557                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 558
 559                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 560                 *attrs &= ~(1ULL << OVS_KEY_ATTR_RECIRC_ID);
 561         }
 562
 563         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
 564                 SW_FLOW_KEY_PUT(match, phy.priority,
 565                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 566                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
 567         }
 568
 569         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
 570                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 571
 572                 if (is_mask) {
 573                         in_port = 0xffffffff; /* Always exact match in_port. */
 574                 } else if (in_port >= DP_MAX_PORTS) {
 575                         OVS_NLERR("Input port (%d) exceeds maximum allowable (%d).\n",
 576                                   in_port, DP_MAX_PORTS);
 577                         return -EINVAL;
 578                 }
 579
 580                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 581                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
 582         } else if (!is_mask) {
 583                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 584         }
 585
 586         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
 587                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 588
 589                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 590                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
 591         }
 592         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
 593                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 594                                          is_mask))
 595                         return -EINVAL;
 596                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
 597         }
 598         return 0;
 599 }
 600
 601 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
 602                                 const struct nlattr **a, bool is_mask)
 603 {
 604         int err;
 605
 606         err = metadata_from_nlattrs(match, &attrs, a, is_mask);
 607         if (err)
 608                 return err;
 609
 610         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
 611                 const struct ovs_key_ethernet *eth_key;
 612
 613                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 614                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 615                                 eth_key->eth_src, ETH_ALEN, is_mask);
 616                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 617                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 618                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
 619         }
 620
 621         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
 622                 __be16 tci;
 623
 624                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 625                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 626                         if (is_mask)
 627                                 OVS_NLERR("VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.\n");
 628                         else
 629                                 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
 630
 631                         return -EINVAL;
 632                 }
 633
 634                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 635                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
 636         } else if (!is_mask)
 637                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
 638
 639         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
 640                 __be16 eth_type;
 641
 642                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 643                 if (is_mask) {
 644                         /* Always exact match EtherType. */
 645                         eth_type = htons(0xffff);
 646                 } else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
 647                         OVS_NLERR("EtherType is less than minimum (type=%x, min=%x).\n",
 648                                         ntohs(eth_type), ETH_P_802_3_MIN);
 649                         return -EINVAL;
 650                 }
 651
 652                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 653                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
 654         } else if (!is_mask) {
 655                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 656         }
 657
 658         if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
 659                 const struct ovs_key_ipv4 *ipv4_key;
 660
 661                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 662                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 663                         OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
 664                                 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 665                         return -EINVAL;
 666                 }
 667                 SW_FLOW_KEY_PUT(match, ip.proto,
 668                                 ipv4_key->ipv4_proto, is_mask);
 669                 SW_FLOW_KEY_PUT(match, ip.tos,
 670                                 ipv4_key->ipv4_tos, is_mask);
 671                 SW_FLOW_KEY_PUT(match, ip.ttl,
 672                                 ipv4_key->ipv4_ttl, is_mask);
 673                 SW_FLOW_KEY_PUT(match, ip.frag,
 674                                 ipv4_key->ipv4_frag, is_mask);
 675                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 676                                 ipv4_key->ipv4_src, is_mask);
 677                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 678                                 ipv4_key->ipv4_dst, is_mask);
 679                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
 680         }
 681
 682         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
 683                 const struct ovs_key_ipv6 *ipv6_key;
 684
 685                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 686                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 687                         OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
 688                                 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 689                         return -EINVAL;
 690                 }
 691                 SW_FLOW_KEY_PUT(match, ipv6.label,
 692                                 ipv6_key->ipv6_label, is_mask);
 693                 SW_FLOW_KEY_PUT(match, ip.proto,
 694                                 ipv6_key->ipv6_proto, is_mask);
 695                 SW_FLOW_KEY_PUT(match, ip.tos,
 696                                 ipv6_key->ipv6_tclass, is_mask);
 697                 SW_FLOW_KEY_PUT(match, ip.ttl,
 698                                 ipv6_key->ipv6_hlimit, is_mask);
 699                 SW_FLOW_KEY_PUT(match, ip.frag,
 700                                 ipv6_key->ipv6_frag, is_mask);
 701                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
 702                                 ipv6_key->ipv6_src,
 703                                 sizeof(match->key->ipv6.addr.src),
 704                                 is_mask);
 705                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
 706                                 ipv6_key->ipv6_dst,
 707                                 sizeof(match->key->ipv6.addr.dst),
 708                                 is_mask);
 709
 710                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
 711         }
 712
 713         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
 714                 const struct ovs_key_arp *arp_key;
 715
 716                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
 717                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
 718                         OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
 719                                   arp_key->arp_op);
 720                         return -EINVAL;
 721                 }
 722
 723                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 724                                 arp_key->arp_sip, is_mask);
 725                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 726                         arp_key->arp_tip, is_mask);
 727                 SW_FLOW_KEY_PUT(match, ip.proto,
