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;
  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
  64                 range = &match->mask->range;
  65
  66         if (range->start == range->end) {
  67                 range->start = start;
  68                 range->end = end;
  69                 return;
  70         }
  71
  72         if (range->start > start)
  73                 range->start = start;
  74
  75         if (range->end < end)
  76                 range->end = end;
  77 }
  78
  79 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  80         do { \
  81                 update_range(match, offsetof(struct sw_flow_key, field),    \
  82                              sizeof((match)->key->field), is_mask);         \
  83                 if (is_mask)                                                \
  84                         (match)->mask->key.field = value;                   \
  85                 else                                                        \
  86                         (match)->key->field = value;                        \
  87         } while (0)
  88
  89 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
  90         do {                                                                \
  91                 update_range(match, offset, len, is_mask);                  \
  92                 if (is_mask)                                                \
  93                         memcpy((u8 *)&(match)->mask->key + offset, value_p, 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 = 1ULL << 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          */
 123         mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
 124                         | (1ULL << OVS_KEY_ATTR_IPV6)
 125                         | (1ULL << OVS_KEY_ATTR_TCP)
 126                         | (1ULL << OVS_KEY_ATTR_TCP_FLAGS)
 127                         | (1ULL << OVS_KEY_ATTR_UDP)
 128                         | (1ULL << OVS_KEY_ATTR_SCTP)
 129                         | (1ULL << OVS_KEY_ATTR_ICMP)
 130                         | (1ULL << OVS_KEY_ATTR_ICMPV6)
 131                         | (1ULL << OVS_KEY_ATTR_ARP)
 132                         | (1ULL << OVS_KEY_ATTR_ND)
 133                         | (1ULL << OVS_KEY_ATTR_MPLS));
 134
 135         /* Always allowed mask fields. */
 136         mask_allowed |= ((1ULL << OVS_KEY_ATTR_TUNNEL)
 137                        | (1ULL << OVS_KEY_ATTR_IN_PORT)
 138                        | (1ULL << OVS_KEY_ATTR_ETHERTYPE));
 139
 140         /* Check key attributes. */
 141         if (match->key->eth.type == htons(ETH_P_ARP)
 142                         || match->key->eth.type == htons(ETH_P_RARP)) {
 143                 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
 144                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 145                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
 146         }
 147
 148
 149         if (eth_p_mpls(match->key->eth.type)) {
 150                 key_expected |= 1ULL << OVS_KEY_ATTR_MPLS;
 151                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 152                         mask_allowed |= 1ULL << OVS_KEY_ATTR_MPLS;
 153         }
 154
 155         if (match->key->eth.type == htons(ETH_P_IP)) {
 156                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
 157                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 158                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
 159
 160                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 161                         if (match->key->ip.proto == IPPROTO_UDP) {
 162                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 163                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 164                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 165                         }
 166
 167                         if (match->key->ip.proto == IPPROTO_SCTP) {
 168                                 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
 169                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 170                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
 171                         }
 172
 173                         if (match->key->ip.proto == IPPROTO_TCP) {
 174                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 175                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 176                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 177                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 178                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 179                                 }
 180                         }
 181
 182                         if (match->key->ip.proto == IPPROTO_ICMP) {
 183                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
 184                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 185                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
 186                         }
 187                 }
 188         }
 189
 190         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 191                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
 192                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 193                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
 194
 195                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 196                         if (match->key->ip.proto == IPPROTO_UDP) {
 197                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 198                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 199                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 200                         }
 201
 202                         if (match->key->ip.proto == IPPROTO_SCTP) {
 203                                 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
 204                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 205                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
 206                         }
 207
 208                         if (match->key->ip.proto == IPPROTO_TCP) {
 209                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 210                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 211                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 212                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 213                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 214                                 }
 215                         }
 216
 217                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 218                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 219                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 220                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 221
 222                                 if (match->key->tp.src ==
 223                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 224                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 225                                         key_expected |= 1ULL << OVS_KEY_ATTR_ND;
 226                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 227                                                 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
 228                                 }
 229                         }
 230                 }
 231         }
 232
 233         if ((key_attrs & key_expected) != key_expected) {
 234                 /* Key attributes check failed. */
 235                 OVS_NLERR(log,
 236                           "Missing expected key attributes (key_attrs=%llx, expected=%llx).",
 237                           (unsigned long long)key_attrs,
 238                           (unsigned long long)key_expected);
 239                 return false;
 240         }
 241
 242         if ((mask_attrs & mask_allowed) != mask_attrs) {
 243                 /* Mask attributes check failed. */
 244                 OVS_NLERR(log,
 245                           "Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).",
 246                           (unsigned long long)mask_attrs,
 247                           (unsigned long long)mask_allowed);
 248                 return false;
 249         }
 250
 251         return true;
 252 }
 253
 254 size_t ovs_tun_key_attr_size(void)
 255 {
 256         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 257          * updating this function.
 258          */
 259         return    nla_total_size(8)    /* OVS_TUNNEL_KEY_ATTR_ID */
 260                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 261                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 262                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
 263                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
 264                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 265                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
 266                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
 267                 + nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 268                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
 269                 + nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
 270 }
 271
 272 size_t ovs_key_attr_size(void)
 273 {
 274         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 275          * updating this function.
