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_64bit(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                          OVS_TUNNEL_KEY_ATTR_PAD))
 725                 return -EMSGSIZE;
 726         switch (tun_proto) {
 727         case AF_INET:
 728                 if (output->u.ipv4.src &&
 729                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 730                                     output->u.ipv4.src))
 731                         return -EMSGSIZE;
 732                 if (output->u.ipv4.dst &&
 733                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 734                                     output->u.ipv4.dst))
 735                         return -EMSGSIZE;
 736                 break;
 737         case AF_INET6:
 738                 if (!ipv6_addr_any(&output->u.ipv6.src) &&
 739                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
 740                                      &output->u.ipv6.src))
 741                         return -EMSGSIZE;
 742                 if (!ipv6_addr_any(&output->u.ipv6.dst) &&
 743                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
 744                                      &output->u.ipv6.dst))
 745                         return -EMSGSIZE;
 746                 break;
 747         }
 748         if (output->tos &&
 749             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 750                 return -EMSGSIZE;
 751         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 752                 return -EMSGSIZE;
 753         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 754             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 755                 return -EMSGSIZE;
 756         if ((output->tun_flags & TUNNEL_CSUM) &&
 757             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 758                 return -EMSGSIZE;
 759         if (output->tp_src &&
 760             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 761                 return -EMSGSIZE;
 762         if (output->tp_dst &&
 763             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 764                 return -EMSGSIZE;
 765         if ((output->tun_flags & TUNNEL_OAM) &&
 766             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 767                 return -EMSGSIZE;
 768         if (swkey_tun_opts_len) {
 769                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 770                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 771                             swkey_tun_opts_len, tun_opts))
 772                         return -EMSGSIZE;
 773                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 774                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 775                         return -EMSGSIZE;
 776         }
 777
 778         return 0;
 779 }
 780
 781 static int ip_tun_to_nlattr(struct sk_buff *skb,
 782                             const struct ip_tunnel_key *output,
 783                             const void *tun_opts, int swkey_tun_opts_len,
 784                             unsigned short tun_proto)
 785 {
 786         struct nlattr *nla;
 787         int err;
 788
 789         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 790         if (!nla)
 791                 return -EMSGSIZE;
 792
 793         err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
 794                                  tun_proto);
 795         if (err)
 796                 return err;
 797
 798         nla_nest_end(skb, nla);
 799         return 0;
 800 }
 801
 802 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
 803                             struct ip_tunnel_info *tun_info)
 804 {
 805         return __ip_tun_to_nlattr(skb, &tun_info->key,
 806                                   ip_tunnel_info_opts(tun_info),
 807                                   tun_info->options_len,
 808                                   ip_tunnel_info_af(tun_info));
 809 }
 810
 811 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
 812                                  u64 *attrs, const struct nlattr **a,
 813                                  bool is_mask, bool log)
 814 {
 815         if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
 816                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 817
 818                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 819                 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
 820         }
 821
 822         if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
 823                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 824
 825                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 826                 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
 827         }
 828
 829         if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 830                 SW_FLOW_KEY_PUT(match, phy.priority,
 831                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 832                 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 833         }
 834
 835         if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 836                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 837
 838                 if (is_mask) {
 839                         in_port = 0xffffffff; /* Always exact match in_port. */
 840                 } else if (in_port >= DP_MAX_PORTS) {
 841                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 842                                   in_port, DP_MAX_PORTS);
 843                         return -EINVAL;
 844                 }
 845
 846                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 847                 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
 848         } else if (!is_mask) {
 849                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 850         }
 851
 852         if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
 853                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 854
 855                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 856                 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
 857         }
 858         if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
 859                 if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 860                                        is_mask, log) < 0)
 861                         return -EINVAL;
 862                 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
 863         }
 864
 865         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
 866             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
 867                 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
 868
 869                 if (ct_state & ~CT_SUPPORTED_MASK) {
 870                         OVS_NLERR(log, "ct_state flags %08x unsupported",
 871                                   ct_state);
 872                         return -EINVAL;
 873                 }
 874
 875                 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
 876                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
 877         }
 878         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
 879             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
 880                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
 881
 882                 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
 883                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
 884         }
 885         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
 886             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
 887                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
 888
 889                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
 890                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
 891         }
 892         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
 893             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
 894                 const struct ovs_key_ct_labels *cl;
 895
 896                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
 897                 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
 898                                    sizeof(*cl), is_mask);
 899                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
 900         }
 901         return 0;
 902 }
 903
 904 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
 905                                 u64 attrs, const struct nlattr **a,
 906                                 bool is_mask, bool log)
 907 {
 908         int err;
 909
 910         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
 911         if (err)
 912                 return err;
 913
 914         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
 915                 const struct ovs_key_ethernet *eth_key;
 916
 917                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 918                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 919                                 eth_key->eth_src, ETH_ALEN, is_mask);
 920                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 921                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 922                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
 923         }
 924
 925         if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
 926                 __be16 tci;
 927
 928                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 929                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 930                         if (is_mask)
 931                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 932                         else
 933                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 934
 935                         return -EINVAL;
 936                 }
 937
 938                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 939                 attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
 940         }
 941
 942         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
 943                 __be16 eth_type;
 944
 945                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 946                 if (is_mask) {
 947                         /* Always exact match EtherType. */
 948                         eth_type = htons(0xffff);
 949                 } else if (!eth_proto_is_802_3(eth_type)) {
 950                         OVS_NLERR(log, "EtherType %x is less than min %x",
 951                                   ntohs(eth_type), ETH_P_802_3_MIN);
 952                         return -EINVAL;
 953                 }
 954
