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 #define OVS_ATTR_VARIABLE -2
  61
  62 static void update_range(struct sw_flow_match *match,
  63                          size_t offset, size_t size, bool is_mask)
  64 {
  65         struct sw_flow_key_range *range;
  66         size_t start = rounddown(offset, sizeof(long));
  67         size_t end = roundup(offset + size, sizeof(long));
  68
  69         if (!is_mask)
  70                 range = &match->range;
  71         else
  72                 range = &match->mask->range;
  73
  74         if (range->start == range->end) {
  75                 range->start = start;
  76                 range->end = end;
  77                 return;
  78         }
  79
  80         if (range->start > start)
  81                 range->start = start;
  82
  83         if (range->end < end)
  84                 range->end = end;
  85 }
  86
  87 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  88         do { \
  89                 update_range(match, offsetof(struct sw_flow_key, field),    \
  90                              sizeof((match)->key->field), is_mask);         \
  91                 if (is_mask)                                                \
  92                         (match)->mask->key.field = value;                   \
  93                 else                                                        \
  94                         (match)->key->field = value;                        \
  95         } while (0)
  96
  97 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
  98         do {                                                                \
  99                 update_range(match, offset, len, is_mask);                  \
 100                 if (is_mask)                                                \
 101                         memcpy((u8 *)&(match)->mask->key + offset, value_p, \
 102                                len);                                       \
 103                 else                                                        \
 104                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
 105         } while (0)
 106
 107 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
 108         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 109                                   value_p, len, is_mask)
 110
 111 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 112         do {                                                                \
 113                 update_range(match, offsetof(struct sw_flow_key, field),    \
 114                              sizeof((match)->key->field), is_mask);         \
 115                 if (is_mask)                                                \
 116                         memset((u8 *)&(match)->mask->key.field, value,      \
 117                                sizeof((match)->mask->key.field));           \
 118                 else                                                        \
 119                         memset((u8 *)&(match)->key->field, value,           \
 120                                sizeof((match)->key->field));                \
 121         } while (0)
 122
 123 static bool match_validate(const struct sw_flow_match *match,
 124                            u64 key_attrs, u64 mask_attrs, bool log)
 125 {
 126         u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
 127         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 128
 129         /* The following mask attributes allowed only if they
 130          * pass the validation tests. */
 131         mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
 132                         | (1 << OVS_KEY_ATTR_IPV6)
 133                         | (1 << OVS_KEY_ATTR_TCP)
 134                         | (1 << OVS_KEY_ATTR_TCP_FLAGS)
 135                         | (1 << OVS_KEY_ATTR_UDP)
 136                         | (1 << OVS_KEY_ATTR_SCTP)
 137                         | (1 << OVS_KEY_ATTR_ICMP)
 138                         | (1 << OVS_KEY_ATTR_ICMPV6)
 139                         | (1 << OVS_KEY_ATTR_ARP)
 140                         | (1 << OVS_KEY_ATTR_ND)
 141                         | (1 << OVS_KEY_ATTR_MPLS));
 142
 143         /* Always allowed mask fields. */
 144         mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
 145                        | (1 << OVS_KEY_ATTR_IN_PORT)
 146                        | (1 << OVS_KEY_ATTR_ETHERTYPE));
 147
 148         /* Check key attributes. */
 149         if (match->key->eth.type == htons(ETH_P_ARP)
 150                         || match->key->eth.type == htons(ETH_P_RARP)) {
 151                 key_expected |= 1 << OVS_KEY_ATTR_ARP;
 152                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 153                         mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
 154         }
 155
 156         if (eth_p_mpls(match->key->eth.type)) {
 157                 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
 158                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 159                         mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
 160         }
 161
 162         if (match->key->eth.type == htons(ETH_P_IP)) {
 163                 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
 164                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 165                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
 166
 167                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 168                         if (match->key->ip.proto == IPPROTO_UDP) {
 169                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 170                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 171                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 172                         }
 173
 174                         if (match->key->ip.proto == IPPROTO_SCTP) {
 175                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 176                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 177                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 178                         }
 179
 180                         if (match->key->ip.proto == IPPROTO_TCP) {
 181                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 182                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 183                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 184                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 185                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 186                                 }
 187                         }
 188
 189                         if (match->key->ip.proto == IPPROTO_ICMP) {
 190                                 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
 191                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 192                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
 193                         }
 194                 }
 195         }
 196
 197         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 198                 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
 199                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 200                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
 201
 202                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 203                         if (match->key->ip.proto == IPPROTO_UDP) {
 204                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 205                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 206                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 207                         }
 208
 209                         if (match->key->ip.proto == IPPROTO_SCTP) {
 210                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 211                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 212                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 213                         }
 214
 215                         if (match->key->ip.proto == IPPROTO_TCP) {
 216                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 217                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 218                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 219                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 220                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 221                                 }
 222                         }
 223
 224                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 225                                 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
 226                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 227                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
 228
 229                                 if (match->key->tp.src ==
 230                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 231                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 232                                         key_expected |= 1 << OVS_KEY_ATTR_ND;
 233                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 234                                                 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
 235                                 }
 236                         }
 237                 }
 238         }
 239
 240         if ((key_attrs & key_expected) != key_expected) {
 241                 /* Key attributes check failed. */
 242                 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
 243                           (unsigned long long)key_attrs,
 244                           (unsigned long long)key_expected);
 245                 return false;
 246         }
 247
 248         if ((mask_attrs & mask_allowed) != mask_attrs) {
 249                 /* Mask attributes check failed. */
 250                 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
 251                           (unsigned long long)mask_attrs,
 252                           (unsigned long long)mask_allowed);
 253                 return false;
 254         }
 255
 256         return true;
 257 }
 258
 259 size_t ovs_tun_key_attr_size(void)
 260 {
 261         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 262          * updating this function.
 263          */
 264         return    nla_total_size(8)    /* OVS_TUNNEL_KEY_ATTR_ID */
 265                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
 266                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
 267                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
 268                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
 269                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 270                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
 271                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
 272                 + nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 273                 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
 274                  * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
 275                  */
 276                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
 277                 + nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
 278 }
 279
 280 size_t ovs_key_attr_size(void)
 281 {
 282         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 283          * updating this function.
