net/openvswitch/flow_netlink.c

   1 /*
   2  * Copyright (c) 2007-2017 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include "flow.h"
  22 #include "datapath.h"
  23 #include <linux/uaccess.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <net/llc_pdu.h>
  29 #include <linux/kernel.h>
  30 #include <linux/jhash.h>
  31 #include <linux/jiffies.h>
  32 #include <linux/llc.h>
  33 #include <linux/module.h>
  34 #include <linux/in.h>
  35 #include <linux/rcupdate.h>
  36 #include <linux/if_arp.h>
  37 #include <linux/ip.h>
  38 #include <linux/ipv6.h>
  39 #include <linux/sctp.h>
  40 #include <linux/tcp.h>
  41 #include <linux/udp.h>
  42 #include <linux/icmp.h>
  43 #include <linux/icmpv6.h>
  44 #include <linux/rculist.h>
  45 #include <net/geneve.h>
  46 #include <net/ip.h>
  47 #include <net/ipv6.h>
  48 #include <net/ndisc.h>
  49 #include <net/mpls.h>
  50 #include <net/vxlan.h>
  51 #include <net/tun_proto.h>
  52 #include <net/erspan.h>
  53
  54 #include "flow_netlink.h"
  55
  56 struct ovs_len_tbl {
  57         int len;
  58         const struct ovs_len_tbl *next;
  59 };
  60
  61 #define OVS_ATTR_NESTED -1
  62 #define OVS_ATTR_VARIABLE -2
  63
  64 static bool actions_may_change_flow(const struct nlattr *actions)
  65 {
  66         struct nlattr *nla;
  67         int rem;
  68
  69         nla_for_each_nested(nla, actions, rem) {
  70                 u16 action = nla_type(nla);
  71
  72                 switch (action) {
  73                 case OVS_ACTION_ATTR_OUTPUT:
  74                 case OVS_ACTION_ATTR_RECIRC:
  75                 case OVS_ACTION_ATTR_TRUNC:
  76                 case OVS_ACTION_ATTR_USERSPACE:
  77                         break;
  78
  79                 case OVS_ACTION_ATTR_CT:
  80                 case OVS_ACTION_ATTR_CT_CLEAR:
  81                 case OVS_ACTION_ATTR_HASH:
  82                 case OVS_ACTION_ATTR_POP_ETH:
  83                 case OVS_ACTION_ATTR_POP_MPLS:
  84                 case OVS_ACTION_ATTR_POP_NSH:
  85                 case OVS_ACTION_ATTR_POP_VLAN:
  86                 case OVS_ACTION_ATTR_PUSH_ETH:
  87                 case OVS_ACTION_ATTR_PUSH_MPLS:
  88                 case OVS_ACTION_ATTR_PUSH_NSH:
  89                 case OVS_ACTION_ATTR_PUSH_VLAN:
  90                 case OVS_ACTION_ATTR_SAMPLE:
  91                 case OVS_ACTION_ATTR_SET:
  92                 case OVS_ACTION_ATTR_SET_MASKED:
  93                 case OVS_ACTION_ATTR_METER:
  94                 case OVS_ACTION_ATTR_CHECK_PKT_LEN:
  95                 default:
  96                         return true;
  97                 }
  98         }
  99         return false;
 100 }
 101
 102 static void update_range(struct sw_flow_match *match,
 103                          size_t offset, size_t size, bool is_mask)
 104 {
 105         struct sw_flow_key_range *range;
 106         size_t start = rounddown(offset, sizeof(long));
 107         size_t end = roundup(offset + size, sizeof(long));
 108
 109         if (!is_mask)
 110                 range = &match->range;
 111         else
 112                 range = &match->mask->range;
 113
 114         if (range->start == range->end) {
 115                 range->start = start;
 116                 range->end = end;
 117                 return;
 118         }
 119
 120         if (range->start > start)
 121                 range->start = start;
 122
 123         if (range->end < end)
 124                 range->end = end;
 125 }
 126
 127 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
 128         do { \
 129                 update_range(match, offsetof(struct sw_flow_key, field),    \
 130                              sizeof((match)->key->field), is_mask);         \
 131                 if (is_mask)                                                \
 132                         (match)->mask->key.field = value;                   \
 133                 else                                                        \
 134                         (match)->key->field = value;                        \
 135         } while (0)
 136
 137 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
 138         do {                                                                \
 139                 update_range(match, offset, len, is_mask);                  \
 140                 if (is_mask)                                                \
 141                         memcpy((u8 *)&(match)->mask->key + offset, value_p, \
 142                                len);                                       \
 143                 else                                                        \
 144                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
 145         } while (0)
 146
 147 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
 148         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 149                                   value_p, len, is_mask)
 150
 151 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 152         do {                                                                \
 153                 update_range(match, offsetof(struct sw_flow_key, field),    \
 154                              sizeof((match)->key->field), is_mask);         \
 155                 if (is_mask)                                                \
 156                         memset((u8 *)&(match)->mask->key.field, value,      \
 157                                sizeof((match)->mask->key.field));           \
 158                 else                                                        \
 159                         memset((u8 *)&(match)->key->field, value,           \
 160                                sizeof((match)->key->field));                \
 161         } while (0)
 162
 163 static bool match_validate(const struct sw_flow_match *match,
 164                            u64 key_attrs, u64 mask_attrs, bool log)
 165 {
 166         u64 key_expected = 0;
 167         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 168
 169         /* The following mask attributes allowed only if they
 170          * pass the validation tests. */
 171         mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
 172                         | (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)
