net/openvswitch/flow_netlink.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include "flow.h"
  22 #include "datapath.h"
  23 #include <linux/uaccess.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <net/llc_pdu.h>
  29 #include <linux/kernel.h>
  30 #include <linux/jhash.h>
  31 #include <linux/jiffies.h>
  32 #include <linux/llc.h>
  33 #include <linux/module.h>
  34 #include <linux/in.h>
  35 #include <linux/rcupdate.h>
  36 #include <linux/if_arp.h>
  37 #include <linux/ip.h>
  38 #include <linux/ipv6.h>
  39 #include <linux/sctp.h>
  40 #include <linux/tcp.h>
  41 #include <linux/udp.h>
  42 #include <linux/icmp.h>
  43 #include <linux/icmpv6.h>
  44 #include <linux/rculist.h>
  45 #include <net/geneve.h>
  46 #include <net/ip.h>
  47 #include <net/ipv6.h>
  48 #include <net/ndisc.h>
  49 #include <net/mpls.h>
  50 #include <net/vxlan.h>
  51
  52 #include "flow_netlink.h"
  53
  54 struct ovs_len_tbl {
  55         int len;
  56         const struct ovs_len_tbl *next;
  57 };
  58
  59 #define OVS_ATTR_NESTED -1
  60 #define OVS_ATTR_VARIABLE -2
  61
  62 static void update_range(struct sw_flow_match *match,
  63                          size_t offset, size_t size, bool is_mask)
  64 {
  65         struct sw_flow_key_range *range;
  66         size_t start = rounddown(offset, sizeof(long));
  67         size_t end = roundup(offset + size, sizeof(long));
  68
  69         if (!is_mask)
  70                 range = &match->range;
  71         else
  72                 range = &match->mask->range;
  73
  74         if (range->start == range->end) {
  75                 range->start = start;
  76                 range->end = end;
  77                 return;
  78         }
  79
  80         if (range->start > start)
  81                 range->start = start;
  82
  83         if (range->end < end)
  84                 range->end = end;
  85 }
  86
  87 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  88         do { \
  89                 update_range(match, offsetof(struct sw_flow_key, field),    \
  90                              sizeof((match)->key->field), is_mask);         \
  91                 if (is_mask)                                                \
  92                         (match)->mask->key.field = value;                   \
  93                 else                                                        \
  94                         (match)->key->field = value;                        \
  95         } while (0)
  96
  97 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
  98         do {                                                                \
  99                 update_range(match, offset, len, is_mask);                  \
 100                 if (is_mask)                                                \
 101                         memcpy((u8 *)&(match)->mask->key + offset, value_p, \
 102                                len);                                       \
 103                 else                                                        \
 104                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
 105         } while (0)
 106
 107 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
 108         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 109                                   value_p, len, is_mask)
 110
 111 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 112         do {                                                                \
 113                 update_range(match, offsetof(struct sw_flow_key, field),    \
 114                              sizeof((match)->key->field), is_mask);         \
 115                 if (is_mask)                                                \
 116                         memset((u8 *)&(match)->mask->key.field, value,      \
 117                                sizeof((match)->mask->key.field));           \
 118                 else                                                        \
 119                         memset((u8 *)&(match)->key->field, value,           \
 120                                sizeof((match)->key->field));                \
 121         } while (0)
 122
 123 static bool match_validate(const struct sw_flow_match *match,
 124                            u64 key_attrs, u64 mask_attrs, bool log)
 125 {
 126         u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
 127         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 128
 129         /* The following mask attributes allowed only if they
 130          * pass the validation tests. */
 131         mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
 132                         | (1 << OVS_KEY_ATTR_IPV6)
 133                         | (1 << OVS_KEY_ATTR_TCP)
 134                         | (1 << OVS_KEY_ATTR_TCP_FLAGS)
 135                         | (1 << OVS_KEY_ATTR_UDP)
 136                         | (1 << OVS_KEY_ATTR_SCTP)
 137                         | (1 << OVS_KEY_ATTR_ICMP)
 138                         | (1 << OVS_KEY_ATTR_ICMPV6)
 139                         | (1 << OVS_KEY_ATTR_ARP)
 140                         | (1 << OVS_KEY_ATTR_ND)
 141                         | (1 << OVS_KEY_ATTR_MPLS));
 142
 143         /* Always allowed mask fields. */
 144         mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
 145                        | (1 << OVS_KEY_ATTR_IN_PORT)
 146                        | (1 << OVS_KEY_ATTR_ETHERTYPE));
 147
 148         /* Check key attributes. */
 149         if (match->key->eth.type == htons(ETH_P_ARP)
 150                         || match->key->eth.type == htons(ETH_P_RARP)) {
 151                 key_expected |= 1 << OVS_KEY_ATTR_ARP;
 152                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 153                         mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
 154         }
 155
 156         if (eth_p_mpls(match->key->eth.type)) {
 157                 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
 158                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 159                         mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
 160         }
 161
 162         if (match->key->eth.type == htons(ETH_P_IP)) {
 163                 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
 164                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 165                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
 166
 167                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 168                         if (match->key->ip.proto == IPPROTO_UDP) {
 169                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 170                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 171                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 172                         }
 173
 174                         if (match->key->ip.proto == IPPROTO_SCTP) {
 175                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 176                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 177                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 178                         }
 179
 180                         if (match->key->ip.proto == IPPROTO_TCP) {
 181                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 182                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 183                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 184                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 185                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 186                                 }
 187                         }
 188
 189                         if (match->key->ip.proto == IPPROTO_ICMP) {
 190                                 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
 191                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 192                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
 193                         }
 194                 }
 195         }
 196
 197         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 198                 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
 199                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 200                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
 201
 202                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 203                         if (match->key->ip.proto == IPPROTO_UDP) {
 204                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 205                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 206                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 207                         }
 208
 209                         if (match->key->ip.proto == IPPROTO_SCTP) {
 210                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 211                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 212                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 213                         }
 214
 215                         if (match->key->ip.proto == IPPROTO_TCP) {
 216                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 217                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 218                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 219                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 220                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 221                                 }
 222                         }
 223
 224                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 225                                 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
 226                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 227                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
 228
 229                                 if (match->key->tp.src ==
 230                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 231                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 232                                         key_expected |= 1 << OVS_KEY_ATTR_ND;
 233                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 234                                                 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
 235                                 }
 236                         }
 237                 }
 238         }
 239
 240         if ((key_attrs & key_expected) != key_expected) {
 241                 /* Key attributes check failed. */
 242                 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
 243                           (unsigned long long)key_attrs,
 244                           (unsigned long long)key_expected);
 245                 return false;
 246         }
 247
 248         if ((mask_attrs & mask_allowed) != mask_attrs) {
 249                 /* Mask attributes check failed. */
 250                 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
 251                           (unsigned long long)mask_attrs,
 252                           (unsigned long long)mask_allowed);
 253                 return false;
 254         }
 255
 256         return true;
 257 }
 258
 259 size_t ovs_tun_key_attr_size(void)
 260 {
 261         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 262          * updating this function.
