net/openvswitch/flow_netlink.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2007-2017 Nicira, Inc.
   4  */
   5
   6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8 #include "flow.h"
   9 #include "datapath.h"
  10 #include <linux/uaccess.h>
  11 #include <linux/netdevice.h>
  12 #include <linux/etherdevice.h>
  13 #include <linux/if_ether.h>
  14 #include <linux/if_vlan.h>
  15 #include <net/llc_pdu.h>
  16 #include <linux/kernel.h>
  17 #include <linux/jhash.h>
  18 #include <linux/jiffies.h>
  19 #include <linux/llc.h>
  20 #include <linux/module.h>
  21 #include <linux/in.h>
  22 #include <linux/rcupdate.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/ip.h>
  25 #include <linux/ipv6.h>
  26 #include <linux/sctp.h>
  27 #include <linux/tcp.h>
  28 #include <linux/udp.h>
  29 #include <linux/icmp.h>
  30 #include <linux/icmpv6.h>
  31 #include <linux/rculist.h>
  32 #include <net/geneve.h>
  33 #include <net/ip.h>
  34 #include <net/ipv6.h>
  35 #include <net/ndisc.h>
  36 #include <net/mpls.h>
  37 #include <net/vxlan.h>
  38 #include <net/tun_proto.h>
  39 #include <net/erspan.h>
  40
  41 #include "flow_netlink.h"
  42
  43 struct ovs_len_tbl {
  44         int len;
  45         const struct ovs_len_tbl *next;
  46 };
  47
  48 #define OVS_ATTR_NESTED -1
  49 #define OVS_ATTR_VARIABLE -2
  50 #define OVS_COPY_ACTIONS_MAX_DEPTH 16
  51
  52 static bool actions_may_change_flow(const struct nlattr *actions)
  53 {
  54         struct nlattr *nla;
  55         int rem;
  56
  57         nla_for_each_nested(nla, actions, rem) {
  58                 u16 action = nla_type(nla);
  59
  60                 switch (action) {
  61                 case OVS_ACTION_ATTR_OUTPUT:
  62                 case OVS_ACTION_ATTR_RECIRC:
  63                 case OVS_ACTION_ATTR_TRUNC:
  64                 case OVS_ACTION_ATTR_USERSPACE:
  65                         break;
  66
  67                 case OVS_ACTION_ATTR_CT:
  68                 case OVS_ACTION_ATTR_CT_CLEAR:
  69                 case OVS_ACTION_ATTR_HASH:
  70                 case OVS_ACTION_ATTR_POP_ETH:
  71                 case OVS_ACTION_ATTR_POP_MPLS:
  72                 case OVS_ACTION_ATTR_POP_NSH:
  73                 case OVS_ACTION_ATTR_POP_VLAN:
  74                 case OVS_ACTION_ATTR_PUSH_ETH:
  75                 case OVS_ACTION_ATTR_PUSH_MPLS:
  76                 case OVS_ACTION_ATTR_PUSH_NSH:
  77                 case OVS_ACTION_ATTR_PUSH_VLAN:
  78                 case OVS_ACTION_ATTR_SAMPLE:
  79                 case OVS_ACTION_ATTR_SET:
  80                 case OVS_ACTION_ATTR_SET_MASKED:
  81                 case OVS_ACTION_ATTR_METER:
  82                 case OVS_ACTION_ATTR_CHECK_PKT_LEN:
  83                 case OVS_ACTION_ATTR_ADD_MPLS:
  84                 case OVS_ACTION_ATTR_DEC_TTL:
  85                 default:
  86                         return true;
  87                 }
  88         }
  89         return false;
  90 }
  91
  92 static void update_range(struct sw_flow_match *match,
  93                          size_t offset, size_t size, bool is_mask)
  94 {
  95         struct sw_flow_key_range *range;
  96         size_t start = rounddown(offset, sizeof(long));
  97         size_t end = roundup(offset + size, sizeof(long));
  98
  99         if (!is_mask)
 100                 range = &match->range;
 101         else
 102                 range = &match->mask->range;
 103
 104         if (range->start == range->end) {
 105                 range->start = start;
 106                 range->end = end;
 107                 return;
 108         }
 109
 110         if (range->start > start)
 111                 range->start = start;
 112
 113         if (range->end < end)
 114                 range->end = end;
 115 }
 116
 117 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
 118         do { \
 119                 update_range(match, offsetof(struct sw_flow_key, field),    \
 120                              sizeof((match)->key->field), is_mask);         \
 121                 if (is_mask)                                                \
 122                         (match)->mask->key.field = value;                   \
 123                 else                                                        \
 124                         (match)->key->field = value;                        \
 125         } while (0)
 126
 127 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
 128         do {                                                                \
 129                 update_range(match, offset, len, is_mask);                  \
 130                 if (is_mask)                                                \
 131                         memcpy((u8 *)&(match)->mask->key + offset, value_p, \
 132                                len);                                       \
 133                 else                                                        \
 134                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
 135         } while (0)
 136
 137 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
 138         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 139                                   value_p, len, is_mask)
 140
 141 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 142         do {                                                                \
 143                 update_range(match, offsetof(struct sw_flow_key, field),    \
 144                              sizeof((match)->key->field), is_mask);         \
 145                 if (is_mask)                                                \
 146                         memset((u8 *)&(match)->mask->key.field, value,      \
 147                                sizeof((match)->mask->key.field));           \
 148                 else                                                        \
 149                         memset((u8 *)&(match)->key->field, value,           \
 150                                sizeof((match)->key->field));                \
 151         } while (0)
 152
 153 static bool match_validate(const struct sw_flow_match *match,
 154                            u64 key_attrs, u64 mask_attrs, bool log)
 155 {
 156         u64 key_expected = 0;
 157         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 158
 159         /* The following mask attributes allowed only if they
 160          * pass the validation tests. */
 161         mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
 162                         | (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)
 163                         | (1 << OVS_KEY_ATTR_IPV6)
 164                         | (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)
 165                         | (1 << OVS_KEY_ATTR_TCP)
 166                         | (1 << OVS_KEY_ATTR_TCP_FLAGS)
 167                         | (1 << OVS_KEY_ATTR_UDP)
 168                         | (1 << OVS_KEY_ATTR_SCTP)
 169                         | (1 << OVS_KEY_ATTR_ICMP)
 170                         | (1 << OVS_KEY_ATTR_ICMPV6)
 171                         | (1 << OVS_KEY_ATTR_ARP)
 172                         | (1 << OVS_KEY_ATTR_ND)
 173                         | (1 << OVS_KEY_ATTR_MPLS)
 174                         | (1 << OVS_KEY_ATTR_NSH));
 175
 176         /* Always allowed mask fields. */
 177         mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
 178                        | (1 << OVS_KEY_ATTR_IN_PORT)
 179                        | (1 << OVS_KEY_ATTR_ETHERTYPE));
 180
 181         /* Check key attributes. */
 182         if (match->key->eth.type == htons(ETH_P_ARP)
 183                         || match->key->eth.type == htons(ETH_P_RARP)) {
 184                 key_expected |= 1 << OVS_KEY_ATTR_ARP;
 185                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 186                         mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
 187         }
 188
 189         if (eth_p_mpls(match->key->eth.type)) {
 190                 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
 191                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 192                         mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
 193         }
 194
 195         if (match->key->eth.type == htons(ETH_P_IP)) {
 196                 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
 197                 if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
 198                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
 199                         mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4;
 200                 }
 201
 202                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 203                         if (match->key->ip.proto == IPPROTO_UDP) {
 204                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 205                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 206                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 207                         }
 208
 209                         if (match->key->ip.proto == IPPROTO_SCTP) {
 210                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 211                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 212                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 213                         }
 214
 215                         if (match->key->ip.proto == IPPROTO_TCP) {
 216                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 217                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 218                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 219                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 220                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 221                                 }
 222                         }
 223
 224                         if (match->key->ip.proto == IPPROTO_ICMP) {
 225                                 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
 226                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 227                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
 228                         }
 229                 }
 230         }
 231
 232         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 233                 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
 234                 if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
 235                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
 236                         mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6;
 237                 }
 238
 239                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 240                         if (match->key->ip.proto == IPPROTO_UDP) {
 241                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 242                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 243                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 244                         }
 245
 246                         if (match->key->ip.proto == IPPROTO_SCTP) {
 247                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 248                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 249                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 250                         }
 251
 252                         if (match->key->ip.proto == IPPROTO_TCP) {
 253                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 254                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 255                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 256                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 257                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 258                                 }
 259                         }
 260
 261                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 262                                 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
 263                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 264                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
 265
 266                                 if (match->key->tp.src ==
 267                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 268                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 269                                         key_expected |= 1 << OVS_KEY_ATTR_ND;
 270                                         /* Original direction conntrack tuple
 271                                          * uses the same space as the ND fields
 272                                          * in the key, so both are not allowed
 273                                          * at the same time.
 274                                          */
 275                                         mask_allowed &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
 276                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 277                                                 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
 278                                 }
 279                         }
 280                 }
 281         }
 282
 283         if (match->key->eth.type == htons(ETH_P_NSH)) {
 284                 key_expected |= 1 << OVS_KEY_ATTR_NSH;
 285                 if (match->mask &&
 286                     match->mask->key.eth.type == htons(0xffff)) {
 287                         mask_allowed |= 1 << OVS_KEY_ATTR_NSH;
 288                 }
 289         }
 290
 291         if ((key_attrs & key_expected) != key_expected) {
 292                 /* Key attributes check failed. */
 293                 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
 294                           (unsigned long long)key_attrs,
 295                           (unsigned long long)key_expected);
 296                 return false;
 297         }
 298
 299         if ((mask_attrs & mask_allowed) != mask_attrs) {
 300                 /* Mask attributes check failed. */
 301                 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
 302                           (unsigned long long)mask_attrs,
 303                           (unsigned long long)mask_allowed);
 304                 return false;
 305         }
 306
 307         return true;
 308 }
 309
 310 size_t ovs_tun_key_attr_size(void)
 311 {
 312         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 313          * updating this function.
 314          */
 315         return    nla_total_size_64bit(8) /* OVS_TUNNEL_KEY_ATTR_ID */
 316                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
 317                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
 318                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
 319                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
 320                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 321                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
 322                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
 323                 + nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 324                 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS and
 325                  * OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS is mutually exclusive with
 326                  * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
 327                  */
 328                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
 329                 + nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
 330 }
 331
 332 static size_t ovs_nsh_key_attr_size(void)
 333 {
 334         /* Whenever adding new OVS_NSH_KEY_ FIELDS, we should consider
 335          * updating this function.
 336          */
 337         return  nla_total_size(NSH_BASE_HDR_LEN) /* OVS_NSH_KEY_ATTR_BASE */
 338                 /* OVS_NSH_KEY_ATTR_MD1 and OVS_NSH_KEY_ATTR_MD2 are
 339                  * mutually exclusive, so the bigger one can cover
 340                  * the small one.
 341                  */
 342                 + nla_total_size(NSH_CTX_HDRS_MAX_LEN);
 343 }
 344
 345 size_t ovs_key_attr_size(void)
 346 {
 347         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 348          * updating this function.
 349          */
 350         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 29);
 351
 352         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 353                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 354                   + ovs_tun_key_attr_size()
 355                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 356                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 357                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 358                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 359                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
 360                 + nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
 361                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
 362                 + nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
 363                 + nla_total_size(40)  /* OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6 */
 364                 + nla_total_size(0)   /* OVS_KEY_ATTR_NSH */
 365                   + ovs_nsh_key_attr_size()
 366                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 367                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 368                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 369                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 370                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 371                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 372                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 373                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 374 }
 375
 376 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
 377         [OVS_VXLAN_EXT_GBP]         = { .len = sizeof(u32) },
 378 };
 379
 380 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 381         [OVS_TUNNEL_KEY_ATTR_ID]            = { .len = sizeof(u64) },
 382         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC]      = { .len = sizeof(u32) },
 383         [OVS_TUNNEL_KEY_ATTR_IPV4_DST]      = { .len = sizeof(u32) },
 384         [OVS_TUNNEL_KEY_ATTR_TOS]           = { .len = 1 },
 385         [OVS_TUNNEL_KEY_ATTR_TTL]           = { .len = 1 },
 386         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
 387         [OVS_TUNNEL_KEY_ATTR_CSUM]          = { .len = 0 },
 388         [OVS_TUNNEL_KEY_ATTR_TP_SRC]        = { .len = sizeof(u16) },
 389         [OVS_TUNNEL_KEY_ATTR_TP_DST]        = { .len = sizeof(u16) },
 390         [OVS_TUNNEL_KEY_ATTR_OAM]           = { .len = 0 },
 391         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
 392         [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
 393                                                 .next = ovs_vxlan_ext_key_lens },
 394         [OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
 395         [OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
 396         [OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS]   = { .len = OVS_ATTR_VARIABLE },
 397         [OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE]   = { .len = 0 },
 398 };
 399
 400 static const struct ovs_len_tbl
 401 ovs_nsh_key_attr_lens[OVS_NSH_KEY_ATTR_MAX + 1] = {
 402         [OVS_NSH_KEY_ATTR_BASE] = { .len = sizeof(struct ovs_nsh_key_base) },
 403         [OVS_NSH_KEY_ATTR_MD1]  = { .len = sizeof(struct ovs_nsh_key_md1) },
 404         [OVS_NSH_KEY_ATTR_MD2]  = { .len = OVS_ATTR_VARIABLE },
 405 };
 406
 407 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 408 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 409         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 410         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 411         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 412         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 413         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 414         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 415         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 416         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 417         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 418         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 419         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 420         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 421         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 422         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 423         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 424         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 425         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 426         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 427         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 428         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 429                                      .next = ovs_tunnel_key_lens, },
 430         [OVS_KEY_ATTR_MPLS]      = { .len = OVS_ATTR_VARIABLE },
 431         [OVS_KEY_ATTR_CT_STATE]  = { .len = sizeof(u32) },
 432         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 433         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 434         [OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
 435         [OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4] = {
 436                 .len = sizeof(struct ovs_key_ct_tuple_ipv4) },
 437         [OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6] = {
 438                 .len = sizeof(struct ovs_key_ct_tuple_ipv6) },
 439         [OVS_KEY_ATTR_NSH]       = { .len = OVS_ATTR_NESTED,
 440                                      .next = ovs_nsh_key_attr_lens, },
 441 };
 442
 443 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
 444 {
 445         return expected_len == attr_len ||
 446                expected_len == OVS_ATTR_NESTED ||
 447                expected_len == OVS_ATTR_VARIABLE;
 448 }
 449
 450 static bool is_all_zero(const u8 *fp, size_t size)
 451 {
 452         int i;
 453
 454         if (!fp)
 455                 return false;
 456
 457         for (i = 0; i < size; i++)
 458                 if (fp[i])
 459                         return false;
 460
 461         return true;
 462 }
 463
 464 static int __parse_flow_nlattrs(const struct nlattr *attr,
 465                                 const struct nlattr *a[],
 466                                 u64 *attrsp, bool log, bool nz)
 467 {
 468         const struct nlattr *nla;
 469         u64 attrs;
 470         int rem;
 471
 472         attrs = *attrsp;
 473         nla_for_each_nested(nla, attr, rem) {
 474                 u16 type = nla_type(nla);
 475                 int expected_len;
 476
 477                 if (type > OVS_KEY_ATTR_MAX) {
 478                         OVS_NLERR(log, "Key type %d is out of range max %d",
 479                                   type, OVS_KEY_ATTR_MAX);
 480                         return -EINVAL;
 481                 }
 482
 483                 if (attrs & (1 << type)) {
 484                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 485                         return -EINVAL;
 486                 }
 487
 488                 expected_len = ovs_key_lens[type].len;
 489                 if (!check_attr_len(nla_len(nla), expected_len)) {
 490                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 491                                   type, nla_len(nla), expected_len);
 492                         return -EINVAL;
 493                 }
 494
 495                 if (!nz || !is_all_zero(nla_data(nla), nla_len(nla))) {
 496                         attrs |= 1 << type;
 497                         a[type] = nla;
 498                 }
 499         }
 500         if (rem) {
 501                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 502                 return -EINVAL;
 503         }
 504
 505         *attrsp = attrs;
 506         return 0;
 507 }
 508
 509 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 510                                    const struct nlattr *a[], u64 *attrsp,
 511                                    bool log)
 512 {
 513         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 514 }
 515
 516 int parse_flow_nlattrs(const struct nlattr *attr, const struct nlattr *a[],
 517                        u64 *attrsp, bool log)
 518 {
 519         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 520 }
 521
 522 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 523                                      struct sw_flow_match *match, bool is_mask,
 524                                      bool log)
 525 {
 526         unsigned long opt_key_offset;
 527
 528         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 529                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 530                           nla_len(a), sizeof(match->key->tun_opts));
 531                 return -EINVAL;
 532         }
 533
 534         if (nla_len(a) % 4 != 0) {
 535                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 536                           nla_len(a));
 537                 return -EINVAL;
 538         }
 539
 540         /* We need to record the length of the options passed
 541          * down, otherwise packets with the same format but
 542          * additional options will be silently matched.