 173                         | (1 << OVS_KEY_ATTR_IPV6)
 174                         | (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)
 175                         | (1 << OVS_KEY_ATTR_TCP)
 176                         | (1 << OVS_KEY_ATTR_TCP_FLAGS)
 177                         | (1 << OVS_KEY_ATTR_UDP)
 178                         | (1 << OVS_KEY_ATTR_SCTP)
 179                         | (1 << OVS_KEY_ATTR_ICMP)
 180                         | (1 << OVS_KEY_ATTR_ICMPV6)
 181                         | (1 << OVS_KEY_ATTR_ARP)
 182                         | (1 << OVS_KEY_ATTR_ND)
 183                         | (1 << OVS_KEY_ATTR_MPLS)
 184                         | (1 << OVS_KEY_ATTR_NSH));
 185
 186         /* Always allowed mask fields. */
 187         mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
 188                        | (1 << OVS_KEY_ATTR_IN_PORT)
 189                        | (1 << OVS_KEY_ATTR_ETHERTYPE));
 190
 191         /* Check key attributes. */
 192         if (match->key->eth.type == htons(ETH_P_ARP)
 193                         || match->key->eth.type == htons(ETH_P_RARP)) {
 194                 key_expected |= 1 << OVS_KEY_ATTR_ARP;
 195                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 196                         mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
 197         }
 198
 199         if (eth_p_mpls(match->key->eth.type)) {
 200                 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
 201                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 202                         mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
 203         }
 204
 205         if (match->key->eth.type == htons(ETH_P_IP)) {
 206                 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
 207                 if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
 208                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
 209                         mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4;
 210                 }
 211
 212                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 213                         if (match->key->ip.proto == IPPROTO_UDP) {
 214                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 215                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 216                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 217                         }
 218
 219                         if (match->key->ip.proto == IPPROTO_SCTP) {
 220                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 221                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 222                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 223                         }
 224
 225                         if (match->key->ip.proto == IPPROTO_TCP) {
 226                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 227                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 228                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 229                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 230                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 231                                 }
 232                         }
 233
 234                         if (match->key->ip.proto == IPPROTO_ICMP) {
 235                                 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
 236                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 237                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
 238                         }
 239                 }
 240         }
 241
 242         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 243                 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
 244                 if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
 245                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
 246                         mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6;
 247                 }
 248
 249                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 250                         if (match->key->ip.proto == IPPROTO_UDP) {
 251                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 252                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 253                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 254                         }
 255
 256                         if (match->key->ip.proto == IPPROTO_SCTP) {
 257                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 258                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 259                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 260                         }
 261
 262                         if (match->key->ip.proto == IPPROTO_TCP) {
 263                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 264                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 265                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 266                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 267                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 268                                 }
 269                         }
 270
 271                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 272                                 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
 273                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 274                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
 275
 276                                 if (match->key->tp.src ==
 277                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 278                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 279                                         key_expected |= 1 << OVS_KEY_ATTR_ND;
 280                                         /* Original direction conntrack tuple
 281                                          * uses the same space as the ND fields
 282                                          * in the key, so both are not allowed
 283                                          * at the same time.
