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 #include <net/vxlan.h>
  51
  52 #include "flow_netlink.h"
  53
  54 struct ovs_len_tbl {
  55         int len;
  56         const struct ovs_len_tbl *next;
  57 };
  58
  59 #define OVS_ATTR_NESTED -1
  60
  61 static void update_range(struct sw_flow_match *match,
  62                          size_t offset, size_t size, bool is_mask)
  63 {
  64         struct sw_flow_key_range *range;
  65         size_t start = rounddown(offset, sizeof(long));
  66         size_t end = roundup(offset + size, sizeof(long));
  67
  68         if (!is_mask)
  69                 range = &match->range;
  70         else
  71                 range = &match->mask->range;
  72
  73         if (range->start == range->end) {
  74                 range->start = start;
  75                 range->end = end;
  76                 return;
  77         }
  78
  79         if (range->start > start)
  80                 range->start = start;
  81
  82         if (range->end < end)
  83                 range->end = end;
  84 }
  85
  86 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  87         do { \
  88                 update_range(match, offsetof(struct sw_flow_key, field),    \
  89                              sizeof((match)->key->field), is_mask);         \
  90                 if (is_mask)                                                \
  91                         (match)->mask->key.field = value;                   \
  92                 else                                                        \
  93                         (match)->key->field = value;                        \
  94         } while (0)
  95
  96 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
  97         do {                                                                \
  98                 update_range(match, offset, len, is_mask);                  \
  99                 if (is_mask)                                                \
 100                         memcpy((u8 *)&(match)->mask->key + offset, value_p, \
 101                                len);                                       \
 102                 else                                                        \
 103                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
 104         } while (0)
 105
 106 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
 107         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 108                                   value_p, len, is_mask)
 109
 110 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 111         do {                                                                \
 112                 update_range(match, offsetof(struct sw_flow_key, field),    \
 113                              sizeof((match)->key->field), is_mask);         \
 114                 if (is_mask)                                                \
 115                         memset((u8 *)&(match)->mask->key.field, value,      \
 116                                sizeof((match)->mask->key.field));           \
 117                 else                                                        \
 118                         memset((u8 *)&(match)->key->field, value,           \
 119                                sizeof((match)->key->field));                \
 120         } while (0)
 121
 122 static bool match_validate(const struct sw_flow_match *match,
 123                            u64 key_attrs, u64 mask_attrs, bool log)
 124 {
 125         u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
 126         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 127
 128         /* The following mask attributes allowed only if they
 129          * pass the validation tests. */
 130         mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
 131                         | (1 << OVS_KEY_ATTR_IPV6)
 132                         | (1 << OVS_KEY_ATTR_TCP)
 133                         | (1 << OVS_KEY_ATTR_TCP_FLAGS)
 134                         | (1 << OVS_KEY_ATTR_UDP)
 135                         | (1 << OVS_KEY_ATTR_SCTP)
 136                         | (1 << OVS_KEY_ATTR_ICMP)
 137                         | (1 << OVS_KEY_ATTR_ICMPV6)
 138                         | (1 << OVS_KEY_ATTR_ARP)
 139                         | (1 << OVS_KEY_ATTR_ND)
 140                         | (1 << OVS_KEY_ATTR_MPLS));
 141
 142         /* Always allowed mask fields. */
 143         mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
 144                        | (1 << OVS_KEY_ATTR_IN_PORT)
 145                        | (1 << OVS_KEY_ATTR_ETHERTYPE));
 146
 147         /* Check key attributes. */
 148         if (match->key->eth.type == htons(ETH_P_ARP)
 149                         || match->key->eth.type == htons(ETH_P_RARP)) {
 150                 key_expected |= 1 << OVS_KEY_ATTR_ARP;
 151                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 152                         mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
 153         }
 154
 155         if (eth_p_mpls(match->key->eth.type)) {
 156                 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
 157                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 158                         mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
 159         }
 160
 161         if (match->key->eth.type == htons(ETH_P_IP)) {
 162                 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
 163                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 164                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
 165
 166                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 167                         if (match->key->ip.proto == IPPROTO_UDP) {
 168                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 169                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 170                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 171                         }
 172
 173                         if (match->key->ip.proto == IPPROTO_SCTP) {
 174                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 175                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 176                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 177                         }
 178
 179                         if (match->key->ip.proto == IPPROTO_TCP) {
 180                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 181                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 182                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 183                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 184                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 185                                 }
 186                         }
 187
 188                         if (match->key->ip.proto == IPPROTO_ICMP) {
 189                                 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
 190                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 191                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
 192                         }
 193                 }
 194         }
 195
 196         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 197                 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
 198                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 199                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
 200
 201                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 202                         if (match->key->ip.proto == IPPROTO_UDP) {
 203                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 204                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 205                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 206                         }
 207
 208                         if (match->key->ip.proto == IPPROTO_SCTP) {
 209                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 210                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 211                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 212                         }
 213
 214                         if (match->key->ip.proto == IPPROTO_TCP) {
 215                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 216                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 217                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 218                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 219                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 220                                 }
 221                         }
 222
 223                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 224                                 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
 225                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 226                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
 227
 228                                 if (match->key->tp.src ==
 229                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 230                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 231                                         key_expected |= 1 << OVS_KEY_ATTR_ND;
 232                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 233                                                 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
 234                                 }
 235                         }
 236                 }
 237         }
 238
 239         if ((key_attrs & key_expected) != key_expected) {
 240                 /* Key attributes check failed. */
 241                 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
 242                           (unsigned long long)key_attrs,
 243                           (unsigned long long)key_expected);
 244                 return false;
 245         }
 246
 247         if ((mask_attrs & mask_allowed) != mask_attrs) {
 248                 /* Mask attributes check failed. */
 249                 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
 250                           (unsigned long long)mask_attrs,
 251                           (unsigned long long)mask_allowed);
 252                 return false;
 253         }
 254
 255         return true;
 256 }
 257
 258 size_t ovs_tun_key_attr_size(void)
 259 {
 260         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 261          * updating this function.
 262          */
 263         return    nla_total_size(8)    /* OVS_TUNNEL_KEY_ATTR_ID */
 264                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 265                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 266                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
 267                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
 268                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 269                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
 270                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
 271                 + nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 272                 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
 273                  * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
 274                  */
 275                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
 276                 + nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
 277 }
 278
 279 size_t ovs_key_attr_size(void)
 280 {
 281         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 282          * updating this function.