 543          */
 544         if (!is_mask) {
 545                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 546                                 false);
 547         } else {
 548                 /* This is somewhat unusual because it looks at
 549                  * both the key and mask while parsing the
 550                  * attributes (and by extension assumes the key
 551                  * is parsed first). Normally, we would verify
 552                  * that each is the correct length and that the
 553                  * attributes line up in the validate function.
 554                  * However, that is difficult because this is
 555                  * variable length and we won't have the
 556                  * information later.
 557                  */
 558                 if (match->key->tun_opts_len != nla_len(a)) {
 559                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 560                                   match->key->tun_opts_len, nla_len(a));
 561                         return -EINVAL;
 562                 }
 563
 564                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 565         }
 566
 567         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 568         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 569                                   nla_len(a), is_mask);
 570         return 0;
 571 }
 572
 573 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
 574                                      struct sw_flow_match *match, bool is_mask,
 575                                      bool log)
 576 {
 577         struct nlattr *a;
 578         int rem;
 579         unsigned long opt_key_offset;
 580         struct vxlan_metadata opts;
 581
 582         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 583
 584         memset(&opts, 0, sizeof(opts));
 585         nla_for_each_nested(a, attr, rem) {
 586                 int type = nla_type(a);
 587
 588                 if (type > OVS_VXLAN_EXT_MAX) {
 589                         OVS_NLERR(log, "VXLAN extension %d out of range max %d",
 590                                   type, OVS_VXLAN_EXT_MAX);
 591                         return -EINVAL;
 592                 }
 593
 594                 if (!check_attr_len(nla_len(a),
 595                                     ovs_vxlan_ext_key_lens[type].len)) {
 596                         OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
 597                                   type, nla_len(a),
 598                                   ovs_vxlan_ext_key_lens[type].len);
 599                         return -EINVAL;
 600                 }
 601
 602                 switch (type) {
 603                 case OVS_VXLAN_EXT_GBP:
 604                         opts.gbp = nla_get_u32(a);
 605                         break;
 606                 default:
 607                         OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
 608                                   type);
 609                         return -EINVAL;
 610                 }
 611         }
 612         if (rem) {
 613                 OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
 614                           rem);
 615                 return -EINVAL;
 616         }
 617
 618         if (!is_mask)
 619                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 620         else
 621                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 622
 623         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 624         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 625                                   is_mask);
 626         return 0;
 627 }
 628
 629 static int erspan_tun_opt_from_nlattr(const struct nlattr *a,
 630                                       struct sw_flow_match *match, bool is_mask,
 631                                       bool log)
 632 {
 633         unsigned long opt_key_offset;
 634
 635         BUILD_BUG_ON(sizeof(struct erspan_metadata) >
 636                      sizeof(match->key->tun_opts));
 637
 638         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 639                 OVS_NLERR(log, "ERSPAN option length err (len %d, max %zu).",
 640                           nla_len(a), sizeof(match->key->tun_opts));
 641                 return -EINVAL;
 642         }
 643
 644         if (!is_mask)
 645                 SW_FLOW_KEY_PUT(match, tun_opts_len,
 646                                 sizeof(struct erspan_metadata), false);
 647         else
 648                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 649
 650         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 651         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 652                                   nla_len(a), is_mask);
 653         return 0;
 654 }
 655
 656 static int ip_tun_from_nlattr(const struct nlattr *attr,
 657                               struct sw_flow_match *match, bool is_mask,
 658                               bool log)
 659 {
 660         bool ttl = false, ipv4 = false, ipv6 = false;
 661         bool info_bridge_mode = false;
 662         __be16 tun_flags = 0;
 663         int opts_type = 0;
 664         struct nlattr *a;
 665         int rem;
 666
 667         nla_for_each_nested(a, attr, rem) {
 668                 int type = nla_type(a);
 669                 int err;
 670
 671                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 672                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 673                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 674                         return -EINVAL;
 675                 }
 676
 677                 if (!check_attr_len(nla_len(a),
 678                                     ovs_tunnel_key_lens[type].len)) {
 679                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 680                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 681                         return -EINVAL;
 682                 }
 683
 684                 switch (type) {
 685                 case OVS_TUNNEL_KEY_ATTR_ID:
 686                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 687                                         nla_get_be64(a), is_mask);
 688                         tun_flags |= TUNNEL_KEY;
 689                         break;
 690                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 691                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
 692                                         nla_get_in_addr(a), is_mask);
 693                         ipv4 = true;
 694                         break;
 695                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 696                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
 697                                         nla_get_in_addr(a), is_mask);
 698                         ipv4 = true;
 699                         break;
 700                 case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
 701                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.src,
 702                                         nla_get_in6_addr(a), is_mask);
 703                         ipv6 = true;
 704                         break;
 705                 case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
 706                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
 707                                         nla_get_in6_addr(a), is_mask);
 708                         ipv6 = true;
 709                         break;
 710                 case OVS_TUNNEL_KEY_ATTR_TOS:
 711                         SW_FLOW_KEY_PUT(match, tun_key.tos,
 712                                         nla_get_u8(a), is_mask);
 713                         break;
 714                 case OVS_TUNNEL_KEY_ATTR_TTL:
 715                         SW_FLOW_KEY_PUT(match, tun_key.ttl,
 716                                         nla_get_u8(a), is_mask);
 717                         ttl = true;
 718                         break;
 719                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 720                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 721                         break;
 722                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 723                         tun_flags |= TUNNEL_CSUM;
 724                         break;
 725                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 726                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 727                                         nla_get_be16(a), is_mask);
 728                         break;
 729                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 730                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 731                                         nla_get_be16(a), is_mask);
 732                         break;
 733                 case OVS_TUNNEL_KEY_ATTR_OAM:
 734                         tun_flags |= TUNNEL_OAM;
 735                         break;
 736                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 737                         if (opts_type) {
 738                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 739                                 return -EINVAL;
 740                         }
 741
 742                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 743                         if (err)
 744                                 return err;
 745
 746                         tun_flags |= TUNNEL_GENEVE_OPT;
 747                         opts_type = type;
 748                         break;
 749                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 750                         if (opts_type) {
 751                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 752                                 return -EINVAL;
 753                         }
 754
 755                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 756                         if (err)
 757                                 return err;
 758
 759                         tun_flags |= TUNNEL_VXLAN_OPT;
 760                         opts_type = type;
 761                         break;
 762                 case OVS_TUNNEL_KEY_ATTR_PAD:
 763                         break;
 764                 case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
 765                         if (opts_type) {
 766                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 767                                 return -EINVAL;
 768                         }
 769
 770                         err = erspan_tun_opt_from_nlattr(a, match, is_mask,
 771                                                          log);
 772                         if (err)
 773                                 return err;
 774
 775                         tun_flags |= TUNNEL_ERSPAN_OPT;
 776                         opts_type = type;
 777                         break;
 778                 case OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE:
 779                         info_bridge_mode = true;
 780                         ipv4 = true;
 781                         break;
 782                 default:
 783                         OVS_NLERR(log, "Unknown IP tunnel attribute %d",
 784                                   type);
 785                         return -EINVAL;
 786                 }
 787         }
 788
 789         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 790         if (is_mask)
 791                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
 792         else
 793                 SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
 794                                 false);
 795
 796         if (rem > 0) {
 797                 OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
 798                           rem);
 799                 return -EINVAL;
 800         }
 801
 802         if (ipv4 && ipv6) {
 803                 OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
 804                 return -EINVAL;
 805         }
 806
 807         if (!is_mask) {
 808                 if (!ipv4 && !ipv6) {
 809                         OVS_NLERR(log, "IP tunnel dst address not specified");
 810                         return -EINVAL;
 811                 }
 812                 if (ipv4) {
 813                         if (info_bridge_mode) {
 814                                 if (match->key->tun_key.u.ipv4.src ||
 815                                     match->key->tun_key.u.ipv4.dst ||
 816                                     match->key->tun_key.tp_src ||
 817                                     match->key->tun_key.tp_dst ||
 818                                     match->key->tun_key.ttl ||
 819                                     match->key->tun_key.tos ||
 820                                     tun_flags & ~TUNNEL_KEY) {
 821                                         OVS_NLERR(log, "IPv4 tun info is not correct");
 822                                         return -EINVAL;
 823                                 }
 824                         } else if (!match->key->tun_key.u.ipv4.dst) {
 825                                 OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 826                                 return -EINVAL;
 827                         }
 828                 }
 829                 if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
 830                         OVS_NLERR(log, "IPv6 tunnel dst address is zero");
 831                         return -EINVAL;
 832                 }
 833
 834                 if (!ttl && !info_bridge_mode) {
 835                         OVS_NLERR(log, "IP tunnel TTL not specified.");
 836                         return -EINVAL;
 837                 }
 838         }
 839
 840         return opts_type;
 841 }
 842
 843 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 844                                const void *tun_opts, int swkey_tun_opts_len)
 845 {
 846         const struct vxlan_metadata *opts = tun_opts;
 847         struct nlattr *nla;
 848
 849         nla = nla_nest_start_noflag(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 850         if (!nla)
 851                 return -EMSGSIZE;
 852
 853         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 854                 return -EMSGSIZE;
 855
 856         nla_nest_end(skb, nla);
 857         return 0;
 858 }
 859
 860 static int __ip_tun_to_nlattr(struct sk_buff *skb,
 861                               const struct ip_tunnel_key *output,
 862                               const void *tun_opts, int swkey_tun_opts_len,
 863                               unsigned short tun_proto, u8 mode)
 864 {
 865         if (output->tun_flags & TUNNEL_KEY &&
 866             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id,
 867                          OVS_TUNNEL_KEY_ATTR_PAD))
 868                 return -EMSGSIZE;
 869
 870         if (mode & IP_TUNNEL_INFO_BRIDGE)
 871                 return nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE)
 872                        ? -EMSGSIZE : 0;
 873
 874         switch (tun_proto) {
 875         case AF_INET:
 876                 if (output->u.ipv4.src &&
 877                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 878                                     output->u.ipv4.src))
 879                         return -EMSGSIZE;
 880                 if (output->u.ipv4.dst &&
 881                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 882                                     output->u.ipv4.dst))
 883                         return -EMSGSIZE;
 884                 break;
 885         case AF_INET6:
 886                 if (!ipv6_addr_any(&output->u.ipv6.src) &&
 887                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
 888                                      &output->u.ipv6.src))
 889                         return -EMSGSIZE;
 890                 if (!ipv6_addr_any(&output->u.ipv6.dst) &&
 891                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
 892                                      &output->u.ipv6.dst))
 893                         return -EMSGSIZE;
 894                 break;
 895         }
 896         if (output->tos &&
 897             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 898                 return -EMSGSIZE;
 899         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 900                 return -EMSGSIZE;
 901         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 902             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 903                 return -EMSGSIZE;
 904         if ((output->tun_flags & TUNNEL_CSUM) &&
 905             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 906                 return -EMSGSIZE;
 907         if (output->tp_src &&
 908             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 909                 return -EMSGSIZE;
 910         if (output->tp_dst &&
 911             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 912                 return -EMSGSIZE;
 913         if ((output->tun_flags & TUNNEL_OAM) &&
 914             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 915                 return -EMSGSIZE;
 916         if (swkey_tun_opts_len) {
 917                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 918                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 919                             swkey_tun_opts_len, tun_opts))
 920                         return -EMSGSIZE;
 921                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 922                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 923                         return -EMSGSIZE;
 924                 else if (output->tun_flags & TUNNEL_ERSPAN_OPT &&
 925                          nla_put(skb, OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
 926                                  swkey_tun_opts_len, tun_opts))
 927                         return -EMSGSIZE;
 928         }
 929
 930         return 0;
 931 }
 932
 933 static int ip_tun_to_nlattr(struct sk_buff *skb,
 934                             const struct ip_tunnel_key *output,
 935                             const void *tun_opts, int swkey_tun_opts_len,
 936                             unsigned short tun_proto, u8 mode)
 937 {
 938         struct nlattr *nla;
 939         int err;
 940
 941         nla = nla_nest_start_noflag(skb, OVS_KEY_ATTR_TUNNEL);
 942         if (!nla)
 943                 return -EMSGSIZE;
 944
 945         err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
 946                                  tun_proto, mode);
 947         if (err)
 948                 return err;
 949
 950         nla_nest_end(skb, nla);
 951         return 0;
 952 }
 953
 954 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
 955                             struct ip_tunnel_info *tun_info)
 956 {
 957         return __ip_tun_to_nlattr(skb, &tun_info->key,
 958                                   ip_tunnel_info_opts(tun_info),
 959                                   tun_info->options_len,
 960                                   ip_tunnel_info_af(tun_info), tun_info->mode);
 961 }
 962
 963 static int encode_vlan_from_nlattrs(struct sw_flow_match *match,
 964                                     const struct nlattr *a[],
 965                                     bool is_mask, bool inner)
 966 {
 967         __be16 tci = 0;
 968         __be16 tpid = 0;
 969
 970         if (a[OVS_KEY_ATTR_VLAN])
 971                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 972
 973         if (a[OVS_KEY_ATTR_ETHERTYPE])
 974                 tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 975
 976         if (likely(!inner)) {
 977                 SW_FLOW_KEY_PUT(match, eth.vlan.tpid, tpid, is_mask);
 978                 SW_FLOW_KEY_PUT(match, eth.vlan.tci, tci, is_mask);
 979         } else {
 980                 SW_FLOW_KEY_PUT(match, eth.cvlan.tpid, tpid, is_mask);
 981                 SW_FLOW_KEY_PUT(match, eth.cvlan.tci, tci, is_mask);
 982         }
 983         return 0;
 984 }
 985
 986 static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
 987                                       u64 key_attrs, bool inner,
 988                                       const struct nlattr **a, bool log)
 989 {
 990         __be16 tci = 0;
 991
 992         if (!((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
 993               (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
 994                eth_type_vlan(nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE])))) {
 995                 /* Not a VLAN. */
 996                 return 0;
 997         }
 998
 999         if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1000               (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1001                 OVS_NLERR(log, "Invalid %s frame", (inner) ? "C-VLAN" : "VLAN");
