net/openvswitch/flow_netlink.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include "flow.h"
  22 #include "datapath.h"
  23 #include <linux/uaccess.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <net/llc_pdu.h>
  29 #include <linux/kernel.h>
  30 #include <linux/jhash.h>
  31 #include <linux/jiffies.h>
  32 #include <linux/llc.h>
  33 #include <linux/module.h>
  34 #include <linux/in.h>
  35 #include <linux/rcupdate.h>
  36 #include <linux/if_arp.h>
  37 #include <linux/ip.h>
  38 #include <linux/ipv6.h>
  39 #include <linux/sctp.h>
  40 #include <linux/tcp.h>
  41 #include <linux/udp.h>
  42 #include <linux/icmp.h>
  43 #include <linux/icmpv6.h>
  44 #include <linux/rculist.h>
  45 #include <net/geneve.h>
  46 #include <net/ip.h>
  47 #include <net/ipv6.h>
  48 #include <net/ndisc.h>
  49 #include <net/mpls.h>
  50 #include <net/vxlan.h>
  51
  52 #include "flow_netlink.h"
  53
  54 struct ovs_len_tbl {
  55         int len;
  56         const struct ovs_len_tbl *next;
  57 };
  58
  59 #define OVS_ATTR_NESTED -1
  60
  61 static void update_range(struct sw_flow_match *match,
  62                          size_t offset, size_t size, bool is_mask)
  63 {
  64         struct sw_flow_key_range *range;
  65         size_t start = rounddown(offset, sizeof(long));
  66         size_t end = roundup(offset + size, sizeof(long));
  67
  68         if (!is_mask)
  69                 range = &match->range;
  70         else
  71                 range = &match->mask->range;
  72
  73         if (range->start == range->end) {
  74                 range->start = start;
  75                 range->end = end;
  76                 return;
  77         }
  78
  79         if (range->start > start)
  80                 range->start = start;
  81
  82         if (range->end < end)
  83                 range->end = end;
  84 }
  85
  86 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  87         do { \
  88                 update_range(match, offsetof(struct sw_flow_key, field),    \
  89                              sizeof((match)->key->field), is_mask);         \
  90                 if (is_mask)                                                \
  91                         (match)->mask->key.field = value;                   \
  92                 else                                                        \
  93                         (match)->key->field = value;                        \
  94         } while (0)
  95
  96 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
  97         do {                                                                \
  98                 update_range(match, offset, len, is_mask);                  \
  99                 if (is_mask)                                                \
 100                         memcpy((u8 *)&(match)->mask->key + offset, value_p, \
 101                                len);                                       \
 102                 else                                                        \
 103                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
 104         } while (0)
 105
 106 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
 107         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 108                                   value_p, len, is_mask)
 109
 110 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 111         do {                                                                \
 112                 update_range(match, offsetof(struct sw_flow_key, field),    \
 113                              sizeof((match)->key->field), is_mask);         \
 114                 if (is_mask)                                                \
 115                         memset((u8 *)&(match)->mask->key.field, value,      \
 116                                sizeof((match)->mask->key.field));           \
 117                 else                                                        \
 118                         memset((u8 *)&(match)->key->field, value,           \
 119                                sizeof((match)->key->field));                \
 120         } while (0)
 121
 122 static bool match_validate(const struct sw_flow_match *match,
 123                            u64 key_attrs, u64 mask_attrs, bool log)
 124 {
 125         u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
 126         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 127
 128         /* The following mask attributes allowed only if they
 129          * pass the validation tests. */
 130         mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
 131                         | (1 << OVS_KEY_ATTR_IPV6)
 132                         | (1 << OVS_KEY_ATTR_TCP)
 133                         | (1 << OVS_KEY_ATTR_TCP_FLAGS)
 134                         | (1 << OVS_KEY_ATTR_UDP)
 135                         | (1 << OVS_KEY_ATTR_SCTP)
 136                         | (1 << OVS_KEY_ATTR_ICMP)
 137                         | (1 << OVS_KEY_ATTR_ICMPV6)
 138                         | (1 << OVS_KEY_ATTR_ARP)
 139                         | (1 << OVS_KEY_ATTR_ND)
 140                         | (1 << OVS_KEY_ATTR_MPLS));
 141
 142         /* Always allowed mask fields. */
 143         mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
 144                        | (1 << OVS_KEY_ATTR_IN_PORT)
 145                        | (1 << OVS_KEY_ATTR_ETHERTYPE));
 146
 147         /* Check key attributes. */
 148         if (match->key->eth.type == htons(ETH_P_ARP)
 149                         || match->key->eth.type == htons(ETH_P_RARP)) {
 150                 key_expected |= 1 << OVS_KEY_ATTR_ARP;
 151                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 152                         mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
 153         }
 154
 155         if (eth_p_mpls(match->key->eth.type)) {
 156                 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
 157                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 158                         mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
 159         }
 160
 161         if (match->key->eth.type == htons(ETH_P_IP)) {
 162                 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
 163                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 164                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
 165
 166                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 167                         if (match->key->ip.proto == IPPROTO_UDP) {
 168                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 169                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 170                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 171                         }
 172
 173                         if (match->key->ip.proto == IPPROTO_SCTP) {
 174                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 175                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 176                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 177                         }
 178
 179                         if (match->key->ip.proto == IPPROTO_TCP) {
 180                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 181                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 182                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 183                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 184                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 185                                 }
 186                         }
 187
 188                         if (match->key->ip.proto == IPPROTO_ICMP) {
 189                                 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
 190                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 191                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
 192                         }
 193                 }
 194         }
 195
 196         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 197                 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
 198                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 199                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
 200
 201                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 202                         if (match->key->ip.proto == IPPROTO_UDP) {
 203                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 204                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 205                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 206                         }
 207
 208                         if (match->key->ip.proto == IPPROTO_SCTP) {
 209                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 210                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 211                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 212                         }
 213
 214                         if (match->key->ip.proto == IPPROTO_TCP) {
 215                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 216                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 217                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 218                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 219                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 220                                 }
 221                         }
 222
 223                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 224                                 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
 225                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 226                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
 227
 228                                 if (match->key->tp.src ==
 229                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 230                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 231                                         key_expected |= 1 << OVS_KEY_ATTR_ND;
 232                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 233                                                 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
 234                                 }
 235                         }
 236                 }
 237         }
 238
 239         if ((key_attrs & key_expected) != key_expected) {
 240                 /* Key attributes check failed. */
 241                 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
 242                           (unsigned long long)key_attrs,
 243                           (unsigned long long)key_expected);
 244                 return false;
 245         }
 246
 247         if ((mask_attrs & mask_allowed) != mask_attrs) {
 248                 /* Mask attributes check failed. */
 249                 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
 250                           (unsigned long long)mask_attrs,
 251                           (unsigned long long)mask_allowed);
 252                 return false;
 253         }
 254
 255         return true;
 256 }
 257
 258 size_t ovs_tun_key_attr_size(void)
 259 {
 260         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 261          * updating this function.
 262          */
 263         return    nla_total_size(8)    /* OVS_TUNNEL_KEY_ATTR_ID */
 264                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 265                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 266                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
 267                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
 268                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 269                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
 270                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
 271                 + nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 272                 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
 273                  * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
 274                  */
 275                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
 276                 + nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
 277 }
 278
 279 size_t ovs_key_attr_size(void)
 280 {
 281         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 282          * updating this function.
