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(1)   /* 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_LABEL */
 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(u8) },
 353         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 354         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 355         [OVS_KEY_ATTR_CT_LABEL]  = { .len = sizeof(struct ovs_key_ct_label) },
 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                 u8 ct_state = nla_get_u8(a[OVS_KEY_ATTR_CT_STATE]);
 818
 819                 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
 820                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
 821         }
 822         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
 823             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
 824                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
 825
 826                 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
 827                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
 828         }
 829         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
 830             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
 831                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
 832
 833                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
 834                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
 835         }
 836         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABEL) &&
 837             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABEL)) {
 838                 const struct ovs_key_ct_label *cl;
 839
 840                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABEL]);
 841                 SW_FLOW_KEY_MEMCPY(match, ct.label, cl->ct_label,
 842                                    sizeof(*cl), is_mask);
 843                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABEL);
 844         }
 845         return 0;
 846 }
 847
 848 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
 849                                 u64 attrs, const struct nlattr **a,
 850                                 bool is_mask, bool log)
 851 {
 852         int err;
 853
 854         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
 855         if (err)
 856                 return err;
 857
 858         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
 859                 const struct ovs_key_ethernet *eth_key;
 860
 861                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 862                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 863                                 eth_key->eth_src, ETH_ALEN, is_mask);
 864                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 865                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 866                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
 867         }
 868
 869         if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
 870                 __be16 tci;
 871
 872                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 873                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 874                         if (is_mask)
 875                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 876                         else
 877                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 878
 879                         return -EINVAL;
 880                 }
 881
 882                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 883                 attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
 884         }
 885
 886         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
 887                 __be16 eth_type;
 888
 889                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 890                 if (is_mask) {
 891                         /* Always exact match EtherType. */
 892                         eth_type = htons(0xffff);
 893                 } else if (!eth_proto_is_802_3(eth_type)) {
 894                         OVS_NLERR(log, "EtherType %x is less than min %x",
 895                                   ntohs(eth_type), ETH_P_802_3_MIN);
 896                         return -EINVAL;
 897                 }
 898
 899                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 900                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
 901         } else if (!is_mask) {
 902                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 903         }
 904
 905         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
 906                 const struct ovs_key_ipv4 *ipv4_key;
 907
 908                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 909                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 910                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 911                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 912                         return -EINVAL;
 913                 }
 914                 SW_FLOW_KEY_PUT(match, ip.proto,
 915                                 ipv4_key->ipv4_proto, is_mask);
 916                 SW_FLOW_KEY_PUT(match, ip.tos,
 917                                 ipv4_key->ipv4_tos, is_mask);
 918                 SW_FLOW_KEY_PUT(match, ip.ttl,
 919                                 ipv4_key->ipv4_ttl, is_mask);
 920                 SW_FLOW_KEY_PUT(match, ip.frag,
 921                                 ipv4_key->ipv4_frag, is_mask);
 922                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 923                                 ipv4_key->ipv4_src, is_mask);
 924                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 925                                 ipv4_key->ipv4_dst, is_mask);
 926                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
 927         }
 928
 929         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
 930                 const struct ovs_key_ipv6 *ipv6_key;
 931
 932                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 933                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 934                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 935                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 936                         return -EINVAL;
 937                 }
 938
 939                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 940                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
 941                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 942                         return -EINVAL;
 943                 }
 944
 945                 SW_FLOW_KEY_PUT(match, ipv6.label,
 946                                 ipv6_key->ipv6_label, is_mask);
 947                 SW_FLOW_KEY_PUT(match, ip.proto,
 948                                 ipv6_key->ipv6_proto, is_mask);
 949                 SW_FLOW_KEY_PUT(match, ip.tos,
 950                                 ipv6_key->ipv6_tclass, is_mask);
 951                 SW_FLOW_KEY_PUT(match, ip.ttl,
 952                                 ipv6_key->ipv6_hlimit, is_mask);
 953                 SW_FLOW_KEY_PUT(match, ip.frag,
 954                                 ipv6_key->ipv6_frag, is_mask);
 955                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
 956                                 ipv6_key->ipv6_src,
 957                                 sizeof(match->key->ipv6.addr.src),
 958                                 is_mask);
 959                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
 960                                 ipv6_key->ipv6_dst,
 961                                 sizeof(match->key->ipv6.addr.dst),
 962                                 is_mask);
 963
 964                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
 965         }
 966
 967         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
 968                 const struct ovs_key_arp *arp_key;
 969
 970                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
 971                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
 972                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
 973                                   arp_key->arp_op);
 974                         return -EINVAL;
 975                 }
 976
 977                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 978                                 arp_key->arp_sip, is_mask);
 979                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 980                         arp_key->arp_tip, is_mask);
 981                 SW_FLOW_KEY_PUT(match, ip.proto,
 982                                 ntohs(arp_key->arp_op), is_mask);
 983                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
 984                                 arp_key->arp_sha, ETH_ALEN, is_mask);
 985                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
 986                                 arp_key->arp_tha, ETH_ALEN, is_mask);
 987
 988                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
 989         }
 990
 991         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
 992                 const struct ovs_key_mpls *mpls_key;
 993
 994                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
