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 static bool is_all_zero(const u8 *fp, size_t size)
 344 {
 345         int i;
 346
 347         if (!fp)
 348                 return false;
 349
 350         for (i = 0; i < size; i++)
 351                 if (fp[i])
 352                         return false;
 353
 354         return true;
 355 }
 356
 357 static int __parse_flow_nlattrs(const struct nlattr *attr,
 358                                 const struct nlattr *a[],
 359                                 u64 *attrsp, bool log, bool nz)
 360 {
 361         const struct nlattr *nla;
 362         u64 attrs;
 363         int rem;
 364
 365         attrs = *attrsp;
 366         nla_for_each_nested(nla, attr, rem) {
 367                 u16 type = nla_type(nla);
 368                 int expected_len;
 369
 370                 if (type > OVS_KEY_ATTR_MAX) {
 371                         OVS_NLERR(log, "Key type %d is out of range max %d",
 372                                   type, OVS_KEY_ATTR_MAX);
 373                         return -EINVAL;
 374                 }
 375
 376                 if (attrs & (1ULL << type)) {
 377                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 378                         return -EINVAL;
 379                 }
 380
 381                 expected_len = ovs_key_lens[type].len;
 382                 if (nla_len(nla) != expected_len && expected_len != OVS_ATTR_NESTED) {
 383                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 384                                   type, nla_len(nla), expected_len);
 385                         return -EINVAL;
 386                 }
 387
 388                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 389                         attrs |= 1ULL << type;
 390                         a[type] = nla;
 391                 }
 392         }
 393         if (rem) {
 394                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 395                 return -EINVAL;
 396         }
 397
 398         *attrsp = attrs;
 399         return 0;
 400 }
 401
 402 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 403                                    const struct nlattr *a[], u64 *attrsp,
 404                                    bool log)
 405 {
 406         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 407 }
 408
 409 static int parse_flow_nlattrs(const struct nlattr *attr,
 410                               const struct nlattr *a[], u64 *attrsp,
 411                               bool log)
 412 {
 413         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 414 }
 415
 416 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 417                                      struct sw_flow_match *match, bool is_mask,
 418                                      bool log)
 419 {
 420         unsigned long opt_key_offset;
 421
 422         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 423                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 424                           nla_len(a), sizeof(match->key->tun_opts));
 425                 return -EINVAL;
 426         }
 427
 428         if (nla_len(a) % 4 != 0) {
 429                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 430                           nla_len(a));
 431                 return -EINVAL;
 432         }
 433
 434         /* We need to record the length of the options passed
 435          * down, otherwise packets with the same format but
 436          * additional options will be silently matched.
 437          */
 438         if (!is_mask) {
 439                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 440                                 false);
 441         } else {
 442                 /* This is somewhat unusual because it looks at
 443                  * both the key and mask while parsing the
 444                  * attributes (and by extension assumes the key
 445                  * is parsed first). Normally, we would verify
 446                  * that each is the correct length and that the
 447                  * attributes line up in the validate function.
 448                  * However, that is difficult because this is
 449                  * variable length and we won't have the
 450                  * information later.
 451                  */
 452                 if (match->key->tun_opts_len != nla_len(a)) {
 453                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 454                                   match->key->tun_opts_len, nla_len(a));
 455                         return -EINVAL;
 456                 }
 457
 458                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 459         }
 460
 461         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 462         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 463                                   nla_len(a), is_mask);
 464         return 0;
 465 }
 466
 467 static const struct nla_policy vxlan_opt_policy[OVS_VXLAN_EXT_MAX + 1] = {
 468         [OVS_VXLAN_EXT_GBP]     = { .type = NLA_U32 },
 469 };
 470
 471 static int vxlan_tun_opt_from_nlattr(const struct nlattr *a,
 472                                      struct sw_flow_match *match, bool is_mask,
 473                                      bool log)
 474 {
 475         struct nlattr *tb[OVS_VXLAN_EXT_MAX+1];
 476         unsigned long opt_key_offset;
 477         struct ovs_vxlan_opts opts;
 478         int err;
 479
 480         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 481
 482         err = nla_parse_nested(tb, OVS_VXLAN_EXT_MAX, a, vxlan_opt_policy);
 483         if (err < 0)
 484                 return err;
 485
 486         memset(&opts, 0, sizeof(opts));
 487
 488         if (tb[OVS_VXLAN_EXT_GBP])
 489                 opts.gbp = nla_get_u32(tb[OVS_VXLAN_EXT_GBP]);
 490
 491         if (!is_mask)
 492                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 493         else
 494                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 495
 496         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 497         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 498                                   is_mask);
 499         return 0;
 500 }
 501
 502 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
 503                                 struct sw_flow_match *match, bool is_mask,
 504                                 bool log)
 505 {
 506         struct nlattr *a;
 507         int rem;
 508         bool ttl = false;
 509         __be16 tun_flags = 0;
 510         int opts_type = 0;
 511
 512         nla_for_each_nested(a, attr, rem) {
 513                 int type = nla_type(a);
 514                 int err;
 515
 516                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 517                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 518                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 519                         return -EINVAL;
 520                 }
 521
 522                 if (ovs_tunnel_key_lens[type].len != nla_len(a) &&
 523                    ovs_tunnel_key_lens[type].len != OVS_ATTR_NESTED) {
 524                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 525                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 526                         return -EINVAL;
 527                 }
 528
 529                 switch (type) {
 530                 case OVS_TUNNEL_KEY_ATTR_ID:
 531                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 532                                         nla_get_be64(a), is_mask);
 533                         tun_flags |= TUNNEL_KEY;
 534                         break;
 535                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 536                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
 537                                         nla_get_be32(a), is_mask);
 538                         break;
 539                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 540                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
 541                                         nla_get_be32(a), is_mask);
 542                         break;
 543                 case OVS_TUNNEL_KEY_ATTR_TOS:
 544                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
 545                                         nla_get_u8(a), is_mask);
 546                         break;
 547                 case OVS_TUNNEL_KEY_ATTR_TTL:
 548                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
 549                                         nla_get_u8(a), is_mask);
 550                         ttl = true;
 551                         break;
 552                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 553                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 554                         break;
 555                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 556                         tun_flags |= TUNNEL_CSUM;
 557                         break;
 558                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 559                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 560                                         nla_get_be16(a), is_mask);
 561                         break;
 562                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 563                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 564                                         nla_get_be16(a), is_mask);
 565                         break;
 566                 case OVS_TUNNEL_KEY_ATTR_OAM:
 567                         tun_flags |= TUNNEL_OAM;
 568                         break;
 569                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 570                         if (opts_type) {
 571                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 572                                 return -EINVAL;
 573                         }
 574
 575                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 576                         if (err)
 577                                 return err;
 578
 579                         tun_flags |= TUNNEL_GENEVE_OPT;
 580                         opts_type = type;
 581                         break;
 582                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 583                         if (opts_type) {