 284          */
 285         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 26);
 286
 287         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 288                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 289                   + ovs_tun_key_attr_size()
 290                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 291                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 292                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 293                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 294                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
 295                 + nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
 296                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
 297                 + nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
 298                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 299                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 300                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 301                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 302                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 303                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 304                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 305                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 306 }
 307
 308 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
 309         [OVS_VXLAN_EXT_GBP]         = { .len = sizeof(u32) },
 310 };
 311
 312 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 313         [OVS_TUNNEL_KEY_ATTR_ID]            = { .len = sizeof(u64) },
 314         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC]      = { .len = sizeof(u32) },
 315         [OVS_TUNNEL_KEY_ATTR_IPV4_DST]      = { .len = sizeof(u32) },
 316         [OVS_TUNNEL_KEY_ATTR_TOS]           = { .len = 1 },
 317         [OVS_TUNNEL_KEY_ATTR_TTL]           = { .len = 1 },
 318         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
 319         [OVS_TUNNEL_KEY_ATTR_CSUM]          = { .len = 0 },
 320         [OVS_TUNNEL_KEY_ATTR_TP_SRC]        = { .len = sizeof(u16) },
 321         [OVS_TUNNEL_KEY_ATTR_TP_DST]        = { .len = sizeof(u16) },
 322         [OVS_TUNNEL_KEY_ATTR_OAM]           = { .len = 0 },
 323         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
 324         [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
 325                                                 .next = ovs_vxlan_ext_key_lens },
 326         [OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
 327         [OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
 328 };
 329
 330 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 331 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 332         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 333         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 334         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 335         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 336         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 337         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 338         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 339         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 340         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 341         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 342         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 343         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 344         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 345         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 346         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 347         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 348         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 349         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 350         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 351         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 352                                      .next = ovs_tunnel_key_lens, },
 353         [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 354         [OVS_KEY_ATTR_CT_STATE]  = { .len = sizeof(u32) },
 355         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 356         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 357         [OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
 358 };
 359
 360 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
 361 {
 362         return expected_len == attr_len ||
 363                expected_len == OVS_ATTR_NESTED ||
 364                expected_len == OVS_ATTR_VARIABLE;
 365 }
 366
 367 static bool is_all_zero(const u8 *fp, size_t size)
 368 {
 369         int i;
 370
 371         if (!fp)
 372                 return false;
 373
 374         for (i = 0; i < size; i++)
 375                 if (fp[i])
 376                         return false;
 377
 378         return true;
 379 }
 380
 381 static int __parse_flow_nlattrs(const struct nlattr *attr,
 382                                 const struct nlattr *a[],
 383                                 u64 *attrsp, bool log, bool nz)
 384 {
 385         const struct nlattr *nla;
 386         u64 attrs;
 387         int rem;
 388
 389         attrs = *attrsp;
 390         nla_for_each_nested(nla, attr, rem) {
 391                 u16 type = nla_type(nla);
 392                 int expected_len;
 393
 394                 if (type > OVS_KEY_ATTR_MAX) {
 395                         OVS_NLERR(log, "Key type %d is out of range max %d",
 396                                   type, OVS_KEY_ATTR_MAX);
 397                         return -EINVAL;
 398                 }
 399
 400                 if (attrs & (1 << type)) {
 401                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 402                         return -EINVAL;
 403                 }
 404
 405                 expected_len = ovs_key_lens[type].len;
 406                 if (!check_attr_len(nla_len(nla), expected_len)) {
 407                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 408                                   type, nla_len(nla), expected_len);
 409                         return -EINVAL;
 410                 }
 411
 412                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 413                         attrs |= 1 << type;
 414                         a[type] = nla;
 415                 }
 416         }
 417         if (rem) {
 418                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 419                 return -EINVAL;
 420         }
 421
 422         *attrsp = attrs;
 423         return 0;
 424 }
 425
 426 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 427                                    const struct nlattr *a[], u64 *attrsp,
 428                                    bool log)
 429 {
 430         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 431 }
 432
 433 static int parse_flow_nlattrs(const struct nlattr *attr,
 434                               const struct nlattr *a[], u64 *attrsp,
 435                               bool log)
 436 {
 437         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 438 }
 439
 440 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 441                                      struct sw_flow_match *match, bool is_mask,
 442                                      bool log)
 443 {
 444         unsigned long opt_key_offset;
 445
 446         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 447                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 448                           nla_len(a), sizeof(match->key->tun_opts));
 449                 return -EINVAL;
 450         }
 451
 452         if (nla_len(a) % 4 != 0) {
 453                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 454                           nla_len(a));
 455                 return -EINVAL;
 456         }
 457
 458         /* We need to record the length of the options passed
 459          * down, otherwise packets with the same format but
 460          * additional options will be silently matched.
 461          */
 462         if (!is_mask) {
 463                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 464                                 false);
 465         } else {
 466                 /* This is somewhat unusual because it looks at
 467                  * both the key and mask while parsing the
 468                  * attributes (and by extension assumes the key
 469                  * is parsed first). Normally, we would verify
 470                  * that each is the correct length and that the
 471                  * attributes line up in the validate function.
 472                  * However, that is difficult because this is
 473                  * variable length and we won't have the
 474                  * information later.
 475                  */
 476                 if (match->key->tun_opts_len != nla_len(a)) {
 477                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 478                                   match->key->tun_opts_len, nla_len(a));
 479                         return -EINVAL;
 480                 }
 481
 482                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 483         }
 484
 485         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 486         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 487                                   nla_len(a), is_mask);
 488         return 0;
 489 }
 490
 491 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
 492                                      struct sw_flow_match *match, bool is_mask,
 493                                      bool log)
 494 {
 495         struct nlattr *a;
 496         int rem;
 497         unsigned long opt_key_offset;
 498         struct vxlan_metadata opts;
 499
 500         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 501
 502         memset(&opts, 0, sizeof(opts));
 503         nla_for_each_nested(a, attr, rem) {
 504                 int type = nla_type(a);
 505
 506                 if (type > OVS_VXLAN_EXT_MAX) {
 507                         OVS_NLERR(log, "VXLAN extension %d out of range max %d",
 508                                   type, OVS_VXLAN_EXT_MAX);
 509                         return -EINVAL;
 510                 }
 511
 512                 if (!check_attr_len(nla_len(a),
 513                                     ovs_vxlan_ext_key_lens[type].len)) {
 514                         OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
 515                                   type, nla_len(a),
 516                                   ovs_vxlan_ext_key_lens[type].len);
 517                         return -EINVAL;
 518                 }
 519
 520                 switch (type) {
 521                 case OVS_VXLAN_EXT_GBP:
 522                         opts.gbp = nla_get_u32(a);
 523                         break;
 524                 default:
 525                         OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
 526                                   type);
 527                         return -EINVAL;
 528                 }
 529         }
 530         if (rem) {
 531                 OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
 532                           rem);
 533                 return -EINVAL;
 534         }
 535
 536         if (!is_mask)
 537                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 538         else
 539                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 540
 541         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 542         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 543                                   is_mask);
 544         return 0;
 545 }
 546
 547 static int ip_tun_from_nlattr(const struct nlattr *attr,
 548                               struct sw_flow_match *match, bool is_mask,
 549                               bool log)
 550 {
 551         bool ttl = false, ipv4 = false, ipv6 = false;
 552         __be16 tun_flags = 0;
 553         int opts_type = 0;
 554         struct nlattr *a;
 555         int rem;
 556
 557         nla_for_each_nested(a, attr, rem) {
 558                 int type = nla_type(a);
 559                 int err;
 560
 561                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 562                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 563                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 564                         return -EINVAL;
 565                 }
 566
 567                 if (!check_attr_len(nla_len(a),
 568                                     ovs_tunnel_key_lens[type].len)) {
 569                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 570                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 571                         return -EINVAL;
 572                 }
 573
 574                 switch (type) {
 575                 case OVS_TUNNEL_KEY_ATTR_ID:
 576                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 577                                         nla_get_be64(a), is_mask);
 578                         tun_flags |= TUNNEL_KEY;
 579                         break;
 580                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 581                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
 582                                         nla_get_in_addr(a), is_mask);
 583                         ipv4 = true;
 584                         break;
 585                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 586                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
 587                                         nla_get_in_addr(a), is_mask);
 588                         ipv4 = true;
 589                         break;
 590                 case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
 591                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
 592                                         nla_get_in6_addr(a), is_mask);
 593                         ipv6 = true;
 594                         break;
 595                 case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
 596                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
 597                                         nla_get_in6_addr(a), is_mask);
 598                         ipv6 = true;
 599                         break;
 600                 case OVS_TUNNEL_KEY_ATTR_TOS:
 601                         SW_FLOW_KEY_PUT(match, tun_key.tos,
 602                                         nla_get_u8(a), is_mask);
 603                         break;
 604                 case OVS_TUNNEL_KEY_ATTR_TTL:
 605                         SW_FLOW_KEY_PUT(match, tun_key.ttl,
 606                                         nla_get_u8(a), is_mask);
 607                         ttl = true;
 608                         break;
 609                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 610                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 611                         break;
 612                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 613                         tun_flags |= TUNNEL_CSUM;
 614                         break;
 615                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 616                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 617                                         nla_get_be16(a), is_mask);
 618                         break;
 619                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 620                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 621                                         nla_get_be16(a), is_mask);
 622                         break;
 623                 case OVS_TUNNEL_KEY_ATTR_OAM:
 624                         tun_flags |= TUNNEL_OAM;
 625                         break;
 626                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 627                         if (opts_type) {