 955                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 956                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
 957         } else if (!is_mask) {
 958                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 959         }
 960
 961         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
 962                 const struct ovs_key_ipv4 *ipv4_key;
 963
 964                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 965                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 966                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 967                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 968                         return -EINVAL;
 969                 }
 970                 SW_FLOW_KEY_PUT(match, ip.proto,
 971                                 ipv4_key->ipv4_proto, is_mask);
 972                 SW_FLOW_KEY_PUT(match, ip.tos,
 973                                 ipv4_key->ipv4_tos, is_mask);
 974                 SW_FLOW_KEY_PUT(match, ip.ttl,
 975                                 ipv4_key->ipv4_ttl, is_mask);
 976                 SW_FLOW_KEY_PUT(match, ip.frag,
 977                                 ipv4_key->ipv4_frag, is_mask);
 978                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 979                                 ipv4_key->ipv4_src, is_mask);
 980                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 981                                 ipv4_key->ipv4_dst, is_mask);
 982                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
 983         }
 984
 985         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
 986                 const struct ovs_key_ipv6 *ipv6_key;
 987
 988                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 989                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 990                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 991                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 992                         return -EINVAL;
 993                 }
 994
 995                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 996                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
 997                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 998                         return -EINVAL;
 999                 }
1000
1001                 SW_FLOW_KEY_PUT(match, ipv6.label,
1002                                 ipv6_key->ipv6_label, is_mask);
1003                 SW_FLOW_KEY_PUT(match, ip.proto,
1004                                 ipv6_key->ipv6_proto, is_mask);
1005                 SW_FLOW_KEY_PUT(match, ip.tos,
1006                                 ipv6_key->ipv6_tclass, is_mask);
1007                 SW_FLOW_KEY_PUT(match, ip.ttl,
1008                                 ipv6_key->ipv6_hlimit, is_mask);
1009                 SW_FLOW_KEY_PUT(match, ip.frag,
1010                                 ipv6_key->ipv6_frag, is_mask);
1011                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1012                                 ipv6_key->ipv6_src,
1013                                 sizeof(match->key->ipv6.addr.src),
1014                                 is_mask);
1015                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1016                                 ipv6_key->ipv6_dst,
1017                                 sizeof(match->key->ipv6.addr.dst),
1018                                 is_mask);
1019
1020                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1021         }
1022
1023         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1024                 const struct ovs_key_arp *arp_key;
1025
1026                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1027                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1028                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1029                                   arp_key->arp_op);
1030                         return -EINVAL;
1031                 }
1032
1033                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1034                                 arp_key->arp_sip, is_mask);
1035                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1036                         arp_key->arp_tip, is_mask);
1037                 SW_FLOW_KEY_PUT(match, ip.proto,
1038                                 ntohs(arp_key->arp_op), is_mask);
1039                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1040                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1041                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1042                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1043
1044                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1045         }
1046
1047         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1048                 const struct ovs_key_mpls *mpls_key;
1049
1050                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1051                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1052                                 mpls_key->mpls_lse, is_mask);
1053
1054                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1055          }
1056
1057         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1058                 const struct ovs_key_tcp *tcp_key;
1059
1060                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1061                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1062                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1063                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1064         }
1065
1066         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1067                 SW_FLOW_KEY_PUT(match, tp.flags,
1068                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1069                                 is_mask);
1070                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1071         }
1072
1073         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1074                 const struct ovs_key_udp *udp_key;
1075
1076                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1077                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1078                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1079                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1080         }
1081
1082         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1083                 const struct ovs_key_sctp *sctp_key;
1084
1085                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1086                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1087                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1088                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1089         }
1090
1091         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1092                 const struct ovs_key_icmp *icmp_key;
1093
1094                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1095                 SW_FLOW_KEY_PUT(match, tp.src,
1096                                 htons(icmp_key->icmp_type), is_mask);
1097                 SW_FLOW_KEY_PUT(match, tp.dst,
1098                                 htons(icmp_key->icmp_code), is_mask);
1099                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1100         }
1101
1102         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1103                 const struct ovs_key_icmpv6 *icmpv6_key;
1104
1105                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1106                 SW_FLOW_KEY_PUT(match, tp.src,
1107                                 htons(icmpv6_key->icmpv6_type), is_mask);
1108                 SW_FLOW_KEY_PUT(match, tp.dst,
1109                                 htons(icmpv6_key->icmpv6_code), is_mask);
1110                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1111         }
1112
1113         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1114                 const struct ovs_key_nd *nd_key;
1115
1116                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1117                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1118                         nd_key->nd_target,
1119                         sizeof(match->key->ipv6.nd.target),
1120                         is_mask);
1121                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1122                         nd_key->nd_sll, ETH_ALEN, is_mask);
1123                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1124                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1125                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1126         }
1127
1128         if (attrs != 0) {
1129                 OVS_NLERR(log, "Unknown key attributes %llx",
1130                           (unsigned long long)attrs);
1131                 return -EINVAL;
1132         }
1133
1134         return 0;
1135 }
1136
1137 static void nlattr_set(struct nlattr *attr, u8 val,
1138                        const struct ovs_len_tbl *tbl)
1139 {
1140         struct nlattr *nla;
1141         int rem;
1142
1143         /* The nlattr stream should already have been validated */
1144         nla_for_each_nested(nla, attr, rem) {
1145                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1146                         if (tbl[nla_type(nla)].next)
1147                                 tbl = tbl[nla_type(nla)].next;
1148                         nlattr_set(nla, val, tbl);
1149                 } else {
1150                         memset(nla_data(nla), val, nla_len(nla));
1151                 }
1152
1153                 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1154                         *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1155         }
1156 }
1157
1158 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1159 {
1160         nlattr_set(attr, val, ovs_key_lens);
1161 }
1162
1163 /**
1164  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1165  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1166  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1167  * does not include any don't care bit.
1168  * @net: Used to determine per-namespace field support.
1169  * @match: receives the extracted flow match information.
1170  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1171  * sequence. The fields should of the packet that triggered the creation
1172  * of this flow.
1173  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1174  * attribute specifies the mask field of the wildcarded flow.
1175  * @log: Boolean to allow kernel error logging.  Normally true, but when
1176  * probing for feature compatibility this should be passed in as false to
1177  * suppress unnecessary error logging.