 276          */
 277         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 22);
 278
 279         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 280                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 281                   + ovs_tun_key_attr_size()
 282                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 283                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 284                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 285                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 286                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 287                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 288                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 289                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 290                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 291                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 292                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 293                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 294 }
 295
 296 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 297 static const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 298         [OVS_KEY_ATTR_ENCAP] = -1,
 299         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 300         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 301         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
 302         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 303         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 304         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 305         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 306         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 307         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 308         [OVS_KEY_ATTR_TCP_FLAGS] = sizeof(__be16),
 309         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 310         [OVS_KEY_ATTR_SCTP] = sizeof(struct ovs_key_sctp),
 311         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 312         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 313         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 314         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 315         [OVS_KEY_ATTR_DP_HASH] = sizeof(u32),
 316         [OVS_KEY_ATTR_RECIRC_ID] = sizeof(u32),
 317         [OVS_KEY_ATTR_TUNNEL] = -1,
 318         [OVS_KEY_ATTR_MPLS] = sizeof(struct ovs_key_mpls),
 319 };
 320
 321 static bool is_all_zero(const u8 *fp, size_t size)
 322 {
 323         int i;
 324
 325         if (!fp)
 326                 return false;
 327
 328         for (i = 0; i < size; i++)
 329                 if (fp[i])
 330                         return false;
 331
 332         return true;
 333 }
 334
 335 static int __parse_flow_nlattrs(const struct nlattr *attr,
 336                                 const struct nlattr *a[],
 337                                 u64 *attrsp, bool log, bool nz)
 338 {
 339         const struct nlattr *nla;
 340         u64 attrs;
 341         int rem;
 342
 343         attrs = *attrsp;
 344         nla_for_each_nested(nla, attr, rem) {
 345                 u16 type = nla_type(nla);
 346                 int expected_len;
 347
 348                 if (type > OVS_KEY_ATTR_MAX) {
 349                         OVS_NLERR(log,
 350                                   "Unknown key attribute (type=%d, max=%d).",
 351                                   type, OVS_KEY_ATTR_MAX);
 352                         return -EINVAL;
 353                 }
 354
 355                 if (attrs & (1ULL << type)) {
 356                         OVS_NLERR(log, "Duplicate key attribute (type %d).",
 357                                   type);
 358                         return -EINVAL;
 359                 }
 360
 361                 expected_len = ovs_key_lens[type];
 362                 if (nla_len(nla) != expected_len && expected_len != -1) {
 363                         OVS_NLERR(log, "Key attribute has unexpected length "
 364                                   "(type=%d, length=%d, expected=%d).", type,
 365                                   nla_len(nla), expected_len);
 366                         return -EINVAL;
 367                 }
 368
 369                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 370                         attrs |= 1ULL << type;
 371                         a[type] = nla;
 372                 }
 373         }
 374         if (rem) {
 375                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 376                 return -EINVAL;
 377         }
 378
 379         *attrsp = attrs;
 380         return 0;
 381 }
 382
 383 static int parse_flow_mask_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, true);
 388 }
 389
 390 static int parse_flow_nlattrs(const struct nlattr *attr,
 391                               const struct nlattr *a[], u64 *attrsp,
 392                               bool log)
 393 {
 394         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 395 }
 396
 397 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
 398                                 struct sw_flow_match *match, bool is_mask,
 399                                 bool log)
 400 {
 401         struct nlattr *a;
 402         int rem;
 403         bool ttl = false;
 404         __be16 tun_flags = 0;
 405
 406         nla_for_each_nested(a, attr, rem) {
 407                 int type = nla_type(a);
 408                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 409                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
 410                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
 411                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
 412                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
 413                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
 414                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
 415                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
 416                         [OVS_TUNNEL_KEY_ATTR_TP_SRC] = sizeof(u16),
 417                         [OVS_TUNNEL_KEY_ATTR_TP_DST] = sizeof(u16),
 418                         [OVS_TUNNEL_KEY_ATTR_OAM] = 0,
 419                         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS] = -1,
 420                 };
 421
 422                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 423                         OVS_NLERR(log,
 424                                   "Unknown tunnel attribute (type=%d, max=%d).",
 425                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 426                         return -EINVAL;
 427                 }
 428
 429                 if (ovs_tunnel_key_lens[type] != nla_len(a) &&
 430                     ovs_tunnel_key_lens[type] != -1) {
 431                         OVS_NLERR(log,
 432                                   "Tunnel attribute has unexpected length "
 433                                   "(type=%d, length=%d, expected=%d).",
 434                                   type, nla_len(a),
 435                                   ovs_tunnel_key_lens[type]);
 436                         return -EINVAL;
 437                 }
 438
 439                 switch (type) {
 440                 case OVS_TUNNEL_KEY_ATTR_ID:
 441                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 442                                         nla_get_be64(a), is_mask);
 443                         tun_flags |= TUNNEL_KEY;
 444                         break;
 445                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 446                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
 447                                         nla_get_be32(a), is_mask);
 448                         break;
 449                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 450                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
 451                                         nla_get_be32(a), is_mask);
 452                         break;
 453                 case OVS_TUNNEL_KEY_ATTR_TOS:
 454                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
 455                                         nla_get_u8(a), is_mask);
 456                         break;
 457                 case OVS_TUNNEL_KEY_ATTR_TTL:
 458                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
 459                                         nla_get_u8(a), is_mask);
 460                         ttl = true;
 461                         break;
 462                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 463                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 464                         break;
 465                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 466                         tun_flags |= TUNNEL_CSUM;
 467                         break;
 468                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 469                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 470                                         nla_get_be16(a), is_mask);
 471                         break;
 472                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 473                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 474                                         nla_get_be16(a), is_mask);
 475                         break;
 476                 case OVS_TUNNEL_KEY_ATTR_OAM:
 477                         tun_flags |= TUNNEL_OAM;
 478                         break;
 479                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 480                         tun_flags |= TUNNEL_OPTIONS_PRESENT;
 481                         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 482                                 OVS_NLERR(log, "Geneve option too long "
 483                                           "(len %d, max %zu).",
 484                                           nla_len(a),
 485                                           sizeof(match->key->tun_opts));
 486                                 return -EINVAL;
 487                         }
 488
 489                         if (nla_len(a) % 4 != 0) {
 490                                 OVS_NLERR(log, "Geneve option length is not "
 491                                           "a multiple of 4 (len %d).",
 492                                           nla_len(a));
 493                                 return -EINVAL;
 494                         }
 495
 496                         /* We need to record the length of the options passed
 497                          * down, otherwise packets with the same format but
 498                          * additional options will be silently matched.
 499                          */
 500                         if (!is_mask) {
 501                                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 502                                                 false);
 503                         } else {
 504                                 /* This is somewhat unusual because it looks at
 505                                  * both the key and mask while parsing the
 506                                  * attributes (and by extension assumes the key
 507                                  * is parsed first). Normally, we would verify
 508                                  * that each is the correct length and that the
 509                                  * attributes line up in the validate function.
 510                                  * However, that is difficult because this is
 511                                  * variable length and we won't have the
 512                                  * information later.
 513                                  */
 514                                 if (match->key->tun_opts_len != nla_len(a)) {
 515                                         OVS_NLERR(log,
 516                                                   "Geneve option key length (%d)"
 517                                                   " is different from mask length (%d).",
 518                                                   match->key->tun_opts_len,
