net/openvswitch/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 <linux/uaccess.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <net/llc_pdu.h>
  29 #include <linux/kernel.h>
  30 #include <linux/jhash.h>
  31 #include <linux/jiffies.h>
  32 #include <linux/llc.h>
  33 #include <linux/module.h>
  34 #include <linux/in.h>
  35 #include <linux/rcupdate.h>
  36 #include <linux/if_arp.h>
  37 #include <linux/ip.h>
  38 #include <linux/ipv6.h>
  39 #include <linux/sctp.h>
  40 #include <linux/tcp.h>
  41 #include <linux/udp.h>
  42 #include <linux/icmp.h>
  43 #include <linux/icmpv6.h>
  44 #include <linux/rculist.h>
  45 #include <net/geneve.h>
  46 #include <net/ip.h>
  47 #include <net/ipv6.h>
  48 #include <net/ndisc.h>
  49 #include <net/mpls.h>
  50
  51 #include "flow_netlink.h"
  52
  53 static void update_range(struct sw_flow_match *match,
  54                          size_t offset, size_t size, bool is_mask)
  55 {
  56         struct sw_flow_key_range *range;
  57         size_t start = rounddown(offset, sizeof(long));
  58         size_t end = roundup(offset + size, sizeof(long));
  59
  60         if (!is_mask)
  61                 range = &match->range;
  62         else
  63                 range = &match->mask->range;
  64
  65         if (range->start == range->end) {
  66                 range->start = start;
  67                 range->end = end;
  68                 return;
  69         }
  70
  71         if (range->start > start)
  72                 range->start = start;
  73
  74         if (range->end < end)
  75                 range->end = end;
  76 }
  77
  78 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  79         do { \
  80                 update_range(match, offsetof(struct sw_flow_key, field),    \
  81                              sizeof((match)->key->field), is_mask);         \
  82                 if (is_mask)                                                \
  83                         (match)->mask->key.field = value;                   \
  84                 else                                                        \
  85                         (match)->key->field = value;                        \
  86         } while (0)
  87
  88 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
  89         do {                                                                \
  90                 update_range(match, offset, len, is_mask);                  \
  91                 if (is_mask)                                                \
  92                         memcpy((u8 *)&(match)->mask->key + offset, value_p, \
  93                                len);                                       \
  94                 else                                                        \
  95                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
  96         } while (0)
  97
  98 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
  99         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 100                                   value_p, len, is_mask)
 101
 102 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 103         do {                                                                \
 104                 update_range(match, offsetof(struct sw_flow_key, field),    \
 105                              sizeof((match)->key->field), is_mask);         \
 106                 if (is_mask)                                                \
 107                         memset((u8 *)&(match)->mask->key.field, value,      \
 108                                sizeof((match)->mask->key.field));           \
 109                 else                                                        \
 110                         memset((u8 *)&(match)->key->field, value,           \
 111                                sizeof((match)->key->field));                \
 112         } while (0)
 113
 114 static bool match_validate(const struct sw_flow_match *match,
 115                            u64 key_attrs, u64 mask_attrs, bool log)
 116 {
 117         u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
 118         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 119
 120         /* The following mask attributes allowed only if they
 121          * pass the validation tests. */
 122         mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
 123                         | (1 << OVS_KEY_ATTR_IPV6)
 124                         | (1 << OVS_KEY_ATTR_TCP)
 125                         | (1 << OVS_KEY_ATTR_TCP_FLAGS)
 126                         | (1 << OVS_KEY_ATTR_UDP)
 127                         | (1 << OVS_KEY_ATTR_SCTP)
 128                         | (1 << OVS_KEY_ATTR_ICMP)
 129                         | (1 << OVS_KEY_ATTR_ICMPV6)
 130                         | (1 << OVS_KEY_ATTR_ARP)
 131                         | (1 << OVS_KEY_ATTR_ND)
 132                         | (1 << OVS_KEY_ATTR_MPLS));
 133
 134         /* Always allowed mask fields. */
 135         mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
 136                        | (1 << OVS_KEY_ATTR_IN_PORT)
 137                        | (1 << OVS_KEY_ATTR_ETHERTYPE));
 138
 139         /* Check key attributes. */
 140         if (match->key->eth.type == htons(ETH_P_ARP)
 141                         || match->key->eth.type == htons(ETH_P_RARP)) {
 142                 key_expected |= 1 << OVS_KEY_ATTR_ARP;
 143                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 144                         mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
 145         }
 146
 147         if (eth_p_mpls(match->key->eth.type)) {
 148                 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
 149                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 150                         mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
 151         }
 152
 153         if (match->key->eth.type == htons(ETH_P_IP)) {
 154                 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
 155                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 156                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
 157
 158                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 159                         if (match->key->ip.proto == IPPROTO_UDP) {
 160                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 161                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 162                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 163                         }
 164
 165                         if (match->key->ip.proto == IPPROTO_SCTP) {
 166                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 167                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 168                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 169                         }
 170
 171                         if (match->key->ip.proto == IPPROTO_TCP) {
 172                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 173                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 174                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 175                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 176                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 177                                 }
 178                         }
 179
 180                         if (match->key->ip.proto == IPPROTO_ICMP) {
 181                                 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
 182                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 183                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
 184                         }
 185                 }
 186         }
 187
 188         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 189                 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
 190                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 191                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
 192
 193                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 194                         if (match->key->ip.proto == IPPROTO_UDP) {
 195                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 196                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 197                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 198                         }
 199
 200                         if (match->key->ip.proto == IPPROTO_SCTP) {
 201                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 202                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 203                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 204                         }
 205
 206                         if (match->key->ip.proto == IPPROTO_TCP) {
 207                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 208                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 209                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 210                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 211                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 212                                 }
 213                         }
 214
 215                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 216                                 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
 217                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 218                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
 219
 220                                 if (match->key->tp.src ==
 221                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 222                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 223                                         key_expected |= 1 << OVS_KEY_ATTR_ND;
 224                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 225                                                 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
 226                                 }
 227                         }
 228                 }
 229         }
 230
 231         if ((key_attrs & key_expected) != key_expected) {
 232                 /* Key attributes check failed. */
 233                 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
 234                           (unsigned long long)key_attrs,
 235                           (unsigned long long)key_expected);
 236                 return false;
 237         }
 238
 239         if ((mask_attrs & mask_allowed) != mask_attrs) {
 240                 /* Mask attributes check failed. */
 241                 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
 242                           (unsigned long long)mask_attrs,
 243                           (unsigned long long)mask_allowed);
 244                 return false;
 245         }
 246
 247         return true;
 248 }
 249
 250 size_t ovs_tun_key_attr_size(void)
 251 {
 252         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 253          * updating this function.
 254          */
 255         return    nla_total_size(8)    /* OVS_TUNNEL_KEY_ATTR_ID */
 256                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 257                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 258                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
 259                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
 260                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 261                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
 262                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
 263                 + nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 264                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
 265                 + nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
 266 }
 267
 268 size_t ovs_key_attr_size(void)
 269 {
 270         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 271          * updating this function.