 584                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 585                                 return -EINVAL;
 586                         }
 587
 588                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 589                         if (err)
 590                                 return err;
 591
 592                         tun_flags |= TUNNEL_VXLAN_OPT;
 593                         opts_type = type;
 594                         break;
 595                 default:
 596                         OVS_NLERR(log, "Unknown IPv4 tunnel attribute %d",
 597                                   type);
 598                         return -EINVAL;
 599                 }
 600         }
 601
 602         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 603
 604         if (rem > 0) {
 605                 OVS_NLERR(log, "IPv4 tunnel attribute has %d unknown bytes.",
 606                           rem);
 607                 return -EINVAL;
 608         }
 609
 610         if (!is_mask) {
 611                 if (!match->key->tun_key.ipv4_dst) {
 612                         OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 613                         return -EINVAL;
 614                 }
 615
 616                 if (!ttl) {
 617                         OVS_NLERR(log, "IPv4 tunnel TTL not specified.");
 618                         return -EINVAL;
 619                 }
 620         }
 621
 622         return opts_type;
 623 }
 624
 625 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 626                                const void *tun_opts, int swkey_tun_opts_len)
 627 {
 628         const struct ovs_vxlan_opts *opts = tun_opts;
 629         struct nlattr *nla;
 630
 631         nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 632         if (!nla)
 633                 return -EMSGSIZE;
 634
 635         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 636                 return -EMSGSIZE;
 637
 638         nla_nest_end(skb, nla);
 639         return 0;
 640 }
 641
 642 static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
 643                                 const struct ovs_key_ipv4_tunnel *output,
 644                                 const void *tun_opts, int swkey_tun_opts_len)
 645 {
 646         if (output->tun_flags & TUNNEL_KEY &&
 647             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 648                 return -EMSGSIZE;
 649         if (output->ipv4_src &&
 650             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
 651                 return -EMSGSIZE;
 652         if (output->ipv4_dst &&
 653             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
 654                 return -EMSGSIZE;
 655         if (output->ipv4_tos &&
 656             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
 657                 return -EMSGSIZE;
 658         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
 659                 return -EMSGSIZE;
 660         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 661             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 662                 return -EMSGSIZE;
 663         if ((output->tun_flags & TUNNEL_CSUM) &&
 664             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 665                 return -EMSGSIZE;
 666         if (output->tp_src &&
 667             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 668                 return -EMSGSIZE;
 669         if (output->tp_dst &&
 670             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 671                 return -EMSGSIZE;
 672         if ((output->tun_flags & TUNNEL_OAM) &&
 673             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 674                 return -EMSGSIZE;
 675         if (tun_opts) {
 676                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 677                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 678                             swkey_tun_opts_len, tun_opts))
 679                         return -EMSGSIZE;
 680                else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 681                         vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 682                         return -EMSGSIZE;
 683         }
 684
 685         return 0;
 686 }
 687
 688 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
 689                               const struct ovs_key_ipv4_tunnel *output,
 690                               const void *tun_opts, int swkey_tun_opts_len)
 691 {
 692         struct nlattr *nla;
 693         int err;
 694
 695         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 696         if (!nla)
 697                 return -EMSGSIZE;
 698
 699         err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
 700         if (err)
 701                 return err;
 702
 703         nla_nest_end(skb, nla);
 704         return 0;
 705 }
 706
 707 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
 708                                   const struct ovs_tunnel_info *egress_tun_info)
 709 {
 710         return __ipv4_tun_to_nlattr(skb, &egress_tun_info->tunnel,
 711                                     egress_tun_info->options,
 712                                     egress_tun_info->options_len);
 713 }
 714
 715 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
 716                                  const struct nlattr **a, bool is_mask,
 717                                  bool log)
 718 {
 719         if (*attrs & (1ULL << OVS_KEY_ATTR_DP_HASH)) {
 720                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 721
 722                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 723                 *attrs &= ~(1ULL << OVS_KEY_ATTR_DP_HASH);
 724         }
 725
 726         if (*attrs & (1ULL << OVS_KEY_ATTR_RECIRC_ID)) {
 727                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 728
 729                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 730                 *attrs &= ~(1ULL << OVS_KEY_ATTR_RECIRC_ID);
 731         }
 732
 733         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
 734                 SW_FLOW_KEY_PUT(match, phy.priority,
 735                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 736                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
 737         }
 738
 739         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
 740                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 741
 742                 if (is_mask) {
 743                         in_port = 0xffffffff; /* Always exact match in_port. */
 744                 } else if (in_port >= DP_MAX_PORTS) {
 745                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 746                                   in_port, DP_MAX_PORTS);
 747                         return -EINVAL;
 748                 }
 749
 750                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 751                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
 752         } else if (!is_mask) {
 753                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 754         }
 755
 756         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
 757                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 758
 759                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 760                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
 761         }
 762         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
 763                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 764                                          is_mask, log) < 0)
 765                         return -EINVAL;
 766                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
 767         }
 768         return 0;
 769 }
 770
 771 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
 772                                 const struct nlattr **a, bool is_mask,
 773                                 bool log)
 774 {
 775         int err;
 776
 777         err = metadata_from_nlattrs(match, &attrs, a, is_mask, log);
 778         if (err)
 779                 return err;
 780
 781         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
 782                 const struct ovs_key_ethernet *eth_key;
 783
 784                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 785                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 786                                 eth_key->eth_src, ETH_ALEN, is_mask);
 787                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 788                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 789                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
 790         }
 791
 792         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
 793                 __be16 tci;
 794
 795                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 796                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 797                         if (is_mask)
 798                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 799                         else
 800                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 801
 802                         return -EINVAL;
 803                 }
 804
 805                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 806                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
 807         }
 808
 809         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
 810                 __be16 eth_type;
 811
 812                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 813                 if (is_mask) {
 814                         /* Always exact match EtherType. */
 815                         eth_type = htons(0xffff);
 816                 } else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
 817                         OVS_NLERR(log, "EtherType %x is less than min %x",
 818                                   ntohs(eth_type), ETH_P_802_3_MIN);
 819                         return -EINVAL;
 820                 }
 821
 822                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 823                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
 824         } else if (!is_mask) {
 825                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 826         }