 519                                                   nla_len(a));
 520                                         return -EINVAL;
 521                                 }
 522
 523                                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff,
 524                                                 true);
 525                         }
 526
 527                         SW_FLOW_KEY_MEMCPY_OFFSET(match,
 528                                 (unsigned long)GENEVE_OPTS((struct sw_flow_key *)0,
 529                                                            nla_len(a)),
 530                                 nla_data(a), nla_len(a), is_mask);
 531                         break;
 532                 default:
 533                         OVS_NLERR(log, "Unknown IPv4 tunnel attribute (%d).",
 534                                   type);
 535                         return -EINVAL;
 536                 }
 537         }
 538
 539         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 540
 541         if (rem > 0) {
 542                 OVS_NLERR(log, "IPv4 tunnel attribute has %d unknown bytes.",
 543                           rem);
 544                 return -EINVAL;
 545         }
 546
 547         if (!is_mask) {
 548                 if (!match->key->tun_key.ipv4_dst) {
 549                         OVS_NLERR(log,
 550                                   "IPv4 tunnel destination address is zero.");
 551                         return -EINVAL;
 552                 }
 553
 554                 if (!ttl) {
 555                         OVS_NLERR(log, "IPv4 tunnel TTL not specified.");
 556                         return -EINVAL;
 557                 }
 558         }
 559
 560         return 0;
 561 }
 562
 563 static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
 564                                 const struct ovs_key_ipv4_tunnel *output,
 565                                 const struct geneve_opt *tun_opts,
 566                                 int swkey_tun_opts_len)
 567 {
 568         if (output->tun_flags & TUNNEL_KEY &&
 569             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 570                 return -EMSGSIZE;
 571         if (output->ipv4_src &&
 572                 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
 573                 return -EMSGSIZE;
 574         if (output->ipv4_dst &&
 575                 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
 576                 return -EMSGSIZE;
 577         if (output->ipv4_tos &&
 578                 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
 579                 return -EMSGSIZE;
 580         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
 581                 return -EMSGSIZE;
 582         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 583                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 584                 return -EMSGSIZE;
 585         if ((output->tun_flags & TUNNEL_CSUM) &&
 586                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 587                 return -EMSGSIZE;
 588         if (output->tp_src &&
 589                 nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 590                 return -EMSGSIZE;
 591         if (output->tp_dst &&
 592                 nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 593                 return -EMSGSIZE;
 594         if ((output->tun_flags & TUNNEL_OAM) &&
 595                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 596                 return -EMSGSIZE;
 597         if (tun_opts &&
 598             nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 599                     swkey_tun_opts_len, tun_opts))
 600                 return -EMSGSIZE;
 601
 602         return 0;
 603 }
 604
 605
 606 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
 607                               const struct ovs_key_ipv4_tunnel *output,
 608                               const struct geneve_opt *tun_opts,
 609                               int swkey_tun_opts_len)
 610 {
 611         struct nlattr *nla;
 612         int err;
 613
 614         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 615         if (!nla)
 616                 return -EMSGSIZE;
 617
 618         err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
 619         if (err)
 620                 return err;
 621
 622         nla_nest_end(skb, nla);
 623         return 0;
 624 }
 625
 626 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
 627                                   const struct ovs_tunnel_info *egress_tun_info)
 628 {
 629         return __ipv4_tun_to_nlattr(skb, &egress_tun_info->tunnel,
 630                                     egress_tun_info->options,
 631                                     egress_tun_info->options_len);
 632 }
 633
 634 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
 635                                  const struct nlattr **a, bool is_mask,
 636                                  bool log)
 637 {
 638         if (*attrs & (1ULL << OVS_KEY_ATTR_DP_HASH)) {
 639                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 640
 641                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 642                 *attrs &= ~(1ULL << OVS_KEY_ATTR_DP_HASH);
 643         }
 644
 645         if (*attrs & (1ULL << OVS_KEY_ATTR_RECIRC_ID)) {
 646                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 647
 648                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 649                 *attrs &= ~(1ULL << OVS_KEY_ATTR_RECIRC_ID);
 650         }
 651
 652         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
 653                 SW_FLOW_KEY_PUT(match, phy.priority,
 654                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 655                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
 656         }
 657
 658         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
 659                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 660
 661                 if (is_mask) {
 662                         in_port = 0xffffffff; /* Always exact match in_port. */
 663                 } else if (in_port >= DP_MAX_PORTS) {
 664                         OVS_NLERR(log,
 665                                   "Input port (%d) exceeds maximum allowable (%d).",
 666                                   in_port, DP_MAX_PORTS);
 667                         return -EINVAL;
 668                 }
 669
 670                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 671                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
 672         } else if (!is_mask) {
 673                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 674         }
 675
 676         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
 677                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 678
 679                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 680                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
 681         }
 682         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
 683                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 684                                          is_mask, log))
 685                         return -EINVAL;
 686                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
 687         }
 688         return 0;
 689 }
 690
 691 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
 692                                 const struct nlattr **a, bool is_mask,
 693                                 bool log)
 694 {
 695         int err;
 696
 697         err = metadata_from_nlattrs(match, &attrs, a, is_mask, log);
 698         if (err)
 699                 return err;
 700
 701         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
 702                 const struct ovs_key_ethernet *eth_key;
 703
 704                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 705                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 706                                 eth_key->eth_src, ETH_ALEN, is_mask);
 707                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 708                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 709                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
 710         }
 711
 712         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
 713                 __be16 tci;
 714
 715                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 716                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 717                         if (is_mask)
 718                                 OVS_NLERR(log,
 719                                           "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 720                         else
 721                                 OVS_NLERR(log,
 722                                           "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 723
 724                         return -EINVAL;
 725                 }
 726
 727                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 728                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
 729         }
 730
 731         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
 732                 __be16 eth_type;
 733
 734                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 735                 if (is_mask) {
 736                         /* Always exact match EtherType. */
 737                         eth_type = htons(0xffff);
 738                 } else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
 739                         OVS_NLERR(log,
 740                                   "EtherType is less than minimum (type=%x, min=%x).",
 741                                   ntohs(eth_type), ETH_P_802_3_MIN);
 742                         return -EINVAL;
 743                 }
 744
 745                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 746                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
 747         } else if (!is_mask) {
 748                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 749         }
 750
 751         if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
 752                 const struct ovs_key_ipv4 *ipv4_key;
 753
 754                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 755                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 756                         OVS_NLERR(log,
 757                                   "Unknown IPv4 fragment type (value=%d, max=%d).",
 758                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 759                         return -EINVAL;
 760                 }
 761                 SW_FLOW_KEY_PUT(match, ip.proto,
 762                                 ipv4_key->ipv4_proto, is_mask);
 763                 SW_FLOW_KEY_PUT(match, ip.tos,
 764                                 ipv4_key->ipv4_tos, is_mask);
 765                 SW_FLOW_KEY_PUT(match, ip.ttl,
 766                                 ipv4_key->ipv4_ttl, is_mask);
 767                 SW_FLOW_KEY_PUT(match, ip.frag,
 768                                 ipv4_key->ipv4_frag, is_mask);
 769                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 770                                 ipv4_key->ipv4_src, is_mask);
 771                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 772                                 ipv4_key->ipv4_dst, is_mask);
 773                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
 774         }
 775
 776         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
 777                 const struct ovs_key_ipv6 *ipv6_key;
 778
 779                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 780                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 781                         OVS_NLERR(log,
 782                                   "Unknown IPv6 fragment type (value=%d, max=%d).",
 783                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 784                         return -EINVAL;
 785                 }
 786                 if (ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 787                         OVS_NLERR(log,
 788                                   "Invalid IPv6 flow label value (value=%x, max=%x).",