 272          */
 273         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 22);
 274
 275         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 276                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 277                   + ovs_tun_key_attr_size()
 278                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 279                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 280                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 281                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 282                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 283                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 284                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 285                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 286                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 287                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 288                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 289                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 290 }
 291
 292 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 293 static const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 294         [OVS_KEY_ATTR_ENCAP] = -1,
 295         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 296         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 297         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
 298         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 299         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 300         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 301         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 302         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 303         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 304         [OVS_KEY_ATTR_TCP_FLAGS] = sizeof(__be16),
 305         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 306         [OVS_KEY_ATTR_SCTP] = sizeof(struct ovs_key_sctp),
 307         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 308         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 309         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 310         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 311         [OVS_KEY_ATTR_RECIRC_ID] = sizeof(u32),
 312         [OVS_KEY_ATTR_DP_HASH] = sizeof(u32),
 313         [OVS_KEY_ATTR_TUNNEL] = -1,
 314         [OVS_KEY_ATTR_MPLS] = sizeof(struct ovs_key_mpls),
 315 };
 316
 317 static bool is_all_zero(const u8 *fp, size_t size)
 318 {
 319         int i;
 320
 321         if (!fp)
 322                 return false;
 323
 324         for (i = 0; i < size; i++)
 325                 if (fp[i])
 326                         return false;
 327
 328         return true;
 329 }
 330
 331 static int __parse_flow_nlattrs(const struct nlattr *attr,
 332                                 const struct nlattr *a[],
 333                                 u64 *attrsp, bool log, bool nz)
 334 {
 335         const struct nlattr *nla;
 336         u64 attrs;
 337         int rem;
 338
 339         attrs = *attrsp;
 340         nla_for_each_nested(nla, attr, rem) {
 341                 u16 type = nla_type(nla);
 342                 int expected_len;
 343
 344                 if (type > OVS_KEY_ATTR_MAX) {
 345                         OVS_NLERR(log, "Key type %d is out of range max %d",
 346                                   type, OVS_KEY_ATTR_MAX);
 347                         return -EINVAL;
 348                 }
 349
 350                 if (attrs & (1 << type)) {
 351                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 352                         return -EINVAL;
 353                 }
 354
 355                 expected_len = ovs_key_lens[type];
 356                 if (nla_len(nla) != expected_len && expected_len != -1) {
 357                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 358                                   type, nla_len(nla), expected_len);
 359                         return -EINVAL;
 360                 }
 361
 362                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 363                         attrs |= 1 << type;
 364                         a[type] = nla;
 365                 }
 366         }
 367         if (rem) {
 368                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 369                 return -EINVAL;
 370         }
 371
 372         *attrsp = attrs;
 373         return 0;
 374 }
 375
 376 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 377                                    const struct nlattr *a[], u64 *attrsp,
 378                                    bool log)
 379 {
 380         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 381 }
 382
 383 static int parse_flow_nlattrs(const struct nlattr *attr,
 384                               const struct nlattr *a[], u64 *attrsp,
 385                               bool log)
 386 {
 387         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 388 }
 389
 390 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 391                                      struct sw_flow_match *match, bool is_mask,
 392                                      bool log)
 393 {
 394         unsigned long opt_key_offset;
 395
 396         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 397                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 398                           nla_len(a), sizeof(match->key->tun_opts));
 399                 return -EINVAL;
 400         }
 401
 402         if (nla_len(a) % 4 != 0) {
 403                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 404                           nla_len(a));
 405                 return -EINVAL;
 406         }
 407
 408         /* We need to record the length of the options passed
 409          * down, otherwise packets with the same format but
 410          * additional options will be silently matched.
 411          */
 412         if (!is_mask) {
 413                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 414                                 false);
 415         } else {
 416                 /* This is somewhat unusual because it looks at
 417                  * both the key and mask while parsing the
 418                  * attributes (and by extension assumes the key
 419                  * is parsed first). Normally, we would verify
 420                  * that each is the correct length and that the
 421                  * attributes line up in the validate function.
 422                  * However, that is difficult because this is
 423                  * variable length and we won't have the
 424                  * information later.
 425                  */
 426                 if (match->key->tun_opts_len != nla_len(a)) {
 427                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 428                                   match->key->tun_opts_len, nla_len(a));
 429                         return -EINVAL;
 430                 }
 431
 432                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 433         }
 434
 435         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 436         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 437                                   nla_len(a), is_mask);
 438         return 0;
 439 }
 440
 441 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
 442                                 struct sw_flow_match *match, bool is_mask,
 443                                 bool log)
 444 {
 445         struct nlattr *a;
 446         int rem;
 447         bool ttl = false;
 448         __be16 tun_flags = 0;
 449
 450         nla_for_each_nested(a, attr, rem) {
 451                 int type = nla_type(a);
 452                 int err;
 453
 454                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 455                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
 456                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
 457                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
 458                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
 459                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
 460                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
 461                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
 462                         [OVS_TUNNEL_KEY_ATTR_TP_SRC] = sizeof(u16),
 463                         [OVS_TUNNEL_KEY_ATTR_TP_DST] = sizeof(u16),
 464                         [OVS_TUNNEL_KEY_ATTR_OAM] = 0,
 465                         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS] = -1,
 466                 };
 467
 468                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 469                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 470                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 471                         return -EINVAL;
 472                 }
 473
 474                 if (ovs_tunnel_key_lens[type] != nla_len(a) &&
 475                     ovs_tunnel_key_lens[type] != -1) {
 476                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 477                                   type, nla_len(a), ovs_tunnel_key_lens[type]);
 478                         return -EINVAL;
 479                 }
 480
 481                 switch (type) {
 482                 case OVS_TUNNEL_KEY_ATTR_ID:
 483                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 484                                         nla_get_be64(a), is_mask);
 485                         tun_flags |= TUNNEL_KEY;
 486                         break;
 487                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 488                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
 489                                         nla_get_be32(a), is_mask);
 490                         break;
 491                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 492                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
 493                                         nla_get_be32(a), is_mask);
 494                         break;
 495                 case OVS_TUNNEL_KEY_ATTR_TOS:
 496                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
 497                                         nla_get_u8(a), is_mask);
 498                         break;
 499                 case OVS_TUNNEL_KEY_ATTR_TTL:
 500                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
 501                                         nla_get_u8(a), is_mask);
 502                         ttl = true;
 503                         break;
 504                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 505                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 506                         break;
 507                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 508                         tun_flags |= TUNNEL_CSUM;
 509                         break;
 510                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 511                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 512                                         nla_get_be16(a), is_mask);
 513                         break;
 514                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 515                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 516                                         nla_get_be16(a), is_mask);
 517                         break;
 518                 case OVS_TUNNEL_KEY_ATTR_OAM:
 519                         tun_flags |= TUNNEL_OAM;
 520                         break;