1002                 return -EINVAL;
1003         }
1004
1005         if (a[OVS_KEY_ATTR_VLAN])
1006                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1007
1008         if (!(tci & htons(VLAN_CFI_MASK))) {
1009                 if (tci) {
1010                         OVS_NLERR(log, "%s TCI does not have VLAN_CFI_MASK bit set.",
1011                                   (inner) ? "C-VLAN" : "VLAN");
1012                         return -EINVAL;
1013                 } else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
1014                         /* Corner case for truncated VLAN header. */
1015                         OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
1016                                   (inner) ? "C-VLAN" : "VLAN");
1017                         return -EINVAL;
1018                 }
1019         }
1020
1021         return 1;
1022 }
1023
1024 static int validate_vlan_mask_from_nlattrs(const struct sw_flow_match *match,
1025                                            u64 key_attrs, bool inner,
1026                                            const struct nlattr **a, bool log)
1027 {
1028         __be16 tci = 0;
1029         __be16 tpid = 0;
1030         bool encap_valid = !!(match->key->eth.vlan.tci &
1031                               htons(VLAN_CFI_MASK));
1032         bool i_encap_valid = !!(match->key->eth.cvlan.tci &
1033                                 htons(VLAN_CFI_MASK));
1034
1035         if (!(key_attrs & (1 << OVS_KEY_ATTR_ENCAP))) {
1036                 /* Not a VLAN. */
1037                 return 0;
1038         }
1039
1040         if ((!inner && !encap_valid) || (inner && !i_encap_valid)) {
1041                 OVS_NLERR(log, "Encap mask attribute is set for non-%s frame.",
1042                           (inner) ? "C-VLAN" : "VLAN");
1043                 return -EINVAL;
1044         }
1045
1046         if (a[OVS_KEY_ATTR_VLAN])
1047                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1048
1049         if (a[OVS_KEY_ATTR_ETHERTYPE])
1050                 tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1051
1052         if (tpid != htons(0xffff)) {
1053                 OVS_NLERR(log, "Must have an exact match on %s TPID (mask=%x).",
1054                           (inner) ? "C-VLAN" : "VLAN", ntohs(tpid));
1055                 return -EINVAL;
1056         }
1057         if (!(tci & htons(VLAN_CFI_MASK))) {
1058                 OVS_NLERR(log, "%s TCI mask does not have exact match for VLAN_CFI_MASK bit.",
1059                           (inner) ? "C-VLAN" : "VLAN");
1060                 return -EINVAL;
1061         }
1062
1063         return 1;
1064 }
1065
1066 static int __parse_vlan_from_nlattrs(struct sw_flow_match *match,
1067                                      u64 *key_attrs, bool inner,
1068                                      const struct nlattr **a, bool is_mask,
1069                                      bool log)
1070 {
1071         int err;
1072         const struct nlattr *encap;
1073
1074         if (!is_mask)
1075                 err = validate_vlan_from_nlattrs(match, *key_attrs, inner,
1076                                                  a, log);
1077         else
1078                 err = validate_vlan_mask_from_nlattrs(match, *key_attrs, inner,
1079                                                       a, log);
1080         if (err <= 0)
1081                 return err;
1082
1083         err = encode_vlan_from_nlattrs(match, a, is_mask, inner);
1084         if (err)
1085                 return err;
1086
1087         *key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1088         *key_attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
1089         *key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1090
1091         encap = a[OVS_KEY_ATTR_ENCAP];
1092
1093         if (!is_mask)
1094                 err = parse_flow_nlattrs(encap, a, key_attrs, log);
1095         else
1096                 err = parse_flow_mask_nlattrs(encap, a, key_attrs, log);
1097
1098         return err;
1099 }
1100
1101 static int parse_vlan_from_nlattrs(struct sw_flow_match *match,
1102                                    u64 *key_attrs, const struct nlattr **a,
1103                                    bool is_mask, bool log)
1104 {
1105         int err;
1106         bool encap_valid = false;
1107
1108         err = __parse_vlan_from_nlattrs(match, key_attrs, false, a,
1109                                         is_mask, log);
1110         if (err)
1111                 return err;
1112
1113         encap_valid = !!(match->key->eth.vlan.tci & htons(VLAN_CFI_MASK));
1114         if (encap_valid) {
1115                 err = __parse_vlan_from_nlattrs(match, key_attrs, true, a,
1116                                                 is_mask, log);
1117                 if (err)
1118                         return err;
1119         }
1120
1121         return 0;
1122 }
1123
1124 static int parse_eth_type_from_nlattrs(struct sw_flow_match *match,
1125                                        u64 *attrs, const struct nlattr **a,
1126                                        bool is_mask, bool log)
1127 {
1128         __be16 eth_type;
1129
1130         eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1131         if (is_mask) {
1132                 /* Always exact match EtherType. */
1133                 eth_type = htons(0xffff);
1134         } else if (!eth_proto_is_802_3(eth_type)) {
1135                 OVS_NLERR(log, "EtherType %x is less than min %x",
1136                                 ntohs(eth_type), ETH_P_802_3_MIN);
1137                 return -EINVAL;
1138         }
1139
1140         SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1141         *attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1142         return 0;
1143 }
1144
1145 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
1146                                  u64 *attrs, const struct nlattr **a,
1147                                  bool is_mask, bool log)
1148 {
1149         u8 mac_proto = MAC_PROTO_ETHERNET;
1150
1151         if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
1152                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
1153
1154                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
1155                 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
1156         }
1157
1158         if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
1159                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
1160
1161                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
1162                 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
1163         }
1164
1165         if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1166                 SW_FLOW_KEY_PUT(match, phy.priority,
1167                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1168                 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1169         }
1170
1171         if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1172                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1173
1174                 if (is_mask) {
1175                         in_port = 0xffffffff; /* Always exact match in_port. */
1176                 } else if (in_port >= DP_MAX_PORTS) {
1177                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
1178                                   in_port, DP_MAX_PORTS);
1179                         return -EINVAL;
1180                 }
1181
1182                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1183                 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1184         } else if (!is_mask) {
1185                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1186         }
1187
1188         if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1189                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1190
1191                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1192                 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1193         }
1194         if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1195                 if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1196                                        is_mask, log) < 0)
1197                         return -EINVAL;
1198                 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1199         }
1200
1201         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
1202             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
1203                 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
1204
1205                 if (ct_state & ~CT_SUPPORTED_MASK) {
1206                         OVS_NLERR(log, "ct_state flags %08x unsupported",
1207                                   ct_state);
1208                         return -EINVAL;
1209                 }
1210
1211                 SW_FLOW_KEY_PUT(match, ct_state, ct_state, is_mask);
1212                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
1213         }
1214         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
1215             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
1216                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
1217
1218                 SW_FLOW_KEY_PUT(match, ct_zone, ct_zone, is_mask);
1219                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
1220         }
1221         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
1222             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
1223                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
1224
1225                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
1226                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
1227         }
1228         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
1229             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
1230                 const struct ovs_key_ct_labels *cl;
1231
1232                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
1233                 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
1234                                    sizeof(*cl), is_mask);
1235                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
1236         }
1237         if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)) {
1238                 const struct ovs_key_ct_tuple_ipv4 *ct;
1239
1240                 ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4]);
1241
1242                 SW_FLOW_KEY_PUT(match, ipv4.ct_orig.src, ct->ipv4_src, is_mask);
1243                 SW_FLOW_KEY_PUT(match, ipv4.ct_orig.dst, ct->ipv4_dst, is_mask);
1244                 SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1245                 SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1246                 SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv4_proto, is_mask);
1247                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4);
1248         }
1249         if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)) {
1250                 const struct ovs_key_ct_tuple_ipv6 *ct;
1251
1252                 ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6]);
1253
1254                 SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.src, &ct->ipv6_src,
1255                                    sizeof(match->key->ipv6.ct_orig.src),
1256                                    is_mask);
1257                 SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.dst, &ct->ipv6_dst,
1258                                    sizeof(match->key->ipv6.ct_orig.dst),
1259                                    is_mask);
1260                 SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1261                 SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1262                 SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv6_proto, is_mask);
1263                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
1264         }
1265
1266         /* For layer 3 packets the Ethernet type is provided
1267          * and treated as metadata but no MAC addresses are provided.
1268          */
1269         if (!(*attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1270             (*attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)))
1271                 mac_proto = MAC_PROTO_NONE;
1272
1273         /* Always exact match mac_proto */
1274         SW_FLOW_KEY_PUT(match, mac_proto, is_mask ? 0xff : mac_proto, is_mask);
1275
1276         if (mac_proto == MAC_PROTO_NONE)
1277                 return parse_eth_type_from_nlattrs(match, attrs, a, is_mask,
1278                                                    log);
1279
1280         return 0;
1281 }
1282
1283 int nsh_hdr_from_nlattr(const struct nlattr *attr,
1284                         struct nshhdr *nh, size_t size)
1285 {
1286         struct nlattr *a;
1287         int rem;
1288         u8 flags = 0;
1289         u8 ttl = 0;
1290         int mdlen = 0;
1291
1292         /* validate_nsh has check this, so we needn't do duplicate check here
1293          */
1294         if (size < NSH_BASE_HDR_LEN)
1295                 return -ENOBUFS;
1296
1297         nla_for_each_nested(a, attr, rem) {
1298                 int type = nla_type(a);
1299
1300                 switch (type) {
1301                 case OVS_NSH_KEY_ATTR_BASE: {
1302                         const struct ovs_nsh_key_base *base = nla_data(a);
1303
1304                         flags = base->flags;
1305                         ttl = base->ttl;
1306                         nh->np = base->np;
1307                         nh->mdtype = base->mdtype;
1308                         nh->path_hdr = base->path_hdr;
1309                         break;
1310                 }
1311                 case OVS_NSH_KEY_ATTR_MD1:
1312                         mdlen = nla_len(a);
1313                         if (mdlen > size - NSH_BASE_HDR_LEN)
1314                                 return -ENOBUFS;
1315                         memcpy(&nh->md1, nla_data(a), mdlen);
1316                         break;
1317
1318                 case OVS_NSH_KEY_ATTR_MD2:
1319                         mdlen = nla_len(a);
1320                         if (mdlen > size - NSH_BASE_HDR_LEN)
1321                                 return -ENOBUFS;
1322                         memcpy(&nh->md2, nla_data(a), mdlen);
1323                         break;
1324
1325                 default:
1326                         return -EINVAL;
1327                 }
1328         }
1329
1330         /* nsh header length  = NSH_BASE_HDR_LEN + mdlen */
1331         nh->ver_flags_ttl_len = 0;
1332         nsh_set_flags_ttl_len(nh, flags, ttl, NSH_BASE_HDR_LEN + mdlen);
1333
1334         return 0;
1335 }
1336
1337 int nsh_key_from_nlattr(const struct nlattr *attr,
1338                         struct ovs_key_nsh *nsh, struct ovs_key_nsh *nsh_mask)
1339 {
1340         struct nlattr *a;
1341         int rem;
1342
1343         /* validate_nsh has check this, so we needn't do duplicate check here
1344          */
1345         nla_for_each_nested(a, attr, rem) {
1346                 int type = nla_type(a);
1347
1348                 switch (type) {
1349                 case OVS_NSH_KEY_ATTR_BASE: {
1350                         const struct ovs_nsh_key_base *base = nla_data(a);
1351                         const struct ovs_nsh_key_base *base_mask = base + 1;
1352
1353                         nsh->base = *base;
1354                         nsh_mask->base = *base_mask;
1355                         break;
1356                 }
1357                 case OVS_NSH_KEY_ATTR_MD1: {
1358                         const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1359                         const struct ovs_nsh_key_md1 *md1_mask = md1 + 1;
1360
1361                         memcpy(nsh->context, md1->context, sizeof(*md1));
1362                         memcpy(nsh_mask->context, md1_mask->context,
1363                                sizeof(*md1_mask));
1364                         break;
1365                 }
1366                 case OVS_NSH_KEY_ATTR_MD2:
1367                         /* Not supported yet */
1368                         return -ENOTSUPP;
1369                 default:
1370                         return -EINVAL;
1371                 }
1372         }
1373
1374         return 0;
1375 }
1376
1377 static int nsh_key_put_from_nlattr(const struct nlattr *attr,
1378                                    struct sw_flow_match *match, bool is_mask,
1379                                    bool is_push_nsh, bool log)
1380 {
1381         struct nlattr *a;
1382         int rem;
1383         bool has_base = false;
1384         bool has_md1 = false;
1385         bool has_md2 = false;
1386         u8 mdtype = 0;
1387         int mdlen = 0;
1388
1389         if (WARN_ON(is_push_nsh && is_mask))
1390                 return -EINVAL;
1391
1392         nla_for_each_nested(a, attr, rem) {
1393                 int type = nla_type(a);
1394                 int i;
1395
1396                 if (type > OVS_NSH_KEY_ATTR_MAX) {
1397                         OVS_NLERR(log, "nsh attr %d is out of range max %d",
1398                                   type, OVS_NSH_KEY_ATTR_MAX);
1399                         return -EINVAL;
1400                 }
1401
1402                 if (!check_attr_len(nla_len(a),
1403                                     ovs_nsh_key_attr_lens[type].len)) {
1404                         OVS_NLERR(
1405                             log,
1406                             "nsh attr %d has unexpected len %d expected %d",
1407                             type,
1408                             nla_len(a),
1409                             ovs_nsh_key_attr_lens[type].len
1410                         );
1411                         return -EINVAL;
1412                 }
1413
1414                 switch (type) {
1415                 case OVS_NSH_KEY_ATTR_BASE: {
1416                         const struct ovs_nsh_key_base *base = nla_data(a);
1417
1418                         has_base = true;
1419                         mdtype = base->mdtype;
1420                         SW_FLOW_KEY_PUT(match, nsh.base.flags,
1421                                         base->flags, is_mask);
1422                         SW_FLOW_KEY_PUT(match, nsh.base.ttl,
1423                                         base->ttl, is_mask);
1424                         SW_FLOW_KEY_PUT(match, nsh.base.mdtype,
1425                                         base->mdtype, is_mask);
1426                         SW_FLOW_KEY_PUT(match, nsh.base.np,
1427                                         base->np, is_mask);
1428                         SW_FLOW_KEY_PUT(match, nsh.base.path_hdr,
1429                                         base->path_hdr, is_mask);
1430                         break;
1431                 }
1432                 case OVS_NSH_KEY_ATTR_MD1: {
1433                         const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1434
1435                         has_md1 = true;
1436                         for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++)
1437                                 SW_FLOW_KEY_PUT(match, nsh.context[i],
1438                                                 md1->context[i], is_mask);
1439                         break;
1440                 }
1441                 case OVS_NSH_KEY_ATTR_MD2:
1442                         if (!is_push_nsh) /* Not supported MD type 2 yet */
1443                                 return -ENOTSUPP;
1444
1445                         has_md2 = true;
1446                         mdlen = nla_len(a);
1447                         if (mdlen > NSH_CTX_HDRS_MAX_LEN || mdlen <= 0) {
1448                                 OVS_NLERR(
1449                                     log,
1450                                     "Invalid MD length %d for MD type %d",
1451                                     mdlen,
1452                                     mdtype
1453                                 );
1454                                 return -EINVAL;
1455                         }
1456                         break;
1457                 default:
1458                         OVS_NLERR(log, "Unknown nsh attribute %d",
1459                                   type);
1460                         return -EINVAL;
1461                 }
1462         }
1463
1464         if (rem > 0) {
1465                 OVS_NLERR(log, "nsh attribute has %d unknown bytes.", rem);
1466                 return -EINVAL;
1467         }
1468
1469         if (has_md1 && has_md2) {
1470                 OVS_NLERR(
1471                     1,
1472                     "invalid nsh attribute: md1 and md2 are exclusive."