 789                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 790                         return -EINVAL;
 791                 }
 792                 SW_FLOW_KEY_PUT(match, ipv6.label,
 793                                 ipv6_key->ipv6_label, is_mask);
 794                 SW_FLOW_KEY_PUT(match, ip.proto,
 795                                 ipv6_key->ipv6_proto, is_mask);
 796                 SW_FLOW_KEY_PUT(match, ip.tos,
 797                                 ipv6_key->ipv6_tclass, is_mask);
 798                 SW_FLOW_KEY_PUT(match, ip.ttl,
 799                                 ipv6_key->ipv6_hlimit, is_mask);
 800                 SW_FLOW_KEY_PUT(match, ip.frag,
 801                                 ipv6_key->ipv6_frag, is_mask);
 802                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
 803                                 ipv6_key->ipv6_src,
 804                                 sizeof(match->key->ipv6.addr.src),
 805                                 is_mask);
 806                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
 807                                 ipv6_key->ipv6_dst,
 808                                 sizeof(match->key->ipv6.addr.dst),
 809                                 is_mask);
 810
 811                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
 812         }
 813
 814         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
 815                 const struct ovs_key_arp *arp_key;
 816
 817                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
 818                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
 819                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
 820                                   arp_key->arp_op);
 821                         return -EINVAL;
 822                 }
 823
 824                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 825                                 arp_key->arp_sip, is_mask);
 826                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 827                         arp_key->arp_tip, is_mask);
 828                 SW_FLOW_KEY_PUT(match, ip.proto,
 829                                 ntohs(arp_key->arp_op), is_mask);
 830                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
 831                                 arp_key->arp_sha, ETH_ALEN, is_mask);
 832                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
 833                                 arp_key->arp_tha, ETH_ALEN, is_mask);
 834
 835                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
 836         }
 837
 838         if (attrs & (1ULL << OVS_KEY_ATTR_MPLS)) {
 839                 const struct ovs_key_mpls *mpls_key;
 840
 841                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
 842                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
 843                                 mpls_key->mpls_lse, is_mask);
 844
 845                 attrs &= ~(1ULL << OVS_KEY_ATTR_MPLS);
 846         }
 847
 848         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
 849                 const struct ovs_key_tcp *tcp_key;
 850
 851                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 852                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
 853                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
 854                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
 855         }
 856
 857         if (attrs & (1ULL << OVS_KEY_ATTR_TCP_FLAGS)) {
 858                 SW_FLOW_KEY_PUT(match, tp.flags,
 859                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
 860                                 is_mask);
 861                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP_FLAGS);
 862         }
 863
 864         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
 865                 const struct ovs_key_udp *udp_key;
 866
 867                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 868                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
 869                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
 870                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
 871         }
 872
 873         if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
 874                 const struct ovs_key_sctp *sctp_key;
 875
 876                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
 877                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
 878                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
 879                 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
 880         }
 881
 882         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
 883                 const struct ovs_key_icmp *icmp_key;
 884
 885                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 886                 SW_FLOW_KEY_PUT(match, tp.src,
 887                                 htons(icmp_key->icmp_type), is_mask);
 888                 SW_FLOW_KEY_PUT(match, tp.dst,
 889                                 htons(icmp_key->icmp_code), is_mask);
 890                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
 891         }
 892
 893         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
 894                 const struct ovs_key_icmpv6 *icmpv6_key;
 895
 896                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 897                 SW_FLOW_KEY_PUT(match, tp.src,
 898                                 htons(icmpv6_key->icmpv6_type), is_mask);
 899                 SW_FLOW_KEY_PUT(match, tp.dst,
 900                                 htons(icmpv6_key->icmpv6_code), is_mask);
 901                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
 902         }
 903
 904         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
 905                 const struct ovs_key_nd *nd_key;
 906
 907                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 908                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
 909                         nd_key->nd_target,
 910                         sizeof(match->key->ipv6.nd.target),
 911                         is_mask);
 912                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
 913                         nd_key->nd_sll, ETH_ALEN, is_mask);
 914                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
 915                                 nd_key->nd_tll, ETH_ALEN, is_mask);
 916                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
 917         }
 918
 919         if (attrs != 0) {
 920                 OVS_NLERR(log, "Unknown key attributes (%llx).",
 921                           (unsigned long long)attrs);
 922                 return -EINVAL;
 923         }
 924
 925         return 0;
 926 }
 927
 928 static void nlattr_set(struct nlattr *attr, u8 val, bool is_attr_mask_key)
 929 {
 930         struct nlattr *nla;
 931         int rem;
 932
 933         /* The nlattr stream should already have been validated */
 934         nla_for_each_nested(nla, attr, rem) {
 935                 /* We assume that ovs_key_lens[type] == -1 means that type is a
 936                  * nested attribute
 937                  */
 938                 if (is_attr_mask_key && ovs_key_lens[nla_type(nla)] == -1)
 939                         nlattr_set(nla, val, false);
 940                 else
 941                         memset(nla_data(nla), val, nla_len(nla));
 942         }
 943 }
 944
 945 static void mask_set_nlattr(struct nlattr *attr, u8 val)
 946 {
 947         nlattr_set(attr, val, true);
 948 }
 949
 950 /**
 951  * ovs_nla_get_match - parses Netlink attributes into a flow key and
 952  * mask. In case the 'mask' is NULL, the flow is treated as exact match
 953  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
 954  * does not include any don't care bit.
 955  * @match: receives the extracted flow match information.
 956  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 957  * sequence. The fields should of the packet that triggered the creation
 958  * of this flow.
 959  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
 960  * attribute specifies the mask field of the wildcarded flow.
 961  * @log: Boolean to allow kernel error logging.  Normally true, but when
 962  * probing for feature compatibility this should be passed in as false to
 963  * suppress unnecessary error logging.
 964  */
 965 int ovs_nla_get_match(struct sw_flow_match *match,
 966                       const struct nlattr *nla_key,
 967                       const struct nlattr *nla_mask,
 968                       bool log)
 969 {
 970         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 971         const struct nlattr *encap;
 972         struct nlattr *newmask = NULL;
 973         u64 key_attrs = 0;
 974         u64 mask_attrs = 0;
 975         bool encap_valid = false;
 976         int err;
 977
 978         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
 979         if (err)
 980                 return err;
 981
 982         if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
 983             (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
 984             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
 985                 __be16 tci;
 986
 987                 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
 988                       (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
 989                         OVS_NLERR(log, "Invalid Vlan frame.");
 990                         return -EINVAL;
 991                 }
 992
 993                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
 994                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 995                 encap = a[OVS_KEY_ATTR_ENCAP];
 996                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
 997                 encap_valid = true;
 998
 999                 if (tci & htons(VLAN_TAG_PRESENT)) {
1000                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1001                         if (err)
1002                                 return err;
1003                 } else if (!tci) {
1004                         /* Corner case for truncated 802.1Q header. */
1005                         if (nla_len(encap)) {
1006                                 OVS_NLERR(log,
1007                                           "Truncated 802.1Q header has non-zero encap attribute.");
1008                                 return -EINVAL;
1009                         }
1010                 } else {
1011                         OVS_NLERR(log,
1012                                   "Encap attribute is set for a non-VLAN frame.");
1013                         return  -EINVAL;
1014                 }
1015         }
1016
1017         err = ovs_key_from_nlattrs(match, key_attrs, a, false, log);
1018         if (err)
1019                 return err;
1020
1021         if (match->mask) {
1022
1023                 if (!nla_mask) {
1024                         /* Create an exact match mask. We need to set to 0xff
1025                          * all the 'match->mask' fields that have been touched
1026                          * in 'match->key'. We cannot simply memset
1027                          * 'match->mask', because padding bytes and fields not
1028                          * specified in 'match->key' should be left to 0.
1029                          * Instead, we use a stream of netlink attributes,
1030                          * copied from 'key' and set to 0xff: ovs_key_from_nlattrs()
1031                          * will take care of filling 'match->mask'
1032                          * appropriately.
1033                          */
1034                         newmask = kmemdup(nla_key,
1035                                           nla_total_size(nla_len(nla_key)),
1036                                           GFP_KERNEL);
1037                         if (!newmask)
1038                                 return -ENOMEM;
1039
1040                         mask_set_nlattr(newmask, 0xff);
1041
1042                         /* The userspace does not send tunnel attributes that
1043                          * are 0, but we should not wildcard them nonetheless.