 521                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 522                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 523                         if (err)
 524                                 return err;
 525
 526                         tun_flags |= TUNNEL_OPTIONS_PRESENT;
 527                         break;
 528                 default:
 529                         OVS_NLERR(log, "Unknown IPv4 tunnel attribute %d",
 530                                   type);
 531                         return -EINVAL;
 532                 }
 533         }
 534
 535         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 536
 537         if (rem > 0) {
 538                 OVS_NLERR(log, "IPv4 tunnel attribute has %d unknown bytes.",
 539                           rem);
 540                 return -EINVAL;
 541         }
 542
 543         if (!is_mask) {
 544                 if (!match->key->tun_key.ipv4_dst) {
 545                         OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 546                         return -EINVAL;
 547                 }
 548
 549                 if (!ttl) {
 550                         OVS_NLERR(log, "IPv4 tunnel TTL not specified.");
 551                         return -EINVAL;
 552                 }
 553         }
 554
 555         return 0;
 556 }
 557
 558 static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
 559                                 const struct ovs_key_ipv4_tunnel *output,
 560                                 const void *tun_opts, int swkey_tun_opts_len)
 561 {
 562         if (output->tun_flags & TUNNEL_KEY &&
 563             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 564                 return -EMSGSIZE;
 565         if (output->ipv4_src &&
 566             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
 567                 return -EMSGSIZE;
 568         if (output->ipv4_dst &&
 569             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
 570                 return -EMSGSIZE;
 571         if (output->ipv4_tos &&
 572             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
 573                 return -EMSGSIZE;
 574         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
 575                 return -EMSGSIZE;
 576         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 577             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 578                 return -EMSGSIZE;
 579         if ((output->tun_flags & TUNNEL_CSUM) &&
 580             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 581                 return -EMSGSIZE;
 582         if (output->tp_src &&
 583             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 584                 return -EMSGSIZE;
 585         if (output->tp_dst &&
 586             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 587                 return -EMSGSIZE;
 588         if ((output->tun_flags & TUNNEL_OAM) &&
 589             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 590                 return -EMSGSIZE;
 591         if (tun_opts &&
 592             nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 593                     swkey_tun_opts_len, tun_opts))
 594                 return -EMSGSIZE;
 595
 596         return 0;
 597 }
 598
 599 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
 600                               const struct ovs_key_ipv4_tunnel *output,
 601                               const void *tun_opts, int swkey_tun_opts_len)
 602 {
 603         struct nlattr *nla;
 604         int err;
 605
 606         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 607         if (!nla)
 608                 return -EMSGSIZE;
 609
 610         err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
 611         if (err)
 612                 return err;
 613
 614         nla_nest_end(skb, nla);
 615         return 0;
 616 }
 617
 618 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
 619                                   const struct ovs_tunnel_info *egress_tun_info)
 620 {
 621         return __ipv4_tun_to_nlattr(skb, &egress_tun_info->tunnel,
 622                                     egress_tun_info->options,
 623                                     egress_tun_info->options_len);
 624 }
 625
 626 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
 627                                  const struct nlattr **a, bool is_mask,
 628                                  bool log)
 629 {
 630         if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
 631                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 632
 633                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 634                 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
 635         }
 636
 637         if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
 638                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 639
 640                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 641                 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
 642         }
 643
 644         if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 645                 SW_FLOW_KEY_PUT(match, phy.priority,
 646                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 647                 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 648         }
 649
 650         if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 651                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 652
 653                 if (is_mask) {
 654                         in_port = 0xffffffff; /* Always exact match in_port. */
 655                 } else if (in_port >= DP_MAX_PORTS) {
 656                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 657                                   in_port, DP_MAX_PORTS);
 658                         return -EINVAL;
 659                 }
 660
 661                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 662                 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
 663         } else if (!is_mask) {
 664                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 665         }
 666
 667         if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
 668                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 669
 670                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 671                 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
 672         }
 673         if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
 674                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 675                                          is_mask, log))
 676                         return -EINVAL;
 677                 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
 678         }
 679         return 0;
 680 }
 681
 682 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
 683                                 const struct nlattr **a, bool is_mask,
 684                                 bool log)
 685 {
 686         int err;
 687
 688         err = metadata_from_nlattrs(match, &attrs, a, is_mask, log);
 689         if (err)
 690                 return err;
 691
 692         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
 693                 const struct ovs_key_ethernet *eth_key;
 694
 695                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 696                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 697                                 eth_key->eth_src, ETH_ALEN, is_mask);
 698                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 699                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 700                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
 701         }
 702
 703         if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
 704                 __be16 tci;
 705
 706                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 707                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 708                         if (is_mask)
 709                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 710                         else
 711                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 712
 713                         return -EINVAL;
 714                 }
 715
 716                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 717                 attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
 718         }
 719
 720         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
 721                 __be16 eth_type;
 722
 723                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 724                 if (is_mask) {
 725                         /* Always exact match EtherType. */
 726                         eth_type = htons(0xffff);
 727                 } else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
 728                         OVS_NLERR(log, "EtherType %x is less than min %x",
 729                                   ntohs(eth_type), ETH_P_802_3_MIN);
 730                         return -EINVAL;
 731                 }
 732
 733                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 734                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
 735         } else if (!is_mask) {
 736                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 737         }
 738
 739         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
 740                 const struct ovs_key_ipv4 *ipv4_key;
 741
 742                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 743                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 744                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 745                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 746                         return -EINVAL;
 747                 }
 748                 SW_FLOW_KEY_PUT(match, ip.proto,
 749                                 ipv4_key->ipv4_proto, is_mask);
 750                 SW_FLOW_KEY_PUT(match, ip.tos,
 751                                 ipv4_key->ipv4_tos, is_mask);
 752                 SW_FLOW_KEY_PUT(match, ip.ttl,
 753                                 ipv4_key->ipv4_ttl, is_mask);
 754                 SW_FLOW_KEY_PUT(match, ip.frag,
 755                                 ipv4_key->ipv4_frag, is_mask);
 756                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 757                                 ipv4_key->ipv4_src, is_mask);
 758                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 759                                 ipv4_key->ipv4_dst, is_mask);
 760                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
 761         }
 762
 763         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
 764                 const struct ovs_key_ipv6 *ipv6_key;
 765
 766                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 767                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 768                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 769                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 770                         return -EINVAL;
 771                 }
 772
 773                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 774                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
 775                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 776                         return -EINVAL;
 777                 }
 778
 779                 SW_FLOW_KEY_PUT(match, ipv6.label,
 780                                 ipv6_key->ipv6_label, is_mask);
 781                 SW_FLOW_KEY_PUT(match, ip.proto,
 782                                 ipv6_key->ipv6_proto, is_mask);