 283          */
 284         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 26);
 285
 286         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 287                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 288                   + ovs_tun_key_attr_size()
 289                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 290                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 291                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 292                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 293                 + nla_total_size(1)   /* OVS_KEY_ATTR_CT_STATE */
 294                 + nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
 295                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
 296                 + nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABEL */
 297                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 298                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 299                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 300                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 301                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 302                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 303                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 304                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 305 }
 306
 307 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 308         [OVS_TUNNEL_KEY_ATTR_ID]            = { .len = sizeof(u64) },
 309         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC]      = { .len = sizeof(u32) },
 310         [OVS_TUNNEL_KEY_ATTR_IPV4_DST]      = { .len = sizeof(u32) },
 311         [OVS_TUNNEL_KEY_ATTR_TOS]           = { .len = 1 },
 312         [OVS_TUNNEL_KEY_ATTR_TTL]           = { .len = 1 },
 313         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
 314         [OVS_TUNNEL_KEY_ATTR_CSUM]          = { .len = 0 },
 315         [OVS_TUNNEL_KEY_ATTR_TP_SRC]        = { .len = sizeof(u16) },
 316         [OVS_TUNNEL_KEY_ATTR_TP_DST]        = { .len = sizeof(u16) },
 317         [OVS_TUNNEL_KEY_ATTR_OAM]           = { .len = 0 },
 318         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_NESTED },
 319         [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED },
 320 };
 321
 322 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 323 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 324         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 325         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 326         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 327         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 328         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 329         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 330         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 331         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 332         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 333         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 334         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 335         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 336         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 337         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 338         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 339         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 340         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 341         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 342         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 343         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 344                                      .next = ovs_tunnel_key_lens, },
 345         [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 346         [OVS_KEY_ATTR_CT_STATE]  = { .len = sizeof(u8) },
 347         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 348         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 349         [OVS_KEY_ATTR_CT_LABEL]  = { .len = sizeof(struct ovs_key_ct_label) },
 350 };
 351
 352 static bool is_all_zero(const u8 *fp, size_t size)
 353 {
 354         int i;
 355
 356         if (!fp)
 357                 return false;
 358
 359         for (i = 0; i < size; i++)
 360                 if (fp[i])
 361                         return false;
 362
 363         return true;
 364 }
 365
 366 static int __parse_flow_nlattrs(const struct nlattr *attr,
 367                                 const struct nlattr *a[],
 368                                 u64 *attrsp, bool log, bool nz)
 369 {
 370         const struct nlattr *nla;
 371         u64 attrs;
 372         int rem;
 373
 374         attrs = *attrsp;
 375         nla_for_each_nested(nla, attr, rem) {
 376                 u16 type = nla_type(nla);
 377                 int expected_len;
 378
 379                 if (type > OVS_KEY_ATTR_MAX) {
 380                         OVS_NLERR(log, "Key type %d is out of range max %d",
 381                                   type, OVS_KEY_ATTR_MAX);
 382                         return -EINVAL;
 383                 }
 384
 385                 if (attrs & (1 << type)) {
 386                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 387                         return -EINVAL;
 388                 }
 389
 390                 expected_len = ovs_key_lens[type].len;
 391                 if (nla_len(nla) != expected_len && expected_len != OVS_ATTR_NESTED) {
 392                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 393                                   type, nla_len(nla), expected_len);
 394                         return -EINVAL;
 395                 }
 396
 397                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 398                         attrs |= 1 << type;
 399                         a[type] = nla;
 400                 }
 401         }
 402         if (rem) {
 403                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 404                 return -EINVAL;
 405         }
 406
 407         *attrsp = attrs;
 408         return 0;
 409 }
 410
 411 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 412                                    const struct nlattr *a[], u64 *attrsp,
 413                                    bool log)
 414 {
 415         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 416 }
 417
 418 static int parse_flow_nlattrs(const struct nlattr *attr,
 419                               const struct nlattr *a[], u64 *attrsp,
 420                               bool log)
 421 {
 422         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 423 }
 424
 425 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 426                                      struct sw_flow_match *match, bool is_mask,
 427                                      bool log)
 428 {
 429         unsigned long opt_key_offset;
 430
 431         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 432                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 433                           nla_len(a), sizeof(match->key->tun_opts));
 434                 return -EINVAL;
 435         }
 436
 437         if (nla_len(a) % 4 != 0) {
 438                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 439                           nla_len(a));
 440                 return -EINVAL;
 441         }
 442
 443         /* We need to record the length of the options passed
 444          * down, otherwise packets with the same format but
 445          * additional options will be silently matched.
 446          */
 447         if (!is_mask) {
 448                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 449                                 false);
 450         } else {
 451                 /* This is somewhat unusual because it looks at
 452                  * both the key and mask while parsing the
 453                  * attributes (and by extension assumes the key
 454                  * is parsed first). Normally, we would verify
 455                  * that each is the correct length and that the
 456                  * attributes line up in the validate function.
 457                  * However, that is difficult because this is
 458                  * variable length and we won't have the
 459                  * information later.
 460                  */
 461                 if (match->key->tun_opts_len != nla_len(a)) {
 462                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 463                                   match->key->tun_opts_len, nla_len(a));
 464                         return -EINVAL;
 465                 }
 466
 467                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 468         }
 469
 470         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 471         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 472                                   nla_len(a), is_mask);
 473         return 0;
 474 }
 475
 476 static const struct nla_policy vxlan_opt_policy[OVS_VXLAN_EXT_MAX + 1] = {
 477         [OVS_VXLAN_EXT_GBP]     = { .type = NLA_U32 },
 478 };
 479
 480 static int vxlan_tun_opt_from_nlattr(const struct nlattr *a,
 481                                      struct sw_flow_match *match, bool is_mask,
 482                                      bool log)
 483 {
 484         struct nlattr *tb[OVS_VXLAN_EXT_MAX+1];
 485         unsigned long opt_key_offset;
 486         struct vxlan_metadata opts;
 487         int err;
 488
 489         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 490
 491         err = nla_parse_nested(tb, OVS_VXLAN_EXT_MAX, a, vxlan_opt_policy);
 492         if (err < 0)
 493                 return err;
 494
 495         memset(&opts, 0, sizeof(opts));
 496
 497         if (tb[OVS_VXLAN_EXT_GBP])
 498                 opts.gbp = nla_get_u32(tb[OVS_VXLAN_EXT_GBP]);
 499
 500         if (!is_mask)
 501                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 502         else
 503                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 504
 505         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 506         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 507                                   is_mask);
 508         return 0;
 509 }
 510
 511 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
 512                                 struct sw_flow_match *match, bool is_mask,
 513                                 bool log)
 514 {
 515         struct nlattr *a;
 516         int rem;
 517         bool ttl = false;
 518         __be16 tun_flags = 0;
 519         int opts_type = 0;
 520
 521         nla_for_each_nested(a, attr, rem) {
 522                 int type = nla_type(a);
 523                 int err;
 524
 525                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 526                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 527                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 528                         return -EINVAL;
 529                 }
 530
 531                 if (ovs_tunnel_key_lens[type].len != nla_len(a) &&
 532                     ovs_tunnel_key_lens[type].len != OVS_ATTR_NESTED) {
 533                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 534                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 535                         return -EINVAL;
 536                 }
 537
 538                 switch (type) {
 539                 case OVS_TUNNEL_KEY_ATTR_ID:
 540                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 541                                         nla_get_be64(a), is_mask);
 542                         tun_flags |= TUNNEL_KEY;
 543                         break;
 544                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 545                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
 546                                         nla_get_in_addr(a), is_mask);
 547                         break;
 548                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 549                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
 550                                         nla_get_in_addr(a), is_mask);
 551                         break;
 552                 case OVS_TUNNEL_KEY_ATTR_TOS:
 553                         SW_FLOW_KEY_PUT(match, tun_key.tos,
 554                                         nla_get_u8(a), is_mask);
 555                         break;
 556                 case OVS_TUNNEL_KEY_ATTR_TTL:
 557                         SW_FLOW_KEY_PUT(match, tun_key.ttl,
 558                                         nla_get_u8(a), is_mask);
 559                         ttl = true;
 560                         break;
 561                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 562                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 563                         break;
 564                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 565                         tun_flags |= TUNNEL_CSUM;
 566                         break;
 567                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 568                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 569                                         nla_get_be16(a), is_mask);
 570                         break;
 571                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 572                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 573                                         nla_get_be16(a), is_mask);
 574                         break;
 575                 case OVS_TUNNEL_KEY_ATTR_OAM:
 576                         tun_flags |= TUNNEL_OAM;
 577                         break;
 578                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 579                         if (opts_type) {
 580                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 581                                 return -EINVAL;
 582                         }
 583
 584                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 585                         if (err)
 586                                 return err;
 587
 588                         tun_flags |= TUNNEL_GENEVE_OPT;
 589                         opts_type = type;
 590                         break;
 591                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 592                         if (opts_type) {
 593                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 594                                 return -EINVAL;
 595                         }
 596