1178  */
1179 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1180                       const struct nlattr *nla_key,
1181                       const struct nlattr *nla_mask,
1182                       bool log)
1183 {
1184         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1185         const struct nlattr *encap;
1186         struct nlattr *newmask = NULL;
1187         u64 key_attrs = 0;
1188         u64 mask_attrs = 0;
1189         bool encap_valid = false;
1190         int err;
1191
1192         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1193         if (err)
1194                 return err;
1195
1196         if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1197             (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1198             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1199                 __be16 tci;
1200
1201                 if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1202                       (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1203                         OVS_NLERR(log, "Invalid Vlan frame.");
1204                         return -EINVAL;
1205                 }
1206
1207                 key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1208                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1209                 encap = a[OVS_KEY_ATTR_ENCAP];
1210                 key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1211                 encap_valid = true;
1212
1213                 if (tci & htons(VLAN_TAG_PRESENT)) {
1214                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1215                         if (err)
1216                                 return err;
1217                 } else if (!tci) {
1218                         /* Corner case for truncated 802.1Q header. */
1219                         if (nla_len(encap)) {
1220                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1221                                 return -EINVAL;
1222                         }
1223                 } else {
1224                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1225                         return  -EINVAL;
1226                 }
1227         }
1228
1229         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1230         if (err)
1231                 return err;
1232
1233         if (match->mask) {
1234                 if (!nla_mask) {
1235                         /* Create an exact match mask. We need to set to 0xff
1236                          * all the 'match->mask' fields that have been touched
1237                          * in 'match->key'. We cannot simply memset
1238                          * 'match->mask', because padding bytes and fields not
1239                          * specified in 'match->key' should be left to 0.
1240                          * Instead, we use a stream of netlink attributes,
1241                          * copied from 'key' and set to 0xff.
1242                          * ovs_key_from_nlattrs() will take care of filling
1243                          * 'match->mask' appropriately.
1244                          */
1245                         newmask = kmemdup(nla_key,
1246                                           nla_total_size(nla_len(nla_key)),
1247                                           GFP_KERNEL);
1248                         if (!newmask)
1249                                 return -ENOMEM;
1250
1251                         mask_set_nlattr(newmask, 0xff);
1252
1253                         /* The userspace does not send tunnel attributes that
1254                          * are 0, but we should not wildcard them nonetheless.
1255                          */
1256                         if (match->key->tun_proto)
1257                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1258                                                          0xff, true);
1259
1260                         nla_mask = newmask;
1261                 }
1262
1263                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1264                 if (err)
1265                         goto free_newmask;
1266
1267                 /* Always match on tci. */
1268                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1269
1270                 if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
1271                         __be16 eth_type = 0;
1272                         __be16 tci = 0;
1273
1274                         if (!encap_valid) {
1275                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1276                                 err = -EINVAL;
1277                                 goto free_newmask;
1278                         }
1279
1280                         mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1281                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1282                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1283
1284                         if (eth_type == htons(0xffff)) {
1285                                 mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1286                                 encap = a[OVS_KEY_ATTR_ENCAP];
1287                                 err = parse_flow_mask_nlattrs(encap, a,
1288                                                               &mask_attrs, log);
1289                                 if (err)
1290                                         goto free_newmask;
1291                         } else {
1292                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1293                                           ntohs(eth_type));
1294                                 err = -EINVAL;
1295                                 goto free_newmask;
1296                         }
1297
1298                         if (a[OVS_KEY_ATTR_VLAN])
1299                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1300
1301                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1302                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1303                                           ntohs(tci));
1304                                 err = -EINVAL;
1305                                 goto free_newmask;
1306                         }
1307                 }
1308
1309                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1310                                            log);
1311                 if (err)
1312                         goto free_newmask;
1313         }
1314
1315         if (!match_validate(match, key_attrs, mask_attrs, log))
1316                 err = -EINVAL;
1317
1318 free_newmask:
1319         kfree(newmask);
1320         return err;
1321 }
1322
1323 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1324 {
1325         size_t len;
1326
1327         if (!attr)
1328                 return 0;
1329
1330         len = nla_len(attr);
1331         if (len < 1 || len > MAX_UFID_LENGTH) {
1332                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1333                           nla_len(attr), MAX_UFID_LENGTH);
1334                 return 0;
1335         }
1336
1337         return len;
1338 }
1339
1340 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1341  * or false otherwise.
1342  */
1343 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1344                       bool log)
1345 {
1346         sfid->ufid_len = get_ufid_len(attr, log);
1347         if (sfid->ufid_len)
1348                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1349
1350         return sfid->ufid_len;
1351 }
1352
1353 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1354                            const struct sw_flow_key *key, bool log)
1355 {
1356         struct sw_flow_key *new_key;
1357
1358         if (ovs_nla_get_ufid(sfid, ufid, log))
1359                 return 0;
1360
1361         /* If UFID was not provided, use unmasked key. */
1362         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1363         if (!new_key)
1364                 return -ENOMEM;
1365         memcpy(new_key, key, sizeof(*key));
1366         sfid->unmasked_key = new_key;
1367
1368         return 0;
1369 }
1370
1371 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1372 {
1373         return attr ? nla_get_u32(attr) : 0;
1374 }
1375
1376 /**
1377  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1378  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1379  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1380  * sequence.
1381  * @log: Boolean to allow kernel error logging.  Normally true, but when
1382  * probing for feature compatibility this should be passed in as false to
1383  * suppress unnecessary error logging.
1384  *
1385  * This parses a series of Netlink attributes that form a flow key, which must
1386  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1387  * get the metadata, that is, the parts of the flow key that cannot be
1388  * extracted from the packet itself.
1389  */
1390
1391 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1392                               struct sw_flow_key *key,
1393                               bool log)
1394 {
1395         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1396         struct sw_flow_match match;
1397         u64 attrs = 0;
1398         int err;
1399
1400         err = parse_flow_nlattrs(attr, a, &attrs, log);
1401         if (err)
1402                 return -EINVAL;
1403
1404         memset(&match, 0, sizeof(match));
1405         match.key = key;
1406
1407         memset(&key->ct, 0, sizeof(key->ct));
1408         key->phy.in_port = DP_MAX_PORTS;
1409
1410         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1411 }
1412
1413 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1414                              const struct sw_flow_key *output, bool is_mask,
1415                              struct sk_buff *skb)
1416 {
1417         struct ovs_key_ethernet *eth_key;
1418         struct nlattr *nla, *encap;
1419
1420         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1421                 goto nla_put_failure;
1422
1423         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1424                 goto nla_put_failure;
1425
1426         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1427                 goto nla_put_failure;
1428
1429         if ((swkey->tun_proto || is_mask)) {
1430                 const void *opts = NULL;
1431
1432                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1433                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1434
1435                 if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
1436                                      swkey->tun_opts_len, swkey->tun_proto))
1437                         goto nla_put_failure;
1438         }
1439
1440         if (swkey->phy.in_port == DP_MAX_PORTS) {
1441                 if (is_mask && (output->phy.in_port == 0xffff))
1442                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1443                                 goto nla_put_failure;
1444         } else {
1445                 u16 upper_u16;
1446                 upper_u16 = !is_mask ? 0 : 0xffff;
1447
1448                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1449                                 (upper_u16 << 16) | output->phy.in_port))
1450                         goto nla_put_failure;
1451         }
1452
1453         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1454                 goto nla_put_failure;
1455
1456         if (ovs_ct_put_key(output, skb))
1457                 goto nla_put_failure;
1458
1459         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1460         if (!nla)
1461                 goto nla_put_failure;
1462
1463         eth_key = nla_data(nla);
1464         ether_addr_copy(eth_key->eth_src, output->eth.src);
1465         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1466
1467         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1468                 __be16 eth_type;
1469                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1470                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1471                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1472                         goto nla_put_failure;
1473                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1474                 if (!swkey->eth.tci)
1475                         goto unencap;
1476         } else
1477                 encap = NULL;
1478
1479         if (swkey->eth.type == htons(ETH_P_802_2)) {
1480                 /*
1481                  * Ethertype 802.2 is represented in the netlink with omitted
1482                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1483                  * 0xffff in the mask attribute.  Ethertype can also
1484                  * be wildcarded.