 783                 SW_FLOW_KEY_PUT(match, ip.tos,
 784                                 ipv6_key->ipv6_tclass, is_mask);
 785                 SW_FLOW_KEY_PUT(match, ip.ttl,
 786                                 ipv6_key->ipv6_hlimit, is_mask);
 787                 SW_FLOW_KEY_PUT(match, ip.frag,
 788                                 ipv6_key->ipv6_frag, is_mask);
 789                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
 790                                 ipv6_key->ipv6_src,
 791                                 sizeof(match->key->ipv6.addr.src),
 792                                 is_mask);
 793                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
 794                                 ipv6_key->ipv6_dst,
 795                                 sizeof(match->key->ipv6.addr.dst),
 796                                 is_mask);
 797
 798                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
 799         }
 800
 801         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
 802                 const struct ovs_key_arp *arp_key;
 803
 804                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
 805                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
 806                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
 807                                   arp_key->arp_op);
 808                         return -EINVAL;
 809                 }
 810
 811                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 812                                 arp_key->arp_sip, is_mask);
 813                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 814                         arp_key->arp_tip, is_mask);
 815                 SW_FLOW_KEY_PUT(match, ip.proto,
 816                                 ntohs(arp_key->arp_op), is_mask);
 817                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
 818                                 arp_key->arp_sha, ETH_ALEN, is_mask);
 819                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
 820                                 arp_key->arp_tha, ETH_ALEN, is_mask);
 821
 822                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
 823         }
 824
 825         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
 826                 const struct ovs_key_mpls *mpls_key;
 827
 828                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
 829                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
 830                                 mpls_key->mpls_lse, is_mask);
 831
 832                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
 833          }
 834
 835         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
 836                 const struct ovs_key_tcp *tcp_key;
 837
 838                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 839                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
 840                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
 841                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 842         }
 843
 844         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
 845                 SW_FLOW_KEY_PUT(match, tp.flags,
 846                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
 847                                 is_mask);
 848                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
 849         }
 850
 851         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
 852                 const struct ovs_key_udp *udp_key;
 853
 854                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 855                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
 856                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
 857                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 858         }
 859
 860         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
 861                 const struct ovs_key_sctp *sctp_key;
 862
 863                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
 864                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
 865                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
 866                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
 867         }
 868
 869         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
 870                 const struct ovs_key_icmp *icmp_key;
 871
 872                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 873                 SW_FLOW_KEY_PUT(match, tp.src,
 874                                 htons(icmp_key->icmp_type), is_mask);
 875                 SW_FLOW_KEY_PUT(match, tp.dst,
 876                                 htons(icmp_key->icmp_code), is_mask);
 877                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 878         }
 879
 880         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
 881                 const struct ovs_key_icmpv6 *icmpv6_key;
 882
 883                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 884                 SW_FLOW_KEY_PUT(match, tp.src,
 885                                 htons(icmpv6_key->icmpv6_type), is_mask);
 886                 SW_FLOW_KEY_PUT(match, tp.dst,
 887                                 htons(icmpv6_key->icmpv6_code), is_mask);
 888                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 889         }
 890
 891         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
 892                 const struct ovs_key_nd *nd_key;
 893
 894                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 895                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
 896                         nd_key->nd_target,
 897                         sizeof(match->key->ipv6.nd.target),
 898                         is_mask);
 899                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
 900                         nd_key->nd_sll, ETH_ALEN, is_mask);
 901                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
 902                                 nd_key->nd_tll, ETH_ALEN, is_mask);
 903                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
 904         }
 905
 906         if (attrs != 0) {
 907                 OVS_NLERR(log, "Unknown key attributes %llx",
 908                           (unsigned long long)attrs);
 909                 return -EINVAL;
 910         }
 911
 912         return 0;
 913 }
 914
 915 static void nlattr_set(struct nlattr *attr, u8 val, bool is_attr_mask_key)
 916 {
 917         struct nlattr *nla;
 918         int rem;
 919
 920         /* The nlattr stream should already have been validated */
 921         nla_for_each_nested(nla, attr, rem) {
 922                 /* We assume that ovs_key_lens[type] == -1 means that type is a
 923                  * nested attribute
 924                  */
 925                 if (is_attr_mask_key && ovs_key_lens[nla_type(nla)] == -1)
 926                         nlattr_set(nla, val, false);
 927                 else
 928                         memset(nla_data(nla), val, nla_len(nla));
 929         }
 930 }
 931
 932 static void mask_set_nlattr(struct nlattr *attr, u8 val)
 933 {
 934         nlattr_set(attr, val, true);
 935 }
 936
 937 /**
 938  * ovs_nla_get_match - parses Netlink attributes into a flow key and
 939  * mask. In case the 'mask' is NULL, the flow is treated as exact match
 940  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
 941  * does not include any don't care bit.
 942  * @match: receives the extracted flow match information.
 943  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 944  * sequence. The fields should of the packet that triggered the creation
 945  * of this flow.
 946  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
 947  * attribute specifies the mask field of the wildcarded flow.
 948  * @log: Boolean to allow kernel error logging.  Normally true, but when
 949  * probing for feature compatibility this should be passed in as false to
 950  * suppress unnecessary error logging.
 951  */
 952 int ovs_nla_get_match(struct sw_flow_match *match,
 953                       const struct nlattr *nla_key,
 954                       const struct nlattr *nla_mask,
 955                       bool log)
 956 {
 957         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 958         const struct nlattr *encap;
 959         struct nlattr *newmask = NULL;
 960         u64 key_attrs = 0;
 961         u64 mask_attrs = 0;
 962         bool encap_valid = false;
 963         int err;
 964
 965         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
 966         if (err)
 967                 return err;
 968
 969         if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
 970             (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
 971             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
 972                 __be16 tci;
 973
 974                 if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
 975                       (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
 976                         OVS_NLERR(log, "Invalid Vlan frame.");
 977                         return -EINVAL;
 978                 }
 979
 980                 key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
 981                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 982                 encap = a[OVS_KEY_ATTR_ENCAP];
 983                 key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
 984                 encap_valid = true;
 985
 986                 if (tci & htons(VLAN_TAG_PRESENT)) {
 987                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
 988                         if (err)
 989                                 return err;
 990                 } else if (!tci) {
 991                         /* Corner case for truncated 802.1Q header. */
 992                         if (nla_len(encap)) {
 993                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
 994                                 return -EINVAL;
 995                         }
 996                 } else {
 997                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
 998                         return  -EINVAL;
 999                 }
1000         }
1001
1002         err = ovs_key_from_nlattrs(match, key_attrs, a, false, log);
1003         if (err)
1004                 return err;
1005
1006         if (match->mask) {
1007                 if (!nla_mask) {
1008                         /* Create an exact match mask. We need to set to 0xff
1009                          * all the 'match->mask' fields that have been touched
1010                          * in 'match->key'. We cannot simply memset
1011                          * 'match->mask', because padding bytes and fields not
1012                          * specified in 'match->key' should be left to 0.
1013                          * Instead, we use a stream of netlink attributes,
1014                          * copied from 'key' and set to 0xff.
1015                          * ovs_key_from_nlattrs() will take care of filling
1016                          * 'match->mask' appropriately.
1017                          */
1018                         newmask = kmemdup(nla_key,
1019                                           nla_total_size(nla_len(nla_key)),
1020                                           GFP_KERNEL);
1021                         if (!newmask)
1022                                 return -ENOMEM;
1023
1024                         mask_set_nlattr(newmask, 0xff);
1025
1026                         /* The userspace does not send tunnel attributes that
1027                          * are 0, but we should not wildcard them nonetheless.
