net/openvswitch/flow_netlink.c

   1 /*
   2  * Copyright (c) 2007-2017 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include "flow.h"
  22 #include "datapath.h"
  23 #include <linux/uaccess.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <net/llc_pdu.h>
  29 #include <linux/kernel.h>
  30 #include <linux/jhash.h>
  31 #include <linux/jiffies.h>
  32 #include <linux/llc.h>
  33 #include <linux/module.h>
  34 #include <linux/in.h>
  35 #include <linux/rcupdate.h>
  36 #include <linux/if_arp.h>
  37 #include <linux/ip.h>
  38 #include <linux/ipv6.h>
  39 #include <linux/sctp.h>
  40 #include <linux/tcp.h>
  41 #include <linux/udp.h>
  42 #include <linux/icmp.h>
  43 #include <linux/icmpv6.h>
  44 #include <linux/rculist.h>
  45 #include <net/geneve.h>
  46 #include <net/ip.h>
  47 #include <net/ipv6.h>
  48 #include <net/ndisc.h>
  49 #include <net/mpls.h>
  50 #include <net/vxlan.h>
  51 #include <net/erspan.h>
  52
  53 #include "flow_netlink.h"
  54
  55 struct ovs_len_tbl {
  56         int len;
  57         const struct ovs_len_tbl *next;
  58 };
  59
  60 #define OVS_ATTR_NESTED -1
  61 #define OVS_ATTR_VARIABLE -2
  62
  63 static bool actions_may_change_flow(const struct nlattr *actions)
  64 {
  65         struct nlattr *nla;
  66         int rem;
  67
  68         nla_for_each_nested(nla, actions, rem) {
  69                 u16 action = nla_type(nla);
  70
  71                 switch (action) {
  72                 case OVS_ACTION_ATTR_OUTPUT:
  73                 case OVS_ACTION_ATTR_RECIRC:
  74                 case OVS_ACTION_ATTR_TRUNC:
  75                 case OVS_ACTION_ATTR_USERSPACE:
  76                         break;
  77
  78                 case OVS_ACTION_ATTR_CT:
  79                 case OVS_ACTION_ATTR_CT_CLEAR:
  80                 case OVS_ACTION_ATTR_HASH:
  81                 case OVS_ACTION_ATTR_POP_ETH:
  82                 case OVS_ACTION_ATTR_POP_MPLS:
  83                 case OVS_ACTION_ATTR_POP_VLAN:
  84                 case OVS_ACTION_ATTR_PUSH_ETH:
  85                 case OVS_ACTION_ATTR_PUSH_MPLS:
  86                 case OVS_ACTION_ATTR_PUSH_VLAN:
  87                 case OVS_ACTION_ATTR_SAMPLE:
  88                 case OVS_ACTION_ATTR_SET:
  89                 case OVS_ACTION_ATTR_SET_MASKED:
  90                 default:
  91                         return true;
  92                 }
  93         }
  94         return false;
  95 }
  96
  97 static void update_range(struct sw_flow_match *match,
  98                          size_t offset, size_t size, bool is_mask)
  99 {
 100         struct sw_flow_key_range *range;
 101         size_t start = rounddown(offset, sizeof(long));
 102         size_t end = roundup(offset + size, sizeof(long));
 103
 104         if (!is_mask)
 105                 range = &match->range;
 106         else
 107                 range = &match->mask->range;
 108
 109         if (range->start == range->end) {
 110                 range->start = start;
 111                 range->end = end;
 112                 return;
 113         }
 114
 115         if (range->start > start)
 116                 range->start = start;
 117
 118         if (range->end < end)
 119                 range->end = end;
 120 }
 121
 122 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
 123         do { \
 124                 update_range(match, offsetof(struct sw_flow_key, field),    \
 125                              sizeof((match)->key->field), is_mask);         \
 126                 if (is_mask)                                                \
 127                         (match)->mask->key.field = value;                   \
 128                 else                                                        \
 129                         (match)->key->field = value;                        \
 130         } while (0)
 131
 132 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
 133         do {                                                                \
 134                 update_range(match, offset, len, is_mask);                  \
 135                 if (is_mask)                                                \
 136                         memcpy((u8 *)&(match)->mask->key + offset, value_p, \
 137                                len);                                       \
 138                 else                                                        \
 139                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
 140         } while (0)
 141
 142 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
 143         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 144                                   value_p, len, is_mask)
 145
 146 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 147         do {                                                                \
 148                 update_range(match, offsetof(struct sw_flow_key, field),    \
 149                              sizeof((match)->key->field), is_mask);         \
 150                 if (is_mask)                                                \
 151                         memset((u8 *)&(match)->mask->key.field, value,      \
 152                                sizeof((match)->mask->key.field));           \
 153                 else                                                        \
 154                         memset((u8 *)&(match)->key->field, value,           \
 155                                sizeof((match)->key->field));                \
 156         } while (0)
 157
 158 static bool match_validate(const struct sw_flow_match *match,
 159                            u64 key_attrs, u64 mask_attrs, bool log)
 160 {
 161         u64 key_expected = 0;
 162         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 163
 164         /* The following mask attributes allowed only if they
 165          * pass the validation tests. */
 166         mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
 167                         | (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)
 168                         | (1 << OVS_KEY_ATTR_IPV6)
 169                         | (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)
 170                         | (1 << OVS_KEY_ATTR_TCP)
 171                         | (1 << OVS_KEY_ATTR_TCP_FLAGS)
 172                         | (1 << OVS_KEY_ATTR_UDP)
 173                         | (1 << OVS_KEY_ATTR_SCTP)
 174                         | (1 << OVS_KEY_ATTR_ICMP)
 175                         | (1 << OVS_KEY_ATTR_ICMPV6)
 176                         | (1 << OVS_KEY_ATTR_ARP)
 177                         | (1 << OVS_KEY_ATTR_ND)
 178                         | (1 << OVS_KEY_ATTR_MPLS));
 179
 180         /* Always allowed mask fields. */
 181         mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
 182                        | (1 << OVS_KEY_ATTR_IN_PORT)
 183                        | (1 << OVS_KEY_ATTR_ETHERTYPE));
 184
 185         /* Check key attributes. */
 186         if (match->key->eth.type == htons(ETH_P_ARP)
 187                         || match->key->eth.type == htons(ETH_P_RARP)) {
 188                 key_expected |= 1 << OVS_KEY_ATTR_ARP;
 189                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 190                         mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
 191         }
 192
 193         if (eth_p_mpls(match->key->eth.type)) {
 194                 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
 195                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 196                         mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
 197         }
 198
 199         if (match->key->eth.type == htons(ETH_P_IP)) {
 200                 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
 201                 if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
 202                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
 203                         mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4;
 204                 }
 205
 206                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 207                         if (match->key->ip.proto == IPPROTO_UDP) {
 208                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 209                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 210                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 211                         }
 212
 213                         if (match->key->ip.proto == IPPROTO_SCTP) {
 214                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 215                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 216                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 217                         }
 218
 219                         if (match->key->ip.proto == IPPROTO_TCP) {
 220                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 221                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 222                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 223                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 224                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 225                                 }
 226                         }
 227
 228                         if (match->key->ip.proto == IPPROTO_ICMP) {
 229                                 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
 230                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 231                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
 232                         }
 233                 }
 234         }
 235
 236         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 237                 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
 238                 if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
 239                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
 240                         mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6;
 241                 }
 242
 243                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 244                         if (match->key->ip.proto == IPPROTO_UDP) {
 245                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 246                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 247                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 248                         }
 249
 250                         if (match->key->ip.proto == IPPROTO_SCTP) {
 251                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 252                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 253                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 254                         }
 255
 256                         if (match->key->ip.proto == IPPROTO_TCP) {
 257                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 258                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 259                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 260                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 261                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 262                                 }
 263                         }
 264
 265                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 266                                 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
 267                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 268                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
 269
 270                                 if (match->key->tp.src ==
 271                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 272                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 273                                         key_expected |= 1 << OVS_KEY_ATTR_ND;
 274                                         /* Original direction conntrack tuple
 275                                          * uses the same space as the ND fields
 276                                          * in the key, so both are not allowed
 277                                          * at the same time.
 278                                          */
 279                                         mask_allowed &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
 280                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 281                                                 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
 282                                 }
 283                         }
 284                 }
 285         }
 286
 287         if ((key_attrs & key_expected) != key_expected) {
 288                 /* Key attributes check failed. */
 289                 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
 290                           (unsigned long long)key_attrs,
 291                           (unsigned long long)key_expected);
 292                 return false;
 293         }
 294
 295         if ((mask_attrs & mask_allowed) != mask_attrs) {
 296                 /* Mask attributes check failed. */
 297                 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
 298                           (unsigned long long)mask_attrs,
 299                           (unsigned long long)mask_allowed);
 300                 return false;
 301         }
 302
 303         return true;
 304 }
 305
 306 size_t ovs_tun_key_attr_size(void)
 307 {
 308         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 309          * updating this function.
 310          */
 311         return    nla_total_size_64bit(8) /* OVS_TUNNEL_KEY_ATTR_ID */
 312                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
 313                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
 314                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
 315                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
 316                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 317                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
 318                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
 319                 + nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 320                 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
 321                  * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
 322                  */
 323                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
 324                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_DST */
 325                 + nla_total_size(4);   /* OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS */
 326 }
 327
 328 size_t ovs_key_attr_size(void)
 329 {
 330         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 331          * updating this function.