 284                                          */
 285                                         mask_allowed &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
 286                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 287                                                 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
 288                                 }
 289                         }
 290                 }
 291         }
 292
 293         if (match->key->eth.type == htons(ETH_P_NSH)) {
 294                 key_expected |= 1 << OVS_KEY_ATTR_NSH;
 295                 if (match->mask &&
 296                     match->mask->key.eth.type == htons(0xffff)) {
 297                         mask_allowed |= 1 << OVS_KEY_ATTR_NSH;
 298                 }
 299         }
 300
 301         if ((key_attrs & key_expected) != key_expected) {
 302                 /* Key attributes check failed. */
 303                 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
 304                           (unsigned long long)key_attrs,
 305                           (unsigned long long)key_expected);
 306                 return false;
 307         }
 308
 309         if ((mask_attrs & mask_allowed) != mask_attrs) {
 310                 /* Mask attributes check failed. */
 311                 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
 312                           (unsigned long long)mask_attrs,
 313                           (unsigned long long)mask_allowed);
 314                 return false;
 315         }
 316
 317         return true;
 318 }
 319
 320 size_t ovs_tun_key_attr_size(void)
 321 {
 322         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 323          * updating this function.
 324          */
 325         return    nla_total_size_64bit(8) /* OVS_TUNNEL_KEY_ATTR_ID */
 326                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
 327                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
 328                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
 329                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
 330                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 331                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
 332                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
 333                 + nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 334                 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS and
 335                  * OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS is mutually exclusive with
 336                  * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
 337                  */
 338                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
 339                 + nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
 340 }
 341
 342 static size_t ovs_nsh_key_attr_size(void)
 343 {
 344         /* Whenever adding new OVS_NSH_KEY_ FIELDS, we should consider
 345          * updating this function.
 346          */
 347         return  nla_total_size(NSH_BASE_HDR_LEN) /* OVS_NSH_KEY_ATTR_BASE */
 348                 /* OVS_NSH_KEY_ATTR_MD1 and OVS_NSH_KEY_ATTR_MD2 are
 349                  * mutually exclusive, so the bigger one can cover
 350                  * the small one.
 351                  */
 352                 + nla_total_size(NSH_CTX_HDRS_MAX_LEN);
 353 }
 354
 355 size_t ovs_key_attr_size(void)
 356 {
 357         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 358          * updating this function.