 827
 828         if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
 829                 const struct ovs_key_ipv4 *ipv4_key;
 830
 831                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 832                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 833                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 834                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 835                         return -EINVAL;
 836                 }
 837                 SW_FLOW_KEY_PUT(match, ip.proto,
 838                                 ipv4_key->ipv4_proto, is_mask);
 839                 SW_FLOW_KEY_PUT(match, ip.tos,
 840                                 ipv4_key->ipv4_tos, is_mask);
 841                 SW_FLOW_KEY_PUT(match, ip.ttl,
 842                                 ipv4_key->ipv4_ttl, is_mask);
 843                 SW_FLOW_KEY_PUT(match, ip.frag,
 844                                 ipv4_key->ipv4_frag, is_mask);
 845                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 846                                 ipv4_key->ipv4_src, is_mask);
 847                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 848                                 ipv4_key->ipv4_dst, is_mask);
 849                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
 850         }
 851
 852         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
 853                 const struct ovs_key_ipv6 *ipv6_key;
 854
 855                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 856                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 857                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 858                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 859                         return -EINVAL;
 860                 }
 861                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 862                         OVS_NLERR(log,
 863                                   "Invalid IPv6 flow label value (value=%x, max=%x).",
 864                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 865                         return -EINVAL;
 866                 }
 867                 SW_FLOW_KEY_PUT(match, ipv6.label,
 868                                 ipv6_key->ipv6_label, is_mask);
 869                 SW_FLOW_KEY_PUT(match, ip.proto,
 870                                 ipv6_key->ipv6_proto, is_mask);
 871                 SW_FLOW_KEY_PUT(match, ip.tos,
 872                                 ipv6_key->ipv6_tclass, is_mask);
 873                 SW_FLOW_KEY_PUT(match, ip.ttl,
 874                                 ipv6_key->ipv6_hlimit, is_mask);
 875                 SW_FLOW_KEY_PUT(match, ip.frag,
 876                                 ipv6_key->ipv6_frag, is_mask);
 877                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
 878                                 ipv6_key->ipv6_src,
 879                                 sizeof(match->key->ipv6.addr.src),
 880                                 is_mask);
 881                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
 882                                 ipv6_key->ipv6_dst,
 883                                 sizeof(match->key->ipv6.addr.dst),
 884                                 is_mask);
 885
 886                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
 887         }
 888
 889         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
 890                 const struct ovs_key_arp *arp_key;
 891
 892                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
 893                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
 894                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
 895                                   arp_key->arp_op);
 896                         return -EINVAL;
 897                 }
 898
 899                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 900                                 arp_key->arp_sip, is_mask);
 901                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 902                         arp_key->arp_tip, is_mask);
 903                 SW_FLOW_KEY_PUT(match, ip.proto,
 904                                 ntohs(arp_key->arp_op), is_mask);
 905                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
 906                                 arp_key->arp_sha, ETH_ALEN, is_mask);
 907                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
 908                                 arp_key->arp_tha, ETH_ALEN, is_mask);
 909
 910                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
 911         }
 912
 913         if (attrs & (1ULL << OVS_KEY_ATTR_MPLS)) {
 914                 const struct ovs_key_mpls *mpls_key;
 915
 916                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
 917                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
 918                                 mpls_key->mpls_lse, is_mask);
 919
 920                 attrs &= ~(1ULL << OVS_KEY_ATTR_MPLS);
 921         }
 922
 923         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
 924                 const struct ovs_key_tcp *tcp_key;
 925
 926                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 927                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
 928                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
 929                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
 930         }
 931
 932         if (attrs & (1ULL << OVS_KEY_ATTR_TCP_FLAGS)) {
 933                 SW_FLOW_KEY_PUT(match, tp.flags,
 934                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
 935                                 is_mask);
 936                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP_FLAGS);
 937         }
 938
 939         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
 940                 const struct ovs_key_udp *udp_key;
 941
 942                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 943                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
 944                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
 945                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
 946         }
 947
 948         if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
 949                 const struct ovs_key_sctp *sctp_key;
 950
 951                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
 952                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
 953                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
 954                 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
 955         }
 956
 957         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
 958                 const struct ovs_key_icmp *icmp_key;
 959
 960                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 961                 SW_FLOW_KEY_PUT(match, tp.src,
 962                                 htons(icmp_key->icmp_type), is_mask);
 963                 SW_FLOW_KEY_PUT(match, tp.dst,
 964                                 htons(icmp_key->icmp_code), is_mask);
 965                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
 966         }
 967
 968         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
 969                 const struct ovs_key_icmpv6 *icmpv6_key;
 970
 971                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 972                 SW_FLOW_KEY_PUT(match, tp.src,
 973                                 htons(icmpv6_key->icmpv6_type), is_mask);
 974                 SW_FLOW_KEY_PUT(match, tp.dst,
 975                                 htons(icmpv6_key->icmpv6_code), is_mask);
 976                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
 977         }
 978
 979         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
 980                 const struct ovs_key_nd *nd_key;
 981
 982                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 983                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
 984                         nd_key->nd_target,
 985                         sizeof(match->key->ipv6.nd.target),
 986                         is_mask);
 987                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
 988                         nd_key->nd_sll, ETH_ALEN, is_mask);
 989                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
 990                                 nd_key->nd_tll, ETH_ALEN, is_mask);
 991                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
 992         }
 993
 994         if (attrs != 0) {
 995                 OVS_NLERR(log, "Unknown key attributes %llx",
 996                           (unsigned long long)attrs);
 997                 return -EINVAL;
 998         }
 999
1000         return 0;
1001 }
1002
1003 static void nlattr_set(struct nlattr *attr, u8 val,
1004                        const struct ovs_len_tbl *tbl)
1005 {
1006         struct nlattr *nla;
1007         int rem;
1008
1009         /* The nlattr stream should already have been validated */
1010         nla_for_each_nested(nla, attr, rem) {
1011                 if (tbl && tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1012                         nlattr_set(nla, val, tbl[nla_type(nla)].next);
1013                 else
1014                         memset(nla_data(nla), val, nla_len(nla));
1015         }
1016 }
1017
1018 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1019 {
1020         nlattr_set(attr, val, ovs_key_lens);
1021 }
1022
1023 /**
1024  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1025  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1026  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1027  * does not include any don't care bit.
1028  * @match: receives the extracted flow match information.
1029  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1030  * sequence. The fields should of the packet that triggered the creation
1031  * of this flow.
1032  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1033  * attribute specifies the mask field of the wildcarded flow.
1034  * @log: Boolean to allow kernel error logging.  Normally true, but when
1035  * probing for feature compatibility this should be passed in as false to
1036  * suppress unnecessary error logging.