 628                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 629                                 return -EINVAL;
 630                         }
 631
 632                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 633                         if (err)
 634                                 return err;
 635
 636                         tun_flags |= TUNNEL_GENEVE_OPT;
 637                         opts_type = type;
 638                         break;
 639                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 640                         if (opts_type) {
 641                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 642                                 return -EINVAL;
 643                         }
 644
 645                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 646                         if (err)
 647                                 return err;
 648
 649                         tun_flags |= TUNNEL_VXLAN_OPT;
 650                         opts_type = type;
 651                         break;
 652                 default:
 653                         OVS_NLERR(log, "Unknown IP tunnel attribute %d",
 654                                   type);
 655                         return -EINVAL;
 656                 }
 657         }
 658
 659         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 660         if (is_mask)
 661                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
 662         else
 663                 SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
 664                                 false);
 665
 666         if (rem > 0) {
 667                 OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
 668                           rem);
 669                 return -EINVAL;
 670         }
 671
 672         if (ipv4 && ipv6) {
 673                 OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
 674                 return -EINVAL;
 675         }
 676
 677         if (!is_mask) {
 678                 if (!ipv4 && !ipv6) {
 679                         OVS_NLERR(log, "IP tunnel dst address not specified");
 680                         return -EINVAL;
 681                 }
 682                 if (ipv4 && !match->key->tun_key.u.ipv4.dst) {
 683                         OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 684                         return -EINVAL;
 685                 }
 686                 if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
 687                         OVS_NLERR(log, "IPv6 tunnel dst address is zero");
 688                         return -EINVAL;
 689                 }
 690
 691                 if (!ttl) {
 692                         OVS_NLERR(log, "IP tunnel TTL not specified.");
 693                         return -EINVAL;
 694                 }
 695         }
 696
 697         return opts_type;
 698 }
 699
 700 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 701                                const void *tun_opts, int swkey_tun_opts_len)
 702 {
 703         const struct vxlan_metadata *opts = tun_opts;
 704         struct nlattr *nla;
 705
 706         nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 707         if (!nla)
 708                 return -EMSGSIZE;
 709
 710         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 711                 return -EMSGSIZE;
 712
 713         nla_nest_end(skb, nla);
 714         return 0;
 715 }
 716
 717 static int __ip_tun_to_nlattr(struct sk_buff *skb,
 718                               const struct ip_tunnel_key *output,
 719                               const void *tun_opts, int swkey_tun_opts_len,
 720                               unsigned short tun_proto)
 721 {
 722         if (output->tun_flags & TUNNEL_KEY &&
 723             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 724                 return -EMSGSIZE;
 725         switch (tun_proto) {
 726         case AF_INET:
 727                 if (output->u.ipv4.src &&
 728                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 729                                     output->u.ipv4.src))
 730                         return -EMSGSIZE;
 731                 if (output->u.ipv4.dst &&
 732                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 733                                     output->u.ipv4.dst))
 734                         return -EMSGSIZE;
 735                 break;
 736         case AF_INET6:
 737                 if (!ipv6_addr_any(&output->u.ipv6.src) &&
 738                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
 739                                      &output->u.ipv6.src))
 740                         return -EMSGSIZE;
 741                 if (!ipv6_addr_any(&output->u.ipv6.dst) &&
 742                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
 743                                      &output->u.ipv6.dst))
 744                         return -EMSGSIZE;
 745                 break;
 746         }
 747         if (output->tos &&
 748             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 749                 return -EMSGSIZE;
 750         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 751                 return -EMSGSIZE;
 752         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 753             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 754                 return -EMSGSIZE;
 755         if ((output->tun_flags & TUNNEL_CSUM) &&
 756             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 757                 return -EMSGSIZE;
 758         if (output->tp_src &&
 759             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 760                 return -EMSGSIZE;
 761         if (output->tp_dst &&
 762             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 763                 return -EMSGSIZE;
 764         if ((output->tun_flags & TUNNEL_OAM) &&
 765             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 766                 return -EMSGSIZE;
 767         if (swkey_tun_opts_len) {
 768                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 769                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 770                             swkey_tun_opts_len, tun_opts))
 771                         return -EMSGSIZE;
 772                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 773                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 774                         return -EMSGSIZE;
 775         }
 776
 777         return 0;
 778 }
 779
 780 static int ip_tun_to_nlattr(struct sk_buff *skb,
 781                             const struct ip_tunnel_key *output,
 782                             const void *tun_opts, int swkey_tun_opts_len,
 783                             unsigned short tun_proto)
 784 {
 785         struct nlattr *nla;
 786         int err;
 787
 788         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 789         if (!nla)
 790                 return -EMSGSIZE;
 791
 792         err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
 793                                  tun_proto);
 794         if (err)
 795                 return err;
 796
 797         nla_nest_end(skb, nla);
 798         return 0;
 799 }
 800
 801 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
 802                             struct ip_tunnel_info *tun_info)
 803 {
 804         return __ip_tun_to_nlattr(skb, &tun_info->key,
 805                                   ip_tunnel_info_opts(tun_info),
 806                                   tun_info->options_len,
 807                                   ip_tunnel_info_af(tun_info));
 808 }
 809
 810 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
 811                                  u64 *attrs, const struct nlattr **a,
 812                                  bool is_mask, bool log)
 813 {
 814         if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
 815                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 816
 817                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 818                 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
 819         }
 820
 821         if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
 822                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 823
 824                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 825                 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
 826         }
 827
 828         if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 829                 SW_FLOW_KEY_PUT(match, phy.priority,
 830                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 831                 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 832         }
 833
 834         if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 835                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 836
 837                 if (is_mask) {
 838                         in_port = 0xffffffff; /* Always exact match in_port. */
 839                 } else if (in_port >= DP_MAX_PORTS) {
 840                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 841                                   in_port, DP_MAX_PORTS);
 842                         return -EINVAL;
 843                 }
 844
 845                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 846                 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
 847         } else if (!is_mask) {
 848                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 849         }
 850
 851         if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
 852                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 853
 854                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 855                 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
 856         }
 857         if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
 858                 if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 859                                        is_mask, log) < 0)
 860                         return -EINVAL;
 861                 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
 862         }
 863
 864         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
 865             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
 866                 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
 867
 868                 if (ct_state & ~CT_SUPPORTED_MASK) {
 869                         OVS_NLERR(log, "ct_state flags %08x unsupported",
 870                                   ct_state);
 871                         return -EINVAL;
 872                 }
 873
 874                 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
 875                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
 876         }
 877         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
 878             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
 879                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
 880
 881                 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
 882                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
 883         }
 884         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
 885             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
 886                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
 887
 888                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
 889                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
 890         }
 891         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
 892             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
 893                 const struct ovs_key_ct_labels *cl;
 894
 895                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
 896                 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
 897                                    sizeof(*cl), is_mask);
 898                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
 899         }
 900         return 0;
 901 }
 902
 903 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
 904                                 u64 attrs, const struct nlattr **a,
 905                                 bool is_mask, bool log)
 906 {
 907         int err;
 908
 909         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
 910         if (err)
 911                 return err;
 912
 913         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
 914                 const struct ovs_key_ethernet *eth_key;
 915
 916                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 917                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 918                                 eth_key->eth_src, ETH_ALEN, is_mask);
 919                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 920                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 921                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
 922         }
 923
 924         if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
 925                 __be16 tci;
 926
 927                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 928                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 929                         if (is_mask)
 930                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 931                         else
 932                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 933
 934                         return -EINVAL;
 935                 }
 936
 937                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 938                 attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
 939         }
 940
 941         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
 942                 __be16 eth_type;
 943
 944                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 945                 if (is_mask) {
 946                         /* Always exact match EtherType. */
 947                         eth_type = htons(0xffff);
 948                 } else if (!eth_proto_is_802_3(eth_type)) {
 949                         OVS_NLERR(log, "EtherType %x is less than min %x",
 950                                   ntohs(eth_type), ETH_P_802_3_MIN);
 951                         return -EINVAL;
 952                 }
 953
 954                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 955                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
 956         } else if (!is_mask) {
 957                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 958         }
 959
 960         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
 961                 const struct ovs_key_ipv4 *ipv4_key;
 962
 963                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 964                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 965                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 966                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 967                         return -EINVAL;
 968                 }
 969                 SW_FLOW_KEY_PUT(match, ip.proto,
 970                                 ipv4_key->ipv4_proto, is_mask);
 971                 SW_FLOW_KEY_PUT(match, ip.tos,
 972                                 ipv4_key->ipv4_tos, is_mask);
 973                 SW_FLOW_KEY_PUT(match, ip.ttl,
 974                                 ipv4_key->ipv4_ttl, is_mask);
 975                 SW_FLOW_KEY_PUT(match, ip.frag,
 976                                 ipv4_key->ipv4_frag, is_mask);
 977                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 978                                 ipv4_key->ipv4_src, is_mask);
 979                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 980                                 ipv4_key->ipv4_dst, is_mask);
 981                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
 982         }
 983
 984         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
 985                 const struct ovs_key_ipv6 *ipv6_key;
 986