 332          */
 333         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 28);
 334
 335         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 336                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 337                   + ovs_tun_key_attr_size()
 338                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 339                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 340                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 341                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 342                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
 343                 + nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
 344                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
 345                 + nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
 346                 + nla_total_size(40)  /* OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6 */
 347                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 348                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 349                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 350                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 351                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 352                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 353                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 354                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 355 }
 356
 357 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
 358         [OVS_VXLAN_EXT_GBP]         = { .len = sizeof(u32) },
 359 };
 360
 361 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 362         [OVS_TUNNEL_KEY_ATTR_ID]            = { .len = sizeof(u64) },
 363         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC]      = { .len = sizeof(u32) },
 364         [OVS_TUNNEL_KEY_ATTR_IPV4_DST]      = { .len = sizeof(u32) },
 365         [OVS_TUNNEL_KEY_ATTR_TOS]           = { .len = 1 },
 366         [OVS_TUNNEL_KEY_ATTR_TTL]           = { .len = 1 },
 367         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
 368         [OVS_TUNNEL_KEY_ATTR_CSUM]          = { .len = 0 },
 369         [OVS_TUNNEL_KEY_ATTR_TP_SRC]        = { .len = sizeof(u16) },
 370         [OVS_TUNNEL_KEY_ATTR_TP_DST]        = { .len = sizeof(u16) },
 371         [OVS_TUNNEL_KEY_ATTR_OAM]           = { .len = 0 },
 372         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
 373         [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
 374                                                 .next = ovs_vxlan_ext_key_lens },
 375         [OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
 376         [OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
 377         [OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS]   = { .len = sizeof(u32) },
 378 };
 379
 380 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 381 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 382         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 383         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 384         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 385         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 386         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 387         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 388         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 389         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 390         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 391         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 392         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 393         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 394         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 395         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 396         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 397         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 398         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 399         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 400         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 401         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 402                                      .next = ovs_tunnel_key_lens, },
 403         [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 404         [OVS_KEY_ATTR_CT_STATE]  = { .len = sizeof(u32) },
 405         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 406         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 407         [OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
 408         [OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4] = {
 409                 .len = sizeof(struct ovs_key_ct_tuple_ipv4) },
 410         [OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6] = {
 411                 .len = sizeof(struct ovs_key_ct_tuple_ipv6) },
 412 };
 413
 414 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
 415 {
 416         return expected_len == attr_len ||
 417                expected_len == OVS_ATTR_NESTED ||
 418                expected_len == OVS_ATTR_VARIABLE;
 419 }
 420
 421 static bool is_all_zero(const u8 *fp, size_t size)
 422 {
 423         int i;
 424
 425         if (!fp)
 426                 return false;
 427
 428         for (i = 0; i < size; i++)
 429                 if (fp[i])
 430                         return false;
 431
 432         return true;
 433 }
 434
 435 static int __parse_flow_nlattrs(const struct nlattr *attr,
 436                                 const struct nlattr *a[],
 437                                 u64 *attrsp, bool log, bool nz)
 438 {
 439         const struct nlattr *nla;
 440         u64 attrs;
 441         int rem;
 442
 443         attrs = *attrsp;
 444         nla_for_each_nested(nla, attr, rem) {
 445                 u16 type = nla_type(nla);
 446                 int expected_len;
 447
 448                 if (type > OVS_KEY_ATTR_MAX) {
 449                         OVS_NLERR(log, "Key type %d is out of range max %d",
 450                                   type, OVS_KEY_ATTR_MAX);
 451                         return -EINVAL;
 452                 }
 453
 454                 if (attrs & (1 << type)) {
 455                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 456                         return -EINVAL;
 457                 }
 458
 459                 expected_len = ovs_key_lens[type].len;
 460                 if (!check_attr_len(nla_len(nla), expected_len)) {
 461                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 462                                   type, nla_len(nla), expected_len);
 463                         return -EINVAL;
 464                 }
 465
 466                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 467                         attrs |= 1 << type;
 468                         a[type] = nla;
 469                 }
 470         }
 471         if (rem) {
 472                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 473                 return -EINVAL;
 474         }
 475
 476         *attrsp = attrs;
 477         return 0;
 478 }
 479
 480 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 481                                    const struct nlattr *a[], u64 *attrsp,
 482                                    bool log)
 483 {
 484         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 485 }
 486
 487 int parse_flow_nlattrs(const struct nlattr *attr, const struct nlattr *a[],
 488                        u64 *attrsp, bool log)
 489 {
 490         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 491 }
 492
 493 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 494                                      struct sw_flow_match *match, bool is_mask,
 495                                      bool log)
 496 {
 497         unsigned long opt_key_offset;
 498
 499         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 500                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 501                           nla_len(a), sizeof(match->key->tun_opts));
 502                 return -EINVAL;
 503         }
 504
 505         if (nla_len(a) % 4 != 0) {
 506                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 507                           nla_len(a));
 508                 return -EINVAL;
 509         }
 510
 511         /* We need to record the length of the options passed
 512          * down, otherwise packets with the same format but
 513          * additional options will be silently matched.
 514          */
 515         if (!is_mask) {
 516                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 517                                 false);
 518         } else {
 519                 /* This is somewhat unusual because it looks at
 520                  * both the key and mask while parsing the
 521                  * attributes (and by extension assumes the key
 522                  * is parsed first). Normally, we would verify
 523                  * that each is the correct length and that the
 524                  * attributes line up in the validate function.
 525                  * However, that is difficult because this is
 526                  * variable length and we won't have the
 527                  * information later.
 528                  */
 529                 if (match->key->tun_opts_len != nla_len(a)) {
 530                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 531                                   match->key->tun_opts_len, nla_len(a));
 532                         return -EINVAL;
 533                 }
 534
 535                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 536         }
 537
 538         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 539         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 540                                   nla_len(a), is_mask);
 541         return 0;
 542 }
 543
 544 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
 545                                      struct sw_flow_match *match, bool is_mask,
 546                                      bool log)
 547 {
 548         struct nlattr *a;
 549         int rem;
 550         unsigned long opt_key_offset;
 551         struct vxlan_metadata opts;
 552
 553         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 554
 555         memset(&opts, 0, sizeof(opts));
 556         nla_for_each_nested(a, attr, rem) {
 557                 int type = nla_type(a);
 558
 559                 if (type > OVS_VXLAN_EXT_MAX) {
 560                         OVS_NLERR(log, "VXLAN extension %d out of range max %d",
 561                                   type, OVS_VXLAN_EXT_MAX);
 562                         return -EINVAL;
 563                 }
 564
 565                 if (!check_attr_len(nla_len(a),
 566                                     ovs_vxlan_ext_key_lens[type].len)) {
 567                         OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
 568                                   type, nla_len(a),
 569                                   ovs_vxlan_ext_key_lens[type].len);
 570                         return -EINVAL;
 571                 }
 572
 573                 switch (type) {
 574                 case OVS_VXLAN_EXT_GBP:
 575                         opts.gbp = nla_get_u32(a);
 576                         break;
 577                 default:
 578                         OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
 579                                   type);
 580                         return -EINVAL;
 581                 }
 582         }
 583         if (rem) {
 584                 OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
 585                           rem);
 586                 return -EINVAL;
 587         }
 588
 589         if (!is_mask)
 590                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 591         else
 592                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 593
 594         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 595         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 596                                   is_mask);
 597         return 0;
 598 }
 599
 600 static int erspan_tun_opt_from_nlattr(const struct nlattr *attr,
 601                                       struct sw_flow_match *match, bool is_mask,
 602                                       bool log)
 603 {
 604         unsigned long opt_key_offset;
 605         struct erspan_metadata opts;
 606
 607         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 608
 609         memset(&opts, 0, sizeof(opts));
 610         opts.index = nla_get_be32(attr);
 611
 612         /* Index has only 20-bit */
 613         if (ntohl(opts.index) & ~INDEX_MASK) {
 614                 OVS_NLERR(log, "ERSPAN index number %x too large.",
 615                           ntohl(opts.index));
 616                 return -EINVAL;
 617         }
 618
 619         SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), is_mask);
 620         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 621         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 622                                   is_mask);
 623
 624         return 0;
 625 }
 626
 627 static int ip_tun_from_nlattr(const struct nlattr *attr,
 628                               struct sw_flow_match *match, bool is_mask,
 629                               bool log)
 630 {
 631         bool ttl = false, ipv4 = false, ipv6 = false;
 632         __be16 tun_flags = 0;
 633         int opts_type = 0;
 634         struct nlattr *a;
 635         int rem;
 636
 637         nla_for_each_nested(a, attr, rem) {
 638                 int type = nla_type(a);
 639                 int err;
 640
 641                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 642                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 643                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 644                         return -EINVAL;
 645                 }
 646
 647                 if (!check_attr_len(nla_len(a),
 648                                     ovs_tunnel_key_lens[type].len)) {
 649                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 650                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 651                         return -EINVAL;
 652                 }
 653
 654                 switch (type) {
 655                 case OVS_TUNNEL_KEY_ATTR_ID:
 656                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 657                                         nla_get_be64(a), is_mask);
 658                         tun_flags |= TUNNEL_KEY;
 659                         break;
 660                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 661                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
 662                                         nla_get_in_addr(a), is_mask);
 663                         ipv4 = true;
 664                         break;
 665                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 666                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
 667                                         nla_get_in_addr(a), is_mask);
 668                         ipv4 = true;
 669                         break;
 670                 case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
 671                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.src,
 672                                         nla_get_in6_addr(a), is_mask);
 673                         ipv6 = true;
 674                         break;
 675                 case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
 676                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
 677                                         nla_get_in6_addr(a), is_mask);
 678                         ipv6 = true;
 679                         break;
 680                 case OVS_TUNNEL_KEY_ATTR_TOS:
 681                         SW_FLOW_KEY_PUT(match, tun_key.tos,
 682                                         nla_get_u8(a), is_mask);
 683                         break;
 684                 case OVS_TUNNEL_KEY_ATTR_TTL:
 685                         SW_FLOW_KEY_PUT(match, tun_key.ttl,
 686                                         nla_get_u8(a), is_mask);
 687                         ttl = true;
 688                         break;
 689                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 690                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 691                         break;
 692                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 693                         tun_flags |= TUNNEL_CSUM;
 694                         break;
 695                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 696                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 697                                         nla_get_be16(a), is_mask);
 698                         break;
 699                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 700                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 701                                         nla_get_be16(a), is_mask);
 702                         break;
 703                 case OVS_TUNNEL_KEY_ATTR_OAM:
 704                         tun_flags |= TUNNEL_OAM;
 705                         break;
 706                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 707                         if (opts_type) {