 263          */
 264         return    nla_total_size(8)    /* OVS_TUNNEL_KEY_ATTR_ID */
 265                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 266                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_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 };
 327
 328 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 329 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 330         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 331         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 332         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 333         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 334         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 335         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 336         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 337         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 338         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 339         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 340         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 341         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 342         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 343         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 344         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 345         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 346         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 347         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 348         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 349         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 350                                      .next = ovs_tunnel_key_lens, },
 351         [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 352         [OVS_KEY_ATTR_CT_STATE]  = { .len = sizeof(u32) },
 353         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 354         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 355         [OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
 356 };
 357
 358 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
 359 {
 360         return expected_len == attr_len ||
 361                expected_len == OVS_ATTR_NESTED ||
 362                expected_len == OVS_ATTR_VARIABLE;
 363 }
 364
 365 static bool is_all_zero(const u8 *fp, size_t size)
 366 {
 367         int i;
 368
 369         if (!fp)
 370                 return false;
 371
 372         for (i = 0; i < size; i++)
 373                 if (fp[i])
 374                         return false;
 375
 376         return true;
 377 }
 378
 379 static int __parse_flow_nlattrs(const struct nlattr *attr,
 380                                 const struct nlattr *a[],
 381                                 u64 *attrsp, bool log, bool nz)
 382 {
 383         const struct nlattr *nla;
 384         u64 attrs;
 385         int rem;
 386
 387         attrs = *attrsp;
 388         nla_for_each_nested(nla, attr, rem) {
 389                 u16 type = nla_type(nla);
 390                 int expected_len;
 391
 392                 if (type > OVS_KEY_ATTR_MAX) {
 393                         OVS_NLERR(log, "Key type %d is out of range max %d",
 394                                   type, OVS_KEY_ATTR_MAX);
 395                         return -EINVAL;
 396                 }
 397
 398                 if (attrs & (1 << type)) {
 399                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 400                         return -EINVAL;
 401                 }
 402
 403                 expected_len = ovs_key_lens[type].len;
 404                 if (!check_attr_len(nla_len(nla), expected_len)) {
 405                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 406                                   type, nla_len(nla), expected_len);
 407                         return -EINVAL;
 408                 }
 409
 410                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 411                         attrs |= 1 << type;
 412                         a[type] = nla;
 413                 }
 414         }
 415         if (rem) {
 416                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 417                 return -EINVAL;
 418         }
 419
 420         *attrsp = attrs;
 421         return 0;
 422 }
 423
 424 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 425                                    const struct nlattr *a[], u64 *attrsp,
 426                                    bool log)
 427 {
 428         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 429 }
 430
 431 static int parse_flow_nlattrs(const struct nlattr *attr,
 432                               const struct nlattr *a[], u64 *attrsp,
 433                               bool log)
 434 {
 435         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 436 }
 437
 438 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 439                                      struct sw_flow_match *match, bool is_mask,
 440                                      bool log)
 441 {
 442         unsigned long opt_key_offset;
 443
 444         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 445                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 446                           nla_len(a), sizeof(match->key->tun_opts));
 447                 return -EINVAL;
 448         }
 449
 450         if (nla_len(a) % 4 != 0) {
 451                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 452                           nla_len(a));
 453                 return -EINVAL;
 454         }
 455
 456         /* We need to record the length of the options passed
 457          * down, otherwise packets with the same format but
 458          * additional options will be silently matched.
 459          */
 460         if (!is_mask) {
 461                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 462                                 false);
 463         } else {
 464                 /* This is somewhat unusual because it looks at
 465                  * both the key and mask while parsing the
 466                  * attributes (and by extension assumes the key
 467                  * is parsed first). Normally, we would verify
 468                  * that each is the correct length and that the
 469                  * attributes line up in the validate function.
 470                  * However, that is difficult because this is
 471                  * variable length and we won't have the
 472                  * information later.
 473                  */
 474                 if (match->key->tun_opts_len != nla_len(a)) {
 475                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 476                                   match->key->tun_opts_len, nla_len(a));
 477                         return -EINVAL;
 478                 }
 479
 480                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 481         }
 482
 483         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 484         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 485                                   nla_len(a), is_mask);
 486         return 0;
 487 }
 488
 489 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
 490                                      struct sw_flow_match *match, bool is_mask,
 491                                      bool log)
 492 {
 493         struct nlattr *a;
 494         int rem;
 495         unsigned long opt_key_offset;
 496         struct vxlan_metadata opts;
 497
 498         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 499
 500         memset(&opts, 0, sizeof(opts));
 501         nla_for_each_nested(a, attr, rem) {
 502                 int type = nla_type(a);
 503
 504                 if (type > OVS_VXLAN_EXT_MAX) {
 505                         OVS_NLERR(log, "VXLAN extension %d out of range max %d",
 506                                   type, OVS_VXLAN_EXT_MAX);
 507                         return -EINVAL;
 508                 }
 509
 510                 if (!check_attr_len(nla_len(a),
 511                                     ovs_vxlan_ext_key_lens[type].len)) {
 512                         OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
 513                                   type, nla_len(a),
 514                                   ovs_vxlan_ext_key_lens[type].len);
 515                         return -EINVAL;
 516                 }
 517
 518                 switch (type) {
 519                 case OVS_VXLAN_EXT_GBP:
 520                         opts.gbp = nla_get_u32(a);
 521                         break;
 522                 default:
 523                         OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
 524                                   type);
 525                         return -EINVAL;
 526                 }
 527         }
 528         if (rem) {
 529                 OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
 530                           rem);
 531                 return -EINVAL;
 532         }
 533
 534         if (!is_mask)
 535                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 536         else
 537                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 538
 539         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 540         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 541                                   is_mask);
 542         return 0;
 543 }
 544
 545 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
 546                                 struct sw_flow_match *match, bool is_mask,
 547                                 bool log)
 548 {
 549         struct nlattr *a;
 550         int rem;
 551         bool ttl = false;
 552         __be16 tun_flags = 0;
 553         int opts_type = 0;
 554
 555         nla_for_each_nested(a, attr, rem) {
 556                 int type = nla_type(a);
 557                 int err;
 558
 559                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 560                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 561                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 562                         return -EINVAL;
 563                 }
 564
 565                 if (!check_attr_len(nla_len(a),
 566                                     ovs_tunnel_key_lens[type].len)) {
 567                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 568                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 569                         return -EINVAL;
 570                 }
 571
 572                 switch (type) {
 573                 case OVS_TUNNEL_KEY_ATTR_ID:
 574                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 575                                         nla_get_be64(a), is_mask);
 576                         tun_flags |= TUNNEL_KEY;
 577                         break;
 578                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 579                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
 580                                         nla_get_in_addr(a), is_mask);
 581                         break;
 582                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 583                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
 584                                         nla_get_in_addr(a), is_mask);
 585                         break;
 586                 case OVS_TUNNEL_KEY_ATTR_TOS:
 587                         SW_FLOW_KEY_PUT(match, tun_key.tos,
 588                                         nla_get_u8(a), is_mask);
 589                         break;
 590                 case OVS_TUNNEL_KEY_ATTR_TTL:
 591                         SW_FLOW_KEY_PUT(match, tun_key.ttl,
 592                                         nla_get_u8(a), is_mask);
 593                         ttl = true;
 594                         break;
 595                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 596                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 597                         break;
 598                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 599                         tun_flags |= TUNNEL_CSUM;
 600                         break;
 601                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 602                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 603                                         nla_get_be16(a), is_mask);
 604                         break;
 605                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 606                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 607                                         nla_get_be16(a), is_mask);
 608                         break;
 609                 case OVS_TUNNEL_KEY_ATTR_OAM:
 610                         tun_flags |= TUNNEL_OAM;
 611                         break;
 612                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 613                         if (opts_type) {
 614                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 615                                 return -EINVAL;
 616                         }
 617
 618                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 619                         if (err)
 620                                 return err;
 621
 622                         tun_flags |= TUNNEL_GENEVE_OPT;
 623                         opts_type = type;
 624                         break;
 625                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 626                         if (opts_type) {
 627                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 628                                 return -EINVAL;
 629                         }
 630
 631                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 632                         if (err)
 633                                 return err;
 634
 635                         tun_flags |= TUNNEL_VXLAN_OPT;
 636                         opts_type = type;
 637                         break;
 638                 default:
 639                         OVS_NLERR(log, "Unknown IPv4 tunnel attribute %d",
 640                                   type);
 641                         return -EINVAL;
 642                 }
 643         }
 644
 645         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 646
 647         if (rem > 0) {
 648                 OVS_NLERR(log, "IPv4 tunnel attribute has %d unknown bytes.",
 649                           rem);
 650                 return -EINVAL;
 651         }
 652
 653         if (!is_mask) {
 654                 if (!match->key->tun_key.u.ipv4.dst) {
 655                         OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 656                         return -EINVAL;
 657                 }
 658
 659                 if (!ttl) {
 660                         OVS_NLERR(log, "IPv4 tunnel TTL not specified.");
 661                         return -EINVAL;
 662                 }
 663         }
 664
 665         return opts_type;