1044                          */
1045                         if (match->key->tun_key.ipv4_dst)
1046                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1047                                                          0xff, true);
1048
1049                         nla_mask = newmask;
1050                 }
1051
1052                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1053                 if (err)
1054                         goto free_newmask;
1055
1056                 /* Always match on tci. */
1057                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1058
1059                 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1060                         __be16 eth_type = 0;
1061                         __be16 tci = 0;
1062
1063                         if (!encap_valid) {
1064                                 OVS_NLERR(log,
1065                                           "Encap mask attribute is set for non-VLAN frame.");
1066                                 err = -EINVAL;
1067                                 goto free_newmask;
1068                         }
1069
1070                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1071                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1072                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1073
1074                         if (eth_type == htons(0xffff)) {
1075                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1076                                 encap = a[OVS_KEY_ATTR_ENCAP];
1077                                 err = parse_flow_mask_nlattrs(encap, a,
1078                                                               &mask_attrs, log);
1079                                 if (err)
1080                                         goto free_newmask;
1081                         } else {
1082                                 OVS_NLERR(log,
1083                                           "VLAN frames must have an exact match on the TPID (mask=%x).",
1084                                           ntohs(eth_type));
1085                                 err = -EINVAL;
1086                                 goto free_newmask;
1087                         }
1088
1089                         if (a[OVS_KEY_ATTR_VLAN])
1090                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1091
1092                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1093                                 OVS_NLERR(log,
1094                                           "VLAN tag present bit must have an exact match (tci_mask=%x).",
1095                                           ntohs(tci));
1096                                 err = -EINVAL;
1097                                 goto free_newmask;
1098                         }
1099                 }
1100
1101                 err = ovs_key_from_nlattrs(match, mask_attrs, a, true, log);
1102                 if (err)
1103                         goto free_newmask;
1104         }
1105
1106         if (!match_validate(match, key_attrs, mask_attrs, log))
1107                 err = -EINVAL;
1108
1109 free_newmask:
1110         kfree(newmask);
1111         return err;
1112 }
1113
1114 /**
1115  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1116  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1117  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1118  * sequence.
1119  * @log: Boolean to allow kernel error logging.  Normally true, but when
1120  * probing for feature compatibility this should be passed in as false to
1121  * suppress unnecessary error logging.
1122  *
1123  * This parses a series of Netlink attributes that form a flow key, which must
1124  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1125  * get the metadata, that is, the parts of the flow key that cannot be
1126  * extracted from the packet itself.
1127  */
1128 int ovs_nla_get_flow_metadata(const struct nlattr *attr,
1129                               struct sw_flow_key *key,
1130                               bool log)
1131 {
1132         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1133         struct sw_flow_match match;
1134         u64 attrs = 0;
1135         int err;
1136
1137         err = parse_flow_nlattrs(attr, a, &attrs, log);
1138         if (err)
1139                 return -EINVAL;
1140
1141         memset(&match, 0, sizeof(match));
1142         match.key = key;
1143
1144         memset(key, 0, OVS_SW_FLOW_KEY_METADATA_SIZE);
1145         key->phy.in_port = DP_MAX_PORTS;
1146
1147         return metadata_from_nlattrs(&match, &attrs, a, false, log);
1148 }
1149
1150 int ovs_nla_put_flow(const struct sw_flow_key *swkey,
1151                      const struct sw_flow_key *output, struct sk_buff *skb)
1152 {
1153         struct ovs_key_ethernet *eth_key;
1154         struct nlattr *nla, *encap;
1155         bool is_mask = (swkey != output);
1156
1157         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1158                 goto nla_put_failure;
1159
1160         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1161                 goto nla_put_failure;
1162
1163         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1164                 goto nla_put_failure;
1165
1166         if ((swkey->tun_key.ipv4_dst || is_mask)) {
1167                 const struct geneve_opt *opts = NULL;
1168
1169                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1170                         opts = GENEVE_OPTS(output, swkey->tun_opts_len);
1171
1172                 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1173                                         swkey->tun_opts_len))
1174                         goto nla_put_failure;
1175         }
1176
1177         if (swkey->phy.in_port == DP_MAX_PORTS) {
1178                 if (is_mask && (output->phy.in_port == 0xffff))
1179                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1180                                 goto nla_put_failure;
1181         } else {
1182                 u16 upper_u16;
1183                 upper_u16 = !is_mask ? 0 : 0xffff;
1184
1185                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1186                                 (upper_u16 << 16) | output->phy.in_port))
1187                         goto nla_put_failure;
1188         }
1189
1190         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1191                 goto nla_put_failure;
1192
1193         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1194         if (!nla)
1195                 goto nla_put_failure;
1196
1197         eth_key = nla_data(nla);
1198         ether_addr_copy(eth_key->eth_src, output->eth.src);
1199         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1200
1201         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1202                 __be16 eth_type;
1203                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1204                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1205                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1206                         goto nla_put_failure;
1207                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1208                 if (!swkey->eth.tci)
1209                         goto unencap;
1210         } else
1211                 encap = NULL;
1212
1213         if (swkey->eth.type == htons(ETH_P_802_2)) {
1214                 /*
1215                  * Ethertype 802.2 is represented in the netlink with omitted
1216                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1217                  * 0xffff in the mask attribute.  Ethertype can also
1218                  * be wildcarded.
1219                  */
1220                 if (is_mask && output->eth.type)
1221                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1222                                                 output->eth.type))
1223                                 goto nla_put_failure;
1224                 goto unencap;
1225         }
1226
1227         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1228                 goto nla_put_failure;
1229
1230         if (swkey->eth.type == htons(ETH_P_IP)) {
1231                 struct ovs_key_ipv4 *ipv4_key;
1232
1233                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1234                 if (!nla)
1235                         goto nla_put_failure;
1236                 ipv4_key = nla_data(nla);
1237                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1238                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1239                 ipv4_key->ipv4_proto = output->ip.proto;
1240                 ipv4_key->ipv4_tos = output->ip.tos;
1241                 ipv4_key->ipv4_ttl = output->ip.ttl;
1242                 ipv4_key->ipv4_frag = output->ip.frag;
1243         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1244                 struct ovs_key_ipv6 *ipv6_key;
1245
1246                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1247                 if (!nla)
1248                         goto nla_put_failure;
1249                 ipv6_key = nla_data(nla);
1250                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1251                                 sizeof(ipv6_key->ipv6_src));
1252                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1253                                 sizeof(ipv6_key->ipv6_dst));
1254                 ipv6_key->ipv6_label = output->ipv6.label;
1255                 ipv6_key->ipv6_proto = output->ip.proto;
1256                 ipv6_key->ipv6_tclass = output->ip.tos;
1257                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1258                 ipv6_key->ipv6_frag = output->ip.frag;
1259         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1260                    swkey->eth.type == htons(ETH_P_RARP)) {
1261                 struct ovs_key_arp *arp_key;
1262
1263                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1264                 if (!nla)
1265                         goto nla_put_failure;
1266                 arp_key = nla_data(nla);
1267                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1268                 arp_key->arp_sip = output->ipv4.addr.src;
1269                 arp_key->arp_tip = output->ipv4.addr.dst;
1270                 arp_key->arp_op = htons(output->ip.proto);
1271                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1272                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1273         } else if (eth_p_mpls(swkey->eth.type)) {
1274                 struct ovs_key_mpls *mpls_key;
1275
1276                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1277                 if (!nla)
1278                         goto nla_put_failure;
1279                 mpls_key = nla_data(nla);
1280                 mpls_key->mpls_lse = output->mpls.top_lse;
1281         }
1282
1283         if ((swkey->eth.type == htons(ETH_P_IP) ||
1284              swkey->eth.type == htons(ETH_P_IPV6)) &&
1285              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1286
1287                 if (swkey->ip.proto == IPPROTO_TCP) {
1288                         struct ovs_key_tcp *tcp_key;
1289
1290                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1291                         if (!nla)
1292                                 goto nla_put_failure;
1293                         tcp_key = nla_data(nla);
1294                         tcp_key->tcp_src = output->tp.src;
1295                         tcp_key->tcp_dst = output->tp.dst;
1296                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1297                                          output->tp.flags))
1298                                 goto nla_put_failure;
1299                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1300                         struct ovs_key_udp *udp_key;
1301
1302                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1303                         if (!nla)
1304                                 goto nla_put_failure;
1305                         udp_key = nla_data(nla);
1306                         udp_key->udp_src = output->tp.src;
1307                         udp_key->udp_dst = output->tp.dst;
1308                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1309                         struct ovs_key_sctp *sctp_key;