 728                                 ntohs(arp_key->arp_op), is_mask);
 729                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
 730                                 arp_key->arp_sha, ETH_ALEN, is_mask);
 731                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
 732                                 arp_key->arp_tha, ETH_ALEN, is_mask);
 733
 734                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
 735         }
 736
 737         if (attrs & (1ULL << OVS_KEY_ATTR_MPLS)) {
 738                 const struct ovs_key_mpls *mpls_key;
 739
 740                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
 741                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
 742                                 mpls_key->mpls_lse, is_mask);
 743
 744                 attrs &= ~(1ULL << OVS_KEY_ATTR_MPLS);
 745         }
 746
 747         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
 748                 const struct ovs_key_tcp *tcp_key;
 749
 750                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 751                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
 752                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
 753                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
 754         }
 755
 756         if (attrs & (1ULL << OVS_KEY_ATTR_TCP_FLAGS)) {
 757                 SW_FLOW_KEY_PUT(match, tp.flags,
 758                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
 759                                 is_mask);
 760                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP_FLAGS);
 761         }
 762
 763         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
 764                 const struct ovs_key_udp *udp_key;
 765
 766                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 767                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
 768                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
 769                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
 770         }
 771
 772         if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
 773                 const struct ovs_key_sctp *sctp_key;
 774
 775                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
 776                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
 777                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
 778                 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
 779         }
 780
 781         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
 782                 const struct ovs_key_icmp *icmp_key;
 783
 784                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 785                 SW_FLOW_KEY_PUT(match, tp.src,
 786                                 htons(icmp_key->icmp_type), is_mask);
 787                 SW_FLOW_KEY_PUT(match, tp.dst,
 788                                 htons(icmp_key->icmp_code), is_mask);
 789                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
 790         }
 791
 792         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
 793                 const struct ovs_key_icmpv6 *icmpv6_key;
 794
 795                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 796                 SW_FLOW_KEY_PUT(match, tp.src,
 797                                 htons(icmpv6_key->icmpv6_type), is_mask);
 798                 SW_FLOW_KEY_PUT(match, tp.dst,
 799                                 htons(icmpv6_key->icmpv6_code), is_mask);
 800                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
 801         }
 802
 803         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
 804                 const struct ovs_key_nd *nd_key;
 805
 806                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 807                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
 808                         nd_key->nd_target,
 809                         sizeof(match->key->ipv6.nd.target),
 810                         is_mask);
 811                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
 812                         nd_key->nd_sll, ETH_ALEN, is_mask);
 813                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
 814                                 nd_key->nd_tll, ETH_ALEN, is_mask);
 815                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
 816         }
 817
 818         if (attrs != 0) {
 819                 OVS_NLERR("Unknown key attributes (%llx).\n",
 820                           (unsigned long long)attrs);
 821                 return -EINVAL;
 822         }
 823
 824         return 0;
 825 }
 826
 827 static void nlattr_set(struct nlattr *attr, u8 val, bool is_attr_mask_key)
 828 {
 829         struct nlattr *nla;
 830         int rem;
 831
 832         /* The nlattr stream should already have been validated */
 833         nla_for_each_nested(nla, attr, rem) {
 834                 /* We assume that ovs_key_lens[type] == -1 means that type is a
 835                  * nested attribute
 836                  */
 837                 if (is_attr_mask_key && ovs_key_lens[nla_type(nla)] == -1)
 838                         nlattr_set(nla, val, false);
 839                 else
 840                         memset(nla_data(nla), val, nla_len(nla));
 841         }
 842 }
 843
 844 static void mask_set_nlattr(struct nlattr *attr, u8 val)
 845 {
 846         nlattr_set(attr, val, true);
 847 }
 848
 849 /**
 850  * ovs_nla_get_match - parses Netlink attributes into a flow key and
 851  * mask. In case the 'mask' is NULL, the flow is treated as exact match
 852  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
 853  * does not include any don't care bit.
 854  * @match: receives the extracted flow match information.
 855  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 856  * sequence. The fields should of the packet that triggered the creation
 857  * of this flow.
 858  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
 859  * attribute specifies the mask field of the wildcarded flow.
 860  */
 861 int ovs_nla_get_match(struct sw_flow_match *match,
 862                       const struct nlattr *key,
 863                       const struct nlattr *mask)
 864 {
 865         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 866         const struct nlattr *encap;
 867         struct nlattr *newmask = NULL;
 868         u64 key_attrs = 0;
 869         u64 mask_attrs = 0;
 870         bool encap_valid = false;
 871         int err;
 872
 873         err = parse_flow_nlattrs(key, a, &key_attrs);
 874         if (err)
 875                 return err;
 876
 877         if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
 878             (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
 879             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
 880                 __be16 tci;
 881
 882                 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
 883                       (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
 884                         OVS_NLERR("Invalid Vlan frame.\n");
 885                         return -EINVAL;
 886                 }
 887
 888                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
 889                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 890                 encap = a[OVS_KEY_ATTR_ENCAP];
 891                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
 892                 encap_valid = true;
 893
 894                 if (tci & htons(VLAN_TAG_PRESENT)) {
 895                         err = parse_flow_nlattrs(encap, a, &key_attrs);
 896                         if (err)
 897                                 return err;
 898                 } else if (!tci) {
 899                         /* Corner case for truncated 802.1Q header. */
 900                         if (nla_len(encap)) {
 901                                 OVS_NLERR("Truncated 802.1Q header has non-zero encap attribute.\n");
 902                                 return -EINVAL;
 903                         }
 904                 } else {
 905                         OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
 906                         return  -EINVAL;
 907                 }
 908         }
 909
 910         err = ovs_key_from_nlattrs(match, key_attrs, a, false);
 911         if (err)
 912                 return err;
 913
 914         if (match->mask && !mask) {
 915                 /* Create an exact match mask. We need to set to 0xff all the
 916                  * 'match->mask' fields that have been touched in 'match->key'.