 987                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 988                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 989                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 990                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 991                         return -EINVAL;
 992                 }
 993
 994                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 995                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
 996                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 997                         return -EINVAL;
 998                 }
 999
1000                 SW_FLOW_KEY_PUT(match, ipv6.label,
1001                                 ipv6_key->ipv6_label, is_mask);
1002                 SW_FLOW_KEY_PUT(match, ip.proto,
1003                                 ipv6_key->ipv6_proto, is_mask);
1004                 SW_FLOW_KEY_PUT(match, ip.tos,
1005                                 ipv6_key->ipv6_tclass, is_mask);
1006                 SW_FLOW_KEY_PUT(match, ip.ttl,
1007                                 ipv6_key->ipv6_hlimit, is_mask);
1008                 SW_FLOW_KEY_PUT(match, ip.frag,
1009                                 ipv6_key->ipv6_frag, is_mask);
1010                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1011                                 ipv6_key->ipv6_src,
1012                                 sizeof(match->key->ipv6.addr.src),
1013                                 is_mask);
1014                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1015                                 ipv6_key->ipv6_dst,
1016                                 sizeof(match->key->ipv6.addr.dst),
1017                                 is_mask);
1018
1019                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1020         }
1021
1022         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1023                 const struct ovs_key_arp *arp_key;
1024
1025                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1026                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1027                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1028                                   arp_key->arp_op);
1029                         return -EINVAL;
1030                 }
1031
1032                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1033                                 arp_key->arp_sip, is_mask);
1034                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1035                         arp_key->arp_tip, is_mask);
1036                 SW_FLOW_KEY_PUT(match, ip.proto,
1037                                 ntohs(arp_key->arp_op), is_mask);
1038                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1039                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1040                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1041                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1042
1043                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1044         }
1045
1046         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1047                 const struct ovs_key_mpls *mpls_key;
1048
1049                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1050                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1051                                 mpls_key->mpls_lse, is_mask);
1052
1053                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1054          }
1055
1056         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1057                 const struct ovs_key_tcp *tcp_key;
1058
1059                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1060                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1061                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1062                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1063         }
1064
1065         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1066                 SW_FLOW_KEY_PUT(match, tp.flags,
1067                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1068                                 is_mask);
1069                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1070         }
1071
1072         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1073                 const struct ovs_key_udp *udp_key;
1074
1075                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1076                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1077                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1078                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1079         }
1080
1081         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1082                 const struct ovs_key_sctp *sctp_key;
1083
1084                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1085                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1086                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1087                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1088         }
1089
1090         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1091                 const struct ovs_key_icmp *icmp_key;
1092
1093                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1094                 SW_FLOW_KEY_PUT(match, tp.src,
1095                                 htons(icmp_key->icmp_type), is_mask);
1096                 SW_FLOW_KEY_PUT(match, tp.dst,
1097                                 htons(icmp_key->icmp_code), is_mask);
1098                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1099         }
1100
1101         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1102                 const struct ovs_key_icmpv6 *icmpv6_key;
1103
1104                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1105                 SW_FLOW_KEY_PUT(match, tp.src,
1106                                 htons(icmpv6_key->icmpv6_type), is_mask);
1107                 SW_FLOW_KEY_PUT(match, tp.dst,
1108                                 htons(icmpv6_key->icmpv6_code), is_mask);
1109                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1110         }
1111
1112         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1113                 const struct ovs_key_nd *nd_key;
1114
1115                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1116                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1117                         nd_key->nd_target,
1118                         sizeof(match->key->ipv6.nd.target),
1119                         is_mask);
1120                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1121                         nd_key->nd_sll, ETH_ALEN, is_mask);
1122                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1123                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1124                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1125         }
1126
1127         if (attrs != 0) {
1128                 OVS_NLERR(log, "Unknown key attributes %llx",
1129                           (unsigned long long)attrs);
1130                 return -EINVAL;
1131         }
1132
1133         return 0;
1134 }
1135
1136 static void nlattr_set(struct nlattr *attr, u8 val,
1137                        const struct ovs_len_tbl *tbl)
1138 {
1139         struct nlattr *nla;
1140         int rem;
1141
1142         /* The nlattr stream should already have been validated */
1143         nla_for_each_nested(nla, attr, rem) {
1144                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1145                         if (tbl[nla_type(nla)].next)
1146                                 tbl = tbl[nla_type(nla)].next;
1147                         nlattr_set(nla, val, tbl);
1148                 } else {
1149                         memset(nla_data(nla), val, nla_len(nla));
1150                 }
1151
1152                 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1153                         *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1154         }
1155 }
1156
1157 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1158 {
1159         nlattr_set(attr, val, ovs_key_lens);
1160 }
1161
1162 /**
1163  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1164  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1165  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1166  * does not include any don't care bit.
1167  * @net: Used to determine per-namespace field support.
1168  * @match: receives the extracted flow match information.
1169  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1170  * sequence. The fields should of the packet that triggered the creation
1171  * of this flow.
1172  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1173  * attribute specifies the mask field of the wildcarded flow.
1174  * @log: Boolean to allow kernel error logging.  Normally true, but when
1175  * probing for feature compatibility this should be passed in as false to
1176  * suppress unnecessary error logging.
1177  */
1178 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1179                       const struct nlattr *nla_key,
1180                       const struct nlattr *nla_mask,
1181                       bool log)
1182 {
1183         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1184         const struct nlattr *encap;
1185         struct nlattr *newmask = NULL;
1186         u64 key_attrs = 0;
1187         u64 mask_attrs = 0;
1188         bool encap_valid = false;
1189         int err;
1190
1191         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1192         if (err)
1193                 return err;
1194
1195         if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1196             (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1197             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1198                 __be16 tci;
1199
1200                 if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1201                       (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1202                         OVS_NLERR(log, "Invalid Vlan frame.");
1203                         return -EINVAL;
1204                 }
1205
1206                 key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1207                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1208                 encap = a[OVS_KEY_ATTR_ENCAP];
1209                 key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1210                 encap_valid = true;
1211
1212                 if (tci & htons(VLAN_TAG_PRESENT)) {
1213                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1214                         if (err)
1215                                 return err;
1216                 } else if (!tci) {
1217                         /* Corner case for truncated 802.1Q header. */
1218                         if (nla_len(encap)) {
1219                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1220                                 return -EINVAL;
1221                         }
1222                 } else {
1223                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1224                         return  -EINVAL;
1225                 }
1226         }
1227
1228         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1229         if (err)
1230                 return err;
1231
1232         if (match->mask) {
1233                 if (!nla_mask) {
1234                         /* Create an exact match mask. We need to set to 0xff
1235                          * all the 'match->mask' fields that have been touched
1236                          * in 'match->key'. We cannot simply memset
1237                          * 'match->mask', because padding bytes and fields not
1238                          * specified in 'match->key' should be left to 0.
1239                          * Instead, we use a stream of netlink attributes,
1240                          * copied from 'key' and set to 0xff.
1241                          * ovs_key_from_nlattrs() will take care of filling
1242                          * 'match->mask' appropriately.
1243                          */
1244                         newmask = kmemdup(nla_key,
1245                                           nla_total_size(nla_len(nla_key)),
1246                                           GFP_KERNEL);
1247                         if (!newmask)
1248                                 return -ENOMEM;
1249
1250                         mask_set_nlattr(newmask, 0xff);
1251
1252                         /* The userspace does not send tunnel attributes that
1253                          * are 0, but we should not wildcard them nonetheless.
1254                          */
1255                         if (match->key->tun_proto)
1256                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1257                                                          0xff, true);
1258
1259                         nla_mask = newmask;
1260                 }
1261
1262                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1263                 if (err)
1264                         goto free_newmask;
1265
1266                 /* Always match on tci. */
1267                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1268
1269                 if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
1270                         __be16 eth_type = 0;
1271                         __be16 tci = 0;
1272
1273                         if (!encap_valid) {
1274                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1275                                 err = -EINVAL;
1276                                 goto free_newmask;
1277                         }
1278
1279                         mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1280                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1281                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1282
1283                         if (eth_type == htons(0xffff)) {
1284                                 mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1285                                 encap = a[OVS_KEY_ATTR_ENCAP];
1286                                 err = parse_flow_mask_nlattrs(encap, a,
1287                                                               &mask_attrs, log);
1288                                 if (err)
1289                                         goto free_newmask;
1290                         } else {
1291                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1292                                           ntohs(eth_type));
1293                                 err = -EINVAL;
1294                                 goto free_newmask;
1295                         }
1296
1297                         if (a[OVS_KEY_ATTR_VLAN])
1298                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1299
1300                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1301                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1302                                           ntohs(tci));
1303                                 err = -EINVAL;
1304                                 goto free_newmask;
1305                         }
1306                 }
1307
1308                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1309                                            log);
1310                 if (err)
1311                         goto free_newmask;
1312         }
1313
1314         if (!match_validate(match, key_attrs, mask_attrs, log))
1315                 err = -EINVAL;
1316
1317 free_newmask:
1318         kfree(newmask);
1319         return err;
1320 }
1321
1322 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1323 {
1324         size_t len;
1325
1326         if (!attr)
1327                 return 0;
1328
1329         len = nla_len(attr);
1330         if (len < 1 || len > MAX_UFID_LENGTH) {
1331                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1332                           nla_len(attr), MAX_UFID_LENGTH);
1333                 return 0;
1334         }
1335
1336         return len;
1337 }
1338
1339 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1340  * or false otherwise.