 359          */
 360         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 29);
 361
 362         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 363                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 364                   + ovs_tun_key_attr_size()
 365                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 366                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 367                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 368                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 369                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
 370                 + nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
 371                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
 372                 + nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
 373                 + nla_total_size(40)  /* OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6 */
 374                 + nla_total_size(0)   /* OVS_KEY_ATTR_NSH */
 375                   + ovs_nsh_key_attr_size()
 376                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 377                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 378                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 379                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 380                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 381                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 382                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 383                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 384 }
 385
 386 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
 387         [OVS_VXLAN_EXT_GBP]         = { .len = sizeof(u32) },
 388 };
 389
 390 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 391         [OVS_TUNNEL_KEY_ATTR_ID]            = { .len = sizeof(u64) },
 392         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC]      = { .len = sizeof(u32) },
 393         [OVS_TUNNEL_KEY_ATTR_IPV4_DST]      = { .len = sizeof(u32) },
 394         [OVS_TUNNEL_KEY_ATTR_TOS]           = { .len = 1 },
 395         [OVS_TUNNEL_KEY_ATTR_TTL]           = { .len = 1 },
 396         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
 397         [OVS_TUNNEL_KEY_ATTR_CSUM]          = { .len = 0 },
 398         [OVS_TUNNEL_KEY_ATTR_TP_SRC]        = { .len = sizeof(u16) },
 399         [OVS_TUNNEL_KEY_ATTR_TP_DST]        = { .len = sizeof(u16) },
 400         [OVS_TUNNEL_KEY_ATTR_OAM]           = { .len = 0 },
 401         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
 402         [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
 403                                                 .next = ovs_vxlan_ext_key_lens },
 404         [OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
 405         [OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
 406         [OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS]   = { .len = OVS_ATTR_VARIABLE },
 407         [OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE]   = { .len = 0 },
 408 };
 409
 410 static const struct ovs_len_tbl
 411 ovs_nsh_key_attr_lens[OVS_NSH_KEY_ATTR_MAX + 1] = {
 412         [OVS_NSH_KEY_ATTR_BASE] = { .len = sizeof(struct ovs_nsh_key_base) },
 413         [OVS_NSH_KEY_ATTR_MD1]  = { .len = sizeof(struct ovs_nsh_key_md1) },
 414         [OVS_NSH_KEY_ATTR_MD2]  = { .len = OVS_ATTR_VARIABLE },
 415 };
 416
 417 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 418 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 419         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 420         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 421         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 422         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 423         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 424         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 425         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 426         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 427         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 428         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 429         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 430         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 431         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 432         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 433         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 434         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 435         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 436         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 437         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 438         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 439                                      .next = ovs_tunnel_key_lens, },
 440         [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 441         [OVS_KEY_ATTR_CT_STATE]  = { .len = sizeof(u32) },
 442         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 443         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 444         [OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
 445         [OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4] = {
 446                 .len = sizeof(struct ovs_key_ct_tuple_ipv4) },
 447         [OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6] = {
 448                 .len = sizeof(struct ovs_key_ct_tuple_ipv6) },
 449         [OVS_KEY_ATTR_NSH]       = { .len = OVS_ATTR_NESTED,
 450                                      .next = ovs_nsh_key_attr_lens, },
 451 };
 452
 453 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
 454 {
 455         return expected_len == attr_len ||
 456                expected_len == OVS_ATTR_NESTED ||
 457                expected_len == OVS_ATTR_VARIABLE;
 458 }
 459
 460 static bool is_all_zero(const u8 *fp, size_t size)
 461 {
 462         int i;
 463
 464         if (!fp)
 465                 return false;
 466
 467         for (i = 0; i < size; i++)
 468                 if (fp[i])