 708                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 709                                 return -EINVAL;
 710                         }
 711
 712                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 713                         if (err)
 714                                 return err;
 715
 716                         tun_flags |= TUNNEL_GENEVE_OPT;
 717                         opts_type = type;
 718                         break;
 719                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 720                         if (opts_type) {
 721                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 722                                 return -EINVAL;
 723                         }
 724
 725                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 726                         if (err)
 727                                 return err;
 728
 729                         tun_flags |= TUNNEL_VXLAN_OPT;
 730                         opts_type = type;
 731                         break;
 732                 case OVS_TUNNEL_KEY_ATTR_PAD:
 733                         break;
 734                 case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
 735                         if (opts_type) {
 736                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 737                                 return -EINVAL;
 738                         }
 739
 740                         err = erspan_tun_opt_from_nlattr(a, match, is_mask, log);
 741                         if (err)
 742                                 return err;
 743
 744                         tun_flags |= TUNNEL_ERSPAN_OPT;
 745                         opts_type = type;
 746                         break;
 747                 default:
 748                         OVS_NLERR(log, "Unknown IP tunnel attribute %d",
 749                                   type);
 750                         return -EINVAL;
 751                 }
 752         }
 753
 754         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 755         if (is_mask)
 756                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
 757         else
 758                 SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
 759                                 false);
 760
 761         if (rem > 0) {
 762                 OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
 763                           rem);
 764                 return -EINVAL;
 765         }
 766
 767         if (ipv4 && ipv6) {
 768                 OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
 769                 return -EINVAL;
 770         }
 771
 772         if (!is_mask) {
 773                 if (!ipv4 && !ipv6) {
 774                         OVS_NLERR(log, "IP tunnel dst address not specified");
 775                         return -EINVAL;
 776                 }
 777                 if (ipv4 && !match->key->tun_key.u.ipv4.dst) {
 778                         OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 779                         return -EINVAL;
 780                 }
 781                 if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
 782                         OVS_NLERR(log, "IPv6 tunnel dst address is zero");
 783                         return -EINVAL;
 784                 }
 785
 786                 if (!ttl) {
 787                         OVS_NLERR(log, "IP tunnel TTL not specified.");
 788                         return -EINVAL;
 789                 }
 790         }
 791
 792         return opts_type;
 793 }
 794
 795 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 796                                const void *tun_opts, int swkey_tun_opts_len)
 797 {
 798         const struct vxlan_metadata *opts = tun_opts;
 799         struct nlattr *nla;
 800
 801         nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 802         if (!nla)
 803                 return -EMSGSIZE;
 804
 805         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 806                 return -EMSGSIZE;
 807
 808         nla_nest_end(skb, nla);
 809         return 0;
 810 }
 811
 812 static int __ip_tun_to_nlattr(struct sk_buff *skb,
 813                               const struct ip_tunnel_key *output,
 814                               const void *tun_opts, int swkey_tun_opts_len,
 815                               unsigned short tun_proto)
 816 {
 817         if (output->tun_flags & TUNNEL_KEY &&
 818             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id,
 819                          OVS_TUNNEL_KEY_ATTR_PAD))
 820                 return -EMSGSIZE;
 821         switch (tun_proto) {
 822         case AF_INET:
 823                 if (output->u.ipv4.src &&
 824                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 825                                     output->u.ipv4.src))
 826                         return -EMSGSIZE;
 827                 if (output->u.ipv4.dst &&
 828                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 829                                     output->u.ipv4.dst))
 830                         return -EMSGSIZE;
 831                 break;
 832         case AF_INET6:
 833                 if (!ipv6_addr_any(&output->u.ipv6.src) &&
 834                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
 835                                      &output->u.ipv6.src))
 836                         return -EMSGSIZE;
 837                 if (!ipv6_addr_any(&output->u.ipv6.dst) &&
 838                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
 839                                      &output->u.ipv6.dst))
 840                         return -EMSGSIZE;
 841                 break;
 842         }
 843         if (output->tos &&
 844             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 845                 return -EMSGSIZE;
 846         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 847                 return -EMSGSIZE;
 848         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 849             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 850                 return -EMSGSIZE;
 851         if ((output->tun_flags & TUNNEL_CSUM) &&
 852             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 853                 return -EMSGSIZE;
 854         if (output->tp_src &&
 855             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 856                 return -EMSGSIZE;
 857         if (output->tp_dst &&
 858             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 859                 return -EMSGSIZE;
 860         if ((output->tun_flags & TUNNEL_OAM) &&
 861             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 862                 return -EMSGSIZE;
 863         if (swkey_tun_opts_len) {
 864                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 865                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 866                             swkey_tun_opts_len, tun_opts))
 867                         return -EMSGSIZE;
 868                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 869                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 870                         return -EMSGSIZE;
 871                 else if (output->tun_flags & TUNNEL_ERSPAN_OPT &&
 872                          nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
 873                                       ((struct erspan_metadata *)tun_opts)->index))
 874                         return -EMSGSIZE;
 875         }
 876
 877         return 0;
 878 }
 879
 880 static int ip_tun_to_nlattr(struct sk_buff *skb,
 881                             const struct ip_tunnel_key *output,
 882                             const void *tun_opts, int swkey_tun_opts_len,
 883                             unsigned short tun_proto)
 884 {
 885         struct nlattr *nla;
 886         int err;
 887
 888         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 889         if (!nla)
 890                 return -EMSGSIZE;
 891
 892         err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
 893                                  tun_proto);
 894         if (err)
 895                 return err;
 896
 897         nla_nest_end(skb, nla);
 898         return 0;
 899 }
 900
 901 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
 902                             struct ip_tunnel_info *tun_info)
 903 {
 904         return __ip_tun_to_nlattr(skb, &tun_info->key,
 905                                   ip_tunnel_info_opts(tun_info),
 906                                   tun_info->options_len,
 907                                   ip_tunnel_info_af(tun_info));
 908 }
 909
 910 static int encode_vlan_from_nlattrs(struct sw_flow_match *match,
 911                                     const struct nlattr *a[],
 912                                     bool is_mask, bool inner)
 913 {
 914         __be16 tci = 0;
 915         __be16 tpid = 0;
 916
 917         if (a[OVS_KEY_ATTR_VLAN])
 918                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 919
 920         if (a[OVS_KEY_ATTR_ETHERTYPE])
 921                 tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 922
 923         if (likely(!inner)) {
 924                 SW_FLOW_KEY_PUT(match, eth.vlan.tpid, tpid, is_mask);
 925                 SW_FLOW_KEY_PUT(match, eth.vlan.tci, tci, is_mask);
 926         } else {
 927                 SW_FLOW_KEY_PUT(match, eth.cvlan.tpid, tpid, is_mask);
 928                 SW_FLOW_KEY_PUT(match, eth.cvlan.tci, tci, is_mask);
 929         }
 930         return 0;
 931 }
 932
 933 static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
 934                                       u64 key_attrs, bool inner,
 935                                       const struct nlattr **a, bool log)
 936 {
 937         __be16 tci = 0;
 938
 939         if (!((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
 940               (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
 941                eth_type_vlan(nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE])))) {
 942                 /* Not a VLAN. */
 943                 return 0;
 944         }
 945
 946         if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
 947               (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
 948                 OVS_NLERR(log, "Invalid %s frame", (inner) ? "C-VLAN" : "VLAN");
 949                 return -EINVAL;
 950         }
 951
 952         if (a[OVS_KEY_ATTR_VLAN])
 953                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 954
 955         if (!(tci & htons(VLAN_TAG_PRESENT))) {
 956                 if (tci) {
 957                         OVS_NLERR(log, "%s TCI does not have VLAN_TAG_PRESENT bit set.",
 958                                   (inner) ? "C-VLAN" : "VLAN");
 959                         return -EINVAL;
 960                 } else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
 961                         /* Corner case for truncated VLAN header. */
 962                         OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
 963                                   (inner) ? "C-VLAN" : "VLAN");
 964                         return -EINVAL;
 965                 }
 966         }
 967
 968         return 1;
 969 }
 970
 971 static int validate_vlan_mask_from_nlattrs(const struct sw_flow_match *match,
 972                                            u64 key_attrs, bool inner,
 973                                            const struct nlattr **a, bool log)
 974 {
 975         __be16 tci = 0;
 976         __be16 tpid = 0;
 977         bool encap_valid = !!(match->key->eth.vlan.tci &
 978                               htons(VLAN_TAG_PRESENT));
 979         bool i_encap_valid = !!(match->key->eth.cvlan.tci &
 980                                 htons(VLAN_TAG_PRESENT));
 981
 982         if (!(key_attrs & (1 << OVS_KEY_ATTR_ENCAP))) {
 983                 /* Not a VLAN. */
 984                 return 0;
 985         }
 986
 987         if ((!inner && !encap_valid) || (inner && !i_encap_valid)) {
 988                 OVS_NLERR(log, "Encap mask attribute is set for non-%s frame.",
 989                           (inner) ? "C-VLAN" : "VLAN");
 990                 return -EINVAL;
 991         }
 992
 993         if (a[OVS_KEY_ATTR_VLAN])
 994                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 995
 996         if (a[OVS_KEY_ATTR_ETHERTYPE])
 997                 tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 998
 999         if (tpid != htons(0xffff)) {
1000                 OVS_NLERR(log, "Must have an exact match on %s TPID (mask=%x).",
1001                           (inner) ? "C-VLAN" : "VLAN", ntohs(tpid));
1002                 return -EINVAL;
1003         }
1004         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1005                 OVS_NLERR(log, "%s TCI mask does not have exact match for VLAN_TAG_PRESENT bit.",
1006                           (inner) ? "C-VLAN" : "VLAN");
1007                 return -EINVAL;
1008         }
1009
1010         return 1;
1011 }
1012
1013 static int __parse_vlan_from_nlattrs(struct sw_flow_match *match,
1014                                      u64 *key_attrs, bool inner,
1015                                      const struct nlattr **a, bool is_mask,
1016                                      bool log)
1017 {
1018         int err;
1019         const struct nlattr *encap;
1020
1021         if (!is_mask)
1022                 err = validate_vlan_from_nlattrs(match, *key_attrs, inner,
1023                                                  a, log);
1024         else
1025                 err = validate_vlan_mask_from_nlattrs(match, *key_attrs, inner,
1026                                                       a, log);
1027         if (err <= 0)
1028                 return err;
1029
1030         err = encode_vlan_from_nlattrs(match, a, is_mask, inner);
1031         if (err)
1032                 return err;
1033
1034         *key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1035         *key_attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
1036         *key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1037
1038         encap = a[OVS_KEY_ATTR_ENCAP];
1039
1040         if (!is_mask)
1041                 err = parse_flow_nlattrs(encap, a, key_attrs, log);
1042         else
1043                 err = parse_flow_mask_nlattrs(encap, a, key_attrs, log);
1044
1045         return err;
1046 }
1047
1048 static int parse_vlan_from_nlattrs(struct sw_flow_match *match,
1049                                    u64 *key_attrs, const struct nlattr **a,
1050                                    bool is_mask, bool log)
1051 {
1052         int err;
1053         bool encap_valid = false;
1054
1055         err = __parse_vlan_from_nlattrs(match, key_attrs, false, a,
1056                                         is_mask, log);
1057         if (err)
1058                 return err;
1059
1060         encap_valid = !!(match->key->eth.vlan.tci & htons(VLAN_TAG_PRESENT));
1061         if (encap_valid) {
1062                 err = __parse_vlan_from_nlattrs(match, key_attrs, true, a,
1063                                                 is_mask, log);
1064                 if (err)
1065                         return err;
1066         }
1067
1068         return 0;
1069 }
1070
1071 static int parse_eth_type_from_nlattrs(struct sw_flow_match *match,
1072                                        u64 *attrs, const struct nlattr **a,
1073                                        bool is_mask, bool log)
1074 {
1075         __be16 eth_type;
1076
1077         eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1078         if (is_mask) {
1079                 /* Always exact match EtherType. */
1080                 eth_type = htons(0xffff);
1081         } else if (!eth_proto_is_802_3(eth_type)) {
1082                 OVS_NLERR(log, "EtherType %x is less than min %x",
1083                                 ntohs(eth_type), ETH_P_802_3_MIN);
1084                 return -EINVAL;
1085         }
1086
1087         SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1088         *attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1089         return 0;
1090 }
1091
1092 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
1093                                  u64 *attrs, const struct nlattr **a,
1094                                  bool is_mask, bool log)
1095 {
1096         u8 mac_proto = MAC_PROTO_ETHERNET;
1097
1098         if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
1099                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
1100
1101                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
1102                 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
1103         }
1104
1105         if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
1106                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
1107
1108                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
1109                 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
1110         }
1111
1112         if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1113                 SW_FLOW_KEY_PUT(match, phy.priority,
1114                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1115                 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1116         }
1117
1118         if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1119                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1120
1121                 if (is_mask) {