1341  */
1342 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1343                       bool log)
1344 {
1345         sfid->ufid_len = get_ufid_len(attr, log);
1346         if (sfid->ufid_len)
1347                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1348
1349         return sfid->ufid_len;
1350 }
1351
1352 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1353                            const struct sw_flow_key *key, bool log)
1354 {
1355         struct sw_flow_key *new_key;
1356
1357         if (ovs_nla_get_ufid(sfid, ufid, log))
1358                 return 0;
1359
1360         /* If UFID was not provided, use unmasked key. */
1361         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1362         if (!new_key)
1363                 return -ENOMEM;
1364         memcpy(new_key, key, sizeof(*key));
1365         sfid->unmasked_key = new_key;
1366
1367         return 0;
1368 }
1369
1370 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1371 {
1372         return attr ? nla_get_u32(attr) : 0;
1373 }
1374
1375 /**
1376  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1377  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1378  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1379  * sequence.
1380  * @log: Boolean to allow kernel error logging.  Normally true, but when
1381  * probing for feature compatibility this should be passed in as false to
1382  * suppress unnecessary error logging.
1383  *
1384  * This parses a series of Netlink attributes that form a flow key, which must
1385  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1386  * get the metadata, that is, the parts of the flow key that cannot be
1387  * extracted from the packet itself.
1388  */
1389
1390 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1391                               struct sw_flow_key *key,
1392                               bool log)
1393 {
1394         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1395         struct sw_flow_match match;
1396         u64 attrs = 0;
1397         int err;
1398
1399         err = parse_flow_nlattrs(attr, a, &attrs, log);
1400         if (err)
1401                 return -EINVAL;
1402
1403         memset(&match, 0, sizeof(match));
1404         match.key = key;
1405
1406         memset(&key->ct, 0, sizeof(key->ct));
1407         key->phy.in_port = DP_MAX_PORTS;
1408
1409         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1410 }
1411
1412 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1413                              const struct sw_flow_key *output, bool is_mask,
1414                              struct sk_buff *skb)
1415 {
1416         struct ovs_key_ethernet *eth_key;
1417         struct nlattr *nla, *encap;
1418
1419         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1420                 goto nla_put_failure;
1421
1422         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1423                 goto nla_put_failure;
1424
1425         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1426                 goto nla_put_failure;
1427
1428         if ((swkey->tun_proto || is_mask)) {
1429                 const void *opts = NULL;
1430
1431                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1432                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1433
1434                 if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
1435                                      swkey->tun_opts_len, swkey->tun_proto))
1436                         goto nla_put_failure;
1437         }
1438
1439         if (swkey->phy.in_port == DP_MAX_PORTS) {
1440                 if (is_mask && (output->phy.in_port == 0xffff))
1441                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1442                                 goto nla_put_failure;
1443         } else {
1444                 u16 upper_u16;
1445                 upper_u16 = !is_mask ? 0 : 0xffff;
1446
1447                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1448                                 (upper_u16 << 16) | output->phy.in_port))
1449                         goto nla_put_failure;
1450         }
1451
1452         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1453                 goto nla_put_failure;
1454
1455         if (ovs_ct_put_key(output, skb))
1456                 goto nla_put_failure;
1457
1458         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1459         if (!nla)
1460                 goto nla_put_failure;
1461
1462         eth_key = nla_data(nla);
1463         ether_addr_copy(eth_key->eth_src, output->eth.src);
1464         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1465
1466         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1467                 __be16 eth_type;
1468                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1469                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1470                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1471                         goto nla_put_failure;
1472                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1473                 if (!swkey->eth.tci)
1474                         goto unencap;
1475         } else
1476                 encap = NULL;
1477
1478         if (swkey->eth.type == htons(ETH_P_802_2)) {
1479                 /*
1480                  * Ethertype 802.2 is represented in the netlink with omitted
1481                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1482                  * 0xffff in the mask attribute.  Ethertype can also
1483                  * be wildcarded.
1484                  */
1485                 if (is_mask && output->eth.type)
1486                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1487                                                 output->eth.type))
1488                                 goto nla_put_failure;
1489                 goto unencap;
1490         }
1491
1492         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1493                 goto nla_put_failure;
1494
1495         if (swkey->eth.type == htons(ETH_P_IP)) {
1496                 struct ovs_key_ipv4 *ipv4_key;
1497
1498                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1499                 if (!nla)
1500                         goto nla_put_failure;
1501                 ipv4_key = nla_data(nla);
1502                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1503                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1504                 ipv4_key->ipv4_proto = output->ip.proto;
1505                 ipv4_key->ipv4_tos = output->ip.tos;
1506                 ipv4_key->ipv4_ttl = output->ip.ttl;
1507                 ipv4_key->ipv4_frag = output->ip.frag;
1508         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1509                 struct ovs_key_ipv6 *ipv6_key;
1510
1511                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1512                 if (!nla)
1513                         goto nla_put_failure;
1514                 ipv6_key = nla_data(nla);
1515                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1516                                 sizeof(ipv6_key->ipv6_src));
1517                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1518                                 sizeof(ipv6_key->ipv6_dst));
1519                 ipv6_key->ipv6_label = output->ipv6.label;
1520                 ipv6_key->ipv6_proto = output->ip.proto;
1521                 ipv6_key->ipv6_tclass = output->ip.tos;
1522                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1523                 ipv6_key->ipv6_frag = output->ip.frag;
1524         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1525                    swkey->eth.type == htons(ETH_P_RARP)) {
1526                 struct ovs_key_arp *arp_key;
1527
1528                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1529                 if (!nla)
1530                         goto nla_put_failure;
1531                 arp_key = nla_data(nla);
1532                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1533                 arp_key->arp_sip = output->ipv4.addr.src;
1534                 arp_key->arp_tip = output->ipv4.addr.dst;
1535                 arp_key->arp_op = htons(output->ip.proto);
1536                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1537                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1538         } else if (eth_p_mpls(swkey->eth.type)) {
1539                 struct ovs_key_mpls *mpls_key;
1540
1541                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1542                 if (!nla)
1543                         goto nla_put_failure;
1544                 mpls_key = nla_data(nla);
1545                 mpls_key->mpls_lse = output->mpls.top_lse;
1546         }
1547
1548         if ((swkey->eth.type == htons(ETH_P_IP) ||
1549              swkey->eth.type == htons(ETH_P_IPV6)) &&
1550              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1551
1552                 if (swkey->ip.proto == IPPROTO_TCP) {
1553                         struct ovs_key_tcp *tcp_key;
1554
1555                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1556                         if (!nla)
1557                                 goto nla_put_failure;
1558                         tcp_key = nla_data(nla);
1559                         tcp_key->tcp_src = output->tp.src;
1560                         tcp_key->tcp_dst = output->tp.dst;
1561                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1562                                          output->tp.flags))
1563                                 goto nla_put_failure;
1564                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1565                         struct ovs_key_udp *udp_key;
1566
1567                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1568                         if (!nla)
1569                                 goto nla_put_failure;
1570                         udp_key = nla_data(nla);
1571                         udp_key->udp_src = output->tp.src;
1572                         udp_key->udp_dst = output->tp.dst;
1573                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1574                         struct ovs_key_sctp *sctp_key;
1575
1576                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1577                         if (!nla)
1578                                 goto nla_put_failure;
1579                         sctp_key = nla_data(nla);
1580                         sctp_key->sctp_src = output->tp.src;
1581                         sctp_key->sctp_dst = output->tp.dst;
1582                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1583                            swkey->ip.proto == IPPROTO_ICMP) {
1584                         struct ovs_key_icmp *icmp_key;
1585
1586                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1587                         if (!nla)
1588                                 goto nla_put_failure;
1589                         icmp_key = nla_data(nla);
1590                         icmp_key->icmp_type = ntohs(output->tp.src);
1591                         icmp_key->icmp_code = ntohs(output->tp.dst);
1592                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1593                            swkey->ip.proto == IPPROTO_ICMPV6) {
1594                         struct ovs_key_icmpv6 *icmpv6_key;
1595
1596                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1597                                                 sizeof(*icmpv6_key));
1598                         if (!nla)
1599                                 goto nla_put_failure;
1600                         icmpv6_key = nla_data(nla);
1601                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1602                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1603
1604                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1605                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1606                                 struct ovs_key_nd *nd_key;
1607
1608                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1609                                 if (!nla)
1610                                         goto nla_put_failure;
1611                                 nd_key = nla_data(nla);
1612                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1613                                                         sizeof(nd_key->nd_target));
1614                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1615                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1616                         }