1310
1311                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1312                         if (!nla)
1313                                 goto nla_put_failure;
1314                         sctp_key = nla_data(nla);
1315                         sctp_key->sctp_src = output->tp.src;
1316                         sctp_key->sctp_dst = output->tp.dst;
1317                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1318                            swkey->ip.proto == IPPROTO_ICMP) {
1319                         struct ovs_key_icmp *icmp_key;
1320
1321                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1322                         if (!nla)
1323                                 goto nla_put_failure;
1324                         icmp_key = nla_data(nla);
1325                         icmp_key->icmp_type = ntohs(output->tp.src);
1326                         icmp_key->icmp_code = ntohs(output->tp.dst);
1327                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1328                            swkey->ip.proto == IPPROTO_ICMPV6) {
1329                         struct ovs_key_icmpv6 *icmpv6_key;
1330
1331                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1332                                                 sizeof(*icmpv6_key));
1333                         if (!nla)
1334                                 goto nla_put_failure;
1335                         icmpv6_key = nla_data(nla);
1336                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1337                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1338
1339                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1340                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1341                                 struct ovs_key_nd *nd_key;
1342
1343                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1344                                 if (!nla)
1345                                         goto nla_put_failure;
1346                                 nd_key = nla_data(nla);
1347                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1348                                                         sizeof(nd_key->nd_target));
1349                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1350                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1351                         }
1352                 }
1353         }
1354
1355 unencap:
1356         if (encap)
1357                 nla_nest_end(skb, encap);
1358
1359         return 0;
1360
1361 nla_put_failure:
1362         return -EMSGSIZE;
1363 }
1364
1365 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1366
1367 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1368 {
1369         struct sw_flow_actions *sfa;
1370
1371         if (size > MAX_ACTIONS_BUFSIZE) {
1372                 OVS_NLERR(log, "Flow action size (%u bytes) exceeds maximum "
1373                           "(%u bytes)", size, MAX_ACTIONS_BUFSIZE);
1374                 return ERR_PTR(-EINVAL);
1375         }
1376
1377         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1378         if (!sfa)
1379                 return ERR_PTR(-ENOMEM);
1380
1381         sfa->actions_len = 0;
1382         return sfa;
1383 }
1384
1385 /* RCU callback used by ovs_nla_free_flow_actions. */
1386 static void rcu_free_acts_callback(struct rcu_head *rcu)
1387 {
1388         struct sw_flow_actions *sf_acts = container_of(rcu,
1389                         struct sw_flow_actions, rcu);
1390         kfree(sf_acts);
1391 }
1392
1393 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1394  * The caller must hold rcu_read_lock for this to be sensible.
1395  */
1396 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1397 {
1398         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
1399 }
1400
1401 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1402                                        int attr_len, bool log)
1403 {
1404
1405         struct sw_flow_actions *acts;
1406         int new_acts_size;
1407         int req_size = NLA_ALIGN(attr_len);
1408         int next_offset = offsetof(struct sw_flow_actions, actions) +
1409                                         (*sfa)->actions_len;
1410
1411         if (req_size <= (ksize(*sfa) - next_offset))
1412                 goto out;
1413
1414         new_acts_size = ksize(*sfa) * 2;
1415
1416         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1417                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1418                         return ERR_PTR(-EMSGSIZE);
1419                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1420         }
1421
1422         acts = nla_alloc_flow_actions(new_acts_size, log);
1423         if (IS_ERR(acts))
1424                 return (void *)acts;
1425
1426         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1427         acts->actions_len = (*sfa)->actions_len;
1428         kfree(*sfa);
1429         *sfa = acts;
1430
1431 out:
1432         (*sfa)->actions_len += req_size;
1433         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1434 }
1435
1436 static struct nlattr *__add_action(struct sw_flow_actions **sfa, int attrtype,
1437                                    void *data, int len, bool log)
1438 {
1439         struct nlattr *a;
1440
1441         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1442         if (IS_ERR(a))
1443                 return a;
1444
1445         a->nla_type = attrtype;
1446         a->nla_len = nla_attr_size(len);
1447
1448         if (data)
1449                 memcpy(nla_data(a), data, len);
1450         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1451
1452         return a;
1453 }
1454
1455 static int add_action(struct sw_flow_actions **sfa, int attrtype,
1456                       void *data, int len, bool log)
1457 {
1458         struct nlattr *a;
1459
1460         a = __add_action(sfa, attrtype, data, len, log);
1461         if (IS_ERR(a))
1462                 return PTR_ERR(a);
1463
1464         return 0;
1465 }
1466
1467 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1468                                           int attrtype, bool log)
1469 {
1470         int used = (*sfa)->actions_len;
1471         int err;
1472
1473         err = add_action(sfa, attrtype, NULL, 0, log);
1474         if (err)
1475                 return err;
1476
1477         return used;
1478 }
1479
1480 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1481                                          int st_offset)
1482 {
1483         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1484                                                                st_offset);
1485
1486         a->nla_len = sfa->actions_len - st_offset;
1487 }
1488
1489 static int __ovs_nla_copy_actions(const struct nlattr *attr,
1490                                   const struct sw_flow_key *key,
1491                                   int depth, struct sw_flow_actions **sfa,
1492                                   __be16 eth_type, __be16 vlan_tci, bool log);
1493
1494 static int validate_and_copy_sample(const struct nlattr *attr,
1495                                     const struct sw_flow_key *key, int depth,
1496                                     struct sw_flow_actions **sfa,
1497                                     __be16 eth_type, __be16 vlan_tci, bool log)
1498 {
1499         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1500         const struct nlattr *probability, *actions;
1501         const struct nlattr *a;
1502         int rem, start, err, st_acts;
1503
1504         memset(attrs, 0, sizeof(attrs));
1505         nla_for_each_nested(a, attr, rem) {
1506                 int type = nla_type(a);
1507                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1508                         return -EINVAL;
1509                 attrs[type] = a;
1510         }
1511         if (rem)
1512                 return -EINVAL;
1513
1514         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1515         if (!probability || nla_len(probability) != sizeof(u32))
1516                 return -EINVAL;
1517
1518         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1519         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1520                 return -EINVAL;
1521
1522         /* validation done, copy sample action. */
1523         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1524         if (start < 0)
1525                 return start;
1526         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1527                          nla_data(probability), sizeof(u32), log);
1528         if (err)
1529                 return err;
1530         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1531         if (st_acts < 0)
1532                 return st_acts;
1533
1534         err = __ovs_nla_copy_actions(actions, key, depth + 1, sfa,
1535                                      eth_type, vlan_tci, log);
1536         if (err)
1537                 return err;
1538
1539         add_nested_action_end(*sfa, st_acts);
1540         add_nested_action_end(*sfa, start);
1541
1542         return 0;
1543 }
1544
1545 static int validate_tp_port(const struct sw_flow_key *flow_key,
1546                             __be16 eth_type)
1547 {
1548         if ((eth_type == htons(ETH_P_IP) || eth_type == htons(ETH_P_IPV6)) &&
1549             (flow_key->tp.src || flow_key->tp.dst))
1550                 return 0;
1551
1552         return -EINVAL;
1553 }
1554
1555 void ovs_match_init(struct sw_flow_match *match,
1556                     struct sw_flow_key *key,
1557                     struct sw_flow_mask *mask)
1558 {
1559         memset(match, 0, sizeof(*match));
1560         match->key = key;
1561         match->mask = mask;
1562
1563         memset(key, 0, sizeof(*key));
1564
1565         if (mask) {
1566                 memset(&mask->key, 0, sizeof(mask->key));
1567                 mask->range.start = mask->range.end = 0;
1568         }
1569 }
1570
1571 static int validate_and_copy_set_tun(const struct nlattr *attr,
1572                                      struct sw_flow_actions **sfa, bool log)
1573 {
1574         struct sw_flow_match match;
1575         struct sw_flow_key key;
1576         struct ovs_tunnel_info *tun_info;
1577         struct nlattr *a;
1578         int err, start;
1579
1580         ovs_match_init(&match, &key, NULL);
1581         err = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1582         if (err)
1583                 return err;
1584
1585         if (key.tun_opts_len) {
1586                 struct geneve_opt *option = GENEVE_OPTS(&key,
1587                                                         key.tun_opts_len);
1588                 int opts_len = key.tun_opts_len;
1589                 bool crit_opt = false;
1590
1591                 while (opts_len > 0) {
1592                         int len;
1593
1594                         if (opts_len < sizeof(*option))
1595                                 return -EINVAL;
1596
1597                         len = sizeof(*option) + option->length * 4;
1598                         if (len > opts_len)
1599                                 return -EINVAL;
1600
1601                         crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1602
1603                         option = (struct geneve_opt *)((u8 *)option + len);
1604                         opts_len -= len;
1605                 };
1606
1607                 key.tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1608         };
1609
1610         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1611         if (start < 0)
1612                 return start;
1613
1614         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1615                          sizeof(*tun_info) + key.tun_opts_len, log);
1616         if (IS_ERR(a))
1617                 return PTR_ERR(a);
1618
1619         tun_info = nla_data(a);
1620         tun_info->tunnel = key.tun_key;
1621         tun_info->options_len = key.tun_opts_len;
1622
1623         if (tun_info->options_len) {
1624                 /* We need to store the options in the action itself since
1625                  * everything else will go away after flow setup. We can append
1626                  * it to tun_info and then point there.