1485                  */
1486                 if (is_mask && output->eth.type)
1487                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1488                                                 output->eth.type))
1489                                 goto nla_put_failure;
1490                 goto unencap;
1491         }
1492
1493         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1494                 goto nla_put_failure;
1495
1496         if (swkey->eth.type == htons(ETH_P_IP)) {
1497                 struct ovs_key_ipv4 *ipv4_key;
1498
1499                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1500                 if (!nla)
1501                         goto nla_put_failure;
1502                 ipv4_key = nla_data(nla);
1503                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1504                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1505                 ipv4_key->ipv4_proto = output->ip.proto;
1506                 ipv4_key->ipv4_tos = output->ip.tos;
1507                 ipv4_key->ipv4_ttl = output->ip.ttl;
1508                 ipv4_key->ipv4_frag = output->ip.frag;
1509         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1510                 struct ovs_key_ipv6 *ipv6_key;
1511
1512                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1513                 if (!nla)
1514                         goto nla_put_failure;
1515                 ipv6_key = nla_data(nla);
1516                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1517                                 sizeof(ipv6_key->ipv6_src));
1518                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1519                                 sizeof(ipv6_key->ipv6_dst));
1520                 ipv6_key->ipv6_label = output->ipv6.label;
1521                 ipv6_key->ipv6_proto = output->ip.proto;
1522                 ipv6_key->ipv6_tclass = output->ip.tos;
1523                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1524                 ipv6_key->ipv6_frag = output->ip.frag;
1525         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1526                    swkey->eth.type == htons(ETH_P_RARP)) {
1527                 struct ovs_key_arp *arp_key;
1528
1529                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1530                 if (!nla)
1531                         goto nla_put_failure;
1532                 arp_key = nla_data(nla);
1533                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1534                 arp_key->arp_sip = output->ipv4.addr.src;
1535                 arp_key->arp_tip = output->ipv4.addr.dst;
1536                 arp_key->arp_op = htons(output->ip.proto);
1537                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1538                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1539         } else if (eth_p_mpls(swkey->eth.type)) {
1540                 struct ovs_key_mpls *mpls_key;
1541
1542                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1543                 if (!nla)
1544                         goto nla_put_failure;
1545                 mpls_key = nla_data(nla);
1546                 mpls_key->mpls_lse = output->mpls.top_lse;
1547         }
1548
1549         if ((swkey->eth.type == htons(ETH_P_IP) ||
1550              swkey->eth.type == htons(ETH_P_IPV6)) &&
1551              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1552
1553                 if (swkey->ip.proto == IPPROTO_TCP) {
1554                         struct ovs_key_tcp *tcp_key;
1555
1556                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1557                         if (!nla)
1558                                 goto nla_put_failure;
1559                         tcp_key = nla_data(nla);
1560                         tcp_key->tcp_src = output->tp.src;
1561                         tcp_key->tcp_dst = output->tp.dst;
1562                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1563                                          output->tp.flags))
1564                                 goto nla_put_failure;
1565                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1566                         struct ovs_key_udp *udp_key;
1567
1568                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1569                         if (!nla)
1570                                 goto nla_put_failure;
1571                         udp_key = nla_data(nla);
1572                         udp_key->udp_src = output->tp.src;
1573                         udp_key->udp_dst = output->tp.dst;
1574                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1575                         struct ovs_key_sctp *sctp_key;
1576
1577                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1578                         if (!nla)
1579                                 goto nla_put_failure;
1580                         sctp_key = nla_data(nla);
1581                         sctp_key->sctp_src = output->tp.src;
1582                         sctp_key->sctp_dst = output->tp.dst;
1583                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1584                            swkey->ip.proto == IPPROTO_ICMP) {
1585                         struct ovs_key_icmp *icmp_key;
1586
1587                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1588                         if (!nla)
1589                                 goto nla_put_failure;
1590                         icmp_key = nla_data(nla);
1591                         icmp_key->icmp_type = ntohs(output->tp.src);
1592                         icmp_key->icmp_code = ntohs(output->tp.dst);
1593                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1594                            swkey->ip.proto == IPPROTO_ICMPV6) {
1595                         struct ovs_key_icmpv6 *icmpv6_key;
1596
1597                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1598                                                 sizeof(*icmpv6_key));
1599                         if (!nla)
1600                                 goto nla_put_failure;
1601                         icmpv6_key = nla_data(nla);
1602                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1603                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1604
1605                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1606                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1607                                 struct ovs_key_nd *nd_key;
1608
1609                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1610                                 if (!nla)
1611                                         goto nla_put_failure;
1612                                 nd_key = nla_data(nla);
1613                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1614                                                         sizeof(nd_key->nd_target));
1615                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1616                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1617                         }
1618                 }
1619         }
1620
1621 unencap:
1622         if (encap)
1623                 nla_nest_end(skb, encap);
1624
1625         return 0;
1626
1627 nla_put_failure:
1628         return -EMSGSIZE;
1629 }
1630
1631 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1632                     const struct sw_flow_key *output, int attr, bool is_mask,
1633                     struct sk_buff *skb)
1634 {
1635         int err;
1636         struct nlattr *nla;
1637
1638         nla = nla_nest_start(skb, attr);
1639         if (!nla)
1640                 return -EMSGSIZE;
1641         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1642         if (err)
1643                 return err;
1644         nla_nest_end(skb, nla);
1645
1646         return 0;
1647 }
1648
1649 /* Called with ovs_mutex or RCU read lock. */
1650 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1651 {
1652         if (ovs_identifier_is_ufid(&flow->id))
1653                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1654                                flow->id.ufid);