1617                 }
1618         }
1619
1620 unencap:
1621         if (encap)
1622                 nla_nest_end(skb, encap);
1623
1624         return 0;
1625
1626 nla_put_failure:
1627         return -EMSGSIZE;
1628 }
1629
1630 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1631                     const struct sw_flow_key *output, int attr, bool is_mask,
1632                     struct sk_buff *skb)
1633 {
1634         int err;
1635         struct nlattr *nla;
1636
1637         nla = nla_nest_start(skb, attr);
1638         if (!nla)
1639                 return -EMSGSIZE;
1640         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1641         if (err)
1642                 return err;
1643         nla_nest_end(skb, nla);
1644
1645         return 0;
1646 }
1647
1648 /* Called with ovs_mutex or RCU read lock. */
1649 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1650 {
1651         if (ovs_identifier_is_ufid(&flow->id))
1652                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1653                                flow->id.ufid);
1654
1655         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1656                                OVS_FLOW_ATTR_KEY, false, skb);
1657 }
1658
1659 /* Called with ovs_mutex or RCU read lock. */
1660 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1661 {
1662         return ovs_nla_put_key(&flow->key, &flow->key,
1663                                 OVS_FLOW_ATTR_KEY, false, skb);
1664 }
1665
1666 /* Called with ovs_mutex or RCU read lock. */
1667 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1668 {
1669         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1670                                 OVS_FLOW_ATTR_MASK, true, skb);
1671 }
1672
1673 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1674
1675 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1676 {
1677         struct sw_flow_actions *sfa;
1678
1679         if (size > MAX_ACTIONS_BUFSIZE) {
1680                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1681                 return ERR_PTR(-EINVAL);
1682         }
1683
1684         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1685         if (!sfa)
1686                 return ERR_PTR(-ENOMEM);
1687
1688         sfa->actions_len = 0;
1689         return sfa;
1690 }
1691
1692 static void ovs_nla_free_set_action(const struct nlattr *a)
1693 {
1694         const struct nlattr *ovs_key = nla_data(a);
1695         struct ovs_tunnel_info *ovs_tun;
1696
1697         switch (nla_type(ovs_key)) {
1698         case OVS_KEY_ATTR_TUNNEL_INFO:
1699                 ovs_tun = nla_data(ovs_key);
1700                 dst_release((struct dst_entry *)ovs_tun->tun_dst);
1701                 break;
1702         }
1703 }
1704
1705 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1706 {
1707         const struct nlattr *a;
1708         int rem;
1709
1710         if (!sf_acts)
1711                 return;
1712
1713         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1714                 switch (nla_type(a)) {
1715                 case OVS_ACTION_ATTR_SET:
1716                         ovs_nla_free_set_action(a);
1717                         break;
1718                 case OVS_ACTION_ATTR_CT:
1719                         ovs_ct_free_action(a);
1720                         break;
1721                 }
1722         }
1723
1724         kfree(sf_acts);
1725 }
1726
1727 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1728 {
1729         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1730 }
1731
1732 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1733  * The caller must hold rcu_read_lock for this to be sensible. */
1734 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1735 {
1736         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1737 }
1738
1739 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1740                                        int attr_len, bool log)
1741 {
1742
1743         struct sw_flow_actions *acts;
1744         int new_acts_size;
1745         int req_size = NLA_ALIGN(attr_len);
1746         int next_offset = offsetof(struct sw_flow_actions, actions) +
1747                                         (*sfa)->actions_len;
1748
1749         if (req_size <= (ksize(*sfa) - next_offset))
1750                 goto out;
1751
1752         new_acts_size = ksize(*sfa) * 2;
1753
1754         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1755                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1756                         return ERR_PTR(-EMSGSIZE);
1757                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1758         }
1759
1760         acts = nla_alloc_flow_actions(new_acts_size, log);
1761         if (IS_ERR(acts))
1762                 return (void *)acts;
1763
1764         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1765         acts->actions_len = (*sfa)->actions_len;
1766         acts->orig_len = (*sfa)->orig_len;
1767         kfree(*sfa);
1768         *sfa = acts;
1769
1770 out:
1771         (*sfa)->actions_len += req_size;
1772         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1773 }
1774
1775 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1776                                    int attrtype, void *data, int len, bool log)
1777 {
1778         struct nlattr *a;
1779
1780         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1781         if (IS_ERR(a))
1782                 return a;
1783
1784         a->nla_type = attrtype;
1785         a->nla_len = nla_attr_size(len);
1786
1787         if (data)
1788                 memcpy(nla_data(a), data, len);
1789         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1790
1791         return a;
1792 }
1793
1794 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1795                        int len, bool log)
1796 {
1797         struct nlattr *a;
1798
1799         a = __add_action(sfa, attrtype, data, len, log);
1800
1801         return PTR_ERR_OR_ZERO(a);
1802 }
1803
1804 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1805                                           int attrtype, bool log)
1806 {
1807         int used = (*sfa)->actions_len;
1808         int err;
1809
1810         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1811         if (err)
1812                 return err;
1813
1814         return used;
1815 }
1816
1817 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1818                                          int st_offset)
1819 {
1820         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1821                                                                st_offset);
1822
1823         a->nla_len = sfa->actions_len - st_offset;
1824 }
1825
1826 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1827                                   const struct sw_flow_key *key,
1828                                   int depth, struct sw_flow_actions **sfa,
1829                                   __be16 eth_type, __be16 vlan_tci, bool log);
1830
1831 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1832                                     const struct sw_flow_key *key, int depth,
1833                                     struct sw_flow_actions **sfa,
1834                                     __be16 eth_type, __be16 vlan_tci, bool log)
1835 {
1836         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1837         const struct nlattr *probability, *actions;
1838         const struct nlattr *a;
1839         int rem, start, err, st_acts;
1840
1841         memset(attrs, 0, sizeof(attrs));
1842         nla_for_each_nested(a, attr, rem) {
1843                 int type = nla_type(a);
1844                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1845                         return -EINVAL;
1846                 attrs[type] = a;
1847         }
1848         if (rem)
1849                 return -EINVAL;
1850
1851         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1852         if (!probability || nla_len(probability) != sizeof(u32))
1853                 return -EINVAL;
1854
1855         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1856         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1857                 return -EINVAL;
1858
1859         /* validation done, copy sample action. */
1860         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1861         if (start < 0)
1862                 return start;
1863         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1864                                  nla_data(probability), sizeof(u32), log);
1865         if (err)
1866                 return err;
1867         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1868         if (st_acts < 0)
1869                 return st_acts;
1870
1871         err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1872                                      eth_type, vlan_tci, log);
1873         if (err)
1874                 return err;
1875
1876         add_nested_action_end(*sfa, st_acts);
1877         add_nested_action_end(*sfa, start);
1878
1879         return 0;
1880 }
1881
1882 void ovs_match_init(struct sw_flow_match *match,
1883                     struct sw_flow_key *key,
1884                     struct sw_flow_mask *mask)
1885 {
1886         memset(match, 0, sizeof(*match));
1887         match->key = key;
1888         match->mask = mask;
1889
1890         memset(key, 0, sizeof(*key));
1891
1892         if (mask) {
1893                 memset(&mask->key, 0, sizeof(mask->key));
1894                 mask->range.start = mask->range.end = 0;
1895         }
1896 }
1897
1898 static int validate_geneve_opts(struct sw_flow_key *key)
1899 {
1900         struct geneve_opt *option;
1901         int opts_len = key->tun_opts_len;
1902         bool crit_opt = false;
1903
1904         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1905         while (opts_len > 0) {
1906                 int len;
1907
1908                 if (opts_len < sizeof(*option))
1909                         return -EINVAL;
1910
1911                 len = sizeof(*option) + option->length * 4;
1912                 if (len > opts_len)
1913                         return -EINVAL;
1914
1915                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1916
1917                 option = (struct geneve_opt *)((u8 *)option + len);
1918                 opts_len -= len;
1919         };
1920
1921         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1922
1923         return 0;
1924 }
1925
1926 static int validate_and_copy_set_tun(const struct nlattr *attr,
1927                                      struct sw_flow_actions **sfa, bool log)
1928 {
1929         struct sw_flow_match match;
1930         struct sw_flow_key key;
1931         struct metadata_dst *tun_dst;
1932         struct ip_tunnel_info *tun_info;
1933         struct ovs_tunnel_info *ovs_tun;
1934         struct nlattr *a;
1935         int err = 0, start, opts_type;
1936
1937         ovs_match_init(&match, &key, NULL);
1938         opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
1939         if (opts_type < 0)
1940                 return opts_type;
1941
1942         if (key.tun_opts_len) {
1943                 switch (opts_type) {
1944                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1945                         err = validate_geneve_opts(&key);
1946                         if (err < 0)
1947                                 return err;
1948                         break;
1949                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1950                         break;
1951                 }
1952         };
1953
1954         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1955         if (start < 0)
1956                 return start;
1957
1958         tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1959         if (!tun_dst)
1960                 return -ENOMEM;
1961
1962         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1963                          sizeof(*ovs_tun), log);
1964         if (IS_ERR(a)) {
1965                 dst_release((struct dst_entry *)tun_dst);
1966                 return PTR_ERR(a);
1967         }
1968
1969         ovs_tun = nla_data(a);
1970         ovs_tun->tun_dst = tun_dst;
1971
1972         tun_info = &tun_dst->u.tun_info;
1973         tun_info->mode = IP_TUNNEL_INFO_TX;
1974         if (key.tun_proto == AF_INET6)
1975                 tun_info->mode |= IP_TUNNEL_INFO_IPV6;
1976         tun_info->key = key.tun_key;
1977
1978         /* We need to store the options in the action itself since
1979          * everything else will go away after flow setup. We can append
1980          * it to tun_info and then point there.
