datapath/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 <linux/uaccess.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <net/llc_pdu.h>
  27 #include <linux/kernel.h>
  28 #include <linux/jhash.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/llc.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/rcupdate.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/sctp.h>
  38 #include <linux/tcp.h>
  39 #include <linux/udp.h>
  40 #include <linux/icmp.h>
  41 #include <linux/icmpv6.h>
  42 #include <linux/rculist.h>
  43 #include <net/geneve.h>
  44 #include <net/ip.h>
  45 #include <net/ipv6.h>
  46 #include <net/ndisc.h>
  47 #include <net/mpls.h>
  48
  49 #include "datapath.h"
  50 #include "flow.h"
  51 #include "flow_netlink.h"
  52 #include "vport-vxlan.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, len);\
 101                 else                                                        \
 102                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
 103         } while (0)
 104
 105 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
 106         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 107                                   value_p, len, is_mask)
 108
 109 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 110         do {                                                                \
 111                 update_range(match, offsetof(struct sw_flow_key, field),    \
 112                              sizeof((match)->key->field), is_mask);         \
 113                 if (is_mask)                                                \
 114                         memset((u8 *)&(match)->mask->key.field, value,      \
 115                                sizeof((match)->mask->key.field));           \
 116                 else                                                        \
 117                         memset((u8 *)&(match)->key->field, value,           \
 118                                sizeof((match)->key->field));                \
 119         } while (0)
 120
 121 static bool match_validate(const struct sw_flow_match *match,
 122                            u64 key_attrs, u64 mask_attrs, bool log)
 123 {
 124         u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
 125         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 126
 127         /* The following mask attributes allowed only if they
 128          * pass the validation tests.
 129          */
 130         mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
 131                         | (1ULL << OVS_KEY_ATTR_IPV6)
 132                         | (1ULL << OVS_KEY_ATTR_TCP)
 133                         | (1ULL << OVS_KEY_ATTR_TCP_FLAGS)
 134                         | (1ULL << OVS_KEY_ATTR_UDP)
 135                         | (1ULL << OVS_KEY_ATTR_SCTP)
 136                         | (1ULL << OVS_KEY_ATTR_ICMP)
 137                         | (1ULL << OVS_KEY_ATTR_ICMPV6)
 138                         | (1ULL << OVS_KEY_ATTR_ARP)
 139                         | (1ULL << OVS_KEY_ATTR_ND)
 140                         | (1ULL << OVS_KEY_ATTR_MPLS));
 141
 142         /* Always allowed mask fields. */
 143         mask_allowed |= ((1ULL << OVS_KEY_ATTR_TUNNEL)
 144                        | (1ULL << OVS_KEY_ATTR_IN_PORT)
 145                        | (1ULL << 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 |= 1ULL << OVS_KEY_ATTR_ARP;
 151                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 152                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
 153         }
 154
 155         if (eth_p_mpls(match->key->eth.type)) {
 156                 key_expected |= 1ULL << OVS_KEY_ATTR_MPLS;
 157                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 158                         mask_allowed |= 1ULL << OVS_KEY_ATTR_MPLS;
 159         }
 160
 161         if (match->key->eth.type == htons(ETH_P_IP)) {
 162                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
 163                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 164                         mask_allowed |= 1ULL << 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 |= 1ULL << OVS_KEY_ATTR_UDP;
 169                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 170                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 171                         }
 172
 173                         if (match->key->ip.proto == IPPROTO_SCTP) {
 174                                 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
 175                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 176                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
 177                         }
 178
 179                         if (match->key->ip.proto == IPPROTO_TCP) {
 180                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 181                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 182                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 183                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 184                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 185                                 }
 186                         }
 187
 188                         if (match->key->ip.proto == IPPROTO_ICMP) {
 189                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
 190                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 191                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
 192                         }
 193                 }
 194         }
 195
 196         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 197                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
 198                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 199                         mask_allowed |= 1ULL << 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 |= 1ULL << OVS_KEY_ATTR_UDP;
 204                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 205                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 206                         }
 207
 208                         if (match->key->ip.proto == IPPROTO_SCTP) {
 209                                 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
 210                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 211                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
 212                         }
 213
 214                         if (match->key->ip.proto == IPPROTO_TCP) {
 215                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 216                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 217                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 218                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 219                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 220                                 }
 221                         }
 222
 223                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 224                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 225                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 226                                         mask_allowed |= 1ULL << 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 |= 1ULL << OVS_KEY_ATTR_ND;
 232                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 233                                                 mask_allowed |= 1ULL << 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 != 22);
 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(12)  /* OVS_KEY_ATTR_ETHERNET */
 294                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 295                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 296                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 297                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 298                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 299                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 300                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 301 }
 302
 303 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 304         [OVS_TUNNEL_KEY_ATTR_ID]            = { .len = sizeof(u64) },
 305         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC]      = { .len = sizeof(u32) },
 306         [OVS_TUNNEL_KEY_ATTR_IPV4_DST]      = { .len = sizeof(u32) },
 307         [OVS_TUNNEL_KEY_ATTR_TOS]           = { .len = 1 },
 308         [OVS_TUNNEL_KEY_ATTR_TTL]           = { .len = 1 },
 309         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
 310         [OVS_TUNNEL_KEY_ATTR_CSUM]          = { .len = 0 },
 311         [OVS_TUNNEL_KEY_ATTR_TP_SRC]        = { .len = sizeof(u16) },
 312         [OVS_TUNNEL_KEY_ATTR_TP_DST]        = { .len = sizeof(u16) },
 313         [OVS_TUNNEL_KEY_ATTR_OAM]           = { .len = 0 },
 314         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_NESTED },
 315         [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED },
 316 };
 317
 318 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 319 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 320         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 321         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 322         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 323         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 324         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 325         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 326         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 327         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 328         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 329         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 330         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 331         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 332         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 333         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 334         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 335         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 336         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 337         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 338         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 339         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 340                                      .next = ovs_tunnel_key_lens, },
 341         [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 342 };
 343
 344 static bool is_all_zero(const u8 *fp, size_t size)
 345 {
 346         int i;
 347
 348         if (!fp)
 349                 return false;
 350
 351         for (i = 0; i < size; i++)
 352                 if (fp[i])
 353                         return false;
 354
 355         return true;
 356 }
 357
 358 static int __parse_flow_nlattrs(const struct nlattr *attr,
 359                                 const struct nlattr *a[],
 360                                 u64 *attrsp, bool log, bool nz)
 361 {
 362         const struct nlattr *nla;
 363         u64 attrs;
 364         int rem;
 365
 366         attrs = *attrsp;
 367         nla_for_each_nested(nla, attr, rem) {
 368                 u16 type = nla_type(nla);
 369                 int expected_len;
 370
 371                 if (type > OVS_KEY_ATTR_MAX) {
 372                         OVS_NLERR(log, "Key type %d is out of range max %d",
 373                                   type, OVS_KEY_ATTR_MAX);
 374                         return -EINVAL;
 375                 }
 376
 377                 if (attrs & (1ULL << type)) {
 378                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 379                         return -EINVAL;
 380                 }
 381
 382                 expected_len = ovs_key_lens[type].len;
 383                 if (nla_len(nla) != expected_len && expected_len != OVS_ATTR_NESTED) {
 384                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 385                                   type, nla_len(nla), expected_len);
 386                         return -EINVAL;
 387                 }
 388
 389                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 390                         attrs |= 1ULL << type;
 391                         a[type] = nla;
 392                 }
 393         }
 394         if (rem) {
 395                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 396                 return -EINVAL;
 397         }
 398
 399         *attrsp = attrs;
 400         return 0;
 401 }
 402
 403 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 404                                    const struct nlattr *a[], u64 *attrsp,
 405                                    bool log)
 406 {
 407         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 408 }
 409
 410 static int parse_flow_nlattrs(const struct nlattr *attr,
 411                               const struct nlattr *a[], u64 *attrsp,
 412                               bool log)
 413 {
 414         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 415 }
 416
 417 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 418                                      struct sw_flow_match *match, bool is_mask,
 419                                      bool log)
 420 {
 421         unsigned long opt_key_offset;
 422
 423         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 424                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 425                           nla_len(a), sizeof(match->key->tun_opts));
 426                 return -EINVAL;
 427         }
 428
 429         if (nla_len(a) % 4 != 0) {
 430                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 431                           nla_len(a));
 432                 return -EINVAL;
 433         }
 434
 435         /* We need to record the length of the options passed
 436          * down, otherwise packets with the same format but
 437          * additional options will be silently matched.