1655
1656         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1657                                OVS_FLOW_ATTR_KEY, false, skb);
1658 }
1659
1660 /* Called with ovs_mutex or RCU read lock. */
1661 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1662 {
1663         return ovs_nla_put_key(&flow->key, &flow->key,
1664                                 OVS_FLOW_ATTR_KEY, false, skb);
1665 }
1666
1667 /* Called with ovs_mutex or RCU read lock. */
1668 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1669 {
1670         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1671                                 OVS_FLOW_ATTR_MASK, true, skb);
1672 }
1673
1674 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1675
1676 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1677 {
1678         struct sw_flow_actions *sfa;
1679
1680         if (size > MAX_ACTIONS_BUFSIZE) {
1681                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1682                 return ERR_PTR(-EINVAL);
1683         }
1684
1685         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1686         if (!sfa)
1687                 return ERR_PTR(-ENOMEM);
1688
1689         sfa->actions_len = 0;
1690         return sfa;
1691 }
1692
1693 static void ovs_nla_free_set_action(const struct nlattr *a)
1694 {
1695         const struct nlattr *ovs_key = nla_data(a);
1696         struct ovs_tunnel_info *ovs_tun;
1697
1698         switch (nla_type(ovs_key)) {
1699         case OVS_KEY_ATTR_TUNNEL_INFO:
1700                 ovs_tun = nla_data(ovs_key);
1701                 dst_release((struct dst_entry *)ovs_tun->tun_dst);
1702                 break;
1703         }
1704 }
1705
1706 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1707 {
1708         const struct nlattr *a;
1709         int rem;
1710
1711         if (!sf_acts)
1712                 return;
1713
1714         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1715                 switch (nla_type(a)) {
1716                 case OVS_ACTION_ATTR_SET:
1717                         ovs_nla_free_set_action(a);
1718                         break;
1719                 case OVS_ACTION_ATTR_CT:
1720                         ovs_ct_free_action(a);
1721                         break;
1722                 }
1723         }
1724
1725         kfree(sf_acts);
1726 }
1727
1728 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1729 {
1730         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1731 }
1732
1733 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1734  * The caller must hold rcu_read_lock for this to be sensible. */
1735 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1736 {
1737         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1738 }
1739
1740 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1741                                        int attr_len, bool log)
1742 {
1743
1744         struct sw_flow_actions *acts;
1745         int new_acts_size;
1746         int req_size = NLA_ALIGN(attr_len);
1747         int next_offset = offsetof(struct sw_flow_actions, actions) +
1748                                         (*sfa)->actions_len;
1749
1750         if (req_size <= (ksize(*sfa) - next_offset))
1751                 goto out;
1752
1753         new_acts_size = ksize(*sfa) * 2;
1754
1755         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1756                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1757                         return ERR_PTR(-EMSGSIZE);
1758                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1759         }
1760
1761         acts = nla_alloc_flow_actions(new_acts_size, log);
1762         if (IS_ERR(acts))
1763                 return (void *)acts;
1764
1765         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1766         acts->actions_len = (*sfa)->actions_len;
1767         acts->orig_len = (*sfa)->orig_len;
1768         kfree(*sfa);
1769         *sfa = acts;
1770
1771 out:
1772         (*sfa)->actions_len += req_size;
1773         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1774 }
1775
1776 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1777                                    int attrtype, void *data, int len, bool log)
1778 {
1779         struct nlattr *a;
1780
1781         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1782         if (IS_ERR(a))
1783                 return a;
1784
1785         a->nla_type = attrtype;
1786         a->nla_len = nla_attr_size(len);
1787
1788         if (data)
1789                 memcpy(nla_data(a), data, len);
1790         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1791
1792         return a;
1793 }
1794
1795 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1796                        int len, bool log)
1797 {
1798         struct nlattr *a;
1799
1800         a = __add_action(sfa, attrtype, data, len, log);
1801
1802         return PTR_ERR_OR_ZERO(a);
1803 }
1804
1805 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1806                                           int attrtype, bool log)
1807 {
1808         int used = (*sfa)->actions_len;
1809         int err;
1810
1811         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1812         if (err)
1813                 return err;
1814
1815         return used;
1816 }
1817
1818 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1819                                          int st_offset)
1820 {
1821         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1822                                                                st_offset);
1823
1824         a->nla_len = sfa->actions_len - st_offset;
1825 }
1826
1827 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1828                                   const struct sw_flow_key *key,
1829                                   int depth, struct sw_flow_actions **sfa,
1830                                   __be16 eth_type, __be16 vlan_tci, bool log);
1831
1832 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1833                                     const struct sw_flow_key *key, int depth,
1834                                     struct sw_flow_actions **sfa,
1835                                     __be16 eth_type, __be16 vlan_tci, bool log)
1836 {
1837         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1838         const struct nlattr *probability, *actions;
1839         const struct nlattr *a;
1840         int rem, start, err, st_acts;
1841
1842         memset(attrs, 0, sizeof(attrs));
1843         nla_for_each_nested(a, attr, rem) {
1844                 int type = nla_type(a);
1845                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1846                         return -EINVAL;
1847                 attrs[type] = a;
1848         }
1849         if (rem)
1850                 return -EINVAL;
1851
1852         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1853         if (!probability || nla_len(probability) != sizeof(u32))
1854                 return -EINVAL;
1855
1856         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1857         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1858                 return -EINVAL;
1859
1860         /* validation done, copy sample action. */
1861         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1862         if (start < 0)
1863                 return start;
1864         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1865                                  nla_data(probability), sizeof(u32), log);
1866         if (err)
1867                 return err;
1868         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1869         if (st_acts < 0)
1870                 return st_acts;
1871
1872         err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1873                                      eth_type, vlan_tci, log);
1874         if (err)
1875                 return err;
1876
1877         add_nested_action_end(*sfa, st_acts);
1878         add_nested_action_end(*sfa, start);
1879
1880         return 0;
1881 }
1882
1883 void ovs_match_init(struct sw_flow_match *match,
1884                     struct sw_flow_key *key,