 469                         return false;
 470
 471         return true;
 472 }
 473
 474 static int __parse_flow_nlattrs(const struct nlattr *attr,
 475                                 const struct nlattr *a[],
 476                                 u64 *attrsp, bool log, bool nz)
 477 {
 478         const struct nlattr *nla;
 479         u64 attrs;
 480         int rem;
 481
 482         attrs = *attrsp;
 483         nla_for_each_nested(nla, attr, rem) {
 484                 u16 type = nla_type(nla);
 485                 int expected_len;
 486
 487                 if (type > OVS_KEY_ATTR_MAX) {
 488                         OVS_NLERR(log, "Key type %d is out of range max %d",
 489                                   type, OVS_KEY_ATTR_MAX);
 490                         return -EINVAL;
 491                 }
 492
 493                 if (attrs & (1 << type)) {
 494                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 495                         return -EINVAL;
 496                 }
 497
 498                 expected_len = ovs_key_lens[type].len;
 499                 if (!check_attr_len(nla_len(nla), expected_len)) {
 500                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 501                                   type, nla_len(nla), expected_len);
 502                         return -EINVAL;
 503                 }
 504
 505                 if (!nz || !is_all_zero(nla_data(nla), nla_len(nla))) {
 506                         attrs |= 1 << type;
 507                         a[type] = nla;
 508                 }
 509         }
 510         if (rem) {
 511                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 512                 return -EINVAL;
 513         }
 514
 515         *attrsp = attrs;
 516         return 0;
 517 }
 518
 519 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 520                                    const struct nlattr *a[], u64 *attrsp,
 521                                    bool log)
 522 {
 523         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 524 }
 525
 526 int parse_flow_nlattrs(const struct nlattr *attr, const struct nlattr *a[],
 527                        u64 *attrsp, bool log)
 528 {
 529         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 530 }
 531
 532 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 533                                      struct sw_flow_match *match, bool is_mask,
 534                                      bool log)
 535 {
 536         unsigned long opt_key_offset;
 537
 538         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 539                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 540                           nla_len(a), sizeof(match->key->tun_opts));
 541                 return -EINVAL;
 542         }
 543
 544         if (nla_len(a) % 4 != 0) {
 545                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 546                           nla_len(a));
 547                 return -EINVAL;
 548         }
 549
 550         /* We need to record the length of the options passed
 551          * down, otherwise packets with the same format but
 552          * additional options will be silently matched.
 553          */
 554         if (!is_mask) {
 555                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 556                                 false);
 557         } else {
 558                 /* This is somewhat unusual because it looks at
 559                  * both the key and mask while parsing the
 560                  * attributes (and by extension assumes the key
 561                  * is parsed first). Normally, we would verify
 562                  * that each is the correct length and that the
 563                  * attributes line up in the validate function.
 564                  * However, that is difficult because this is
 565                  * variable length and we won't have the
 566                  * information later.
 567                  */
 568                 if (match->key->tun_opts_len != nla_len(a)) {
 569                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 570                                   match->key->tun_opts_len, nla_len(a));
 571                         return -EINVAL;
 572                 }
 573
 574                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 575         }
 576
 577         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 578         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 579                                   nla_len(a), is_mask);
 580         return 0;
 581 }
 582
 583 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
 584                                      struct sw_flow_match *match, bool is_mask,
 585                                      bool log)
 586 {
 587         struct nlattr *a;
 588         int rem;
 589         unsigned long opt_key_offset;
 590         struct vxlan_metadata opts;
 591
 592         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 593
 594         memset(&opts, 0, sizeof(opts));
 595         nla_for_each_nested(a, attr, rem) {
 596                 int type = nla_type(a);
 597
 598                 if (type > OVS_VXLAN_EXT_MAX) {
 599                         OVS_NLERR(log, "VXLAN extension %d out of range max %d",
 600                                   type, OVS_VXLAN_EXT_MAX);
 601                         return -EINVAL;
 602                 }
 603
 604                 if (!check_attr_len(nla_len(a),
 605                                     ovs_vxlan_ext_key_lens[type].len)) {
 606                         OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
 607                                   type, nla_len(a),
 608                                   ovs_vxlan_ext_key_lens[type].len);
 609                         return -EINVAL;
 610                 }
 611
 612                 switch (type) {
 613                 case OVS_VXLAN_EXT_GBP:
 614                         opts.gbp = nla_get_u32(a);
 615                         break;
 616                 default:
 617                         OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
 618                                   type);
 619                         return -EINVAL;
 620                 }