 995                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
 996                                 mpls_key->mpls_lse, is_mask);
 997
 998                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
 999          }
1000
1001         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1002                 const struct ovs_key_tcp *tcp_key;
1003
1004                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1005                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1006                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1007                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1008         }
1009
1010         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1011                 SW_FLOW_KEY_PUT(match, tp.flags,
1012                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1013                                 is_mask);
1014                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1015         }
1016
1017         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1018                 const struct ovs_key_udp *udp_key;
1019
1020                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1021                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1022                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1023                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1024         }
1025
1026         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1027                 const struct ovs_key_sctp *sctp_key;
1028
1029                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1030                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1031                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1032                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1033         }
1034
1035         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1036                 const struct ovs_key_icmp *icmp_key;
1037
1038                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1039                 SW_FLOW_KEY_PUT(match, tp.src,
1040                                 htons(icmp_key->icmp_type), is_mask);
1041                 SW_FLOW_KEY_PUT(match, tp.dst,
1042                                 htons(icmp_key->icmp_code), is_mask);
1043                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1044         }
1045
1046         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1047                 const struct ovs_key_icmpv6 *icmpv6_key;
1048
1049                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1050                 SW_FLOW_KEY_PUT(match, tp.src,
1051                                 htons(icmpv6_key->icmpv6_type), is_mask);
1052                 SW_FLOW_KEY_PUT(match, tp.dst,
1053                                 htons(icmpv6_key->icmpv6_code), is_mask);
1054                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1055         }
1056
1057         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1058                 const struct ovs_key_nd *nd_key;
1059
1060                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1061                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1062                         nd_key->nd_target,
1063                         sizeof(match->key->ipv6.nd.target),
1064                         is_mask);
1065                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1066                         nd_key->nd_sll, ETH_ALEN, is_mask);
1067                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1068                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1069                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1070         }
1071
1072         if (attrs != 0) {
1073                 OVS_NLERR(log, "Unknown key attributes %llx",
1074                           (unsigned long long)attrs);
1075                 return -EINVAL;
1076         }
1077
1078         return 0;
1079 }
1080
1081 static void nlattr_set(struct nlattr *attr, u8 val,
1082                        const struct ovs_len_tbl *tbl)
1083 {
1084         struct nlattr *nla;
1085         int rem;
1086
1087         /* The nlattr stream should already have been validated */
1088         nla_for_each_nested(nla, attr, rem) {
1089                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1090                         if (tbl[nla_type(nla)].next)
1091                                 tbl = tbl[nla_type(nla)].next;
1092                         nlattr_set(nla, val, tbl);
1093                 } else {
1094                         memset(nla_data(nla), val, nla_len(nla));
1095                 }
1096         }
1097 }
1098
1099 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1100 {
1101         nlattr_set(attr, val, ovs_key_lens);
1102 }
1103
1104 /**
1105  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1106  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1107  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1108  * does not include any don't care bit.
1109  * @net: Used to determine per-namespace field support.
1110  * @match: receives the extracted flow match information.
1111  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1112  * sequence. The fields should of the packet that triggered the creation
1113  * of this flow.
1114  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1115  * attribute specifies the mask field of the wildcarded flow.
1116  * @log: Boolean to allow kernel error logging.  Normally true, but when
1117  * probing for feature compatibility this should be passed in as false to
1118  * suppress unnecessary error logging.
1119  */
1120 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1121                       const struct nlattr *nla_key,
1122                       const struct nlattr *nla_mask,
1123                       bool log)
1124 {
1125         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1126         const struct nlattr *encap;
1127         struct nlattr *newmask = NULL;
1128         u64 key_attrs = 0;
1129         u64 mask_attrs = 0;
1130         bool encap_valid = false;
1131         int err;
1132
1133         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1134         if (err)
1135                 return err;
1136
1137         if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1138             (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1139             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1140                 __be16 tci;
1141
1142                 if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1143                       (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1144                         OVS_NLERR(log, "Invalid Vlan frame.");
1145                         return -EINVAL;
1146                 }
1147
1148                 key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1149                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1150                 encap = a[OVS_KEY_ATTR_ENCAP];
1151                 key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1152                 encap_valid = true;
1153
1154                 if (tci & htons(VLAN_TAG_PRESENT)) {
1155                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1156                         if (err)
1157                                 return err;
1158                 } else if (!tci) {
1159                         /* Corner case for truncated 802.1Q header. */
1160                         if (nla_len(encap)) {
1161                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1162                                 return -EINVAL;
1163                         }
1164                 } else {
1165                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1166                         return  -EINVAL;
1167                 }
1168         }
1169
1170         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1171         if (err)
1172                 return err;
1173
1174         if (match->mask) {
1175                 if (!nla_mask) {
1176                         /* Create an exact match mask. We need to set to 0xff
1177                          * all the 'match->mask' fields that have been touched
1178                          * in 'match->key'. We cannot simply memset
1179                          * 'match->mask', because padding bytes and fields not
1180                          * specified in 'match->key' should be left to 0.
1181                          * Instead, we use a stream of netlink attributes,
1182                          * copied from 'key' and set to 0xff.
1183                          * ovs_key_from_nlattrs() will take care of filling
1184                          * 'match->mask' appropriately.
1185                          */
1186                         newmask = kmemdup(nla_key,
1187                                           nla_total_size(nla_len(nla_key)),
1188                                           GFP_KERNEL);
1189                         if (!newmask)
1190                                 return -ENOMEM;
1191
1192                         mask_set_nlattr(newmask, 0xff);
1193
1194                         /* The userspace does not send tunnel attributes that
1195                          * are 0, but we should not wildcard them nonetheless.