 283          */
 284         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 25);
 285
 286         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 287                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 288                   + ovs_tun_key_attr_size()
 289                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 290                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 291                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 292                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 293                 + nla_total_size(1)   /* OVS_KEY_ATTR_CT_STATE */
 294                 + nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
 295                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
 296                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 297                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 298                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 299                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 300                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 301                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 302                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 303                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 304 }
 305
 306 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 307         [OVS_TUNNEL_KEY_ATTR_ID]            = { .len = sizeof(u64) },
 308         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC]      = { .len = sizeof(u32) },
 309         [OVS_TUNNEL_KEY_ATTR_IPV4_DST]      = { .len = sizeof(u32) },
 310         [OVS_TUNNEL_KEY_ATTR_TOS]           = { .len = 1 },
 311         [OVS_TUNNEL_KEY_ATTR_TTL]           = { .len = 1 },
 312         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
 313         [OVS_TUNNEL_KEY_ATTR_CSUM]          = { .len = 0 },
 314         [OVS_TUNNEL_KEY_ATTR_TP_SRC]        = { .len = sizeof(u16) },
 315         [OVS_TUNNEL_KEY_ATTR_TP_DST]        = { .len = sizeof(u16) },
 316         [OVS_TUNNEL_KEY_ATTR_OAM]           = { .len = 0 },
 317         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_NESTED },
 318         [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED },
 319 };
 320
 321 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 322 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 323         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 324         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 325         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 326         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 327         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 328         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 329         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 330         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 331         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 332         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 333         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 334         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 335         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 336         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 337         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 338         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 339         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 340         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 341         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 342         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 343                                      .next = ovs_tunnel_key_lens, },
 344         [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 345         [OVS_KEY_ATTR_CT_STATE]  = { .len = sizeof(u8) },
 346         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 347         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 348 };
 349
 350 static bool is_all_zero(const u8 *fp, size_t size)
 351 {
 352         int i;
 353
 354         if (!fp)
 355                 return false;
 356
 357         for (i = 0; i < size; i++)
 358                 if (fp[i])
 359                         return false;
 360
 361         return true;
 362 }
 363
 364 static int __parse_flow_nlattrs(const struct nlattr *attr,
 365                                 const struct nlattr *a[],
 366                                 u64 *attrsp, bool log, bool nz)
 367 {
 368         const struct nlattr *nla;
 369         u64 attrs;
 370         int rem;
 371
 372         attrs = *attrsp;
 373         nla_for_each_nested(nla, attr, rem) {
 374                 u16 type = nla_type(nla);
 375                 int expected_len;
 376
 377                 if (type > OVS_KEY_ATTR_MAX) {
 378                         OVS_NLERR(log, "Key type %d is out of range max %d",
 379                                   type, OVS_KEY_ATTR_MAX);
 380                         return -EINVAL;
 381                 }
 382
 383                 if (attrs & (1 << type)) {
 384                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 385                         return -EINVAL;
 386                 }
 387
 388                 expected_len = ovs_key_lens[type].len;
 389                 if (nla_len(nla) != expected_len && expected_len != OVS_ATTR_NESTED) {
 390                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 391                                   type, nla_len(nla), expected_len);
 392                         return -EINVAL;
 393                 }
 394
 395                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 396                         attrs |= 1 << type;
 397                         a[type] = nla;
 398                 }
 399         }
 400         if (rem) {
 401                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 402                 return -EINVAL;
 403         }
 404
 405         *attrsp = attrs;
 406         return 0;
 407 }
 408
 409 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 410                                    const struct nlattr *a[], u64 *attrsp,
 411                                    bool log)
 412 {
 413         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 414 }
 415
 416 static int parse_flow_nlattrs(const struct nlattr *attr,
 417                               const struct nlattr *a[], u64 *attrsp,
 418                               bool log)
 419 {
 420         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 421 }
 422
 423 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 424                                      struct sw_flow_match *match, bool is_mask,
 425                                      bool log)
 426 {
 427         unsigned long opt_key_offset;
 428
 429         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 430                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 431                           nla_len(a), sizeof(match->key->tun_opts));
 432                 return -EINVAL;
 433         }
 434
 435         if (nla_len(a) % 4 != 0) {
 436                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 437                           nla_len(a));
 438                 return -EINVAL;
 439         }
 440
 441         /* We need to record the length of the options passed
 442          * down, otherwise packets with the same format but
 443          * additional options will be silently matched.
 444          */
 445         if (!is_mask) {
 446                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 447                                 false);
 448         } else {
 449                 /* This is somewhat unusual because it looks at
 450                  * both the key and mask while parsing the
 451                  * attributes (and by extension assumes the key
 452                  * is parsed first). Normally, we would verify
 453                  * that each is the correct length and that the
 454                  * attributes line up in the validate function.
 455                  * However, that is difficult because this is
 456                  * variable length and we won't have the
 457                  * information later.
 458                  */
 459                 if (match->key->tun_opts_len != nla_len(a)) {
 460                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 461                                   match->key->tun_opts_len, nla_len(a));
 462                         return -EINVAL;
 463                 }
 464
 465                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 466         }
 467
 468         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 469         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 470                                   nla_len(a), is_mask);
 471         return 0;
 472 }
 473
 474 static const struct nla_policy vxlan_opt_policy[OVS_VXLAN_EXT_MAX + 1] = {
 475         [OVS_VXLAN_EXT_GBP]     = { .type = NLA_U32 },
 476 };
 477
 478 static int vxlan_tun_opt_from_nlattr(const struct nlattr *a,
 479                                      struct sw_flow_match *match, bool is_mask,
 480                                      bool log)
 481 {
 482         struct nlattr *tb[OVS_VXLAN_EXT_MAX+1];
 483         unsigned long opt_key_offset;
 484         struct vxlan_metadata opts;
 485         int err;
 486
 487         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 488
 489         err = nla_parse_nested(tb, OVS_VXLAN_EXT_MAX, a, vxlan_opt_policy);
 490         if (err < 0)
 491                 return err;
 492
 493         memset(&opts, 0, sizeof(opts));
 494
 495         if (tb[OVS_VXLAN_EXT_GBP])
 496                 opts.gbp = nla_get_u32(tb[OVS_VXLAN_EXT_GBP]);
 497
 498         if (!is_mask)
 499                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 500         else
 501                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 502
 503         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 504         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 505                                   is_mask);
 506         return 0;
 507 }
 508
 509 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
 510                                 struct sw_flow_match *match, bool is_mask,
 511                                 bool log)
 512 {
 513         struct nlattr *a;
 514         int rem;
 515         bool ttl = false;
 516         __be16 tun_flags = 0;
 517         int opts_type = 0;
 518
 519         nla_for_each_nested(a, attr, rem) {
 520                 int type = nla_type(a);
 521                 int err;
 522
 523                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 524                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 525                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 526                         return -EINVAL;
 527                 }
 528
 529                 if (ovs_tunnel_key_lens[type].len != nla_len(a) &&
 530                     ovs_tunnel_key_lens[type].len != OVS_ATTR_NESTED) {
 531                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 532                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 533                         return -EINVAL;
 534                 }
 535
 536                 switch (type) {
 537                 case OVS_TUNNEL_KEY_ATTR_ID:
 538                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 539                                         nla_get_be64(a), is_mask);
 540                         tun_flags |= TUNNEL_KEY;
 541                         break;
 542                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 543                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
 544                                         nla_get_in_addr(a), is_mask);
 545                         break;
 546                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 547                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
 548                                         nla_get_in_addr(a), is_mask);
 549                         break;
 550                 case OVS_TUNNEL_KEY_ATTR_TOS:
 551                         SW_FLOW_KEY_PUT(match, tun_key.tos,
 552                                         nla_get_u8(a), is_mask);
 553                         break;
 554                 case OVS_TUNNEL_KEY_ATTR_TTL:
 555                         SW_FLOW_KEY_PUT(match, tun_key.ttl,
 556                                         nla_get_u8(a), is_mask);
 557                         ttl = true;
 558                         break;
 559                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 560                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 561                         break;
 562                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 563                         tun_flags |= TUNNEL_CSUM;
 564                         break;
 565                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 566                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 567                                         nla_get_be16(a), is_mask);
 568                         break;
 569                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 570                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 571                                         nla_get_be16(a), is_mask);
 572                         break;
 573                 case OVS_TUNNEL_KEY_ATTR_OAM:
 574                         tun_flags |= TUNNEL_OAM;
 575                         break;
 576                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 577                         if (opts_type) {
 578                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 579                                 return -EINVAL;
 580                         }
 581
 582                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 583                         if (err)
 584                                 return err;
 585
 586                         tun_flags |= TUNNEL_GENEVE_OPT;
 587                         opts_type = type;
 588                         break;
 589                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 590                         if (opts_type) {
 591                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 592                                 return -EINVAL;
 593                         }
 594
 595                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 596                         if (err)
 597                                 return err;
 598
 599                         tun_flags |= TUNNEL_VXLAN_OPT;
 600                         opts_type = type;
 601                         break;
 602                 default:
 603                         OVS_NLERR(log, "Unknown IPv4 tunnel attribute %d",
 604                                   type);
 605                         return -EINVAL;
 606                 }
 607         }
 608
 609         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 610
 611         if (rem > 0) {
 612                 OVS_NLERR(log, "IPv4 tunnel attribute has %d unknown bytes.",
 613                           rem);
 614                 return -EINVAL;
 615         }
 616
 617         if (!is_mask) {
 618                 if (!match->key->tun_key.u.ipv4.dst) {
 619                         OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 620                         return -EINVAL;
 621                 }
 622
 623                 if (!ttl) {
 624                         OVS_NLERR(log, "IPv4 tunnel TTL not specified.");
 625                         return -EINVAL;
 626                 }
 627         }
 628
 629         return opts_type;
 630 }
 631
 632 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 633                                const void *tun_opts, int swkey_tun_opts_len)
 634 {
 635         const struct vxlan_metadata *opts = tun_opts;
 636         struct nlattr *nla;
 637
 638         nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 639         if (!nla)
 640                 return -EMSGSIZE;
 641
 642         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 643                 return -EMSGSIZE;
 644
 645         nla_nest_end(skb, nla);
 646         return 0;
 647 }
 648
 649 static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
 650                                 const struct ip_tunnel_key *output,
 651                                 const void *tun_opts, int swkey_tun_opts_len)
 652 {
 653         if (output->tun_flags & TUNNEL_KEY &&
 654             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 655                 return -EMSGSIZE;
 656         if (output->u.ipv4.src &&
 657             nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 658                             output->u.ipv4.src))
 659                 return -EMSGSIZE;
 660         if (output->u.ipv4.dst &&
 661             nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 662                             output->u.ipv4.dst))
 663                 return -EMSGSIZE;
 664         if (output->tos &&
 665             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 666                 return -EMSGSIZE;
 667         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 668                 return -EMSGSIZE;
 669         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 670             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 671                 return -EMSGSIZE;
 672         if ((output->tun_flags & TUNNEL_CSUM) &&
 673             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 674                 return -EMSGSIZE;
 675         if (output->tp_src &&
 676             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 677                 return -EMSGSIZE;
 678         if (output->tp_dst &&
 679             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 680                 return -EMSGSIZE;
 681         if ((output->tun_flags & TUNNEL_OAM) &&
 682             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 683                 return -EMSGSIZE;
 684         if (tun_opts) {
 685                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 686                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 687                             swkey_tun_opts_len, tun_opts))
 688                         return -EMSGSIZE;
 689                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 690                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 691                         return -EMSGSIZE;
 692         }
 693
 694         return 0;
 695 }
 696
 697 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
 698                               const struct ip_tunnel_key *output,
 699                               const void *tun_opts, int swkey_tun_opts_len)
 700 {
 701         struct nlattr *nla;
 702         int err;
 703
 704         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 705         if (!nla)
 706                 return -EMSGSIZE;
 707
 708         err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
 709         if (err)
 710                 return err;
 711
 712         nla_nest_end(skb, nla);
 713         return 0;
 714 }
 715
 716 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
 717                                   const struct ip_tunnel_info *egress_tun_info)
 718 {
 719         return __ipv4_tun_to_nlattr(skb, &egress_tun_info->key,
 720                                     egress_tun_info->options,
 721                                     egress_tun_info->options_len);
 722 }
 723
 724 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
 725                                  const struct nlattr **a, bool is_mask,
 726                                  bool log)
 727 {
 728         if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
 729                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 730
 731                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 732                 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
 733         }
 734
 735         if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
 736                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 737
 738                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 739                 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
 740         }
 741
 742         if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 743                 SW_FLOW_KEY_PUT(match, phy.priority,
 744                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 745                 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 746         }
 747
 748         if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 749                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 750
 751                 if (is_mask) {
 752                         in_port = 0xffffffff; /* Always exact match in_port. */
 753                 } else if (in_port >= DP_MAX_PORTS) {
 754                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 755                                   in_port, DP_MAX_PORTS);
 756                         return -EINVAL;
 757                 }
 758
 759                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 760                 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
 761         } else if (!is_mask) {
 762                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 763         }
 764
 765         if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
 766                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 767
 768                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 769                 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
 770         }
 771         if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
 772                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 773                                          is_mask, log) < 0)
 774                         return -EINVAL;
 775                 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
 776         }
 777
 778         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
 779             ovs_ct_verify(OVS_KEY_ATTR_CT_STATE)) {
 780                 u8 ct_state = nla_get_u8(a[OVS_KEY_ATTR_CT_STATE]);
 781
 782                 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
 783                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
 784         }
 785         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
 786             ovs_ct_verify(OVS_KEY_ATTR_CT_ZONE)) {
 787                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
 788
 789                 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
 790                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
 791         }
 792         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
 793             ovs_ct_verify(OVS_KEY_ATTR_CT_MARK)) {
 794                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
 795
 796                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
 797                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
 798         }
 799         return 0;
 800 }
 801
 802 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
 803                                 const struct nlattr **a, bool is_mask,
 804                                 bool log)
 805 {
 806         int err;
 807
 808         err = metadata_from_nlattrs(match, &attrs, a, is_mask, log);
 809         if (err)
 810                 return err;
 811
 812         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
 813                 const struct ovs_key_ethernet *eth_key;
 814
 815                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 816                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 817                                 eth_key->eth_src, ETH_ALEN, is_mask);
 818                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 819                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 820                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
 821         }
 822
 823         if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
 824                 __be16 tci;
 825
 826                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 827                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 828                         if (is_mask)
 829                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 830                         else
 831                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 832
 833                         return -EINVAL;
 834                 }
 835
 836                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 837                 attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
 838         }
 839
 840         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
 841                 __be16 eth_type;
 842
 843                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 844                 if (is_mask) {
 845                         /* Always exact match EtherType. */
 846                         eth_type = htons(0xffff);
 847                 } else if (!eth_proto_is_802_3(eth_type)) {
 848                         OVS_NLERR(log, "EtherType %x is less than min %x",
 849                                   ntohs(eth_type), ETH_P_802_3_MIN);
 850                         return -EINVAL;
 851                 }
 852
 853                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 854                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
 855         } else if (!is_mask) {
 856                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 857         }
 858
 859         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
 860                 const struct ovs_key_ipv4 *ipv4_key;
 861
 862                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 863                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 864                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 865                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 866                         return -EINVAL;
 867                 }
 868                 SW_FLOW_KEY_PUT(match, ip.proto,
 869                                 ipv4_key->ipv4_proto, is_mask);
 870                 SW_FLOW_KEY_PUT(match, ip.tos,
 871                                 ipv4_key->ipv4_tos, is_mask);
 872                 SW_FLOW_KEY_PUT(match, ip.ttl,
 873                                 ipv4_key->ipv4_ttl, is_mask);
 874                 SW_FLOW_KEY_PUT(match, ip.frag,
 875                                 ipv4_key->ipv4_frag, is_mask);
 876                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 877                                 ipv4_key->ipv4_src, is_mask);
 878                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 879                                 ipv4_key->ipv4_dst, is_mask);
 880                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
 881         }
 882
 883         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
 884                 const struct ovs_key_ipv6 *ipv6_key;
 885
 886                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 887                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 888                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 889                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 890                         return -EINVAL;
 891                 }
 892
 893                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 894                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
 895                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 896                         return -EINVAL;
 897                 }
 898
 899                 SW_FLOW_KEY_PUT(match, ipv6.label,
 900                                 ipv6_key->ipv6_label, is_mask);
 901                 SW_FLOW_KEY_PUT(match, ip.proto,
 902                                 ipv6_key->ipv6_proto, is_mask);
 903                 SW_FLOW_KEY_PUT(match, ip.tos,
 904                                 ipv6_key->ipv6_tclass, is_mask);
 905                 SW_FLOW_KEY_PUT(match, ip.ttl,
 906                                 ipv6_key->ipv6_hlimit, is_mask);
 907                 SW_FLOW_KEY_PUT(match, ip.frag,
 908                                 ipv6_key->ipv6_frag, is_mask);
 909                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
 910                                 ipv6_key->ipv6_src,
 911                                 sizeof(match->key->ipv6.addr.src),
 912                                 is_mask);
 913                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
 914                                 ipv6_key->ipv6_dst,
 915                                 sizeof(match->key->ipv6.addr.dst),
 916                                 is_mask);
 917
 918                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
 919         }
 920
 921         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
 922                 const struct ovs_key_arp *arp_key;
 923
 924                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
 925                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
 926                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
 927                                   arp_key->arp_op);
 928                         return -EINVAL;
 929                 }
 930