1122                         in_port = 0xffffffff; /* Always exact match in_port. */
1123                 } else if (in_port >= DP_MAX_PORTS) {
1124                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
1125                                   in_port, DP_MAX_PORTS);
1126                         return -EINVAL;
1127                 }
1128
1129                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1130                 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1131         } else if (!is_mask) {
1132                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1133         }
1134
1135         if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1136                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1137
1138                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1139                 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1140         }
1141         if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1142                 if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1143                                        is_mask, log) < 0)
1144                         return -EINVAL;
1145                 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1146         }
1147
1148         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
1149             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
1150                 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
1151
1152                 if (ct_state & ~CT_SUPPORTED_MASK) {
1153                         OVS_NLERR(log, "ct_state flags %08x unsupported",
1154                                   ct_state);
1155                         return -EINVAL;
1156                 }
1157
1158                 SW_FLOW_KEY_PUT(match, ct_state, ct_state, is_mask);
1159                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
1160         }
1161         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
1162             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
1163                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
1164
1165                 SW_FLOW_KEY_PUT(match, ct_zone, ct_zone, is_mask);
1166                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
1167         }
1168         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
1169             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
1170                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
1171
1172                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
1173                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
1174         }
1175         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
1176             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
1177                 const struct ovs_key_ct_labels *cl;
1178
1179                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
1180                 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
1181                                    sizeof(*cl), is_mask);
1182                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
1183         }
1184         if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)) {
1185                 const struct ovs_key_ct_tuple_ipv4 *ct;
1186
1187                 ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4]);
1188
1189                 SW_FLOW_KEY_PUT(match, ipv4.ct_orig.src, ct->ipv4_src, is_mask);
1190                 SW_FLOW_KEY_PUT(match, ipv4.ct_orig.dst, ct->ipv4_dst, is_mask);
1191                 SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1192                 SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1193                 SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv4_proto, is_mask);
1194                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4);
1195         }
1196         if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)) {
1197                 const struct ovs_key_ct_tuple_ipv6 *ct;
1198
1199                 ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6]);
1200
1201                 SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.src, &ct->ipv6_src,
1202                                    sizeof(match->key->ipv6.ct_orig.src),
1203                                    is_mask);
1204                 SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.dst, &ct->ipv6_dst,
1205                                    sizeof(match->key->ipv6.ct_orig.dst),
1206                                    is_mask);
1207                 SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1208                 SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1209                 SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv6_proto, is_mask);
1210                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
1211         }
1212
1213         /* For layer 3 packets the Ethernet type is provided
1214          * and treated as metadata but no MAC addresses are provided.
1215          */
1216         if (!(*attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1217             (*attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)))
1218                 mac_proto = MAC_PROTO_NONE;
1219
1220         /* Always exact match mac_proto */
1221         SW_FLOW_KEY_PUT(match, mac_proto, is_mask ? 0xff : mac_proto, is_mask);
1222
1223         if (mac_proto == MAC_PROTO_NONE)
1224                 return parse_eth_type_from_nlattrs(match, attrs, a, is_mask,
1225                                                    log);
1226
1227         return 0;
1228 }
1229
1230 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
1231                                 u64 attrs, const struct nlattr **a,
1232                                 bool is_mask, bool log)
1233 {
1234         int err;
1235
1236         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
1237         if (err)
1238                 return err;
1239
1240         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1241                 const struct ovs_key_ethernet *eth_key;
1242
1243                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1244                 SW_FLOW_KEY_MEMCPY(match, eth.src,
1245                                 eth_key->eth_src, ETH_ALEN, is_mask);
1246                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1247                                 eth_key->eth_dst, ETH_ALEN, is_mask);
1248                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1249
1250                 if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1251                         /* VLAN attribute is always parsed before getting here since it
1252                          * may occur multiple times.
1253                          */
1254                         OVS_NLERR(log, "VLAN attribute unexpected.");
1255                         return -EINVAL;
1256                 }
1257
1258                 if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1259                         err = parse_eth_type_from_nlattrs(match, &attrs, a, is_mask,
1260                                                           log);
1261                         if (err)
1262                                 return err;
1263                 } else if (!is_mask) {
1264                         SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1265                 }
1266         } else if (!match->key->eth.type) {
1267                 OVS_NLERR(log, "Either Ethernet header or EtherType is required.");
1268                 return -EINVAL;
1269         }
1270
1271         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1272                 const struct ovs_key_ipv4 *ipv4_key;
1273
1274                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1275                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1276                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
1277                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1278                         return -EINVAL;
1279                 }
1280                 SW_FLOW_KEY_PUT(match, ip.proto,
1281                                 ipv4_key->ipv4_proto, is_mask);
1282                 SW_FLOW_KEY_PUT(match, ip.tos,
1283                                 ipv4_key->ipv4_tos, is_mask);
1284                 SW_FLOW_KEY_PUT(match, ip.ttl,
1285                                 ipv4_key->ipv4_ttl, is_mask);
1286                 SW_FLOW_KEY_PUT(match, ip.frag,
1287                                 ipv4_key->ipv4_frag, is_mask);
1288                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1289                                 ipv4_key->ipv4_src, is_mask);
1290                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1291                                 ipv4_key->ipv4_dst, is_mask);
1292                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1293         }
1294
1295         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1296                 const struct ovs_key_ipv6 *ipv6_key;
1297
1298                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1299                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1300                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
1301                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1302                         return -EINVAL;
1303                 }
1304
1305                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
1306                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x)",
1307                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1308                         return -EINVAL;
1309                 }
1310
1311                 SW_FLOW_KEY_PUT(match, ipv6.label,
1312                                 ipv6_key->ipv6_label, is_mask);
1313                 SW_FLOW_KEY_PUT(match, ip.proto,
1314                                 ipv6_key->ipv6_proto, is_mask);
1315                 SW_FLOW_KEY_PUT(match, ip.tos,
1316                                 ipv6_key->ipv6_tclass, is_mask);
1317                 SW_FLOW_KEY_PUT(match, ip.ttl,
1318                                 ipv6_key->ipv6_hlimit, is_mask);
1319                 SW_FLOW_KEY_PUT(match, ip.frag,
1320                                 ipv6_key->ipv6_frag, is_mask);
1321                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1322                                 ipv6_key->ipv6_src,
1323                                 sizeof(match->key->ipv6.addr.src),
1324                                 is_mask);
1325                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1326                                 ipv6_key->ipv6_dst,
1327                                 sizeof(match->key->ipv6.addr.dst),
1328                                 is_mask);
1329
1330                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1331         }
1332
1333         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1334                 const struct ovs_key_arp *arp_key;
1335
1336                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1337                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1338                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1339                                   arp_key->arp_op);
1340                         return -EINVAL;
1341                 }
1342
1343                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1344                                 arp_key->arp_sip, is_mask);
1345                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1346                         arp_key->arp_tip, is_mask);
1347                 SW_FLOW_KEY_PUT(match, ip.proto,
1348                                 ntohs(arp_key->arp_op), is_mask);
1349                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1350                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1351                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1352                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1353
1354                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1355         }
1356
1357         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1358                 const struct ovs_key_mpls *mpls_key;
1359
1360                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1361                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1362                                 mpls_key->mpls_lse, is_mask);
1363
1364                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1365          }
1366
1367         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1368                 const struct ovs_key_tcp *tcp_key;
1369
1370                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1371                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1372                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1373                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1374         }
1375
1376         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1377                 SW_FLOW_KEY_PUT(match, tp.flags,
1378                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1379                                 is_mask);
1380                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1381         }
1382
1383         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1384                 const struct ovs_key_udp *udp_key;
1385
1386                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1387                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1388                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1389                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1390         }
1391
1392         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1393                 const struct ovs_key_sctp *sctp_key;
1394
1395                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1396                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1397                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1398                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1399         }
1400
1401         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1402                 const struct ovs_key_icmp *icmp_key;
1403
1404                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1405                 SW_FLOW_KEY_PUT(match, tp.src,
1406                                 htons(icmp_key->icmp_type), is_mask);
1407                 SW_FLOW_KEY_PUT(match, tp.dst,
1408                                 htons(icmp_key->icmp_code), is_mask);
1409                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1410         }
1411
1412         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1413                 const struct ovs_key_icmpv6 *icmpv6_key;
1414
1415                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1416                 SW_FLOW_KEY_PUT(match, tp.src,
1417                                 htons(icmpv6_key->icmpv6_type), is_mask);
1418                 SW_FLOW_KEY_PUT(match, tp.dst,
1419                                 htons(icmpv6_key->icmpv6_code), is_mask);
1420                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1421         }
1422
1423         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1424                 const struct ovs_key_nd *nd_key;
1425
1426                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1427                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1428                         nd_key->nd_target,
1429                         sizeof(match->key->ipv6.nd.target),
1430                         is_mask);
1431                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1432                         nd_key->nd_sll, ETH_ALEN, is_mask);
1433                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1434                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1435                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1436         }
1437
1438         if (attrs != 0) {
1439                 OVS_NLERR(log, "Unknown key attributes %llx",
1440                           (unsigned long long)attrs);
1441                 return -EINVAL;
1442         }
1443
1444         return 0;
1445 }
1446
1447 static void nlattr_set(struct nlattr *attr, u8 val,
1448                        const struct ovs_len_tbl *tbl)
1449 {
1450         struct nlattr *nla;
1451         int rem;
1452
1453         /* The nlattr stream should already have been validated */
1454         nla_for_each_nested(nla, attr, rem) {
1455                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1456                         if (tbl[nla_type(nla)].next)
1457                                 tbl = tbl[nla_type(nla)].next;
1458                         nlattr_set(nla, val, tbl);
1459                 } else {
1460                         memset(nla_data(nla), val, nla_len(nla));
1461                 }
1462
1463                 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1464                         *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1465         }
1466 }
1467
1468 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1469 {
1470         nlattr_set(attr, val, ovs_key_lens);
1471 }
1472
1473 /**
1474  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1475  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1476  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1477  * does not include any don't care bit.