 597                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 598                         if (err)
 599                                 return err;
 600
 601                         tun_flags |= TUNNEL_VXLAN_OPT;
 602                         opts_type = type;
 603                         break;
 604                 default:
 605                         OVS_NLERR(log, "Unknown IPv4 tunnel attribute %d",
 606                                   type);
 607                         return -EINVAL;
 608                 }
 609         }
 610
 611         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 612
 613         if (rem > 0) {
 614                 OVS_NLERR(log, "IPv4 tunnel attribute has %d unknown bytes.",
 615                           rem);
 616                 return -EINVAL;
 617         }
 618
 619         if (!is_mask) {
 620                 if (!match->key->tun_key.u.ipv4.dst) {
 621                         OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 622                         return -EINVAL;
 623                 }
 624
 625                 if (!ttl) {
 626                         OVS_NLERR(log, "IPv4 tunnel TTL not specified.");
 627                         return -EINVAL;
 628                 }
 629         }
 630
 631         return opts_type;
 632 }
 633
 634 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 635                                const void *tun_opts, int swkey_tun_opts_len)
 636 {
 637         const struct vxlan_metadata *opts = tun_opts;
 638         struct nlattr *nla;
 639
 640         nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 641         if (!nla)
 642                 return -EMSGSIZE;
 643
 644         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 645                 return -EMSGSIZE;
 646
 647         nla_nest_end(skb, nla);
 648         return 0;
 649 }
 650
 651 static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
 652                                 const struct ip_tunnel_key *output,
 653                                 const void *tun_opts, int swkey_tun_opts_len)
 654 {
 655         if (output->tun_flags & TUNNEL_KEY &&
 656             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 657                 return -EMSGSIZE;
 658         if (output->u.ipv4.src &&
 659             nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 660                             output->u.ipv4.src))
 661                 return -EMSGSIZE;
 662         if (output->u.ipv4.dst &&
 663             nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 664                             output->u.ipv4.dst))
 665                 return -EMSGSIZE;
 666         if (output->tos &&
 667             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 668                 return -EMSGSIZE;
 669         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 670                 return -EMSGSIZE;
 671         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 672             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 673                 return -EMSGSIZE;
 674         if ((output->tun_flags & TUNNEL_CSUM) &&
 675             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 676                 return -EMSGSIZE;
 677         if (output->tp_src &&
 678             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 679                 return -EMSGSIZE;
 680         if (output->tp_dst &&
 681             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 682                 return -EMSGSIZE;
 683         if ((output->tun_flags & TUNNEL_OAM) &&
 684             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 685                 return -EMSGSIZE;
 686         if (tun_opts) {
 687                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 688                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 689                             swkey_tun_opts_len, tun_opts))
 690                         return -EMSGSIZE;
 691                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 692                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 693                         return -EMSGSIZE;
 694         }
 695
 696         return 0;
 697 }
 698
 699 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
 700                               const struct ip_tunnel_key *output,
 701                               const void *tun_opts, int swkey_tun_opts_len)
 702 {
 703         struct nlattr *nla;
 704         int err;
 705
 706         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 707         if (!nla)
 708                 return -EMSGSIZE;
 709
 710         err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
 711         if (err)
 712                 return err;
 713
 714         nla_nest_end(skb, nla);
 715         return 0;
 716 }
 717
 718 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
 719                                   const struct ip_tunnel_info *egress_tun_info)
 720 {
 721         return __ipv4_tun_to_nlattr(skb, &egress_tun_info->key,
 722                                     egress_tun_info->options,
 723                                     egress_tun_info->options_len);
 724 }
 725
 726 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
 727                                  u64 *attrs, const struct nlattr **a,
 728                                  bool is_mask, bool log)
 729 {
 730         if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
 731                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 732
 733                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 734                 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
 735         }
 736
 737         if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
 738                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 739
 740                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 741                 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
 742         }
 743
 744         if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 745                 SW_FLOW_KEY_PUT(match, phy.priority,
 746                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 747                 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 748         }
 749
 750         if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 751                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 752
 753                 if (is_mask) {
 754                         in_port = 0xffffffff; /* Always exact match in_port. */
 755                 } else if (in_port >= DP_MAX_PORTS) {
 756                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 757                                   in_port, DP_MAX_PORTS);
 758                         return -EINVAL;
 759                 }
 760
 761                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 762                 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
 763         } else if (!is_mask) {
 764                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 765         }
 766
 767         if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
 768                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 769
 770                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 771                 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
 772         }
 773         if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
 774                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 775                                          is_mask, log) < 0)
 776                         return -EINVAL;
 777                 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
 778         }
 779
 780         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
 781             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
 782                 u8 ct_state = nla_get_u8(a[OVS_KEY_ATTR_CT_STATE]);
 783
 784                 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
 785                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
 786         }
 787         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
 788             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
 789                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
 790
 791                 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
 792                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
 793         }
 794         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
 795             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
 796                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
 797
 798                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
 799                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
 800         }
 801         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABEL) &&
 802             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABEL)) {
 803                 const struct ovs_key_ct_label *cl;
 804
 805                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABEL]);
 806                 SW_FLOW_KEY_MEMCPY(match, ct.label, cl->ct_label,
 807                                    sizeof(*cl), is_mask);
 808                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABEL);
 809         }
 810         return 0;
 811 }
 812
 813 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
 814                                 u64 attrs, const struct nlattr **a,
 815                                 bool is_mask, bool log)
 816 {
 817         int err;
 818
 819         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
 820         if (err)
 821                 return err;
 822
 823         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
 824                 const struct ovs_key_ethernet *eth_key;
 825
 826                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 827                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 828                                 eth_key->eth_src, ETH_ALEN, is_mask);
 829                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 830                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 831                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
 832         }
 833
 834         if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
 835                 __be16 tci;
 836
 837                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 838                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 839                         if (is_mask)
 840                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 841                         else
 842                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 843
 844                         return -EINVAL;
 845                 }
 846
 847                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 848                 attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
 849         }
 850
 851         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
 852                 __be16 eth_type;
 853
 854                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 855                 if (is_mask) {
 856                         /* Always exact match EtherType. */
 857                         eth_type = htons(0xffff);
 858                 } else if (!eth_proto_is_802_3(eth_type)) {
 859                         OVS_NLERR(log, "EtherType %x is less than min %x",
 860                                   ntohs(eth_type), ETH_P_802_3_MIN);
 861                         return -EINVAL;
 862                 }
 863
 864                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 865                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
 866         } else if (!is_mask) {
 867                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 868         }
 869
 870         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
 871                 const struct ovs_key_ipv4 *ipv4_key;
 872
 873                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 874                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 875                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 876                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 877                         return -EINVAL;
 878                 }
 879                 SW_FLOW_KEY_PUT(match, ip.proto,
 880                                 ipv4_key->ipv4_proto, is_mask);
 881                 SW_FLOW_KEY_PUT(match, ip.tos,
 882                                 ipv4_key->ipv4_tos, is_mask);
 883                 SW_FLOW_KEY_PUT(match, ip.ttl,
 884                                 ipv4_key->ipv4_ttl, is_mask);
 885                 SW_FLOW_KEY_PUT(match, ip.frag,
 886                                 ipv4_key->ipv4_frag, is_mask);
 887                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 888                                 ipv4_key->ipv4_src, is_mask);
 889                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 890                                 ipv4_key->ipv4_dst, is_mask);
 891                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
 892         }
 893
 894         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
 895                 const struct ovs_key_ipv6 *ipv6_key;
 896
 897                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 898                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 899                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 900                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 901                         return -EINVAL;
 902                 }
 903
 904                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 905                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
 906                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 907                         return -EINVAL;
 908                 }
 909
 910                 SW_FLOW_KEY_PUT(match, ipv6.label,
 911                                 ipv6_key->ipv6_label, is_mask);
 912                 SW_FLOW_KEY_PUT(match, ip.proto,
 913                                 ipv6_key->ipv6_proto, is_mask);
 914                 SW_FLOW_KEY_PUT(match, ip.tos,
 915                                 ipv6_key->ipv6_tclass, is_mask);
 916                 SW_FLOW_KEY_PUT(match, ip.ttl,
 917                                 ipv6_key->ipv6_hlimit, is_mask);
 918                 SW_FLOW_KEY_PUT(match, ip.frag,
 919                                 ipv6_key->ipv6_frag, is_mask);
 920                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