1885                     struct sw_flow_mask *mask)
1886 {
1887         memset(match, 0, sizeof(*match));
1888         match->key = key;
1889         match->mask = mask;
1890
1891         memset(key, 0, sizeof(*key));
1892
1893         if (mask) {
1894                 memset(&mask->key, 0, sizeof(mask->key));
1895                 mask->range.start = mask->range.end = 0;
1896         }
1897 }
1898
1899 static int validate_geneve_opts(struct sw_flow_key *key)
1900 {
1901         struct geneve_opt *option;
1902         int opts_len = key->tun_opts_len;
1903         bool crit_opt = false;
1904
1905         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1906         while (opts_len > 0) {
1907                 int len;
1908
1909                 if (opts_len < sizeof(*option))
1910                         return -EINVAL;
1911
1912                 len = sizeof(*option) + option->length * 4;
1913                 if (len > opts_len)
1914                         return -EINVAL;
1915
1916                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1917
1918                 option = (struct geneve_opt *)((u8 *)option + len);
1919                 opts_len -= len;
1920         };
1921
1922         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1923
1924         return 0;
1925 }
1926
1927 static int validate_and_copy_set_tun(const struct nlattr *attr,
1928                                      struct sw_flow_actions **sfa, bool log)
1929 {
1930         struct sw_flow_match match;
1931         struct sw_flow_key key;
1932         struct metadata_dst *tun_dst;
1933         struct ip_tunnel_info *tun_info;
1934         struct ovs_tunnel_info *ovs_tun;
1935         struct nlattr *a;
1936         int err = 0, start, opts_type;
1937
1938         ovs_match_init(&match, &key, NULL);
1939         opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
1940         if (opts_type < 0)
1941                 return opts_type;
1942
1943         if (key.tun_opts_len) {
1944                 switch (opts_type) {
1945                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1946                         err = validate_geneve_opts(&key);
1947                         if (err < 0)
1948                                 return err;
1949                         break;
1950                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1951                         break;
1952                 }
1953         };
1954
1955         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1956         if (start < 0)
1957                 return start;
1958
1959         tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1960         if (!tun_dst)
1961                 return -ENOMEM;
1962
1963         err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
1964         if (err) {
1965                 dst_release((struct dst_entry *)tun_dst);
1966                 return err;
1967         }
1968
1969         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1970                          sizeof(*ovs_tun), log);
1971         if (IS_ERR(a)) {
1972                 dst_release((struct dst_entry *)tun_dst);
1973                 return PTR_ERR(a);
1974         }
1975
1976         ovs_tun = nla_data(a);
1977         ovs_tun->tun_dst = tun_dst;
1978
1979         tun_info = &tun_dst->u.tun_info;
1980         tun_info->mode = IP_TUNNEL_INFO_TX;
1981         if (key.tun_proto == AF_INET6)
1982                 tun_info->mode |= IP_TUNNEL_INFO_IPV6;
1983         tun_info->key = key.tun_key;
1984
1985         /* We need to store the options in the action itself since
1986          * everything else will go away after flow setup. We can append
1987          * it to tun_info and then point there.
1988          */
1989         ip_tunnel_info_opts_set(tun_info,
1990                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
1991                                 key.tun_opts_len);
1992         add_nested_action_end(*sfa, start);
1993
1994         return err;
1995 }
1996
1997 /* Return false if there are any non-masked bits set.
1998  * Mask follows data immediately, before any netlink padding.
1999  */
2000 static bool validate_masked(u8 *data, int len)
2001 {
2002         u8 *mask = data + len;
2003
2004         while (len--)
2005                 if (*data++ & ~*mask++)
2006                         return false;
2007
2008         return true;
2009 }
2010
2011 static int validate_set(const struct nlattr *a,
2012                         const struct sw_flow_key *flow_key,
2013                         struct sw_flow_actions **sfa,
2014                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
2015 {
2016         const struct nlattr *ovs_key = nla_data(a);
2017         int key_type = nla_type(ovs_key);
2018         size_t key_len;
2019
2020         /* There can be only one key in a action */
2021         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2022                 return -EINVAL;
2023
2024         key_len = nla_len(ovs_key);
2025         if (masked)
2026                 key_len /= 2;
2027
2028         if (key_type > OVS_KEY_ATTR_MAX ||
2029             !check_attr_len(key_len, ovs_key_lens[key_type].len))
2030                 return -EINVAL;
2031
2032         if (masked && !validate_masked(nla_data(ovs_key), key_len))
2033                 return -EINVAL;
2034
2035         switch (key_type) {
2036         const struct ovs_key_ipv4 *ipv4_key;
2037         const struct ovs_key_ipv6 *ipv6_key;
2038         int err;
2039
2040         case OVS_KEY_ATTR_PRIORITY:
2041         case OVS_KEY_ATTR_SKB_MARK:
2042         case OVS_KEY_ATTR_CT_MARK:
2043         case OVS_KEY_ATTR_CT_LABELS:
2044         case OVS_KEY_ATTR_ETHERNET:
2045                 break;
2046
2047         case OVS_KEY_ATTR_TUNNEL:
2048                 if (masked)
2049                         return -EINVAL; /* Masked tunnel set not supported. */
2050
2051                 *skip_copy = true;
2052                 err = validate_and_copy_set_tun(a, sfa, log);
2053                 if (err)
2054                         return err;
2055                 break;
2056
2057         case OVS_KEY_ATTR_IPV4:
2058                 if (eth_type != htons(ETH_P_IP))
2059                         return -EINVAL;
2060
2061                 ipv4_key = nla_data(ovs_key);
2062
2063                 if (masked) {
2064                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2065
2066                         /* Non-writeable fields. */
2067                         if (mask->ipv4_proto || mask->ipv4_frag)
2068                                 return -EINVAL;
2069                 } else {
2070                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2071                                 return -EINVAL;
2072
2073                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2074                                 return -EINVAL;
2075                 }
2076                 break;
2077
2078         case OVS_KEY_ATTR_IPV6:
2079                 if (eth_type != htons(ETH_P_IPV6))
2080                         return -EINVAL;
2081
2082                 ipv6_key = nla_data(ovs_key);
2083
2084                 if (masked) {
2085                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2086
2087                         /* Non-writeable fields. */
2088                         if (mask->ipv6_proto || mask->ipv6_frag)
2089                                 return -EINVAL;
2090
2091                         /* Invalid bits in the flow label mask? */
2092                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2093                                 return -EINVAL;
2094                 } else {
2095                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2096                                 return -EINVAL;
2097
2098                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2099                                 return -EINVAL;
2100                 }
2101                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2102                         return -EINVAL;
2103
2104                 break;
2105
2106         case OVS_KEY_ATTR_TCP:
2107                 if ((eth_type != htons(ETH_P_IP) &&