1028                          */
1029                         if (match->key->tun_key.ipv4_dst)
1030                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1031                                                          0xff, true);
1032
1033                         nla_mask = newmask;
1034                 }
1035
1036                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1037                 if (err)
1038                         goto free_newmask;
1039
1040                 /* Always match on tci. */
1041                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1042
1043                 if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
1044                         __be16 eth_type = 0;
1045                         __be16 tci = 0;
1046
1047                         if (!encap_valid) {
1048                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1049                                 err = -EINVAL;
1050                                 goto free_newmask;
1051                         }
1052
1053                         mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1054                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1055                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1056
1057                         if (eth_type == htons(0xffff)) {
1058                                 mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1059                                 encap = a[OVS_KEY_ATTR_ENCAP];
1060                                 err = parse_flow_mask_nlattrs(encap, a,
1061                                                               &mask_attrs, log);
1062                                 if (err)
1063                                         goto free_newmask;
1064                         } else {
1065                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1066                                           ntohs(eth_type));
1067                                 err = -EINVAL;
1068                                 goto free_newmask;
1069                         }
1070
1071                         if (a[OVS_KEY_ATTR_VLAN])
1072                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1073
1074                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1075                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1076                                           ntohs(tci));
1077                                 err = -EINVAL;
1078                                 goto free_newmask;
1079                         }
1080                 }
1081
1082                 err = ovs_key_from_nlattrs(match, mask_attrs, a, true, log);
1083                 if (err)
1084                         goto free_newmask;
1085         }
1086
1087         if (!match_validate(match, key_attrs, mask_attrs, log))
1088                 err = -EINVAL;
1089
1090 free_newmask:
1091         kfree(newmask);
1092         return err;
1093 }
1094
1095 /**
1096  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1097  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1098  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1099  * sequence.
1100  * @log: Boolean to allow kernel error logging.  Normally true, but when
1101  * probing for feature compatibility this should be passed in as false to
1102  * suppress unnecessary error logging.
1103  *
1104  * This parses a series of Netlink attributes that form a flow key, which must
1105  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1106  * get the metadata, that is, the parts of the flow key that cannot be
1107  * extracted from the packet itself.
1108  */
1109
1110 int ovs_nla_get_flow_metadata(const struct nlattr *attr,
1111                               struct sw_flow_key *key,
1112                               bool log)
1113 {
1114         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1115         struct sw_flow_match match;
1116         u64 attrs = 0;
1117         int err;
1118
1119         err = parse_flow_nlattrs(attr, a, &attrs, log);
1120         if (err)
1121                 return -EINVAL;
1122
1123         memset(&match, 0, sizeof(match));
1124         match.key = key;
1125
1126         key->phy.in_port = DP_MAX_PORTS;
1127
1128         return metadata_from_nlattrs(&match, &attrs, a, false, log);
1129 }
1130
1131 int ovs_nla_put_flow(const struct sw_flow_key *swkey,
1132                      const struct sw_flow_key *output, struct sk_buff *skb)
1133 {
1134         struct ovs_key_ethernet *eth_key;
1135         struct nlattr *nla, *encap;
1136         bool is_mask = (swkey != output);
1137
1138         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1139                 goto nla_put_failure;
1140
1141         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1142                 goto nla_put_failure;
1143
1144         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1145                 goto nla_put_failure;
1146
1147         if ((swkey->tun_key.ipv4_dst || is_mask)) {
1148                 const void *opts = NULL;
1149
1150                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1151                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1152
1153                 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1154                                        swkey->tun_opts_len))
1155                         goto nla_put_failure;
1156         }
1157
1158         if (swkey->phy.in_port == DP_MAX_PORTS) {
1159                 if (is_mask && (output->phy.in_port == 0xffff))
1160                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1161                                 goto nla_put_failure;
1162         } else {
1163                 u16 upper_u16;
1164                 upper_u16 = !is_mask ? 0 : 0xffff;
1165
1166                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1167                                 (upper_u16 << 16) | output->phy.in_port))
1168                         goto nla_put_failure;
1169         }
1170
1171         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1172                 goto nla_put_failure;
1173
1174         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1175         if (!nla)
1176                 goto nla_put_failure;
1177
1178         eth_key = nla_data(nla);
1179         ether_addr_copy(eth_key->eth_src, output->eth.src);
1180         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1181
1182         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1183                 __be16 eth_type;
1184                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1185                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1186                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1187                         goto nla_put_failure;
1188                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1189                 if (!swkey->eth.tci)
1190                         goto unencap;
1191         } else
1192                 encap = NULL;
1193
1194         if (swkey->eth.type == htons(ETH_P_802_2)) {
1195                 /*
1196                  * Ethertype 802.2 is represented in the netlink with omitted
1197                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1198                  * 0xffff in the mask attribute.  Ethertype can also
1199                  * be wildcarded.
1200                  */
1201                 if (is_mask && output->eth.type)
1202                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1203                                                 output->eth.type))
1204                                 goto nla_put_failure;
1205                 goto unencap;
1206         }
1207
1208         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1209                 goto nla_put_failure;
1210
1211         if (swkey->eth.type == htons(ETH_P_IP)) {
1212                 struct ovs_key_ipv4 *ipv4_key;
1213
1214                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1215                 if (!nla)
1216                         goto nla_put_failure;
1217                 ipv4_key = nla_data(nla);
1218                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1219                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1220                 ipv4_key->ipv4_proto = output->ip.proto;
1221                 ipv4_key->ipv4_tos = output->ip.tos;
1222                 ipv4_key->ipv4_ttl = output->ip.ttl;
1223                 ipv4_key->ipv4_frag = output->ip.frag;
1224         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1225                 struct ovs_key_ipv6 *ipv6_key;
1226
1227                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1228                 if (!nla)
1229                         goto nla_put_failure;
1230                 ipv6_key = nla_data(nla);
1231                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1232                                 sizeof(ipv6_key->ipv6_src));
1233                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1234                                 sizeof(ipv6_key->ipv6_dst));
1235                 ipv6_key->ipv6_label = output->ipv6.label;
1236                 ipv6_key->ipv6_proto = output->ip.proto;
1237                 ipv6_key->ipv6_tclass = output->ip.tos;
1238                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1239                 ipv6_key->ipv6_frag = output->ip.frag;
1240         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1241                    swkey->eth.type == htons(ETH_P_RARP)) {
1242                 struct ovs_key_arp *arp_key;
1243
1244                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1245                 if (!nla)
1246                         goto nla_put_failure;
1247                 arp_key = nla_data(nla);
1248                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1249                 arp_key->arp_sip = output->ipv4.addr.src;
1250                 arp_key->arp_tip = output->ipv4.addr.dst;
1251                 arp_key->arp_op = htons(output->ip.proto);
1252                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1253                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1254         } else if (eth_p_mpls(swkey->eth.type)) {
1255                 struct ovs_key_mpls *mpls_key;
1256
1257                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1258                 if (!nla)
1259                         goto nla_put_failure;
1260                 mpls_key = nla_data(nla);
1261                 mpls_key->mpls_lse = output->mpls.top_lse;
1262         }
1263
1264         if ((swkey->eth.type == htons(ETH_P_IP) ||
1265              swkey->eth.type == htons(ETH_P_IPV6)) &&
1266              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1267
1268                 if (swkey->ip.proto == IPPROTO_TCP) {
1269                         struct ovs_key_tcp *tcp_key;
1270
1271                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1272                         if (!nla)
1273                                 goto nla_put_failure;
1274                         tcp_key = nla_data(nla);
1275                         tcp_key->tcp_src = output->tp.src;
1276                         tcp_key->tcp_dst = output->tp.dst;
1277                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1278                                          output->tp.flags))
1279                                 goto nla_put_failure;
1280                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1281                         struct ovs_key_udp *udp_key;
1282
1283                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1284                         if (!nla)
1285                                 goto nla_put_failure;
1286                         udp_key = nla_data(nla);
1287                         udp_key->udp_src = output->tp.src;
1288                         udp_key->udp_dst = output->tp.dst;