 921                                 ipv6_key->ipv6_src,
 922                                 sizeof(match->key->ipv6.addr.src),
 923                                 is_mask);
 924                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
 925                                 ipv6_key->ipv6_dst,
 926                                 sizeof(match->key->ipv6.addr.dst),
 927                                 is_mask);
 928
 929                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
 930         }
 931
 932         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
 933                 const struct ovs_key_arp *arp_key;
 934
 935                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
 936                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
 937                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
 938                                   arp_key->arp_op);
 939                         return -EINVAL;
 940                 }
 941
 942                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 943                                 arp_key->arp_sip, is_mask);
 944                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 945                         arp_key->arp_tip, is_mask);
 946                 SW_FLOW_KEY_PUT(match, ip.proto,
 947                                 ntohs(arp_key->arp_op), is_mask);
 948                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
 949                                 arp_key->arp_sha, ETH_ALEN, is_mask);
 950                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
 951                                 arp_key->arp_tha, ETH_ALEN, is_mask);
 952
 953                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
 954         }
 955
 956         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
 957                 const struct ovs_key_mpls *mpls_key;
 958
 959                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
 960                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
 961                                 mpls_key->mpls_lse, is_mask);
 962
 963                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
 964          }
 965
 966         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
 967                 const struct ovs_key_tcp *tcp_key;
 968
 969                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 970                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
 971                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
 972                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 973         }
 974
 975         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
 976                 SW_FLOW_KEY_PUT(match, tp.flags,
 977                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
 978                                 is_mask);
 979                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
 980         }
 981
 982         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
 983                 const struct ovs_key_udp *udp_key;
 984
 985                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 986                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
 987                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
 988                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 989         }
 990
 991         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
 992                 const struct ovs_key_sctp *sctp_key;
 993
 994                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
 995                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
 996                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
 997                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
 998         }
 999
1000         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1001                 const struct ovs_key_icmp *icmp_key;
1002
1003                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1004                 SW_FLOW_KEY_PUT(match, tp.src,
1005                                 htons(icmp_key->icmp_type), is_mask);
1006                 SW_FLOW_KEY_PUT(match, tp.dst,
1007                                 htons(icmp_key->icmp_code), is_mask);
1008                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1009         }
1010
1011         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1012                 const struct ovs_key_icmpv6 *icmpv6_key;
1013
1014                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1015                 SW_FLOW_KEY_PUT(match, tp.src,
1016                                 htons(icmpv6_key->icmpv6_type), is_mask);
1017                 SW_FLOW_KEY_PUT(match, tp.dst,
1018                                 htons(icmpv6_key->icmpv6_code), is_mask);
1019                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1020         }
1021
1022         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1023                 const struct ovs_key_nd *nd_key;
1024
1025                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1026                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1027                         nd_key->nd_target,
1028                         sizeof(match->key->ipv6.nd.target),
1029                         is_mask);
1030                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1031                         nd_key->nd_sll, ETH_ALEN, is_mask);
1032                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1033                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1034                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1035         }
1036
1037         if (attrs != 0) {
1038                 OVS_NLERR(log, "Unknown key attributes %llx",
1039                           (unsigned long long)attrs);
1040                 return -EINVAL;
1041         }
1042
1043         return 0;
1044 }
1045
1046 static void nlattr_set(struct nlattr *attr, u8 val,
1047                        const struct ovs_len_tbl *tbl)
1048 {
1049         struct nlattr *nla;
1050         int rem;
1051
1052         /* The nlattr stream should already have been validated */
1053         nla_for_each_nested(nla, attr, rem) {
1054                 if (tbl && tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1055                         nlattr_set(nla, val, tbl[nla_type(nla)].next);
1056                 else
1057                         memset(nla_data(nla), val, nla_len(nla));
1058         }
1059 }
1060
1061 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1062 {
1063         nlattr_set(attr, val, ovs_key_lens);
1064 }
1065
1066 /**
1067  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1068  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1069  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1070  * does not include any don't care bit.
1071  * @net: Used to determine per-namespace field support.
1072  * @match: receives the extracted flow match information.
1073  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1074  * sequence. The fields should of the packet that triggered the creation
1075  * of this flow.
1076  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1077  * attribute specifies the mask field of the wildcarded flow.
1078  * @log: Boolean to allow kernel error logging.  Normally true, but when
1079  * probing for feature compatibility this should be passed in as false to
1080  * suppress unnecessary error logging.
1081  */
1082 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1083                       const struct nlattr *nla_key,
1084                       const struct nlattr *nla_mask,
1085                       bool log)
1086 {
1087         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1088         const struct nlattr *encap;
1089         struct nlattr *newmask = NULL;
1090         u64 key_attrs = 0;
1091         u64 mask_attrs = 0;
1092         bool encap_valid = false;
1093         int err;
1094
1095         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1096         if (err)
1097                 return err;
1098
1099         if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1100             (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1101             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1102                 __be16 tci;
1103
1104                 if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1105                       (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1106                         OVS_NLERR(log, "Invalid Vlan frame.");
1107                         return -EINVAL;
1108                 }
1109
1110                 key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1111                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1112                 encap = a[OVS_KEY_ATTR_ENCAP];
1113                 key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1114                 encap_valid = true;
1115
1116                 if (tci & htons(VLAN_TAG_PRESENT)) {
1117                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1118                         if (err)
1119                                 return err;
1120                 } else if (!tci) {
1121                         /* Corner case for truncated 802.1Q header. */
1122                         if (nla_len(encap)) {
1123                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1124                                 return -EINVAL;
1125                         }
1126                 } else {
1127                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1128                         return  -EINVAL;
1129                 }
1130         }
1131
1132         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1133         if (err)
1134                 return err;
1135
1136         if (match->mask) {
1137                 if (!nla_mask) {
1138                         /* Create an exact match mask. We need to set to 0xff
1139                          * all the 'match->mask' fields that have been touched
1140                          * in 'match->key'. We cannot simply memset
1141                          * 'match->mask', because padding bytes and fields not
1142                          * specified in 'match->key' should be left to 0.
1143                          * Instead, we use a stream of netlink attributes,
1144                          * copied from 'key' and set to 0xff.
1145                          * ovs_key_from_nlattrs() will take care of filling
1146                          * 'match->mask' appropriately.
1147                          */
1148                         newmask = kmemdup(nla_key,
1149                                           nla_total_size(nla_len(nla_key)),
1150                                           GFP_KERNEL);
1151                         if (!newmask)
1152                                 return -ENOMEM;
1153
1154                         mask_set_nlattr(newmask, 0xff);
1155
1156                         /* The userspace does not send tunnel attributes that
1157                          * are 0, but we should not wildcard them nonetheless.
1158                          */
1159                         if (match->key->tun_key.u.ipv4.dst)
1160                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1161                                                          0xff, true);
1162
1163                         nla_mask = newmask;
1164                 }
1165
1166                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1167                 if (err)
1168                         goto free_newmask;
1169
1170                 /* Always match on tci. */
1171                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1172
1173                 if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
1174                         __be16 eth_type = 0;
1175                         __be16 tci = 0;
1176
1177                         if (!encap_valid) {
1178                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1179                                 err = -EINVAL;
1180                                 goto free_newmask;
1181                         }
1182
1183                         mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1184                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1185                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1186
1187                         if (eth_type == htons(0xffff)) {
1188                                 mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1189                                 encap = a[OVS_KEY_ATTR_ENCAP];
1190                                 err = parse_flow_mask_nlattrs(encap, a,
1191                                                               &mask_attrs, log);
1192                                 if (err)
1193                                         goto free_newmask;
1194                         } else {
1195                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1196                                           ntohs(eth_type));
1197                                 err = -EINVAL;
1198                                 goto free_newmask;
1199                         }
1200
1201                         if (a[OVS_KEY_ATTR_VLAN])
1202                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1203
1204                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1205                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1206                                           ntohs(tci));
1207                                 err = -EINVAL;
1208                                 goto free_newmask;
1209                         }
1210                 }
1211
1212                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1213                                            log);
1214                 if (err)
1215                         goto free_newmask;
1216         }
1217
1218         if (!match_validate(match, key_attrs, mask_attrs, log))
1219                 err = -EINVAL;
1220
1221 free_newmask:
1222         kfree(newmask);
1223         return err;
1224 }
1225
1226 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1227 {
1228         size_t len;
1229
1230         if (!attr)
1231                 return 0;
1232
1233         len = nla_len(attr);
1234         if (len < 1 || len > MAX_UFID_LENGTH) {
1235                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1236                           nla_len(attr), MAX_UFID_LENGTH);
1237                 return 0;
1238         }
1239
1240         return len;
1241 }
1242
1243 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1244  * or false otherwise.