1478  * @net: Used to determine per-namespace field support.
1479  * @match: receives the extracted flow match information.
1480  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1481  * sequence. The fields should of the packet that triggered the creation
1482  * of this flow.
1483  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1484  * attribute specifies the mask field of the wildcarded flow.
1485  * @log: Boolean to allow kernel error logging.  Normally true, but when
1486  * probing for feature compatibility this should be passed in as false to
1487  * suppress unnecessary error logging.
1488  */
1489 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1490                       const struct nlattr *nla_key,
1491                       const struct nlattr *nla_mask,
1492                       bool log)
1493 {
1494         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1495         struct nlattr *newmask = NULL;
1496         u64 key_attrs = 0;
1497         u64 mask_attrs = 0;
1498         int err;
1499
1500         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1501         if (err)
1502                 return err;
1503
1504         err = parse_vlan_from_nlattrs(match, &key_attrs, a, false, log);
1505         if (err)
1506                 return err;
1507
1508         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1509         if (err)
1510                 return err;
1511
1512         if (match->mask) {
1513                 if (!nla_mask) {
1514                         /* Create an exact match mask. We need to set to 0xff
1515                          * all the 'match->mask' fields that have been touched
1516                          * in 'match->key'. We cannot simply memset
1517                          * 'match->mask', because padding bytes and fields not
1518                          * specified in 'match->key' should be left to 0.
1519                          * Instead, we use a stream of netlink attributes,
1520                          * copied from 'key' and set to 0xff.
1521                          * ovs_key_from_nlattrs() will take care of filling
1522                          * 'match->mask' appropriately.
1523                          */
1524                         newmask = kmemdup(nla_key,
1525                                           nla_total_size(nla_len(nla_key)),
1526                                           GFP_KERNEL);
1527                         if (!newmask)
1528                                 return -ENOMEM;
1529
1530                         mask_set_nlattr(newmask, 0xff);
1531
1532                         /* The userspace does not send tunnel attributes that
1533                          * are 0, but we should not wildcard them nonetheless.
1534                          */
1535                         if (match->key->tun_proto)
1536                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1537                                                          0xff, true);
1538
1539                         nla_mask = newmask;
1540                 }
1541
1542                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1543                 if (err)
1544                         goto free_newmask;
1545
1546                 /* Always match on tci. */
1547                 SW_FLOW_KEY_PUT(match, eth.vlan.tci, htons(0xffff), true);
1548                 SW_FLOW_KEY_PUT(match, eth.cvlan.tci, htons(0xffff), true);
1549
1550                 err = parse_vlan_from_nlattrs(match, &mask_attrs, a, true, log);
1551                 if (err)
1552                         goto free_newmask;
1553
1554                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1555                                            log);
1556                 if (err)
1557                         goto free_newmask;
1558         }
1559
1560         if (!match_validate(match, key_attrs, mask_attrs, log))
1561                 err = -EINVAL;
1562
1563 free_newmask:
1564         kfree(newmask);
1565         return err;
1566 }
1567
1568 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1569 {
1570         size_t len;
1571
1572         if (!attr)
1573                 return 0;
1574
1575         len = nla_len(attr);
1576         if (len < 1 || len > MAX_UFID_LENGTH) {
1577                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1578                           nla_len(attr), MAX_UFID_LENGTH);
1579                 return 0;
1580         }
1581
1582         return len;
1583 }
1584
1585 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1586  * or false otherwise.
1587  */
1588 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1589                       bool log)
1590 {
1591         sfid->ufid_len = get_ufid_len(attr, log);
1592         if (sfid->ufid_len)
1593                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1594
1595         return sfid->ufid_len;
1596 }
1597
1598 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1599                            const struct sw_flow_key *key, bool log)
1600 {
1601         struct sw_flow_key *new_key;
1602
1603         if (ovs_nla_get_ufid(sfid, ufid, log))
1604                 return 0;
1605
1606         /* If UFID was not provided, use unmasked key. */
1607         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1608         if (!new_key)
1609                 return -ENOMEM;
1610         memcpy(new_key, key, sizeof(*key));
1611         sfid->unmasked_key = new_key;
1612
1613         return 0;
1614 }
1615
1616 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1617 {
1618         return attr ? nla_get_u32(attr) : 0;
1619 }
1620
1621 /**
1622  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1623  * @net: Network namespace.
1624  * @key: Receives extracted in_port, priority, tun_key, skb_mark and conntrack
1625  * metadata.
1626  * @a: Array of netlink attributes holding parsed %OVS_KEY_ATTR_* Netlink
1627  * attributes.
1628  * @attrs: Bit mask for the netlink attributes included in @a.
1629  * @log: Boolean to allow kernel error logging.  Normally true, but when
1630  * probing for feature compatibility this should be passed in as false to
1631  * suppress unnecessary error logging.
1632  *
1633  * This parses a series of Netlink attributes that form a flow key, which must
1634  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1635  * get the metadata, that is, the parts of the flow key that cannot be
1636  * extracted from the packet itself.
1637  *
1638  * This must be called before the packet key fields are filled in 'key'.
1639  */
1640
1641 int ovs_nla_get_flow_metadata(struct net *net,
1642                               const struct nlattr *a[OVS_KEY_ATTR_MAX + 1],
1643                               u64 attrs, struct sw_flow_key *key, bool log)
1644 {
1645         struct sw_flow_match match;
1646
1647         memset(&match, 0, sizeof(match));
1648         match.key = key;
1649
1650         key->ct_state = 0;
1651         key->ct_zone = 0;
1652         key->ct_orig_proto = 0;
1653         memset(&key->ct, 0, sizeof(key->ct));
1654         memset(&key->ipv4.ct_orig, 0, sizeof(key->ipv4.ct_orig));
1655         memset(&key->ipv6.ct_orig, 0, sizeof(key->ipv6.ct_orig));
1656
1657         key->phy.in_port = DP_MAX_PORTS;
1658
1659         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1660 }
1661
1662 static int ovs_nla_put_vlan(struct sk_buff *skb, const struct vlan_head *vh,
1663                             bool is_mask)
1664 {
1665         __be16 eth_type = !is_mask ? vh->tpid : htons(0xffff);
1666
1667         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1668             nla_put_be16(skb, OVS_KEY_ATTR_VLAN, vh->tci))
1669                 return -EMSGSIZE;
1670         return 0;
1671 }
1672
1673 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1674                              const struct sw_flow_key *output, bool is_mask,
1675                              struct sk_buff *skb)
1676 {
1677         struct ovs_key_ethernet *eth_key;
1678         struct nlattr *nla;
1679         struct nlattr *encap = NULL;
1680         struct nlattr *in_encap = NULL;
1681
1682         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1683                 goto nla_put_failure;
1684
1685         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1686                 goto nla_put_failure;
1687
1688         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1689                 goto nla_put_failure;
1690
1691         if ((swkey->tun_proto || is_mask)) {
1692                 const void *opts = NULL;
1693
1694                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1695                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1696
1697                 if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
1698                                      swkey->tun_opts_len, swkey->tun_proto))
1699                         goto nla_put_failure;
1700         }
1701
1702         if (swkey->phy.in_port == DP_MAX_PORTS) {
1703                 if (is_mask && (output->phy.in_port == 0xffff))
1704                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1705                                 goto nla_put_failure;
1706         } else {
1707                 u16 upper_u16;
1708                 upper_u16 = !is_mask ? 0 : 0xffff;
1709
1710                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1711                                 (upper_u16 << 16) | output->phy.in_port))
1712                         goto nla_put_failure;
1713         }
1714
1715         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1716                 goto nla_put_failure;
1717
1718         if (ovs_ct_put_key(swkey, output, skb))
1719                 goto nla_put_failure;
1720
1721         if (ovs_key_mac_proto(swkey) == MAC_PROTO_ETHERNET) {
1722                 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1723                 if (!nla)
1724                         goto nla_put_failure;
1725
1726                 eth_key = nla_data(nla);
1727                 ether_addr_copy(eth_key->eth_src, output->eth.src);
1728                 ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1729
1730                 if (swkey->eth.vlan.tci || eth_type_vlan(swkey->eth.type)) {
1731                         if (ovs_nla_put_vlan(skb, &output->eth.vlan, is_mask))
1732                                 goto nla_put_failure;
1733                         encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1734                         if (!swkey->eth.vlan.tci)
1735                                 goto unencap;
1736
1737                         if (swkey->eth.cvlan.tci || eth_type_vlan(swkey->eth.type)) {
1738                                 if (ovs_nla_put_vlan(skb, &output->eth.cvlan, is_mask))
1739                                         goto nla_put_failure;
1740                                 in_encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1741                                 if (!swkey->eth.cvlan.tci)
1742                                         goto unencap;
1743                         }
1744                 }
1745
1746                 if (swkey->eth.type == htons(ETH_P_802_2)) {
1747                         /*
1748                         * Ethertype 802.2 is represented in the netlink with omitted
1749                         * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1750                         * 0xffff in the mask attribute.  Ethertype can also
1751                         * be wildcarded.
1752                         */
1753                         if (is_mask && output->eth.type)
1754                                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1755                                                         output->eth.type))
1756                                         goto nla_put_failure;
1757                         goto unencap;
1758                 }
1759         }
1760
1761         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1762                 goto nla_put_failure;
1763
1764         if (eth_type_vlan(swkey->eth.type)) {
1765                 /* There are 3 VLAN tags, we don't know anything about the rest
1766                  * of the packet, so truncate here.