 917                  * We cannot simply memset 'match->mask', because padding bytes
 918                  * and fields not specified in 'match->key' should be left to 0.
 919                  * Instead, we use a stream of netlink attributes, copied from
 920                  * 'key' and set to 0xff: ovs_key_from_nlattrs() will take care
 921                  * of filling 'match->mask' appropriately.
 922                  */
 923                 newmask = kmemdup(key, nla_total_size(nla_len(key)),
 924                                   GFP_KERNEL);
 925                 if (!newmask)
 926                         return -ENOMEM;
 927
 928                 mask_set_nlattr(newmask, 0xff);
 929
 930                 /* The userspace does not send tunnel attributes that are 0,
 931                  * but we should not wildcard them nonetheless. */
 932                 if (match->key->tun_key.ipv4_dst)
 933                         SW_FLOW_KEY_MEMSET_FIELD(match, tun_key, 0xff, true);
 934
 935                 mask = newmask;
 936         }
 937
 938         if (mask) {
 939                 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
 940                 if (err)
 941                         goto free_newmask;
 942
 943                 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
 944                         __be16 eth_type = 0;
 945                         __be16 tci = 0;
 946
 947                         if (!encap_valid) {
 948                                 OVS_NLERR("Encap mask attribute is set for non-VLAN frame.\n");
 949                                 err = -EINVAL;
 950                                 goto free_newmask;
 951                         }
 952
 953                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
 954                         if (a[OVS_KEY_ATTR_ETHERTYPE])
 955                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 956
 957                         if (eth_type == htons(0xffff)) {
 958                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
 959                                 encap = a[OVS_KEY_ATTR_ENCAP];
 960                                 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
 961                                 if (err)
 962                                         goto free_newmask;
 963                         } else {
 964                                 OVS_NLERR("VLAN frames must have an exact match on the TPID (mask=%x).\n",
 965                                                 ntohs(eth_type));
 966                                 err = -EINVAL;
 967                                 goto free_newmask;
 968                         }
 969
 970                         if (a[OVS_KEY_ATTR_VLAN])
 971                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 972
 973                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
 974                                 OVS_NLERR("VLAN tag present bit must have an exact match (tci_mask=%x).\n", ntohs(tci));
 975                                 err = -EINVAL;
 976                                 goto free_newmask;
 977                         }
 978                 }
 979
 980                 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
 981                 if (err)
 982                         goto free_newmask;
 983         }
 984
 985         if (!match_validate(match, key_attrs, mask_attrs))
 986                 err = -EINVAL;
 987
 988 free_newmask:
 989         kfree(newmask);
 990         return err;
 991 }
 992
 993 /**
 994  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
 995  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
 996  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 997  * sequence.
 998  *
 999  * This parses a series of Netlink attributes that form a flow key, which must
1000  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1001  * get the metadata, that is, the parts of the flow key that cannot be
1002  * extracted from the packet itself.
1003  */
1004 int ovs_nla_get_flow_metadata(const struct nlattr *attr,
1005                               struct sw_flow_key *key)
1006 {
1007         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1008         struct sw_flow_match match;
1009         u64 attrs = 0;
1010         int err;
1011
1012         err = parse_flow_nlattrs(attr, a, &attrs);
1013         if (err)
1014                 return -EINVAL;
1015
1016         memset(&match, 0, sizeof(match));
1017         match.key = key;
1018
1019         key->tun_opts_len = 0;
1020         memset(&key->tun_key, 0, sizeof(key->tun_key));
1021         key->phy.priority = 0;
1022         key->phy.skb_mark = 0;
1023         key->phy.in_port = DP_MAX_PORTS;
1024         key->ovs_flow_hash = 0;
1025         key->recirc_id = 0;
1026
1027         return metadata_from_nlattrs(&match, &attrs, a, false);
1028 }
1029
1030 int ovs_nla_put_flow(struct datapath *dp, const struct sw_flow_key *swkey,
1031                      const struct sw_flow_key *output, struct sk_buff *skb)
1032 {
1033         struct ovs_key_ethernet *eth_key;
1034         struct nlattr *nla, *encap;
1035         bool is_mask = (swkey != output);
1036
1037         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1038                 goto nla_put_failure;
1039
1040         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1041                 goto nla_put_failure;
1042
1043         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1044                 goto nla_put_failure;
1045
1046         if ((swkey->tun_key.ipv4_dst || is_mask)) {
1047                 const struct geneve_opt *opts = NULL;
1048
1049                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1050                         opts = GENEVE_OPTS(output, swkey->tun_opts_len);
1051
1052                 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1053                                         swkey->tun_opts_len))
1054                         goto nla_put_failure;
1055         }
1056
1057         if (swkey->phy.in_port == DP_MAX_PORTS) {
1058                 if (is_mask && (output->phy.in_port == 0xffff))
1059                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1060                                 goto nla_put_failure;
1061         } else {
1062                 u16 upper_u16;
1063                 upper_u16 = !is_mask ? 0 : 0xffff;
1064
1065                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1066                                 (upper_u16 << 16) | output->phy.in_port))
1067                         goto nla_put_failure;
1068         }
1069
1070         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1071                 goto nla_put_failure;
1072
1073         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1074         if (!nla)
1075                 goto nla_put_failure;
1076
1077         eth_key = nla_data(nla);
1078         ether_addr_copy(eth_key->eth_src, output->eth.src);
1079         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1080
1081         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1082                 __be16 eth_type;
1083                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1084                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1085                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1086                         goto nla_put_failure;
1087                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1088                 if (!swkey->eth.tci)
1089                         goto unencap;
1090         } else
1091                 encap = NULL;
1092
1093         if (swkey->eth.type == htons(ETH_P_802_2)) {
1094                 /*
1095                  * Ethertype 802.2 is represented in the netlink with omitted
1096                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1097                  * 0xffff in the mask attribute.  Ethertype can also
1098                  * be wildcarded.