1981          */
1982         ip_tunnel_info_opts_set(tun_info,
1983                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
1984                                 key.tun_opts_len);
1985         add_nested_action_end(*sfa, start);
1986
1987         return err;
1988 }
1989
1990 /* Return false if there are any non-masked bits set.
1991  * Mask follows data immediately, before any netlink padding.
1992  */
1993 static bool validate_masked(u8 *data, int len)
1994 {
1995         u8 *mask = data + len;
1996
1997         while (len--)
1998                 if (*data++ & ~*mask++)
1999                         return false;
2000
2001         return true;
2002 }
2003
2004 static int validate_set(const struct nlattr *a,
2005                         const struct sw_flow_key *flow_key,
2006                         struct sw_flow_actions **sfa,
2007                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
2008 {
2009         const struct nlattr *ovs_key = nla_data(a);
2010         int key_type = nla_type(ovs_key);
2011         size_t key_len;
2012
2013         /* There can be only one key in a action */
2014         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2015                 return -EINVAL;
2016
2017         key_len = nla_len(ovs_key);
2018         if (masked)
2019                 key_len /= 2;
2020
2021         if (key_type > OVS_KEY_ATTR_MAX ||
2022             !check_attr_len(key_len, ovs_key_lens[key_type].len))
2023                 return -EINVAL;
2024
2025         if (masked && !validate_masked(nla_data(ovs_key), key_len))
2026                 return -EINVAL;
2027
2028         switch (key_type) {
2029         const struct ovs_key_ipv4 *ipv4_key;
2030         const struct ovs_key_ipv6 *ipv6_key;
2031         int err;
2032
2033         case OVS_KEY_ATTR_PRIORITY:
2034         case OVS_KEY_ATTR_SKB_MARK:
2035         case OVS_KEY_ATTR_CT_MARK:
2036         case OVS_KEY_ATTR_CT_LABELS:
2037         case OVS_KEY_ATTR_ETHERNET:
2038                 break;
2039
2040         case OVS_KEY_ATTR_TUNNEL:
2041                 if (eth_p_mpls(eth_type))
2042                         return -EINVAL;
2043
2044                 if (masked)
2045                         return -EINVAL; /* Masked tunnel set not supported. */
2046
2047                 *skip_copy = true;
2048                 err = validate_and_copy_set_tun(a, sfa, log);
2049                 if (err)
2050                         return err;
2051                 break;
2052
2053         case OVS_KEY_ATTR_IPV4:
2054                 if (eth_type != htons(ETH_P_IP))
2055                         return -EINVAL;
2056
2057                 ipv4_key = nla_data(ovs_key);
2058
2059                 if (masked) {
2060                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2061
2062                         /* Non-writeable fields. */
2063                         if (mask->ipv4_proto || mask->ipv4_frag)
2064                                 return -EINVAL;
2065                 } else {
2066                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2067                                 return -EINVAL;
2068
2069                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2070                                 return -EINVAL;
2071                 }
2072                 break;
2073
2074         case OVS_KEY_ATTR_IPV6:
2075                 if (eth_type != htons(ETH_P_IPV6))
2076                         return -EINVAL;
2077
2078                 ipv6_key = nla_data(ovs_key);
2079
2080                 if (masked) {
2081                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2082
2083                         /* Non-writeable fields. */
2084                         if (mask->ipv6_proto || mask->ipv6_frag)
2085                                 return -EINVAL;
2086
2087                         /* Invalid bits in the flow label mask? */
2088                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2089                                 return -EINVAL;
2090                 } else {
2091                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2092                                 return -EINVAL;
2093
2094                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2095                                 return -EINVAL;
2096                 }
2097                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2098                         return -EINVAL;
2099
2100                 break;
2101
2102         case OVS_KEY_ATTR_TCP:
2103                 if ((eth_type != htons(ETH_P_IP) &&
2104                      eth_type != htons(ETH_P_IPV6)) ||
2105                     flow_key->ip.proto != IPPROTO_TCP)
2106                         return -EINVAL;
2107
2108                 break;
2109
2110         case OVS_KEY_ATTR_UDP:
2111                 if ((eth_type != htons(ETH_P_IP) &&
2112                      eth_type != htons(ETH_P_IPV6)) ||
2113                     flow_key->ip.proto != IPPROTO_UDP)
2114                         return -EINVAL;
2115
2116                 break;
2117
2118         case OVS_KEY_ATTR_MPLS:
2119                 if (!eth_p_mpls(eth_type))
2120                         return -EINVAL;
2121                 break;
2122
2123         case OVS_KEY_ATTR_SCTP:
2124                 if ((eth_type != htons(ETH_P_IP) &&
2125                      eth_type != htons(ETH_P_IPV6)) ||
2126                     flow_key->ip.proto != IPPROTO_SCTP)
2127                         return -EINVAL;
2128
2129                 break;
2130
2131         default:
2132                 return -EINVAL;
2133         }
2134
2135         /* Convert non-masked non-tunnel set actions to masked set actions. */
2136         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2137                 int start, len = key_len * 2;
2138                 struct nlattr *at;
2139
2140                 *skip_copy = true;
2141
2142                 start = add_nested_action_start(sfa,
2143                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2144                                                 log);
2145                 if (start < 0)
2146                         return start;
2147
2148                 at = __add_action(sfa, key_type, NULL, len, log);
2149                 if (IS_ERR(at))
2150                         return PTR_ERR(at);
2151
2152                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2153                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2154                 /* Clear non-writeable bits from otherwise writeable fields. */
2155                 if (key_type == OVS_KEY_ATTR_IPV6) {
2156                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2157
2158                         mask->ipv6_label &= htonl(0x000FFFFF);
2159                 }
2160                 add_nested_action_end(*sfa, start);
2161         }
2162
2163         return 0;
2164 }
2165
2166 static int validate_userspace(const struct nlattr *attr)
2167 {
2168         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2169                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2170                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2171                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2172         };
2173         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2174         int error;
2175
2176         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2177                                  attr, userspace_policy);
2178         if (error)
2179                 return error;
2180
2181         if (!a[OVS_USERSPACE_ATTR_PID] ||
2182             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2183                 return -EINVAL;
2184
2185         return 0;
2186 }
2187
2188 static int copy_action(const struct nlattr *from,
2189                        struct sw_flow_actions **sfa, bool log)
2190 {
2191         int totlen = NLA_ALIGN(from->nla_len);
2192         struct nlattr *to;
2193
2194         to = reserve_sfa_size(sfa, from->nla_len, log);
2195         if (IS_ERR(to))
2196                 return PTR_ERR(to);
2197
2198         memcpy(to, from, totlen);
2199         return 0;
2200 }
2201
2202 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2203                                   const struct sw_flow_key *key,
2204                                   int depth, struct sw_flow_actions **sfa,
2205                                   __be16 eth_type, __be16 vlan_tci, bool log)
2206 {
2207         const struct nlattr *a;
2208         int rem, err;
2209
2210         if (depth >= SAMPLE_ACTION_DEPTH)
2211                 return -EOVERFLOW;
2212
2213         nla_for_each_nested(a, attr, rem) {
2214                 /* Expected argument lengths, (u32)-1 for variable length. */
2215                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2216                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2217                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2218                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2219                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2220                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2221                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2222                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2223                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2224                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2225                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2226                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2227                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2228                 };
2229                 const struct ovs_action_push_vlan *vlan;
2230                 int type = nla_type(a);
2231                 bool skip_copy;
2232
2233                 if (type > OVS_ACTION_ATTR_MAX ||
2234                     (action_lens[type] != nla_len(a) &&
2235                      action_lens[type] != (u32)-1))
2236                         return -EINVAL;
2237
2238                 skip_copy = false;
2239                 switch (type) {
2240                 case OVS_ACTION_ATTR_UNSPEC:
2241                         return -EINVAL;
2242
2243                 case OVS_ACTION_ATTR_USERSPACE:
2244                         err = validate_userspace(a);
2245                         if (err)
2246                                 return err;
2247                         break;
2248
2249                 case OVS_ACTION_ATTR_OUTPUT:
2250                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2251                                 return -EINVAL;
2252                         break;
2253
2254                 case OVS_ACTION_ATTR_HASH: {
2255                         const struct ovs_action_hash *act_hash = nla_data(a);
2256
2257                         switch (act_hash->hash_alg) {
2258                         case OVS_HASH_ALG_L4:
2259                                 break;
2260                         default:
2261                                 return  -EINVAL;
2262                         }
2263
2264                         break;
2265                 }
2266
2267                 case OVS_ACTION_ATTR_POP_VLAN:
2268                         vlan_tci = htons(0);
2269                         break;
2270
2271                 case OVS_ACTION_ATTR_PUSH_VLAN:
2272                         vlan = nla_data(a);
2273                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2274                                 return -EINVAL;
2275                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2276                                 return -EINVAL;
2277                         vlan_tci = vlan->vlan_tci;
2278                         break;
2279
2280                 case OVS_ACTION_ATTR_RECIRC:
2281                         break;
2282
2283                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2284                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2285
2286                         if (!eth_p_mpls(mpls->mpls_ethertype))
2287                                 return -EINVAL;
2288                         /* Prohibit push MPLS other than to a white list
2289                          * for packets that have a known tag order.