 666 }
 667
 668 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 669                                const void *tun_opts, int swkey_tun_opts_len)
 670 {
 671         const struct vxlan_metadata *opts = tun_opts;
 672         struct nlattr *nla;
 673
 674         nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 675         if (!nla)
 676                 return -EMSGSIZE;
 677
 678         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 679                 return -EMSGSIZE;
 680
 681         nla_nest_end(skb, nla);
 682         return 0;
 683 }
 684
 685 static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
 686                                 const struct ip_tunnel_key *output,
 687                                 const void *tun_opts, int swkey_tun_opts_len)
 688 {
 689         if (output->tun_flags & TUNNEL_KEY &&
 690             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 691                 return -EMSGSIZE;
 692         if (output->u.ipv4.src &&
 693             nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 694                             output->u.ipv4.src))
 695                 return -EMSGSIZE;
 696         if (output->u.ipv4.dst &&
 697             nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 698                             output->u.ipv4.dst))
 699                 return -EMSGSIZE;
 700         if (output->tos &&
 701             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 702                 return -EMSGSIZE;
 703         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 704                 return -EMSGSIZE;
 705         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 706             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 707                 return -EMSGSIZE;
 708         if ((output->tun_flags & TUNNEL_CSUM) &&
 709             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 710                 return -EMSGSIZE;
 711         if (output->tp_src &&
 712             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 713                 return -EMSGSIZE;
 714         if (output->tp_dst &&
 715             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 716                 return -EMSGSIZE;
 717         if ((output->tun_flags & TUNNEL_OAM) &&
 718             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 719                 return -EMSGSIZE;
 720         if (tun_opts) {
 721                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 722                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 723                             swkey_tun_opts_len, tun_opts))
 724                         return -EMSGSIZE;
 725                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 726                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 727                         return -EMSGSIZE;
 728         }
 729
 730         return 0;
 731 }
 732
 733 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
 734                               const struct ip_tunnel_key *output,
 735                               const void *tun_opts, int swkey_tun_opts_len)
 736 {
 737         struct nlattr *nla;
 738         int err;
 739
 740         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 741         if (!nla)
 742                 return -EMSGSIZE;
 743
 744         err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
 745         if (err)
 746                 return err;
 747
 748         nla_nest_end(skb, nla);
 749         return 0;
 750 }
 751
 752 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
 753                                   const struct ip_tunnel_info *egress_tun_info,
 754                                   const void *egress_tun_opts)
 755 {
 756         return __ipv4_tun_to_nlattr(skb, &egress_tun_info->key,
 757                                     egress_tun_opts,
 758                                     egress_tun_info->options_len);
 759 }
 760
 761 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
 762                                  u64 *attrs, const struct nlattr **a,
 763                                  bool is_mask, bool log)
 764 {
 765         if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
 766                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 767
 768                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 769                 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
 770         }
 771
 772         if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
 773                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 774
 775                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 776                 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
 777         }
 778
 779         if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 780                 SW_FLOW_KEY_PUT(match, phy.priority,
 781                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 782                 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 783         }
 784
 785         if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 786                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 787
 788                 if (is_mask) {
 789                         in_port = 0xffffffff; /* Always exact match in_port. */
 790                 } else if (in_port >= DP_MAX_PORTS) {
 791                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 792                                   in_port, DP_MAX_PORTS);
 793                         return -EINVAL;
 794                 }
 795
 796                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 797                 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
 798         } else if (!is_mask) {
 799                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 800         }
 801
 802         if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
 803                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 804
 805                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 806                 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
 807         }
 808         if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
 809                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 810                                          is_mask, log) < 0)
 811                         return -EINVAL;
 812                 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
 813         }
 814
 815         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
 816             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
 817                 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
 818
 819                 if (ct_state & ~CT_SUPPORTED_MASK) {
 820                         OVS_NLERR(log, "ct_state flags %08x unsupported",
 821                                   ct_state);
 822                         return -EINVAL;
 823                 }
 824
 825                 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
 826                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
 827         }
 828         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
 829             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
 830                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
 831
 832                 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
 833                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
 834         }
 835         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
 836             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
 837                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
 838
 839                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
 840                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
 841         }
 842         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
 843             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
 844                 const struct ovs_key_ct_labels *cl;
 845
 846                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
 847                 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
 848                                    sizeof(*cl), is_mask);
 849                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
 850         }
 851         return 0;
 852 }
 853
 854 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
 855                                 u64 attrs, const struct nlattr **a,
 856                                 bool is_mask, bool log)
 857 {
 858         int err;
 859
 860         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
 861         if (err)
 862                 return err;
 863
 864         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
 865                 const struct ovs_key_ethernet *eth_key;
 866
 867                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 868                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 869                                 eth_key->eth_src, ETH_ALEN, is_mask);
 870                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 871                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 872                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
 873         }
 874
 875         if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
 876                 __be16 tci;
 877
 878                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 879                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 880                         if (is_mask)
 881                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 882                         else
 883                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 884
 885                         return -EINVAL;
 886                 }
 887
 888                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 889                 attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
 890         }
 891
 892         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
 893                 __be16 eth_type;
 894
 895                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 896                 if (is_mask) {
 897                         /* Always exact match EtherType. */
 898                         eth_type = htons(0xffff);
 899                 } else if (!eth_proto_is_802_3(eth_type)) {
 900                         OVS_NLERR(log, "EtherType %x is less than min %x",
 901                                   ntohs(eth_type), ETH_P_802_3_MIN);
 902                         return -EINVAL;
 903                 }
 904
 905                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 906                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
 907         } else if (!is_mask) {
 908                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 909         }
 910
 911         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
 912                 const struct ovs_key_ipv4 *ipv4_key;
 913
 914                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 915                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 916                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 917                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 918                         return -EINVAL;
 919                 }
 920                 SW_FLOW_KEY_PUT(match, ip.proto,
 921                                 ipv4_key->ipv4_proto, is_mask);
 922                 SW_FLOW_KEY_PUT(match, ip.tos,
 923                                 ipv4_key->ipv4_tos, is_mask);
 924                 SW_FLOW_KEY_PUT(match, ip.ttl,
 925                                 ipv4_key->ipv4_ttl, is_mask);
 926                 SW_FLOW_KEY_PUT(match, ip.frag,
 927                                 ipv4_key->ipv4_frag, is_mask);
 928                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 929                                 ipv4_key->ipv4_src, is_mask);
 930                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 931                                 ipv4_key->ipv4_dst, is_mask);
 932                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
 933         }
 934
 935         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
 936                 const struct ovs_key_ipv6 *ipv6_key;
 937
 938                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 939                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 940                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 941                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 942                         return -EINVAL;
 943                 }
 944
 945                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 946                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
 947                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 948                         return -EINVAL;
 949                 }
 950
 951                 SW_FLOW_KEY_PUT(match, ipv6.label,
 952                                 ipv6_key->ipv6_label, is_mask);
 953                 SW_FLOW_KEY_PUT(match, ip.proto,
 954                                 ipv6_key->ipv6_proto, is_mask);
 955                 SW_FLOW_KEY_PUT(match, ip.tos,
 956                                 ipv6_key->ipv6_tclass, is_mask);
 957                 SW_FLOW_KEY_PUT(match, ip.ttl,
 958                                 ipv6_key->ipv6_hlimit, is_mask);
 959                 SW_FLOW_KEY_PUT(match, ip.frag,
 960                                 ipv6_key->ipv6_frag, is_mask);
 961                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
 962                                 ipv6_key->ipv6_src,
 963                                 sizeof(match->key->ipv6.addr.src),
 964                                 is_mask);
 965                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
 966                                 ipv6_key->ipv6_dst,
 967                                 sizeof(match->key->ipv6.addr.dst),
 968                                 is_mask);
 969
 970                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
 971         }
 972
 973         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
 974                 const struct ovs_key_arp *arp_key;
 975
 976                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
 977                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
 978                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
 979                                   arp_key->arp_op);
 980                         return -EINVAL;
 981                 }
 982
 983                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 984                                 arp_key->arp_sip, is_mask);
 985                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 986                         arp_key->arp_tip, is_mask);