1627                  */
1628                 memcpy((tun_info + 1), GENEVE_OPTS(&key, key.tun_opts_len),
1629                         key.tun_opts_len);
1630                 tun_info->options = (struct geneve_opt *)(tun_info + 1);
1631         } else {
1632                 tun_info->options = NULL;
1633         }
1634
1635         add_nested_action_end(*sfa, start);
1636
1637         return err;
1638 }
1639
1640 static int validate_set(const struct nlattr *a,
1641                         const struct sw_flow_key *flow_key,
1642                         struct sw_flow_actions **sfa,
1643                         bool *set_tun, __be16 eth_type, bool log)
1644 {
1645         const struct nlattr *ovs_key = nla_data(a);
1646         int key_type = nla_type(ovs_key);
1647
1648         /* There can be only one key in a action */
1649         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1650                 return -EINVAL;
1651
1652         if (key_type > OVS_KEY_ATTR_MAX ||
1653             (ovs_key_lens[key_type] != nla_len(ovs_key) &&
1654              ovs_key_lens[key_type] != -1))
1655                 return -EINVAL;
1656
1657         switch (key_type) {
1658         const struct ovs_key_ipv4 *ipv4_key;
1659         const struct ovs_key_ipv6 *ipv6_key;
1660         int err;
1661
1662         case OVS_KEY_ATTR_PRIORITY:
1663         case OVS_KEY_ATTR_SKB_MARK:
1664         case OVS_KEY_ATTR_ETHERNET:
1665                 break;
1666
1667         case OVS_KEY_ATTR_TUNNEL:
1668                 *set_tun = true;
1669                 err = validate_and_copy_set_tun(a, sfa, log);
1670                 if (err)
1671                         return err;
1672                 break;
1673
1674         case OVS_KEY_ATTR_IPV4:
1675                 if (eth_type != htons(ETH_P_IP))
1676                         return -EINVAL;
1677
1678                 if (!flow_key->ip.proto)
1679                         return -EINVAL;
1680
1681                 ipv4_key = nla_data(ovs_key);
1682                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1683                         return -EINVAL;
1684
1685                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1686                         return -EINVAL;
1687
1688                 break;
1689
1690         case OVS_KEY_ATTR_IPV6:
1691                 if (eth_type != htons(ETH_P_IPV6))
1692                         return -EINVAL;
1693
1694                 if (!flow_key->ip.proto)
1695                         return -EINVAL;
1696
1697                 ipv6_key = nla_data(ovs_key);
1698                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1699                         return -EINVAL;
1700
1701                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1702                         return -EINVAL;
1703
1704                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
1705                         return -EINVAL;
1706
1707                 break;
1708
1709         case OVS_KEY_ATTR_TCP:
1710                 if (flow_key->ip.proto != IPPROTO_TCP)
1711                         return -EINVAL;
1712
1713                 return validate_tp_port(flow_key, eth_type);
1714
1715         case OVS_KEY_ATTR_UDP:
1716                 if (flow_key->ip.proto != IPPROTO_UDP)
1717                         return -EINVAL;
1718
1719                 return validate_tp_port(flow_key, eth_type);
1720
1721         case OVS_KEY_ATTR_MPLS:
1722                 if (!eth_p_mpls(eth_type))
1723                         return -EINVAL;
1724                 break;
1725
1726         case OVS_KEY_ATTR_SCTP:
1727                 if (flow_key->ip.proto != IPPROTO_SCTP)
1728                         return -EINVAL;
1729
1730                 return validate_tp_port(flow_key, eth_type);
1731
1732         default:
1733                 return -EINVAL;
1734         }
1735
1736         return 0;
1737 }
1738
1739 static int validate_userspace(const struct nlattr *attr)
1740 {
1741         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
1742                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
1743                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
1744                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
1745         };
1746         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
1747         int error;
1748
1749         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
1750                                  attr, userspace_policy);
1751         if (error)
1752                 return error;
1753
1754         if (!a[OVS_USERSPACE_ATTR_PID] ||
1755             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
1756                 return -EINVAL;
1757
1758         return 0;
1759 }
1760
1761 static int copy_action(const struct nlattr *from,
1762                        struct sw_flow_actions **sfa, bool log)
1763 {
1764         int totlen = NLA_ALIGN(from->nla_len);
1765         struct nlattr *to;
1766
1767         to = reserve_sfa_size(sfa, from->nla_len, log);
1768         if (IS_ERR(to))
1769                 return PTR_ERR(to);
1770
1771         memcpy(to, from, totlen);
1772         return 0;
1773 }
1774
1775 static int __ovs_nla_copy_actions(const struct nlattr *attr,
1776                                   const struct sw_flow_key *key,
1777                                   int depth, struct sw_flow_actions **sfa,
1778                                   __be16 eth_type, __be16 vlan_tci, bool log)
1779 {
1780         const struct nlattr *a;
1781         int rem, err;
1782
1783         if (depth >= SAMPLE_ACTION_DEPTH)
1784                 return -EOVERFLOW;
1785
1786         nla_for_each_nested(a, attr, rem) {
1787                 /* Expected argument lengths, (u32)-1 for variable length. */
1788                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
1789                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
1790                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
1791                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
1792                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
1793                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
1794                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
1795                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
1796                         [OVS_ACTION_ATTR_SET] = (u32)-1,
1797                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
1798                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash)
1799                 };
1800                 const struct ovs_action_push_vlan *vlan;
1801                 int type = nla_type(a);
1802                 bool skip_copy;
1803
1804                 if (type > OVS_ACTION_ATTR_MAX ||
1805                     (action_lens[type] != nla_len(a) &&
1806                      action_lens[type] != (u32)-1))
1807                         return -EINVAL;
1808
1809                 skip_copy = false;
1810                 switch (type) {
1811                 case OVS_ACTION_ATTR_UNSPEC:
1812                         return -EINVAL;
1813
1814                 case OVS_ACTION_ATTR_USERSPACE:
1815                         err = validate_userspace(a);
1816                         if (err)
1817                                 return err;
1818                         break;
1819
1820                 case OVS_ACTION_ATTR_OUTPUT:
1821                         if (nla_get_u32(a) >= DP_MAX_PORTS)
1822                                 return -EINVAL;
1823                         break;
1824
1825                 case OVS_ACTION_ATTR_HASH: {
1826                         const struct ovs_action_hash *act_hash = nla_data(a);
1827
1828                         switch (act_hash->hash_alg) {
1829                         case OVS_HASH_ALG_L4:
1830                                 break;
1831                         default:
1832                                 return  -EINVAL;
1833                         }