1245  */
1246 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1247                       bool log)
1248 {
1249         sfid->ufid_len = get_ufid_len(attr, log);
1250         if (sfid->ufid_len)
1251                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1252
1253         return sfid->ufid_len;
1254 }
1255
1256 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1257                            const struct sw_flow_key *key, bool log)
1258 {
1259         struct sw_flow_key *new_key;
1260
1261         if (ovs_nla_get_ufid(sfid, ufid, log))
1262                 return 0;
1263
1264         /* If UFID was not provided, use unmasked key. */
1265         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1266         if (!new_key)
1267                 return -ENOMEM;
1268         memcpy(new_key, key, sizeof(*key));
1269         sfid->unmasked_key = new_key;
1270
1271         return 0;
1272 }
1273
1274 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1275 {
1276         return attr ? nla_get_u32(attr) : 0;
1277 }
1278
1279 /**
1280  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1281  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1282  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1283  * sequence.
1284  * @log: Boolean to allow kernel error logging.  Normally true, but when
1285  * probing for feature compatibility this should be passed in as false to
1286  * suppress unnecessary error logging.
1287  *
1288  * This parses a series of Netlink attributes that form a flow key, which must
1289  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1290  * get the metadata, that is, the parts of the flow key that cannot be
1291  * extracted from the packet itself.
1292  */
1293
1294 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1295                               struct sw_flow_key *key,
1296                               bool log)
1297 {
1298         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1299         struct sw_flow_match match;
1300         u64 attrs = 0;
1301         int err;
1302
1303         err = parse_flow_nlattrs(attr, a, &attrs, log);
1304         if (err)
1305                 return -EINVAL;
1306
1307         memset(&match, 0, sizeof(match));
1308         match.key = key;
1309
1310         memset(&key->ct, 0, sizeof(key->ct));
1311         key->phy.in_port = DP_MAX_PORTS;
1312
1313         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1314 }
1315
1316 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1317                              const struct sw_flow_key *output, bool is_mask,
1318                              struct sk_buff *skb)
1319 {
1320         struct ovs_key_ethernet *eth_key;
1321         struct nlattr *nla, *encap;
1322
1323         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1324                 goto nla_put_failure;
1325
1326         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1327                 goto nla_put_failure;
1328
1329         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1330                 goto nla_put_failure;
1331
1332         if ((swkey->tun_key.u.ipv4.dst || is_mask)) {
1333                 const void *opts = NULL;
1334
1335                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1336                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1337
1338                 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1339                                        swkey->tun_opts_len))
1340                         goto nla_put_failure;
1341         }
1342
1343         if (swkey->phy.in_port == DP_MAX_PORTS) {
1344                 if (is_mask && (output->phy.in_port == 0xffff))
1345                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1346                                 goto nla_put_failure;
1347         } else {
1348                 u16 upper_u16;
1349                 upper_u16 = !is_mask ? 0 : 0xffff;
1350
1351                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1352                                 (upper_u16 << 16) | output->phy.in_port))
1353                         goto nla_put_failure;
1354         }
1355
1356         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1357                 goto nla_put_failure;
1358
1359         if (ovs_ct_put_key(output, skb))
1360                 goto nla_put_failure;
1361
1362         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1363         if (!nla)
1364                 goto nla_put_failure;
1365
1366         eth_key = nla_data(nla);
1367         ether_addr_copy(eth_key->eth_src, output->eth.src);
1368         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1369
1370         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1371                 __be16 eth_type;
1372                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1373                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1374                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1375                         goto nla_put_failure;
1376                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1377                 if (!swkey->eth.tci)
1378                         goto unencap;
1379         } else
1380                 encap = NULL;
1381
1382         if (swkey->eth.type == htons(ETH_P_802_2)) {
1383                 /*
1384                  * Ethertype 802.2 is represented in the netlink with omitted
1385                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1386                  * 0xffff in the mask attribute.  Ethertype can also
1387                  * be wildcarded.
1388                  */
1389                 if (is_mask && output->eth.type)
1390                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1391                                                 output->eth.type))
1392                                 goto nla_put_failure;
1393                 goto unencap;
1394         }
1395
1396         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1397                 goto nla_put_failure;
1398
1399         if (swkey->eth.type == htons(ETH_P_IP)) {
1400                 struct ovs_key_ipv4 *ipv4_key;
1401
1402                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1403                 if (!nla)
1404                         goto nla_put_failure;
1405                 ipv4_key = nla_data(nla);
1406                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1407                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1408                 ipv4_key->ipv4_proto = output->ip.proto;
1409                 ipv4_key->ipv4_tos = output->ip.tos;
1410                 ipv4_key->ipv4_ttl = output->ip.ttl;
1411                 ipv4_key->ipv4_frag = output->ip.frag;
1412         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1413                 struct ovs_key_ipv6 *ipv6_key;
1414
1415                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1416                 if (!nla)
1417                         goto nla_put_failure;
1418                 ipv6_key = nla_data(nla);
1419                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1420                                 sizeof(ipv6_key->ipv6_src));
1421                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1422                                 sizeof(ipv6_key->ipv6_dst));
1423                 ipv6_key->ipv6_label = output->ipv6.label;
1424                 ipv6_key->ipv6_proto = output->ip.proto;
1425                 ipv6_key->ipv6_tclass = output->ip.tos;
1426                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1427                 ipv6_key->ipv6_frag = output->ip.frag;
1428         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1429                    swkey->eth.type == htons(ETH_P_RARP)) {
1430                 struct ovs_key_arp *arp_key;
1431
1432                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1433                 if (!nla)
1434                         goto nla_put_failure;
1435                 arp_key = nla_data(nla);
1436                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1437                 arp_key->arp_sip = output->ipv4.addr.src;
1438                 arp_key->arp_tip = output->ipv4.addr.dst;
1439                 arp_key->arp_op = htons(output->ip.proto);
1440                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1441                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1442         } else if (eth_p_mpls(swkey->eth.type)) {
1443                 struct ovs_key_mpls *mpls_key;
1444
1445                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1446                 if (!nla)
1447                         goto nla_put_failure;
1448                 mpls_key = nla_data(nla);
1449                 mpls_key->mpls_lse = output->mpls.top_lse;
1450         }
1451
1452         if ((swkey->eth.type == htons(ETH_P_IP) ||
1453              swkey->eth.type == htons(ETH_P_IPV6)) &&
1454              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1455
1456                 if (swkey->ip.proto == IPPROTO_TCP) {
1457                         struct ovs_key_tcp *tcp_key;
1458
1459                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1460                         if (!nla)
1461                                 goto nla_put_failure;
1462                         tcp_key = nla_data(nla);
1463                         tcp_key->tcp_src = output->tp.src;
1464                         tcp_key->tcp_dst = output->tp.dst;
1465                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1466                                          output->tp.flags))
1467                                 goto nla_put_failure;
1468                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1469                         struct ovs_key_udp *udp_key;
1470
1471                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1472                         if (!nla)
1473                                 goto nla_put_failure;
1474                         udp_key = nla_data(nla);
1475                         udp_key->udp_src = output->tp.src;
1476                         udp_key->udp_dst = output->tp.dst;
1477                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1478                         struct ovs_key_sctp *sctp_key;
1479
1480                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1481                         if (!nla)
1482                                 goto nla_put_failure;
1483                         sctp_key = nla_data(nla);
1484                         sctp_key->sctp_src = output->tp.src;
1485                         sctp_key->sctp_dst = output->tp.dst;
1486                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1487                            swkey->ip.proto == IPPROTO_ICMP) {
1488                         struct ovs_key_icmp *icmp_key;
1489
1490                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1491                         if (!nla)
1492                                 goto nla_put_failure;
1493                         icmp_key = nla_data(nla);
1494                         icmp_key->icmp_type = ntohs(output->tp.src);
1495                         icmp_key->icmp_code = ntohs(output->tp.dst);
1496                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1497                            swkey->ip.proto == IPPROTO_ICMPV6) {
1498                         struct ovs_key_icmpv6 *icmpv6_key;
1499
1500                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1501                                                 sizeof(*icmpv6_key));
1502                         if (!nla)
1503                                 goto nla_put_failure;
1504                         icmpv6_key = nla_data(nla);
1505                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1506                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1507
1508                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1509                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1510                                 struct ovs_key_nd *nd_key;
1511
1512                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1513                                 if (!nla)
1514                                         goto nla_put_failure;
1515                                 nd_key = nla_data(nla);
1516                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1517                                                         sizeof(nd_key->nd_target));
1518                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1519                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1520                         }