 931                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 932                                 arp_key->arp_sip, is_mask);
 933                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 934                         arp_key->arp_tip, is_mask);
 935                 SW_FLOW_KEY_PUT(match, ip.proto,
 936                                 ntohs(arp_key->arp_op), is_mask);
 937                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
 938                                 arp_key->arp_sha, ETH_ALEN, is_mask);
 939                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
 940                                 arp_key->arp_tha, ETH_ALEN, is_mask);
 941
 942                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
 943         }
 944
 945         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
 946                 const struct ovs_key_mpls *mpls_key;
 947
 948                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
 949                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
 950                                 mpls_key->mpls_lse, is_mask);
 951
 952                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
 953          }
 954
 955         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
 956                 const struct ovs_key_tcp *tcp_key;
 957
 958                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 959                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
 960                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
 961                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 962         }
 963
 964         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
 965                 SW_FLOW_KEY_PUT(match, tp.flags,
 966                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
 967                                 is_mask);
 968                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
 969         }
 970
 971         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
 972                 const struct ovs_key_udp *udp_key;
 973
 974                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 975                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
 976                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
 977                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 978         }
 979
 980         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
 981                 const struct ovs_key_sctp *sctp_key;
 982
 983                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
 984                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
 985                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
 986                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
 987         }
 988
 989         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
 990                 const struct ovs_key_icmp *icmp_key;
 991
 992                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 993                 SW_FLOW_KEY_PUT(match, tp.src,
 994                                 htons(icmp_key->icmp_type), is_mask);
 995                 SW_FLOW_KEY_PUT(match, tp.dst,
 996                                 htons(icmp_key->icmp_code), is_mask);
 997                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 998         }
 999
1000         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1001                 const struct ovs_key_icmpv6 *icmpv6_key;
1002
1003                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1004                 SW_FLOW_KEY_PUT(match, tp.src,
1005                                 htons(icmpv6_key->icmpv6_type), is_mask);
1006                 SW_FLOW_KEY_PUT(match, tp.dst,
1007                                 htons(icmpv6_key->icmpv6_code), is_mask);
1008                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1009         }
1010
1011         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1012                 const struct ovs_key_nd *nd_key;
1013
1014                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1015                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1016                         nd_key->nd_target,
1017                         sizeof(match->key->ipv6.nd.target),
1018                         is_mask);
1019                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1020                         nd_key->nd_sll, ETH_ALEN, is_mask);
1021                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1022                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1023                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1024         }
1025
1026         if (attrs != 0) {
1027                 OVS_NLERR(log, "Unknown key attributes %llx",
1028                           (unsigned long long)attrs);
1029                 return -EINVAL;
1030         }
1031
1032         return 0;
1033 }
1034
1035 static void nlattr_set(struct nlattr *attr, u8 val,
1036                        const struct ovs_len_tbl *tbl)
1037 {
1038         struct nlattr *nla;
1039         int rem;
1040
1041         /* The nlattr stream should already have been validated */
1042         nla_for_each_nested(nla, attr, rem) {
1043                 if (tbl && tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1044                         nlattr_set(nla, val, tbl[nla_type(nla)].next);
1045                 else
1046                         memset(nla_data(nla), val, nla_len(nla));
1047         }
1048 }
1049
1050 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1051 {
1052         nlattr_set(attr, val, ovs_key_lens);
1053 }
1054
1055 /**
1056  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1057  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1058  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1059  * does not include any don't care bit.
1060  * @match: receives the extracted flow match information.
1061  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1062  * sequence. The fields should of the packet that triggered the creation
1063  * of this flow.
1064  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1065  * attribute specifies the mask field of the wildcarded flow.
1066  * @log: Boolean to allow kernel error logging.  Normally true, but when
1067  * probing for feature compatibility this should be passed in as false to
1068  * suppress unnecessary error logging.
1069  */
1070 int ovs_nla_get_match(struct sw_flow_match *match,
1071                       const struct nlattr *nla_key,
1072                       const struct nlattr *nla_mask,
1073                       bool log)
1074 {
1075         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1076         const struct nlattr *encap;
1077         struct nlattr *newmask = NULL;
1078         u64 key_attrs = 0;
1079         u64 mask_attrs = 0;
1080         bool encap_valid = false;
1081         int err;
1082
1083         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1084         if (err)
1085                 return err;
1086
1087         if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1088             (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1089             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1090                 __be16 tci;
1091
1092                 if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1093                       (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1094                         OVS_NLERR(log, "Invalid Vlan frame.");
1095                         return -EINVAL;
1096                 }
1097
1098                 key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1099                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1100                 encap = a[OVS_KEY_ATTR_ENCAP];
1101                 key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1102                 encap_valid = true;
1103
1104                 if (tci & htons(VLAN_TAG_PRESENT)) {
1105                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1106                         if (err)
1107                                 return err;
1108                 } else if (!tci) {
1109                         /* Corner case for truncated 802.1Q header. */
1110                         if (nla_len(encap)) {
1111                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1112                                 return -EINVAL;
1113                         }
1114                 } else {
1115                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1116                         return  -EINVAL;
1117                 }
1118         }
1119
1120         err = ovs_key_from_nlattrs(match, key_attrs, a, false, log);
1121         if (err)
1122                 return err;
1123
1124         if (match->mask) {
1125                 if (!nla_mask) {
1126                         /* Create an exact match mask. We need to set to 0xff
1127                          * all the 'match->mask' fields that have been touched
1128                          * in 'match->key'. We cannot simply memset
1129                          * 'match->mask', because padding bytes and fields not
1130                          * specified in 'match->key' should be left to 0.
1131                          * Instead, we use a stream of netlink attributes,
1132                          * copied from 'key' and set to 0xff.
1133                          * ovs_key_from_nlattrs() will take care of filling
1134                          * 'match->mask' appropriately.
1135                          */
1136                         newmask = kmemdup(nla_key,
1137                                           nla_total_size(nla_len(nla_key)),
1138                                           GFP_KERNEL);
1139                         if (!newmask)
1140                                 return -ENOMEM;
1141
1142                         mask_set_nlattr(newmask, 0xff);
1143
1144                         /* The userspace does not send tunnel attributes that
1145                          * are 0, but we should not wildcard them nonetheless.
1146                          */
1147                         if (match->key->tun_key.u.ipv4.dst)
1148                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1149                                                          0xff, true);
1150
1151                         nla_mask = newmask;
1152                 }
1153
1154                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1155                 if (err)
1156                         goto free_newmask;
1157
1158                 /* Always match on tci. */
1159                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1160
1161                 if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
1162                         __be16 eth_type = 0;
1163                         __be16 tci = 0;
1164
1165                         if (!encap_valid) {
1166                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1167                                 err = -EINVAL;
1168                                 goto free_newmask;
1169                         }
1170
1171                         mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1172                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1173                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1174
1175                         if (eth_type == htons(0xffff)) {
1176                                 mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1177                                 encap = a[OVS_KEY_ATTR_ENCAP];
1178                                 err = parse_flow_mask_nlattrs(encap, a,
1179                                                               &mask_attrs, log);
1180                                 if (err)
1181                                         goto free_newmask;
1182                         } else {
1183                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1184                                           ntohs(eth_type));
1185                                 err = -EINVAL;
1186                                 goto free_newmask;
1187                         }
1188
1189                         if (a[OVS_KEY_ATTR_VLAN])
1190                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1191
1192                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1193                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1194                                           ntohs(tci));
1195                                 err = -EINVAL;
1196                                 goto free_newmask;
1197                         }
1198                 }
1199
1200                 err = ovs_key_from_nlattrs(match, mask_attrs, a, true, log);
1201                 if (err)
1202                         goto free_newmask;
1203         }
1204
1205         if (!match_validate(match, key_attrs, mask_attrs, log))
1206                 err = -EINVAL;
1207
1208 free_newmask:
1209         kfree(newmask);
1210         return err;
1211 }
1212
1213 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1214 {
1215         size_t len;
1216
1217         if (!attr)
1218                 return 0;
1219
1220         len = nla_len(attr);
1221         if (len < 1 || len > MAX_UFID_LENGTH) {
1222                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1223                           nla_len(attr), MAX_UFID_LENGTH);
1224                 return 0;
1225         }
1226
1227         return len;
1228 }
1229
1230 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1231  * or false otherwise.
1232  */
1233 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1234                       bool log)
1235 {
1236         sfid->ufid_len = get_ufid_len(attr, log);
1237         if (sfid->ufid_len)
1238                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1239
1240         return sfid->ufid_len;
1241 }
1242
1243 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1244                            const struct sw_flow_key *key, bool log)
1245 {
1246         struct sw_flow_key *new_key;
1247
1248         if (ovs_nla_get_ufid(sfid, ufid, log))
1249                 return 0;
1250
1251         /* If UFID was not provided, use unmasked key. */
1252         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1253         if (!new_key)
1254                 return -ENOMEM;
1255         memcpy(new_key, key, sizeof(*key));
1256         sfid->unmasked_key = new_key;
1257
1258         return 0;
1259 }
1260
1261 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1262 {
1263         return attr ? nla_get_u32(attr) : 0;
1264 }
1265
1266 /**
1267  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1268  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1269  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1270  * sequence.