1289                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1290                         struct ovs_key_sctp *sctp_key;
1291
1292                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1293                         if (!nla)
1294                                 goto nla_put_failure;
1295                         sctp_key = nla_data(nla);
1296                         sctp_key->sctp_src = output->tp.src;
1297                         sctp_key->sctp_dst = output->tp.dst;
1298                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1299                            swkey->ip.proto == IPPROTO_ICMP) {
1300                         struct ovs_key_icmp *icmp_key;
1301
1302                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1303                         if (!nla)
1304                                 goto nla_put_failure;
1305                         icmp_key = nla_data(nla);
1306                         icmp_key->icmp_type = ntohs(output->tp.src);
1307                         icmp_key->icmp_code = ntohs(output->tp.dst);
1308                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1309                            swkey->ip.proto == IPPROTO_ICMPV6) {
1310                         struct ovs_key_icmpv6 *icmpv6_key;
1311
1312                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1313                                                 sizeof(*icmpv6_key));
1314                         if (!nla)
1315                                 goto nla_put_failure;
1316                         icmpv6_key = nla_data(nla);
1317                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1318                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1319
1320                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1321                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1322                                 struct ovs_key_nd *nd_key;
1323
1324                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1325                                 if (!nla)
1326                                         goto nla_put_failure;
1327                                 nd_key = nla_data(nla);
1328                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1329                                                         sizeof(nd_key->nd_target));
1330                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1331                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1332                         }
1333                 }
1334         }
1335
1336 unencap:
1337         if (encap)
1338                 nla_nest_end(skb, encap);
1339
1340         return 0;
1341
1342 nla_put_failure:
1343         return -EMSGSIZE;
1344 }
1345
1346 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1347
1348 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1349 {
1350         struct sw_flow_actions *sfa;
1351
1352         if (size > MAX_ACTIONS_BUFSIZE) {
1353                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1354                 return ERR_PTR(-EINVAL);
1355         }
1356
1357         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1358         if (!sfa)
1359                 return ERR_PTR(-ENOMEM);
1360
1361         sfa->actions_len = 0;
1362         return sfa;
1363 }
1364
1365 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1366  * The caller must hold rcu_read_lock for this to be sensible. */
1367 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1368 {
1369         kfree_rcu(sf_acts, rcu);
1370 }
1371
1372 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1373                                        int attr_len, bool log)
1374 {
1375
1376         struct sw_flow_actions *acts;
1377         int new_acts_size;
1378         int req_size = NLA_ALIGN(attr_len);
1379         int next_offset = offsetof(struct sw_flow_actions, actions) +
1380                                         (*sfa)->actions_len;
1381
1382         if (req_size <= (ksize(*sfa) - next_offset))
1383                 goto out;
1384
1385         new_acts_size = ksize(*sfa) * 2;
1386
1387         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1388                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1389                         return ERR_PTR(-EMSGSIZE);
1390                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1391         }
1392
1393         acts = nla_alloc_flow_actions(new_acts_size, log);
1394         if (IS_ERR(acts))
1395                 return (void *)acts;
1396
1397         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1398         acts->actions_len = (*sfa)->actions_len;
1399         kfree(*sfa);
1400         *sfa = acts;
1401
1402 out:
1403         (*sfa)->actions_len += req_size;
1404         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1405 }
1406
1407 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1408                                    int attrtype, void *data, int len, bool log)
1409 {
1410         struct nlattr *a;
1411
1412         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1413         if (IS_ERR(a))
1414                 return a;
1415
1416         a->nla_type = attrtype;
1417         a->nla_len = nla_attr_size(len);
1418
1419         if (data)
1420                 memcpy(nla_data(a), data, len);
1421         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1422
1423         return a;
1424 }
1425
1426 static int add_action(struct sw_flow_actions **sfa, int attrtype,
1427                       void *data, int len, bool log)
1428 {
1429         struct nlattr *a;
1430
1431         a = __add_action(sfa, attrtype, data, len, log);
1432
1433         return PTR_ERR_OR_ZERO(a);
1434 }
1435
1436 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1437                                           int attrtype, bool log)
1438 {
1439         int used = (*sfa)->actions_len;
1440         int err;
1441
1442         err = add_action(sfa, attrtype, NULL, 0, log);
1443         if (err)
1444                 return err;
1445
1446         return used;
1447 }
1448
1449 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1450                                          int st_offset)
1451 {
1452         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1453                                                                st_offset);
1454
1455         a->nla_len = sfa->actions_len - st_offset;
1456 }
1457
1458 static int __ovs_nla_copy_actions(const struct nlattr *attr,
1459                                   const struct sw_flow_key *key,
1460                                   int depth, struct sw_flow_actions **sfa,
1461                                   __be16 eth_type, __be16 vlan_tci, bool log);
1462
1463 static int validate_and_copy_sample(const struct nlattr *attr,
1464                                     const struct sw_flow_key *key, int depth,
1465                                     struct sw_flow_actions **sfa,
1466                                     __be16 eth_type, __be16 vlan_tci, bool log)
1467 {
1468         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1469         const struct nlattr *probability, *actions;
1470         const struct nlattr *a;
1471         int rem, start, err, st_acts;
1472
1473         memset(attrs, 0, sizeof(attrs));
1474         nla_for_each_nested(a, attr, rem) {
1475                 int type = nla_type(a);
1476                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1477                         return -EINVAL;
1478                 attrs[type] = a;
1479         }
1480         if (rem)
1481                 return -EINVAL;
1482
1483         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1484         if (!probability || nla_len(probability) != sizeof(u32))
1485                 return -EINVAL;
1486
1487         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1488         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1489                 return -EINVAL;
1490
1491         /* validation done, copy sample action. */
1492         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1493         if (start < 0)
1494                 return start;
1495         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1496                          nla_data(probability), sizeof(u32), log);
1497         if (err)
1498                 return err;
1499         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1500         if (st_acts < 0)
1501                 return st_acts;
1502
1503         err = __ovs_nla_copy_actions(actions, key, depth + 1, sfa,
1504                                      eth_type, vlan_tci, log);
1505         if (err)
1506                 return err;
1507
1508         add_nested_action_end(*sfa, st_acts);
1509         add_nested_action_end(*sfa, start);
1510
1511         return 0;
1512 }
1513
1514 static int validate_tp_port(const struct sw_flow_key *flow_key,
1515                             __be16 eth_type)
1516 {
1517         if ((eth_type == htons(ETH_P_IP) || eth_type == htons(ETH_P_IPV6)) &&
1518             (flow_key->tp.src || flow_key->tp.dst))
1519                 return 0;
1520
1521         return -EINVAL;
1522 }
1523
1524 void ovs_match_init(struct sw_flow_match *match,
1525                     struct sw_flow_key *key,
1526                     struct sw_flow_mask *mask)
1527 {
1528         memset(match, 0, sizeof(*match));
1529         match->key = key;
1530         match->mask = mask;
1531
1532         memset(key, 0, sizeof(*key));
1533
1534         if (mask) {
1535                 memset(&mask->key, 0, sizeof(mask->key));
1536                 mask->range.start = mask->range.end = 0;
1537         }
1538 }
1539
1540 static int validate_geneve_opts(struct sw_flow_key *key)
1541 {
1542         struct geneve_opt *option;
1543         int opts_len = key->tun_opts_len;
1544         bool crit_opt = false;
1545
1546         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1547         while (opts_len > 0) {
1548                 int len;
1549
1550                 if (opts_len < sizeof(*option))
1551                         return -EINVAL;
1552
1553                 len = sizeof(*option) + option->length * 4;
1554                 if (len > opts_len)
1555                         return -EINVAL;
1556
1557                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1558
1559                 option = (struct geneve_opt *)((u8 *)option + len);
1560                 opts_len -= len;
1561         };
1562
1563         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1564
1565         return 0;
1566 }
1567
1568 static int validate_and_copy_set_tun(const struct nlattr *attr,
1569                                      struct sw_flow_actions **sfa, bool log)
1570 {
1571         struct sw_flow_match match;
1572         struct sw_flow_key key;
1573         struct ovs_tunnel_info *tun_info;
1574         struct nlattr *a;
1575         int err, start;
1576
1577         ovs_match_init(&match, &key, NULL);
1578         err = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1579         if (err)
1580                 return err;
1581
1582         if (key.tun_opts_len) {
1583                 err = validate_geneve_opts(&key);
1584                 if (err < 0)
1585                         return err;
1586         };
1587
1588         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1589         if (start < 0)
1590                 return start;
1591
1592         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1593                          sizeof(*tun_info) + key.tun_opts_len, log);
1594         if (IS_ERR(a))
1595                 return PTR_ERR(a);
1596
1597         tun_info = nla_data(a);
1598         tun_info->tunnel = key.tun_key;
1599         tun_info->options_len = key.tun_opts_len;
1600
1601         if (tun_info->options_len) {
1602                 /* We need to store the options in the action itself since
1603                  * everything else will go away after flow setup. We can append
1604                  * it to tun_info and then point there.