1196                          */
1197                         if (match->key->tun_key.u.ipv4.dst)
1198                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1199                                                          0xff, true);
1200
1201                         nla_mask = newmask;
1202                 }
1203
1204                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1205                 if (err)
1206                         goto free_newmask;
1207
1208                 /* Always match on tci. */
1209                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1210
1211                 if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
1212                         __be16 eth_type = 0;
1213                         __be16 tci = 0;
1214
1215                         if (!encap_valid) {
1216                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1217                                 err = -EINVAL;
1218                                 goto free_newmask;
1219                         }
1220
1221                         mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1222                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1223                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1224
1225                         if (eth_type == htons(0xffff)) {
1226                                 mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1227                                 encap = a[OVS_KEY_ATTR_ENCAP];
1228                                 err = parse_flow_mask_nlattrs(encap, a,
1229                                                               &mask_attrs, log);
1230                                 if (err)
1231                                         goto free_newmask;
1232                         } else {
1233                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1234                                           ntohs(eth_type));
1235                                 err = -EINVAL;
1236                                 goto free_newmask;
1237                         }
1238
1239                         if (a[OVS_KEY_ATTR_VLAN])
1240                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1241
1242                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1243                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1244                                           ntohs(tci));
1245                                 err = -EINVAL;
1246                                 goto free_newmask;
1247                         }
1248                 }
1249
1250                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1251                                            log);
1252                 if (err)
1253                         goto free_newmask;
1254         }
1255
1256         if (!match_validate(match, key_attrs, mask_attrs, log))
1257                 err = -EINVAL;
1258
1259 free_newmask:
1260         kfree(newmask);
1261         return err;
1262 }
1263
1264 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1265 {
1266         size_t len;
1267
1268         if (!attr)
1269                 return 0;
1270
1271         len = nla_len(attr);
1272         if (len < 1 || len > MAX_UFID_LENGTH) {
1273                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1274                           nla_len(attr), MAX_UFID_LENGTH);
1275                 return 0;
1276         }
1277
1278         return len;
1279 }
1280
1281 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1282  * or false otherwise.
1283  */
1284 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1285                       bool log)
1286 {
1287         sfid->ufid_len = get_ufid_len(attr, log);
1288         if (sfid->ufid_len)
1289                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1290
1291         return sfid->ufid_len;
1292 }
1293
1294 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1295                            const struct sw_flow_key *key, bool log)
1296 {
1297         struct sw_flow_key *new_key;
1298
1299         if (ovs_nla_get_ufid(sfid, ufid, log))
1300                 return 0;
1301
1302         /* If UFID was not provided, use unmasked key. */
1303         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1304         if (!new_key)
1305                 return -ENOMEM;
1306         memcpy(new_key, key, sizeof(*key));
1307         sfid->unmasked_key = new_key;
1308
1309         return 0;
1310 }
1311
1312 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1313 {
1314         return attr ? nla_get_u32(attr) : 0;
1315 }
1316
1317 /**
1318  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1319  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1320  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1321  * sequence.
1322  * @log: Boolean to allow kernel error logging.  Normally true, but when
1323  * probing for feature compatibility this should be passed in as false to
1324  * suppress unnecessary error logging.
1325  *
1326  * This parses a series of Netlink attributes that form a flow key, which must
1327  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1328  * get the metadata, that is, the parts of the flow key that cannot be
1329  * extracted from the packet itself.
1330  */
1331
1332 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1333                               struct sw_flow_key *key,
1334                               bool log)
1335 {
1336         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1337         struct sw_flow_match match;
1338         u64 attrs = 0;
1339         int err;
1340
1341         err = parse_flow_nlattrs(attr, a, &attrs, log);
1342         if (err)
1343                 return -EINVAL;
1344
1345         memset(&match, 0, sizeof(match));
1346         match.key = key;
1347
1348         memset(&key->ct, 0, sizeof(key->ct));
1349         key->phy.in_port = DP_MAX_PORTS;
1350
1351         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1352 }
1353
1354 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1355                              const struct sw_flow_key *output, bool is_mask,
1356                              struct sk_buff *skb)
1357 {
1358         struct ovs_key_ethernet *eth_key;
1359         struct nlattr *nla, *encap;
1360
1361         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1362                 goto nla_put_failure;
1363
1364         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1365                 goto nla_put_failure;
1366
1367         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1368                 goto nla_put_failure;
1369
1370         if ((swkey->tun_key.u.ipv4.dst || is_mask)) {
1371                 const void *opts = NULL;
1372
1373                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1374                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1375
1376                 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1377                                        swkey->tun_opts_len))
1378                         goto nla_put_failure;
1379         }
1380
1381         if (swkey->phy.in_port == DP_MAX_PORTS) {
1382                 if (is_mask && (output->phy.in_port == 0xffff))
1383                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1384                                 goto nla_put_failure;
1385         } else {
1386                 u16 upper_u16;
1387                 upper_u16 = !is_mask ? 0 : 0xffff;
1388
1389                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1390                                 (upper_u16 << 16) | output->phy.in_port))
1391                         goto nla_put_failure;
1392         }
1393
1394         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1395                 goto nla_put_failure;
1396
1397         if (ovs_ct_put_key(output, skb))
1398                 goto nla_put_failure;
1399
1400         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1401         if (!nla)
1402                 goto nla_put_failure;
1403
1404         eth_key = nla_data(nla);
1405         ether_addr_copy(eth_key->eth_src, output->eth.src);
1406         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1407
1408         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1409                 __be16 eth_type;
1410                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1411                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1412                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1413                         goto nla_put_failure;
1414                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1415                 if (!swkey->eth.tci)
1416                         goto unencap;
1417         } else
1418                 encap = NULL;
1419
1420         if (swkey->eth.type == htons(ETH_P_802_2)) {
1421                 /*
1422                  * Ethertype 802.2 is represented in the netlink with omitted
1423                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1424                  * 0xffff in the mask attribute.  Ethertype can also
1425                  * be wildcarded.