1271  * @log: Boolean to allow kernel error logging.  Normally true, but when
1272  * probing for feature compatibility this should be passed in as false to
1273  * suppress unnecessary error logging.
1274  *
1275  * This parses a series of Netlink attributes that form a flow key, which must
1276  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1277  * get the metadata, that is, the parts of the flow key that cannot be
1278  * extracted from the packet itself.
1279  */
1280
1281 int ovs_nla_get_flow_metadata(const struct nlattr *attr,
1282                               struct sw_flow_key *key,
1283                               bool log)
1284 {
1285         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1286         struct sw_flow_match match;
1287         u64 attrs = 0;
1288         int err;
1289
1290         err = parse_flow_nlattrs(attr, a, &attrs, log);
1291         if (err)
1292                 return -EINVAL;
1293
1294         memset(&match, 0, sizeof(match));
1295         match.key = key;
1296
1297         memset(&key->ct, 0, sizeof(key->ct));
1298         key->phy.in_port = DP_MAX_PORTS;
1299
1300         return metadata_from_nlattrs(&match, &attrs, a, false, log);
1301 }
1302
1303 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1304                              const struct sw_flow_key *output, bool is_mask,
1305                              struct sk_buff *skb)
1306 {
1307         struct ovs_key_ethernet *eth_key;
1308         struct nlattr *nla, *encap;
1309
1310         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1311                 goto nla_put_failure;
1312
1313         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1314                 goto nla_put_failure;
1315
1316         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1317                 goto nla_put_failure;
1318
1319         if ((swkey->tun_key.u.ipv4.dst || is_mask)) {
1320                 const void *opts = NULL;
1321
1322                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1323                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1324
1325                 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1326                                        swkey->tun_opts_len))
1327                         goto nla_put_failure;
1328         }
1329
1330         if (swkey->phy.in_port == DP_MAX_PORTS) {
1331                 if (is_mask && (output->phy.in_port == 0xffff))
1332                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1333                                 goto nla_put_failure;
1334         } else {
1335                 u16 upper_u16;
1336                 upper_u16 = !is_mask ? 0 : 0xffff;
1337
1338                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1339                                 (upper_u16 << 16) | output->phy.in_port))
1340                         goto nla_put_failure;
1341         }
1342
1343         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1344                 goto nla_put_failure;
1345
1346         if (ovs_ct_put_key(output, skb))
1347                 goto nla_put_failure;
1348
1349         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1350         if (!nla)
1351                 goto nla_put_failure;
1352
1353         eth_key = nla_data(nla);
1354         ether_addr_copy(eth_key->eth_src, output->eth.src);
1355         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1356
1357         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1358                 __be16 eth_type;
1359                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1360                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1361                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1362                         goto nla_put_failure;
1363                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1364                 if (!swkey->eth.tci)
1365                         goto unencap;
1366         } else
1367                 encap = NULL;
1368
1369         if (swkey->eth.type == htons(ETH_P_802_2)) {
1370                 /*
1371                  * Ethertype 802.2 is represented in the netlink with omitted
1372                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1373                  * 0xffff in the mask attribute.  Ethertype can also
1374                  * be wildcarded.
1375                  */
1376                 if (is_mask && output->eth.type)
1377                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1378                                                 output->eth.type))
1379                                 goto nla_put_failure;
1380                 goto unencap;
1381         }
1382
1383         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1384                 goto nla_put_failure;
1385
1386         if (swkey->eth.type == htons(ETH_P_IP)) {
1387                 struct ovs_key_ipv4 *ipv4_key;
1388
1389                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1390                 if (!nla)
1391                         goto nla_put_failure;
1392                 ipv4_key = nla_data(nla);
1393                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1394                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1395                 ipv4_key->ipv4_proto = output->ip.proto;
1396                 ipv4_key->ipv4_tos = output->ip.tos;
1397                 ipv4_key->ipv4_ttl = output->ip.ttl;
1398                 ipv4_key->ipv4_frag = output->ip.frag;
1399         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1400                 struct ovs_key_ipv6 *ipv6_key;
1401
1402                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1403                 if (!nla)
1404                         goto nla_put_failure;
1405                 ipv6_key = nla_data(nla);
1406                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1407                                 sizeof(ipv6_key->ipv6_src));
1408                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1409                                 sizeof(ipv6_key->ipv6_dst));
1410                 ipv6_key->ipv6_label = output->ipv6.label;
1411                 ipv6_key->ipv6_proto = output->ip.proto;
1412                 ipv6_key->ipv6_tclass = output->ip.tos;
1413                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1414                 ipv6_key->ipv6_frag = output->ip.frag;
1415         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1416                    swkey->eth.type == htons(ETH_P_RARP)) {
1417                 struct ovs_key_arp *arp_key;
1418
1419                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1420                 if (!nla)
1421                         goto nla_put_failure;
1422                 arp_key = nla_data(nla);
1423                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1424                 arp_key->arp_sip = output->ipv4.addr.src;
1425                 arp_key->arp_tip = output->ipv4.addr.dst;
1426                 arp_key->arp_op = htons(output->ip.proto);
1427                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1428                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1429         } else if (eth_p_mpls(swkey->eth.type)) {
1430                 struct ovs_key_mpls *mpls_key;
1431
1432                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1433                 if (!nla)
1434                         goto nla_put_failure;
1435                 mpls_key = nla_data(nla);
1436                 mpls_key->mpls_lse = output->mpls.top_lse;
1437         }
1438
1439         if ((swkey->eth.type == htons(ETH_P_IP) ||
1440              swkey->eth.type == htons(ETH_P_IPV6)) &&
1441              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1442
1443                 if (swkey->ip.proto == IPPROTO_TCP) {
1444                         struct ovs_key_tcp *tcp_key;
1445
1446                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1447                         if (!nla)
1448                                 goto nla_put_failure;
1449                         tcp_key = nla_data(nla);
1450                         tcp_key->tcp_src = output->tp.src;
1451                         tcp_key->tcp_dst = output->tp.dst;
1452                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1453                                          output->tp.flags))
1454                                 goto nla_put_failure;
1455                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1456                         struct ovs_key_udp *udp_key;
1457
1458                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1459                         if (!nla)
1460                                 goto nla_put_failure;
1461                         udp_key = nla_data(nla);
1462                         udp_key->udp_src = output->tp.src;
1463                         udp_key->udp_dst = output->tp.dst;
1464                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1465                         struct ovs_key_sctp *sctp_key;
1466
1467                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1468                         if (!nla)
1469                                 goto nla_put_failure;
1470                         sctp_key = nla_data(nla);
1471                         sctp_key->sctp_src = output->tp.src;
1472                         sctp_key->sctp_dst = output->tp.dst;
1473                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1474                            swkey->ip.proto == IPPROTO_ICMP) {
1475                         struct ovs_key_icmp *icmp_key;
1476
1477                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1478                         if (!nla)
1479                                 goto nla_put_failure;
1480                         icmp_key = nla_data(nla);
1481                         icmp_key->icmp_type = ntohs(output->tp.src);
1482                         icmp_key->icmp_code = ntohs(output->tp.dst);
1483                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1484                            swkey->ip.proto == IPPROTO_ICMPV6) {
1485                         struct ovs_key_icmpv6 *icmpv6_key;
1486
1487                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1488                                                 sizeof(*icmpv6_key));
1489                         if (!nla)
1490                                 goto nla_put_failure;
1491                         icmpv6_key = nla_data(nla);
1492                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1493                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1494
1495                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1496                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1497                                 struct ovs_key_nd *nd_key;
1498
1499                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1500                                 if (!nla)
1501                                         goto nla_put_failure;
1502                                 nd_key = nla_data(nla);
1503                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1504                                                         sizeof(nd_key->nd_target));
1505                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1506                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1507                         }
1508                 }
1509         }
1510
1511 unencap:
1512         if (encap)
1513                 nla_nest_end(skb, encap);
1514
1515         return 0;
1516
1517 nla_put_failure:
1518         return -EMSGSIZE;
1519 }
1520
1521 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1522                     const struct sw_flow_key *output, int attr, bool is_mask,
1523                     struct sk_buff *skb)
1524 {
1525         int err;
1526         struct nlattr *nla;
1527
1528         nla = nla_nest_start(skb, attr);
1529         if (!nla)
1530                 return -EMSGSIZE;