1767                  */
1768                 WARN_ON_ONCE(!(encap && in_encap));
1769                 goto unencap;
1770         }
1771
1772         if (swkey->eth.type == htons(ETH_P_IP)) {
1773                 struct ovs_key_ipv4 *ipv4_key;
1774
1775                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1776                 if (!nla)
1777                         goto nla_put_failure;
1778                 ipv4_key = nla_data(nla);
1779                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1780                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1781                 ipv4_key->ipv4_proto = output->ip.proto;
1782                 ipv4_key->ipv4_tos = output->ip.tos;
1783                 ipv4_key->ipv4_ttl = output->ip.ttl;
1784                 ipv4_key->ipv4_frag = output->ip.frag;
1785         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1786                 struct ovs_key_ipv6 *ipv6_key;
1787
1788                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1789                 if (!nla)
1790                         goto nla_put_failure;
1791                 ipv6_key = nla_data(nla);
1792                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1793                                 sizeof(ipv6_key->ipv6_src));
1794                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1795                                 sizeof(ipv6_key->ipv6_dst));
1796                 ipv6_key->ipv6_label = output->ipv6.label;
1797                 ipv6_key->ipv6_proto = output->ip.proto;
1798                 ipv6_key->ipv6_tclass = output->ip.tos;
1799                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1800                 ipv6_key->ipv6_frag = output->ip.frag;
1801         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1802                    swkey->eth.type == htons(ETH_P_RARP)) {
1803                 struct ovs_key_arp *arp_key;
1804
1805                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1806                 if (!nla)
1807                         goto nla_put_failure;
1808                 arp_key = nla_data(nla);
1809                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1810                 arp_key->arp_sip = output->ipv4.addr.src;
1811                 arp_key->arp_tip = output->ipv4.addr.dst;
1812                 arp_key->arp_op = htons(output->ip.proto);
1813                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1814                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1815         } else if (eth_p_mpls(swkey->eth.type)) {
1816                 struct ovs_key_mpls *mpls_key;
1817
1818                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1819                 if (!nla)
1820                         goto nla_put_failure;
1821                 mpls_key = nla_data(nla);
1822                 mpls_key->mpls_lse = output->mpls.top_lse;
1823         }
1824
1825         if ((swkey->eth.type == htons(ETH_P_IP) ||
1826              swkey->eth.type == htons(ETH_P_IPV6)) &&
1827              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1828
1829                 if (swkey->ip.proto == IPPROTO_TCP) {
1830                         struct ovs_key_tcp *tcp_key;
1831
1832                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1833                         if (!nla)
1834                                 goto nla_put_failure;
1835                         tcp_key = nla_data(nla);
1836                         tcp_key->tcp_src = output->tp.src;
1837                         tcp_key->tcp_dst = output->tp.dst;
1838                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1839                                          output->tp.flags))
1840                                 goto nla_put_failure;
1841                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1842                         struct ovs_key_udp *udp_key;
1843
1844                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1845                         if (!nla)
1846                                 goto nla_put_failure;
1847                         udp_key = nla_data(nla);
1848                         udp_key->udp_src = output->tp.src;
1849                         udp_key->udp_dst = output->tp.dst;
1850                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1851                         struct ovs_key_sctp *sctp_key;
1852
1853                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1854                         if (!nla)
1855                                 goto nla_put_failure;
1856                         sctp_key = nla_data(nla);
1857                         sctp_key->sctp_src = output->tp.src;
1858                         sctp_key->sctp_dst = output->tp.dst;
1859                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1860                            swkey->ip.proto == IPPROTO_ICMP) {
1861                         struct ovs_key_icmp *icmp_key;
1862
1863                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1864                         if (!nla)
1865                                 goto nla_put_failure;
1866                         icmp_key = nla_data(nla);
1867                         icmp_key->icmp_type = ntohs(output->tp.src);
1868                         icmp_key->icmp_code = ntohs(output->tp.dst);
1869                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1870                            swkey->ip.proto == IPPROTO_ICMPV6) {
1871                         struct ovs_key_icmpv6 *icmpv6_key;
1872
1873                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1874                                                 sizeof(*icmpv6_key));
1875                         if (!nla)
1876                                 goto nla_put_failure;
1877                         icmpv6_key = nla_data(nla);
1878                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1879                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1880
1881                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1882                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1883                                 struct ovs_key_nd *nd_key;
1884
1885                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1886                                 if (!nla)
1887                                         goto nla_put_failure;
1888                                 nd_key = nla_data(nla);
1889                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1890                                                         sizeof(nd_key->nd_target));
1891                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1892                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1893                         }
1894                 }
1895         }
1896
1897 unencap:
1898         if (in_encap)
1899                 nla_nest_end(skb, in_encap);
1900         if (encap)
1901                 nla_nest_end(skb, encap);
1902
1903         return 0;
1904
1905 nla_put_failure:
1906         return -EMSGSIZE;
1907 }
1908
1909 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1910                     const struct sw_flow_key *output, int attr, bool is_mask,
1911                     struct sk_buff *skb)
1912 {
1913         int err;
1914         struct nlattr *nla;
1915
1916         nla = nla_nest_start(skb, attr);
1917         if (!nla)
1918                 return -EMSGSIZE;
1919         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1920         if (err)
1921                 return err;
1922         nla_nest_end(skb, nla);
1923
1924         return 0;
1925 }
1926
1927 /* Called with ovs_mutex or RCU read lock. */
1928 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1929 {
1930         if (ovs_identifier_is_ufid(&flow->id))
1931                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1932                                flow->id.ufid);
1933
1934         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1935                                OVS_FLOW_ATTR_KEY, false, skb);
1936 }
1937
1938 /* Called with ovs_mutex or RCU read lock. */
1939 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1940 {
1941         return ovs_nla_put_key(&flow->key, &flow->key,
1942                                 OVS_FLOW_ATTR_KEY, false, skb);
1943 }
1944
1945 /* Called with ovs_mutex or RCU read lock. */
1946 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1947 {
1948         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1949                                 OVS_FLOW_ATTR_MASK, true, skb);
1950 }
1951
1952 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1953
1954 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1955 {
1956         struct sw_flow_actions *sfa;
1957
1958         if (size > MAX_ACTIONS_BUFSIZE) {
1959                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1960                 return ERR_PTR(-EINVAL);
1961         }
1962
1963         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1964         if (!sfa)
1965                 return ERR_PTR(-ENOMEM);
1966
1967         sfa->actions_len = 0;
1968         return sfa;
1969 }
1970
1971 static void ovs_nla_free_set_action(const struct nlattr *a)
1972 {
1973         const struct nlattr *ovs_key = nla_data(a);
1974         struct ovs_tunnel_info *ovs_tun;
1975
1976         switch (nla_type(ovs_key)) {
1977         case OVS_KEY_ATTR_TUNNEL_INFO:
1978                 ovs_tun = nla_data(ovs_key);
1979                 dst_release((struct dst_entry *)ovs_tun->tun_dst);
1980                 break;
1981         }
1982 }
1983
1984 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1985 {
1986         const struct nlattr *a;
1987         int rem;
1988
1989         if (!sf_acts)
1990                 return;
1991
1992         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1993                 switch (nla_type(a)) {
1994                 case OVS_ACTION_ATTR_SET:
1995                         ovs_nla_free_set_action(a);
1996                         break;
1997                 case OVS_ACTION_ATTR_CT:
1998                         ovs_ct_free_action(a);
1999                         break;
2000                 }
2001         }
2002
2003         kfree(sf_acts);
2004 }
2005
2006 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
2007 {
2008         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
2009 }
2010
2011 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
2012  * The caller must hold rcu_read_lock for this to be sensible. */
2013 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
2014 {
2015         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
2016 }
2017
2018 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
2019                                        int attr_len, bool log)
2020 {
2021
2022         struct sw_flow_actions *acts;
2023         int new_acts_size;
2024         int req_size = NLA_ALIGN(attr_len);
2025         int next_offset = offsetof(struct sw_flow_actions, actions) +
2026                                         (*sfa)->actions_len;
2027
2028         if (req_size <= (ksize(*sfa) - next_offset))
2029                 goto out;
2030
2031         new_acts_size = ksize(*sfa) * 2;
2032
2033         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
2034                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
2035                         return ERR_PTR(-EMSGSIZE);
2036                 new_acts_size = MAX_ACTIONS_BUFSIZE;
2037         }
2038
2039         acts = nla_alloc_flow_actions(new_acts_size, log);
2040         if (IS_ERR(acts))
2041                 return (void *)acts;
2042
2043         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
2044         acts->actions_len = (*sfa)->actions_len;
2045         acts->orig_len = (*sfa)->orig_len;
2046         kfree(*sfa);
2047         *sfa = acts;
2048
2049 out:
2050         (*sfa)->actions_len += req_size;
2051         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
2052 }
2053
2054 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
2055                                    int attrtype, void *data, int len, bool log)
2056 {
2057         struct nlattr *a;
2058
2059         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
2060         if (IS_ERR(a))
2061                 return a;
2062
2063         a->nla_type = attrtype;
2064         a->nla_len = nla_attr_size(len);
2065
2066         if (data)
2067                 memcpy(nla_data(a), data, len);
2068         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
2069
2070         return a;
2071 }
2072
2073 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
2074                        int len, bool log)
2075 {
2076         struct nlattr *a;
2077
2078         a = __add_action(sfa, attrtype, data, len, log);
2079
2080         return PTR_ERR_OR_ZERO(a);
2081 }
2082
2083 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
2084                                           int attrtype, bool log)
2085 {
2086         int used = (*sfa)->actions_len;
2087         int err;
2088
2089         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
2090         if (err)
2091                 return err;
2092
2093         return used;
2094 }
2095
2096 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
2097                                          int st_offset)
2098 {
2099         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
2100                                                                st_offset);
2101
2102         a->nla_len = sfa->actions_len - st_offset;
2103 }
2104
2105 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2106                                   const struct sw_flow_key *key,
2107                                   struct sw_flow_actions **sfa,
2108                                   __be16 eth_type, __be16 vlan_tci, bool log);
2109
2110 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
2111                                     const struct sw_flow_key *key,
2112                                     struct sw_flow_actions **sfa,
2113                                     __be16 eth_type, __be16 vlan_tci,
2114                                     bool log, bool last)
2115 {
2116         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
2117         const struct nlattr *probability, *actions;
2118         const struct nlattr *a;
2119         int rem, start, err;
2120         struct sample_arg arg;
2121
2122         memset(attrs, 0, sizeof(attrs));
2123         nla_for_each_nested(a, attr, rem) {
2124                 int type = nla_type(a);
2125                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
2126                         return -EINVAL;
2127                 attrs[type] = a;
2128         }
2129         if (rem)
2130                 return -EINVAL;
2131
2132         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
2133         if (!probability || nla_len(probability) != sizeof(u32))
2134                 return -EINVAL;
2135
2136         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
2137         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
2138                 return -EINVAL;
2139
2140         /* validation done, copy sample action. */
2141         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
2142         if (start < 0)
2143                 return start;
2144
2145         /* When both skb and flow may be changed, put the sample
2146          * into a deferred fifo. On the other hand, if only skb
2147          * may be modified, the actions can be executed in place.
2148          *
2149          * Do this analysis at the flow installation time.
2150          * Set 'clone_action->exec' to true if the actions can be
2151          * executed without being deferred.
2152          *
2153          * If the sample is the last action, it can always be excuted
2154          * rather than deferred.