 987                 SW_FLOW_KEY_PUT(match, ip.proto,
 988                                 ntohs(arp_key->arp_op), is_mask);
 989                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
 990                                 arp_key->arp_sha, ETH_ALEN, is_mask);
 991                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
 992                                 arp_key->arp_tha, ETH_ALEN, is_mask);
 993
 994                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
 995         }
 996
 997         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
 998                 const struct ovs_key_mpls *mpls_key;
 999
1000                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1001                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1002                                 mpls_key->mpls_lse, is_mask);
1003
1004                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1005          }
1006
1007         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1008                 const struct ovs_key_tcp *tcp_key;
1009
1010                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1011                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1012                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1013                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1014         }
1015
1016         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1017                 SW_FLOW_KEY_PUT(match, tp.flags,
1018                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1019                                 is_mask);
1020                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1021         }
1022
1023         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1024                 const struct ovs_key_udp *udp_key;
1025
1026                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1027                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1028                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1029                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1030         }
1031
1032         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1033                 const struct ovs_key_sctp *sctp_key;
1034
1035                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1036                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1037                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1038                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1039         }
1040
1041         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1042                 const struct ovs_key_icmp *icmp_key;
1043
1044                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1045                 SW_FLOW_KEY_PUT(match, tp.src,
1046                                 htons(icmp_key->icmp_type), is_mask);
1047                 SW_FLOW_KEY_PUT(match, tp.dst,
1048                                 htons(icmp_key->icmp_code), is_mask);
1049                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1050         }
1051
1052         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1053                 const struct ovs_key_icmpv6 *icmpv6_key;
1054
1055                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1056                 SW_FLOW_KEY_PUT(match, tp.src,
1057                                 htons(icmpv6_key->icmpv6_type), is_mask);
1058                 SW_FLOW_KEY_PUT(match, tp.dst,
1059                                 htons(icmpv6_key->icmpv6_code), is_mask);
1060                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1061         }
1062
1063         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1064                 const struct ovs_key_nd *nd_key;
1065
1066                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1067                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1068                         nd_key->nd_target,
1069                         sizeof(match->key->ipv6.nd.target),
1070                         is_mask);
1071                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1072                         nd_key->nd_sll, ETH_ALEN, is_mask);
1073                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1074                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1075                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1076         }
1077
1078         if (attrs != 0) {
1079                 OVS_NLERR(log, "Unknown key attributes %llx",
1080                           (unsigned long long)attrs);
1081                 return -EINVAL;
1082         }
1083
1084         return 0;
1085 }
1086
1087 static void nlattr_set(struct nlattr *attr, u8 val,
1088                        const struct ovs_len_tbl *tbl)
1089 {
1090         struct nlattr *nla;
1091         int rem;
1092
1093         /* The nlattr stream should already have been validated */
1094         nla_for_each_nested(nla, attr, rem) {
1095                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1096                         if (tbl[nla_type(nla)].next)
1097                                 tbl = tbl[nla_type(nla)].next;
1098                         nlattr_set(nla, val, tbl);
1099                 } else {
1100                         memset(nla_data(nla), val, nla_len(nla));
1101                 }
1102
1103                 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1104                         *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1105         }
1106 }
1107
1108 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1109 {
1110         nlattr_set(attr, val, ovs_key_lens);
1111 }
1112
1113 /**
1114  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1115  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1116  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1117  * does not include any don't care bit.
1118  * @net: Used to determine per-namespace field support.
1119  * @match: receives the extracted flow match information.
1120  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1121  * sequence. The fields should of the packet that triggered the creation
1122  * of this flow.
1123  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1124  * attribute specifies the mask field of the wildcarded flow.
1125  * @log: Boolean to allow kernel error logging.  Normally true, but when
1126  * probing for feature compatibility this should be passed in as false to
1127  * suppress unnecessary error logging.
1128  */
1129 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1130                       const struct nlattr *nla_key,
1131                       const struct nlattr *nla_mask,
1132                       bool log)
1133 {
1134         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1135         const struct nlattr *encap;
1136         struct nlattr *newmask = NULL;
1137         u64 key_attrs = 0;
1138         u64 mask_attrs = 0;
1139         bool encap_valid = false;
1140         int err;
1141
1142         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1143         if (err)
1144                 return err;
1145
1146         if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1147             (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1148             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1149                 __be16 tci;
1150
1151                 if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1152                       (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1153                         OVS_NLERR(log, "Invalid Vlan frame.");
1154                         return -EINVAL;
1155                 }
1156
1157                 key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1158                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1159                 encap = a[OVS_KEY_ATTR_ENCAP];
1160                 key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1161                 encap_valid = true;
1162
1163                 if (tci & htons(VLAN_TAG_PRESENT)) {
1164                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1165                         if (err)
1166                                 return err;
1167                 } else if (!tci) {
1168                         /* Corner case for truncated 802.1Q header. */
1169                         if (nla_len(encap)) {
1170                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1171                                 return -EINVAL;
1172                         }
1173                 } else {
1174                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1175                         return  -EINVAL;
1176                 }
1177         }
1178
1179         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1180         if (err)
1181                 return err;
1182
1183         if (match->mask) {
1184                 if (!nla_mask) {
1185                         /* Create an exact match mask. We need to set to 0xff
1186                          * all the 'match->mask' fields that have been touched
1187                          * in 'match->key'. We cannot simply memset
1188                          * 'match->mask', because padding bytes and fields not
1189                          * specified in 'match->key' should be left to 0.
1190                          * Instead, we use a stream of netlink attributes,
1191                          * copied from 'key' and set to 0xff.
1192                          * ovs_key_from_nlattrs() will take care of filling
1193                          * 'match->mask' appropriately.
1194                          */
1195                         newmask = kmemdup(nla_key,
1196                                           nla_total_size(nla_len(nla_key)),
1197                                           GFP_KERNEL);
1198                         if (!newmask)
1199                                 return -ENOMEM;
1200
1201                         mask_set_nlattr(newmask, 0xff);
1202
1203                         /* The userspace does not send tunnel attributes that
1204                          * are 0, but we should not wildcard them nonetheless.
1205                          */
1206                         if (match->key->tun_key.u.ipv4.dst)
1207                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1208                                                          0xff, true);
1209
1210                         nla_mask = newmask;
1211                 }
1212
1213                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1214                 if (err)
1215                         goto free_newmask;
1216
1217                 /* Always match on tci. */
1218                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1219
1220                 if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
1221                         __be16 eth_type = 0;
1222                         __be16 tci = 0;
1223
1224                         if (!encap_valid) {
1225                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1226                                 err = -EINVAL;
1227                                 goto free_newmask;
1228                         }
1229
1230                         mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1231                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1232                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1233
1234                         if (eth_type == htons(0xffff)) {
1235                                 mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1236                                 encap = a[OVS_KEY_ATTR_ENCAP];
1237                                 err = parse_flow_mask_nlattrs(encap, a,
1238                                                               &mask_attrs, log);
1239                                 if (err)
1240                                         goto free_newmask;
1241                         } else {
1242                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1243                                           ntohs(eth_type));
1244                                 err = -EINVAL;
1245                                 goto free_newmask;
1246                         }
1247
1248                         if (a[OVS_KEY_ATTR_VLAN])
1249                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1250
1251                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1252                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1253                                           ntohs(tci));
1254                                 err = -EINVAL;
1255                                 goto free_newmask;
1256                         }
1257                 }
1258
1259                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1260                                            log);
1261                 if (err)
1262                         goto free_newmask;
1263         }
1264
1265         if (!match_validate(match, key_attrs, mask_attrs, log))
1266                 err = -EINVAL;
1267
1268 free_newmask:
1269         kfree(newmask);
1270         return err;
1271 }
1272
1273 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1274 {
1275         size_t len;
1276
1277         if (!attr)
1278                 return 0;
1279
1280         len = nla_len(attr);
1281         if (len < 1 || len > MAX_UFID_LENGTH) {
1282                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1283                           nla_len(attr), MAX_UFID_LENGTH);
1284                 return 0;
1285         }
1286
1287         return len;
1288 }
1289
1290 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1291  * or false otherwise.
1292  */
1293 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1294                       bool log)
1295 {
1296         sfid->ufid_len = get_ufid_len(attr, log);
1297         if (sfid->ufid_len)
1298                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1299
1300         return sfid->ufid_len;
1301 }
1302
1303 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1304                            const struct sw_flow_key *key, bool log)
1305 {
1306         struct sw_flow_key *new_key;
1307
1308         if (ovs_nla_get_ufid(sfid, ufid, log))
1309                 return 0;
1310
1311         /* If UFID was not provided, use unmasked key. */
1312         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1313         if (!new_key)
1314                 return -ENOMEM;
1315         memcpy(new_key, key, sizeof(*key));
1316         sfid->unmasked_key = new_key;
1317
1318         return 0;
1319 }
1320
1321 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1322 {
1323         return attr ? nla_get_u32(attr) : 0;
1324 }
1325
1326 /**
1327  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1328  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1329  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1330  * sequence.