1473                 );
1474                 return -EINVAL;
1475         }
1476
1477         if (!is_mask) {
1478                 if ((has_md1 && mdtype != NSH_M_TYPE1) ||
1479                     (has_md2 && mdtype != NSH_M_TYPE2)) {
1480                         OVS_NLERR(1, "nsh attribute has unmatched MD type %d.",
1481                                   mdtype);
1482                         return -EINVAL;
1483                 }
1484
1485                 if (is_push_nsh &&
1486                     (!has_base || (!has_md1 && !has_md2))) {
1487                         OVS_NLERR(
1488                             1,
1489                             "push_nsh: missing base or metadata attributes"
1490                         );
1491                         return -EINVAL;
1492                 }
1493         }
1494
1495         return 0;
1496 }
1497
1498 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
1499                                 u64 attrs, const struct nlattr **a,
1500                                 bool is_mask, bool log)
1501 {
1502         int err;
1503
1504         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
1505         if (err)
1506                 return err;
1507
1508         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1509                 const struct ovs_key_ethernet *eth_key;
1510
1511                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1512                 SW_FLOW_KEY_MEMCPY(match, eth.src,
1513                                 eth_key->eth_src, ETH_ALEN, is_mask);
1514                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1515                                 eth_key->eth_dst, ETH_ALEN, is_mask);
1516                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1517
1518                 if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1519                         /* VLAN attribute is always parsed before getting here since it
1520                          * may occur multiple times.
1521                          */
1522                         OVS_NLERR(log, "VLAN attribute unexpected.");
1523                         return -EINVAL;
1524                 }
1525
1526                 if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1527                         err = parse_eth_type_from_nlattrs(match, &attrs, a, is_mask,
1528                                                           log);
1529                         if (err)
1530                                 return err;
1531                 } else if (!is_mask) {
1532                         SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1533                 }
1534         } else if (!match->key->eth.type) {
1535                 OVS_NLERR(log, "Either Ethernet header or EtherType is required.");
1536                 return -EINVAL;
1537         }
1538
1539         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1540                 const struct ovs_key_ipv4 *ipv4_key;
1541
1542                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1543                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1544                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
1545                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1546                         return -EINVAL;
1547                 }
1548                 SW_FLOW_KEY_PUT(match, ip.proto,
1549                                 ipv4_key->ipv4_proto, is_mask);
1550                 SW_FLOW_KEY_PUT(match, ip.tos,
1551                                 ipv4_key->ipv4_tos, is_mask);
1552                 SW_FLOW_KEY_PUT(match, ip.ttl,
1553                                 ipv4_key->ipv4_ttl, is_mask);
1554                 SW_FLOW_KEY_PUT(match, ip.frag,
1555                                 ipv4_key->ipv4_frag, is_mask);
1556                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1557                                 ipv4_key->ipv4_src, is_mask);
1558                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1559                                 ipv4_key->ipv4_dst, is_mask);
1560                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1561         }
1562
1563         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1564                 const struct ovs_key_ipv6 *ipv6_key;
1565
1566                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1567                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1568                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
1569                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1570                         return -EINVAL;
1571                 }
1572
1573                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
1574                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x)",
1575                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1576                         return -EINVAL;
1577                 }
1578
1579                 SW_FLOW_KEY_PUT(match, ipv6.label,
1580                                 ipv6_key->ipv6_label, is_mask);
1581                 SW_FLOW_KEY_PUT(match, ip.proto,
1582                                 ipv6_key->ipv6_proto, is_mask);
1583                 SW_FLOW_KEY_PUT(match, ip.tos,
1584                                 ipv6_key->ipv6_tclass, is_mask);
1585                 SW_FLOW_KEY_PUT(match, ip.ttl,
1586                                 ipv6_key->ipv6_hlimit, is_mask);
1587                 SW_FLOW_KEY_PUT(match, ip.frag,
1588                                 ipv6_key->ipv6_frag, is_mask);
1589                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1590                                 ipv6_key->ipv6_src,
1591                                 sizeof(match->key->ipv6.addr.src),
1592                                 is_mask);
1593                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1594                                 ipv6_key->ipv6_dst,
1595                                 sizeof(match->key->ipv6.addr.dst),
1596                                 is_mask);
1597
1598                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1599         }
1600
1601         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1602                 const struct ovs_key_arp *arp_key;
1603
1604                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1605                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1606                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1607                                   arp_key->arp_op);
1608                         return -EINVAL;
1609                 }
1610
1611                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1612                                 arp_key->arp_sip, is_mask);
1613                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1614                         arp_key->arp_tip, is_mask);
1615                 SW_FLOW_KEY_PUT(match, ip.proto,
1616                                 ntohs(arp_key->arp_op), is_mask);
1617                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1618                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1619                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1620                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1621
1622                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1623         }
1624
1625         if (attrs & (1 << OVS_KEY_ATTR_NSH)) {
1626                 if (nsh_key_put_from_nlattr(a[OVS_KEY_ATTR_NSH], match,
1627                                             is_mask, false, log) < 0)
1628                         return -EINVAL;
1629                 attrs &= ~(1 << OVS_KEY_ATTR_NSH);
1630         }
1631
1632         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1633                 const struct ovs_key_mpls *mpls_key;
1634                 u32 hdr_len;
1635                 u32 label_count, label_count_mask, i;
1636
1637                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1638                 hdr_len = nla_len(a[OVS_KEY_ATTR_MPLS]);
1639                 label_count = hdr_len / sizeof(struct ovs_key_mpls);
1640
1641                 if (label_count == 0 || label_count > MPLS_LABEL_DEPTH ||
1642                     hdr_len % sizeof(struct ovs_key_mpls))
1643                         return -EINVAL;
1644
1645                 label_count_mask =  GENMASK(label_count - 1, 0);
1646
1647                 for (i = 0 ; i < label_count; i++)
1648                         SW_FLOW_KEY_PUT(match, mpls.lse[i],
1649                                         mpls_key[i].mpls_lse, is_mask);
1650
1651                 SW_FLOW_KEY_PUT(match, mpls.num_labels_mask,
1652                                 label_count_mask, is_mask);
1653
1654                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1655          }
1656
1657         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1658                 const struct ovs_key_tcp *tcp_key;
1659
1660                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1661                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1662                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1663                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1664         }
1665
1666         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1667                 SW_FLOW_KEY_PUT(match, tp.flags,
1668                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1669                                 is_mask);
1670                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1671         }
1672
1673         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1674                 const struct ovs_key_udp *udp_key;
1675
1676                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1677                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1678                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1679                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1680         }
1681
1682         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1683                 const struct ovs_key_sctp *sctp_key;
1684
1685                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1686                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1687                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1688                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1689         }
1690
1691         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1692                 const struct ovs_key_icmp *icmp_key;
1693
1694                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1695                 SW_FLOW_KEY_PUT(match, tp.src,
1696                                 htons(icmp_key->icmp_type), is_mask);
1697                 SW_FLOW_KEY_PUT(match, tp.dst,
1698                                 htons(icmp_key->icmp_code), is_mask);
1699                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1700         }
1701
1702         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1703                 const struct ovs_key_icmpv6 *icmpv6_key;
1704
1705                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1706                 SW_FLOW_KEY_PUT(match, tp.src,
1707                                 htons(icmpv6_key->icmpv6_type), is_mask);
1708                 SW_FLOW_KEY_PUT(match, tp.dst,
1709                                 htons(icmpv6_key->icmpv6_code), is_mask);
1710                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1711         }
1712
1713         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1714                 const struct ovs_key_nd *nd_key;
1715
1716                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1717                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1718                         nd_key->nd_target,
1719                         sizeof(match->key->ipv6.nd.target),
1720                         is_mask);
1721                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1722                         nd_key->nd_sll, ETH_ALEN, is_mask);
1723                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1724                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1725                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1726         }
1727
1728         if (attrs != 0) {
1729                 OVS_NLERR(log, "Unknown key attributes %llx",
1730                           (unsigned long long)attrs);
1731                 return -EINVAL;
1732         }
1733
1734         return 0;
1735 }
1736
1737 static void nlattr_set(struct nlattr *attr, u8 val,
1738                        const struct ovs_len_tbl *tbl)
1739 {
1740         struct nlattr *nla;
1741         int rem;
1742
1743         /* The nlattr stream should already have been validated */
1744         nla_for_each_nested(nla, attr, rem) {
1745                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1746                         nlattr_set(nla, val, tbl[nla_type(nla)].next ? : tbl);
1747                 else
1748                         memset(nla_data(nla), val, nla_len(nla));
1749
1750                 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1751                         *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1752         }
1753 }
1754
1755 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1756 {
1757         nlattr_set(attr, val, ovs_key_lens);
1758 }
1759
1760 /**
1761  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1762  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1763  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1764  * does not include any don't care bit.
1765  * @net: Used to determine per-namespace field support.
1766  * @match: receives the extracted flow match information.
1767  * @nla_key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1768  * sequence. The fields should of the packet that triggered the creation
1769  * of this flow.
1770  * @nla_mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_*
1771  * Netlink attribute specifies the mask field of the wildcarded flow.
1772  * @log: Boolean to allow kernel error logging.  Normally true, but when
1773  * probing for feature compatibility this should be passed in as false to
1774  * suppress unnecessary error logging.
1775  */
1776 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1777                       const struct nlattr *nla_key,
1778                       const struct nlattr *nla_mask,
1779                       bool log)
1780 {
1781         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1782         struct nlattr *newmask = NULL;
1783         u64 key_attrs = 0;
1784         u64 mask_attrs = 0;
1785         int err;
1786
1787         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1788         if (err)
1789                 return err;
1790
1791         err = parse_vlan_from_nlattrs(match, &key_attrs, a, false, log);
1792         if (err)
1793                 return err;
1794
1795         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1796         if (err)
1797                 return err;
1798
1799         if (match->mask) {
1800                 if (!nla_mask) {
1801                         /* Create an exact match mask. We need to set to 0xff
1802                          * all the 'match->mask' fields that have been touched
1803                          * in 'match->key'. We cannot simply memset
1804                          * 'match->mask', because padding bytes and fields not
1805                          * specified in 'match->key' should be left to 0.
1806                          * Instead, we use a stream of netlink attributes,
1807                          * copied from 'key' and set to 0xff.
1808                          * ovs_key_from_nlattrs() will take care of filling
1809                          * 'match->mask' appropriately.
1810                          */
1811                         newmask = kmemdup(nla_key,
1812                                           nla_total_size(nla_len(nla_key)),
1813                                           GFP_KERNEL);
1814                         if (!newmask)
1815                                 return -ENOMEM;
1816
1817                         mask_set_nlattr(newmask, 0xff);
1818
1819                         /* The userspace does not send tunnel attributes that
1820                          * are 0, but we should not wildcard them nonetheless.
1821                          */
1822                         if (match->key->tun_proto)
1823                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1824                                                          0xff, true);
1825
1826                         nla_mask = newmask;
1827                 }
1828
1829                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1830                 if (err)
1831                         goto free_newmask;
1832
1833                 /* Always match on tci. */
1834                 SW_FLOW_KEY_PUT(match, eth.vlan.tci, htons(0xffff), true);
1835                 SW_FLOW_KEY_PUT(match, eth.cvlan.tci, htons(0xffff), true);
1836
1837                 err = parse_vlan_from_nlattrs(match, &mask_attrs, a, true, log);
1838                 if (err)
1839                         goto free_newmask;
1840
1841                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1842                                            log);
1843                 if (err)
1844                         goto free_newmask;
1845         }
1846
1847         if (!match_validate(match, key_attrs, mask_attrs, log))
1848                 err = -EINVAL;
1849
1850 free_newmask:
1851         kfree(newmask);
1852         return err;
1853 }
1854
1855 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1856 {
1857         size_t len;
1858
1859         if (!attr)
1860                 return 0;
1861
1862         len = nla_len(attr);
1863         if (len < 1 || len > MAX_UFID_LENGTH) {
1864                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1865                           nla_len(attr), MAX_UFID_LENGTH);
1866                 return 0;
1867         }
1868
1869         return len;
1870 }
1871
1872 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1873  * or false otherwise.
1874  */
1875 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1876                       bool log)
1877 {
1878         sfid->ufid_len = get_ufid_len(attr, log);
1879         if (sfid->ufid_len)
1880                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1881
1882         return sfid->ufid_len;
1883 }
1884
1885 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1886                            const struct sw_flow_key *key, bool log)
1887 {
1888         struct sw_flow_key *new_key;
1889
1890         if (ovs_nla_get_ufid(sfid, ufid, log))
1891                 return 0;
1892
1893         /* If UFID was not provided, use unmasked key. */
1894         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1895         if (!new_key)
1896                 return -ENOMEM;
1897         memcpy(new_key, key, sizeof(*key));
1898         sfid->unmasked_key = new_key;
1899
1900         return 0;
1901 }
1902
1903 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1904 {
1905         return attr ? nla_get_u32(attr) : 0;
1906 }
1907
1908 /**
1909  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1910  * @net: Network namespace.
1911  * @key: Receives extracted in_port, priority, tun_key, skb_mark and conntrack
1912  * metadata.
1913  * @a: Array of netlink attributes holding parsed %OVS_KEY_ATTR_* Netlink
1914  * attributes.
1915  * @attrs: Bit mask for the netlink attributes included in @a.
1916  * @log: Boolean to allow kernel error logging.  Normally true, but when
1917  * probing for feature compatibility this should be passed in as false to
1918  * suppress unnecessary error logging.
1919  *
1920  * This parses a series of Netlink attributes that form a flow key, which must
1921  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1922  * get the metadata, that is, the parts of the flow key that cannot be
1923  * extracted from the packet itself.
1924  *
1925  * This must be called before the packet key fields are filled in 'key'.