 438          */
 439         if (!is_mask) {
 440                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 441                                 false);
 442         } else {
 443                 /* This is somewhat unusual because it looks at
 444                  * both the key and mask while parsing the
 445                  * attributes (and by extension assumes the key
 446                  * is parsed first). Normally, we would verify
 447                  * that each is the correct length and that the
 448                  * attributes line up in the validate function.
 449                  * However, that is difficult because this is
 450                  * variable length and we won't have the
 451                  * information later.
 452                  */
 453                 if (match->key->tun_opts_len != nla_len(a)) {
 454                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 455                                   match->key->tun_opts_len, nla_len(a));
 456                         return -EINVAL;
 457                 }
 458
 459                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 460         }
 461
 462         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 463         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 464                                   nla_len(a), is_mask);
 465         return 0;
 466 }
 467
 468 static const struct nla_policy vxlan_opt_policy[OVS_VXLAN_EXT_MAX + 1] = {
 469         [OVS_VXLAN_EXT_GBP]     = { .type = NLA_U32 },
 470 };
 471
 472 static int vxlan_tun_opt_from_nlattr(const struct nlattr *a,
 473                                      struct sw_flow_match *match, bool is_mask,
 474                                      bool log)
 475 {
 476         struct nlattr *tb[OVS_VXLAN_EXT_MAX+1];
 477         unsigned long opt_key_offset;
 478         struct ovs_vxlan_opts opts;
 479         int err;
 480
 481         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 482
 483         err = nla_parse_nested(tb, OVS_VXLAN_EXT_MAX, a, vxlan_opt_policy);
 484         if (err < 0)
 485                 return err;
 486
 487         memset(&opts, 0, sizeof(opts));
 488
 489         if (tb[OVS_VXLAN_EXT_GBP])
 490                 opts.gbp = nla_get_u32(tb[OVS_VXLAN_EXT_GBP]);
 491
 492         if (!is_mask)
 493                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 494         else
 495                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 496
 497         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 498         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 499                                   is_mask);
 500         return 0;
 501 }
 502
 503 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
 504                                 struct sw_flow_match *match, bool is_mask,
 505                                 bool log)
 506 {
 507         struct nlattr *a;
 508         int rem;
 509         bool ttl = false;
 510         __be16 tun_flags = 0;
 511         int opts_type = 0;
 512
 513         nla_for_each_nested(a, attr, rem) {
 514                 int type = nla_type(a);
 515                 int err;
 516
 517                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 518                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 519                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 520                         return -EINVAL;
 521                 }
 522
 523                 if (ovs_tunnel_key_lens[type].len != nla_len(a) &&
 524                    ovs_tunnel_key_lens[type].len != OVS_ATTR_NESTED) {
 525                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 526                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 527                         return -EINVAL;
 528                 }
 529
 530                 switch (type) {
 531                 case OVS_TUNNEL_KEY_ATTR_ID:
 532                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 533                                         nla_get_be64(a), is_mask);
 534                         tun_flags |= TUNNEL_KEY;
 535                         break;
 536                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 537                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
 538                                         nla_get_in_addr(a), is_mask);
 539                         break;
 540                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 541                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
 542                                         nla_get_in_addr(a), is_mask);
 543                         break;
 544                 case OVS_TUNNEL_KEY_ATTR_TOS:
 545                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
 546                                         nla_get_u8(a), is_mask);
 547                         break;
 548                 case OVS_TUNNEL_KEY_ATTR_TTL:
 549                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
 550                                         nla_get_u8(a), is_mask);
 551                         ttl = true;
 552                         break;
 553                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 554                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 555                         break;
 556                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 557                         tun_flags |= TUNNEL_CSUM;
 558                         break;
 559                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 560                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 561                                         nla_get_be16(a), is_mask);
 562                         break;
 563                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 564                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 565                                         nla_get_be16(a), is_mask);
 566                         break;
 567                 case OVS_TUNNEL_KEY_ATTR_OAM:
 568                         tun_flags |= TUNNEL_OAM;
 569                         break;
 570                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 571                         if (opts_type) {
 572                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 573                                 return -EINVAL;
 574                         }
 575
 576                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 577                         if (err)
 578                                 return err;
 579
 580                         tun_flags |= TUNNEL_GENEVE_OPT;
 581                         opts_type = type;
 582                         break;
 583                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 584                         if (opts_type) {
 585                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 586                                 return -EINVAL;
 587                         }
 588
 589                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 590                         if (err)
 591                                 return err;
 592
 593                         tun_flags |= TUNNEL_VXLAN_OPT;
 594                         opts_type = type;
 595                         break;
 596                 default:
 597                         OVS_NLERR(log, "Unknown IPv4 tunnel attribute %d",
 598                                   type);
 599                         return -EINVAL;
 600                 }
 601         }
 602
 603         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 604
 605         if (rem > 0) {
 606                 OVS_NLERR(log, "IPv4 tunnel attribute has %d unknown bytes.",
 607                           rem);
 608                 return -EINVAL;
 609         }
 610
 611         if (!is_mask) {
 612                 if (!match->key->tun_key.ipv4_dst) {
 613                         OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 614                         return -EINVAL;
 615                 }
 616
 617                 if (!ttl) {
 618                         OVS_NLERR(log, "IPv4 tunnel TTL not specified.");
 619                         return -EINVAL;
 620                 }
 621         }
 622
 623         return opts_type;
 624 }
 625
 626 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 627                                const void *tun_opts, int swkey_tun_opts_len)
 628 {
 629         const struct ovs_vxlan_opts *opts = tun_opts;
 630         struct nlattr *nla;
 631
 632         nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 633         if (!nla)
 634                 return -EMSGSIZE;
 635
 636         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 637                 return -EMSGSIZE;
 638
 639         nla_nest_end(skb, nla);
 640         return 0;
 641 }
 642
 643 static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
 644                                 const struct ovs_key_ipv4_tunnel *output,
 645                                 const void *tun_opts, int swkey_tun_opts_len)
 646 {
 647         if (output->tun_flags & TUNNEL_KEY &&
 648             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 649                 return -EMSGSIZE;
 650         if (output->ipv4_src &&
 651             nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
 652                 return -EMSGSIZE;
 653         if (output->ipv4_dst &&
 654             nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
 655                 return -EMSGSIZE;
 656         if (output->ipv4_tos &&
 657             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
 658                 return -EMSGSIZE;
 659         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
 660                 return -EMSGSIZE;
 661         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 662             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 663                 return -EMSGSIZE;
 664         if ((output->tun_flags & TUNNEL_CSUM) &&
 665             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 666                 return -EMSGSIZE;
 667         if (output->tp_src &&
 668             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 669                 return -EMSGSIZE;
 670         if (output->tp_dst &&
 671             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 672                 return -EMSGSIZE;
 673         if ((output->tun_flags & TUNNEL_OAM) &&
 674             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 675                 return -EMSGSIZE;
 676         if (tun_opts) {
 677                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 678                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 679                             swkey_tun_opts_len, tun_opts))
 680                         return -EMSGSIZE;
 681                else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 682                         vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 683                         return -EMSGSIZE;
 684         }
 685
 686         return 0;
 687 }
 688
 689 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
 690                               const struct ovs_key_ipv4_tunnel *output,
 691                               const void *tun_opts, int swkey_tun_opts_len)
 692 {
 693         struct nlattr *nla;
 694         int err;
 695
 696         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 697         if (!nla)
 698                 return -EMSGSIZE;
 699
 700         err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
 701         if (err)
 702                 return err;
 703
 704         nla_nest_end(skb, nla);
 705         return 0;
 706 }
 707
 708 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
 709                                   const struct ovs_tunnel_info *egress_tun_info)
 710 {
 711         return __ipv4_tun_to_nlattr(skb, &egress_tun_info->tunnel,
 712                                     egress_tun_info->options,
 713                                     egress_tun_info->options_len);
 714 }
 715
 716 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
 717                                  const struct nlattr **a, bool is_mask,
 718                                  bool log)
 719 {
 720         if (*attrs & (1ULL << OVS_KEY_ATTR_DP_HASH)) {
 721                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 722
 723                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 724                 *attrs &= ~(1ULL << OVS_KEY_ATTR_DP_HASH);
 725         }
 726
 727         if (*attrs & (1ULL << OVS_KEY_ATTR_RECIRC_ID)) {
 728                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 729
 730                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 731                 *attrs &= ~(1ULL << OVS_KEY_ATTR_RECIRC_ID);
 732         }
 733
 734         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
 735                 SW_FLOW_KEY_PUT(match, phy.priority,
 736                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 737                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
 738         }
 739
 740         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
 741                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 742
 743                 if (is_mask) {
 744                         in_port = 0xffffffff; /* Always exact match in_port. */
 745                 } else if (in_port >= DP_MAX_PORTS) {