2108                      eth_type != htons(ETH_P_IPV6)) ||
2109                     flow_key->ip.proto != IPPROTO_TCP)
2110                         return -EINVAL;
2111
2112                 break;
2113
2114         case OVS_KEY_ATTR_UDP:
2115                 if ((eth_type != htons(ETH_P_IP) &&
2116                      eth_type != htons(ETH_P_IPV6)) ||
2117                     flow_key->ip.proto != IPPROTO_UDP)
2118                         return -EINVAL;
2119
2120                 break;
2121
2122         case OVS_KEY_ATTR_MPLS:
2123                 if (!eth_p_mpls(eth_type))
2124                         return -EINVAL;
2125                 break;
2126
2127         case OVS_KEY_ATTR_SCTP:
2128                 if ((eth_type != htons(ETH_P_IP) &&
2129                      eth_type != htons(ETH_P_IPV6)) ||
2130                     flow_key->ip.proto != IPPROTO_SCTP)
2131                         return -EINVAL;
2132
2133                 break;
2134
2135         default:
2136                 return -EINVAL;
2137         }
2138
2139         /* Convert non-masked non-tunnel set actions to masked set actions. */
2140         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2141                 int start, len = key_len * 2;
2142                 struct nlattr *at;
2143
2144                 *skip_copy = true;
2145
2146                 start = add_nested_action_start(sfa,
2147                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2148                                                 log);
2149                 if (start < 0)
2150                         return start;
2151
2152                 at = __add_action(sfa, key_type, NULL, len, log);
2153                 if (IS_ERR(at))
2154                         return PTR_ERR(at);
2155
2156                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2157                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2158                 /* Clear non-writeable bits from otherwise writeable fields. */
2159                 if (key_type == OVS_KEY_ATTR_IPV6) {
2160                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2161
2162                         mask->ipv6_label &= htonl(0x000FFFFF);
2163                 }
2164                 add_nested_action_end(*sfa, start);
2165         }
2166
2167         return 0;
2168 }
2169
2170 static int validate_userspace(const struct nlattr *attr)
2171 {
2172         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2173                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2174                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2175                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2176         };
2177         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2178         int error;
2179
2180         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2181                                  attr, userspace_policy);
2182         if (error)
2183                 return error;
2184
2185         if (!a[OVS_USERSPACE_ATTR_PID] ||
2186             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2187                 return -EINVAL;
2188
2189         return 0;
2190 }
2191
2192 static int copy_action(const struct nlattr *from,
2193                        struct sw_flow_actions **sfa, bool log)
2194 {
2195         int totlen = NLA_ALIGN(from->nla_len);
2196         struct nlattr *to;
2197
2198         to = reserve_sfa_size(sfa, from->nla_len, log);
2199         if (IS_ERR(to))
2200                 return PTR_ERR(to);
2201
2202         memcpy(to, from, totlen);
2203         return 0;
2204 }
2205
2206 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2207                                   const struct sw_flow_key *key,
2208                                   int depth, struct sw_flow_actions **sfa,
2209                                   __be16 eth_type, __be16 vlan_tci, bool log)
2210 {
2211         const struct nlattr *a;
2212         int rem, err;
2213
2214         if (depth >= SAMPLE_ACTION_DEPTH)
2215                 return -EOVERFLOW;
2216
2217         nla_for_each_nested(a, attr, rem) {
2218                 /* Expected argument lengths, (u32)-1 for variable length. */
2219                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2220                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2221                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2222                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2223                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2224                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2225                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2226                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2227                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2228                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2229                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2230                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2231                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2232                         [OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
2233                 };
2234                 const struct ovs_action_push_vlan *vlan;
2235                 int type = nla_type(a);
2236                 bool skip_copy;
2237
2238                 if (type > OVS_ACTION_ATTR_MAX ||
2239                     (action_lens[type] != nla_len(a) &&
2240                      action_lens[type] != (u32)-1))
2241                         return -EINVAL;
2242
2243                 skip_copy = false;
2244                 switch (type) {
2245                 case OVS_ACTION_ATTR_UNSPEC:
2246                         return -EINVAL;
2247
2248                 case OVS_ACTION_ATTR_USERSPACE:
2249                         err = validate_userspace(a);
2250                         if (err)
2251                                 return err;
2252                         break;
2253
2254                 case OVS_ACTION_ATTR_OUTPUT:
2255                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2256                                 return -EINVAL;
2257                         break;
2258
2259                 case OVS_ACTION_ATTR_TRUNC: {
2260                         const struct ovs_action_trunc *trunc = nla_data(a);
2261
2262                         if (trunc->max_len < ETH_HLEN)
2263                                 return -EINVAL;
2264                         break;
2265                 }
2266
2267                 case OVS_ACTION_ATTR_HASH: {
2268                         const struct ovs_action_hash *act_hash = nla_data(a);
2269
2270                         switch (act_hash->hash_alg) {
2271                         case OVS_HASH_ALG_L4:
2272                                 break;
2273                         default:
2274                                 return  -EINVAL;
2275                         }
2276
2277                         break;
2278                 }
2279
2280                 case OVS_ACTION_ATTR_POP_VLAN:
2281                         vlan_tci = htons(0);
2282                         break;
2283
2284                 case OVS_ACTION_ATTR_PUSH_VLAN:
2285                         vlan = nla_data(a);
2286                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2287                                 return -EINVAL;
2288                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2289                                 return -EINVAL;
2290                         vlan_tci = vlan->vlan_tci;
2291                         break;
2292
2293                 case OVS_ACTION_ATTR_RECIRC:
2294                         break;
2295
2296                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2297                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2298
2299                         if (!eth_p_mpls(mpls->mpls_ethertype))
2300                                 return -EINVAL;
2301                         /* Prohibit push MPLS other than to a white list
2302                          * for packets that have a known tag order.