1521                 }
1522         }
1523
1524 unencap:
1525         if (encap)
1526                 nla_nest_end(skb, encap);
1527
1528         return 0;
1529
1530 nla_put_failure:
1531         return -EMSGSIZE;
1532 }
1533
1534 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1535                     const struct sw_flow_key *output, int attr, bool is_mask,
1536                     struct sk_buff *skb)
1537 {
1538         int err;
1539         struct nlattr *nla;
1540
1541         nla = nla_nest_start(skb, attr);
1542         if (!nla)
1543                 return -EMSGSIZE;
1544         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1545         if (err)
1546                 return err;
1547         nla_nest_end(skb, nla);
1548
1549         return 0;
1550 }
1551
1552 /* Called with ovs_mutex or RCU read lock. */
1553 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1554 {
1555         if (ovs_identifier_is_ufid(&flow->id))
1556                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1557                                flow->id.ufid);
1558
1559         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1560                                OVS_FLOW_ATTR_KEY, false, skb);
1561 }
1562
1563 /* Called with ovs_mutex or RCU read lock. */
1564 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1565 {
1566         return ovs_nla_put_key(&flow->key, &flow->key,
1567                                 OVS_FLOW_ATTR_KEY, false, skb);
1568 }
1569
1570 /* Called with ovs_mutex or RCU read lock. */
1571 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1572 {
1573         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1574                                 OVS_FLOW_ATTR_MASK, true, skb);
1575 }
1576
1577 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1578
1579 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1580 {
1581         struct sw_flow_actions *sfa;
1582
1583         if (size > MAX_ACTIONS_BUFSIZE) {
1584                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1585                 return ERR_PTR(-EINVAL);
1586         }
1587
1588         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1589         if (!sfa)
1590                 return ERR_PTR(-ENOMEM);
1591
1592         sfa->actions_len = 0;
1593         return sfa;
1594 }
1595
1596 static void ovs_nla_free_set_action(const struct nlattr *a)
1597 {
1598         const struct nlattr *ovs_key = nla_data(a);
1599         struct ovs_tunnel_info *ovs_tun;
1600
1601         switch (nla_type(ovs_key)) {
1602         case OVS_KEY_ATTR_TUNNEL_INFO:
1603                 ovs_tun = nla_data(ovs_key);
1604                 dst_release((struct dst_entry *)ovs_tun->tun_dst);
1605                 break;
1606         }
1607 }
1608
1609 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1610 {
1611         const struct nlattr *a;
1612         int rem;
1613
1614         if (!sf_acts)
1615                 return;
1616
1617         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1618                 switch (nla_type(a)) {
1619                 case OVS_ACTION_ATTR_SET:
1620                         ovs_nla_free_set_action(a);
1621                         break;
1622                 case OVS_ACTION_ATTR_CT:
1623                         ovs_ct_free_action(a);
1624                         break;
1625                 }
1626         }
1627
1628         kfree(sf_acts);
1629 }
1630
1631 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1632 {
1633         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1634 }
1635
1636 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1637  * The caller must hold rcu_read_lock for this to be sensible. */
1638 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1639 {
1640         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1641 }
1642
1643 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1644                                        int attr_len, bool log)
1645 {
1646
1647         struct sw_flow_actions *acts;
1648         int new_acts_size;
1649         int req_size = NLA_ALIGN(attr_len);
1650         int next_offset = offsetof(struct sw_flow_actions, actions) +
1651                                         (*sfa)->actions_len;
1652
1653         if (req_size <= (ksize(*sfa) - next_offset))
1654                 goto out;
1655
1656         new_acts_size = ksize(*sfa) * 2;
1657
1658         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1659                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1660                         return ERR_PTR(-EMSGSIZE);
1661                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1662         }
1663
1664         acts = nla_alloc_flow_actions(new_acts_size, log);
1665         if (IS_ERR(acts))
1666                 return (void *)acts;
1667
1668         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1669         acts->actions_len = (*sfa)->actions_len;
1670         acts->orig_len = (*sfa)->orig_len;
1671         kfree(*sfa);
1672         *sfa = acts;
1673
1674 out:
1675         (*sfa)->actions_len += req_size;
1676         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1677 }
1678
1679 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1680                                    int attrtype, void *data, int len, bool log)
1681 {
1682         struct nlattr *a;
1683
1684         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1685         if (IS_ERR(a))
1686                 return a;
1687
1688         a->nla_type = attrtype;
1689         a->nla_len = nla_attr_size(len);
1690
1691         if (data)
1692                 memcpy(nla_data(a), data, len);
1693         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1694
1695         return a;
1696 }
1697
1698 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1699                        int len, bool log)
1700 {
1701         struct nlattr *a;
1702
1703         a = __add_action(sfa, attrtype, data, len, log);
1704
1705         return PTR_ERR_OR_ZERO(a);
1706 }
1707
1708 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1709                                           int attrtype, bool log)
1710 {
1711         int used = (*sfa)->actions_len;
1712         int err;
1713
1714         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1715         if (err)
1716                 return err;
1717
1718         return used;
1719 }
1720
1721 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1722                                          int st_offset)
1723 {
1724         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1725                                                                st_offset);
1726
1727         a->nla_len = sfa->actions_len - st_offset;
1728 }
1729
1730 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1731                                   const struct sw_flow_key *key,
1732                                   int depth, struct sw_flow_actions **sfa,
1733                                   __be16 eth_type, __be16 vlan_tci, bool log);
1734
1735 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1736                                     const struct sw_flow_key *key, int depth,
1737                                     struct sw_flow_actions **sfa,
1738                                     __be16 eth_type, __be16 vlan_tci, bool log)
1739 {
1740         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1741         const struct nlattr *probability, *actions;
1742         const struct nlattr *a;
1743         int rem, start, err, st_acts;
1744
1745         memset(attrs, 0, sizeof(attrs));
1746         nla_for_each_nested(a, attr, rem) {
1747                 int type = nla_type(a);
1748                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1749                         return -EINVAL;
1750                 attrs[type] = a;
1751         }
1752         if (rem)
1753                 return -EINVAL;
1754
1755         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1756         if (!probability || nla_len(probability) != sizeof(u32))
1757                 return -EINVAL;
1758
1759         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1760         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1761                 return -EINVAL;
1762
1763         /* validation done, copy sample action. */
1764         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1765         if (start < 0)
1766                 return start;
1767         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1768                                  nla_data(probability), sizeof(u32), log);
1769         if (err)
1770                 return err;
1771         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1772         if (st_acts < 0)
1773                 return st_acts;
1774
1775         err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1776                                      eth_type, vlan_tci, log);
1777         if (err)
1778                 return err;
1779
1780         add_nested_action_end(*sfa, st_acts);
1781         add_nested_action_end(*sfa, start);
1782
1783         return 0;
1784 }
1785
1786 void ovs_match_init(struct sw_flow_match *match,
1787                     struct sw_flow_key *key,
1788                     struct sw_flow_mask *mask)
1789 {
1790         memset(match, 0, sizeof(*match));
1791         match->key = key;
1792         match->mask = mask;
1793
1794         memset(key, 0, sizeof(*key));
1795
1796         if (mask) {
1797                 memset(&mask->key, 0, sizeof(mask->key));
1798                 mask->range.start = mask->range.end = 0;
1799         }
1800 }
1801
1802 static int validate_geneve_opts(struct sw_flow_key *key)
1803 {
1804         struct geneve_opt *option;
1805         int opts_len = key->tun_opts_len;
1806         bool crit_opt = false;
1807
1808         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1809         while (opts_len > 0) {
1810                 int len;
1811
1812                 if (opts_len < sizeof(*option))
1813                         return -EINVAL;
1814
1815                 len = sizeof(*option) + option->length * 4;
1816                 if (len > opts_len)
1817                         return -EINVAL;
1818
1819                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1820
1821                 option = (struct geneve_opt *)((u8 *)option + len);
1822                 opts_len -= len;
1823         };
1824
1825         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1826
1827         return 0;
1828 }
1829
1830 static int validate_and_copy_set_tun(const struct nlattr *attr,
1831                                      struct sw_flow_actions **sfa, bool log)
1832 {
1833         struct sw_flow_match match;
1834         struct sw_flow_key key;
1835         struct metadata_dst *tun_dst;
1836         struct ip_tunnel_info *tun_info;
1837         struct ovs_tunnel_info *ovs_tun;
1838         struct nlattr *a;
1839         int err = 0, start, opts_type;
1840
1841         ovs_match_init(&match, &key, NULL);
1842         opts_type = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1843         if (opts_type < 0)
1844                 return opts_type;
1845
1846         if (key.tun_opts_len) {
1847                 switch (opts_type) {
1848                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1849                         err = validate_geneve_opts(&key);
1850                         if (err < 0)
1851                                 return err;
1852                         break;
1853                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1854                         break;
1855                 }
1856         };
1857
1858         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1859         if (start < 0)
1860                 return start;
1861
1862         tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1863         if (!tun_dst)
1864                 return -ENOMEM;
1865
1866         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1867                          sizeof(*ovs_tun), log);
1868         if (IS_ERR(a)) {
1869                 dst_release((struct dst_entry *)tun_dst);
1870                 return PTR_ERR(a);
1871         }
1872
1873         ovs_tun = nla_data(a);
1874         ovs_tun->tun_dst = tun_dst;
1875
1876         tun_info = &tun_dst->u.tun_info;
1877         tun_info->mode = IP_TUNNEL_INFO_TX;
1878         tun_info->key = key.tun_key;
1879         tun_info->options_len = key.tun_opts_len;
1880
1881         if (tun_info->options_len) {
1882                 /* We need to store the options in the action itself since
1883                  * everything else will go away after flow setup. We can append
1884                  * it to tun_info and then point there.