1426                  */
1427                 if (is_mask && output->eth.type)
1428                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1429                                                 output->eth.type))
1430                                 goto nla_put_failure;
1431                 goto unencap;
1432         }
1433
1434         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1435                 goto nla_put_failure;
1436
1437         if (swkey->eth.type == htons(ETH_P_IP)) {
1438                 struct ovs_key_ipv4 *ipv4_key;
1439
1440                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1441                 if (!nla)
1442                         goto nla_put_failure;
1443                 ipv4_key = nla_data(nla);
1444                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1445                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1446                 ipv4_key->ipv4_proto = output->ip.proto;
1447                 ipv4_key->ipv4_tos = output->ip.tos;
1448                 ipv4_key->ipv4_ttl = output->ip.ttl;
1449                 ipv4_key->ipv4_frag = output->ip.frag;
1450         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1451                 struct ovs_key_ipv6 *ipv6_key;
1452
1453                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1454                 if (!nla)
1455                         goto nla_put_failure;
1456                 ipv6_key = nla_data(nla);
1457                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1458                                 sizeof(ipv6_key->ipv6_src));
1459                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1460                                 sizeof(ipv6_key->ipv6_dst));
1461                 ipv6_key->ipv6_label = output->ipv6.label;
1462                 ipv6_key->ipv6_proto = output->ip.proto;
1463                 ipv6_key->ipv6_tclass = output->ip.tos;
1464                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1465                 ipv6_key->ipv6_frag = output->ip.frag;
1466         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1467                    swkey->eth.type == htons(ETH_P_RARP)) {
1468                 struct ovs_key_arp *arp_key;
1469
1470                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1471                 if (!nla)
1472                         goto nla_put_failure;
1473                 arp_key = nla_data(nla);
1474                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1475                 arp_key->arp_sip = output->ipv4.addr.src;
1476                 arp_key->arp_tip = output->ipv4.addr.dst;
1477                 arp_key->arp_op = htons(output->ip.proto);
1478                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1479                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1480         } else if (eth_p_mpls(swkey->eth.type)) {
1481                 struct ovs_key_mpls *mpls_key;
1482
1483                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1484                 if (!nla)
1485                         goto nla_put_failure;
1486                 mpls_key = nla_data(nla);
1487                 mpls_key->mpls_lse = output->mpls.top_lse;
1488         }
1489
1490         if ((swkey->eth.type == htons(ETH_P_IP) ||
1491              swkey->eth.type == htons(ETH_P_IPV6)) &&
1492              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1493
1494                 if (swkey->ip.proto == IPPROTO_TCP) {
1495                         struct ovs_key_tcp *tcp_key;
1496
1497                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1498                         if (!nla)
1499                                 goto nla_put_failure;
1500                         tcp_key = nla_data(nla);
1501                         tcp_key->tcp_src = output->tp.src;
1502                         tcp_key->tcp_dst = output->tp.dst;
1503                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1504                                          output->tp.flags))
1505                                 goto nla_put_failure;
1506                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1507                         struct ovs_key_udp *udp_key;
1508
1509                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1510                         if (!nla)
1511                                 goto nla_put_failure;
1512                         udp_key = nla_data(nla);
1513                         udp_key->udp_src = output->tp.src;
1514                         udp_key->udp_dst = output->tp.dst;
1515                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1516                         struct ovs_key_sctp *sctp_key;
1517
1518                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1519                         if (!nla)
1520                                 goto nla_put_failure;
1521                         sctp_key = nla_data(nla);
1522                         sctp_key->sctp_src = output->tp.src;
1523                         sctp_key->sctp_dst = output->tp.dst;
1524                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1525                            swkey->ip.proto == IPPROTO_ICMP) {
1526                         struct ovs_key_icmp *icmp_key;
1527
1528                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1529                         if (!nla)
1530                                 goto nla_put_failure;
1531                         icmp_key = nla_data(nla);
1532                         icmp_key->icmp_type = ntohs(output->tp.src);
1533                         icmp_key->icmp_code = ntohs(output->tp.dst);
1534                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1535                            swkey->ip.proto == IPPROTO_ICMPV6) {
1536                         struct ovs_key_icmpv6 *icmpv6_key;
1537
1538                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1539                                                 sizeof(*icmpv6_key));
1540                         if (!nla)
1541                                 goto nla_put_failure;
1542                         icmpv6_key = nla_data(nla);
1543                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1544                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1545
1546                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1547                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1548                                 struct ovs_key_nd *nd_key;
1549
1550                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1551                                 if (!nla)
1552                                         goto nla_put_failure;
1553                                 nd_key = nla_data(nla);
1554                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1555                                                         sizeof(nd_key->nd_target));
1556                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1557                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1558                         }
1559                 }
1560         }
1561
1562 unencap:
1563         if (encap)
1564                 nla_nest_end(skb, encap);
1565
1566         return 0;
1567
1568 nla_put_failure:
1569         return -EMSGSIZE;
1570 }
1571
1572 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1573                     const struct sw_flow_key *output, int attr, bool is_mask,
1574                     struct sk_buff *skb)
1575 {
1576         int err;
1577         struct nlattr *nla;
1578
1579         nla = nla_nest_start(skb, attr);
1580         if (!nla)
1581                 return -EMSGSIZE;
1582         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1583         if (err)
1584                 return err;
1585         nla_nest_end(skb, nla);
1586
1587         return 0;
1588 }
1589
1590 /* Called with ovs_mutex or RCU read lock. */
1591 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1592 {
1593         if (ovs_identifier_is_ufid(&flow->id))
1594                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1595                                flow->id.ufid);
1596
1597         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1598                                OVS_FLOW_ATTR_KEY, false, skb);