 746                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 747                                   in_port, DP_MAX_PORTS);
 748                         return -EINVAL;
 749                 }
 750
 751                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 752                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
 753         } else if (!is_mask) {
 754                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 755         }
 756
 757         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
 758                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 759
 760                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 761                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
 762         }
 763         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
 764                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 765                                          is_mask, log) < 0)
 766                         return -EINVAL;
 767                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
 768         }
 769         return 0;
 770 }
 771
 772 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
 773                                 const struct nlattr **a, bool is_mask,
 774                                 bool log)
 775 {
 776         int err;
 777
 778         err = metadata_from_nlattrs(match, &attrs, a, is_mask, log);
 779         if (err)
 780                 return err;
 781
 782         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
 783                 const struct ovs_key_ethernet *eth_key;
 784
 785                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 786                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 787                                 eth_key->eth_src, ETH_ALEN, is_mask);
 788                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 789                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 790                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
 791         }
 792
 793         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
 794                 __be16 tci;
 795
 796                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 797                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 798                         if (is_mask)
 799                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 800                         else
 801                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 802
 803                         return -EINVAL;
 804                 }
 805
 806                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 807                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
 808         }
 809
 810         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
 811                 __be16 eth_type;
 812
 813                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 814                 if (is_mask) {
 815                         /* Always exact match EtherType. */
 816                         eth_type = htons(0xffff);
 817                 } else if (!eth_proto_is_802_3(eth_type)) {
 818                         OVS_NLERR(log, "EtherType %x is less than min %x",
 819                                   ntohs(eth_type), ETH_P_802_3_MIN);
 820                         return -EINVAL;
 821                 }
 822
 823                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 824                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
 825         } else if (!is_mask) {
 826                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 827         }
 828
 829         if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
 830                 const struct ovs_key_ipv4 *ipv4_key;
 831
 832                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 833                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 834                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 835                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 836                         return -EINVAL;
 837                 }
 838                 SW_FLOW_KEY_PUT(match, ip.proto,
 839                                 ipv4_key->ipv4_proto, is_mask);
 840                 SW_FLOW_KEY_PUT(match, ip.tos,
 841                                 ipv4_key->ipv4_tos, is_mask);
 842                 SW_FLOW_KEY_PUT(match, ip.ttl,
 843                                 ipv4_key->ipv4_ttl, is_mask);
 844                 SW_FLOW_KEY_PUT(match, ip.frag,
 845                                 ipv4_key->ipv4_frag, is_mask);
 846                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 847                                 ipv4_key->ipv4_src, is_mask);
 848                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 849                                 ipv4_key->ipv4_dst, is_mask);
 850                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
 851         }
 852
 853         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
 854                 const struct ovs_key_ipv6 *ipv6_key;
 855
 856                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 857                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 858                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 859                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 860                         return -EINVAL;
 861                 }
 862
 863                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 864                         OVS_NLERR(log,
 865                                   "Invalid IPv6 flow label value (value=%x, max=%x).",
 866                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 867                         return -EINVAL;
 868                 }
 869
 870                 SW_FLOW_KEY_PUT(match, ipv6.label,
 871                                 ipv6_key->ipv6_label, is_mask);
 872                 SW_FLOW_KEY_PUT(match, ip.proto,
 873                                 ipv6_key->ipv6_proto, is_mask);
 874                 SW_FLOW_KEY_PUT(match, ip.tos,
 875                                 ipv6_key->ipv6_tclass, is_mask);
 876                 SW_FLOW_KEY_PUT(match, ip.ttl,
 877                                 ipv6_key->ipv6_hlimit, is_mask);
 878                 SW_FLOW_KEY_PUT(match, ip.frag,
 879                                 ipv6_key->ipv6_frag, is_mask);
 880                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
 881                                 ipv6_key->ipv6_src,
 882                                 sizeof(match->key->ipv6.addr.src),
 883                                 is_mask);
 884                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
 885                                 ipv6_key->ipv6_dst,
 886                                 sizeof(match->key->ipv6.addr.dst),
 887                                 is_mask);
 888
 889                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
 890         }
 891
 892         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
 893                 const struct ovs_key_arp *arp_key;
 894
 895                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
 896                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
 897                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
 898                                   arp_key->arp_op);
 899                         return -EINVAL;
 900                 }
 901
 902                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 903                                 arp_key->arp_sip, is_mask);
 904                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 905                         arp_key->arp_tip, is_mask);
 906                 SW_FLOW_KEY_PUT(match, ip.proto,
 907                                 ntohs(arp_key->arp_op), is_mask);
 908                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
 909                                 arp_key->arp_sha, ETH_ALEN, is_mask);
 910                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
 911                                 arp_key->arp_tha, ETH_ALEN, is_mask);
 912
 913                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
 914         }
 915
 916         if (attrs & (1ULL << OVS_KEY_ATTR_MPLS)) {
 917                 const struct ovs_key_mpls *mpls_key;
 918
 919                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
 920                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
 921                                 mpls_key->mpls_lse, is_mask);
 922
 923                 attrs &= ~(1ULL << OVS_KEY_ATTR_MPLS);
 924         }
 925
 926         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
 927                 const struct ovs_key_tcp *tcp_key;
 928
 929                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 930                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
 931                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
 932                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
 933         }
 934
 935         if (attrs & (1ULL << OVS_KEY_ATTR_TCP_FLAGS)) {
 936                 SW_FLOW_KEY_PUT(match, tp.flags,
 937                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
 938                                 is_mask);
 939                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP_FLAGS);
 940         }
 941
 942         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
 943                 const struct ovs_key_udp *udp_key;
 944
 945                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 946                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
 947                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
 948                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
 949         }
 950
 951         if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
 952                 const struct ovs_key_sctp *sctp_key;
 953
 954                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
 955                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
 956                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
 957                 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
 958         }
 959
 960         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
 961                 const struct ovs_key_icmp *icmp_key;
 962
 963                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 964                 SW_FLOW_KEY_PUT(match, tp.src,
 965                                 htons(icmp_key->icmp_type), is_mask);
 966                 SW_FLOW_KEY_PUT(match, tp.dst,
 967                                 htons(icmp_key->icmp_code), is_mask);
 968                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
 969         }
 970
 971         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
 972                 const struct ovs_key_icmpv6 *icmpv6_key;
 973
 974                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 975                 SW_FLOW_KEY_PUT(match, tp.src,
 976                                 htons(icmpv6_key->icmpv6_type), is_mask);
 977                 SW_FLOW_KEY_PUT(match, tp.dst,
 978                                 htons(icmpv6_key->icmpv6_code), is_mask);
 979                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
 980         }
 981
 982         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
 983                 const struct ovs_key_nd *nd_key;
 984
 985                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 986                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
 987                         nd_key->nd_target,
 988                         sizeof(match->key->ipv6.nd.target),
 989                         is_mask);
 990                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
 991                         nd_key->nd_sll, ETH_ALEN, is_mask);
 992                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
 993                                 nd_key->nd_tll, ETH_ALEN, is_mask);
 994                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
 995         }
 996
 997         if (attrs != 0) {
 998                 OVS_NLERR(log, "Unknown key attributes %llx",
 999                           (unsigned long long)attrs);
1000                 return -EINVAL;
1001         }
1002
1003         return 0;
1004 }
1005
1006 static void nlattr_set(struct nlattr *attr, u8 val,
1007                        const struct ovs_len_tbl *tbl)
1008 {
1009         struct nlattr *nla;
1010         int rem;
1011
1012         /* The nlattr stream should already have been validated */
1013         nla_for_each_nested(nla, attr, rem) {
1014                 if (tbl && tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1015                         nlattr_set(nla, val, tbl[nla_type(nla)].next);
1016                 else
1017                         memset(nla_data(nla), val, nla_len(nla));
1018         }
1019 }
1020
1021 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1022 {
1023         nlattr_set(attr, val, ovs_key_lens);
1024 }
1025
1026 /**
1027  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1028  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1029  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1030  * does not include any don't care bit.