1099                  */
1100                 if (is_mask && output->eth.type)
1101                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1102                                                 output->eth.type))
1103                                 goto nla_put_failure;
1104                 goto unencap;
1105         }
1106
1107         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1108                 goto nla_put_failure;
1109
1110         if (swkey->eth.type == htons(ETH_P_IP)) {
1111                 struct ovs_key_ipv4 *ipv4_key;
1112
1113                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1114                 if (!nla)
1115                         goto nla_put_failure;
1116                 ipv4_key = nla_data(nla);
1117                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1118                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1119                 ipv4_key->ipv4_proto = output->ip.proto;
1120                 ipv4_key->ipv4_tos = output->ip.tos;
1121                 ipv4_key->ipv4_ttl = output->ip.ttl;
1122                 ipv4_key->ipv4_frag = output->ip.frag;
1123         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1124                 struct ovs_key_ipv6 *ipv6_key;
1125
1126                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1127                 if (!nla)
1128                         goto nla_put_failure;
1129                 ipv6_key = nla_data(nla);
1130                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1131                                 sizeof(ipv6_key->ipv6_src));
1132                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1133                                 sizeof(ipv6_key->ipv6_dst));
1134                 ipv6_key->ipv6_label = output->ipv6.label;
1135                 ipv6_key->ipv6_proto = output->ip.proto;
1136                 ipv6_key->ipv6_tclass = output->ip.tos;
1137                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1138                 ipv6_key->ipv6_frag = output->ip.frag;
1139         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1140                    swkey->eth.type == htons(ETH_P_RARP)) {
1141                 struct ovs_key_arp *arp_key;
1142
1143                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1144                 if (!nla)
1145                         goto nla_put_failure;
1146                 arp_key = nla_data(nla);
1147                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1148                 arp_key->arp_sip = output->ipv4.addr.src;
1149                 arp_key->arp_tip = output->ipv4.addr.dst;
1150                 arp_key->arp_op = htons(output->ip.proto);
1151                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1152                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1153         } else if (eth_p_mpls(swkey->eth.type)) {
1154                 struct ovs_key_mpls *mpls_key;
1155
1156                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1157                 if (!nla)
1158                         goto nla_put_failure;
1159                 mpls_key = nla_data(nla);
1160                 mpls_key->mpls_lse = output->mpls.top_lse;
1161         }
1162
1163         if ((swkey->eth.type == htons(ETH_P_IP) ||
1164              swkey->eth.type == htons(ETH_P_IPV6)) &&
1165              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1166
1167                 if (swkey->ip.proto == IPPROTO_TCP) {
1168                         struct ovs_key_tcp *tcp_key;
1169
1170                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1171                         if (!nla)
1172                                 goto nla_put_failure;
1173                         tcp_key = nla_data(nla);
1174                         tcp_key->tcp_src = output->tp.src;
1175                         tcp_key->tcp_dst = output->tp.dst;
1176                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1177                                          output->tp.flags))
1178                                 goto nla_put_failure;
1179                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1180                         struct ovs_key_udp *udp_key;
1181
1182                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1183                         if (!nla)
1184                                 goto nla_put_failure;
1185                         udp_key = nla_data(nla);
1186                         udp_key->udp_src = output->tp.src;
1187                         udp_key->udp_dst = output->tp.dst;
1188                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1189                         struct ovs_key_sctp *sctp_key;
1190
1191                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1192                         if (!nla)
1193                                 goto nla_put_failure;
1194                         sctp_key = nla_data(nla);
1195                         sctp_key->sctp_src = output->tp.src;
1196                         sctp_key->sctp_dst = output->tp.dst;
1197                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1198                            swkey->ip.proto == IPPROTO_ICMP) {
1199                         struct ovs_key_icmp *icmp_key;
1200
1201                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1202                         if (!nla)
1203                                 goto nla_put_failure;
1204                         icmp_key = nla_data(nla);
1205                         icmp_key->icmp_type = ntohs(output->tp.src);
1206                         icmp_key->icmp_code = ntohs(output->tp.dst);
1207                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1208                            swkey->ip.proto == IPPROTO_ICMPV6) {
1209                         struct ovs_key_icmpv6 *icmpv6_key;
1210
1211                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1212                                                 sizeof(*icmpv6_key));
1213                         if (!nla)
1214                                 goto nla_put_failure;
1215                         icmpv6_key = nla_data(nla);
1216                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1217                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1218
1219                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1220                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1221                                 struct ovs_key_nd *nd_key;