1331  * @log: Boolean to allow kernel error logging.  Normally true, but when
1332  * probing for feature compatibility this should be passed in as false to
1333  * suppress unnecessary error logging.
1334  *
1335  * This parses a series of Netlink attributes that form a flow key, which must
1336  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1337  * get the metadata, that is, the parts of the flow key that cannot be
1338  * extracted from the packet itself.
1339  */
1340
1341 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1342                               struct sw_flow_key *key,
1343                               bool log)
1344 {
1345         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1346         struct sw_flow_match match;
1347         u64 attrs = 0;
1348         int err;
1349
1350         err = parse_flow_nlattrs(attr, a, &attrs, log);
1351         if (err)
1352                 return -EINVAL;
1353
1354         memset(&match, 0, sizeof(match));
1355         match.key = key;
1356
1357         memset(&key->ct, 0, sizeof(key->ct));
1358         key->phy.in_port = DP_MAX_PORTS;
1359
1360         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1361 }
1362
1363 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1364                              const struct sw_flow_key *output, bool is_mask,
1365                              struct sk_buff *skb)
1366 {
1367         struct ovs_key_ethernet *eth_key;
1368         struct nlattr *nla, *encap;
1369
1370         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1371                 goto nla_put_failure;
1372
1373         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1374                 goto nla_put_failure;
1375
1376         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1377                 goto nla_put_failure;
1378
1379         if ((swkey->tun_key.u.ipv4.dst || is_mask)) {
1380                 const void *opts = NULL;
1381
1382                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1383                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1384
1385                 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1386                                        swkey->tun_opts_len))
1387                         goto nla_put_failure;
1388         }
1389
1390         if (swkey->phy.in_port == DP_MAX_PORTS) {
1391                 if (is_mask && (output->phy.in_port == 0xffff))
1392                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1393                                 goto nla_put_failure;
1394         } else {
1395                 u16 upper_u16;
1396                 upper_u16 = !is_mask ? 0 : 0xffff;
1397
1398                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1399                                 (upper_u16 << 16) | output->phy.in_port))
1400                         goto nla_put_failure;
1401         }
1402
1403         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1404                 goto nla_put_failure;
1405
1406         if (ovs_ct_put_key(output, skb))
1407                 goto nla_put_failure;
1408
1409         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1410         if (!nla)
1411                 goto nla_put_failure;
1412
1413         eth_key = nla_data(nla);
1414         ether_addr_copy(eth_key->eth_src, output->eth.src);
1415         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1416
1417         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1418                 __be16 eth_type;
1419                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1420                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1421                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1422                         goto nla_put_failure;
1423                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1424                 if (!swkey->eth.tci)
1425                         goto unencap;
1426         } else
1427                 encap = NULL;
1428
1429         if (swkey->eth.type == htons(ETH_P_802_2)) {
1430                 /*
1431                  * Ethertype 802.2 is represented in the netlink with omitted
1432                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1433                  * 0xffff in the mask attribute.  Ethertype can also
1434                  * be wildcarded.
1435                  */
1436                 if (is_mask && output->eth.type)
1437                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1438                                                 output->eth.type))
1439                                 goto nla_put_failure;
1440                 goto unencap;
1441         }
1442
1443         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1444                 goto nla_put_failure;
1445
1446         if (swkey->eth.type == htons(ETH_P_IP)) {
1447                 struct ovs_key_ipv4 *ipv4_key;
1448
1449                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1450                 if (!nla)
1451                         goto nla_put_failure;
1452                 ipv4_key = nla_data(nla);
1453                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1454                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1455                 ipv4_key->ipv4_proto = output->ip.proto;
1456                 ipv4_key->ipv4_tos = output->ip.tos;
1457                 ipv4_key->ipv4_ttl = output->ip.ttl;
1458                 ipv4_key->ipv4_frag = output->ip.frag;
1459         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1460                 struct ovs_key_ipv6 *ipv6_key;
1461
1462                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1463                 if (!nla)
1464                         goto nla_put_failure;
1465                 ipv6_key = nla_data(nla);
1466                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1467                                 sizeof(ipv6_key->ipv6_src));
1468                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1469                                 sizeof(ipv6_key->ipv6_dst));
1470                 ipv6_key->ipv6_label = output->ipv6.label;
1471                 ipv6_key->ipv6_proto = output->ip.proto;
1472                 ipv6_key->ipv6_tclass = output->ip.tos;
1473                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1474                 ipv6_key->ipv6_frag = output->ip.frag;
1475         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1476                    swkey->eth.type == htons(ETH_P_RARP)) {
1477                 struct ovs_key_arp *arp_key;
1478
1479                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1480                 if (!nla)
1481                         goto nla_put_failure;
1482                 arp_key = nla_data(nla);
1483                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1484                 arp_key->arp_sip = output->ipv4.addr.src;
1485                 arp_key->arp_tip = output->ipv4.addr.dst;
1486                 arp_key->arp_op = htons(output->ip.proto);
1487                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1488                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1489         } else if (eth_p_mpls(swkey->eth.type)) {
1490                 struct ovs_key_mpls *mpls_key;
1491
1492                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1493                 if (!nla)
1494                         goto nla_put_failure;
1495                 mpls_key = nla_data(nla);
1496                 mpls_key->mpls_lse = output->mpls.top_lse;
1497         }
1498
1499         if ((swkey->eth.type == htons(ETH_P_IP) ||
1500              swkey->eth.type == htons(ETH_P_IPV6)) &&
1501              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1502
1503                 if (swkey->ip.proto == IPPROTO_TCP) {
1504                         struct ovs_key_tcp *tcp_key;
1505
1506                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1507                         if (!nla)
1508                                 goto nla_put_failure;
1509                         tcp_key = nla_data(nla);
1510                         tcp_key->tcp_src = output->tp.src;
1511                         tcp_key->tcp_dst = output->tp.dst;
1512                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1513                                          output->tp.flags))
1514                                 goto nla_put_failure;
1515                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1516                         struct ovs_key_udp *udp_key;
1517
1518                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1519                         if (!nla)
1520                                 goto nla_put_failure;
1521                         udp_key = nla_data(nla);
1522                         udp_key->udp_src = output->tp.src;
1523                         udp_key->udp_dst = output->tp.dst;
1524                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1525                         struct ovs_key_sctp *sctp_key;
1526
1527                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1528                         if (!nla)
1529                                 goto nla_put_failure;
1530                         sctp_key = nla_data(nla);
1531                         sctp_key->sctp_src = output->tp.src;
1532                         sctp_key->sctp_dst = output->tp.dst;
1533                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1534                            swkey->ip.proto == IPPROTO_ICMP) {
1535                         struct ovs_key_icmp *icmp_key;
1536
1537                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1538                         if (!nla)
1539                                 goto nla_put_failure;
1540                         icmp_key = nla_data(nla);
1541                         icmp_key->icmp_type = ntohs(output->tp.src);
1542                         icmp_key->icmp_code = ntohs(output->tp.dst);
1543                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1544                            swkey->ip.proto == IPPROTO_ICMPV6) {
1545                         struct ovs_key_icmpv6 *icmpv6_key;
1546
1547                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1548                                                 sizeof(*icmpv6_key));
1549                         if (!nla)
1550                                 goto nla_put_failure;
1551                         icmpv6_key = nla_data(nla);
1552                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1553                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1554
1555                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1556                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1557                                 struct ovs_key_nd *nd_key;
1558
1559                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1560                                 if (!nla)
1561                                         goto nla_put_failure;
1562                                 nd_key = nla_data(nla);
1563                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1564                                                         sizeof(nd_key->nd_target));
1565                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1566                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1567                         }
1568                 }
1569         }
1570
1571 unencap:
1572         if (encap)
1573                 nla_nest_end(skb, encap);
1574
1575         return 0;
1576
1577 nla_put_failure:
1578         return -EMSGSIZE;
1579 }
1580
1581 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1582                     const struct sw_flow_key *output, int attr, bool is_mask,
1583                     struct sk_buff *skb)
1584 {
1585         int err;
1586         struct nlattr *nla;
1587
1588         nla = nla_nest_start(skb, attr);
1589         if (!nla)
1590                 return -EMSGSIZE;