1031  * @match: receives the extracted flow match information.
1032  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1033  * sequence. The fields should of the packet that triggered the creation
1034  * of this flow.
1035  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1036  * attribute specifies the mask field of the wildcarded flow.
1037  * @log: Boolean to allow kernel error logging.  Normally true, but when
1038  * probing for feature compatibility this should be passed in as false to
1039  * suppress unnecessary error logging.
1040  */
1041 int ovs_nla_get_match(struct sw_flow_match *match,
1042                       const struct nlattr *nla_key,
1043                       const struct nlattr *nla_mask,
1044                       bool log)
1045 {
1046         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1047         const struct nlattr *encap;
1048         struct nlattr *newmask = NULL;
1049         u64 key_attrs = 0;
1050         u64 mask_attrs = 0;
1051         bool encap_valid = false;
1052         int err;
1053
1054         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1055         if (err)
1056                 return err;
1057
1058         if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1059             (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
1060             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1061                 __be16 tci;
1062
1063                 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
1064                       (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
1065                         OVS_NLERR(log, "Invalid Vlan frame.");
1066                         return -EINVAL;
1067                 }
1068
1069                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1070                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1071                 encap = a[OVS_KEY_ATTR_ENCAP];
1072                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1073                 encap_valid = true;
1074
1075                 if (tci & htons(VLAN_TAG_PRESENT)) {
1076                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1077                         if (err)
1078                                 return err;
1079                 } else if (!tci) {
1080                         /* Corner case for truncated 802.1Q header. */
1081                         if (nla_len(encap)) {
1082                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1083                                 return -EINVAL;
1084                         }
1085                 } else {
1086                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1087                         return  -EINVAL;
1088                 }
1089         }
1090
1091         err = ovs_key_from_nlattrs(match, key_attrs, a, false, log);
1092         if (err)
1093                 return err;
1094
1095         if (match->mask) {
1096                 if (!nla_mask) {
1097                         /* Create an exact match mask. We need to set to 0xff
1098                          * all the 'match->mask' fields that have been touched
1099                          * in 'match->key'. We cannot simply memset
1100                          * 'match->mask', because padding bytes and fields not
1101                          * specified in 'match->key' should be left to 0.
1102                          * Instead, we use a stream of netlink attributes,
1103                          * copied from 'key' and set to 0xff.
1104                          * ovs_key_from_nlattrs() will take care of filling
1105                          * 'match->mask' appropriately.
1106                          */
1107                         newmask = kmemdup(nla_key,
1108                                           nla_total_size(nla_len(nla_key)),
1109                                           GFP_KERNEL);
1110                         if (!newmask)
1111                                 return -ENOMEM;
1112
1113                         mask_set_nlattr(newmask, 0xff);
1114
1115                         /* The userspace does not send tunnel attributes that
1116                          * are 0, but we should not wildcard them nonetheless.
1117                          */
1118                         if (match->key->tun_key.ipv4_dst)
1119                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1120                                                          0xff, true);
1121
1122                         nla_mask = newmask;
1123                 }
1124
1125                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1126                 if (err)
1127                         goto free_newmask;
1128
1129                 /* Always match on tci. */
1130                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1131
1132                 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1133                         __be16 eth_type = 0;
1134                         __be16 tci = 0;
1135
1136                         if (!encap_valid) {
1137                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1138                                 err = -EINVAL;
1139                                 goto free_newmask;
1140                         }
1141
1142                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1143                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1144                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1145
1146                         if (eth_type == htons(0xffff)) {
1147                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1148                                 encap = a[OVS_KEY_ATTR_ENCAP];
1149                                 err = parse_flow_mask_nlattrs(encap, a,
1150                                                               &mask_attrs, log);
1151                                 if (err)
1152                                         goto free_newmask;
1153                         } else {
1154                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1155                                           ntohs(eth_type));
1156                                 err = -EINVAL;
1157                                 goto free_newmask;
1158                         }
1159
1160                         if (a[OVS_KEY_ATTR_VLAN])
1161                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1162
1163                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1164                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1165                                           ntohs(tci));
1166                                 err = -EINVAL;
1167                                 goto free_newmask;
1168                         }
1169                 }
1170
1171                 err = ovs_key_from_nlattrs(match, mask_attrs, a, true, log);
1172                 if (err)
1173                         goto free_newmask;
1174         }
1175
1176         if (!match_validate(match, key_attrs, mask_attrs, log))
1177                 err = -EINVAL;
1178
1179 free_newmask:
1180         kfree(newmask);
1181         return err;
1182 }
1183
1184 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1185 {
1186         size_t len;
1187
1188         if (!attr)
1189                 return 0;
1190
1191         len = nla_len(attr);
1192         if (len < 1 || len > MAX_UFID_LENGTH) {
1193                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1194                           nla_len(attr), MAX_UFID_LENGTH);
1195                 return 0;
1196         }
1197
1198         return len;
1199 }
1200
1201 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1202  * or false otherwise.
1203  */
1204 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1205                       bool log)
1206 {
1207         sfid->ufid_len = get_ufid_len(attr, log);
1208         if (sfid->ufid_len)
1209                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1210
1211         return sfid->ufid_len;
1212 }
1213
1214 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1215                            const struct sw_flow_key *key, bool log)
1216 {
1217         struct sw_flow_key *new_key;
1218
1219         if (ovs_nla_get_ufid(sfid, ufid, log))
1220                 return 0;
1221
1222         /* If UFID was not provided, use unmasked key. */
1223         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1224         if (!new_key)
1225                 return -ENOMEM;
1226         memcpy(new_key, key, sizeof(*key));
1227         sfid->unmasked_key = new_key;
1228
1229         return 0;
1230 }
1231
1232 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1233 {
1234         return attr ? nla_get_u32(attr) : 0;
1235 }
1236
1237 /**
1238  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1239  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1240  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1241  * sequence.
1242  * @log: Boolean to allow kernel error logging.  Normally true, but when
1243  * probing for feature compatibility this should be passed in as false to
1244  * suppress unnecessary error logging.
1245  *
1246  * This parses a series of Netlink attributes that form a flow key, which must
1247  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1248  * get the metadata, that is, the parts of the flow key that cannot be
1249  * extracted from the packet itself.