1037  */
1038 int ovs_nla_get_match(struct sw_flow_match *match,
1039                       const struct nlattr *nla_key,
1040                       const struct nlattr *nla_mask,
1041                       bool log)
1042 {
1043         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1044         const struct nlattr *encap;
1045         struct nlattr *newmask = NULL;
1046         u64 key_attrs = 0;
1047         u64 mask_attrs = 0;
1048         bool encap_valid = false;
1049         int err;
1050
1051         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1052         if (err)
1053                 return err;
1054
1055         if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1056             (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
1057             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1058                 __be16 tci;
1059
1060                 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
1061                       (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
1062                         OVS_NLERR(log, "Invalid Vlan frame.");
1063                         return -EINVAL;
1064                 }
1065
1066                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1067                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1068                 encap = a[OVS_KEY_ATTR_ENCAP];
1069                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1070                 encap_valid = true;
1071
1072                 if (tci & htons(VLAN_TAG_PRESENT)) {
1073                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1074                         if (err)
1075                                 return err;
1076                 } else if (!tci) {
1077                         /* Corner case for truncated 802.1Q header. */
1078                         if (nla_len(encap)) {
1079                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1080                                 return -EINVAL;
1081                         }
1082                 } else {
1083                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1084                         return  -EINVAL;
1085                 }
1086         }
1087
1088         err = ovs_key_from_nlattrs(match, key_attrs, a, false, log);
1089         if (err)
1090                 return err;
1091
1092         if (match->mask) {
1093                 if (!nla_mask) {
1094                         /* Create an exact match mask. We need to set to 0xff
1095                          * all the 'match->mask' fields that have been touched
1096                          * in 'match->key'. We cannot simply memset
1097                          * 'match->mask', because padding bytes and fields not
1098                          * specified in 'match->key' should be left to 0.
1099                          * Instead, we use a stream of netlink attributes,
1100                          * copied from 'key' and set to 0xff.
1101                          * ovs_key_from_nlattrs() will take care of filling
1102                          * 'match->mask' appropriately.
1103                          */
1104                         newmask = kmemdup(nla_key,
1105                                           nla_total_size(nla_len(nla_key)),
1106                                           GFP_KERNEL);
1107                         if (!newmask)
1108                                 return -ENOMEM;
1109
1110                         mask_set_nlattr(newmask, 0xff);
1111
1112                         /* The userspace does not send tunnel attributes that
1113                          * are 0, but we should not wildcard them nonetheless.
1114                          */
1115                         if (match->key->tun_key.ipv4_dst)
1116                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1117                                                          0xff, true);
1118
1119                         nla_mask = newmask;
1120                 }
1121
1122                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1123                 if (err)
1124                         goto free_newmask;
1125
1126                 /* Always match on tci. */
1127                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1128
1129                 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1130                         __be16 eth_type = 0;
1131                         __be16 tci = 0;
1132
1133                         if (!encap_valid) {
1134                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1135                                 err = -EINVAL;
1136                                 goto free_newmask;
1137                         }
1138
1139                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1140                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1141                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1142
1143                         if (eth_type == htons(0xffff)) {
1144                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1145                                 encap = a[OVS_KEY_ATTR_ENCAP];
1146                                 err = parse_flow_mask_nlattrs(encap, a,
1147                                                               &mask_attrs, log);
1148                                 if (err)
1149                                         goto free_newmask;
1150                         } else {
1151                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1152                                           ntohs(eth_type));
1153                                 err = -EINVAL;
1154                                 goto free_newmask;
1155                         }
1156
1157                         if (a[OVS_KEY_ATTR_VLAN])
1158                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1159
1160                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1161                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1162                                           ntohs(tci));
1163                                 err = -EINVAL;
1164                                 goto free_newmask;
1165                         }
1166                 }
1167
1168                 err = ovs_key_from_nlattrs(match, mask_attrs, a, true, log);
1169                 if (err)
1170                         goto free_newmask;
1171         }
1172
1173         if (!match_validate(match, key_attrs, mask_attrs, log))
1174                 err = -EINVAL;
1175
1176 free_newmask:
1177         kfree(newmask);
1178         return err;
1179 }
1180
1181 /**
1182  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1183  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1184  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1185  * sequence.
1186  * @log: Boolean to allow kernel error logging.  Normally true, but when
1187  * probing for feature compatibility this should be passed in as false to
1188  * suppress unnecessary error logging.
1189  *
1190  * This parses a series of Netlink attributes that form a flow key, which must
1191  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1192  * get the metadata, that is, the parts of the flow key that cannot be
1193  * extracted from the packet itself.
1194  */
1195
1196 int ovs_nla_get_flow_metadata(const struct nlattr *attr,
1197                               struct sw_flow_key *key,
1198                               bool log)
1199 {
1200         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1201         struct sw_flow_match match;
1202         u64 attrs = 0;
1203         int err;
1204
1205         err = parse_flow_nlattrs(attr, a, &attrs, log);
1206         if (err)
1207                 return -EINVAL;
1208
1209         memset(&match, 0, sizeof(match));
1210         match.key = key;
1211
1212         memset(key, 0, OVS_SW_FLOW_KEY_METADATA_SIZE);
1213         key->phy.in_port = DP_MAX_PORTS;
1214
1215         return metadata_from_nlattrs(&match, &attrs, a, false, log);
1216 }
1217
1218 int ovs_nla_put_flow(const struct sw_flow_key *swkey,
1219                      const struct sw_flow_key *output, struct sk_buff *skb)
1220 {
1221         struct ovs_key_ethernet *eth_key;
1222         struct nlattr *nla, *encap;
1223         bool is_mask = (swkey != output);
1224
1225         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1226                 goto nla_put_failure;
1227
1228         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1229                 goto nla_put_failure;
1230
1231         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1232                 goto nla_put_failure;
1233
1234         if ((swkey->tun_key.ipv4_dst || is_mask)) {
1235                 const void *opts = NULL;
1236
1237                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1238                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1239
1240                 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1241                                        swkey->tun_opts_len))
1242                         goto nla_put_failure;
1243         }
1244
1245         if (swkey->phy.in_port == DP_MAX_PORTS) {
1246                 if (is_mask && (output->phy.in_port == 0xffff))
1247                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1248                                 goto nla_put_failure;
1249         } else {
1250                 u16 upper_u16;
1251                 upper_u16 = !is_mask ? 0 : 0xffff;
1252
1253                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1254                                 (upper_u16 << 16) | output->phy.in_port))
1255                         goto nla_put_failure;
1256         }
1257
1258         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1259                 goto nla_put_failure;
1260
1261         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1262         if (!nla)
1263                 goto nla_put_failure;
1264
1265         eth_key = nla_data(nla);
1266         ether_addr_copy(eth_key->eth_src, output->eth.src);
1267         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1268
1269         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1270                 __be16 eth_type;
1271                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1272                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1273                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1274                         goto nla_put_failure;
1275                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1276                 if (!swkey->eth.tci)
1277                         goto unencap;
1278         } else
1279                 encap = NULL;
1280
1281         if (swkey->eth.type == htons(ETH_P_802_2)) {
1282                 /*
1283                  * Ethertype 802.2 is represented in the netlink with omitted
1284                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1285                  * 0xffff in the mask attribute.  Ethertype can also
1286                  * be wildcarded.