1599 }
1600
1601 /* Called with ovs_mutex or RCU read lock. */
1602 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1603 {
1604         return ovs_nla_put_key(&flow->key, &flow->key,
1605                                 OVS_FLOW_ATTR_KEY, false, skb);
1606 }
1607
1608 /* Called with ovs_mutex or RCU read lock. */
1609 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1610 {
1611         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1612                                 OVS_FLOW_ATTR_MASK, true, skb);
1613 }
1614
1615 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1616
1617 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1618 {
1619         struct sw_flow_actions *sfa;
1620
1621         if (size > MAX_ACTIONS_BUFSIZE) {
1622                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1623                 return ERR_PTR(-EINVAL);
1624         }
1625
1626         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1627         if (!sfa)
1628                 return ERR_PTR(-ENOMEM);
1629
1630         sfa->actions_len = 0;
1631         return sfa;
1632 }
1633
1634 static void ovs_nla_free_set_action(const struct nlattr *a)
1635 {
1636         const struct nlattr *ovs_key = nla_data(a);
1637         struct ovs_tunnel_info *ovs_tun;
1638
1639         switch (nla_type(ovs_key)) {
1640         case OVS_KEY_ATTR_TUNNEL_INFO:
1641                 ovs_tun = nla_data(ovs_key);
1642                 dst_release((struct dst_entry *)ovs_tun->tun_dst);
1643                 break;
1644         }
1645 }
1646
1647 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1648 {
1649         const struct nlattr *a;
1650         int rem;
1651
1652         if (!sf_acts)
1653                 return;
1654
1655         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1656                 switch (nla_type(a)) {
1657                 case OVS_ACTION_ATTR_SET:
1658                         ovs_nla_free_set_action(a);
1659                         break;
1660                 case OVS_ACTION_ATTR_CT:
1661                         ovs_ct_free_action(a);
1662                         break;
1663                 }
1664         }
1665
1666         kfree(sf_acts);
1667 }
1668
1669 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1670 {
1671         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1672 }
1673
1674 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1675  * The caller must hold rcu_read_lock for this to be sensible. */
1676 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1677 {
1678         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1679 }
1680
1681 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1682                                        int attr_len, bool log)
1683 {
1684
1685         struct sw_flow_actions *acts;
1686         int new_acts_size;
1687         int req_size = NLA_ALIGN(attr_len);
1688         int next_offset = offsetof(struct sw_flow_actions, actions) +
1689                                         (*sfa)->actions_len;
1690
1691         if (req_size <= (ksize(*sfa) - next_offset))
1692                 goto out;
1693
1694         new_acts_size = ksize(*sfa) * 2;
1695
1696         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1697                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1698                         return ERR_PTR(-EMSGSIZE);
1699                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1700         }
1701
1702         acts = nla_alloc_flow_actions(new_acts_size, log);
1703         if (IS_ERR(acts))
1704                 return (void *)acts;
1705
1706         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1707         acts->actions_len = (*sfa)->actions_len;
1708         acts->orig_len = (*sfa)->orig_len;
1709         kfree(*sfa);
1710         *sfa = acts;
1711
1712 out:
1713         (*sfa)->actions_len += req_size;
1714         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1715 }
1716
1717 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1718                                    int attrtype, void *data, int len, bool log)
1719 {
1720         struct nlattr *a;
1721
1722         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1723         if (IS_ERR(a))
1724                 return a;
1725
1726         a->nla_type = attrtype;
1727         a->nla_len = nla_attr_size(len);
1728
1729         if (data)
1730                 memcpy(nla_data(a), data, len);
1731         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1732
1733         return a;
1734 }
1735
1736 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1737                        int len, bool log)
1738 {
1739         struct nlattr *a;
1740
1741         a = __add_action(sfa, attrtype, data, len, log);
1742
1743         return PTR_ERR_OR_ZERO(a);
1744 }
1745
1746 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1747                                           int attrtype, bool log)
1748 {
1749         int used = (*sfa)->actions_len;
1750         int err;
1751
1752         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1753         if (err)
1754                 return err;
1755
1756         return used;
1757 }
1758
1759 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1760                                          int st_offset)
1761 {
1762         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1763                                                                st_offset);
1764
1765         a->nla_len = sfa->actions_len - st_offset;
1766 }
1767
1768 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1769                                   const struct sw_flow_key *key,
1770                                   int depth, struct sw_flow_actions **sfa,
1771                                   __be16 eth_type, __be16 vlan_tci, bool log);
1772
1773 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1774                                     const struct sw_flow_key *key, int depth,
1775                                     struct sw_flow_actions **sfa,
1776                                     __be16 eth_type, __be16 vlan_tci, bool log)
1777 {
1778         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1779         const struct nlattr *probability, *actions;
1780         const struct nlattr *a;
1781         int rem, start, err, st_acts;
1782
1783         memset(attrs, 0, sizeof(attrs));
1784         nla_for_each_nested(a, attr, rem) {
1785                 int type = nla_type(a);
1786                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1787                         return -EINVAL;
1788                 attrs[type] = a;
1789         }
1790         if (rem)
1791                 return -EINVAL;
1792
1793         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1794         if (!probability || nla_len(probability) != sizeof(u32))
1795                 return -EINVAL;
1796
1797         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1798         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1799                 return -EINVAL;
1800
1801         /* validation done, copy sample action. */
1802         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1803         if (start < 0)
1804                 return start;
1805         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1806                                  nla_data(probability), sizeof(u32), log);
1807         if (err)
1808                 return err;
1809         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1810         if (st_acts < 0)
1811                 return st_acts;
1812
1813         err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1814                                      eth_type, vlan_tci, log);
1815         if (err)
1816                 return err;
1817
1818         add_nested_action_end(*sfa, st_acts);
1819         add_nested_action_end(*sfa, start);
1820
1821         return 0;
1822 }
1823
1824 void ovs_match_init(struct sw_flow_match *match,
1825                     struct sw_flow_key *key,
1826                     struct sw_flow_mask *mask)
1827 {
1828         memset(match, 0, sizeof(*match));
1829         match->key = key;
1830         match->mask = mask;
1831
1832         memset(key, 0, sizeof(*key));
1833
1834         if (mask) {
1835                 memset(&mask->key, 0, sizeof(mask->key));