1250  */
1251
1252 int ovs_nla_get_flow_metadata(const struct nlattr *attr,
1253                               struct sw_flow_key *key,
1254                               bool log)
1255 {
1256         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1257         struct sw_flow_match match;
1258         u64 attrs = 0;
1259         int err;
1260
1261         err = parse_flow_nlattrs(attr, a, &attrs, log);
1262         if (err)
1263                 return -EINVAL;
1264
1265         memset(&match, 0, sizeof(match));
1266         match.key = key;
1267
1268         memset(key, 0, OVS_SW_FLOW_KEY_METADATA_SIZE);
1269         key->phy.in_port = DP_MAX_PORTS;
1270
1271         return metadata_from_nlattrs(&match, &attrs, a, false, log);
1272 }
1273
1274 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1275                              const struct sw_flow_key *output, bool is_mask,
1276                              struct sk_buff *skb)
1277 {
1278         struct ovs_key_ethernet *eth_key;
1279         struct nlattr *nla, *encap;
1280
1281         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1282                 goto nla_put_failure;
1283
1284         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1285                 goto nla_put_failure;
1286
1287         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1288                 goto nla_put_failure;
1289
1290         if ((swkey->tun_key.ipv4_dst || is_mask)) {
1291                 const void *opts = NULL;
1292
1293                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1294                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1295
1296                 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1297                                        swkey->tun_opts_len))
1298                         goto nla_put_failure;
1299         }
1300
1301         if (swkey->phy.in_port == DP_MAX_PORTS) {
1302                 if (is_mask && (output->phy.in_port == 0xffff))
1303                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1304                                 goto nla_put_failure;
1305         } else {
1306                 u16 upper_u16;
1307                 upper_u16 = !is_mask ? 0 : 0xffff;
1308
1309                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1310                                 (upper_u16 << 16) | output->phy.in_port))
1311                         goto nla_put_failure;
1312         }
1313
1314         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1315                 goto nla_put_failure;
1316
1317         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1318         if (!nla)
1319                 goto nla_put_failure;
1320
1321         eth_key = nla_data(nla);
1322         ether_addr_copy(eth_key->eth_src, output->eth.src);
1323         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1324
1325         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1326                 __be16 eth_type;
1327                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1328                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1329                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1330                         goto nla_put_failure;
1331                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1332                 if (!swkey->eth.tci)
1333                         goto unencap;
1334         } else
1335                 encap = NULL;
1336
1337         if (swkey->eth.type == htons(ETH_P_802_2)) {
1338                 /*
1339                  * Ethertype 802.2 is represented in the netlink with omitted
1340                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1341                  * 0xffff in the mask attribute.  Ethertype can also
1342                  * be wildcarded.
1343                  */
1344                 if (is_mask && output->eth.type)
1345                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1346                                                 output->eth.type))
1347                                 goto nla_put_failure;
1348                 goto unencap;
1349         }
1350
1351         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1352                 goto nla_put_failure;
1353
1354         if (swkey->eth.type == htons(ETH_P_IP)) {
1355                 struct ovs_key_ipv4 *ipv4_key;
1356
1357                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1358                 if (!nla)
1359                         goto nla_put_failure;
1360                 ipv4_key = nla_data(nla);
1361                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1362                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1363                 ipv4_key->ipv4_proto = output->ip.proto;
1364                 ipv4_key->ipv4_tos = output->ip.tos;
1365                 ipv4_key->ipv4_ttl = output->ip.ttl;
1366                 ipv4_key->ipv4_frag = output->ip.frag;
1367         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1368                 struct ovs_key_ipv6 *ipv6_key;
1369
1370                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1371                 if (!nla)
1372                         goto nla_put_failure;
1373                 ipv6_key = nla_data(nla);
1374                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1375                                 sizeof(ipv6_key->ipv6_src));
1376                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1377                                 sizeof(ipv6_key->ipv6_dst));
1378                 ipv6_key->ipv6_label = output->ipv6.label;
1379                 ipv6_key->ipv6_proto = output->ip.proto;
1380                 ipv6_key->ipv6_tclass = output->ip.tos;
1381                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1382                 ipv6_key->ipv6_frag = output->ip.frag;
1383         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1384                    swkey->eth.type == htons(ETH_P_RARP)) {
1385                 struct ovs_key_arp *arp_key;
1386
1387                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1388                 if (!nla)
1389                         goto nla_put_failure;
1390                 arp_key = nla_data(nla);
1391                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1392                 arp_key->arp_sip = output->ipv4.addr.src;
1393                 arp_key->arp_tip = output->ipv4.addr.dst;
1394                 arp_key->arp_op = htons(output->ip.proto);
1395                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1396                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1397         } else if (eth_p_mpls(swkey->eth.type)) {
1398                 struct ovs_key_mpls *mpls_key;
1399
1400                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1401                 if (!nla)
1402                         goto nla_put_failure;
1403                 mpls_key = nla_data(nla);
1404                 mpls_key->mpls_lse = output->mpls.top_lse;
1405         }
1406
1407         if ((swkey->eth.type == htons(ETH_P_IP) ||
1408              swkey->eth.type == htons(ETH_P_IPV6)) &&
1409              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1410
1411                 if (swkey->ip.proto == IPPROTO_TCP) {
1412                         struct ovs_key_tcp *tcp_key;
1413
1414                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1415                         if (!nla)
1416                                 goto nla_put_failure;
1417                         tcp_key = nla_data(nla);
1418                         tcp_key->tcp_src = output->tp.src;
1419                         tcp_key->tcp_dst = output->tp.dst;
1420                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1421                                          output->tp.flags))
1422                                 goto nla_put_failure;
1423                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1424                         struct ovs_key_udp *udp_key;
1425
1426                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1427                         if (!nla)
1428                                 goto nla_put_failure;
1429                         udp_key = nla_data(nla);
1430                         udp_key->udp_src = output->tp.src;
1431                         udp_key->udp_dst = output->tp.dst;
1432                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1433                         struct ovs_key_sctp *sctp_key;
1434
1435                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1436                         if (!nla)
1437                                 goto nla_put_failure;
1438                         sctp_key = nla_data(nla);
1439                         sctp_key->sctp_src = output->tp.src;
1440                         sctp_key->sctp_dst = output->tp.dst;
1441                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1442                            swkey->ip.proto == IPPROTO_ICMP) {
1443                         struct ovs_key_icmp *icmp_key;
1444
1445                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1446                         if (!nla)
1447                                 goto nla_put_failure;
1448                         icmp_key = nla_data(nla);
1449                         icmp_key->icmp_type = ntohs(output->tp.src);
1450                         icmp_key->icmp_code = ntohs(output->tp.dst);
1451                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1452                            swkey->ip.proto == IPPROTO_ICMPV6) {
1453                         struct ovs_key_icmpv6 *icmpv6_key;
1454
1455                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1456                                                 sizeof(*icmpv6_key));
1457                         if (!nla)
1458                                 goto nla_put_failure;
1459                         icmpv6_key = nla_data(nla);
1460                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1461                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1462
1463                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1464                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1465                                 struct ovs_key_nd *nd_key;
1466
1467                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1468                                 if (!nla)
1469                                         goto nla_put_failure;
1470                                 nd_key = nla_data(nla);
1471                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1472                                                         sizeof(nd_key->nd_target));
1473                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1474                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1475                         }
1476                 }
1477         }
1478
1479 unencap:
1480         if (encap)
1481                 nla_nest_end(skb, encap);
1482
1483         return 0;
1484
1485 nla_put_failure:
1486         return -EMSGSIZE;
1487 }
1488
1489 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1490                     const struct sw_flow_key *output, int attr, bool is_mask,
1491                     struct sk_buff *skb)
1492 {
1493         int err;
1494         struct nlattr *nla;
1495
1496         nla = nla_nest_start(skb, attr);
1497         if (!nla)
1498                 return -EMSGSIZE;
1499         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1500         if (err)
1501                 return err;
1502         nla_nest_end(skb, nla);
1503
1504         return 0;
1505 }
1506
1507 /* Called with ovs_mutex or RCU read lock. */
1508 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1509 {
1510         if (ovs_identifier_is_ufid(&flow->id))
1511                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1512                                flow->id.ufid);
1513
1514         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1515                                OVS_FLOW_ATTR_KEY, false, skb);
1516 }
1517
1518 /* Called with ovs_mutex or RCU read lock. */
1519 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1520 {
1521         return ovs_nla_put_key(&flow->key, &flow->key,
1522                                 OVS_FLOW_ATTR_KEY, false, skb);
1523 }
1524
1525 /* Called with ovs_mutex or RCU read lock. */
1526 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1527 {
1528         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1529                                 OVS_FLOW_ATTR_MASK, true, skb);
1530 }
1531
1532 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1533
1534 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1535 {
1536         struct sw_flow_actions *sfa;
1537
1538         if (size > MAX_ACTIONS_BUFSIZE) {
1539                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1540                 return ERR_PTR(-EINVAL);
1541         }
1542
1543         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1544         if (!sfa)
1545                 return ERR_PTR(-ENOMEM);
1546
1547         sfa->actions_len = 0;
1548         return sfa;
1549 }
1550
1551 /* RCU callback used by ovs_nla_free_flow_actions. */
1552 static void rcu_free_acts_callback(struct rcu_head *rcu)
1553 {
1554         struct sw_flow_actions *sf_acts = container_of(rcu,
1555                         struct sw_flow_actions, rcu);
1556         kfree(sf_acts);
1557 }
1558
1559 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1560  * The caller must hold rcu_read_lock for this to be sensible.