1591         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1592         if (err)
1593                 return err;
1594         nla_nest_end(skb, nla);
1595
1596         return 0;
1597 }
1598
1599 /* Called with ovs_mutex or RCU read lock. */
1600 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1601 {
1602         if (ovs_identifier_is_ufid(&flow->id))
1603                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1604                                flow->id.ufid);
1605
1606         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1607                                OVS_FLOW_ATTR_KEY, false, skb);
1608 }
1609
1610 /* Called with ovs_mutex or RCU read lock. */
1611 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1612 {
1613         return ovs_nla_put_key(&flow->key, &flow->key,
1614                                 OVS_FLOW_ATTR_KEY, false, skb);
1615 }
1616
1617 /* Called with ovs_mutex or RCU read lock. */
1618 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1619 {
1620         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1621                                 OVS_FLOW_ATTR_MASK, true, skb);
1622 }
1623
1624 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1625
1626 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1627 {
1628         struct sw_flow_actions *sfa;
1629
1630         if (size > MAX_ACTIONS_BUFSIZE) {
1631                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1632                 return ERR_PTR(-EINVAL);
1633         }
1634
1635         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1636         if (!sfa)
1637                 return ERR_PTR(-ENOMEM);
1638
1639         sfa->actions_len = 0;
1640         return sfa;
1641 }
1642
1643 static void ovs_nla_free_set_action(const struct nlattr *a)
1644 {
1645         const struct nlattr *ovs_key = nla_data(a);
1646         struct ovs_tunnel_info *ovs_tun;
1647
1648         switch (nla_type(ovs_key)) {
1649         case OVS_KEY_ATTR_TUNNEL_INFO:
1650                 ovs_tun = nla_data(ovs_key);
1651                 dst_release((struct dst_entry *)ovs_tun->tun_dst);
1652                 break;
1653         }
1654 }
1655
1656 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1657 {
1658         const struct nlattr *a;
1659         int rem;
1660
1661         if (!sf_acts)
1662                 return;
1663
1664         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1665                 switch (nla_type(a)) {
1666                 case OVS_ACTION_ATTR_SET:
1667                         ovs_nla_free_set_action(a);
1668                         break;
1669                 case OVS_ACTION_ATTR_CT:
1670                         ovs_ct_free_action(a);
1671                         break;
1672                 }
1673         }
1674
1675         kfree(sf_acts);
1676 }
1677
1678 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1679 {
1680         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1681 }
1682
1683 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1684  * The caller must hold rcu_read_lock for this to be sensible. */
1685 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1686 {
1687         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1688 }
1689
1690 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1691                                        int attr_len, bool log)
1692 {
1693
1694         struct sw_flow_actions *acts;
1695         int new_acts_size;
1696         int req_size = NLA_ALIGN(attr_len);
1697         int next_offset = offsetof(struct sw_flow_actions, actions) +
1698                                         (*sfa)->actions_len;
1699
1700         if (req_size <= (ksize(*sfa) - next_offset))
1701                 goto out;
1702
1703         new_acts_size = ksize(*sfa) * 2;
1704
1705         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1706                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1707                         return ERR_PTR(-EMSGSIZE);
1708                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1709         }
1710
1711         acts = nla_alloc_flow_actions(new_acts_size, log);
1712         if (IS_ERR(acts))
1713                 return (void *)acts;
1714
1715         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1716         acts->actions_len = (*sfa)->actions_len;
1717         acts->orig_len = (*sfa)->orig_len;
1718         kfree(*sfa);
1719         *sfa = acts;
1720
1721 out:
1722         (*sfa)->actions_len += req_size;
1723         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1724 }
1725
1726 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1727                                    int attrtype, void *data, int len, bool log)
1728 {
1729         struct nlattr *a;
1730
1731         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1732         if (IS_ERR(a))
1733                 return a;
1734
1735         a->nla_type = attrtype;
1736         a->nla_len = nla_attr_size(len);
1737
1738         if (data)
1739                 memcpy(nla_data(a), data, len);
1740         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1741
1742         return a;
1743 }
1744
1745 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1746                        int len, bool log)
1747 {
1748         struct nlattr *a;
1749
1750         a = __add_action(sfa, attrtype, data, len, log);
1751
1752         return PTR_ERR_OR_ZERO(a);
1753 }
1754
1755 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1756                                           int attrtype, bool log)
1757 {
1758         int used = (*sfa)->actions_len;
1759         int err;
1760
1761         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1762         if (err)
1763                 return err;
1764
1765         return used;
1766 }
1767
1768 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1769                                          int st_offset)
1770 {
1771         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1772                                                                st_offset);
1773
1774         a->nla_len = sfa->actions_len - st_offset;
1775 }
1776
1777 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1778                                   const struct sw_flow_key *key,
1779                                   int depth, struct sw_flow_actions **sfa,
1780                                   __be16 eth_type, __be16 vlan_tci, bool log);
1781
1782 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1783                                     const struct sw_flow_key *key, int depth,
1784                                     struct sw_flow_actions **sfa,
1785                                     __be16 eth_type, __be16 vlan_tci, bool log)
1786 {
1787         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1788         const struct nlattr *probability, *actions;
1789         const struct nlattr *a;
1790         int rem, start, err, st_acts;
1791
1792         memset(attrs, 0, sizeof(attrs));
1793         nla_for_each_nested(a, attr, rem) {
1794                 int type = nla_type(a);
1795                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1796                         return -EINVAL;
1797                 attrs[type] = a;
1798         }
1799         if (rem)
1800                 return -EINVAL;
1801
1802         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1803         if (!probability || nla_len(probability) != sizeof(u32))
1804                 return -EINVAL;
1805
1806         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1807         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1808                 return -EINVAL;
1809
1810         /* validation done, copy sample action. */
1811         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1812         if (start < 0)
1813                 return start;
1814         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1815                                  nla_data(probability), sizeof(u32), log);
1816         if (err)
1817                 return err;
1818         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1819         if (st_acts < 0)
1820                 return st_acts;
1821
1822         err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1823                                      eth_type, vlan_tci, log);
1824         if (err)
1825                 return err;
1826
1827         add_nested_action_end(*sfa, st_acts);
1828         add_nested_action_end(*sfa, start);
1829
1830         return 0;
1831 }
1832
1833 void ovs_match_init(struct sw_flow_match *match,
1834                     struct sw_flow_key *key,
1835                     struct sw_flow_mask *mask)
1836 {
1837         memset(match, 0, sizeof(*match));
1838         match->key = key;
1839         match->mask = mask;
1840
1841         memset(key, 0, sizeof(*key));
1842
1843         if (mask) {
1844                 memset(&mask->key, 0, sizeof(mask->key));
1845                 mask->range.start = mask->range.end = 0;
1846         }
1847 }
1848
1849 static int validate_geneve_opts(struct sw_flow_key *key)
1850 {
1851         struct geneve_opt *option;
1852         int opts_len = key->tun_opts_len;
1853         bool crit_opt = false;
1854
1855         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1856         while (opts_len > 0) {
1857                 int len;
1858
1859                 if (opts_len < sizeof(*option))
1860                         return -EINVAL;
1861
1862                 len = sizeof(*option) + option->length * 4;
1863                 if (len > opts_len)
1864                         return -EINVAL;
1865
1866                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1867
1868                 option = (struct geneve_opt *)((u8 *)option + len);
1869                 opts_len -= len;
1870         };
1871
1872         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1873
1874         return 0;
1875 }
1876
1877 static int validate_and_copy_set_tun(const struct nlattr *attr,
1878                                      struct sw_flow_actions **sfa, bool log)
1879 {
1880         struct sw_flow_match match;
1881         struct sw_flow_key key;
1882         struct metadata_dst *tun_dst;
1883         struct ip_tunnel_info *tun_info;
1884         struct ovs_tunnel_info *ovs_tun;
1885         struct nlattr *a;
1886         int err = 0, start, opts_type;
1887
1888         ovs_match_init(&match, &key, NULL);
1889         opts_type = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1890         if (opts_type < 0)
1891                 return opts_type;
1892
1893         if (key.tun_opts_len) {
1894                 switch (opts_type) {
1895                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1896                         err = validate_geneve_opts(&key);
1897                         if (err < 0)
1898                                 return err;
1899                         break;
1900                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1901                         break;
1902                 }
1903         };
1904
1905         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1906         if (start < 0)
1907                 return start;
1908
1909         tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1910         if (!tun_dst)
1911                 return -ENOMEM;
1912
1913         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1914                          sizeof(*ovs_tun), log);
1915         if (IS_ERR(a)) {
1916                 dst_release((struct dst_entry *)tun_dst);
1917                 return PTR_ERR(a);
1918         }
1919
1920         ovs_tun = nla_data(a);
1921         ovs_tun->tun_dst = tun_dst;
1922
1923         tun_info = &tun_dst->u.tun_info;
1924         tun_info->mode = IP_TUNNEL_INFO_TX;
1925         tun_info->key = key.tun_key;
1926
1927         /* We need to store the options in the action itself since
1928          * everything else will go away after flow setup. We can append
1929          * it to tun_info and then point there.