1222
1223                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1224                                 if (!nla)
1225                                         goto nla_put_failure;
1226                                 nd_key = nla_data(nla);
1227                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1228                                                         sizeof(nd_key->nd_target));
1229                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1230                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1231                         }
1232                 }
1233         }
1234
1235 unencap:
1236         if (encap)
1237                 nla_nest_end(skb, encap);
1238
1239         return 0;
1240
1241 nla_put_failure:
1242         return -EMSGSIZE;
1243 }
1244
1245 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1246
1247 struct sw_flow_actions *ovs_nla_alloc_flow_actions(int size)
1248 {
1249         struct sw_flow_actions *sfa;
1250
1251         if (size > MAX_ACTIONS_BUFSIZE) {
1252                 OVS_NLERR("Flow action size (%u bytes) exceeds maximum "
1253                           "(%u bytes)\n", size, MAX_ACTIONS_BUFSIZE);
1254                 return ERR_PTR(-EINVAL);
1255         }
1256
1257         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1258         if (!sfa)
1259                 return ERR_PTR(-ENOMEM);
1260
1261         sfa->actions_len = 0;
1262         return sfa;
1263 }
1264
1265 /* RCU callback used by ovs_nla_free_flow_actions. */
1266 static void rcu_free_acts_callback(struct rcu_head *rcu)
1267 {
1268         struct sw_flow_actions *sf_acts = container_of(rcu,
1269                         struct sw_flow_actions, rcu);
1270         kfree(sf_acts);
1271 }
1272
1273 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1274  * The caller must hold rcu_read_lock for this to be sensible. */
1275 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1276 {
1277         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
1278 }
1279
1280 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1281                                        int attr_len)
1282 {
1283
1284         struct sw_flow_actions *acts;
1285         int new_acts_size;
1286         int req_size = NLA_ALIGN(attr_len);
1287         int next_offset = offsetof(struct sw_flow_actions, actions) +
1288                                         (*sfa)->actions_len;
1289
1290         if (req_size <= (ksize(*sfa) - next_offset))
1291                 goto out;
1292
1293         new_acts_size = ksize(*sfa) * 2;
1294
1295         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1296                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1297                         return ERR_PTR(-EMSGSIZE);
1298                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1299         }
1300
1301         acts = ovs_nla_alloc_flow_actions(new_acts_size);
1302         if (IS_ERR(acts))
1303                 return (void *)acts;
1304
1305         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1306         acts->actions_len = (*sfa)->actions_len;
1307         kfree(*sfa);
1308         *sfa = acts;
1309
1310 out:
1311         (*sfa)->actions_len += req_size;
1312         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1313 }
1314
1315 static struct nlattr *__add_action(struct sw_flow_actions **sfa, int attrtype,
1316                                    void *data, int len)
1317 {
1318         struct nlattr *a;
1319
1320         a = reserve_sfa_size(sfa, nla_attr_size(len));
1321         if (IS_ERR(a))
1322                 return a;
1323
1324         a->nla_type = attrtype;
1325         a->nla_len = nla_attr_size(len);
1326
1327         if (data)
1328                 memcpy(nla_data(a), data, len);
1329         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1330
1331         return a;
1332 }
1333
1334 static int add_action(struct sw_flow_actions **sfa, int attrtype,
1335                       void *data, int len)
1336 {
1337         struct nlattr *a;
1338
1339         a = __add_action(sfa, attrtype, data, len);
1340         if (IS_ERR(a))
1341                 return PTR_ERR(a);
1342
1343         return 0;
1344 }
1345
1346 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1347                                           int attrtype)
1348 {
1349         int used = (*sfa)->actions_len;
1350         int err;
1351
1352         err = add_action(sfa, attrtype, NULL, 0);
1353         if (err)
1354                 return err;
1355
1356         return used;
1357 }
1358
1359 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1360                                          int st_offset)
1361 {
1362         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1363                                                                st_offset);
1364
1365         a->nla_len = sfa->actions_len - st_offset;
1366 }
1367
1368 static int ovs_nla_copy_actions__(const struct nlattr *attr,
1369                                   const struct sw_flow_key *key,
1370                                   int depth, struct sw_flow_actions **sfa,
1371                                   __be16 eth_type, __be16 vlan_tci);
1372
1373 static int validate_and_copy_sample(const struct nlattr *attr,
1374                                     const struct sw_flow_key *key, int depth,
1375                                     struct sw_flow_actions **sfa,
1376                                     __be16 eth_type, __be16 vlan_tci)
1377 {
1378         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1379         const struct nlattr *probability, *actions;
1380         const struct nlattr *a;
1381         int rem, start, err, st_acts;
1382
1383         memset(attrs, 0, sizeof(attrs));
1384         nla_for_each_nested(a, attr, rem) {
1385                 int type = nla_type(a);
1386                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1387                         return -EINVAL;
1388                 attrs[type] = a;
1389         }
1390         if (rem)
1391                 return -EINVAL;
1392
1393         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1394         if (!probability || nla_len(probability) != sizeof(u32))
1395                 return -EINVAL;
1396
1397         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1398         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1399                 return -EINVAL;
1400