1531         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1532         if (err)
1533                 return err;
1534         nla_nest_end(skb, nla);
1535
1536         return 0;
1537 }
1538
1539 /* Called with ovs_mutex or RCU read lock. */
1540 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1541 {
1542         if (ovs_identifier_is_ufid(&flow->id))
1543                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1544                                flow->id.ufid);
1545
1546         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1547                                OVS_FLOW_ATTR_KEY, false, skb);
1548 }
1549
1550 /* Called with ovs_mutex or RCU read lock. */
1551 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1552 {
1553         return ovs_nla_put_key(&flow->key, &flow->key,
1554                                 OVS_FLOW_ATTR_KEY, false, skb);
1555 }
1556
1557 /* Called with ovs_mutex or RCU read lock. */
1558 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1559 {
1560         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1561                                 OVS_FLOW_ATTR_MASK, true, skb);
1562 }
1563
1564 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1565
1566 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1567 {
1568         struct sw_flow_actions *sfa;
1569
1570         if (size > MAX_ACTIONS_BUFSIZE) {
1571                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1572                 return ERR_PTR(-EINVAL);
1573         }
1574
1575         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1576         if (!sfa)
1577                 return ERR_PTR(-ENOMEM);
1578
1579         sfa->actions_len = 0;
1580         return sfa;
1581 }
1582
1583 static void ovs_nla_free_set_action(const struct nlattr *a)
1584 {
1585         const struct nlattr *ovs_key = nla_data(a);
1586         struct ovs_tunnel_info *ovs_tun;
1587
1588         switch (nla_type(ovs_key)) {
1589         case OVS_KEY_ATTR_TUNNEL_INFO:
1590                 ovs_tun = nla_data(ovs_key);
1591                 dst_release((struct dst_entry *)ovs_tun->tun_dst);
1592                 break;
1593         }
1594 }
1595
1596 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1597 {
1598         const struct nlattr *a;
1599         int rem;
1600
1601         if (!sf_acts)
1602                 return;
1603
1604         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1605                 switch (nla_type(a)) {
1606                 case OVS_ACTION_ATTR_SET:
1607                         ovs_nla_free_set_action(a);
1608                         break;
1609                 case OVS_ACTION_ATTR_CT:
1610                         ovs_ct_free_action(a);
1611                         break;
1612                 }
1613         }
1614
1615         kfree(sf_acts);
1616 }
1617
1618 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1619 {
1620         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1621 }
1622
1623 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1624  * The caller must hold rcu_read_lock for this to be sensible. */
1625 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1626 {
1627         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1628 }
1629
1630 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1631                                        int attr_len, bool log)
1632 {
1633
1634         struct sw_flow_actions *acts;
1635         int new_acts_size;
1636         int req_size = NLA_ALIGN(attr_len);
1637         int next_offset = offsetof(struct sw_flow_actions, actions) +
1638                                         (*sfa)->actions_len;
1639
1640         if (req_size <= (ksize(*sfa) - next_offset))
1641                 goto out;
1642
1643         new_acts_size = ksize(*sfa) * 2;
1644
1645         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1646                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1647                         return ERR_PTR(-EMSGSIZE);
1648                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1649         }
1650
1651         acts = nla_alloc_flow_actions(new_acts_size, log);
1652         if (IS_ERR(acts))
1653                 return (void *)acts;
1654
1655         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1656         acts->actions_len = (*sfa)->actions_len;
1657         acts->orig_len = (*sfa)->orig_len;
1658         kfree(*sfa);
1659         *sfa = acts;
1660
1661 out:
1662         (*sfa)->actions_len += req_size;
1663         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1664 }
1665
1666 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1667                                    int attrtype, void *data, int len, bool log)
1668 {
1669         struct nlattr *a;
1670
1671         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1672         if (IS_ERR(a))
1673                 return a;
1674
1675         a->nla_type = attrtype;
1676         a->nla_len = nla_attr_size(len);
1677
1678         if (data)
1679                 memcpy(nla_data(a), data, len);
1680         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1681
1682         return a;
1683 }
1684
1685 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1686                        int len, bool log)
1687 {
1688         struct nlattr *a;
1689
1690         a = __add_action(sfa, attrtype, data, len, log);
1691
1692         return PTR_ERR_OR_ZERO(a);
1693 }
1694
1695 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1696                                           int attrtype, bool log)
1697 {
1698         int used = (*sfa)->actions_len;
1699         int err;
1700
1701         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1702         if (err)
1703                 return err;
1704
1705         return used;
1706 }
1707
1708 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1709                                          int st_offset)
1710 {
1711         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1712                                                                st_offset);
1713
1714         a->nla_len = sfa->actions_len - st_offset;
1715 }
1716
1717 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1718                                   const struct sw_flow_key *key,
1719                                   int depth, struct sw_flow_actions **sfa,
1720                                   __be16 eth_type, __be16 vlan_tci, bool log);
1721
1722 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1723                                     const struct sw_flow_key *key, int depth,
1724                                     struct sw_flow_actions **sfa,
1725                                     __be16 eth_type, __be16 vlan_tci, bool log)
1726 {
1727         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1728         const struct nlattr *probability, *actions;
1729         const struct nlattr *a;
1730         int rem, start, err, st_acts;
1731
1732         memset(attrs, 0, sizeof(attrs));
1733         nla_for_each_nested(a, attr, rem) {
1734                 int type = nla_type(a);
1735                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1736                         return -EINVAL;
1737                 attrs[type] = a;
1738         }
1739         if (rem)
1740                 return -EINVAL;
1741
1742         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1743         if (!probability || nla_len(probability) != sizeof(u32))
1744                 return -EINVAL;
1745
1746         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1747         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1748                 return -EINVAL;
1749
1750         /* validation done, copy sample action. */
1751         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1752         if (start < 0)
1753                 return start;
1754         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1755                                  nla_data(probability), sizeof(u32), log);
1756         if (err)
1757                 return err;
1758         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1759         if (st_acts < 0)
1760                 return st_acts;
1761
1762         err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1763                                      eth_type, vlan_tci, log);
1764         if (err)
1765                 return err;
1766
1767         add_nested_action_end(*sfa, st_acts);
1768         add_nested_action_end(*sfa, start);
1769
1770         return 0;
1771 }
1772
1773 void ovs_match_init(struct sw_flow_match *match,
1774                     struct sw_flow_key *key,
1775                     struct sw_flow_mask *mask)
1776 {
1777         memset(match, 0, sizeof(*match));
1778         match->key = key;
1779         match->mask = mask;
1780
1781         memset(key, 0, sizeof(*key));
1782
1783         if (mask) {
1784                 memset(&mask->key, 0, sizeof(mask->key));
1785                 mask->range.start = mask->range.end = 0;
1786         }
1787 }
1788
1789 static int validate_geneve_opts(struct sw_flow_key *key)
1790 {
1791         struct geneve_opt *option;
1792         int opts_len = key->tun_opts_len;
1793         bool crit_opt = false;
1794
1795         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1796         while (opts_len > 0) {
1797                 int len;
1798
1799                 if (opts_len < sizeof(*option))
1800                         return -EINVAL;
1801
1802                 len = sizeof(*option) + option->length * 4;
1803                 if (len > opts_len)
1804                         return -EINVAL;
1805
1806                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1807
1808                 option = (struct geneve_opt *)((u8 *)option + len);
1809                 opts_len -= len;
1810         };
1811
1812         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1813
1814         return 0;
1815 }
1816
1817 static int validate_and_copy_set_tun(const struct nlattr *attr,
1818                                      struct sw_flow_actions **sfa, bool log)
1819 {
1820         struct sw_flow_match match;
1821         struct sw_flow_key key;
1822         struct metadata_dst *tun_dst;
1823         struct ip_tunnel_info *tun_info;
1824         struct ovs_tunnel_info *ovs_tun;
1825         struct nlattr *a;
1826         int err = 0, start, opts_type;
1827
1828         ovs_match_init(&match, &key, NULL);
1829         opts_type = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1830         if (opts_type < 0)
1831                 return opts_type;
1832
1833         if (key.tun_opts_len) {
1834                 switch (opts_type) {
1835                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1836                         err = validate_geneve_opts(&key);
1837                         if (err < 0)
1838                                 return err;
1839                         break;
1840                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1841                         break;
1842                 }
1843         };
1844
1845         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1846         if (start < 0)
1847                 return start;
1848
1849         tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1850         if (!tun_dst)
1851                 return -ENOMEM;
1852
1853         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1854                          sizeof(*ovs_tun), log);
1855         if (IS_ERR(a)) {
1856                 dst_release((struct dst_entry *)tun_dst);
1857                 return PTR_ERR(a);
1858         }
1859
1860         ovs_tun = nla_data(a);
1861         ovs_tun->tun_dst = tun_dst;
1862
1863         tun_info = &tun_dst->u.tun_info;
1864         tun_info->mode = IP_TUNNEL_INFO_TX;
1865         tun_info->key = key.tun_key;
1866         tun_info->options_len = key.tun_opts_len;
1867
1868         if (tun_info->options_len) {
1869                 /* We need to store the options in the action itself since
1870                  * everything else will go away after flow setup. We can append
1871                  * it to tun_info and then point there.