1836                 mask->range.start = mask->range.end = 0;
1837         }
1838 }
1839
1840 static int validate_geneve_opts(struct sw_flow_key *key)
1841 {
1842         struct geneve_opt *option;
1843         int opts_len = key->tun_opts_len;
1844         bool crit_opt = false;
1845
1846         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1847         while (opts_len > 0) {
1848                 int len;
1849
1850                 if (opts_len < sizeof(*option))
1851                         return -EINVAL;
1852
1853                 len = sizeof(*option) + option->length * 4;
1854                 if (len > opts_len)
1855                         return -EINVAL;
1856
1857                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1858
1859                 option = (struct geneve_opt *)((u8 *)option + len);
1860                 opts_len -= len;
1861         };
1862
1863         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1864
1865         return 0;
1866 }
1867
1868 static int validate_and_copy_set_tun(const struct nlattr *attr,
1869                                      struct sw_flow_actions **sfa, bool log)
1870 {
1871         struct sw_flow_match match;
1872         struct sw_flow_key key;
1873         struct metadata_dst *tun_dst;
1874         struct ip_tunnel_info *tun_info;
1875         struct ovs_tunnel_info *ovs_tun;
1876         struct nlattr *a;
1877         int err = 0, start, opts_type;
1878
1879         ovs_match_init(&match, &key, NULL);
1880         opts_type = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1881         if (opts_type < 0)
1882                 return opts_type;
1883
1884         if (key.tun_opts_len) {
1885                 switch (opts_type) {
1886                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1887                         err = validate_geneve_opts(&key);
1888                         if (err < 0)
1889                                 return err;
1890                         break;
1891                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1892                         break;
1893                 }
1894         };
1895
1896         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1897         if (start < 0)
1898                 return start;
1899
1900         tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1901         if (!tun_dst)
1902                 return -ENOMEM;
1903
1904         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1905                          sizeof(*ovs_tun), log);
1906         if (IS_ERR(a)) {
1907                 dst_release((struct dst_entry *)tun_dst);
1908                 return PTR_ERR(a);
1909         }
1910
1911         ovs_tun = nla_data(a);
1912         ovs_tun->tun_dst = tun_dst;
1913
1914         tun_info = &tun_dst->u.tun_info;
1915         tun_info->mode = IP_TUNNEL_INFO_TX;
1916         tun_info->key = key.tun_key;
1917
1918         /* We need to store the options in the action itself since
1919          * everything else will go away after flow setup. We can append
1920          * it to tun_info and then point there.
1921          */
1922         ip_tunnel_info_opts_set(tun_info,
1923                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
1924                                 key.tun_opts_len);
1925         add_nested_action_end(*sfa, start);
1926
1927         return err;
1928 }
1929
1930 /* Return false if there are any non-masked bits set.
1931  * Mask follows data immediately, before any netlink padding.
1932  */
1933 static bool validate_masked(u8 *data, int len)
1934 {
1935         u8 *mask = data + len;
1936
1937         while (len--)
1938                 if (*data++ & ~*mask++)
1939                         return false;
1940
1941         return true;
1942 }
1943
1944 static int validate_set(const struct nlattr *a,
1945                         const struct sw_flow_key *flow_key,
1946                         struct sw_flow_actions **sfa,
1947                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
1948 {
1949         const struct nlattr *ovs_key = nla_data(a);
1950         int key_type = nla_type(ovs_key);
1951         size_t key_len;
1952
1953         /* There can be only one key in a action */
1954         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1955                 return -EINVAL;
1956
1957         key_len = nla_len(ovs_key);
1958         if (masked)
1959                 key_len /= 2;
1960
1961         if (key_type > OVS_KEY_ATTR_MAX ||
1962             !check_attr_len(key_len, ovs_key_lens[key_type].len))
1963                 return -EINVAL;
1964
1965         if (masked && !validate_masked(nla_data(ovs_key), key_len))
1966                 return -EINVAL;
1967
1968         switch (key_type) {
1969         const struct ovs_key_ipv4 *ipv4_key;
1970         const struct ovs_key_ipv6 *ipv6_key;
1971         int err;
1972
1973         case OVS_KEY_ATTR_PRIORITY:
1974         case OVS_KEY_ATTR_SKB_MARK:
1975         case OVS_KEY_ATTR_CT_MARK:
1976         case OVS_KEY_ATTR_CT_LABEL:
1977         case OVS_KEY_ATTR_ETHERNET:
1978                 break;
1979
1980         case OVS_KEY_ATTR_TUNNEL:
1981                 if (eth_p_mpls(eth_type))
1982                         return -EINVAL;
1983
1984                 if (masked)
1985                         return -EINVAL; /* Masked tunnel set not supported. */
1986
1987                 *skip_copy = true;
1988                 err = validate_and_copy_set_tun(a, sfa, log);
1989                 if (err)
1990                         return err;
1991                 break;
1992
1993         case OVS_KEY_ATTR_IPV4:
1994                 if (eth_type != htons(ETH_P_IP))
1995                         return -EINVAL;
1996
1997                 ipv4_key = nla_data(ovs_key);
1998
1999                 if (masked) {
2000                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2001
2002                         /* Non-writeable fields. */
2003                         if (mask->ipv4_proto || mask->ipv4_frag)
2004                                 return -EINVAL;
2005                 } else {
2006                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2007                                 return -EINVAL;
2008
2009                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2010                                 return -EINVAL;
2011                 }
2012                 break;
2013
2014         case OVS_KEY_ATTR_IPV6:
2015                 if (eth_type != htons(ETH_P_IPV6))
2016                         return -EINVAL;
2017
2018                 ipv6_key = nla_data(ovs_key);
2019
2020                 if (masked) {
2021                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2022
2023                         /* Non-writeable fields. */
2024                         if (mask->ipv6_proto || mask->ipv6_frag)
2025                                 return -EINVAL;
2026
2027                         /* Invalid bits in the flow label mask? */
2028                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2029                                 return -EINVAL;
2030                 } else {
2031                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2032                                 return -EINVAL;
2033
2034                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2035                                 return -EINVAL;
2036                 }
2037                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2038                         return -EINVAL;
2039
2040                 break;
2041
2042         case OVS_KEY_ATTR_TCP:
2043                 if ((eth_type != htons(ETH_P_IP) &&
2044                      eth_type != htons(ETH_P_IPV6)) ||
2045                     flow_key->ip.proto != IPPROTO_TCP)
2046                         return -EINVAL;
2047
2048                 break;
2049
2050         case OVS_KEY_ATTR_UDP:
2051                 if ((eth_type != htons(ETH_P_IP) &&
2052                      eth_type != htons(ETH_P_IPV6)) ||
2053                     flow_key->ip.proto != IPPROTO_UDP)
2054                         return -EINVAL;
2055
2056                 break;
2057
2058         case OVS_KEY_ATTR_MPLS:
2059                 if (!eth_p_mpls(eth_type))
2060                         return -EINVAL;