1605                  */
1606                 memcpy((tun_info + 1),
1607                        TUN_METADATA_OPTS(&key, key.tun_opts_len), key.tun_opts_len);
1608                 tun_info->options = (tun_info + 1);
1609         } else {
1610                 tun_info->options = NULL;
1611         }
1612
1613         add_nested_action_end(*sfa, start);
1614
1615         return err;
1616 }
1617
1618 static int validate_set(const struct nlattr *a,
1619                         const struct sw_flow_key *flow_key,
1620                         struct sw_flow_actions **sfa,
1621                         bool *set_tun, __be16 eth_type, bool log)
1622 {
1623         const struct nlattr *ovs_key = nla_data(a);
1624         int key_type = nla_type(ovs_key);
1625
1626         /* There can be only one key in a action */
1627         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1628                 return -EINVAL;
1629
1630         if (key_type > OVS_KEY_ATTR_MAX ||
1631             (ovs_key_lens[key_type] != nla_len(ovs_key) &&
1632              ovs_key_lens[key_type] != -1))
1633                 return -EINVAL;
1634
1635         switch (key_type) {
1636         const struct ovs_key_ipv4 *ipv4_key;
1637         const struct ovs_key_ipv6 *ipv6_key;
1638         int err;
1639
1640         case OVS_KEY_ATTR_PRIORITY:
1641         case OVS_KEY_ATTR_SKB_MARK:
1642         case OVS_KEY_ATTR_ETHERNET:
1643                 break;
1644
1645         case OVS_KEY_ATTR_TUNNEL:
1646                 if (eth_p_mpls(eth_type))
1647                         return -EINVAL;
1648
1649                 *set_tun = true;
1650                 err = validate_and_copy_set_tun(a, sfa, log);
1651                 if (err)
1652                         return err;
1653                 break;
1654
1655         case OVS_KEY_ATTR_IPV4:
1656                 if (eth_type != htons(ETH_P_IP))
1657                         return -EINVAL;
1658
1659                 if (!flow_key->ip.proto)
1660                         return -EINVAL;
1661
1662                 ipv4_key = nla_data(ovs_key);
1663                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1664                         return -EINVAL;
1665
1666                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1667                         return -EINVAL;
1668
1669                 break;
1670
1671         case OVS_KEY_ATTR_IPV6:
1672                 if (eth_type != htons(ETH_P_IPV6))
1673                         return -EINVAL;
1674
1675                 if (!flow_key->ip.proto)
1676                         return -EINVAL;
1677
1678                 ipv6_key = nla_data(ovs_key);
1679                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1680                         return -EINVAL;
1681
1682                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1683                         return -EINVAL;
1684
1685                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
1686                         return -EINVAL;
1687
1688                 break;
1689
1690         case OVS_KEY_ATTR_TCP:
1691                 if (flow_key->ip.proto != IPPROTO_TCP)
1692                         return -EINVAL;
1693
1694                 return validate_tp_port(flow_key, eth_type);
1695
1696         case OVS_KEY_ATTR_UDP:
1697                 if (flow_key->ip.proto != IPPROTO_UDP)
1698                         return -EINVAL;
1699
1700                 return validate_tp_port(flow_key, eth_type);
1701
1702         case OVS_KEY_ATTR_MPLS:
1703                 if (!eth_p_mpls(eth_type))
1704                         return -EINVAL;
1705                 break;
1706
1707         case OVS_KEY_ATTR_SCTP:
1708                 if (flow_key->ip.proto != IPPROTO_SCTP)
1709                         return -EINVAL;
1710
1711                 return validate_tp_port(flow_key, eth_type);
1712
1713         default:
1714                 return -EINVAL;
1715         }
1716
1717         return 0;
1718 }
1719
1720 static int validate_userspace(const struct nlattr *attr)
1721 {
1722         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
1723                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
1724                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
1725                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
1726         };
1727         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
1728         int error;
1729
1730         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
1731                                  attr, userspace_policy);
1732         if (error)
1733                 return error;
1734
1735         if (!a[OVS_USERSPACE_ATTR_PID] ||
1736             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
1737                 return -EINVAL;
1738
1739         return 0;
1740 }
1741
1742 static int copy_action(const struct nlattr *from,
1743                        struct sw_flow_actions **sfa, bool log)
1744 {
1745         int totlen = NLA_ALIGN(from->nla_len);
1746         struct nlattr *to;
1747
1748         to = reserve_sfa_size(sfa, from->nla_len, log);
1749         if (IS_ERR(to))
1750                 return PTR_ERR(to);
1751
1752         memcpy(to, from, totlen);
1753         return 0;
1754 }
1755
1756 static int __ovs_nla_copy_actions(const struct nlattr *attr,
1757                                   const struct sw_flow_key *key,
1758                                   int depth, struct sw_flow_actions **sfa,
1759                                   __be16 eth_type, __be16 vlan_tci, bool log)
1760 {
1761         const struct nlattr *a;
1762         int rem, err;
1763
1764         if (depth >= SAMPLE_ACTION_DEPTH)
1765                 return -EOVERFLOW;
1766
1767         nla_for_each_nested(a, attr, rem) {
1768                 /* Expected argument lengths, (u32)-1 for variable length. */
1769                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
1770                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
1771                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
1772                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
1773                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
1774                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
1775                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
1776                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
1777                         [OVS_ACTION_ATTR_SET] = (u32)-1,
1778                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
1779                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash)
1780                 };
1781                 const struct ovs_action_push_vlan *vlan;
1782                 int type = nla_type(a);
1783                 bool skip_copy;
1784
1785                 if (type > OVS_ACTION_ATTR_MAX ||
1786                     (action_lens[type] != nla_len(a) &&
1787                      action_lens[type] != (u32)-1))
1788                         return -EINVAL;
1789
1790                 skip_copy = false;
1791                 switch (type) {
1792                 case OVS_ACTION_ATTR_UNSPEC:
1793                         return -EINVAL;
1794
1795                 case OVS_ACTION_ATTR_USERSPACE:
1796                         err = validate_userspace(a);
1797                         if (err)
1798                                 return err;
1799                         break;
1800
1801                 case OVS_ACTION_ATTR_OUTPUT:
1802                         if (nla_get_u32(a) >= DP_MAX_PORTS)
1803                                 return -EINVAL;
1804                         break;
1805
1806                 case OVS_ACTION_ATTR_HASH: {
1807                         const struct ovs_action_hash *act_hash = nla_data(a);
1808
1809                         switch (act_hash->hash_alg) {
1810                         case OVS_HASH_ALG_L4:
1811                                 break;
1812                         default:
1813                                 return  -EINVAL;
1814                         }
1815
1816                         break;
1817                 }
1818
1819                 case OVS_ACTION_ATTR_POP_VLAN:
1820                         vlan_tci = htons(0);
1821                         break;
1822
1823                 case OVS_ACTION_ATTR_PUSH_VLAN:
1824                         vlan = nla_data(a);
1825                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
1826                                 return -EINVAL;
1827                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
1828                                 return -EINVAL;
1829                         vlan_tci = vlan->vlan_tci;
1830                         break;
1831
1832                 case OVS_ACTION_ATTR_RECIRC:
1833                         break;
1834
1835                 case OVS_ACTION_ATTR_PUSH_MPLS: {
1836                         const struct ovs_action_push_mpls *mpls = nla_data(a);
1837
1838                         if (!eth_p_mpls(mpls->mpls_ethertype))
1839                                 return -EINVAL;
1840                         /* Prohibit push MPLS other than to a white list
1841                          * for packets that have a known tag order.