1287                  */
1288                 if (is_mask && output->eth.type)
1289                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1290                                                 output->eth.type))
1291                                 goto nla_put_failure;
1292                 goto unencap;
1293         }
1294
1295         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1296                 goto nla_put_failure;
1297
1298         if (swkey->eth.type == htons(ETH_P_IP)) {
1299                 struct ovs_key_ipv4 *ipv4_key;
1300
1301                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1302                 if (!nla)
1303                         goto nla_put_failure;
1304                 ipv4_key = nla_data(nla);
1305                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1306                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1307                 ipv4_key->ipv4_proto = output->ip.proto;
1308                 ipv4_key->ipv4_tos = output->ip.tos;
1309                 ipv4_key->ipv4_ttl = output->ip.ttl;
1310                 ipv4_key->ipv4_frag = output->ip.frag;
1311         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1312                 struct ovs_key_ipv6 *ipv6_key;
1313
1314                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1315                 if (!nla)
1316                         goto nla_put_failure;
1317                 ipv6_key = nla_data(nla);
1318                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1319                                 sizeof(ipv6_key->ipv6_src));
1320                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1321                                 sizeof(ipv6_key->ipv6_dst));
1322                 ipv6_key->ipv6_label = output->ipv6.label;
1323                 ipv6_key->ipv6_proto = output->ip.proto;
1324                 ipv6_key->ipv6_tclass = output->ip.tos;
1325                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1326                 ipv6_key->ipv6_frag = output->ip.frag;
1327         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1328                    swkey->eth.type == htons(ETH_P_RARP)) {
1329                 struct ovs_key_arp *arp_key;
1330
1331                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1332                 if (!nla)
1333                         goto nla_put_failure;
1334                 arp_key = nla_data(nla);
1335                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1336                 arp_key->arp_sip = output->ipv4.addr.src;
1337                 arp_key->arp_tip = output->ipv4.addr.dst;
1338                 arp_key->arp_op = htons(output->ip.proto);
1339                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1340                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1341         } else if (eth_p_mpls(swkey->eth.type)) {
1342                 struct ovs_key_mpls *mpls_key;
1343
1344                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1345                 if (!nla)
1346                         goto nla_put_failure;
1347                 mpls_key = nla_data(nla);
1348                 mpls_key->mpls_lse = output->mpls.top_lse;
1349         }
1350
1351         if ((swkey->eth.type == htons(ETH_P_IP) ||
1352              swkey->eth.type == htons(ETH_P_IPV6)) &&
1353              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1354
1355                 if (swkey->ip.proto == IPPROTO_TCP) {
1356                         struct ovs_key_tcp *tcp_key;
1357
1358                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1359                         if (!nla)
1360                                 goto nla_put_failure;
1361                         tcp_key = nla_data(nla);
1362                         tcp_key->tcp_src = output->tp.src;
1363                         tcp_key->tcp_dst = output->tp.dst;
1364                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1365                                          output->tp.flags))
1366                                 goto nla_put_failure;
1367                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1368                         struct ovs_key_udp *udp_key;
1369
1370                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1371                         if (!nla)
1372                                 goto nla_put_failure;
1373                         udp_key = nla_data(nla);
1374                         udp_key->udp_src = output->tp.src;
1375                         udp_key->udp_dst = output->tp.dst;
1376                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1377                         struct ovs_key_sctp *sctp_key;
1378
1379                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1380                         if (!nla)
1381                                 goto nla_put_failure;
1382                         sctp_key = nla_data(nla);
1383                         sctp_key->sctp_src = output->tp.src;
1384                         sctp_key->sctp_dst = output->tp.dst;
1385                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1386                            swkey->ip.proto == IPPROTO_ICMP) {
1387                         struct ovs_key_icmp *icmp_key;
1388
1389                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1390                         if (!nla)
1391                                 goto nla_put_failure;
1392                         icmp_key = nla_data(nla);
1393                         icmp_key->icmp_type = ntohs(output->tp.src);
1394                         icmp_key->icmp_code = ntohs(output->tp.dst);
1395                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1396                            swkey->ip.proto == IPPROTO_ICMPV6) {
1397                         struct ovs_key_icmpv6 *icmpv6_key;
1398
1399                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1400                                                 sizeof(*icmpv6_key));
1401                         if (!nla)
1402                                 goto nla_put_failure;
1403                         icmpv6_key = nla_data(nla);
1404                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1405                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1406
1407                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1408                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1409                                 struct ovs_key_nd *nd_key;
1410
1411                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1412                                 if (!nla)
1413                                         goto nla_put_failure;
1414                                 nd_key = nla_data(nla);
1415                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1416                                                         sizeof(nd_key->nd_target));
1417                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1418                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1419                         }
1420                 }
1421         }
1422
1423 unencap:
1424         if (encap)
1425                 nla_nest_end(skb, encap);
1426
1427         return 0;
1428
1429 nla_put_failure:
1430         return -EMSGSIZE;
1431 }
1432
1433 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1434
1435 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1436 {
1437         struct sw_flow_actions *sfa;
1438
1439         if (size > MAX_ACTIONS_BUFSIZE) {
1440                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1441                 return ERR_PTR(-EINVAL);
1442         }
1443
1444         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1445         if (!sfa)
1446                 return ERR_PTR(-ENOMEM);
1447
1448         sfa->actions_len = 0;
1449         return sfa;
1450 }
1451
1452 /* RCU callback used by ovs_nla_free_flow_actions. */
1453 static void rcu_free_acts_callback(struct rcu_head *rcu)
1454 {
1455         struct sw_flow_actions *sf_acts = container_of(rcu,
1456                         struct sw_flow_actions, rcu);
1457         kfree(sf_acts);
1458 }
1459
1460 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1461  * The caller must hold rcu_read_lock for this to be sensible.