1926  */
1927
1928 int ovs_nla_get_flow_metadata(struct net *net,
1929                               const struct nlattr *a[OVS_KEY_ATTR_MAX + 1],
1930                               u64 attrs, struct sw_flow_key *key, bool log)
1931 {
1932         struct sw_flow_match match;
1933
1934         memset(&match, 0, sizeof(match));
1935         match.key = key;
1936
1937         key->ct_state = 0;
1938         key->ct_zone = 0;
1939         key->ct_orig_proto = 0;
1940         memset(&key->ct, 0, sizeof(key->ct));
1941         memset(&key->ipv4.ct_orig, 0, sizeof(key->ipv4.ct_orig));
1942         memset(&key->ipv6.ct_orig, 0, sizeof(key->ipv6.ct_orig));
1943
1944         key->phy.in_port = DP_MAX_PORTS;
1945
1946         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1947 }
1948
1949 static int ovs_nla_put_vlan(struct sk_buff *skb, const struct vlan_head *vh,
1950                             bool is_mask)
1951 {
1952         __be16 eth_type = !is_mask ? vh->tpid : htons(0xffff);
1953
1954         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1955             nla_put_be16(skb, OVS_KEY_ATTR_VLAN, vh->tci))
1956                 return -EMSGSIZE;
1957         return 0;
1958 }
1959
1960 static int nsh_key_to_nlattr(const struct ovs_key_nsh *nsh, bool is_mask,
1961                              struct sk_buff *skb)
1962 {
1963         struct nlattr *start;
1964
1965         start = nla_nest_start_noflag(skb, OVS_KEY_ATTR_NSH);
1966         if (!start)
1967                 return -EMSGSIZE;
1968
1969         if (nla_put(skb, OVS_NSH_KEY_ATTR_BASE, sizeof(nsh->base), &nsh->base))
1970                 goto nla_put_failure;
1971
1972         if (is_mask || nsh->base.mdtype == NSH_M_TYPE1) {
1973                 if (nla_put(skb, OVS_NSH_KEY_ATTR_MD1,
1974                             sizeof(nsh->context), nsh->context))
1975                         goto nla_put_failure;
1976         }
1977
1978         /* Don't support MD type 2 yet */
1979
1980         nla_nest_end(skb, start);
1981
1982         return 0;
1983
1984 nla_put_failure:
1985         return -EMSGSIZE;
1986 }
1987
1988 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1989                              const struct sw_flow_key *output, bool is_mask,
1990                              struct sk_buff *skb)
1991 {
1992         struct ovs_key_ethernet *eth_key;
1993         struct nlattr *nla;
1994         struct nlattr *encap = NULL;
1995         struct nlattr *in_encap = NULL;
1996
1997         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1998                 goto nla_put_failure;
1999
2000         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
2001                 goto nla_put_failure;
2002
2003         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
2004                 goto nla_put_failure;
2005
2006         if ((swkey->tun_proto || is_mask)) {
2007                 const void *opts = NULL;
2008
2009                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
2010                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
2011
2012                 if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
2013                                      swkey->tun_opts_len, swkey->tun_proto, 0))
2014                         goto nla_put_failure;
2015         }
2016
2017         if (swkey->phy.in_port == DP_MAX_PORTS) {
2018                 if (is_mask && (output->phy.in_port == 0xffff))
2019                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
2020                                 goto nla_put_failure;
2021         } else {
2022                 u16 upper_u16;
2023                 upper_u16 = !is_mask ? 0 : 0xffff;
2024
2025                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
2026                                 (upper_u16 << 16) | output->phy.in_port))
2027                         goto nla_put_failure;
2028         }
2029
2030         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
2031                 goto nla_put_failure;
2032
2033         if (ovs_ct_put_key(swkey, output, skb))
2034                 goto nla_put_failure;
2035
2036         if (ovs_key_mac_proto(swkey) == MAC_PROTO_ETHERNET) {
2037                 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
2038                 if (!nla)
2039                         goto nla_put_failure;
2040
2041                 eth_key = nla_data(nla);
2042                 ether_addr_copy(eth_key->eth_src, output->eth.src);
2043                 ether_addr_copy(eth_key->eth_dst, output->eth.dst);
2044
2045                 if (swkey->eth.vlan.tci || eth_type_vlan(swkey->eth.type)) {
2046                         if (ovs_nla_put_vlan(skb, &output->eth.vlan, is_mask))
2047                                 goto nla_put_failure;
2048                         encap = nla_nest_start_noflag(skb, OVS_KEY_ATTR_ENCAP);
2049                         if (!swkey->eth.vlan.tci)
2050                                 goto unencap;
2051
2052                         if (swkey->eth.cvlan.tci || eth_type_vlan(swkey->eth.type)) {
2053                                 if (ovs_nla_put_vlan(skb, &output->eth.cvlan, is_mask))
2054                                         goto nla_put_failure;
2055                                 in_encap = nla_nest_start_noflag(skb,
2056                                                                  OVS_KEY_ATTR_ENCAP);
2057                                 if (!swkey->eth.cvlan.tci)
2058                                         goto unencap;
2059                         }
2060                 }
2061
2062                 if (swkey->eth.type == htons(ETH_P_802_2)) {
2063                         /*
2064                         * Ethertype 802.2 is represented in the netlink with omitted
2065                         * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
2066                         * 0xffff in the mask attribute.  Ethertype can also
2067                         * be wildcarded.
2068                         */
2069                         if (is_mask && output->eth.type)
2070                                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
2071                                                         output->eth.type))
2072                                         goto nla_put_failure;
2073                         goto unencap;
2074                 }
2075         }
2076
2077         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
2078                 goto nla_put_failure;
2079
2080         if (eth_type_vlan(swkey->eth.type)) {
2081                 /* There are 3 VLAN tags, we don't know anything about the rest
2082                  * of the packet, so truncate here.
2083                  */
2084                 WARN_ON_ONCE(!(encap && in_encap));
2085                 goto unencap;
2086         }
2087
2088         if (swkey->eth.type == htons(ETH_P_IP)) {
2089                 struct ovs_key_ipv4 *ipv4_key;
2090
2091                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
2092                 if (!nla)
2093                         goto nla_put_failure;
2094                 ipv4_key = nla_data(nla);
2095                 ipv4_key->ipv4_src = output->ipv4.addr.src;
2096                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
2097                 ipv4_key->ipv4_proto = output->ip.proto;
2098                 ipv4_key->ipv4_tos = output->ip.tos;
2099                 ipv4_key->ipv4_ttl = output->ip.ttl;
2100                 ipv4_key->ipv4_frag = output->ip.frag;
2101         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
2102                 struct ovs_key_ipv6 *ipv6_key;
2103
2104                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
2105                 if (!nla)
2106                         goto nla_put_failure;
2107                 ipv6_key = nla_data(nla);
2108                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
2109                                 sizeof(ipv6_key->ipv6_src));
2110                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
2111                                 sizeof(ipv6_key->ipv6_dst));
2112                 ipv6_key->ipv6_label = output->ipv6.label;
2113                 ipv6_key->ipv6_proto = output->ip.proto;
2114                 ipv6_key->ipv6_tclass = output->ip.tos;
2115                 ipv6_key->ipv6_hlimit = output->ip.ttl;
2116                 ipv6_key->ipv6_frag = output->ip.frag;
2117         } else if (swkey->eth.type == htons(ETH_P_NSH)) {
2118                 if (nsh_key_to_nlattr(&output->nsh, is_mask, skb))
2119                         goto nla_put_failure;
2120         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
2121                    swkey->eth.type == htons(ETH_P_RARP)) {
2122                 struct ovs_key_arp *arp_key;
2123
2124                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
2125                 if (!nla)
2126                         goto nla_put_failure;
2127                 arp_key = nla_data(nla);
2128                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
2129                 arp_key->arp_sip = output->ipv4.addr.src;
2130                 arp_key->arp_tip = output->ipv4.addr.dst;
2131                 arp_key->arp_op = htons(output->ip.proto);
2132                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
2133                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
2134         } else if (eth_p_mpls(swkey->eth.type)) {
2135                 u8 i, num_labels;
2136                 struct ovs_key_mpls *mpls_key;
2137
2138                 num_labels = hweight_long(output->mpls.num_labels_mask);
2139                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS,
2140                                   num_labels * sizeof(*mpls_key));
2141                 if (!nla)
2142                         goto nla_put_failure;
2143
2144                 mpls_key = nla_data(nla);
2145                 for (i = 0; i < num_labels; i++)
2146                         mpls_key[i].mpls_lse = output->mpls.lse[i];
2147         }
2148
2149         if ((swkey->eth.type == htons(ETH_P_IP) ||
2150              swkey->eth.type == htons(ETH_P_IPV6)) &&
2151              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
2152
2153                 if (swkey->ip.proto == IPPROTO_TCP) {
2154                         struct ovs_key_tcp *tcp_key;
2155
2156                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
2157                         if (!nla)
2158                                 goto nla_put_failure;
2159                         tcp_key = nla_data(nla);
2160                         tcp_key->tcp_src = output->tp.src;
2161                         tcp_key->tcp_dst = output->tp.dst;
2162                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
2163                                          output->tp.flags))
2164                                 goto nla_put_failure;
2165                 } else if (swkey->ip.proto == IPPROTO_UDP) {
2166                         struct ovs_key_udp *udp_key;
2167
2168                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
2169                         if (!nla)
2170                                 goto nla_put_failure;
2171                         udp_key = nla_data(nla);
2172                         udp_key->udp_src = output->tp.src;
2173                         udp_key->udp_dst = output->tp.dst;
2174                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
2175                         struct ovs_key_sctp *sctp_key;
2176
2177                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
2178                         if (!nla)
2179                                 goto nla_put_failure;
2180                         sctp_key = nla_data(nla);
2181                         sctp_key->sctp_src = output->tp.src;
2182                         sctp_key->sctp_dst = output->tp.dst;
2183                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
2184                            swkey->ip.proto == IPPROTO_ICMP) {
2185                         struct ovs_key_icmp *icmp_key;
2186
2187                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
2188                         if (!nla)
2189                                 goto nla_put_failure;
2190                         icmp_key = nla_data(nla);
2191                         icmp_key->icmp_type = ntohs(output->tp.src);
2192                         icmp_key->icmp_code = ntohs(output->tp.dst);
2193                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
2194                            swkey->ip.proto == IPPROTO_ICMPV6) {
2195                         struct ovs_key_icmpv6 *icmpv6_key;
2196
2197                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
2198                                                 sizeof(*icmpv6_key));
2199                         if (!nla)
2200                                 goto nla_put_failure;
2201                         icmpv6_key = nla_data(nla);
2202                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
2203                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
2204
2205                         if (swkey->tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
2206                             swkey->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
2207                                 struct ovs_key_nd *nd_key;
2208
2209                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
2210                                 if (!nla)
2211                                         goto nla_put_failure;
2212                                 nd_key = nla_data(nla);
2213                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
2214                                                         sizeof(nd_key->nd_target));
2215                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
2216                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
2217                         }
2218                 }
2219         }
2220
2221 unencap:
2222         if (in_encap)
2223                 nla_nest_end(skb, in_encap);
2224         if (encap)
2225                 nla_nest_end(skb, encap);
2226
2227         return 0;
2228
2229 nla_put_failure:
2230         return -EMSGSIZE;
2231 }
2232
2233 int ovs_nla_put_key(const struct sw_flow_key *swkey,
2234                     const struct sw_flow_key *output, int attr, bool is_mask,
2235                     struct sk_buff *skb)
2236 {
2237         int err;
2238         struct nlattr *nla;
2239
2240         nla = nla_nest_start_noflag(skb, attr);
2241         if (!nla)
2242                 return -EMSGSIZE;
2243         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
2244         if (err)
2245                 return err;
2246         nla_nest_end(skb, nla);
2247
2248         return 0;
2249 }
2250
2251 /* Called with ovs_mutex or RCU read lock. */
2252 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
2253 {
2254         if (ovs_identifier_is_ufid(&flow->id))
2255                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
2256                                flow->id.ufid);
2257
2258         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
2259                                OVS_FLOW_ATTR_KEY, false, skb);
2260 }
2261
2262 /* Called with ovs_mutex or RCU read lock. */
2263 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
2264 {
2265         return ovs_nla_put_key(&flow->key, &flow->key,
2266                                 OVS_FLOW_ATTR_KEY, false, skb);
2267 }
2268
2269 /* Called with ovs_mutex or RCU read lock. */
2270 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
2271 {
2272         return ovs_nla_put_key(&flow->key, &flow->mask->key,
2273                                 OVS_FLOW_ATTR_MASK, true, skb);
2274 }
2275
2276 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
2277
2278 static struct sw_flow_actions *nla_alloc_flow_actions(int size)
2279 {
2280         struct sw_flow_actions *sfa;
2281
2282         WARN_ON_ONCE(size > MAX_ACTIONS_BUFSIZE);
2283
2284         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
2285         if (!sfa)
2286                 return ERR_PTR(-ENOMEM);
2287
2288         sfa->actions_len = 0;
2289         return sfa;
2290 }
2291
2292 static void ovs_nla_free_nested_actions(const struct nlattr *actions, int len);
2293
2294 static void ovs_nla_free_check_pkt_len_action(const struct nlattr *action)
2295 {
2296         const struct nlattr *a;
2297         int rem;
2298
2299         nla_for_each_nested(a, action, rem) {
2300                 switch (nla_type(a)) {
2301                 case OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL:
2302                 case OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER:
2303                         ovs_nla_free_nested_actions(nla_data(a), nla_len(a));
2304                         break;
2305                 }
2306         }
2307 }
2308
2309 static void ovs_nla_free_clone_action(const struct nlattr *action)
2310 {
2311         const struct nlattr *a = nla_data(action);
2312         int rem = nla_len(action);
2313
2314         switch (nla_type(a)) {
2315         case OVS_CLONE_ATTR_EXEC:
2316                 /* The real list of actions follows this attribute. */
2317                 a = nla_next(a, &rem);
2318                 ovs_nla_free_nested_actions(a, rem);
2319                 break;
2320         }
2321 }
2322
2323 static void ovs_nla_free_dec_ttl_action(const struct nlattr *action)
2324 {
2325         const struct nlattr *a = nla_data(action);
2326
2327         switch (nla_type(a)) {
2328         case OVS_DEC_TTL_ATTR_ACTION:
2329                 ovs_nla_free_nested_actions(nla_data(a), nla_len(a));
2330                 break;
2331         }
2332 }
2333
2334 static void ovs_nla_free_sample_action(const struct nlattr *action)
2335 {
2336         const struct nlattr *a = nla_data(action);
2337         int rem = nla_len(action);
2338
2339         switch (nla_type(a)) {
2340         case OVS_SAMPLE_ATTR_ARG:
2341                 /* The real list of actions follows this attribute. */
2342                 a = nla_next(a, &rem);
2343                 ovs_nla_free_nested_actions(a, rem);
2344                 break;
2345         }
2346 }
2347
2348 static void ovs_nla_free_set_action(const struct nlattr *a)
2349 {
2350         const struct nlattr *ovs_key = nla_data(a);
2351         struct ovs_tunnel_info *ovs_tun;
2352
2353         switch (nla_type(ovs_key)) {
2354         case OVS_KEY_ATTR_TUNNEL_INFO:
2355                 ovs_tun = nla_data(ovs_key);
2356                 dst_release((struct dst_entry *)ovs_tun->tun_dst);
2357                 break;
2358         }
2359 }
2360
2361 static void ovs_nla_free_nested_actions(const struct nlattr *actions, int len)
2362 {
2363         const struct nlattr *a;
2364         int rem;
2365
2366         /* Whenever new actions are added, the need to update this
2367          * function should be considered.