1401         /* validation done, copy sample action. */
1402         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
1403         if (start < 0)
1404                 return start;
1405         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1406                          nla_data(probability), sizeof(u32));
1407         if (err)
1408                 return err;
1409         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
1410         if (st_acts < 0)
1411                 return st_acts;
1412
1413         err = ovs_nla_copy_actions__(actions, key, depth + 1, sfa,
1414                                      eth_type, vlan_tci);
1415         if (err)
1416                 return err;
1417
1418         add_nested_action_end(*sfa, st_acts);
1419         add_nested_action_end(*sfa, start);
1420
1421         return 0;
1422 }
1423
1424 static int validate_tp_port(const struct sw_flow_key *flow_key,
1425                             __be16 eth_type)
1426 {
1427         if ((eth_type == htons(ETH_P_IP) || eth_type == htons(ETH_P_IPV6)) &&
1428             (flow_key->tp.src || flow_key->tp.dst))
1429                 return 0;
1430
1431         return -EINVAL;
1432 }
1433
1434 void ovs_match_init(struct sw_flow_match *match,
1435                     struct sw_flow_key *key,
1436                     struct sw_flow_mask *mask)
1437 {
1438         memset(match, 0, sizeof(*match));
1439         match->key = key;
1440         match->mask = mask;
1441
1442         memset(key, 0, sizeof(*key));
1443
1444         if (mask) {
1445                 memset(&mask->key, 0, sizeof(mask->key));
1446                 mask->range.start = mask->range.end = 0;
1447         }
1448 }
1449
1450 static int validate_and_copy_set_tun(const struct nlattr *attr,
1451                                      struct sw_flow_actions **sfa)
1452 {
1453         struct sw_flow_match match;
1454         struct sw_flow_key key;
1455         struct ovs_tunnel_info *tun_info;
1456         struct nlattr *a;
1457         int err, start;
1458
1459         ovs_match_init(&match, &key, NULL);
1460         err = ipv4_tun_from_nlattr(nla_data(attr), &match, false);
1461         if (err)
1462                 return err;
1463
1464         if (key.tun_opts_len) {
1465                 struct geneve_opt *option = GENEVE_OPTS(&key,
1466                                                         key.tun_opts_len);
1467                 int opts_len = key.tun_opts_len;
1468                 bool crit_opt = false;
1469
1470                 while (opts_len > 0) {
1471                         int len;
1472
1473                         if (opts_len < sizeof(*option))
1474                                 return -EINVAL;
1475
1476                         len = sizeof(*option) + option->length * 4;
1477                         if (len > opts_len)
1478                                 return -EINVAL;
1479
1480                         crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1481
1482                         option = (struct geneve_opt *)((u8 *)option + len);
1483                         opts_len -= len;
1484                 };
1485
1486                 key.tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1487         };
1488
1489         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
1490         if (start < 0)
1491                 return start;
1492
1493         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1494                         sizeof(*tun_info) + key.tun_opts_len);
1495         if (IS_ERR(a))
1496                 return PTR_ERR(a);
1497
1498         tun_info = nla_data(a);
1499         tun_info->tunnel = key.tun_key;
1500         tun_info->options_len = key.tun_opts_len;
1501
1502         if (tun_info->options_len) {
1503                 /* We need to store the options in the action itself since
1504                  * everything else will go away after flow setup. We can append
1505                  * it to tun_info and then point there.
1506                  */
1507                 tun_info->options = (struct geneve_opt *)(tun_info + 1);
1508                 memcpy(tun_info->options, GENEVE_OPTS(&key, key.tun_opts_len),
1509                         key.tun_opts_len);
1510         } else {
1511                 tun_info->options = NULL;
1512         }
1513
1514         add_nested_action_end(*sfa, start);
1515
1516         return err;
1517 }
1518
1519 static int validate_set(const struct nlattr *a,
1520                         const struct sw_flow_key *flow_key,
1521                         struct sw_flow_actions **sfa,
1522                         bool *set_tun, __be16 eth_type)
1523 {
1524         const struct nlattr *ovs_key = nla_data(a);
1525         int key_type = nla_type(ovs_key);
1526
1527         /* There can be only one key in a action */
1528         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1529                 return -EINVAL;
1530
1531         if (key_type > OVS_KEY_ATTR_MAX ||
1532             (ovs_key_lens[key_type] != nla_len(ovs_key) &&
1533              ovs_key_lens[key_type] != -1))
1534                 return -EINVAL;
1535
1536         switch (key_type) {
1537         const struct ovs_key_ipv4 *ipv4_key;
1538         const struct ovs_key_ipv6 *ipv6_key;
1539         int err;
1540
1541         case OVS_KEY_ATTR_PRIORITY:
1542         case OVS_KEY_ATTR_SKB_MARK:
1543         case OVS_KEY_ATTR_ETHERNET:
1544                 break;
1545
1546         case OVS_KEY_ATTR_TUNNEL:
1547                 *set_tun = true;
1548                 err = validate_and_copy_set_tun(a, sfa);
1549                 if (err)
1550                         return err;
1551                 break;
1552
1553         case OVS_KEY_ATTR_IPV4:
1554                 if (eth_type != htons(ETH_P_IP))
1555                         return -EINVAL;
1556
1557                 if (!flow_key->ip.proto)
1558                         return -EINVAL;
1559
1560                 ipv4_key = nla_data(ovs_key);
1561                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1562                         return -EINVAL;
1563
1564                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1565                         return -EINVAL;
1566
1567                 break;
1568
1569         case OVS_KEY_ATTR_IPV6:
1570                 if (eth_type != htons(ETH_P_IPV6))
1571                         return -EINVAL;
1572
1573                 if (!flow_key->ip.proto)
1574                         return -EINVAL;
1575
1576                 ipv6_key = nla_data(ovs_key);