1561  */
1562 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1563 {
1564         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
1565 }
1566
1567 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1568                                        int attr_len, bool log)
1569 {
1570
1571         struct sw_flow_actions *acts;
1572         int new_acts_size;
1573         int req_size = NLA_ALIGN(attr_len);
1574         int next_offset = offsetof(struct sw_flow_actions, actions) +
1575                                         (*sfa)->actions_len;
1576
1577         if (req_size <= (ksize(*sfa) - next_offset))
1578                 goto out;
1579
1580         new_acts_size = ksize(*sfa) * 2;
1581
1582         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1583                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1584                         return ERR_PTR(-EMSGSIZE);
1585                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1586         }
1587
1588         acts = nla_alloc_flow_actions(new_acts_size, log);
1589         if (IS_ERR(acts))
1590                 return (void *)acts;
1591
1592         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1593         acts->actions_len = (*sfa)->actions_len;
1594         kfree(*sfa);
1595         *sfa = acts;
1596
1597 out:
1598         (*sfa)->actions_len += req_size;
1599         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1600 }
1601
1602 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1603                                    int attrtype, void *data, int len, bool log)
1604 {
1605         struct nlattr *a;
1606
1607         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1608         if (IS_ERR(a))
1609                 return a;
1610
1611         a->nla_type = attrtype;
1612         a->nla_len = nla_attr_size(len);
1613
1614         if (data)
1615                 memcpy(nla_data(a), data, len);
1616         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1617
1618         return a;
1619 }
1620
1621 static int add_action(struct sw_flow_actions **sfa, int attrtype,
1622                       void *data, int len, bool log)
1623 {
1624         struct nlattr *a;
1625
1626         a = __add_action(sfa, attrtype, data, len, log);
1627         if (IS_ERR(a))
1628                 return PTR_ERR(a);
1629
1630         return 0;
1631 }
1632
1633 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1634                                           int attrtype, bool log)
1635 {
1636         int used = (*sfa)->actions_len;
1637         int err;
1638
1639         err = add_action(sfa, attrtype, NULL, 0, log);
1640         if (err)
1641                 return err;
1642
1643         return used;
1644 }
1645
1646 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1647                                          int st_offset)
1648 {
1649         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1650                                                                st_offset);
1651
1652         a->nla_len = sfa->actions_len - st_offset;
1653 }
1654
1655 static int __ovs_nla_copy_actions(const struct nlattr *attr,
1656                                   const struct sw_flow_key *key,
1657                                   int depth, struct sw_flow_actions **sfa,
1658                                   __be16 eth_type, __be16 vlan_tci, bool log);
1659
1660 static int validate_and_copy_sample(const struct nlattr *attr,
1661                                     const struct sw_flow_key *key, int depth,
1662                                     struct sw_flow_actions **sfa,
1663                                     __be16 eth_type, __be16 vlan_tci, bool log)
1664 {
1665         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1666         const struct nlattr *probability, *actions;
1667         const struct nlattr *a;
1668         int rem, start, err, st_acts;
1669
1670         memset(attrs, 0, sizeof(attrs));
1671         nla_for_each_nested(a, attr, rem) {
1672                 int type = nla_type(a);
1673                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1674                         return -EINVAL;
1675                 attrs[type] = a;
1676         }
1677         if (rem)
1678                 return -EINVAL;
1679
1680         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1681         if (!probability || nla_len(probability) != sizeof(u32))
1682                 return -EINVAL;
1683
1684         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1685         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1686                 return -EINVAL;
1687
1688         /* validation done, copy sample action. */
1689         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1690         if (start < 0)
1691                 return start;
1692         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1693                          nla_data(probability), sizeof(u32), log);
1694         if (err)
1695                 return err;
1696         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1697         if (st_acts < 0)
1698                 return st_acts;
1699
1700         err = __ovs_nla_copy_actions(actions, key, depth + 1, sfa,
1701                                      eth_type, vlan_tci, log);
1702         if (err)
1703                 return err;
1704
1705         add_nested_action_end(*sfa, st_acts);
1706         add_nested_action_end(*sfa, start);
1707
1708         return 0;
1709 }
1710
1711 void ovs_match_init(struct sw_flow_match *match,
1712                     struct sw_flow_key *key,
1713                     struct sw_flow_mask *mask)
1714 {
1715         memset(match, 0, sizeof(*match));
1716         match->key = key;
1717         match->mask = mask;
1718
1719         memset(key, 0, sizeof(*key));
1720
1721         if (mask) {
1722                 memset(&mask->key, 0, sizeof(mask->key));
1723                 mask->range.start = mask->range.end = 0;
1724         }
1725 }
1726
1727 static int validate_geneve_opts(struct sw_flow_key *key)
1728 {
1729         struct geneve_opt *option;
1730         int opts_len = key->tun_opts_len;
1731         bool crit_opt = false;
1732
1733         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1734         while (opts_len > 0) {
1735                 int len;
1736
1737                 if (opts_len < sizeof(*option))
1738                         return -EINVAL;
1739
1740                 len = sizeof(*option) + option->length * 4;
1741                 if (len > opts_len)
1742                         return -EINVAL;
1743
1744                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1745
1746                 option = (struct geneve_opt *)((u8 *)option + len);
1747                 opts_len -= len;
1748         };
1749
1750         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1751
1752         return 0;
1753 }
1754
1755 static int validate_and_copy_set_tun(const struct nlattr *attr,
1756                                      struct sw_flow_actions **sfa, bool log)
1757 {
1758         struct sw_flow_match match;
1759         struct sw_flow_key key;
1760         struct ovs_tunnel_info *tun_info;
1761         struct nlattr *a;
1762         int start, opts_type;
1763         int err = 0;
1764
1765         ovs_match_init(&match, &key, NULL);
1766         opts_type = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1767         if (opts_type < 0)
1768                 return opts_type;
1769
1770         if (key.tun_opts_len) {
1771                 switch (opts_type) {
1772                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1773                         err = validate_geneve_opts(&key);
1774                         if (err < 0)
1775                                 return err;
1776                         break;
1777                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1778                         break;
1779                 }
1780         };
1781
1782         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1783         if (start < 0)
1784                 return start;
1785
1786         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1787                          sizeof(*tun_info) + key.tun_opts_len, log);
1788         if (IS_ERR(a))
1789                 return PTR_ERR(a);
1790
1791         tun_info = nla_data(a);
1792         tun_info->tunnel = key.tun_key;
1793         tun_info->options_len = key.tun_opts_len;
1794
1795         if (tun_info->options_len) {
1796                 /* We need to store the options in the action itself since
1797                  * everything else will go away after flow setup. We can append
1798                  * it to tun_info and then point there.