2303                          */
2304                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2305                             (eth_type != htons(ETH_P_IP) &&
2306                              eth_type != htons(ETH_P_IPV6) &&
2307                              eth_type != htons(ETH_P_ARP) &&
2308                              eth_type != htons(ETH_P_RARP) &&
2309                              !eth_p_mpls(eth_type)))
2310                                 return -EINVAL;
2311                         eth_type = mpls->mpls_ethertype;
2312                         break;
2313                 }
2314
2315                 case OVS_ACTION_ATTR_POP_MPLS:
2316                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2317                             !eth_p_mpls(eth_type))
2318                                 return -EINVAL;
2319
2320                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2321                          * as there is no check here to ensure that the new
2322                          * eth_type is valid and thus set actions could
2323                          * write off the end of the packet or otherwise
2324                          * corrupt it.
2325                          *
2326                          * Support for these actions is planned using packet
2327                          * recirculation.
2328                          */
2329                         eth_type = htons(0);
2330                         break;
2331
2332                 case OVS_ACTION_ATTR_SET:
2333                         err = validate_set(a, key, sfa,
2334                                            &skip_copy, eth_type, false, log);
2335                         if (err)
2336                                 return err;
2337                         break;
2338
2339                 case OVS_ACTION_ATTR_SET_MASKED:
2340                         err = validate_set(a, key, sfa,
2341                                            &skip_copy, eth_type, true, log);
2342                         if (err)
2343                                 return err;
2344                         break;
2345
2346                 case OVS_ACTION_ATTR_SAMPLE:
2347                         err = validate_and_copy_sample(net, a, key, depth, sfa,
2348                                                        eth_type, vlan_tci, log);
2349                         if (err)
2350                                 return err;
2351                         skip_copy = true;
2352                         break;
2353
2354                 case OVS_ACTION_ATTR_CT:
2355                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2356                         if (err)
2357                                 return err;
2358                         skip_copy = true;
2359                         break;
2360
2361                 default:
2362                         OVS_NLERR(log, "Unknown Action type %d", type);
2363                         return -EINVAL;
2364                 }
2365                 if (!skip_copy) {
2366                         err = copy_action(a, sfa, log);
2367                         if (err)
2368                                 return err;
2369                 }
2370         }
2371
2372         if (rem > 0)
2373                 return -EINVAL;
2374
2375         return 0;
2376 }
2377
2378 /* 'key' must be the masked key. */
2379 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2380                          const struct sw_flow_key *key,
2381                          struct sw_flow_actions **sfa, bool log)
2382 {
2383         int err;
2384
2385         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2386         if (IS_ERR(*sfa))
2387                 return PTR_ERR(*sfa);
2388
2389         (*sfa)->orig_len = nla_len(attr);
2390         err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2391                                      key->eth.tci, log);
2392         if (err)
2393                 ovs_nla_free_flow_actions(*sfa);
2394
2395         return err;
2396 }
2397
2398 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2399 {
2400         const struct nlattr *a;
2401         struct nlattr *start;
2402         int err = 0, rem;
2403
2404         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2405         if (!start)
2406                 return -EMSGSIZE;
2407
2408         nla_for_each_nested(a, attr, rem) {
2409                 int type = nla_type(a);
2410                 struct nlattr *st_sample;
2411
2412                 switch (type) {
2413                 case OVS_SAMPLE_ATTR_PROBABILITY:
2414                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2415                                     sizeof(u32), nla_data(a)))
2416                                 return -EMSGSIZE;
2417                         break;
2418                 case OVS_SAMPLE_ATTR_ACTIONS:
2419                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2420                         if (!st_sample)
2421                                 return -EMSGSIZE;
2422                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2423                         if (err)
2424                                 return err;
2425                         nla_nest_end(skb, st_sample);
2426                         break;
2427                 }
2428         }
2429
2430         nla_nest_end(skb, start);
2431         return err;
2432 }
2433
2434 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2435 {
2436         const struct nlattr *ovs_key = nla_data(a);
2437         int key_type = nla_type(ovs_key);
2438         struct nlattr *start;
2439         int err;
2440
2441         switch (key_type) {
2442         case OVS_KEY_ATTR_TUNNEL_INFO: {
2443                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2444                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2445
2446                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2447                 if (!start)
2448                         return -EMSGSIZE;
2449
2450                 err =  ip_tun_to_nlattr(skb, &tun_info->key,
2451                                         ip_tunnel_info_opts(tun_info),
2452                                         tun_info->options_len,
2453                                         ip_tunnel_info_af(tun_info));
2454                 if (err)
2455                         return err;
2456                 nla_nest_end(skb, start);
2457                 break;
2458         }
2459         default:
2460                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2461                         return -EMSGSIZE;
2462                 break;
2463         }
2464
2465         return 0;
2466 }
2467
2468 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2469                                                 struct sk_buff *skb)
2470 {
2471         const struct nlattr *ovs_key = nla_data(a);
2472         struct nlattr *nla;
2473         size_t key_len = nla_len(ovs_key) / 2;
2474
2475         /* Revert the conversion we did from a non-masked set action to
2476          * masked set action.
2477          */
2478         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2479         if (!nla)
2480                 return -EMSGSIZE;
2481
2482         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2483                 return -EMSGSIZE;
2484
2485         nla_nest_end(skb, nla);
2486         return 0;
2487 }
2488
2489 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2490 {
2491         const struct nlattr *a;
2492         int rem, err;
2493
2494         nla_for_each_attr(a, attr, len, rem) {
2495                 int type = nla_type(a);
2496
2497                 switch (type) {
2498                 case OVS_ACTION_ATTR_SET:
2499                         err = set_action_to_attr(a, skb);
2500                         if (err)
2501                                 return err;
2502                         break;
2503
2504                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2505                         err = masked_set_action_to_set_action_attr(a, skb);
2506                         if (err)
2507                                 return err;
2508                         break;
2509
2510                 case OVS_ACTION_ATTR_SAMPLE:
2511                         err = sample_action_to_attr(a, skb);
2512                         if (err)
2513                                 return err;
2514                         break;
2515
2516                 case OVS_ACTION_ATTR_CT:
2517                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2518                         if (err)
2519                                 return err;
2520                         break;
2521
2522                 default:
2523                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2524                                 return -EMSGSIZE;
2525                         break;
2526                 }
2527         }
2528
2529         return 0;
2530 }