1462  */
1463 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1464 {
1465         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
1466 }
1467
1468 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1469                                        int attr_len, bool log)
1470 {
1471
1472         struct sw_flow_actions *acts;
1473         int new_acts_size;
1474         int req_size = NLA_ALIGN(attr_len);
1475         int next_offset = offsetof(struct sw_flow_actions, actions) +
1476                                         (*sfa)->actions_len;
1477
1478         if (req_size <= (ksize(*sfa) - next_offset))
1479                 goto out;
1480
1481         new_acts_size = ksize(*sfa) * 2;
1482
1483         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1484                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1485                         return ERR_PTR(-EMSGSIZE);
1486                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1487         }
1488
1489         acts = nla_alloc_flow_actions(new_acts_size, log);
1490         if (IS_ERR(acts))
1491                 return (void *)acts;
1492
1493         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1494         acts->actions_len = (*sfa)->actions_len;
1495         kfree(*sfa);
1496         *sfa = acts;
1497
1498 out:
1499         (*sfa)->actions_len += req_size;
1500         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1501 }
1502
1503 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1504                                    int attrtype, void *data, int len, bool log)
1505 {
1506         struct nlattr *a;
1507
1508         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1509         if (IS_ERR(a))
1510                 return a;
1511
1512         a->nla_type = attrtype;
1513         a->nla_len = nla_attr_size(len);
1514
1515         if (data)
1516                 memcpy(nla_data(a), data, len);
1517         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1518
1519         return a;
1520 }
1521
1522 static int add_action(struct sw_flow_actions **sfa, int attrtype,
1523                       void *data, int len, bool log)
1524 {
1525         struct nlattr *a;
1526
1527         a = __add_action(sfa, attrtype, data, len, log);
1528         if (IS_ERR(a))
1529                 return PTR_ERR(a);
1530
1531         return 0;
1532 }
1533
1534 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1535                                           int attrtype, bool log)
1536 {
1537         int used = (*sfa)->actions_len;
1538         int err;
1539
1540         err = add_action(sfa, attrtype, NULL, 0, log);
1541         if (err)
1542                 return err;
1543
1544         return used;
1545 }
1546
1547 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1548                                          int st_offset)
1549 {
1550         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1551                                                                st_offset);
1552
1553         a->nla_len = sfa->actions_len - st_offset;
1554 }
1555
1556 static int __ovs_nla_copy_actions(const struct nlattr *attr,
1557                                   const struct sw_flow_key *key,
1558                                   int depth, struct sw_flow_actions **sfa,
1559                                   __be16 eth_type, __be16 vlan_tci, bool log);
1560
1561 static int validate_and_copy_sample(const struct nlattr *attr,
1562                                     const struct sw_flow_key *key, int depth,
1563                                     struct sw_flow_actions **sfa,
1564                                     __be16 eth_type, __be16 vlan_tci, bool log)
1565 {
1566         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1567         const struct nlattr *probability, *actions;
1568         const struct nlattr *a;
1569         int rem, start, err, st_acts;
1570
1571         memset(attrs, 0, sizeof(attrs));
1572         nla_for_each_nested(a, attr, rem) {
1573                 int type = nla_type(a);
1574                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1575                         return -EINVAL;
1576                 attrs[type] = a;
1577         }
1578         if (rem)
1579                 return -EINVAL;
1580
1581         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1582         if (!probability || nla_len(probability) != sizeof(u32))
1583                 return -EINVAL;
1584
1585         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1586         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1587                 return -EINVAL;
1588
1589         /* validation done, copy sample action. */
1590         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1591         if (start < 0)
1592                 return start;
1593         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1594                          nla_data(probability), sizeof(u32), log);
1595         if (err)
1596                 return err;
1597         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1598         if (st_acts < 0)
1599                 return st_acts;
1600
1601         err = __ovs_nla_copy_actions(actions, key, depth + 1, sfa,
1602                                      eth_type, vlan_tci, log);
1603         if (err)
1604                 return err;
1605
1606         add_nested_action_end(*sfa, st_acts);
1607         add_nested_action_end(*sfa, start);
1608
1609         return 0;
1610 }
1611
1612 static int validate_tp_port(const struct sw_flow_key *flow_key,
1613                             __be16 eth_type)
1614 {
1615         if ((eth_type == htons(ETH_P_IP) || eth_type == htons(ETH_P_IPV6)) &&
1616             (flow_key->tp.src || flow_key->tp.dst))
1617                 return 0;
1618
1619         return -EINVAL;
1620 }
1621
1622 void ovs_match_init(struct sw_flow_match *match,
1623                     struct sw_flow_key *key,
1624                     struct sw_flow_mask *mask)
1625 {
1626         memset(match, 0, sizeof(*match));
1627         match->key = key;
1628         match->mask = mask;
1629
1630         memset(key, 0, sizeof(*key));
1631
1632         if (mask) {
1633                 memset(&mask->key, 0, sizeof(mask->key));
1634                 mask->range.start = mask->range.end = 0;
1635         }
1636 }
1637
1638 static int validate_geneve_opts(struct sw_flow_key *key)
1639 {
1640         struct geneve_opt *option;
1641         int opts_len = key->tun_opts_len;
1642         bool crit_opt = false;
1643
1644         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1645         while (opts_len > 0) {
1646                 int len;
1647
1648                 if (opts_len < sizeof(*option))
1649                         return -EINVAL;
1650
1651                 len = sizeof(*option) + option->length * 4;
1652                 if (len > opts_len)
1653                         return -EINVAL;
1654
1655                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1656
1657                 option = (struct geneve_opt *)((u8 *)option + len);
1658                 opts_len -= len;
1659         };
1660
1661         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1662
1663         return 0;
1664 }
1665
1666 static int validate_and_copy_set_tun(const struct nlattr *attr,
1667                                      struct sw_flow_actions **sfa, bool log)
1668 {
1669         struct sw_flow_match match;
1670         struct sw_flow_key key;
1671         struct ovs_tunnel_info *tun_info;
1672         struct nlattr *a;
1673         int err, start, opts_type;
1674
1675         ovs_match_init(&match, &key, NULL);
1676         opts_type = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1677         if (opts_type < 0)
1678                 return opts_type;
1679
1680         if (key.tun_opts_len) {
1681                 switch (opts_type) {
1682                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1683                         err = validate_geneve_opts(&key);
1684                         if (err < 0)
1685                                 return err;
1686                         break;
1687                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1688                         break;
1689                 }
1690         };
1691
1692         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1693         if (start < 0)
1694                 return start;
1695
1696         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1697                          sizeof(*tun_info) + key.tun_opts_len, log);
1698         if (IS_ERR(a))
1699                 return PTR_ERR(a);
1700
1701         tun_info = nla_data(a);
1702         tun_info->tunnel = key.tun_key;
1703         tun_info->options_len = key.tun_opts_len;
1704
1705         if (tun_info->options_len) {
1706                 /* We need to store the options in the action itself since
1707                  * everything else will go away after flow setup. We can append
1708                  * it to tun_info and then point there.