1885                  */
1886                 memcpy((tun_info + 1),
1887                        TUN_METADATA_OPTS(&key, key.tun_opts_len), key.tun_opts_len);
1888                 tun_info->options = (tun_info + 1);
1889         } else {
1890                 tun_info->options = NULL;
1891         }
1892
1893         add_nested_action_end(*sfa, start);
1894
1895         return err;
1896 }
1897
1898 /* Return false if there are any non-masked bits set.
1899  * Mask follows data immediately, before any netlink padding.
1900  */
1901 static bool validate_masked(u8 *data, int len)
1902 {
1903         u8 *mask = data + len;
1904
1905         while (len--)
1906                 if (*data++ & ~*mask++)
1907                         return false;
1908
1909         return true;
1910 }
1911
1912 static int validate_set(const struct nlattr *a,
1913                         const struct sw_flow_key *flow_key,
1914                         struct sw_flow_actions **sfa,
1915                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
1916 {
1917         const struct nlattr *ovs_key = nla_data(a);
1918         int key_type = nla_type(ovs_key);
1919         size_t key_len;
1920
1921         /* There can be only one key in a action */
1922         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1923                 return -EINVAL;
1924
1925         key_len = nla_len(ovs_key);
1926         if (masked)
1927                 key_len /= 2;
1928
1929         if (key_type > OVS_KEY_ATTR_MAX ||
1930             (ovs_key_lens[key_type].len != key_len &&
1931              ovs_key_lens[key_type].len != OVS_ATTR_NESTED))
1932                 return -EINVAL;
1933
1934         if (masked && !validate_masked(nla_data(ovs_key), key_len))
1935                 return -EINVAL;
1936
1937         switch (key_type) {
1938         const struct ovs_key_ipv4 *ipv4_key;
1939         const struct ovs_key_ipv6 *ipv6_key;
1940         int err;
1941
1942         case OVS_KEY_ATTR_PRIORITY:
1943         case OVS_KEY_ATTR_SKB_MARK:
1944         case OVS_KEY_ATTR_CT_MARK:
1945         case OVS_KEY_ATTR_CT_LABEL:
1946         case OVS_KEY_ATTR_ETHERNET:
1947                 break;
1948
1949         case OVS_KEY_ATTR_TUNNEL:
1950                 if (eth_p_mpls(eth_type))
1951                         return -EINVAL;
1952
1953                 if (masked)
1954                         return -EINVAL; /* Masked tunnel set not supported. */
1955
1956                 *skip_copy = true;
1957                 err = validate_and_copy_set_tun(a, sfa, log);
1958                 if (err)
1959                         return err;
1960                 break;
1961
1962         case OVS_KEY_ATTR_IPV4:
1963                 if (eth_type != htons(ETH_P_IP))
1964                         return -EINVAL;
1965
1966                 ipv4_key = nla_data(ovs_key);
1967
1968                 if (masked) {
1969                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
1970
1971                         /* Non-writeable fields. */
1972                         if (mask->ipv4_proto || mask->ipv4_frag)
1973                                 return -EINVAL;
1974                 } else {
1975                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1976                                 return -EINVAL;
1977
1978                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1979                                 return -EINVAL;
1980                 }
1981                 break;
1982
1983         case OVS_KEY_ATTR_IPV6:
1984                 if (eth_type != htons(ETH_P_IPV6))
1985                         return -EINVAL;
1986
1987                 ipv6_key = nla_data(ovs_key);
1988
1989                 if (masked) {
1990                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
1991
1992                         /* Non-writeable fields. */
1993                         if (mask->ipv6_proto || mask->ipv6_frag)
1994                                 return -EINVAL;
1995
1996                         /* Invalid bits in the flow label mask? */
1997                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
1998                                 return -EINVAL;
1999                 } else {
2000                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2001                                 return -EINVAL;
2002
2003                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2004                                 return -EINVAL;
2005                 }
2006                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2007                         return -EINVAL;
2008
2009                 break;
2010
2011         case OVS_KEY_ATTR_TCP:
2012                 if ((eth_type != htons(ETH_P_IP) &&
2013                      eth_type != htons(ETH_P_IPV6)) ||
2014                     flow_key->ip.proto != IPPROTO_TCP)
2015                         return -EINVAL;
2016
2017                 break;
2018
2019         case OVS_KEY_ATTR_UDP:
2020                 if ((eth_type != htons(ETH_P_IP) &&
2021                      eth_type != htons(ETH_P_IPV6)) ||
2022                     flow_key->ip.proto != IPPROTO_UDP)
2023                         return -EINVAL;
2024
2025                 break;
2026
2027         case OVS_KEY_ATTR_MPLS:
2028                 if (!eth_p_mpls(eth_type))
2029                         return -EINVAL;
2030                 break;
2031
2032         case OVS_KEY_ATTR_SCTP:
2033                 if ((eth_type != htons(ETH_P_IP) &&
2034                      eth_type != htons(ETH_P_IPV6)) ||
2035                     flow_key->ip.proto != IPPROTO_SCTP)
2036                         return -EINVAL;
2037
2038                 break;
2039
2040         default:
2041                 return -EINVAL;
2042         }
2043
2044         /* Convert non-masked non-tunnel set actions to masked set actions. */
2045         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2046                 int start, len = key_len * 2;
2047                 struct nlattr *at;
2048
2049                 *skip_copy = true;
2050
2051                 start = add_nested_action_start(sfa,
2052                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2053                                                 log);
2054                 if (start < 0)
2055                         return start;
2056
2057                 at = __add_action(sfa, key_type, NULL, len, log);
2058                 if (IS_ERR(at))
2059                         return PTR_ERR(at);
2060
2061                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2062                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2063                 /* Clear non-writeable bits from otherwise writeable fields. */
2064                 if (key_type == OVS_KEY_ATTR_IPV6) {
2065                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2066
2067                         mask->ipv6_label &= htonl(0x000FFFFF);
2068                 }
2069                 add_nested_action_end(*sfa, start);
2070         }
2071
2072         return 0;
2073 }
2074
2075 static int validate_userspace(const struct nlattr *attr)
2076 {
2077         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2078                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2079                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2080                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2081         };
2082         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2083         int error;
2084
2085         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2086                                  attr, userspace_policy);
2087         if (error)
2088                 return error;
2089
2090         if (!a[OVS_USERSPACE_ATTR_PID] ||
2091             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2092                 return -EINVAL;
2093
2094         return 0;
2095 }
2096
2097 static int copy_action(const struct nlattr *from,
2098                        struct sw_flow_actions **sfa, bool log)
2099 {
2100         int totlen = NLA_ALIGN(from->nla_len);
2101         struct nlattr *to;
2102
2103         to = reserve_sfa_size(sfa, from->nla_len, log);
2104         if (IS_ERR(to))
2105                 return PTR_ERR(to);
2106
2107         memcpy(to, from, totlen);
2108         return 0;
2109 }
2110
2111 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2112                                   const struct sw_flow_key *key,
2113                                   int depth, struct sw_flow_actions **sfa,
2114                                   __be16 eth_type, __be16 vlan_tci, bool log)
2115 {
2116         const struct nlattr *a;
2117         int rem, err;
2118
2119         if (depth >= SAMPLE_ACTION_DEPTH)
2120                 return -EOVERFLOW;
2121
2122         nla_for_each_nested(a, attr, rem) {
2123                 /* Expected argument lengths, (u32)-1 for variable length. */
2124                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2125                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2126                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2127                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2128                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2129                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2130                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2131                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2132                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2133                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2134                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2135                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2136                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2137                 };
2138                 const struct ovs_action_push_vlan *vlan;
2139                 int type = nla_type(a);
2140                 bool skip_copy;
2141
2142                 if (type > OVS_ACTION_ATTR_MAX ||
2143                     (action_lens[type] != nla_len(a) &&
2144                      action_lens[type] != (u32)-1))
2145                         return -EINVAL;
2146
2147                 skip_copy = false;
2148                 switch (type) {
2149                 case OVS_ACTION_ATTR_UNSPEC:
2150                         return -EINVAL;
2151
2152                 case OVS_ACTION_ATTR_USERSPACE:
2153                         err = validate_userspace(a);
2154                         if (err)
2155                                 return err;
2156                         break;
2157
2158                 case OVS_ACTION_ATTR_OUTPUT:
2159                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2160                                 return -EINVAL;
2161                         break;
2162
2163                 case OVS_ACTION_ATTR_HASH: {
2164                         const struct ovs_action_hash *act_hash = nla_data(a);
2165
2166                         switch (act_hash->hash_alg) {
2167                         case OVS_HASH_ALG_L4:
2168                                 break;
2169                         default:
2170                                 return  -EINVAL;
2171                         }
2172
2173                         break;
2174                 }
2175
2176                 case OVS_ACTION_ATTR_POP_VLAN:
2177                         vlan_tci = htons(0);
2178                         break;
2179
2180                 case OVS_ACTION_ATTR_PUSH_VLAN:
2181                         vlan = nla_data(a);
2182                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2183                                 return -EINVAL;
2184                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2185                                 return -EINVAL;
2186                         vlan_tci = vlan->vlan_tci;
2187                         break;
2188
2189                 case OVS_ACTION_ATTR_RECIRC:
2190                         break;
2191
2192                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2193                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2194
2195                         if (!eth_p_mpls(mpls->mpls_ethertype))
2196                                 return -EINVAL;
2197                         /* Prohibit push MPLS other than to a white list
2198                          * for packets that have a known tag order.