1872                  */
1873                 memcpy((tun_info + 1),
1874                        TUN_METADATA_OPTS(&key, key.tun_opts_len), key.tun_opts_len);
1875                 tun_info->options = (tun_info + 1);
1876         } else {
1877                 tun_info->options = NULL;
1878         }
1879
1880         add_nested_action_end(*sfa, start);
1881
1882         return err;
1883 }
1884
1885 /* Return false if there are any non-masked bits set.
1886  * Mask follows data immediately, before any netlink padding.
1887  */
1888 static bool validate_masked(u8 *data, int len)
1889 {
1890         u8 *mask = data + len;
1891
1892         while (len--)
1893                 if (*data++ & ~*mask++)
1894                         return false;
1895
1896         return true;
1897 }
1898
1899 static int validate_set(const struct nlattr *a,
1900                         const struct sw_flow_key *flow_key,
1901                         struct sw_flow_actions **sfa,
1902                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
1903 {
1904         const struct nlattr *ovs_key = nla_data(a);
1905         int key_type = nla_type(ovs_key);
1906         size_t key_len;
1907
1908         /* There can be only one key in a action */
1909         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1910                 return -EINVAL;
1911
1912         key_len = nla_len(ovs_key);
1913         if (masked)
1914                 key_len /= 2;
1915
1916         if (key_type > OVS_KEY_ATTR_MAX ||
1917             (ovs_key_lens[key_type].len != key_len &&
1918              ovs_key_lens[key_type].len != OVS_ATTR_NESTED))
1919                 return -EINVAL;
1920
1921         if (masked && !validate_masked(nla_data(ovs_key), key_len))
1922                 return -EINVAL;
1923
1924         switch (key_type) {
1925         const struct ovs_key_ipv4 *ipv4_key;
1926         const struct ovs_key_ipv6 *ipv6_key;
1927         int err;
1928
1929         case OVS_KEY_ATTR_PRIORITY:
1930         case OVS_KEY_ATTR_SKB_MARK:
1931         case OVS_KEY_ATTR_CT_MARK:
1932         case OVS_KEY_ATTR_ETHERNET:
1933                 break;
1934
1935         case OVS_KEY_ATTR_TUNNEL:
1936                 if (eth_p_mpls(eth_type))
1937                         return -EINVAL;
1938
1939                 if (masked)
1940                         return -EINVAL; /* Masked tunnel set not supported. */
1941
1942                 *skip_copy = true;
1943                 err = validate_and_copy_set_tun(a, sfa, log);
1944                 if (err)
1945                         return err;
1946                 break;
1947
1948         case OVS_KEY_ATTR_IPV4:
1949                 if (eth_type != htons(ETH_P_IP))
1950                         return -EINVAL;
1951
1952                 ipv4_key = nla_data(ovs_key);
1953
1954                 if (masked) {
1955                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
1956
1957                         /* Non-writeable fields. */
1958                         if (mask->ipv4_proto || mask->ipv4_frag)
1959                                 return -EINVAL;
1960                 } else {
1961                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1962                                 return -EINVAL;
1963
1964                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1965                                 return -EINVAL;
1966                 }
1967                 break;
1968
1969         case OVS_KEY_ATTR_IPV6:
1970                 if (eth_type != htons(ETH_P_IPV6))
1971                         return -EINVAL;
1972
1973                 ipv6_key = nla_data(ovs_key);
1974
1975                 if (masked) {
1976                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
1977
1978                         /* Non-writeable fields. */
1979                         if (mask->ipv6_proto || mask->ipv6_frag)
1980                                 return -EINVAL;
1981
1982                         /* Invalid bits in the flow label mask? */
1983                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
1984                                 return -EINVAL;
1985                 } else {
1986                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1987                                 return -EINVAL;
1988
1989                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1990                                 return -EINVAL;
1991                 }
1992                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
1993                         return -EINVAL;
1994
1995                 break;
1996
1997         case OVS_KEY_ATTR_TCP:
1998                 if ((eth_type != htons(ETH_P_IP) &&
1999                      eth_type != htons(ETH_P_IPV6)) ||
2000                     flow_key->ip.proto != IPPROTO_TCP)
2001                         return -EINVAL;
2002
2003                 break;
2004
2005         case OVS_KEY_ATTR_UDP:
2006                 if ((eth_type != htons(ETH_P_IP) &&
2007                      eth_type != htons(ETH_P_IPV6)) ||
2008                     flow_key->ip.proto != IPPROTO_UDP)
2009                         return -EINVAL;
2010
2011                 break;
2012
2013         case OVS_KEY_ATTR_MPLS:
2014                 if (!eth_p_mpls(eth_type))
2015                         return -EINVAL;
2016                 break;
2017
2018         case OVS_KEY_ATTR_SCTP:
2019                 if ((eth_type != htons(ETH_P_IP) &&
2020                      eth_type != htons(ETH_P_IPV6)) ||
2021                     flow_key->ip.proto != IPPROTO_SCTP)
2022                         return -EINVAL;
2023
2024                 break;
2025
2026         default:
2027                 return -EINVAL;
2028         }
2029
2030         /* Convert non-masked non-tunnel set actions to masked set actions. */
2031         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2032                 int start, len = key_len * 2;
2033                 struct nlattr *at;
2034
2035                 *skip_copy = true;
2036
2037                 start = add_nested_action_start(sfa,
2038                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2039                                                 log);
2040                 if (start < 0)
2041                         return start;
2042
2043                 at = __add_action(sfa, key_type, NULL, len, log);
2044                 if (IS_ERR(at))
2045                         return PTR_ERR(at);
2046
2047                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2048                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2049                 /* Clear non-writeable bits from otherwise writeable fields. */
2050                 if (key_type == OVS_KEY_ATTR_IPV6) {
2051                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2052
2053                         mask->ipv6_label &= htonl(0x000FFFFF);
2054                 }
2055                 add_nested_action_end(*sfa, start);
2056         }
2057
2058         return 0;
2059 }
2060
2061 static int validate_userspace(const struct nlattr *attr)
2062 {
2063         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2064                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2065                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2066                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2067         };
2068         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2069         int error;
2070
2071         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2072                                  attr, userspace_policy);
2073         if (error)
2074                 return error;
2075
2076         if (!a[OVS_USERSPACE_ATTR_PID] ||
2077             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2078                 return -EINVAL;
2079
2080         return 0;
2081 }
2082
2083 static int copy_action(const struct nlattr *from,
2084                        struct sw_flow_actions **sfa, bool log)
2085 {
2086         int totlen = NLA_ALIGN(from->nla_len);
2087         struct nlattr *to;
2088
2089         to = reserve_sfa_size(sfa, from->nla_len, log);
2090         if (IS_ERR(to))
2091                 return PTR_ERR(to);
2092
2093         memcpy(to, from, totlen);
2094         return 0;
2095 }
2096
2097 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2098                                   const struct sw_flow_key *key,
2099                                   int depth, struct sw_flow_actions **sfa,
2100                                   __be16 eth_type, __be16 vlan_tci, bool log)
2101 {
2102         const struct nlattr *a;
2103         int rem, err;
2104
2105         if (depth >= SAMPLE_ACTION_DEPTH)
2106                 return -EOVERFLOW;
2107
2108         nla_for_each_nested(a, attr, rem) {
2109                 /* Expected argument lengths, (u32)-1 for variable length. */
2110                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2111                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2112                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2113                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2114                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2115                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2116                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2117                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2118                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2119                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2120                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2121                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2122                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2123                 };
2124                 const struct ovs_action_push_vlan *vlan;
2125                 int type = nla_type(a);
2126                 bool skip_copy;
2127
2128                 if (type > OVS_ACTION_ATTR_MAX ||
2129                     (action_lens[type] != nla_len(a) &&
2130                      action_lens[type] != (u32)-1))
2131                         return -EINVAL;
2132
2133                 skip_copy = false;
2134                 switch (type) {
2135                 case OVS_ACTION_ATTR_UNSPEC:
2136                         return -EINVAL;
2137
2138                 case OVS_ACTION_ATTR_USERSPACE:
2139                         err = validate_userspace(a);
2140                         if (err)
2141                                 return err;
2142                         break;
2143
2144                 case OVS_ACTION_ATTR_OUTPUT:
2145                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2146                                 return -EINVAL;
2147                         break;
2148
2149                 case OVS_ACTION_ATTR_HASH: {
2150                         const struct ovs_action_hash *act_hash = nla_data(a);
2151
2152                         switch (act_hash->hash_alg) {
2153                         case OVS_HASH_ALG_L4:
2154                                 break;
2155                         default:
2156                                 return  -EINVAL;
2157                         }
2158
2159                         break;
2160                 }
2161
2162                 case OVS_ACTION_ATTR_POP_VLAN:
2163                         vlan_tci = htons(0);
2164                         break;
2165
2166                 case OVS_ACTION_ATTR_PUSH_VLAN:
2167                         vlan = nla_data(a);
2168                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2169                                 return -EINVAL;
2170                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2171                                 return -EINVAL;
2172                         vlan_tci = vlan->vlan_tci;
2173                         break;
2174
2175                 case OVS_ACTION_ATTR_RECIRC:
2176                         break;
2177
2178                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2179                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2180
2181                         if (!eth_p_mpls(mpls->mpls_ethertype))
2182                                 return -EINVAL;
2183                         /* Prohibit push MPLS other than to a white list
2184                          * for packets that have a known tag order.