1577                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1578                         return -EINVAL;
1579
1580                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1581                         return -EINVAL;
1582
1583                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
1584                         return -EINVAL;
1585
1586                 break;
1587
1588         case OVS_KEY_ATTR_TCP:
1589                 if (flow_key->ip.proto != IPPROTO_TCP)
1590                         return -EINVAL;
1591
1592                 return validate_tp_port(flow_key, eth_type);
1593
1594         case OVS_KEY_ATTR_UDP:
1595                 if (flow_key->ip.proto != IPPROTO_UDP)
1596                         return -EINVAL;
1597
1598                 return validate_tp_port(flow_key, eth_type);
1599
1600         case OVS_KEY_ATTR_MPLS:
1601                 if (!eth_p_mpls(eth_type))
1602                         return -EINVAL;
1603                 break;
1604
1605         case OVS_KEY_ATTR_SCTP:
1606                 if (flow_key->ip.proto != IPPROTO_SCTP)
1607                         return -EINVAL;
1608
1609                 return validate_tp_port(flow_key, eth_type);
1610
1611         default:
1612                 return -EINVAL;
1613         }
1614
1615         return 0;
1616 }
1617
1618 static int validate_userspace(const struct nlattr *attr)
1619 {
1620         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
1621                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
1622                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
1623         };
1624         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
1625         int error;
1626
1627         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
1628                                  attr, userspace_policy);
1629         if (error)
1630                 return error;
1631
1632         if (!a[OVS_USERSPACE_ATTR_PID] ||
1633             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
1634                 return -EINVAL;
1635
1636         return 0;
1637 }
1638
1639 static int copy_action(const struct nlattr *from,
1640                        struct sw_flow_actions **sfa)
1641 {
1642         int totlen = NLA_ALIGN(from->nla_len);
1643         struct nlattr *to;
1644
1645         to = reserve_sfa_size(sfa, from->nla_len);
1646         if (IS_ERR(to))
1647                 return PTR_ERR(to);
1648
1649         memcpy(to, from, totlen);
1650         return 0;
1651 }
1652
1653 static int ovs_nla_copy_actions__(const struct nlattr *attr,
1654                                   const struct sw_flow_key *key,
1655                                   int depth, struct sw_flow_actions **sfa,
1656                                   __be16 eth_type, __be16 vlan_tci)
1657 {
1658         const struct nlattr *a;
1659         int rem, err;
1660
1661         if (depth >= SAMPLE_ACTION_DEPTH)
1662                 return -EOVERFLOW;
1663
1664         nla_for_each_nested(a, attr, rem) {
1665                 /* Expected argument lengths, (u32)-1 for variable length. */
1666                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
1667                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
1668                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
1669                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
1670                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
1671                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
1672                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
1673                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
1674                         [OVS_ACTION_ATTR_SET] = (u32)-1,
1675                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
1676                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash)
1677                 };
1678                 const struct ovs_action_push_vlan *vlan;
1679                 int type = nla_type(a);
1680                 bool skip_copy;
1681
1682                 if (type > OVS_ACTION_ATTR_MAX ||
1683                     (action_lens[type] != nla_len(a) &&
1684                      action_lens[type] != (u32)-1))
1685                         return -EINVAL;
1686
1687                 skip_copy = false;
1688                 switch (type) {
1689                 case OVS_ACTION_ATTR_UNSPEC:
1690                         return -EINVAL;
1691
1692                 case OVS_ACTION_ATTR_USERSPACE:
1693                         err = validate_userspace(a);
1694                         if (err)
1695                                 return err;
1696                         break;
1697
1698                 case OVS_ACTION_ATTR_OUTPUT:
1699                         if (nla_get_u32(a) >= DP_MAX_PORTS)
1700                                 return -EINVAL;
1701                         break;
1702
1703                 case OVS_ACTION_ATTR_HASH: {
1704                         const struct ovs_action_hash *act_hash = nla_data(a);
1705
1706                         switch (act_hash->hash_alg) {
1707                         case OVS_HASH_ALG_L4:
1708                                 break;
1709                         default:
1710                                 return  -EINVAL;
1711                         }
1712
1713                         break;
1714                 }
1715
1716                 case OVS_ACTION_ATTR_POP_VLAN:
1717                         vlan_tci = htons(0);
1718                         break;
1719
1720                 case OVS_ACTION_ATTR_PUSH_VLAN:
1721                         vlan = nla_data(a);
1722                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
1723                                 return -EINVAL;
1724                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
1725                                 return -EINVAL;
1726                         vlan_tci = vlan->vlan_tci;
1727                         break;
1728
1729                 case OVS_ACTION_ATTR_RECIRC:
1730                         break;
1731
1732                 case OVS_ACTION_ATTR_PUSH_MPLS: {
1733                         const struct ovs_action_push_mpls *mpls = nla_data(a);
1734
1735                         if (!eth_p_mpls(mpls->mpls_ethertype))
1736                                 return -EINVAL;
1737                         /* Prohibit push MPLS other than to a white list
1738                          * for packets that have a known tag order.