2199                          */
2200                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2201                             (eth_type != htons(ETH_P_IP) &&
2202                              eth_type != htons(ETH_P_IPV6) &&
2203                              eth_type != htons(ETH_P_ARP) &&
2204                              eth_type != htons(ETH_P_RARP) &&
2205                              !eth_p_mpls(eth_type)))
2206                                 return -EINVAL;
2207                         eth_type = mpls->mpls_ethertype;
2208                         break;
2209                 }
2210
2211                 case OVS_ACTION_ATTR_POP_MPLS:
2212                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2213                             !eth_p_mpls(eth_type))
2214                                 return -EINVAL;
2215
2216                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2217                          * as there is no check here to ensure that the new
2218                          * eth_type is valid and thus set actions could
2219                          * write off the end of the packet or otherwise
2220                          * corrupt it.
2221                          *
2222                          * Support for these actions is planned using packet
2223                          * recirculation.
2224                          */
2225                         eth_type = htons(0);
2226                         break;
2227
2228                 case OVS_ACTION_ATTR_SET:
2229                         err = validate_set(a, key, sfa,
2230                                            &skip_copy, eth_type, false, log);
2231                         if (err)
2232                                 return err;
2233                         break;
2234
2235                 case OVS_ACTION_ATTR_SET_MASKED:
2236                         err = validate_set(a, key, sfa,
2237                                            &skip_copy, eth_type, true, log);
2238                         if (err)
2239                                 return err;
2240                         break;
2241
2242                 case OVS_ACTION_ATTR_SAMPLE:
2243                         err = validate_and_copy_sample(net, a, key, depth, sfa,
2244                                                        eth_type, vlan_tci, log);
2245                         if (err)
2246                                 return err;
2247                         skip_copy = true;
2248                         break;
2249
2250                 case OVS_ACTION_ATTR_CT:
2251                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2252                         if (err)
2253                                 return err;
2254                         skip_copy = true;
2255                         break;
2256
2257                 default:
2258                         OVS_NLERR(log, "Unknown Action type %d", type);
2259                         return -EINVAL;
2260                 }
2261                 if (!skip_copy) {
2262                         err = copy_action(a, sfa, log);
2263                         if (err)
2264                                 return err;
2265                 }
2266         }
2267
2268         if (rem > 0)
2269                 return -EINVAL;
2270
2271         return 0;
2272 }
2273
2274 /* 'key' must be the masked key. */
2275 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2276                          const struct sw_flow_key *key,
2277                          struct sw_flow_actions **sfa, bool log)
2278 {
2279         int err;
2280
2281         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2282         if (IS_ERR(*sfa))
2283                 return PTR_ERR(*sfa);
2284
2285         (*sfa)->orig_len = nla_len(attr);
2286         err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2287                                      key->eth.tci, log);
2288         if (err)
2289                 ovs_nla_free_flow_actions(*sfa);
2290
2291         return err;
2292 }
2293
2294 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2295 {
2296         const struct nlattr *a;
2297         struct nlattr *start;
2298         int err = 0, rem;
2299
2300         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2301         if (!start)
2302                 return -EMSGSIZE;
2303
2304         nla_for_each_nested(a, attr, rem) {
2305                 int type = nla_type(a);
2306                 struct nlattr *st_sample;
2307
2308                 switch (type) {
2309                 case OVS_SAMPLE_ATTR_PROBABILITY:
2310                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2311                                     sizeof(u32), nla_data(a)))
2312                                 return -EMSGSIZE;
2313                         break;
2314                 case OVS_SAMPLE_ATTR_ACTIONS:
2315                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2316                         if (!st_sample)
2317                                 return -EMSGSIZE;
2318                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2319                         if (err)
2320                                 return err;
2321                         nla_nest_end(skb, st_sample);
2322                         break;
2323                 }
2324         }
2325
2326         nla_nest_end(skb, start);
2327         return err;
2328 }
2329
2330 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2331 {
2332         const struct nlattr *ovs_key = nla_data(a);
2333         int key_type = nla_type(ovs_key);
2334         struct nlattr *start;
2335         int err;
2336
2337         switch (key_type) {
2338         case OVS_KEY_ATTR_TUNNEL_INFO: {
2339                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2340                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2341
2342                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2343                 if (!start)
2344                         return -EMSGSIZE;
2345
2346                 err = ipv4_tun_to_nlattr(skb, &tun_info->key,
2347                                          tun_info->options_len ?
2348                                                 tun_info->options : NULL,
2349                                          tun_info->options_len);
2350                 if (err)
2351                         return err;
2352                 nla_nest_end(skb, start);
2353                 break;
2354         }
2355         default:
2356                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2357                         return -EMSGSIZE;
2358                 break;
2359         }
2360
2361         return 0;
2362 }
2363
2364 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2365                                                 struct sk_buff *skb)
2366 {
2367         const struct nlattr *ovs_key = nla_data(a);
2368         struct nlattr *nla;
2369         size_t key_len = nla_len(ovs_key) / 2;
2370
2371         /* Revert the conversion we did from a non-masked set action to
2372          * masked set action.
2373          */
2374         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2375         if (!nla)
2376                 return -EMSGSIZE;
2377
2378         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2379                 return -EMSGSIZE;
2380
2381         nla_nest_end(skb, nla);
2382         return 0;
2383 }
2384
2385 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2386 {
2387         const struct nlattr *a;
2388         int rem, err;
2389
2390         nla_for_each_attr(a, attr, len, rem) {
2391                 int type = nla_type(a);
2392
2393                 switch (type) {
2394                 case OVS_ACTION_ATTR_SET:
2395                         err = set_action_to_attr(a, skb);
2396                         if (err)
2397                                 return err;
2398                         break;
2399
2400                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2401                         err = masked_set_action_to_set_action_attr(a, skb);
2402                         if (err)
2403                                 return err;
2404                         break;
2405
2406                 case OVS_ACTION_ATTR_SAMPLE:
2407                         err = sample_action_to_attr(a, skb);
2408                         if (err)
2409                                 return err;
2410                         break;
2411
2412                 case OVS_ACTION_ATTR_CT:
2413                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2414                         if (err)
2415                                 return err;
2416                         break;
2417
2418                 default:
2419                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2420                                 return -EMSGSIZE;
2421                         break;
2422                 }
2423         }
2424
2425         return 0;
2426 }