1709                  */
1710                 memcpy((tun_info + 1),
1711                        TUN_METADATA_OPTS(&key, key.tun_opts_len), key.tun_opts_len);
1712                 tun_info->options = (tun_info + 1);
1713
1714         } else {
1715                 tun_info->options = NULL;
1716         }
1717
1718         add_nested_action_end(*sfa, start);
1719
1720         return err;
1721 }
1722
1723 static int validate_set(const struct nlattr *a,
1724                         const struct sw_flow_key *flow_key,
1725                         struct sw_flow_actions **sfa,
1726                         bool *set_tun, __be16 eth_type, bool log)
1727 {
1728         const struct nlattr *ovs_key = nla_data(a);
1729         int key_type = nla_type(ovs_key);
1730
1731         /* There can be only one key in a action */
1732         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1733                 return -EINVAL;
1734
1735         if (key_type > OVS_KEY_ATTR_MAX ||
1736             (ovs_key_lens[key_type].len != nla_len(ovs_key) &&
1737              ovs_key_lens[key_type].len != OVS_ATTR_NESTED))
1738                 return -EINVAL;
1739
1740         switch (key_type) {
1741         const struct ovs_key_ipv4 *ipv4_key;
1742         const struct ovs_key_ipv6 *ipv6_key;
1743         int err;
1744
1745         case OVS_KEY_ATTR_PRIORITY:
1746         case OVS_KEY_ATTR_SKB_MARK:
1747         case OVS_KEY_ATTR_ETHERNET:
1748                 break;
1749
1750         case OVS_KEY_ATTR_TUNNEL:
1751                 if (eth_p_mpls(eth_type))
1752                         return -EINVAL;
1753
1754                 *set_tun = true;
1755                 err = validate_and_copy_set_tun(a, sfa, log);
1756                 if (err)
1757                         return err;
1758                 break;
1759
1760         case OVS_KEY_ATTR_IPV4:
1761                 if (eth_type != htons(ETH_P_IP))
1762                         return -EINVAL;
1763
1764                 if (!flow_key->ip.proto)
1765                         return -EINVAL;
1766
1767                 ipv4_key = nla_data(ovs_key);
1768                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1769                         return -EINVAL;
1770
1771                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1772                         return -EINVAL;
1773
1774                 break;
1775
1776         case OVS_KEY_ATTR_IPV6:
1777                 if (eth_type != htons(ETH_P_IPV6))
1778                         return -EINVAL;
1779
1780                 if (!flow_key->ip.proto)
1781                         return -EINVAL;
1782
1783                 ipv6_key = nla_data(ovs_key);
1784                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1785                         return -EINVAL;
1786
1787                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1788                         return -EINVAL;
1789
1790                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
1791                         return -EINVAL;
1792
1793                 break;
1794
1795         case OVS_KEY_ATTR_TCP:
1796                 if (flow_key->ip.proto != IPPROTO_TCP)
1797                         return -EINVAL;
1798
1799                 return validate_tp_port(flow_key, eth_type);
1800
1801         case OVS_KEY_ATTR_UDP:
1802                 if (flow_key->ip.proto != IPPROTO_UDP)
1803                         return -EINVAL;
1804
1805                 return validate_tp_port(flow_key, eth_type);
1806
1807         case OVS_KEY_ATTR_MPLS:
1808                 if (!eth_p_mpls(eth_type))
1809                         return -EINVAL;
1810                 break;
1811
1812         case OVS_KEY_ATTR_SCTP:
1813                 if (flow_key->ip.proto != IPPROTO_SCTP)
1814                         return -EINVAL;
1815
1816                 return validate_tp_port(flow_key, eth_type);
1817
1818         default:
1819                 return -EINVAL;
1820         }
1821
1822         return 0;
1823 }
1824
1825 static int validate_userspace(const struct nlattr *attr)
1826 {
1827         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
1828                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
1829                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
1830                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
1831         };
1832         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
1833         int error;
1834
1835         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
1836                                  attr, userspace_policy);
1837         if (error)
1838                 return error;
1839
1840         if (!a[OVS_USERSPACE_ATTR_PID] ||
1841             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
1842                 return -EINVAL;
1843
1844         return 0;
1845 }
1846
1847 static int copy_action(const struct nlattr *from,
1848                        struct sw_flow_actions **sfa, bool log)
1849 {
1850         int totlen = NLA_ALIGN(from->nla_len);
1851         struct nlattr *to;
1852
1853         to = reserve_sfa_size(sfa, from->nla_len, log);
1854         if (IS_ERR(to))
1855                 return PTR_ERR(to);
1856
1857         memcpy(to, from, totlen);
1858         return 0;
1859 }
1860
1861 static int __ovs_nla_copy_actions(const struct nlattr *attr,
1862                                   const struct sw_flow_key *key,
1863                                   int depth, struct sw_flow_actions **sfa,
1864                                   __be16 eth_type, __be16 vlan_tci, bool log)
1865 {
1866         const struct nlattr *a;
1867         int rem, err;
1868
1869         if (depth >= SAMPLE_ACTION_DEPTH)
1870                 return -EOVERFLOW;
1871
1872         nla_for_each_nested(a, attr, rem) {
1873                 /* Expected argument lengths, (u32)-1 for variable length. */
1874                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
1875                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
1876                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
1877                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
1878                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
1879                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
1880                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
1881                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
1882                         [OVS_ACTION_ATTR_SET] = (u32)-1,
1883                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
1884                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash)
1885                 };
1886                 const struct ovs_action_push_vlan *vlan;
1887                 int type = nla_type(a);
1888                 bool skip_copy;
1889
1890                 if (type > OVS_ACTION_ATTR_MAX ||
1891                     (action_lens[type] != nla_len(a) &&
1892                      action_lens[type] != (u32)-1))
1893                         return -EINVAL;
1894
1895                 skip_copy = false;
1896                 switch (type) {
1897                 case OVS_ACTION_ATTR_UNSPEC:
1898                         return -EINVAL;
1899
1900                 case OVS_ACTION_ATTR_USERSPACE:
1901                         err = validate_userspace(a);
1902                         if (err)
1903                                 return err;
1904                         break;
1905
1906                 case OVS_ACTION_ATTR_OUTPUT:
1907                         if (nla_get_u32(a) >= DP_MAX_PORTS)
1908                                 return -EINVAL;
1909
1910                         break;
1911
1912                 case OVS_ACTION_ATTR_HASH: {
1913                         const struct ovs_action_hash *act_hash = nla_data(a);
1914
1915                         switch (act_hash->hash_alg) {
1916                         case OVS_HASH_ALG_L4:
1917                                 break;
1918                         default:
1919                                 return  -EINVAL;
1920                         }
1921
1922                         break;
1923                 }
1924
1925                 case OVS_ACTION_ATTR_POP_VLAN:
1926                         vlan_tci = htons(0);
1927                         break;
1928
1929                 case OVS_ACTION_ATTR_PUSH_VLAN:
1930                         vlan = nla_data(a);
1931                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
1932                                 return -EINVAL;
1933                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
1934                                 return -EINVAL;
1935                         vlan_tci = vlan->vlan_tci;
1936                         break;
1937
1938                 case OVS_ACTION_ATTR_RECIRC:
1939                         break;
1940
1941                 case OVS_ACTION_ATTR_PUSH_MPLS: {
1942                         const struct ovs_action_push_mpls *mpls = nla_data(a);
1943
1944                         if (!eth_p_mpls(mpls->mpls_ethertype))
1945                                 return -EINVAL;
1946
1947                         /* Prohibit push MPLS other than to a white list
1948                          * for packets that have a known tag order.