2061                 break;
2062
2063         case OVS_KEY_ATTR_SCTP:
2064                 if ((eth_type != htons(ETH_P_IP) &&
2065                      eth_type != htons(ETH_P_IPV6)) ||
2066                     flow_key->ip.proto != IPPROTO_SCTP)
2067                         return -EINVAL;
2068
2069                 break;
2070
2071         default:
2072                 return -EINVAL;
2073         }
2074
2075         /* Convert non-masked non-tunnel set actions to masked set actions. */
2076         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2077                 int start, len = key_len * 2;
2078                 struct nlattr *at;
2079
2080                 *skip_copy = true;
2081
2082                 start = add_nested_action_start(sfa,
2083                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2084                                                 log);
2085                 if (start < 0)
2086                         return start;
2087
2088                 at = __add_action(sfa, key_type, NULL, len, log);
2089                 if (IS_ERR(at))
2090                         return PTR_ERR(at);
2091
2092                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2093                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2094                 /* Clear non-writeable bits from otherwise writeable fields. */
2095                 if (key_type == OVS_KEY_ATTR_IPV6) {
2096                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2097
2098                         mask->ipv6_label &= htonl(0x000FFFFF);
2099                 }
2100                 add_nested_action_end(*sfa, start);
2101         }
2102
2103         return 0;
2104 }
2105
2106 static int validate_userspace(const struct nlattr *attr)
2107 {
2108         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2109                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2110                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2111                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2112         };
2113         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2114         int error;
2115
2116         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2117                                  attr, userspace_policy);
2118         if (error)
2119                 return error;
2120
2121         if (!a[OVS_USERSPACE_ATTR_PID] ||
2122             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2123                 return -EINVAL;
2124
2125         return 0;
2126 }
2127
2128 static int copy_action(const struct nlattr *from,
2129                        struct sw_flow_actions **sfa, bool log)
2130 {
2131         int totlen = NLA_ALIGN(from->nla_len);
2132         struct nlattr *to;
2133
2134         to = reserve_sfa_size(sfa, from->nla_len, log);
2135         if (IS_ERR(to))
2136                 return PTR_ERR(to);
2137
2138         memcpy(to, from, totlen);
2139         return 0;
2140 }
2141
2142 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2143                                   const struct sw_flow_key *key,
2144                                   int depth, struct sw_flow_actions **sfa,
2145                                   __be16 eth_type, __be16 vlan_tci, bool log)
2146 {
2147         const struct nlattr *a;
2148         int rem, err;
2149
2150         if (depth >= SAMPLE_ACTION_DEPTH)
2151                 return -EOVERFLOW;
2152
2153         nla_for_each_nested(a, attr, rem) {
2154                 /* Expected argument lengths, (u32)-1 for variable length. */
2155                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2156                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2157                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2158                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2159                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2160                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2161                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2162                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2163                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2164                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2165                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2166                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2167                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2168                 };
2169                 const struct ovs_action_push_vlan *vlan;
2170                 int type = nla_type(a);
2171                 bool skip_copy;
2172
2173                 if (type > OVS_ACTION_ATTR_MAX ||
2174                     (action_lens[type] != nla_len(a) &&
2175                      action_lens[type] != (u32)-1))
2176                         return -EINVAL;
2177
2178                 skip_copy = false;
2179                 switch (type) {
2180                 case OVS_ACTION_ATTR_UNSPEC:
2181                         return -EINVAL;
2182
2183                 case OVS_ACTION_ATTR_USERSPACE:
2184                         err = validate_userspace(a);
2185                         if (err)
2186                                 return err;
2187                         break;
2188
2189                 case OVS_ACTION_ATTR_OUTPUT:
2190                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2191                                 return -EINVAL;
2192                         break;
2193
2194                 case OVS_ACTION_ATTR_HASH: {
2195                         const struct ovs_action_hash *act_hash = nla_data(a);
2196
2197                         switch (act_hash->hash_alg) {
2198                         case OVS_HASH_ALG_L4:
2199                                 break;
2200                         default:
2201                                 return  -EINVAL;
2202                         }
2203
2204                         break;
2205                 }
2206
2207                 case OVS_ACTION_ATTR_POP_VLAN:
2208                         vlan_tci = htons(0);
2209                         break;
2210
2211                 case OVS_ACTION_ATTR_PUSH_VLAN:
2212                         vlan = nla_data(a);
2213                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2214                                 return -EINVAL;
2215                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2216                                 return -EINVAL;
2217                         vlan_tci = vlan->vlan_tci;
2218                         break;
2219
2220                 case OVS_ACTION_ATTR_RECIRC:
2221                         break;
2222
2223                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2224                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2225
2226                         if (!eth_p_mpls(mpls->mpls_ethertype))
2227                                 return -EINVAL;
2228                         /* Prohibit push MPLS other than to a white list
2229                          * for packets that have a known tag order.
2230                          */
2231                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2232                             (eth_type != htons(ETH_P_IP) &&
2233                              eth_type != htons(ETH_P_IPV6) &&
2234                              eth_type != htons(ETH_P_ARP) &&
2235                              eth_type != htons(ETH_P_RARP) &&
2236                              !eth_p_mpls(eth_type)))
2237                                 return -EINVAL;
2238                         eth_type = mpls->mpls_ethertype;
2239                         break;
2240                 }
2241
2242                 case OVS_ACTION_ATTR_POP_MPLS:
2243                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2244                             !eth_p_mpls(eth_type))
2245                                 return -EINVAL;
2246
2247                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2248                          * as there is no check here to ensure that the new
2249                          * eth_type is valid and thus set actions could
2250                          * write off the end of the packet or otherwise
2251                          * corrupt it.