1799                  */
1800                 memcpy((tun_info + 1),
1801                        TUN_METADATA_OPTS(&key, key.tun_opts_len), key.tun_opts_len);
1802                 tun_info->options = (tun_info + 1);
1803         } else {
1804                 tun_info->options = NULL;
1805         }
1806
1807         add_nested_action_end(*sfa, start);
1808
1809         return err;
1810 }
1811
1812 /* Return false if there are any non-masked bits set.
1813  * Mask follows data immediately, before any netlink padding.
1814  */
1815 static bool validate_masked(u8 *data, int len)
1816 {
1817         u8 *mask = data + len;
1818
1819         while (len--)
1820                 if (*data++ & ~*mask++)
1821                         return false;
1822
1823         return true;
1824 }
1825
1826 static int validate_set(const struct nlattr *a,
1827                         const struct sw_flow_key *flow_key,
1828                         struct sw_flow_actions **sfa,
1829                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
1830 {
1831         const struct nlattr *ovs_key = nla_data(a);
1832         int key_type = nla_type(ovs_key);
1833         size_t key_len;
1834
1835         /* There can be only one key in a action */
1836         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1837                 return -EINVAL;
1838
1839         key_len = nla_len(ovs_key);
1840         if (masked)
1841                 key_len /= 2;
1842
1843         if (key_type > OVS_KEY_ATTR_MAX ||
1844             (ovs_key_lens[key_type].len != key_len &&
1845              ovs_key_lens[key_type].len != OVS_ATTR_NESTED))
1846                 return -EINVAL;
1847
1848         if (masked && !validate_masked(nla_data(ovs_key), key_len))
1849                 return -EINVAL;
1850
1851         switch (key_type) {
1852         const struct ovs_key_ipv4 *ipv4_key;
1853         const struct ovs_key_ipv6 *ipv6_key;
1854         int err;
1855
1856         case OVS_KEY_ATTR_PRIORITY:
1857         case OVS_KEY_ATTR_SKB_MARK:
1858         case OVS_KEY_ATTR_ETHERNET:
1859                 break;
1860
1861         case OVS_KEY_ATTR_TUNNEL:
1862                 if (eth_p_mpls(eth_type))
1863                         return -EINVAL;
1864
1865                 if (masked)
1866                         return -EINVAL; /* Masked tunnel set not supported. */
1867
1868                 *skip_copy = true;
1869                 err = validate_and_copy_set_tun(a, sfa, log);
1870                 if (err)
1871                         return err;
1872                 break;
1873
1874         case OVS_KEY_ATTR_IPV4:
1875                 if (eth_type != htons(ETH_P_IP))
1876                         return -EINVAL;
1877
1878                 ipv4_key = nla_data(ovs_key);
1879
1880                 if (masked) {
1881                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
1882
1883                         /* Non-writeable fields. */
1884                         if (mask->ipv4_proto || mask->ipv4_frag)
1885                                 return -EINVAL;
1886                 } else {
1887                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1888                                 return -EINVAL;
1889
1890                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1891                                 return -EINVAL;
1892                 }
1893                 break;
1894
1895         case OVS_KEY_ATTR_IPV6:
1896                 if (eth_type != htons(ETH_P_IPV6))
1897                         return -EINVAL;
1898
1899                 ipv6_key = nla_data(ovs_key);
1900
1901                 if (masked) {
1902                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
1903
1904                         /* Non-writeable fields. */
1905                         if (mask->ipv6_proto || mask->ipv6_frag)
1906                                 return -EINVAL;
1907
1908                         /* Invalid bits in the flow label mask? */
1909                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
1910                                 return -EINVAL;
1911                 } else {
1912                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1913                                 return -EINVAL;
1914
1915                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1916                                 return -EINVAL;
1917                 }
1918                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
1919                         return -EINVAL;
1920
1921                 break;
1922
1923         case OVS_KEY_ATTR_TCP:
1924                 if ((eth_type != htons(ETH_P_IP) &&
1925                      eth_type != htons(ETH_P_IPV6)) ||
1926                     flow_key->ip.proto != IPPROTO_TCP)
1927                         return -EINVAL;
1928
1929                 break;
1930
1931         case OVS_KEY_ATTR_UDP:
1932                 if ((eth_type != htons(ETH_P_IP) &&
1933                      eth_type != htons(ETH_P_IPV6)) ||
1934                     flow_key->ip.proto != IPPROTO_UDP)
1935                         return -EINVAL;
1936
1937                 break;
1938
1939         case OVS_KEY_ATTR_MPLS:
1940                 if (!eth_p_mpls(eth_type))
1941                         return -EINVAL;
1942                 break;
1943
1944         case OVS_KEY_ATTR_SCTP:
1945                 if ((eth_type != htons(ETH_P_IP) &&
1946                      eth_type != htons(ETH_P_IPV6)) ||
1947                     flow_key->ip.proto != IPPROTO_SCTP)
1948                         return -EINVAL;
1949
1950                 break;
1951
1952         default:
1953                 return -EINVAL;
1954         }
1955
1956         /* Convert non-masked non-tunnel set actions to masked set actions. */
1957         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
1958                 int start, len = key_len * 2;
1959                 struct nlattr *at;
1960
1961                 *skip_copy = true;
1962
1963                 start = add_nested_action_start(sfa,
1964                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
1965                                                 log);
1966                 if (start < 0)
1967                         return start;
1968
1969                 at = __add_action(sfa, key_type, NULL, len, log);
1970                 if (IS_ERR(at))
1971                         return PTR_ERR(at);
1972
1973                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
1974                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
1975                 /* Clear non-writeable bits from otherwise writeable fields. */
1976                 if (key_type == OVS_KEY_ATTR_IPV6) {
1977                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
1978
1979                         mask->ipv6_label &= htonl(0x000FFFFF);
1980                 }
1981                 add_nested_action_end(*sfa, start);
1982         }
1983
1984         return 0;
1985 }
1986
1987 static int validate_userspace(const struct nlattr *attr)
1988 {
1989         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
1990                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
1991                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
1992                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
1993         };
1994         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
1995         int error;
1996
1997         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
1998                                  attr, userspace_policy);
1999         if (error)
2000                 return error;
2001
2002         if (!a[OVS_USERSPACE_ATTR_PID] ||
2003             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2004                 return -EINVAL;
2005
2006         return 0;
2007 }
2008
2009 static int copy_action(const struct nlattr *from,
2010                        struct sw_flow_actions **sfa, bool log)
2011 {
2012         int totlen = NLA_ALIGN(from->nla_len);
2013         struct nlattr *to;
2014
2015         to = reserve_sfa_size(sfa, from->nla_len, log);
2016         if (IS_ERR(to))
2017                 return PTR_ERR(to);
2018
2019         memcpy(to, from, totlen);
2020         return 0;
2021 }
2022
2023 static int __ovs_nla_copy_actions(const struct nlattr *attr,
2024                                   const struct sw_flow_key *key,
2025                                   int depth, struct sw_flow_actions **sfa,
2026                                   __be16 eth_type, __be16 vlan_tci, bool log)
2027 {
2028         const struct nlattr *a;
2029         int rem, err;
2030
2031         if (depth >= SAMPLE_ACTION_DEPTH)
2032                 return -EOVERFLOW;
2033
2034         nla_for_each_nested(a, attr, rem) {
2035                 /* Expected argument lengths, (u32)-1 for variable length. */
2036                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2037                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2038                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2039                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2040                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2041                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2042                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2043                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2044                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2045                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2046                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2047                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash)
2048                 };
2049                 const struct ovs_action_push_vlan *vlan;
2050                 int type = nla_type(a);
2051                 bool skip_copy;
2052
2053                 if (type > OVS_ACTION_ATTR_MAX ||
2054                     (action_lens[type] != nla_len(a) &&
2055                      action_lens[type] != (u32)-1))
2056                         return -EINVAL;
2057
2058                 skip_copy = false;
2059                 switch (type) {
2060                 case OVS_ACTION_ATTR_UNSPEC:
2061                         return -EINVAL;
2062
2063                 case OVS_ACTION_ATTR_USERSPACE:
2064                         err = validate_userspace(a);
2065                         if (err)
2066                                 return err;
2067                         break;
2068
2069                 case OVS_ACTION_ATTR_OUTPUT:
2070                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2071                                 return -EINVAL;
2072                         break;
2073
2074                 case OVS_ACTION_ATTR_HASH: {
2075                         const struct ovs_action_hash *act_hash = nla_data(a);
2076
2077                         switch (act_hash->hash_alg) {
2078                         case OVS_HASH_ALG_L4:
2079                                 break;
2080                         default:
2081                                 return  -EINVAL;
2082                         }
2083
2084                         break;
2085                 }
2086
2087                 case OVS_ACTION_ATTR_POP_VLAN:
2088                         vlan_tci = htons(0);
2089                         break;
2090
2091                 case OVS_ACTION_ATTR_PUSH_VLAN:
2092                         vlan = nla_data(a);
2093                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2094                                 return -EINVAL;
2095                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2096                                 return -EINVAL;
2097                         vlan_tci = vlan->vlan_tci;
2098                         break;
2099
2100                 case OVS_ACTION_ATTR_RECIRC:
2101                         break;
2102
2103                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2104                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2105
2106                         if (!eth_p_mpls(mpls->mpls_ethertype))
2107                                 return -EINVAL;
2108                         /* Prohibit push MPLS other than to a white list
2109                          * for packets that have a known tag order.