1930          */
1931         ip_tunnel_info_opts_set(tun_info,
1932                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
1933                                 key.tun_opts_len);
1934         add_nested_action_end(*sfa, start);
1935
1936         return err;
1937 }
1938
1939 /* Return false if there are any non-masked bits set.
1940  * Mask follows data immediately, before any netlink padding.
1941  */
1942 static bool validate_masked(u8 *data, int len)
1943 {
1944         u8 *mask = data + len;
1945
1946         while (len--)
1947                 if (*data++ & ~*mask++)
1948                         return false;
1949
1950         return true;
1951 }
1952
1953 static int validate_set(const struct nlattr *a,
1954                         const struct sw_flow_key *flow_key,
1955                         struct sw_flow_actions **sfa,
1956                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
1957 {
1958         const struct nlattr *ovs_key = nla_data(a);
1959         int key_type = nla_type(ovs_key);
1960         size_t key_len;
1961
1962         /* There can be only one key in a action */
1963         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1964                 return -EINVAL;
1965
1966         key_len = nla_len(ovs_key);
1967         if (masked)
1968                 key_len /= 2;
1969
1970         if (key_type > OVS_KEY_ATTR_MAX ||
1971             !check_attr_len(key_len, ovs_key_lens[key_type].len))
1972                 return -EINVAL;
1973
1974         if (masked && !validate_masked(nla_data(ovs_key), key_len))
1975                 return -EINVAL;
1976
1977         switch (key_type) {
1978         const struct ovs_key_ipv4 *ipv4_key;
1979         const struct ovs_key_ipv6 *ipv6_key;
1980         int err;
1981
1982         case OVS_KEY_ATTR_PRIORITY:
1983         case OVS_KEY_ATTR_SKB_MARK:
1984         case OVS_KEY_ATTR_CT_MARK:
1985         case OVS_KEY_ATTR_CT_LABELS:
1986         case OVS_KEY_ATTR_ETHERNET:
1987                 break;
1988
1989         case OVS_KEY_ATTR_TUNNEL:
1990                 if (eth_p_mpls(eth_type))
1991                         return -EINVAL;
1992
1993                 if (masked)
1994                         return -EINVAL; /* Masked tunnel set not supported. */
1995
1996                 *skip_copy = true;
1997                 err = validate_and_copy_set_tun(a, sfa, log);
1998                 if (err)
1999                         return err;
2000                 break;
2001
2002         case OVS_KEY_ATTR_IPV4:
2003                 if (eth_type != htons(ETH_P_IP))
2004                         return -EINVAL;
2005
2006                 ipv4_key = nla_data(ovs_key);
2007
2008                 if (masked) {
2009                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2010
2011                         /* Non-writeable fields. */
2012                         if (mask->ipv4_proto || mask->ipv4_frag)
2013                                 return -EINVAL;
2014                 } else {
2015                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2016                                 return -EINVAL;
2017
2018                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2019                                 return -EINVAL;
2020                 }
2021                 break;
2022
2023         case OVS_KEY_ATTR_IPV6:
2024                 if (eth_type != htons(ETH_P_IPV6))
2025                         return -EINVAL;
2026
2027                 ipv6_key = nla_data(ovs_key);
2028
2029                 if (masked) {
2030                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2031
2032                         /* Non-writeable fields. */
2033                         if (mask->ipv6_proto || mask->ipv6_frag)
2034                                 return -EINVAL;
2035
2036                         /* Invalid bits in the flow label mask? */
2037                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2038                                 return -EINVAL;
2039                 } else {
2040                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2041                                 return -EINVAL;
2042
2043                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2044                                 return -EINVAL;
2045                 }
2046                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2047                         return -EINVAL;
2048
2049                 break;
2050
2051         case OVS_KEY_ATTR_TCP:
2052                 if ((eth_type != htons(ETH_P_IP) &&
2053                      eth_type != htons(ETH_P_IPV6)) ||
2054                     flow_key->ip.proto != IPPROTO_TCP)
2055                         return -EINVAL;
2056
2057                 break;
2058
2059         case OVS_KEY_ATTR_UDP:
2060                 if ((eth_type != htons(ETH_P_IP) &&
2061                      eth_type != htons(ETH_P_IPV6)) ||
2062                     flow_key->ip.proto != IPPROTO_UDP)
2063                         return -EINVAL;
2064
2065                 break;
2066
2067         case OVS_KEY_ATTR_MPLS:
2068                 if (!eth_p_mpls(eth_type))
2069                         return -EINVAL;
2070                 break;
2071
2072         case OVS_KEY_ATTR_SCTP:
2073                 if ((eth_type != htons(ETH_P_IP) &&
2074                      eth_type != htons(ETH_P_IPV6)) ||
2075                     flow_key->ip.proto != IPPROTO_SCTP)
2076                         return -EINVAL;
2077
2078                 break;
2079
2080         default:
2081                 return -EINVAL;
2082         }
2083
2084         /* Convert non-masked non-tunnel set actions to masked set actions. */
2085         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2086                 int start, len = key_len * 2;
2087                 struct nlattr *at;
2088
2089                 *skip_copy = true;
2090
2091                 start = add_nested_action_start(sfa,
2092                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2093                                                 log);
2094                 if (start < 0)
2095                         return start;
2096
2097                 at = __add_action(sfa, key_type, NULL, len, log);
2098                 if (IS_ERR(at))
2099                         return PTR_ERR(at);
2100
2101                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2102                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2103                 /* Clear non-writeable bits from otherwise writeable fields. */
2104                 if (key_type == OVS_KEY_ATTR_IPV6) {
2105                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2106
2107                         mask->ipv6_label &= htonl(0x000FFFFF);
2108                 }
2109                 add_nested_action_end(*sfa, start);
2110         }
2111
2112         return 0;
2113 }
2114
2115 static int validate_userspace(const struct nlattr *attr)
2116 {
2117         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2118                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2119                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2120                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2121         };
2122         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2123         int error;
2124
2125         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2126                                  attr, userspace_policy);
2127         if (error)
2128                 return error;
2129
2130         if (!a[OVS_USERSPACE_ATTR_PID] ||
2131             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2132                 return -EINVAL;
2133
2134         return 0;
2135 }
2136
2137 static int copy_action(const struct nlattr *from,
2138                        struct sw_flow_actions **sfa, bool log)
2139 {
2140         int totlen = NLA_ALIGN(from->nla_len);
2141         struct nlattr *to;
2142
2143         to = reserve_sfa_size(sfa, from->nla_len, log);
2144         if (IS_ERR(to))
2145                 return PTR_ERR(to);
2146
2147         memcpy(to, from, totlen);
2148         return 0;
2149 }
2150
2151 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2152                                   const struct sw_flow_key *key,
2153                                   int depth, struct sw_flow_actions **sfa,
2154                                   __be16 eth_type, __be16 vlan_tci, bool log)
2155 {
2156         const struct nlattr *a;
2157         int rem, err;
2158
2159         if (depth >= SAMPLE_ACTION_DEPTH)
2160                 return -EOVERFLOW;
2161
2162         nla_for_each_nested(a, attr, rem) {
2163                 /* Expected argument lengths, (u32)-1 for variable length. */
2164                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2165                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2166                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2167                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2168                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2169                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2170                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2171                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2172                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2173                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2174                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2175                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2176                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2177                 };
2178                 const struct ovs_action_push_vlan *vlan;
2179                 int type = nla_type(a);
2180                 bool skip_copy;
2181
2182                 if (type > OVS_ACTION_ATTR_MAX ||
2183                     (action_lens[type] != nla_len(a) &&
2184                      action_lens[type] != (u32)-1))
2185                         return -EINVAL;
2186
2187                 skip_copy = false;
2188                 switch (type) {
2189                 case OVS_ACTION_ATTR_UNSPEC:
2190                         return -EINVAL;
2191
2192                 case OVS_ACTION_ATTR_USERSPACE:
2193                         err = validate_userspace(a);
2194                         if (err)
2195                                 return err;
2196                         break;
2197
2198                 case OVS_ACTION_ATTR_OUTPUT:
2199                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2200                                 return -EINVAL;
2201                         break;
2202
2203                 case OVS_ACTION_ATTR_HASH: {
2204                         const struct ovs_action_hash *act_hash = nla_data(a);
2205
2206                         switch (act_hash->hash_alg) {
2207                         case OVS_HASH_ALG_L4:
2208                                 break;
2209                         default:
2210                                 return  -EINVAL;
2211                         }
2212
2213                         break;
2214                 }
2215
2216                 case OVS_ACTION_ATTR_POP_VLAN:
2217                         vlan_tci = htons(0);
2218                         break;
2219
2220                 case OVS_ACTION_ATTR_PUSH_VLAN:
2221                         vlan = nla_data(a);
2222                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2223                                 return -EINVAL;
2224                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2225                                 return -EINVAL;
2226                         vlan_tci = vlan->vlan_tci;
2227                         break;
2228
2229                 case OVS_ACTION_ATTR_RECIRC:
2230                         break;
2231
2232                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2233                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2234
2235                         if (!eth_p_mpls(mpls->mpls_ethertype))
2236                                 return -EINVAL;
2237                         /* Prohibit push MPLS other than to a white list
2238                          * for packets that have a known tag order.