2368          */
2369         BUILD_BUG_ON(OVS_ACTION_ATTR_MAX != 23);
2370
2371         if (!actions)
2372                 return;
2373
2374         nla_for_each_attr(a, actions, len, rem) {
2375                 switch (nla_type(a)) {
2376                 case OVS_ACTION_ATTR_CHECK_PKT_LEN:
2377                         ovs_nla_free_check_pkt_len_action(a);
2378                         break;
2379
2380                 case OVS_ACTION_ATTR_CLONE:
2381                         ovs_nla_free_clone_action(a);
2382                         break;
2383
2384                 case OVS_ACTION_ATTR_CT:
2385                         ovs_ct_free_action(a);
2386                         break;
2387
2388                 case OVS_ACTION_ATTR_DEC_TTL:
2389                         ovs_nla_free_dec_ttl_action(a);
2390                         break;
2391
2392                 case OVS_ACTION_ATTR_SAMPLE:
2393                         ovs_nla_free_sample_action(a);
2394                         break;
2395
2396                 case OVS_ACTION_ATTR_SET:
2397                         ovs_nla_free_set_action(a);
2398                         break;
2399                 }
2400         }
2401 }
2402
2403 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
2404 {
2405         if (!sf_acts)
2406                 return;
2407
2408         ovs_nla_free_nested_actions(sf_acts->actions, sf_acts->actions_len);
2409         kfree(sf_acts);
2410 }
2411
2412 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
2413 {
2414         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
2415 }
2416
2417 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
2418  * The caller must hold rcu_read_lock for this to be sensible. */
2419 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
2420 {
2421         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
2422 }
2423
2424 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
2425                                        int attr_len, bool log)
2426 {
2427
2428         struct sw_flow_actions *acts;
2429         int new_acts_size;
2430         size_t req_size = NLA_ALIGN(attr_len);
2431         int next_offset = offsetof(struct sw_flow_actions, actions) +
2432                                         (*sfa)->actions_len;
2433
2434         if (req_size <= (ksize(*sfa) - next_offset))
2435                 goto out;
2436
2437         new_acts_size = max(next_offset + req_size, ksize(*sfa) * 2);
2438
2439         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
2440                 if ((next_offset + req_size) > MAX_ACTIONS_BUFSIZE) {
2441                         OVS_NLERR(log, "Flow action size exceeds max %u",
2442                                   MAX_ACTIONS_BUFSIZE);
2443                         return ERR_PTR(-EMSGSIZE);
2444                 }
2445                 new_acts_size = MAX_ACTIONS_BUFSIZE;
2446         }
2447
2448         acts = nla_alloc_flow_actions(new_acts_size);
2449         if (IS_ERR(acts))
2450                 return (void *)acts;
2451
2452         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
2453         acts->actions_len = (*sfa)->actions_len;
2454         acts->orig_len = (*sfa)->orig_len;
2455         kfree(*sfa);
2456         *sfa = acts;
2457
2458 out:
2459         (*sfa)->actions_len += req_size;
2460         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
2461 }
2462
2463 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
2464                                    int attrtype, void *data, int len, bool log)
2465 {
2466         struct nlattr *a;
2467
2468         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
2469         if (IS_ERR(a))
2470                 return a;
2471
2472         a->nla_type = attrtype;
2473         a->nla_len = nla_attr_size(len);
2474
2475         if (data)
2476                 memcpy(nla_data(a), data, len);
2477         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
2478
2479         return a;
2480 }
2481
2482 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
2483                        int len, bool log)
2484 {
2485         struct nlattr *a;
2486
2487         a = __add_action(sfa, attrtype, data, len, log);
2488
2489         return PTR_ERR_OR_ZERO(a);
2490 }
2491
2492 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
2493                                           int attrtype, bool log)
2494 {
2495         int used = (*sfa)->actions_len;
2496         int err;
2497
2498         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
2499         if (err)
2500                 return err;
2501
2502         return used;
2503 }
2504
2505 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
2506                                          int st_offset)
2507 {
2508         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
2509                                                                st_offset);
2510
2511         a->nla_len = sfa->actions_len - st_offset;
2512 }
2513
2514 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2515                                   const struct sw_flow_key *key,
2516                                   struct sw_flow_actions **sfa,
2517                                   __be16 eth_type, __be16 vlan_tci,
2518                                   u32 mpls_label_count, bool log,
2519                                   u32 depth);
2520
2521 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
2522                                     const struct sw_flow_key *key,
2523                                     struct sw_flow_actions **sfa,
2524                                     __be16 eth_type, __be16 vlan_tci,
2525                                     u32 mpls_label_count, bool log, bool last,
2526                                     u32 depth)
2527 {
2528         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
2529         const struct nlattr *probability, *actions;
2530         const struct nlattr *a;
2531         int rem, start, err;
2532         struct sample_arg arg;
2533
2534         memset(attrs, 0, sizeof(attrs));
2535         nla_for_each_nested(a, attr, rem) {
2536                 int type = nla_type(a);
2537                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
2538                         return -EINVAL;
2539                 attrs[type] = a;
2540         }
2541         if (rem)
2542                 return -EINVAL;
2543
2544         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
2545         if (!probability || nla_len(probability) != sizeof(u32))
2546                 return -EINVAL;
2547
2548         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
2549         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
2550                 return -EINVAL;
2551
2552         /* validation done, copy sample action. */
2553         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
2554         if (start < 0)
2555                 return start;
2556
2557         /* When both skb and flow may be changed, put the sample
2558          * into a deferred fifo. On the other hand, if only skb
2559          * may be modified, the actions can be executed in place.
2560          *
2561          * Do this analysis at the flow installation time.
2562          * Set 'clone_action->exec' to true if the actions can be
2563          * executed without being deferred.
2564          *
2565          * If the sample is the last action, it can always be excuted
2566          * rather than deferred.
2567          */
2568         arg.exec = last || !actions_may_change_flow(actions);
2569         arg.probability = nla_get_u32(probability);
2570
2571         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_ARG, &arg, sizeof(arg),
2572                                  log);
2573         if (err)
2574                 return err;
2575
2576         err = __ovs_nla_copy_actions(net, actions, key, sfa,
2577                                      eth_type, vlan_tci, mpls_label_count, log,
2578                                      depth + 1);
2579
2580         if (err)
2581                 return err;
2582
2583         add_nested_action_end(*sfa, start);
2584
2585         return 0;
2586 }
2587
2588 static int validate_and_copy_dec_ttl(struct net *net,
2589                                      const struct nlattr *attr,
2590                                      const struct sw_flow_key *key,
2591                                      struct sw_flow_actions **sfa,
2592                                      __be16 eth_type, __be16 vlan_tci,
2593                                      u32 mpls_label_count, bool log,
2594                                      u32 depth)
2595 {
2596         const struct nlattr *attrs[OVS_DEC_TTL_ATTR_MAX + 1];
2597         int start, action_start, err, rem;
2598         const struct nlattr *a, *actions;
2599
2600         memset(attrs, 0, sizeof(attrs));
2601         nla_for_each_nested(a, attr, rem) {
2602                 int type = nla_type(a);
2603
2604                 /* Ignore unknown attributes to be future proof. */
2605                 if (type > OVS_DEC_TTL_ATTR_MAX)
2606                         continue;
2607
2608                 if (!type || attrs[type])
2609                         return -EINVAL;
2610
2611                 attrs[type] = a;
2612         }
2613
2614         actions = attrs[OVS_DEC_TTL_ATTR_ACTION];
2615         if (rem || !actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
2616                 return -EINVAL;
2617
2618         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_DEC_TTL, log);
2619         if (start < 0)
2620                 return start;
2621
2622         action_start = add_nested_action_start(sfa, OVS_DEC_TTL_ATTR_ACTION, log);
2623         if (action_start < 0)
2624                 return action_start;
2625
2626         err = __ovs_nla_copy_actions(net, actions, key, sfa, eth_type,
2627                                      vlan_tci, mpls_label_count, log,
2628                                      depth + 1);
2629         if (err)
2630                 return err;
2631
2632         add_nested_action_end(*sfa, action_start);
2633         add_nested_action_end(*sfa, start);
2634         return 0;
2635 }
2636
2637 static int validate_and_copy_clone(struct net *net,
2638                                    const struct nlattr *attr,
2639                                    const struct sw_flow_key *key,
2640                                    struct sw_flow_actions **sfa,
2641                                    __be16 eth_type, __be16 vlan_tci,
2642                                    u32 mpls_label_count, bool log, bool last,
2643                                    u32 depth)
2644 {
2645         int start, err;
2646         u32 exec;
2647
2648         if (nla_len(attr) && nla_len(attr) < NLA_HDRLEN)
2649                 return -EINVAL;
2650
2651         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CLONE, log);
2652         if (start < 0)
2653                 return start;
2654
2655         exec = last || !actions_may_change_flow(attr);
2656
2657         err = ovs_nla_add_action(sfa, OVS_CLONE_ATTR_EXEC, &exec,
2658                                  sizeof(exec), log);
2659         if (err)
2660                 return err;
2661
2662         err = __ovs_nla_copy_actions(net, attr, key, sfa,
2663                                      eth_type, vlan_tci, mpls_label_count, log,
2664                                      depth + 1);
2665         if (err)
2666                 return err;
2667
2668         add_nested_action_end(*sfa, start);
2669
2670         return 0;
2671 }
2672
2673 void ovs_match_init(struct sw_flow_match *match,
2674                     struct sw_flow_key *key,
2675                     bool reset_key,
2676                     struct sw_flow_mask *mask)
2677 {
2678         memset(match, 0, sizeof(*match));
2679         match->key = key;
2680         match->mask = mask;
2681
2682         if (reset_key)
2683                 memset(key, 0, sizeof(*key));
2684
2685         if (mask) {
2686                 memset(&mask->key, 0, sizeof(mask->key));
2687                 mask->range.start = mask->range.end = 0;
2688         }
2689 }
2690
2691 static int validate_geneve_opts(struct sw_flow_key *key)
2692 {
2693         struct geneve_opt *option;
2694         int opts_len = key->tun_opts_len;
2695         bool crit_opt = false;
2696
2697         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
2698         while (opts_len > 0) {
2699                 int len;
2700
2701                 if (opts_len < sizeof(*option))
2702                         return -EINVAL;
2703
2704                 len = sizeof(*option) + option->length * 4;
2705                 if (len > opts_len)
2706                         return -EINVAL;
2707
2708                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
2709
2710                 option = (struct geneve_opt *)((u8 *)option + len);
2711                 opts_len -= len;
2712         }
2713
2714         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
2715
2716         return 0;
2717 }
2718
2719 static int validate_and_copy_set_tun(const struct nlattr *attr,
2720                                      struct sw_flow_actions **sfa, bool log)
2721 {
2722         struct sw_flow_match match;
2723         struct sw_flow_key key;
2724         struct metadata_dst *tun_dst;
2725         struct ip_tunnel_info *tun_info;
2726         struct ovs_tunnel_info *ovs_tun;
2727         struct nlattr *a;
2728         int err = 0, start, opts_type;
2729         __be16 dst_opt_type;
2730
2731         dst_opt_type = 0;
2732         ovs_match_init(&match, &key, true, NULL);
2733         opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
2734         if (opts_type < 0)
2735                 return opts_type;
2736
2737         if (key.tun_opts_len) {
2738                 switch (opts_type) {
2739                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2740                         err = validate_geneve_opts(&key);
2741                         if (err < 0)
2742                                 return err;
2743                         dst_opt_type = TUNNEL_GENEVE_OPT;
2744                         break;
2745                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2746                         dst_opt_type = TUNNEL_VXLAN_OPT;
2747                         break;
2748                 case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
2749                         dst_opt_type = TUNNEL_ERSPAN_OPT;
2750                         break;
2751                 }
2752         }
2753
2754         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
2755         if (start < 0)
2756                 return start;
2757
2758         tun_dst = metadata_dst_alloc(key.tun_opts_len, METADATA_IP_TUNNEL,
2759                                      GFP_KERNEL);
2760
2761         if (!tun_dst)
2762                 return -ENOMEM;
2763
2764         err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
2765         if (err) {
2766                 dst_release((struct dst_entry *)tun_dst);
2767                 return err;
2768         }
2769
2770         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
2771                          sizeof(*ovs_tun), log);
2772         if (IS_ERR(a)) {
2773                 dst_release((struct dst_entry *)tun_dst);
2774                 return PTR_ERR(a);
2775         }
2776
2777         ovs_tun = nla_data(a);
2778         ovs_tun->tun_dst = tun_dst;
2779
2780         tun_info = &tun_dst->u.tun_info;
2781         tun_info->mode = IP_TUNNEL_INFO_TX;
2782         if (key.tun_proto == AF_INET6)
2783                 tun_info->mode |= IP_TUNNEL_INFO_IPV6;
2784         else if (key.tun_proto == AF_INET && key.tun_key.u.ipv4.dst == 0)
2785                 tun_info->mode |= IP_TUNNEL_INFO_BRIDGE;
2786         tun_info->key = key.tun_key;
2787
2788         /* We need to store the options in the action itself since
2789          * everything else will go away after flow setup. We can append
2790          * it to tun_info and then point there.
2791          */
2792         ip_tunnel_info_opts_set(tun_info,
2793                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
2794                                 key.tun_opts_len, dst_opt_type);
2795         add_nested_action_end(*sfa, start);
2796
2797         return err;
2798 }
2799
2800 static bool validate_nsh(const struct nlattr *attr, bool is_mask,
2801                          bool is_push_nsh, bool log)
2802 {
2803         struct sw_flow_match match;
2804         struct sw_flow_key key;
2805         int ret = 0;
2806
2807         ovs_match_init(&match, &key, true, NULL);
2808         ret = nsh_key_put_from_nlattr(attr, &match, is_mask,
2809                                       is_push_nsh, log);
2810         return !ret;
2811 }
2812
2813 /* Return false if there are any non-masked bits set.
2814  * Mask follows data immediately, before any netlink padding.