2185                          */
2186                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2187                             (eth_type != htons(ETH_P_IP) &&
2188                              eth_type != htons(ETH_P_IPV6) &&
2189                              eth_type != htons(ETH_P_ARP) &&
2190                              eth_type != htons(ETH_P_RARP) &&
2191                              !eth_p_mpls(eth_type)))
2192                                 return -EINVAL;
2193                         eth_type = mpls->mpls_ethertype;
2194                         break;
2195                 }
2196
2197                 case OVS_ACTION_ATTR_POP_MPLS:
2198                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2199                             !eth_p_mpls(eth_type))
2200                                 return -EINVAL;
2201
2202                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2203                          * as there is no check here to ensure that the new
2204                          * eth_type is valid and thus set actions could
2205                          * write off the end of the packet or otherwise
2206                          * corrupt it.
2207                          *
2208                          * Support for these actions is planned using packet
2209                          * recirculation.
2210                          */
2211                         eth_type = htons(0);
2212                         break;
2213
2214                 case OVS_ACTION_ATTR_SET:
2215                         err = validate_set(a, key, sfa,
2216                                            &skip_copy, eth_type, false, log);
2217                         if (err)
2218                                 return err;
2219                         break;
2220
2221                 case OVS_ACTION_ATTR_SET_MASKED:
2222                         err = validate_set(a, key, sfa,
2223                                            &skip_copy, eth_type, true, log);
2224                         if (err)
2225                                 return err;
2226                         break;
2227
2228                 case OVS_ACTION_ATTR_SAMPLE:
2229                         err = validate_and_copy_sample(net, a, key, depth, sfa,
2230                                                        eth_type, vlan_tci, log);
2231                         if (err)
2232                                 return err;
2233                         skip_copy = true;
2234                         break;
2235
2236                 case OVS_ACTION_ATTR_CT:
2237                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2238                         if (err)
2239                                 return err;
2240                         skip_copy = true;
2241                         break;
2242
2243                 default:
2244                         OVS_NLERR(log, "Unknown Action type %d", type);
2245                         return -EINVAL;
2246                 }
2247                 if (!skip_copy) {
2248                         err = copy_action(a, sfa, log);
2249                         if (err)
2250                                 return err;
2251                 }
2252         }
2253
2254         if (rem > 0)
2255                 return -EINVAL;
2256
2257         return 0;
2258 }
2259
2260 /* 'key' must be the masked key. */
2261 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2262                          const struct sw_flow_key *key,
2263                          struct sw_flow_actions **sfa, bool log)
2264 {
2265         int err;
2266
2267         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2268         if (IS_ERR(*sfa))
2269                 return PTR_ERR(*sfa);
2270
2271         (*sfa)->orig_len = nla_len(attr);
2272         err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2273                                      key->eth.tci, log);
2274         if (err)
2275                 ovs_nla_free_flow_actions(*sfa);
2276
2277         return err;
2278 }
2279
2280 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2281 {
2282         const struct nlattr *a;
2283         struct nlattr *start;
2284         int err = 0, rem;
2285
2286         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2287         if (!start)
2288                 return -EMSGSIZE;
2289
2290         nla_for_each_nested(a, attr, rem) {
2291                 int type = nla_type(a);
2292                 struct nlattr *st_sample;
2293
2294                 switch (type) {
2295                 case OVS_SAMPLE_ATTR_PROBABILITY:
2296                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2297                                     sizeof(u32), nla_data(a)))
2298                                 return -EMSGSIZE;
2299                         break;
2300                 case OVS_SAMPLE_ATTR_ACTIONS:
2301                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2302                         if (!st_sample)
2303                                 return -EMSGSIZE;
2304                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2305                         if (err)
2306                                 return err;
2307                         nla_nest_end(skb, st_sample);
2308                         break;
2309                 }
2310         }
2311
2312         nla_nest_end(skb, start);
2313         return err;
2314 }
2315
2316 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2317 {
2318         const struct nlattr *ovs_key = nla_data(a);
2319         int key_type = nla_type(ovs_key);
2320         struct nlattr *start;
2321         int err;
2322
2323         switch (key_type) {
2324         case OVS_KEY_ATTR_TUNNEL_INFO: {
2325                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2326                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2327
2328                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2329                 if (!start)
2330                         return -EMSGSIZE;
2331
2332                 err = ipv4_tun_to_nlattr(skb, &tun_info->key,
2333                                          tun_info->options_len ?
2334                                                 tun_info->options : NULL,
2335                                          tun_info->options_len);
2336                 if (err)
2337                         return err;
2338                 nla_nest_end(skb, start);
2339                 break;
2340         }
2341         default:
2342                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2343                         return -EMSGSIZE;
2344                 break;
2345         }
2346
2347         return 0;
2348 }
2349
2350 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2351                                                 struct sk_buff *skb)
2352 {
2353         const struct nlattr *ovs_key = nla_data(a);
2354         struct nlattr *nla;
2355         size_t key_len = nla_len(ovs_key) / 2;
2356
2357         /* Revert the conversion we did from a non-masked set action to
2358          * masked set action.
2359          */
2360         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2361         if (!nla)
2362                 return -EMSGSIZE;
2363
2364         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2365                 return -EMSGSIZE;
2366
2367         nla_nest_end(skb, nla);
2368         return 0;
2369 }
2370
2371 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2372 {
2373         const struct nlattr *a;
2374         int rem, err;
2375
2376         nla_for_each_attr(a, attr, len, rem) {
2377                 int type = nla_type(a);
2378
2379                 switch (type) {
2380                 case OVS_ACTION_ATTR_SET:
2381                         err = set_action_to_attr(a, skb);
2382                         if (err)
2383                                 return err;
2384                         break;
2385
2386                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2387                         err = masked_set_action_to_set_action_attr(a, skb);
2388                         if (err)
2389                                 return err;
2390                         break;
2391
2392                 case OVS_ACTION_ATTR_SAMPLE:
2393                         err = sample_action_to_attr(a, skb);
2394                         if (err)
2395                                 return err;
2396                         break;
2397
2398                 case OVS_ACTION_ATTR_CT:
2399                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2400                         if (err)
2401                                 return err;
2402                         break;
2403
2404                 default:
2405                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2406                                 return -EMSGSIZE;
2407                         break;
2408                 }
2409         }
2410
2411         return 0;
2412 }