2110                          */
2111                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2112                             (eth_type != htons(ETH_P_IP) &&
2113                              eth_type != htons(ETH_P_IPV6) &&
2114                              eth_type != htons(ETH_P_ARP) &&
2115                              eth_type != htons(ETH_P_RARP) &&
2116                              !eth_p_mpls(eth_type)))
2117                                 return -EINVAL;
2118                         eth_type = mpls->mpls_ethertype;
2119                         break;
2120                 }
2121
2122                 case OVS_ACTION_ATTR_POP_MPLS:
2123                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2124                             !eth_p_mpls(eth_type))
2125                                 return -EINVAL;
2126
2127                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2128                          * as there is no check here to ensure that the new
2129                          * eth_type is valid and thus set actions could
2130                          * write off the end of the packet or otherwise
2131                          * corrupt it.
2132                          *
2133                          * Support for these actions is planned using packet
2134                          * recirculation.
2135                          */
2136                         eth_type = htons(0);
2137                         break;
2138
2139                 case OVS_ACTION_ATTR_SET:
2140                         err = validate_set(a, key, sfa,
2141                                            &skip_copy, eth_type, false, log);
2142                         if (err)
2143                                 return err;
2144                         break;
2145
2146                 case OVS_ACTION_ATTR_SET_MASKED:
2147                         err = validate_set(a, key, sfa,
2148                                            &skip_copy, eth_type, true, log);
2149                         if (err)
2150                                 return err;
2151                         break;
2152
2153                 case OVS_ACTION_ATTR_SAMPLE:
2154                         err = validate_and_copy_sample(a, key, depth, sfa,
2155                                                        eth_type, vlan_tci, log);
2156                         if (err)
2157                                 return err;
2158                         skip_copy = true;
2159                         break;
2160
2161                 default:
2162                         OVS_NLERR(log, "Unknown Action type %d", type);
2163                         return -EINVAL;
2164                 }
2165                 if (!skip_copy) {
2166                         err = copy_action(a, sfa, log);
2167                         if (err)
2168                                 return err;
2169                 }
2170         }
2171
2172         if (rem > 0)
2173                 return -EINVAL;
2174
2175         return 0;
2176 }
2177
2178 /* 'key' must be the masked key. */
2179 int ovs_nla_copy_actions(const struct nlattr *attr,
2180                          const struct sw_flow_key *key,
2181                          struct sw_flow_actions **sfa, bool log)
2182 {
2183         int err;
2184
2185         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2186         if (IS_ERR(*sfa))
2187                 return PTR_ERR(*sfa);
2188
2189         err = __ovs_nla_copy_actions(attr, key, 0, sfa, key->eth.type,
2190                                      key->eth.tci, log);
2191         if (err)
2192                 kfree(*sfa);
2193
2194         return err;
2195 }
2196
2197 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2198 {
2199         const struct nlattr *a;
2200         struct nlattr *start;
2201         int err = 0, rem;
2202
2203         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2204         if (!start)
2205                 return -EMSGSIZE;
2206
2207         nla_for_each_nested(a, attr, rem) {
2208                 int type = nla_type(a);
2209                 struct nlattr *st_sample;
2210
2211                 switch (type) {
2212                 case OVS_SAMPLE_ATTR_PROBABILITY:
2213                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2214                                     sizeof(u32), nla_data(a)))
2215                                 return -EMSGSIZE;
2216                         break;
2217                 case OVS_SAMPLE_ATTR_ACTIONS:
2218                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2219                         if (!st_sample)
2220                                 return -EMSGSIZE;
2221                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2222                         if (err)
2223                                 return err;
2224                         nla_nest_end(skb, st_sample);
2225                         break;
2226                 }
2227         }
2228
2229         nla_nest_end(skb, start);
2230         return err;
2231 }
2232
2233 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2234 {
2235         const struct nlattr *ovs_key = nla_data(a);
2236         int key_type = nla_type(ovs_key);
2237         struct nlattr *start;
2238         int err;
2239
2240         switch (key_type) {
2241         case OVS_KEY_ATTR_TUNNEL_INFO: {
2242                 struct ovs_tunnel_info *tun_info = nla_data(ovs_key);
2243
2244                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2245                 if (!start)
2246                         return -EMSGSIZE;
2247
2248                 err = ipv4_tun_to_nlattr(skb, &tun_info->tunnel,
2249                                          tun_info->options_len ?
2250                                                 tun_info->options : NULL,
2251                                          tun_info->options_len);
2252                 if (err)
2253                         return err;
2254                 nla_nest_end(skb, start);
2255                 break;
2256         }
2257         default:
2258                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2259                         return -EMSGSIZE;
2260                 break;
2261         }
2262
2263         return 0;
2264 }
2265
2266 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2267                                                 struct sk_buff *skb)
2268 {
2269         const struct nlattr *ovs_key = nla_data(a);
2270         size_t key_len = nla_len(ovs_key) / 2;
2271
2272         /* Revert the conversion we did from a non-masked set action to
2273          * masked set action.
2274          */
2275         if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a) - key_len, ovs_key))
2276                 return -EMSGSIZE;
2277
2278         return 0;
2279 }
2280
2281 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2282 {
2283         const struct nlattr *a;
2284         int rem, err;
2285
2286         nla_for_each_attr(a, attr, len, rem) {
2287                 int type = nla_type(a);
2288
2289                 switch (type) {
2290                 case OVS_ACTION_ATTR_SET:
2291                         err = set_action_to_attr(a, skb);
2292                         if (err)
2293                                 return err;
2294                         break;
2295
2296                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2297                         err = masked_set_action_to_set_action_attr(a, skb);
2298                         if (err)
2299                                 return err;
2300                         break;
2301
2302                 case OVS_ACTION_ATTR_SAMPLE:
2303                         err = sample_action_to_attr(a, skb);
2304                         if (err)
2305                                 return err;
2306                         break;
2307                 default:
2308                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2309                                 return -EMSGSIZE;
2310                         break;
2311                 }
2312         }
2313
2314         return 0;
2315 }