2815  */
2816 static bool validate_masked(u8 *data, int len)
2817 {
2818         u8 *mask = data + len;
2819
2820         while (len--)
2821                 if (*data++ & ~*mask++)
2822                         return false;
2823
2824         return true;
2825 }
2826
2827 static int validate_set(const struct nlattr *a,
2828                         const struct sw_flow_key *flow_key,
2829                         struct sw_flow_actions **sfa, bool *skip_copy,
2830                         u8 mac_proto, __be16 eth_type, bool masked, bool log)
2831 {
2832         const struct nlattr *ovs_key = nla_data(a);
2833         int key_type = nla_type(ovs_key);
2834         size_t key_len;
2835
2836         /* There can be only one key in a action */
2837         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2838                 return -EINVAL;
2839
2840         key_len = nla_len(ovs_key);
2841         if (masked)
2842                 key_len /= 2;
2843
2844         if (key_type > OVS_KEY_ATTR_MAX ||
2845             !check_attr_len(key_len, ovs_key_lens[key_type].len))
2846                 return -EINVAL;
2847
2848         if (masked && !validate_masked(nla_data(ovs_key), key_len))
2849                 return -EINVAL;
2850
2851         switch (key_type) {
2852         case OVS_KEY_ATTR_PRIORITY:
2853         case OVS_KEY_ATTR_SKB_MARK:
2854         case OVS_KEY_ATTR_CT_MARK:
2855         case OVS_KEY_ATTR_CT_LABELS:
2856                 break;
2857
2858         case OVS_KEY_ATTR_ETHERNET:
2859                 if (mac_proto != MAC_PROTO_ETHERNET)
2860                         return -EINVAL;
2861                 break;
2862
2863         case OVS_KEY_ATTR_TUNNEL: {
2864                 int err;
2865
2866                 if (masked)
2867                         return -EINVAL; /* Masked tunnel set not supported. */
2868
2869                 *skip_copy = true;
2870                 err = validate_and_copy_set_tun(a, sfa, log);
2871                 if (err)
2872                         return err;
2873                 break;
2874         }
2875         case OVS_KEY_ATTR_IPV4: {
2876                 const struct ovs_key_ipv4 *ipv4_key;
2877
2878                 if (eth_type != htons(ETH_P_IP))
2879                         return -EINVAL;
2880
2881                 ipv4_key = nla_data(ovs_key);
2882
2883                 if (masked) {
2884                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2885
2886                         /* Non-writeable fields. */
2887                         if (mask->ipv4_proto || mask->ipv4_frag)
2888                                 return -EINVAL;
2889                 } else {
2890                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2891                                 return -EINVAL;
2892
2893                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2894                                 return -EINVAL;
2895                 }
2896                 break;
2897         }
2898         case OVS_KEY_ATTR_IPV6: {
2899                 const struct ovs_key_ipv6 *ipv6_key;
2900
2901                 if (eth_type != htons(ETH_P_IPV6))
2902                         return -EINVAL;
2903
2904                 ipv6_key = nla_data(ovs_key);
2905
2906                 if (masked) {
2907                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2908
2909                         /* Non-writeable fields. */
2910                         if (mask->ipv6_proto || mask->ipv6_frag)
2911                                 return -EINVAL;
2912
2913                         /* Invalid bits in the flow label mask? */
2914                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2915                                 return -EINVAL;
2916                 } else {
2917                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2918                                 return -EINVAL;
2919
2920                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2921                                 return -EINVAL;
2922                 }
2923                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2924                         return -EINVAL;
2925
2926                 break;
2927         }
2928         case OVS_KEY_ATTR_TCP:
2929                 if ((eth_type != htons(ETH_P_IP) &&
2930                      eth_type != htons(ETH_P_IPV6)) ||
2931                     flow_key->ip.proto != IPPROTO_TCP)
2932                         return -EINVAL;
2933
2934                 break;
2935
2936         case OVS_KEY_ATTR_UDP:
2937                 if ((eth_type != htons(ETH_P_IP) &&
2938                      eth_type != htons(ETH_P_IPV6)) ||
2939                     flow_key->ip.proto != IPPROTO_UDP)
2940                         return -EINVAL;
2941
2942                 break;
2943
2944         case OVS_KEY_ATTR_MPLS:
2945                 if (!eth_p_mpls(eth_type))
2946                         return -EINVAL;
2947                 break;
2948
2949         case OVS_KEY_ATTR_SCTP:
2950                 if ((eth_type != htons(ETH_P_IP) &&
2951                      eth_type != htons(ETH_P_IPV6)) ||
2952                     flow_key->ip.proto != IPPROTO_SCTP)
2953                         return -EINVAL;
2954
2955                 break;
2956
2957         case OVS_KEY_ATTR_NSH:
2958                 if (eth_type != htons(ETH_P_NSH))
2959                         return -EINVAL;
2960                 if (!validate_nsh(nla_data(a), masked, false, log))
2961                         return -EINVAL;
2962                 break;
2963
2964         default:
2965                 return -EINVAL;
2966         }
2967
2968         /* Convert non-masked non-tunnel set actions to masked set actions. */
2969         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2970                 int start, len = key_len * 2;
2971                 struct nlattr *at;
2972
2973                 *skip_copy = true;
2974
2975                 start = add_nested_action_start(sfa,
2976                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2977                                                 log);
2978                 if (start < 0)
2979                         return start;
2980
2981                 at = __add_action(sfa, key_type, NULL, len, log);
2982                 if (IS_ERR(at))
2983                         return PTR_ERR(at);
2984
2985                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2986                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2987                 /* Clear non-writeable bits from otherwise writeable fields. */
2988                 if (key_type == OVS_KEY_ATTR_IPV6) {
2989                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2990
2991                         mask->ipv6_label &= htonl(0x000FFFFF);
2992                 }
2993                 add_nested_action_end(*sfa, start);
2994         }
2995
2996         return 0;
2997 }
2998
2999 static int validate_userspace(const struct nlattr *attr)
3000 {
3001         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
3002                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
3003                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
3004                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
3005         };
3006         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
3007         int error;
3008
3009         error = nla_parse_nested_deprecated(a, OVS_USERSPACE_ATTR_MAX, attr,
3010                                             userspace_policy, NULL);
3011         if (error)
3012                 return error;
3013
3014         if (!a[OVS_USERSPACE_ATTR_PID] ||
3015             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
3016                 return -EINVAL;
3017
3018         return 0;
3019 }
3020
3021 static const struct nla_policy cpl_policy[OVS_CHECK_PKT_LEN_ATTR_MAX + 1] = {
3022         [OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] = {.type = NLA_U16 },
3023         [OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER] = {.type = NLA_NESTED },
3024         [OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL] = {.type = NLA_NESTED },
3025 };
3026
3027 static int validate_and_copy_check_pkt_len(struct net *net,
3028                                            const struct nlattr *attr,
3029                                            const struct sw_flow_key *key,
3030                                            struct sw_flow_actions **sfa,
3031                                            __be16 eth_type, __be16 vlan_tci,
3032                                            u32 mpls_label_count,
3033                                            bool log, bool last, u32 depth)
3034 {
3035         const struct nlattr *acts_if_greater, *acts_if_lesser_eq;
3036         struct nlattr *a[OVS_CHECK_PKT_LEN_ATTR_MAX + 1];
3037         struct check_pkt_len_arg arg;
3038         int nested_acts_start;
3039         int start, err;
3040
3041         err = nla_parse_deprecated_strict(a, OVS_CHECK_PKT_LEN_ATTR_MAX,
3042                                           nla_data(attr), nla_len(attr),
3043                                           cpl_policy, NULL);
3044         if (err)
3045                 return err;
3046
3047         if (!a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] ||
3048             !nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]))
3049                 return -EINVAL;
3050
3051         acts_if_lesser_eq = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL];
3052         acts_if_greater = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER];
3053
3054         /* Both the nested action should be present. */
3055         if (!acts_if_greater || !acts_if_lesser_eq)
3056                 return -EINVAL;
3057
3058         /* validation done, copy the nested actions. */
3059         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CHECK_PKT_LEN,
3060                                         log);
3061         if (start < 0)
3062                 return start;
3063
3064         arg.pkt_len = nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]);
3065         arg.exec_for_lesser_equal =
3066                 last || !actions_may_change_flow(acts_if_lesser_eq);
3067         arg.exec_for_greater =
3068                 last || !actions_may_change_flow(acts_if_greater);
3069
3070         err = ovs_nla_add_action(sfa, OVS_CHECK_PKT_LEN_ATTR_ARG, &arg,
3071                                  sizeof(arg), log);
3072         if (err)
3073                 return err;
3074
3075         nested_acts_start = add_nested_action_start(sfa,
3076                 OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL, log);
3077         if (nested_acts_start < 0)
3078                 return nested_acts_start;
3079
3080         err = __ovs_nla_copy_actions(net, acts_if_lesser_eq, key, sfa,
3081                                      eth_type, vlan_tci, mpls_label_count, log,
3082                                      depth + 1);
3083
3084         if (err)
3085                 return err;
3086
3087         add_nested_action_end(*sfa, nested_acts_start);
3088
3089         nested_acts_start = add_nested_action_start(sfa,
3090                 OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER, log);
3091         if (nested_acts_start < 0)
3092                 return nested_acts_start;
3093
3094         err = __ovs_nla_copy_actions(net, acts_if_greater, key, sfa,
3095                                      eth_type, vlan_tci, mpls_label_count, log,
3096                                      depth + 1);
3097
3098         if (err)
3099                 return err;
3100
3101         add_nested_action_end(*sfa, nested_acts_start);
3102         add_nested_action_end(*sfa, start);
3103         return 0;
3104 }
3105
3106 static int copy_action(const struct nlattr *from,
3107                        struct sw_flow_actions **sfa, bool log)
3108 {
3109         int totlen = NLA_ALIGN(from->nla_len);
3110         struct nlattr *to;
3111
3112         to = reserve_sfa_size(sfa, from->nla_len, log);
3113         if (IS_ERR(to))
3114                 return PTR_ERR(to);
3115
3116         memcpy(to, from, totlen);
3117         return 0;
3118 }
3119
3120 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3121                                   const struct sw_flow_key *key,
3122                                   struct sw_flow_actions **sfa,
3123                                   __be16 eth_type, __be16 vlan_tci,
3124                                   u32 mpls_label_count, bool log,
3125                                   u32 depth)
3126 {
3127         u8 mac_proto = ovs_key_mac_proto(key);
3128         const struct nlattr *a;
3129         int rem, err;
3130
3131         if (depth > OVS_COPY_ACTIONS_MAX_DEPTH)
3132                 return -EOVERFLOW;
3133
3134         nla_for_each_nested(a, attr, rem) {
3135                 /* Expected argument lengths, (u32)-1 for variable length. */
3136                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
3137                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
3138                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
3139                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
3140                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
3141                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
3142                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
3143                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
3144                         [OVS_ACTION_ATTR_SET] = (u32)-1,
3145                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
3146                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
3147                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
3148                         [OVS_ACTION_ATTR_CT] = (u32)-1,
3149                         [OVS_ACTION_ATTR_CT_CLEAR] = 0,
3150                         [OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
3151                         [OVS_ACTION_ATTR_PUSH_ETH] = sizeof(struct ovs_action_push_eth),
3152                         [OVS_ACTION_ATTR_POP_ETH] = 0,
3153                         [OVS_ACTION_ATTR_PUSH_NSH] = (u32)-1,
3154                         [OVS_ACTION_ATTR_POP_NSH] = 0,
3155                         [OVS_ACTION_ATTR_METER] = sizeof(u32),
3156                         [OVS_ACTION_ATTR_CLONE] = (u32)-1,
3157                         [OVS_ACTION_ATTR_CHECK_PKT_LEN] = (u32)-1,
3158                         [OVS_ACTION_ATTR_ADD_MPLS] = sizeof(struct ovs_action_add_mpls),
3159                         [OVS_ACTION_ATTR_DEC_TTL] = (u32)-1,
3160                 };
3161                 const struct ovs_action_push_vlan *vlan;
3162                 int type = nla_type(a);
3163                 bool skip_copy;
3164
3165                 if (type > OVS_ACTION_ATTR_MAX ||
3166                     (action_lens[type] != nla_len(a) &&
3167                      action_lens[type] != (u32)-1))
3168                         return -EINVAL;
3169
3170                 skip_copy = false;
3171                 switch (type) {
3172                 case OVS_ACTION_ATTR_UNSPEC:
3173                         return -EINVAL;
3174
3175                 case OVS_ACTION_ATTR_USERSPACE:
3176                         err = validate_userspace(a);
3177                         if (err)
3178                                 return err;
3179                         break;
3180
3181                 case OVS_ACTION_ATTR_OUTPUT:
3182                         if (nla_get_u32(a) >= DP_MAX_PORTS)
3183                                 return -EINVAL;
3184                         break;
3185
3186                 case OVS_ACTION_ATTR_TRUNC: {
3187                         const struct ovs_action_trunc *trunc = nla_data(a);
3188
3189                         if (trunc->max_len < ETH_HLEN)
3190                                 return -EINVAL;
3191                         break;
3192                 }
3193
3194                 case OVS_ACTION_ATTR_HASH: {
3195                         const struct ovs_action_hash *act_hash = nla_data(a);
3196
3197                         switch (act_hash->hash_alg) {
3198                         case OVS_HASH_ALG_L4:
3199                                 break;
3200                         default:
3201                                 return  -EINVAL;
3202                         }
3203
3204                         break;
3205                 }
3206
3207                 case OVS_ACTION_ATTR_POP_VLAN:
3208                         if (mac_proto != MAC_PROTO_ETHERNET)
3209                                 return -EINVAL;
3210                         vlan_tci = htons(0);
3211                         break;
3212
3213                 case OVS_ACTION_ATTR_PUSH_VLAN:
3214                         if (mac_proto != MAC_PROTO_ETHERNET)
3215                                 return -EINVAL;
3216                         vlan = nla_data(a);
3217                         if (!eth_type_vlan(vlan->vlan_tpid))
3218                                 return -EINVAL;
3219                         if (!(vlan->vlan_tci & htons(VLAN_CFI_MASK)))
3220                                 return -EINVAL;
3221                         vlan_tci = vlan->vlan_tci;
3222                         break;
3223
3224                 case OVS_ACTION_ATTR_RECIRC:
3225                         break;
3226
3227                 case OVS_ACTION_ATTR_ADD_MPLS: {
3228                         const struct ovs_action_add_mpls *mpls = nla_data(a);
3229
3230                         if (!eth_p_mpls(mpls->mpls_ethertype))
3231                                 return -EINVAL;
3232
3233                         if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK) {
3234                                 if (vlan_tci & htons(VLAN_CFI_MASK) ||
3235                                     (eth_type != htons(ETH_P_IP) &&
3236                                      eth_type != htons(ETH_P_IPV6) &&
3237                                      eth_type != htons(ETH_P_ARP) &&
3238                                      eth_type != htons(ETH_P_RARP) &&
3239                                      !eth_p_mpls(eth_type)))
3240                                         return -EINVAL;
3241                                 mpls_label_count++;
3242                         } else {
3243                                 if (mac_proto == MAC_PROTO_ETHERNET) {
3244                                         mpls_label_count = 1;
3245                                         mac_proto = MAC_PROTO_NONE;
3246                                 } else {
3247                                         mpls_label_count++;
3248                                 }
3249                         }
3250                         eth_type = mpls->mpls_ethertype;
3251                         break;
3252                 }
3253
3254                 case OVS_ACTION_ATTR_PUSH_MPLS: {
3255                         const struct ovs_action_push_mpls *mpls = nla_data(a);
3256
3257                         if (!eth_p_mpls(mpls->mpls_ethertype))
3258                                 return -EINVAL;
3259                         /* Prohibit push MPLS other than to a white list
3260                          * for packets that have a known tag order.
3261                          */
3262                         if (vlan_tci & htons(VLAN_CFI_MASK) ||
3263                             (eth_type != htons(ETH_P_IP) &&
3264                              eth_type != htons(ETH_P_IPV6) &&
3265                              eth_type != htons(ETH_P_ARP) &&
3266                              eth_type != htons(ETH_P_RARP) &&
3267                              !eth_p_mpls(eth_type)))
3268                                 return -EINVAL;
3269                         eth_type = mpls->mpls_ethertype;
3270                         mpls_label_count++;
3271                         break;
3272                 }
3273
3274                 case OVS_ACTION_ATTR_POP_MPLS: {
3275                         __be16  proto;
3276                         if (vlan_tci & htons(VLAN_CFI_MASK) ||
3277                             !eth_p_mpls(eth_type))
3278                                 return -EINVAL;
3279
3280                         /* Disallow subsequent L2.5+ set actions and mpls_pop
3281                          * actions once the last MPLS label in the packet is
3282                          * is popped as there is no check here to ensure that
3283                          * the new eth type is valid and thus set actions could
3284                          * write off the end of the packet or otherwise corrupt
3285                          * it.