1834
1835                         break;
1836                 }
1837
1838                 case OVS_ACTION_ATTR_POP_VLAN:
1839                         vlan_tci = htons(0);
1840                         break;
1841
1842                 case OVS_ACTION_ATTR_PUSH_VLAN:
1843                         vlan = nla_data(a);
1844                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
1845                                 return -EINVAL;
1846                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
1847                                 return -EINVAL;
1848                         vlan_tci = vlan->vlan_tci;
1849                         break;
1850
1851                 case OVS_ACTION_ATTR_RECIRC:
1852                         break;
1853
1854                 case OVS_ACTION_ATTR_PUSH_MPLS: {
1855                         const struct ovs_action_push_mpls *mpls = nla_data(a);
1856
1857                         if (!eth_p_mpls(mpls->mpls_ethertype))
1858                                 return -EINVAL;
1859                         /* Prohibit push MPLS other than to a white list
1860                          * for packets that have a known tag order.
1861                          */
1862                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
1863                             (eth_type != htons(ETH_P_IP) &&
1864                              eth_type != htons(ETH_P_IPV6) &&
1865                              eth_type != htons(ETH_P_ARP) &&
1866                              eth_type != htons(ETH_P_RARP) &&
1867                              !eth_p_mpls(eth_type)))
1868                                 return -EINVAL;
1869                         eth_type = mpls->mpls_ethertype;
1870                         break;
1871                 }
1872
1873                 case OVS_ACTION_ATTR_POP_MPLS:
1874                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
1875                             !eth_p_mpls(eth_type))
1876                                 return -EINVAL;
1877
1878                         /* Disallow subsequent L2.5+ set and mpls_pop actions
1879                          * as there is no check here to ensure that the new
1880                          * eth_type is valid and thus set actions could
1881                          * write off the end of the packet or otherwise
1882                          * corrupt it.
1883                          *
1884                          * Support for these actions is planned using packet
1885                          * recirculation.
1886                          */
1887                         eth_type = htons(0);
1888                         break;
1889
1890                 case OVS_ACTION_ATTR_SET:
1891                         err = validate_set(a, key, sfa, &skip_copy, eth_type,
1892                                            log);
1893                         if (err)
1894                                 return err;
1895                         break;
1896
1897                 case OVS_ACTION_ATTR_SAMPLE:
1898                         err = validate_and_copy_sample(a, key, depth, sfa,
1899                                                        eth_type, vlan_tci, log);
1900                         if (err)
1901                                 return err;
1902                         skip_copy = true;
1903                         break;
1904
1905                 default:
1906                         return -EINVAL;
1907                 }
1908                 if (!skip_copy) {
1909                         err = copy_action(a, sfa, log);
1910                         if (err)
1911                                 return err;
1912                 }
1913         }
1914
1915         if (rem > 0)
1916                 return -EINVAL;
1917
1918         return 0;
1919 }
1920
1921 int ovs_nla_copy_actions(const struct nlattr *attr,
1922                          const struct sw_flow_key *key,
1923                          struct sw_flow_actions **sfa, bool log)
1924 {
1925         int err;
1926
1927         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
1928         if (IS_ERR(*sfa))
1929                 return PTR_ERR(*sfa);
1930
1931         err = __ovs_nla_copy_actions(attr, key, 0, sfa, key->eth.type,
1932                                      key->eth.tci, log);
1933         if (err)
1934                 kfree(*sfa);
1935
1936         return err;
1937 }
1938
1939 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1940 {
1941         const struct nlattr *a;
1942         struct nlattr *start;
1943         int err = 0, rem;
1944
1945         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1946         if (!start)
1947                 return -EMSGSIZE;
1948
1949         nla_for_each_nested(a, attr, rem) {
1950                 int type = nla_type(a);
1951                 struct nlattr *st_sample;
1952
1953                 switch (type) {
1954                 case OVS_SAMPLE_ATTR_PROBABILITY:
1955                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
1956                                     sizeof(u32), nla_data(a)))
1957                                 return -EMSGSIZE;
1958                         break;
1959                 case OVS_SAMPLE_ATTR_ACTIONS:
1960                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1961                         if (!st_sample)
1962                                 return -EMSGSIZE;
1963                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
1964                         if (err)
1965                                 return err;
1966                         nla_nest_end(skb, st_sample);
1967                         break;
1968                 }
1969         }
1970
1971         nla_nest_end(skb, start);
1972         return err;
1973 }
1974
1975 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1976 {
1977         const struct nlattr *ovs_key = nla_data(a);
1978         int key_type = nla_type(ovs_key);
1979         struct nlattr *start;
1980         int err;
1981
1982         switch (key_type) {
1983         case OVS_KEY_ATTR_TUNNEL_INFO: {
1984                 struct ovs_tunnel_info *tun_info = nla_data(ovs_key);
1985
1986                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1987                 if (!start)
1988                         return -EMSGSIZE;
1989
1990                 err = ipv4_tun_to_nlattr(skb, &tun_info->tunnel,
1991                                          tun_info->options_len ?
1992                                                 tun_info->options : NULL,
1993                                          tun_info->options_len);
1994                 if (err)
1995                         return err;
1996                 nla_nest_end(skb, start);
1997                 break;
1998         }
1999         default:
2000                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2001                         return -EMSGSIZE;
2002                 break;
2003         }
2004
2005         return 0;
2006 }
2007
2008 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2009 {
2010         const struct nlattr *a;
2011         int rem, err;
2012
2013         nla_for_each_attr(a, attr, len, rem) {
2014                 int type = nla_type(a);
2015
2016                 switch (type) {
2017                 case OVS_ACTION_ATTR_SET:
2018                         err = set_action_to_attr(a, skb);
2019                         if (err)
2020                                 return err;
2021                         break;
2022
2023                 case OVS_ACTION_ATTR_SAMPLE:
2024                         err = sample_action_to_attr(a, skb);
2025                         if (err)
2026                                 return err;
2027                         break;
2028                 default:
2029                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2030                                 return -EMSGSIZE;
2031                         break;
2032                 }
2033         }
2034
2035         return 0;
2036 }