2252                          *
2253                          * Support for these actions is planned using packet
2254                          * recirculation.
2255                          */
2256                         eth_type = htons(0);
2257                         break;
2258
2259                 case OVS_ACTION_ATTR_SET:
2260                         err = validate_set(a, key, sfa,
2261                                            &skip_copy, eth_type, false, log);
2262                         if (err)
2263                                 return err;
2264                         break;
2265
2266                 case OVS_ACTION_ATTR_SET_MASKED:
2267                         err = validate_set(a, key, sfa,
2268                                            &skip_copy, eth_type, true, log);
2269                         if (err)
2270                                 return err;
2271                         break;
2272
2273                 case OVS_ACTION_ATTR_SAMPLE:
2274                         err = validate_and_copy_sample(net, a, key, depth, sfa,
2275                                                        eth_type, vlan_tci, log);
2276                         if (err)
2277                                 return err;
2278                         skip_copy = true;
2279                         break;
2280
2281                 case OVS_ACTION_ATTR_CT:
2282                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2283                         if (err)
2284                                 return err;
2285                         skip_copy = true;
2286                         break;
2287
2288                 default:
2289                         OVS_NLERR(log, "Unknown Action type %d", type);
2290                         return -EINVAL;
2291                 }
2292                 if (!skip_copy) {
2293                         err = copy_action(a, sfa, log);
2294                         if (err)
2295                                 return err;
2296                 }
2297         }
2298
2299         if (rem > 0)
2300                 return -EINVAL;
2301
2302         return 0;
2303 }
2304
2305 /* 'key' must be the masked key. */
2306 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2307                          const struct sw_flow_key *key,
2308                          struct sw_flow_actions **sfa, bool log)
2309 {
2310         int err;
2311
2312         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2313         if (IS_ERR(*sfa))
2314                 return PTR_ERR(*sfa);
2315
2316         (*sfa)->orig_len = nla_len(attr);
2317         err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2318                                      key->eth.tci, log);
2319         if (err)
2320                 ovs_nla_free_flow_actions(*sfa);
2321
2322         return err;
2323 }
2324
2325 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2326 {
2327         const struct nlattr *a;
2328         struct nlattr *start;
2329         int err = 0, rem;
2330
2331         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2332         if (!start)
2333                 return -EMSGSIZE;
2334
2335         nla_for_each_nested(a, attr, rem) {
2336                 int type = nla_type(a);
2337                 struct nlattr *st_sample;
2338
2339                 switch (type) {
2340                 case OVS_SAMPLE_ATTR_PROBABILITY:
2341                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2342                                     sizeof(u32), nla_data(a)))
2343                                 return -EMSGSIZE;
2344                         break;
2345                 case OVS_SAMPLE_ATTR_ACTIONS:
2346                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2347                         if (!st_sample)
2348                                 return -EMSGSIZE;
2349                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2350                         if (err)
2351                                 return err;
2352                         nla_nest_end(skb, st_sample);
2353                         break;
2354                 }
2355         }
2356
2357         nla_nest_end(skb, start);
2358         return err;
2359 }
2360
2361 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2362 {
2363         const struct nlattr *ovs_key = nla_data(a);
2364         int key_type = nla_type(ovs_key);
2365         struct nlattr *start;
2366         int err;
2367
2368         switch (key_type) {
2369         case OVS_KEY_ATTR_TUNNEL_INFO: {
2370                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2371                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2372
2373                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2374                 if (!start)
2375                         return -EMSGSIZE;
2376
2377                 err = ipv4_tun_to_nlattr(skb, &tun_info->key,
2378                                          tun_info->options_len ?
2379                                              ip_tunnel_info_opts(tun_info) : NULL,
2380                                          tun_info->options_len);
2381                 if (err)
2382                         return err;
2383                 nla_nest_end(skb, start);
2384                 break;
2385         }
2386         default:
2387                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2388                         return -EMSGSIZE;
2389                 break;
2390         }
2391
2392         return 0;
2393 }
2394
2395 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2396                                                 struct sk_buff *skb)
2397 {
2398         const struct nlattr *ovs_key = nla_data(a);
2399         struct nlattr *nla;
2400         size_t key_len = nla_len(ovs_key) / 2;
2401
2402         /* Revert the conversion we did from a non-masked set action to
2403          * masked set action.
2404          */
2405         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2406         if (!nla)
2407                 return -EMSGSIZE;
2408
2409         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2410                 return -EMSGSIZE;
2411
2412         nla_nest_end(skb, nla);
2413         return 0;
2414 }
2415
2416 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2417 {
2418         const struct nlattr *a;
2419         int rem, err;
2420
2421         nla_for_each_attr(a, attr, len, rem) {
2422                 int type = nla_type(a);
2423
2424                 switch (type) {
2425                 case OVS_ACTION_ATTR_SET:
2426                         err = set_action_to_attr(a, skb);
2427                         if (err)
2428                                 return err;
2429                         break;
2430
2431                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2432                         err = masked_set_action_to_set_action_attr(a, skb);
2433                         if (err)
2434                                 return err;
2435                         break;
2436
2437                 case OVS_ACTION_ATTR_SAMPLE:
2438                         err = sample_action_to_attr(a, skb);
2439                         if (err)
2440                                 return err;
2441                         break;
2442
2443                 case OVS_ACTION_ATTR_CT:
2444                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2445                         if (err)
2446                                 return err;
2447                         break;
2448
2449                 default:
2450                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2451                                 return -EMSGSIZE;
2452                         break;
2453                 }
2454         }
2455
2456         return 0;
2457 }