 621         }
 622         if (rem) {
 623                 OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
 624                           rem);
 625                 return -EINVAL;
 626         }
 627
 628         if (!is_mask)
 629                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 630         else
 631                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 632
 633         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 634         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 635                                   is_mask);
 636         return 0;
 637 }
 638
 639 static int erspan_tun_opt_from_nlattr(const struct nlattr *a,
 640                                       struct sw_flow_match *match, bool is_mask,
 641                                       bool log)
 642 {
 643         unsigned long opt_key_offset;
 644
 645         BUILD_BUG_ON(sizeof(struct erspan_metadata) >
 646                      sizeof(match->key->tun_opts));
 647
 648         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 649                 OVS_NLERR(log, "ERSPAN option length err (len %d, max %zu).",
 650                           nla_len(a), sizeof(match->key->tun_opts));
 651                 return -EINVAL;
 652         }
 653
 654         if (!is_mask)
 655                 SW_FLOW_KEY_PUT(match, tun_opts_len,
 656                                 sizeof(struct erspan_metadata), false);
 657         else
 658                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 659
 660         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 661         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 662                                   nla_len(a), is_mask);
 663         return 0;
 664 }
 665
 666 static int ip_tun_from_nlattr(const struct nlattr *attr,
 667                               struct sw_flow_match *match, bool is_mask,
 668                               bool log)
 669 {
 670         bool ttl = false, ipv4 = false, ipv6 = false;
 671         bool info_bridge_mode = false;
 672         __be16 tun_flags = 0;
 673         int opts_type = 0;
 674         struct nlattr *a;
 675         int rem;
 676
 677         nla_for_each_nested(a, attr, rem) {
 678                 int type = nla_type(a);
 679                 int err;
 680
 681                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 682                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 683                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 684                         return -EINVAL;
 685                 }
 686
 687                 if (!check_attr_len(nla_len(a),
 688                                     ovs_tunnel_key_lens[type].len)) {
 689                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 690                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 691                         return -EINVAL;
 692                 }
 693
 694                 switch (type) {
 695                 case OVS_TUNNEL_KEY_ATTR_ID:
 696                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 697                                         nla_get_be64(a), is_mask);
 698                         tun_flags |= TUNNEL_KEY;
 699                         break;
 700                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 701                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
 702                                         nla_get_in_addr(a), is_mask);
 703                         ipv4 = true;
 704                         break;
 705                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 706                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
 707                                         nla_get_in_addr(a), is_mask);
 708                         ipv4 = true;
 709                         break;
 710                 case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
 711                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.src,
 712                                         nla_get_in6_addr(a), is_mask);
 713                         ipv6 = true;
 714                         break;
 715                 case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
 716                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
 717                                         nla_get_in6_addr(a), is_mask);
 718                         ipv6 = true;
 719                         break;
 720                 case OVS_TUNNEL_KEY_ATTR_TOS:
 721                         SW_FLOW_KEY_PUT(match, tun_key.tos,
 722                                         nla_get_u8(a), is_mask);
 723                         break;
 724                 case OVS_TUNNEL_KEY_ATTR_TTL:
 725                         SW_FLOW_KEY_PUT(match, tun_key.ttl,
 726                                         nla_get_u8(a), is_mask);
 727                         ttl = true;
 728                         break;
 729                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 730                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 731                         break;
 732                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 733                         tun_flags |= TUNNEL_CSUM;
 734                         break;
 735                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 736                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 737                                         nla_get_be16(a), is_mask);
 738                         break;
 739                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 740                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 741                                         nla_get_be16(a), is_mask);
 742                         break;
 743                 case OVS_TUNNEL_KEY_ATTR_OAM:
 744                         tun_flags |= TUNNEL_OAM;
 745                         break;
 746                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 747                         if (opts_type) {
 748                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 749                                 return -EINVAL;
 750                         }
 751