1842                          */
1843                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
1844                             (eth_type != htons(ETH_P_IP) &&
1845                              eth_type != htons(ETH_P_IPV6) &&
1846                              eth_type != htons(ETH_P_ARP) &&
1847                              eth_type != htons(ETH_P_RARP) &&
1848                              !eth_p_mpls(eth_type)))
1849                                 return -EINVAL;
1850                         eth_type = mpls->mpls_ethertype;
1851                         break;
1852                 }
1853
1854                 case OVS_ACTION_ATTR_POP_MPLS:
1855                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
1856                             !eth_p_mpls(eth_type))
1857                                 return -EINVAL;
1858
1859                         /* Disallow subsequent L2.5+ set and mpls_pop actions
1860                          * as there is no check here to ensure that the new
1861                          * eth_type is valid and thus set actions could
1862                          * write off the end of the packet or otherwise
1863                          * corrupt it.
1864                          *
1865                          * Support for these actions is planned using packet
1866                          * recirculation.
1867                          */
1868                         eth_type = htons(0);
1869                         break;
1870
1871                 case OVS_ACTION_ATTR_SET:
1872                         err = validate_set(a, key, sfa,
1873                                            &skip_copy, eth_type, log);
1874                         if (err)
1875                                 return err;
1876                         break;
1877
1878                 case OVS_ACTION_ATTR_SAMPLE:
1879                         err = validate_and_copy_sample(a, key, depth, sfa,
1880                                                        eth_type, vlan_tci, log);
1881                         if (err)
1882                                 return err;
1883                         skip_copy = true;
1884                         break;
1885
1886                 default:
1887                         OVS_NLERR(log, "Unknown Action type %d", type);
1888                         return -EINVAL;
1889                 }
1890                 if (!skip_copy) {
1891                         err = copy_action(a, sfa, log);
1892                         if (err)
1893                                 return err;
1894                 }
1895         }
1896
1897         if (rem > 0)
1898                 return -EINVAL;
1899
1900         return 0;
1901 }
1902
1903 int ovs_nla_copy_actions(const struct nlattr *attr,
1904                          const struct sw_flow_key *key,
1905                          struct sw_flow_actions **sfa, bool log)
1906 {
1907         int err;
1908
1909         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
1910         if (IS_ERR(*sfa))
1911                 return PTR_ERR(*sfa);
1912
1913         err = __ovs_nla_copy_actions(attr, key, 0, sfa, key->eth.type,
1914                                      key->eth.tci, log);
1915         if (err)
1916                 kfree(*sfa);
1917
1918         return err;
1919 }
1920
1921 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1922 {
1923         const struct nlattr *a;
1924         struct nlattr *start;
1925         int err = 0, rem;
1926
1927         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1928         if (!start)
1929                 return -EMSGSIZE;
1930
1931         nla_for_each_nested(a, attr, rem) {
1932                 int type = nla_type(a);
1933                 struct nlattr *st_sample;
1934
1935                 switch (type) {
1936                 case OVS_SAMPLE_ATTR_PROBABILITY:
1937                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
1938                                     sizeof(u32), nla_data(a)))
1939                                 return -EMSGSIZE;
1940                         break;
1941                 case OVS_SAMPLE_ATTR_ACTIONS:
1942                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1943                         if (!st_sample)
1944                                 return -EMSGSIZE;
1945                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
1946                         if (err)
1947                                 return err;
1948                         nla_nest_end(skb, st_sample);
1949                         break;
1950                 }
1951         }
1952
1953         nla_nest_end(skb, start);
1954         return err;
1955 }
1956
1957 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1958 {
1959         const struct nlattr *ovs_key = nla_data(a);
1960         int key_type = nla_type(ovs_key);
1961         struct nlattr *start;
1962         int err;
1963
1964         switch (key_type) {
1965         case OVS_KEY_ATTR_TUNNEL_INFO: {
1966                 struct ovs_tunnel_info *tun_info = nla_data(ovs_key);
1967
1968                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1969                 if (!start)
1970                         return -EMSGSIZE;
1971
1972                 err = ipv4_tun_to_nlattr(skb, &tun_info->tunnel,
1973                                          tun_info->options_len ?
1974                                                 tun_info->options : NULL,
1975                                          tun_info->options_len);
1976                 if (err)
1977                         return err;
1978                 nla_nest_end(skb, start);
1979                 break;
1980         }
1981         default:
1982                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1983                         return -EMSGSIZE;
1984                 break;
1985         }
1986
1987         return 0;
1988 }
1989
1990 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
1991 {
1992         const struct nlattr *a;
1993         int rem, err;
1994
1995         nla_for_each_attr(a, attr, len, rem) {
1996                 int type = nla_type(a);
1997
1998                 switch (type) {
1999                 case OVS_ACTION_ATTR_SET:
2000                         err = set_action_to_attr(a, skb);
2001                         if (err)
2002                                 return err;
2003                         break;
2004
2005                 case OVS_ACTION_ATTR_SAMPLE:
2006                         err = sample_action_to_attr(a, skb);
2007                         if (err)
2008                                 return err;
2009                         break;
2010                 default:
2011                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2012                                 return -EMSGSIZE;
2013                         break;
2014                 }
2015         }
2016
2017         return 0;
2018 }