2239                          */
2240                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2241                             (eth_type != htons(ETH_P_IP) &&
2242                              eth_type != htons(ETH_P_IPV6) &&
2243                              eth_type != htons(ETH_P_ARP) &&
2244                              eth_type != htons(ETH_P_RARP) &&
2245                              !eth_p_mpls(eth_type)))
2246                                 return -EINVAL;
2247                         eth_type = mpls->mpls_ethertype;
2248                         break;
2249                 }
2250
2251                 case OVS_ACTION_ATTR_POP_MPLS:
2252                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2253                             !eth_p_mpls(eth_type))
2254                                 return -EINVAL;
2255
2256                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2257                          * as there is no check here to ensure that the new
2258                          * eth_type is valid and thus set actions could
2259                          * write off the end of the packet or otherwise
2260                          * corrupt it.
2261                          *
2262                          * Support for these actions is planned using packet
2263                          * recirculation.
2264                          */
2265                         eth_type = htons(0);
2266                         break;
2267
2268                 case OVS_ACTION_ATTR_SET:
2269                         err = validate_set(a, key, sfa,
2270                                            &skip_copy, eth_type, false, log);
2271                         if (err)
2272                                 return err;
2273                         break;
2274
2275                 case OVS_ACTION_ATTR_SET_MASKED:
2276                         err = validate_set(a, key, sfa,
2277                                            &skip_copy, eth_type, true, log);
2278                         if (err)
2279                                 return err;
2280                         break;
2281
2282                 case OVS_ACTION_ATTR_SAMPLE:
2283                         err = validate_and_copy_sample(net, a, key, depth, sfa,
2284                                                        eth_type, vlan_tci, log);
2285                         if (err)
2286                                 return err;
2287                         skip_copy = true;
2288                         break;
2289
2290                 case OVS_ACTION_ATTR_CT:
2291                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2292                         if (err)
2293                                 return err;
2294                         skip_copy = true;
2295                         break;
2296
2297                 default:
2298                         OVS_NLERR(log, "Unknown Action type %d", type);
2299                         return -EINVAL;
2300                 }
2301                 if (!skip_copy) {
2302                         err = copy_action(a, sfa, log);
2303                         if (err)
2304                                 return err;
2305                 }
2306         }
2307
2308         if (rem > 0)
2309                 return -EINVAL;
2310
2311         return 0;
2312 }
2313
2314 /* 'key' must be the masked key. */
2315 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2316                          const struct sw_flow_key *key,
2317                          struct sw_flow_actions **sfa, bool log)
2318 {
2319         int err;
2320
2321         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2322         if (IS_ERR(*sfa))
2323                 return PTR_ERR(*sfa);
2324
2325         (*sfa)->orig_len = nla_len(attr);
2326         err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2327                                      key->eth.tci, log);
2328         if (err)
2329                 ovs_nla_free_flow_actions(*sfa);
2330
2331         return err;
2332 }
2333
2334 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2335 {
2336         const struct nlattr *a;
2337         struct nlattr *start;
2338         int err = 0, rem;
2339
2340         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2341         if (!start)
2342                 return -EMSGSIZE;
2343
2344         nla_for_each_nested(a, attr, rem) {
2345                 int type = nla_type(a);
2346                 struct nlattr *st_sample;
2347
2348                 switch (type) {
2349                 case OVS_SAMPLE_ATTR_PROBABILITY:
2350                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2351                                     sizeof(u32), nla_data(a)))
2352                                 return -EMSGSIZE;
2353                         break;
2354                 case OVS_SAMPLE_ATTR_ACTIONS:
2355                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2356                         if (!st_sample)
2357                                 return -EMSGSIZE;
2358                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2359                         if (err)
2360                                 return err;
2361                         nla_nest_end(skb, st_sample);
2362                         break;
2363                 }
2364         }
2365
2366         nla_nest_end(skb, start);
2367         return err;
2368 }
2369
2370 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2371 {
2372         const struct nlattr *ovs_key = nla_data(a);
2373         int key_type = nla_type(ovs_key);
2374         struct nlattr *start;
2375         int err;
2376
2377         switch (key_type) {
2378         case OVS_KEY_ATTR_TUNNEL_INFO: {
2379                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2380                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2381
2382                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2383                 if (!start)
2384                         return -EMSGSIZE;
2385
2386                 err = ipv4_tun_to_nlattr(skb, &tun_info->key,
2387                                          tun_info->options_len ?
2388                                              ip_tunnel_info_opts(tun_info) : NULL,
2389                                          tun_info->options_len);
2390                 if (err)
2391                         return err;
2392                 nla_nest_end(skb, start);
2393                 break;
2394         }
2395         default:
2396                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2397                         return -EMSGSIZE;
2398                 break;
2399         }
2400
2401         return 0;
2402 }
2403
2404 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2405                                                 struct sk_buff *skb)
2406 {
2407         const struct nlattr *ovs_key = nla_data(a);
2408         struct nlattr *nla;
2409         size_t key_len = nla_len(ovs_key) / 2;
2410
2411         /* Revert the conversion we did from a non-masked set action to
2412          * masked set action.
2413          */
2414         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2415         if (!nla)
2416                 return -EMSGSIZE;
2417
2418         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2419                 return -EMSGSIZE;
2420
2421         nla_nest_end(skb, nla);
2422         return 0;
2423 }
2424
2425 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2426 {
2427         const struct nlattr *a;
2428         int rem, err;
2429
2430         nla_for_each_attr(a, attr, len, rem) {
2431                 int type = nla_type(a);
2432
2433                 switch (type) {
2434                 case OVS_ACTION_ATTR_SET:
2435                         err = set_action_to_attr(a, skb);
2436                         if (err)
2437                                 return err;
2438                         break;
2439
2440                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2441                         err = masked_set_action_to_set_action_attr(a, skb);
2442                         if (err)
2443                                 return err;
2444                         break;
2445
2446                 case OVS_ACTION_ATTR_SAMPLE:
2447                         err = sample_action_to_attr(a, skb);
2448                         if (err)
2449                                 return err;
2450                         break;
2451
2452                 case OVS_ACTION_ATTR_CT:
2453                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2454                         if (err)
2455                                 return err;
2456                         break;
2457
2458                 default:
2459                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2460                                 return -EMSGSIZE;
2461                         break;
2462                 }
2463         }
2464
2465         return 0;
2466 }