2290                          */
2291                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2292                             (eth_type != htons(ETH_P_IP) &&
2293                              eth_type != htons(ETH_P_IPV6) &&
2294                              eth_type != htons(ETH_P_ARP) &&
2295                              eth_type != htons(ETH_P_RARP) &&
2296                              !eth_p_mpls(eth_type)))
2297                                 return -EINVAL;
2298                         eth_type = mpls->mpls_ethertype;
2299                         break;
2300                 }
2301
2302                 case OVS_ACTION_ATTR_POP_MPLS:
2303                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2304                             !eth_p_mpls(eth_type))
2305                                 return -EINVAL;
2306
2307                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2308                          * as there is no check here to ensure that the new
2309                          * eth_type is valid and thus set actions could
2310                          * write off the end of the packet or otherwise
2311                          * corrupt it.
2312                          *
2313                          * Support for these actions is planned using packet
2314                          * recirculation.
2315                          */
2316                         eth_type = htons(0);
2317                         break;
2318
2319                 case OVS_ACTION_ATTR_SET:
2320                         err = validate_set(a, key, sfa,
2321                                            &skip_copy, eth_type, false, log);
2322                         if (err)
2323                                 return err;
2324                         break;
2325
2326                 case OVS_ACTION_ATTR_SET_MASKED:
2327                         err = validate_set(a, key, sfa,
2328                                            &skip_copy, eth_type, true, log);
2329                         if (err)
2330                                 return err;
2331                         break;
2332
2333                 case OVS_ACTION_ATTR_SAMPLE:
2334                         err = validate_and_copy_sample(net, a, key, depth, sfa,
2335                                                        eth_type, vlan_tci, log);
2336                         if (err)
2337                                 return err;
2338                         skip_copy = true;
2339                         break;
2340
2341                 case OVS_ACTION_ATTR_CT:
2342                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2343                         if (err)
2344                                 return err;
2345                         skip_copy = true;
2346                         break;
2347
2348                 default:
2349                         OVS_NLERR(log, "Unknown Action type %d", type);
2350                         return -EINVAL;
2351                 }
2352                 if (!skip_copy) {
2353                         err = copy_action(a, sfa, log);
2354                         if (err)
2355                                 return err;
2356                 }
2357         }
2358
2359         if (rem > 0)
2360                 return -EINVAL;
2361
2362         return 0;
2363 }
2364
2365 /* 'key' must be the masked key. */
2366 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2367                          const struct sw_flow_key *key,
2368                          struct sw_flow_actions **sfa, bool log)
2369 {
2370         int err;
2371
2372         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2373         if (IS_ERR(*sfa))
2374                 return PTR_ERR(*sfa);
2375
2376         (*sfa)->orig_len = nla_len(attr);
2377         err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2378                                      key->eth.tci, log);
2379         if (err)
2380                 ovs_nla_free_flow_actions(*sfa);
2381
2382         return err;
2383 }
2384
2385 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2386 {
2387         const struct nlattr *a;
2388         struct nlattr *start;
2389         int err = 0, rem;
2390
2391         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2392         if (!start)
2393                 return -EMSGSIZE;
2394
2395         nla_for_each_nested(a, attr, rem) {
2396                 int type = nla_type(a);
2397                 struct nlattr *st_sample;
2398
2399                 switch (type) {
2400                 case OVS_SAMPLE_ATTR_PROBABILITY:
2401                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2402                                     sizeof(u32), nla_data(a)))
2403                                 return -EMSGSIZE;
2404                         break;
2405                 case OVS_SAMPLE_ATTR_ACTIONS:
2406                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2407                         if (!st_sample)
2408                                 return -EMSGSIZE;
2409                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2410                         if (err)
2411                                 return err;
2412                         nla_nest_end(skb, st_sample);
2413                         break;
2414                 }
2415         }
2416
2417         nla_nest_end(skb, start);
2418         return err;
2419 }
2420
2421 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2422 {
2423         const struct nlattr *ovs_key = nla_data(a);
2424         int key_type = nla_type(ovs_key);
2425         struct nlattr *start;
2426         int err;
2427
2428         switch (key_type) {
2429         case OVS_KEY_ATTR_TUNNEL_INFO: {
2430                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2431                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2432
2433                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2434                 if (!start)
2435                         return -EMSGSIZE;
2436
2437                 err = ovs_nla_put_tunnel_info(skb, tun_info);
2438                 if (err)
2439                         return err;
2440                 nla_nest_end(skb, start);
2441                 break;
2442         }
2443         default:
2444                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2445                         return -EMSGSIZE;
2446                 break;
2447         }
2448
2449         return 0;
2450 }
2451
2452 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2453                                                 struct sk_buff *skb)
2454 {
2455         const struct nlattr *ovs_key = nla_data(a);
2456         struct nlattr *nla;
2457         size_t key_len = nla_len(ovs_key) / 2;
2458
2459         /* Revert the conversion we did from a non-masked set action to
2460          * masked set action.
2461          */
2462         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2463         if (!nla)
2464                 return -EMSGSIZE;
2465
2466         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2467                 return -EMSGSIZE;
2468
2469         nla_nest_end(skb, nla);
2470         return 0;
2471 }
2472
2473 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2474 {
2475         const struct nlattr *a;
2476         int rem, err;
2477
2478         nla_for_each_attr(a, attr, len, rem) {
2479                 int type = nla_type(a);
2480
2481                 switch (type) {
2482                 case OVS_ACTION_ATTR_SET:
2483                         err = set_action_to_attr(a, skb);
2484                         if (err)
2485                                 return err;
2486                         break;
2487
2488                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2489                         err = masked_set_action_to_set_action_attr(a, skb);
2490                         if (err)
2491                                 return err;
2492                         break;
2493
2494                 case OVS_ACTION_ATTR_SAMPLE:
2495                         err = sample_action_to_attr(a, skb);
2496                         if (err)
2497                                 return err;
2498                         break;
2499
2500                 case OVS_ACTION_ATTR_CT:
2501                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2502                         if (err)
2503                                 return err;
2504                         break;
2505
2506                 default:
2507                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2508                                 return -EMSGSIZE;
2509                         break;
2510                 }
2511         }
2512
2513         return 0;
2514 }