3286                          *
3287                          * Support for these actions is planned using packet
3288                          * recirculation.
3289                          */
3290                         proto = nla_get_be16(a);
3291
3292                         if (proto == htons(ETH_P_TEB) &&
3293                             mac_proto != MAC_PROTO_NONE)
3294                                 return -EINVAL;
3295
3296                         mpls_label_count--;
3297
3298                         if (!eth_p_mpls(proto) || !mpls_label_count)
3299                                 eth_type = htons(0);
3300                         else
3301                                 eth_type =  proto;
3302
3303                         break;
3304                 }
3305
3306                 case OVS_ACTION_ATTR_SET:
3307                         err = validate_set(a, key, sfa,
3308                                            &skip_copy, mac_proto, eth_type,
3309                                            false, log);
3310                         if (err)
3311                                 return err;
3312                         break;
3313
3314                 case OVS_ACTION_ATTR_SET_MASKED:
3315                         err = validate_set(a, key, sfa,
3316                                            &skip_copy, mac_proto, eth_type,
3317                                            true, log);
3318                         if (err)
3319                                 return err;
3320                         break;
3321
3322                 case OVS_ACTION_ATTR_SAMPLE: {
3323                         bool last = nla_is_last(a, rem);
3324
3325                         err = validate_and_copy_sample(net, a, key, sfa,
3326                                                        eth_type, vlan_tci,
3327                                                        mpls_label_count,
3328                                                        log, last, depth);
3329                         if (err)
3330                                 return err;
3331                         skip_copy = true;
3332                         break;
3333                 }
3334
3335                 case OVS_ACTION_ATTR_CT:
3336                         err = ovs_ct_copy_action(net, a, key, sfa, log);
3337                         if (err)
3338                                 return err;
3339                         skip_copy = true;
3340                         break;
3341
3342                 case OVS_ACTION_ATTR_CT_CLEAR:
3343                         break;
3344
3345                 case OVS_ACTION_ATTR_PUSH_ETH:
3346                         /* Disallow pushing an Ethernet header if one
3347                          * is already present */
3348                         if (mac_proto != MAC_PROTO_NONE)
3349                                 return -EINVAL;
3350                         mac_proto = MAC_PROTO_ETHERNET;
3351                         break;
3352
3353                 case OVS_ACTION_ATTR_POP_ETH:
3354                         if (mac_proto != MAC_PROTO_ETHERNET)
3355                                 return -EINVAL;
3356                         if (vlan_tci & htons(VLAN_CFI_MASK))
3357                                 return -EINVAL;
3358                         mac_proto = MAC_PROTO_NONE;
3359                         break;
3360
3361                 case OVS_ACTION_ATTR_PUSH_NSH:
3362                         if (mac_proto != MAC_PROTO_ETHERNET) {
3363                                 u8 next_proto;
3364
3365                                 next_proto = tun_p_from_eth_p(eth_type);
3366                                 if (!next_proto)
3367                                         return -EINVAL;
3368                         }
3369                         mac_proto = MAC_PROTO_NONE;
3370                         if (!validate_nsh(nla_data(a), false, true, true))
3371                                 return -EINVAL;
3372                         break;
3373
3374                 case OVS_ACTION_ATTR_POP_NSH: {
3375                         __be16 inner_proto;
3376
3377                         if (eth_type != htons(ETH_P_NSH))
3378                                 return -EINVAL;
3379                         inner_proto = tun_p_to_eth_p(key->nsh.base.np);
3380                         if (!inner_proto)
3381                                 return -EINVAL;
3382                         if (key->nsh.base.np == TUN_P_ETHERNET)
3383                                 mac_proto = MAC_PROTO_ETHERNET;
3384                         else
3385                                 mac_proto = MAC_PROTO_NONE;
3386                         break;
3387                 }
3388
3389                 case OVS_ACTION_ATTR_METER:
3390                         /* Non-existent meters are simply ignored.  */
3391                         break;
3392
3393                 case OVS_ACTION_ATTR_CLONE: {
3394                         bool last = nla_is_last(a, rem);
3395
3396                         err = validate_and_copy_clone(net, a, key, sfa,
3397                                                       eth_type, vlan_tci,
3398                                                       mpls_label_count,
3399                                                       log, last, depth);
3400                         if (err)
3401                                 return err;
3402                         skip_copy = true;
3403                         break;
3404                 }
3405
3406                 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
3407                         bool last = nla_is_last(a, rem);
3408
3409                         err = validate_and_copy_check_pkt_len(net, a, key, sfa,
3410                                                               eth_type,
3411                                                               vlan_tci,
3412                                                               mpls_label_count,
3413                                                               log, last,
3414                                                               depth);
3415                         if (err)
3416                                 return err;
3417                         skip_copy = true;
3418                         break;
3419                 }
3420
3421                 case OVS_ACTION_ATTR_DEC_TTL:
3422                         err = validate_and_copy_dec_ttl(net, a, key, sfa,
3423                                                         eth_type, vlan_tci,
3424                                                         mpls_label_count, log,
3425                                                         depth);
3426                         if (err)
3427                                 return err;
3428                         skip_copy = true;
3429                         break;
3430
3431                 default:
3432                         OVS_NLERR(log, "Unknown Action type %d", type);
3433                         return -EINVAL;
3434                 }
3435                 if (!skip_copy) {
3436                         err = copy_action(a, sfa, log);
3437                         if (err)
3438                                 return err;
3439                 }
3440         }
3441
3442         if (rem > 0)
3443                 return -EINVAL;
3444
3445         return 0;
3446 }
3447
3448 /* 'key' must be the masked key. */
3449 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3450                          const struct sw_flow_key *key,
3451                          struct sw_flow_actions **sfa, bool log)
3452 {
3453         int err;
3454         u32 mpls_label_count = 0;
3455
3456         *sfa = nla_alloc_flow_actions(min(nla_len(attr), MAX_ACTIONS_BUFSIZE));
3457         if (IS_ERR(*sfa))
3458                 return PTR_ERR(*sfa);
3459
3460         if (eth_p_mpls(key->eth.type))
3461                 mpls_label_count = hweight_long(key->mpls.num_labels_mask);
3462
3463         (*sfa)->orig_len = nla_len(attr);
3464         err = __ovs_nla_copy_actions(net, attr, key, sfa, key->eth.type,
3465                                      key->eth.vlan.tci, mpls_label_count, log,
3466                                      0);
3467         if (err)
3468                 ovs_nla_free_flow_actions(*sfa);
3469
3470         return err;
3471 }
3472
3473 static int sample_action_to_attr(const struct nlattr *attr,
3474                                  struct sk_buff *skb)
3475 {
3476         struct nlattr *start, *ac_start = NULL, *sample_arg;
3477         int err = 0, rem = nla_len(attr);
3478         const struct sample_arg *arg;
3479         struct nlattr *actions;
3480
3481         start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SAMPLE);
3482         if (!start)
3483                 return -EMSGSIZE;
3484
3485         sample_arg = nla_data(attr);
3486         arg = nla_data(sample_arg);
3487         actions = nla_next(sample_arg, &rem);
3488
3489         if (nla_put_u32(skb, OVS_SAMPLE_ATTR_PROBABILITY, arg->probability)) {
3490                 err = -EMSGSIZE;
3491                 goto out;
3492         }
3493
3494         ac_start = nla_nest_start_noflag(skb, OVS_SAMPLE_ATTR_ACTIONS);
3495         if (!ac_start) {
3496                 err = -EMSGSIZE;
3497                 goto out;
3498         }
3499
3500         err = ovs_nla_put_actions(actions, rem, skb);
3501
3502 out:
3503         if (err) {
3504                 nla_nest_cancel(skb, ac_start);
3505                 nla_nest_cancel(skb, start);
3506         } else {
3507                 nla_nest_end(skb, ac_start);
3508                 nla_nest_end(skb, start);
3509         }
3510
3511         return err;
3512 }
3513
3514 static int clone_action_to_attr(const struct nlattr *attr,
3515                                 struct sk_buff *skb)
3516 {
3517         struct nlattr *start;
3518         int err = 0, rem = nla_len(attr);
3519
3520         start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CLONE);
3521         if (!start)
3522                 return -EMSGSIZE;
3523
3524         /* Skipping the OVS_CLONE_ATTR_EXEC that is always the first attribute. */
3525         attr = nla_next(nla_data(attr), &rem);
3526         err = ovs_nla_put_actions(attr, rem, skb);
3527
3528         if (err)
3529                 nla_nest_cancel(skb, start);
3530         else
3531                 nla_nest_end(skb, start);
3532
3533         return err;
3534 }
3535
3536 static int check_pkt_len_action_to_attr(const struct nlattr *attr,
3537                                         struct sk_buff *skb)
3538 {
3539         struct nlattr *start, *ac_start = NULL;
3540         const struct check_pkt_len_arg *arg;
3541         const struct nlattr *a, *cpl_arg;
3542         int err = 0, rem = nla_len(attr);
3543
3544         start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CHECK_PKT_LEN);
3545         if (!start)
3546                 return -EMSGSIZE;
3547
3548         /* The first nested attribute in 'attr' is always
3549          * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
3550          */
3551         cpl_arg = nla_data(attr);
3552         arg = nla_data(cpl_arg);
3553
3554         if (nla_put_u16(skb, OVS_CHECK_PKT_LEN_ATTR_PKT_LEN, arg->pkt_len)) {
3555                 err = -EMSGSIZE;
3556                 goto out;
3557         }
3558
3559         /* Second nested attribute in 'attr' is always
3560          * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
3561          */
3562         a = nla_next(cpl_arg, &rem);
3563         ac_start =  nla_nest_start_noflag(skb,
3564                                           OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL);
3565         if (!ac_start) {
3566                 err = -EMSGSIZE;
3567                 goto out;
3568         }
3569
3570         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3571         if (err) {
3572                 nla_nest_cancel(skb, ac_start);
3573                 goto out;
3574         } else {
3575                 nla_nest_end(skb, ac_start);
3576         }
3577
3578         /* Third nested attribute in 'attr' is always
3579          * OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER.
3580          */
3581         a = nla_next(a, &rem);
3582         ac_start =  nla_nest_start_noflag(skb,
3583                                           OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER);
3584         if (!ac_start) {
3585                 err = -EMSGSIZE;
3586                 goto out;
3587         }
3588
3589         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3590         if (err) {
3591                 nla_nest_cancel(skb, ac_start);
3592                 goto out;
3593         } else {
3594                 nla_nest_end(skb, ac_start);
3595         }
3596
3597         nla_nest_end(skb, start);
3598         return 0;
3599
3600 out:
3601         nla_nest_cancel(skb, start);
3602         return err;
3603 }
3604
3605 static int dec_ttl_action_to_attr(const struct nlattr *attr,
3606                                   struct sk_buff *skb)
3607 {
3608         struct nlattr *start, *action_start;
3609         const struct nlattr *a;
3610         int err = 0, rem;
3611
3612         start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_DEC_TTL);
3613         if (!start)
3614                 return -EMSGSIZE;
3615
3616         nla_for_each_attr(a, nla_data(attr), nla_len(attr), rem) {
3617                 switch (nla_type(a)) {
3618                 case OVS_DEC_TTL_ATTR_ACTION:
3619
3620                         action_start = nla_nest_start_noflag(skb, OVS_DEC_TTL_ATTR_ACTION);
3621                         if (!action_start) {
3622                                 err = -EMSGSIZE;
3623                                 goto out;
3624                         }
3625
3626                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3627                         if (err)
3628                                 goto out;
3629
3630                         nla_nest_end(skb, action_start);
3631                         break;
3632
3633                 default:
3634                         /* Ignore all other option to be future compatible */
3635                         break;
3636                 }
3637         }
3638
3639         nla_nest_end(skb, start);
3640         return 0;
3641
3642 out:
3643         nla_nest_cancel(skb, start);
3644         return err;
3645 }
3646
3647 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
3648 {
3649         const struct nlattr *ovs_key = nla_data(a);
3650         int key_type = nla_type(ovs_key);
3651         struct nlattr *start;
3652         int err;
3653
3654         switch (key_type) {
3655         case OVS_KEY_ATTR_TUNNEL_INFO: {
3656                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
3657                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
3658
3659                 start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SET);
3660                 if (!start)
3661                         return -EMSGSIZE;
3662
3663                 err =  ip_tun_to_nlattr(skb, &tun_info->key,
3664                                         ip_tunnel_info_opts(tun_info),
3665                                         tun_info->options_len,
3666                                         ip_tunnel_info_af(tun_info), tun_info->mode);
3667                 if (err)
3668                         return err;
3669                 nla_nest_end(skb, start);
3670                 break;
3671         }
3672         default:
3673                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
3674                         return -EMSGSIZE;
3675                 break;
3676         }
3677
3678         return 0;
3679 }
3680
3681 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
3682                                                 struct sk_buff *skb)
3683 {
3684         const struct nlattr *ovs_key = nla_data(a);
3685         struct nlattr *nla;
3686         size_t key_len = nla_len(ovs_key) / 2;
3687
3688         /* Revert the conversion we did from a non-masked set action to
3689          * masked set action.
3690          */
3691         nla = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SET);
3692         if (!nla)
3693                 return -EMSGSIZE;
3694
3695         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
3696                 return -EMSGSIZE;
3697
3698         nla_nest_end(skb, nla);
3699         return 0;
3700 }
3701
3702 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
3703 {
3704         const struct nlattr *a;
3705         int rem, err;
3706
3707         nla_for_each_attr(a, attr, len, rem) {
3708                 int type = nla_type(a);
3709
3710                 switch (type) {
3711                 case OVS_ACTION_ATTR_SET:
3712                         err = set_action_to_attr(a, skb);
3713                         if (err)
3714                                 return err;
3715                         break;
3716
3717                 case OVS_ACTION_ATTR_SET_TO_MASKED:
3718                         err = masked_set_action_to_set_action_attr(a, skb);
3719                         if (err)
3720                                 return err;
3721                         break;
3722
3723                 case OVS_ACTION_ATTR_SAMPLE:
3724                         err = sample_action_to_attr(a, skb);
3725                         if (err)
3726                                 return err;
3727                         break;
3728
3729                 case OVS_ACTION_ATTR_CT:
3730                         err = ovs_ct_action_to_attr(nla_data(a), skb);
3731                         if (err)
3732                                 return err;
3733                         break;
3734
3735                 case OVS_ACTION_ATTR_CLONE:
3736                         err = clone_action_to_attr(a, skb);
3737                         if (err)
3738                                 return err;
3739                         break;
3740
3741                 case OVS_ACTION_ATTR_CHECK_PKT_LEN:
3742                         err = check_pkt_len_action_to_attr(a, skb);
3743                         if (err)
3744                                 return err;
3745                         break;
3746
3747                 case OVS_ACTION_ATTR_DEC_TTL:
3748                         err = dec_ttl_action_to_attr(a, skb);
3749                         if (err)
3750                                 return err;
3751                         break;
3752
3753                 default:
3754                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
3755                                 return -EMSGSIZE;
3756                         break;
3757                 }
3758         }
3759
3760         return 0;
3761 }