1739                          */
1740                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
1741                             (eth_type != htons(ETH_P_IP) &&
1742                              eth_type != htons(ETH_P_IPV6) &&
1743                              eth_type != htons(ETH_P_ARP) &&
1744                              eth_type != htons(ETH_P_RARP) &&
1745                              !eth_p_mpls(eth_type)))
1746                                 return -EINVAL;
1747                         eth_type = mpls->mpls_ethertype;
1748                         break;
1749                 }
1750
1751                 case OVS_ACTION_ATTR_POP_MPLS:
1752                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
1753                             !eth_p_mpls(eth_type))
1754                                 return -EINVAL;
1755
1756                         /* Disallow subsequent L2.5+ set and mpls_pop actions
1757                          * as there is no check here to ensure that the new
1758                          * eth_type is valid and thus set actions could
1759                          * write off the end of the packet or otherwise
1760                          * corrupt it.
1761                          *
1762                          * Support for these actions is planned using packet
1763                          * recirculation.
1764                          */
1765                         eth_type = htons(0);
1766                         break;
1767
1768                 case OVS_ACTION_ATTR_SET:
1769                         err = validate_set(a, key, sfa, &skip_copy, eth_type);
1770                         if (err)
1771                                 return err;
1772                         break;
1773
1774                 case OVS_ACTION_ATTR_SAMPLE:
1775                         err = validate_and_copy_sample(a, key, depth, sfa,
1776                                                        eth_type, vlan_tci);
1777                         if (err)
1778                                 return err;
1779                         skip_copy = true;
1780                         break;
1781
1782                 default:
1783                         return -EINVAL;
1784                 }
1785                 if (!skip_copy) {
1786                         err = copy_action(a, sfa);
1787                         if (err)
1788                                 return err;
1789                 }
1790         }
1791
1792         if (rem > 0)
1793                 return -EINVAL;
1794
1795         return 0;
1796 }
1797
1798 int ovs_nla_copy_actions(const struct nlattr *attr,
1799                          const struct sw_flow_key *key,
1800                          struct sw_flow_actions **sfa)
1801 {
1802         return ovs_nla_copy_actions__(attr, key, 0, sfa, key->eth.type,
1803                                       key->eth.tci);
1804 }
1805
1806 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1807 {
1808         const struct nlattr *a;
1809         struct nlattr *start;
1810         int err = 0, rem;
1811
1812         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1813         if (!start)
1814                 return -EMSGSIZE;
1815
1816         nla_for_each_nested(a, attr, rem) {
1817                 int type = nla_type(a);
1818                 struct nlattr *st_sample;
1819
1820                 switch (type) {
1821                 case OVS_SAMPLE_ATTR_PROBABILITY:
1822                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
1823                                     sizeof(u32), nla_data(a)))
1824                                 return -EMSGSIZE;
1825                         break;
1826                 case OVS_SAMPLE_ATTR_ACTIONS:
1827                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1828                         if (!st_sample)
1829                                 return -EMSGSIZE;
1830                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
1831                         if (err)
1832                                 return err;
1833                         nla_nest_end(skb, st_sample);
1834                         break;
1835                 }
1836         }
1837
1838         nla_nest_end(skb, start);
1839         return err;
1840 }
1841
1842 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1843 {
1844         const struct nlattr *ovs_key = nla_data(a);
1845         int key_type = nla_type(ovs_key);
1846         struct nlattr *start;
1847         int err;
1848
1849         switch (key_type) {
1850         case OVS_KEY_ATTR_TUNNEL_INFO: {
1851                 struct ovs_tunnel_info *tun_info = nla_data(ovs_key);
1852
1853                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1854                 if (!start)
1855                         return -EMSGSIZE;
1856
1857                 err = ipv4_tun_to_nlattr(skb, &tun_info->tunnel,
1858                                          tun_info->options_len ?
1859                                                 tun_info->options : NULL,
1860                                          tun_info->options_len);
1861                 if (err)
1862                         return err;
1863                 nla_nest_end(skb, start);
1864                 break;
1865         }
1866         default:
1867                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1868                         return -EMSGSIZE;
1869                 break;
1870         }
1871
1872         return 0;
1873 }
1874
1875 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
1876 {
1877         const struct nlattr *a;
1878         int rem, err;
1879
1880         nla_for_each_attr(a, attr, len, rem) {
1881                 int type = nla_type(a);
1882
1883                 switch (type) {
1884                 case OVS_ACTION_ATTR_SET:
1885                         err = set_action_to_attr(a, skb);
1886                         if (err)
1887                                 return err;
1888                         break;
1889
1890                 case OVS_ACTION_ATTR_SAMPLE:
1891                         err = sample_action_to_attr(a, skb);
1892                         if (err)
1893                                 return err;
1894                         break;
1895                 default:
1896                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1897                                 return -EMSGSIZE;
1898                         break;
1899                 }
1900         }
1901
1902         return 0;
1903 }