2155          */
2156         arg.exec = last || !actions_may_change_flow(actions);
2157         arg.probability = nla_get_u32(probability);
2158
2159         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_ARG, &arg, sizeof(arg),
2160                                  log);
2161         if (err)
2162                 return err;
2163
2164         err = __ovs_nla_copy_actions(net, actions, key, sfa,
2165                                      eth_type, vlan_tci, log);
2166
2167         if (err)
2168                 return err;
2169
2170         add_nested_action_end(*sfa, start);
2171
2172         return 0;
2173 }
2174
2175 void ovs_match_init(struct sw_flow_match *match,
2176                     struct sw_flow_key *key,
2177                     bool reset_key,
2178                     struct sw_flow_mask *mask)
2179 {
2180         memset(match, 0, sizeof(*match));
2181         match->key = key;
2182         match->mask = mask;
2183
2184         if (reset_key)
2185                 memset(key, 0, sizeof(*key));
2186
2187         if (mask) {
2188                 memset(&mask->key, 0, sizeof(mask->key));
2189                 mask->range.start = mask->range.end = 0;
2190         }
2191 }
2192
2193 static int validate_geneve_opts(struct sw_flow_key *key)
2194 {
2195         struct geneve_opt *option;
2196         int opts_len = key->tun_opts_len;
2197         bool crit_opt = false;
2198
2199         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
2200         while (opts_len > 0) {
2201                 int len;
2202
2203                 if (opts_len < sizeof(*option))
2204                         return -EINVAL;
2205
2206                 len = sizeof(*option) + option->length * 4;
2207                 if (len > opts_len)
2208                         return -EINVAL;
2209
2210                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
2211
2212                 option = (struct geneve_opt *)((u8 *)option + len);
2213                 opts_len -= len;
2214         };
2215
2216         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
2217
2218         return 0;
2219 }
2220
2221 static int validate_and_copy_set_tun(const struct nlattr *attr,
2222                                      struct sw_flow_actions **sfa, bool log)
2223 {
2224         struct sw_flow_match match;
2225         struct sw_flow_key key;
2226         struct metadata_dst *tun_dst;
2227         struct ip_tunnel_info *tun_info;
2228         struct ovs_tunnel_info *ovs_tun;
2229         struct nlattr *a;
2230         int err = 0, start, opts_type;
2231
2232         ovs_match_init(&match, &key, true, NULL);
2233         opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
2234         if (opts_type < 0)
2235                 return opts_type;
2236
2237         if (key.tun_opts_len) {
2238                 switch (opts_type) {
2239                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2240                         err = validate_geneve_opts(&key);
2241                         if (err < 0)
2242                                 return err;
2243                         break;
2244                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2245                         break;
2246                 case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
2247                         break;
2248                 }
2249         };
2250
2251         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
2252         if (start < 0)
2253                 return start;
2254
2255         tun_dst = metadata_dst_alloc(key.tun_opts_len, METADATA_IP_TUNNEL,
2256                                      GFP_KERNEL);
2257
2258         if (!tun_dst)
2259                 return -ENOMEM;
2260
2261         err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
2262         if (err) {
2263                 dst_release((struct dst_entry *)tun_dst);
2264                 return err;
2265         }
2266
2267         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
2268                          sizeof(*ovs_tun), log);
2269         if (IS_ERR(a)) {
2270                 dst_release((struct dst_entry *)tun_dst);
2271                 return PTR_ERR(a);
2272         }
2273
2274         ovs_tun = nla_data(a);
2275         ovs_tun->tun_dst = tun_dst;
2276
2277         tun_info = &tun_dst->u.tun_info;
2278         tun_info->mode = IP_TUNNEL_INFO_TX;
2279         if (key.tun_proto == AF_INET6)
2280                 tun_info->mode |= IP_TUNNEL_INFO_IPV6;
2281         tun_info->key = key.tun_key;
2282
2283         /* We need to store the options in the action itself since
2284          * everything else will go away after flow setup. We can append
2285          * it to tun_info and then point there.
2286          */
2287         ip_tunnel_info_opts_set(tun_info,
2288                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
2289                                 key.tun_opts_len);
2290         add_nested_action_end(*sfa, start);
2291
2292         return err;
2293 }
2294
2295 /* Return false if there are any non-masked bits set.
2296  * Mask follows data immediately, before any netlink padding.
2297  */
2298 static bool validate_masked(u8 *data, int len)
2299 {
2300         u8 *mask = data + len;
2301
2302         while (len--)
2303                 if (*data++ & ~*mask++)
2304                         return false;
2305
2306         return true;
2307 }
2308
2309 static int validate_set(const struct nlattr *a,
2310                         const struct sw_flow_key *flow_key,
2311                         struct sw_flow_actions **sfa, bool *skip_copy,
2312                         u8 mac_proto, __be16 eth_type, bool masked, bool log)
2313 {
2314         const struct nlattr *ovs_key = nla_data(a);
2315         int key_type = nla_type(ovs_key);
2316         size_t key_len;
2317
2318         /* There can be only one key in a action */
2319         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2320                 return -EINVAL;
2321
2322         key_len = nla_len(ovs_key);
2323         if (masked)
2324                 key_len /= 2;
2325
2326         if (key_type > OVS_KEY_ATTR_MAX ||
2327             !check_attr_len(key_len, ovs_key_lens[key_type].len))
2328                 return -EINVAL;
2329
2330         if (masked && !validate_masked(nla_data(ovs_key), key_len))
2331                 return -EINVAL;
2332
2333         switch (key_type) {
2334         const struct ovs_key_ipv4 *ipv4_key;
2335         const struct ovs_key_ipv6 *ipv6_key;
2336         int err;
2337
2338         case OVS_KEY_ATTR_PRIORITY:
2339         case OVS_KEY_ATTR_SKB_MARK:
2340         case OVS_KEY_ATTR_CT_MARK:
2341         case OVS_KEY_ATTR_CT_LABELS:
2342                 break;
2343
2344         case OVS_KEY_ATTR_ETHERNET:
2345                 if (mac_proto != MAC_PROTO_ETHERNET)
2346                         return -EINVAL;
2347                 break;
2348
2349         case OVS_KEY_ATTR_TUNNEL:
2350                 if (masked)
2351                         return -EINVAL; /* Masked tunnel set not supported. */
2352
2353                 *skip_copy = true;
2354                 err = validate_and_copy_set_tun(a, sfa, log);
2355                 if (err)
2356                         return err;
2357                 break;
2358
2359         case OVS_KEY_ATTR_IPV4:
2360                 if (eth_type != htons(ETH_P_IP))
2361                         return -EINVAL;
2362
2363                 ipv4_key = nla_data(ovs_key);
2364
2365                 if (masked) {
2366                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2367
2368                         /* Non-writeable fields. */
2369                         if (mask->ipv4_proto || mask->ipv4_frag)
2370                                 return -EINVAL;
2371                 } else {
2372                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2373                                 return -EINVAL;
2374
2375                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2376                                 return -EINVAL;
2377                 }
2378                 break;
2379
2380         case OVS_KEY_ATTR_IPV6:
2381                 if (eth_type != htons(ETH_P_IPV6))
2382                         return -EINVAL;
2383
2384                 ipv6_key = nla_data(ovs_key);
2385
2386                 if (masked) {
2387                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2388
2389                         /* Non-writeable fields. */
2390                         if (mask->ipv6_proto || mask->ipv6_frag)
2391                                 return -EINVAL;
2392
2393                         /* Invalid bits in the flow label mask? */
2394                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2395                                 return -EINVAL;
2396                 } else {
2397                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2398                                 return -EINVAL;
2399
2400                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2401                                 return -EINVAL;
2402                 }
2403                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2404                         return -EINVAL;
2405
2406                 break;
2407
2408         case OVS_KEY_ATTR_TCP:
2409                 if ((eth_type != htons(ETH_P_IP) &&
2410                      eth_type != htons(ETH_P_IPV6)) ||
2411                     flow_key->ip.proto != IPPROTO_TCP)
2412                         return -EINVAL;
2413
2414                 break;
2415
2416         case OVS_KEY_ATTR_UDP:
2417                 if ((eth_type != htons(ETH_P_IP) &&
2418                      eth_type != htons(ETH_P_IPV6)) ||
2419                     flow_key->ip.proto != IPPROTO_UDP)
2420                         return -EINVAL;
2421
2422                 break;
2423
2424         case OVS_KEY_ATTR_MPLS:
2425                 if (!eth_p_mpls(eth_type))
2426                         return -EINVAL;
2427                 break;
2428
2429         case OVS_KEY_ATTR_SCTP:
2430                 if ((eth_type != htons(ETH_P_IP) &&
2431                      eth_type != htons(ETH_P_IPV6)) ||
2432                     flow_key->ip.proto != IPPROTO_SCTP)
2433                         return -EINVAL;
2434
2435                 break;
2436
2437         default:
2438                 return -EINVAL;
2439         }
2440
2441         /* Convert non-masked non-tunnel set actions to masked set actions. */
2442         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2443                 int start, len = key_len * 2;
2444                 struct nlattr *at;
2445
2446                 *skip_copy = true;
2447
2448                 start = add_nested_action_start(sfa,
2449                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2450                                                 log);
2451                 if (start < 0)
2452                         return start;
2453
2454                 at = __add_action(sfa, key_type, NULL, len, log);
2455                 if (IS_ERR(at))
2456                         return PTR_ERR(at);
2457
2458                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2459                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2460                 /* Clear non-writeable bits from otherwise writeable fields. */
2461                 if (key_type == OVS_KEY_ATTR_IPV6) {
2462                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2463
2464                         mask->ipv6_label &= htonl(0x000FFFFF);
2465                 }
2466                 add_nested_action_end(*sfa, start);
2467         }
2468
2469         return 0;
2470 }
2471
2472 static int validate_userspace(const struct nlattr *attr)
2473 {
2474         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2475                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2476                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2477                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2478         };
2479         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2480         int error;
2481
2482         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, attr,
2483                                  userspace_policy, NULL);
2484         if (error)
2485                 return error;
2486
2487         if (!a[OVS_USERSPACE_ATTR_PID] ||
2488             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2489                 return -EINVAL;
2490
2491         return 0;
2492 }
2493
2494 static int copy_action(const struct nlattr *from,
2495                        struct sw_flow_actions **sfa, bool log)
2496 {
2497         int totlen = NLA_ALIGN(from->nla_len);
2498         struct nlattr *to;
2499
2500         to = reserve_sfa_size(sfa, from->nla_len, log);
2501         if (IS_ERR(to))
2502                 return PTR_ERR(to);
2503
2504         memcpy(to, from, totlen);
2505         return 0;
2506 }
2507
2508 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2509                                   const struct sw_flow_key *key,
2510                                   struct sw_flow_actions **sfa,
2511                                   __be16 eth_type, __be16 vlan_tci, bool log)
2512 {
2513         u8 mac_proto = ovs_key_mac_proto(key);
2514         const struct nlattr *a;
2515         int rem, err;
2516
2517         nla_for_each_nested(a, attr, rem) {
2518                 /* Expected argument lengths, (u32)-1 for variable length. */
2519                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2520                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2521                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2522                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2523                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2524                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2525                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2526                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2527                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2528                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2529                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2530                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2531                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2532                         [OVS_ACTION_ATTR_CT_CLEAR] = 0,
2533                         [OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
2534                         [OVS_ACTION_ATTR_PUSH_ETH] = sizeof(struct ovs_action_push_eth),
2535                         [OVS_ACTION_ATTR_POP_ETH] = 0,
2536                 };
2537                 const struct ovs_action_push_vlan *vlan;
2538                 int type = nla_type(a);
2539                 bool skip_copy;
2540
2541                 if (type > OVS_ACTION_ATTR_MAX ||
2542                     (action_lens[type] != nla_len(a) &&
2543                      action_lens[type] != (u32)-1))
2544                         return -EINVAL;
2545
2546                 skip_copy = false;
2547                 switch (type) {
2548                 case OVS_ACTION_ATTR_UNSPEC:
2549                         return -EINVAL;
2550
2551                 case OVS_ACTION_ATTR_USERSPACE:
2552                         err = validate_userspace(a);
2553                         if (err)
2554                                 return err;
2555                         break;
2556
2557                 case OVS_ACTION_ATTR_OUTPUT:
2558                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2559                                 return -EINVAL;
2560                         break;
2561
2562                 case OVS_ACTION_ATTR_TRUNC: {
2563                         const struct ovs_action_trunc *trunc = nla_data(a);
2564
2565                         if (trunc->max_len < ETH_HLEN)
2566                                 return -EINVAL;
2567                         break;
2568                 }
2569
2570                 case OVS_ACTION_ATTR_HASH: {
2571                         const struct ovs_action_hash *act_hash = nla_data(a);
2572
2573                         switch (act_hash->hash_alg) {
2574                         case OVS_HASH_ALG_L4:
2575                                 break;
2576                         default:
2577                                 return  -EINVAL;
2578                         }
2579
2580                         break;
2581                 }
2582
2583                 case OVS_ACTION_ATTR_POP_VLAN:
2584                         if (mac_proto != MAC_PROTO_ETHERNET)
2585                                 return -EINVAL;
2586                         vlan_tci = htons(0);
2587                         break;
2588
2589                 case OVS_ACTION_ATTR_PUSH_VLAN:
2590                         if (mac_proto != MAC_PROTO_ETHERNET)
2591                                 return -EINVAL;
2592                         vlan = nla_data(a);
2593                         if (!eth_type_vlan(vlan->vlan_tpid))
2594                                 return -EINVAL;
2595                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2596                                 return -EINVAL;
2597                         vlan_tci = vlan->vlan_tci;
2598                         break;
2599
2600                 case OVS_ACTION_ATTR_RECIRC:
2601                         break;
2602
2603                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2604                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2605
2606                         if (!eth_p_mpls(mpls->mpls_ethertype))
2607                                 return -EINVAL;
2608                         /* Prohibit push MPLS other than to a white list
2609                          * for packets that have a known tag order.