1949                          */
1950                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
1951                             (eth_type != htons(ETH_P_IP) &&
1952                              eth_type != htons(ETH_P_IPV6) &&
1953                              eth_type != htons(ETH_P_ARP) &&
1954                              eth_type != htons(ETH_P_RARP) &&
1955                              !eth_p_mpls(eth_type)))
1956                                 return -EINVAL;
1957                         eth_type = mpls->mpls_ethertype;
1958                         break;
1959                 }
1960
1961                 case OVS_ACTION_ATTR_POP_MPLS:
1962                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
1963                             !eth_p_mpls(eth_type))
1964                                 return -EINVAL;
1965
1966                         /* Disallow subsequent L2.5+ set and mpls_pop actions
1967                          * as there is no check here to ensure that the new
1968                          * eth_type is valid and thus set actions could
1969                          * write off the end of the packet or otherwise
1970                          * corrupt it.
1971                          *
1972                          * Support for these actions is planned using packet
1973                          * recirculation.
1974                          */
1975                         eth_type = htons(0);
1976                         break;
1977
1978                 case OVS_ACTION_ATTR_SET:
1979                         err = validate_set(a, key, sfa,
1980                                            &skip_copy, eth_type, log);
1981                         if (err)
1982                                 return err;
1983                         break;
1984
1985                 case OVS_ACTION_ATTR_SAMPLE:
1986                         err = validate_and_copy_sample(a, key, depth, sfa,
1987                                                        eth_type, vlan_tci, log);
1988                         if (err)
1989                                 return err;
1990                         skip_copy = true;
1991                         break;
1992
1993                 default:
1994                         OVS_NLERR(log, "Unknown Action type %d", type);
1995                         return -EINVAL;
1996                 }
1997                 if (!skip_copy) {
1998                         err = copy_action(a, sfa, log);
1999                         if (err)
2000                                 return err;
2001                 }
2002         }
2003
2004         if (rem > 0)
2005                 return -EINVAL;
2006
2007         return 0;
2008 }
2009
2010 int ovs_nla_copy_actions(const struct nlattr *attr,
2011                          const struct sw_flow_key *key,
2012                          struct sw_flow_actions **sfa, bool log)
2013 {
2014         int err;
2015
2016         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2017         if (IS_ERR(*sfa))
2018                 return PTR_ERR(*sfa);
2019
2020         err = __ovs_nla_copy_actions(attr, key, 0, sfa, key->eth.type,
2021                                      key->eth.tci, log);
2022         if (err)
2023                 kfree(*sfa);
2024
2025         return err;
2026 }
2027
2028 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2029 {
2030         const struct nlattr *a;
2031         struct nlattr *start;
2032         int err = 0, rem;
2033
2034         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2035         if (!start)
2036                 return -EMSGSIZE;
2037
2038         nla_for_each_nested(a, attr, rem) {
2039                 int type = nla_type(a);
2040                 struct nlattr *st_sample;
2041
2042                 switch (type) {
2043                 case OVS_SAMPLE_ATTR_PROBABILITY:
2044                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2045                                     sizeof(u32), nla_data(a)))
2046                                 return -EMSGSIZE;
2047                         break;
2048                 case OVS_SAMPLE_ATTR_ACTIONS:
2049                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2050                         if (!st_sample)
2051                                 return -EMSGSIZE;
2052                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2053                         if (err)
2054                                 return err;
2055                         nla_nest_end(skb, st_sample);
2056                         break;
2057                 }
2058         }
2059
2060         nla_nest_end(skb, start);
2061         return err;
2062 }
2063
2064 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2065 {
2066         const struct nlattr *ovs_key = nla_data(a);
2067         int key_type = nla_type(ovs_key);
2068         struct nlattr *start;
2069         int err;
2070
2071         switch (key_type) {
2072         case OVS_KEY_ATTR_TUNNEL_INFO: {
2073                 struct ovs_tunnel_info *tun_info = nla_data(ovs_key);
2074
2075                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2076                 if (!start)
2077                         return -EMSGSIZE;
2078
2079                 err = ipv4_tun_to_nlattr(skb, &tun_info->tunnel,
2080                                          tun_info->options_len ?
2081                                                 tun_info->options : NULL,
2082                                          tun_info->options_len);
2083                 if (err)
2084                         return err;
2085                 nla_nest_end(skb, start);
2086                 break;
2087         }
2088         default:
2089                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2090                         return -EMSGSIZE;
2091                 break;
2092         }
2093
2094         return 0;
2095 }
2096
2097 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2098 {
2099         const struct nlattr *a;
2100         int rem, err;
2101
2102         nla_for_each_attr(a, attr, len, rem) {
2103                 int type = nla_type(a);
2104
2105                 switch (type) {
2106                 case OVS_ACTION_ATTR_SET:
2107                         err = set_action_to_attr(a, skb);
2108                         if (err)
2109                                 return err;
2110                         break;
2111
2112                 case OVS_ACTION_ATTR_SAMPLE:
2113                         err = sample_action_to_attr(a, skb);
2114                         if (err)
2115                                 return err;
2116                         break;
2117                 default:
2118                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2119                                 return -EMSGSIZE;
2120                         break;
2121                 }
2122         }
2123
2124         return 0;
2125 }