2610                          */
2611                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2612                             (eth_type != htons(ETH_P_IP) &&
2613                              eth_type != htons(ETH_P_IPV6) &&
2614                              eth_type != htons(ETH_P_ARP) &&
2615                              eth_type != htons(ETH_P_RARP) &&
2616                              !eth_p_mpls(eth_type)))
2617                                 return -EINVAL;
2618                         eth_type = mpls->mpls_ethertype;
2619                         break;
2620                 }
2621
2622                 case OVS_ACTION_ATTR_POP_MPLS:
2623                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2624                             !eth_p_mpls(eth_type))
2625                                 return -EINVAL;
2626
2627                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2628                          * as there is no check here to ensure that the new
2629                          * eth_type is valid and thus set actions could
2630                          * write off the end of the packet or otherwise
2631                          * corrupt it.
2632                          *
2633                          * Support for these actions is planned using packet
2634                          * recirculation.
2635                          */
2636                         eth_type = htons(0);
2637                         break;
2638
2639                 case OVS_ACTION_ATTR_SET:
2640                         err = validate_set(a, key, sfa,
2641                                            &skip_copy, mac_proto, eth_type,
2642                                            false, log);
2643                         if (err)
2644                                 return err;
2645                         break;
2646
2647                 case OVS_ACTION_ATTR_SET_MASKED:
2648                         err = validate_set(a, key, sfa,
2649                                            &skip_copy, mac_proto, eth_type,
2650                                            true, log);
2651                         if (err)
2652                                 return err;
2653                         break;
2654
2655                 case OVS_ACTION_ATTR_SAMPLE: {
2656                         bool last = nla_is_last(a, rem);
2657
2658                         err = validate_and_copy_sample(net, a, key, sfa,
2659                                                        eth_type, vlan_tci,
2660                                                        log, last);
2661                         if (err)
2662                                 return err;
2663                         skip_copy = true;
2664                         break;
2665                 }
2666
2667                 case OVS_ACTION_ATTR_CT:
2668                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2669                         if (err)
2670                                 return err;
2671                         skip_copy = true;
2672                         break;
2673
2674                 case OVS_ACTION_ATTR_CT_CLEAR:
2675                         break;
2676
2677                 case OVS_ACTION_ATTR_PUSH_ETH:
2678                         /* Disallow pushing an Ethernet header if one
2679                          * is already present */
2680                         if (mac_proto != MAC_PROTO_NONE)
2681                                 return -EINVAL;
2682                         mac_proto = MAC_PROTO_NONE;
2683                         break;
2684
2685                 case OVS_ACTION_ATTR_POP_ETH:
2686                         if (mac_proto != MAC_PROTO_ETHERNET)
2687                                 return -EINVAL;
2688                         if (vlan_tci & htons(VLAN_TAG_PRESENT))
2689                                 return -EINVAL;
2690                         mac_proto = MAC_PROTO_ETHERNET;
2691                         break;
2692
2693                 default:
2694                         OVS_NLERR(log, "Unknown Action type %d", type);
2695                         return -EINVAL;
2696                 }
2697                 if (!skip_copy) {
2698                         err = copy_action(a, sfa, log);
2699                         if (err)
2700                                 return err;
2701                 }
2702         }
2703
2704         if (rem > 0)
2705                 return -EINVAL;
2706
2707         return 0;
2708 }
2709
2710 /* 'key' must be the masked key. */
2711 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2712                          const struct sw_flow_key *key,
2713                          struct sw_flow_actions **sfa, bool log)
2714 {
2715         int err;
2716
2717         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2718         if (IS_ERR(*sfa))
2719                 return PTR_ERR(*sfa);
2720
2721         (*sfa)->orig_len = nla_len(attr);
2722         err = __ovs_nla_copy_actions(net, attr, key, sfa, key->eth.type,
2723                                      key->eth.vlan.tci, log);
2724         if (err)
2725                 ovs_nla_free_flow_actions(*sfa);
2726
2727         return err;
2728 }
2729
2730 static int sample_action_to_attr(const struct nlattr *attr,
2731                                  struct sk_buff *skb)
2732 {
2733         struct nlattr *start, *ac_start = NULL, *sample_arg;
2734         int err = 0, rem = nla_len(attr);
2735         const struct sample_arg *arg;
2736         struct nlattr *actions;
2737
2738         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2739         if (!start)
2740                 return -EMSGSIZE;
2741
2742         sample_arg = nla_data(attr);
2743         arg = nla_data(sample_arg);
2744         actions = nla_next(sample_arg, &rem);
2745
2746         if (nla_put_u32(skb, OVS_SAMPLE_ATTR_PROBABILITY, arg->probability)) {
2747                 err = -EMSGSIZE;
2748                 goto out;
2749         }
2750
2751         ac_start = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2752         if (!ac_start) {
2753                 err = -EMSGSIZE;
2754                 goto out;
2755         }
2756
2757         err = ovs_nla_put_actions(actions, rem, skb);
2758
2759 out:
2760         if (err) {
2761                 nla_nest_cancel(skb, ac_start);
2762                 nla_nest_cancel(skb, start);
2763         } else {
2764                 nla_nest_end(skb, ac_start);
2765                 nla_nest_end(skb, start);
2766         }
2767
2768         return err;
2769 }
2770
2771 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2772 {
2773         const struct nlattr *ovs_key = nla_data(a);
2774         int key_type = nla_type(ovs_key);
2775         struct nlattr *start;
2776         int err;
2777
2778         switch (key_type) {
2779         case OVS_KEY_ATTR_TUNNEL_INFO: {
2780                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2781                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2782
2783                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2784                 if (!start)
2785                         return -EMSGSIZE;
2786
2787                 err =  ip_tun_to_nlattr(skb, &tun_info->key,
2788                                         ip_tunnel_info_opts(tun_info),
2789                                         tun_info->options_len,
2790                                         ip_tunnel_info_af(tun_info));
2791                 if (err)
2792                         return err;
2793                 nla_nest_end(skb, start);
2794                 break;
2795         }
2796         default:
2797                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2798                         return -EMSGSIZE;
2799                 break;
2800         }
2801
2802         return 0;
2803 }
2804
2805 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2806                                                 struct sk_buff *skb)
2807 {
2808         const struct nlattr *ovs_key = nla_data(a);
2809         struct nlattr *nla;
2810         size_t key_len = nla_len(ovs_key) / 2;
2811
2812         /* Revert the conversion we did from a non-masked set action to
2813          * masked set action.
2814          */
2815         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2816         if (!nla)
2817                 return -EMSGSIZE;
2818
2819         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2820                 return -EMSGSIZE;
2821
2822         nla_nest_end(skb, nla);
2823         return 0;
2824 }
2825
2826 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2827 {
2828         const struct nlattr *a;
2829         int rem, err;
2830
2831         nla_for_each_attr(a, attr, len, rem) {
2832                 int type = nla_type(a);
2833
2834                 switch (type) {
2835                 case OVS_ACTION_ATTR_SET:
2836                         err = set_action_to_attr(a, skb);
2837                         if (err)
2838                                 return err;
2839                         break;
2840
2841                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2842                         err = masked_set_action_to_set_action_attr(a, skb);
2843                         if (err)
2844                                 return err;
2845                         break;
2846
2847                 case OVS_ACTION_ATTR_SAMPLE:
2848                         err = sample_action_to_attr(a, skb);
2849                         if (err)
2850                                 return err;
2851                         break;
2852
2853                 case OVS_ACTION_ATTR_CT:
2854                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2855                         if (err)
2856                                 return err;
2857                         break;
2858
2859                 default:
2860                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2861                                 return -EMSGSIZE;
2862                         break;
2863                 }
2864         }
2865
2866         return 0;
2867 }