 752                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 753                         if (err)
 754                                 return err;
 755
 756                         tun_flags |= TUNNEL_GENEVE_OPT;
 757                         opts_type = type;
 758                         break;
 759                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 760                         if (opts_type) {
 761                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 762                                 return -EINVAL;
 763                         }
 764
 765                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 766                         if (err)
 767                                 return err;
 768
 769                         tun_flags |= TUNNEL_VXLAN_OPT;
 770                         opts_type = type;
 771                         break;
 772                 case OVS_TUNNEL_KEY_ATTR_PAD:
 773                         break;
 774                 case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
 775                         if (opts_type) {
 776                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 777                                 return -EINVAL;
 778                         }
 779
 780                         err = erspan_tun_opt_from_nlattr(a, match, is_mask,
 781                                                          log);
 782                         if (err)
 783                                 return err;
 784
 785                         tun_flags |= TUNNEL_ERSPAN_OPT;
 786                         opts_type = type;
 787                         break;
 788                 case OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE:
 789                         info_bridge_mode = true;
 790                         ipv4 = true;
 791                         break;
 792                 default:
 793                         OVS_NLERR(log, "Unknown IP tunnel attribute %d",
 794                                   type);
 795                         return -EINVAL;
 796                 }
 797         }
 798
 799         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 800         if (is_mask)
 801                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
 802         else
 803                 SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
 804                                 false);
 805
 806         if (rem > 0) {
 807                 OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
 808                           rem);
 809                 return -EINVAL;
 810         }
 811
 812         if (ipv4 && ipv6) {
 813                 OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
 814                 return -EINVAL;
 815         }
 816
 817         if (!is_mask) {
 818                 if (!ipv4 && !ipv6) {
 819                         OVS_NLERR(log, "IP tunnel dst address not specified");
 820                         return -EINVAL;
 821                 }
 822                 if (ipv4) {
 823                         if (info_bridge_mode) {
 824                                 if (match->key->tun_key.u.ipv4.src ||
 825                                     match->key->tun_key.u.ipv4.dst ||
 826                                     match->key->tun_key.tp_src ||
 827                                     match->key->tun_key.tp_dst ||
 828                                     match->key->tun_key.ttl ||
 829                                     match->key->tun_key.tos ||
 830                                     tun_flags & ~TUNNEL_KEY) {
 831                                         OVS_NLERR(log, "IPv4 tun info is not correct");
 832                                         return -EINVAL;
 833                                 }
 834                         } else if (!match->key->tun_key.u.ipv4.dst) {
 835                                 OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 836                                 return -EINVAL;
 837                         }
 838                 }
 839                 if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
 840                         OVS_NLERR(log, "IPv6 tunnel dst address is zero");
 841                         return -EINVAL;
 842                 }
 843
 844                 if (!ttl && !info_bridge_mode) {
 845                         OVS_NLERR(log, "IP tunnel TTL not specified.");
 846                         return -EINVAL;
 847                 }
 848         }
 849
 850         return opts_type;
 851 }
 852
 853 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 854                                const void *tun_opts, int swkey_tun_opts_len)
 855 {
 856         const struct vxlan_metadata *opts = tun_opts;
 857         struct nlattr *nla;
 858
 859         nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 860         if (!nla)
 861                 return -EMSGSIZE;
 862
 863         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 864                 return -EMSGSIZE;
 865
 866         nla_nest_end(skb, nla);
 867         return 0;
 868 }
 869
 870 static int __ip_tun_to_nlattr(struct sk_buff *skb,
 871                               const struct ip_tunnel_key *output,
 872                               const void *tun_opts, int swkey_tun_opts_len,
 873                               unsigned short tun_proto, u8 mode)
 874 {
 875         if (output->tun_flags & TUNNEL_KEY &&
 876             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id,
 877                          OVS_TUNNEL_KEY_ATTR_PAD))
 878                 return -EMSGSIZE;
 879
 880         if (mode & IP_TUNNEL_INFO_BRIDGE)
 881                 return nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE)
 882                        ? -EMSGSIZE : 0;
 883
 884         switch (tun_proto) {
 885         case AF_INET:
 886                 if (output->u.ipv4.src &&
 887                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 888                                     output->u.ipv4.src))
 889                         return -EMSGSIZE;
 890                 if (output->u.ipv4.dst &&
 891                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 892                                     output->u.ipv4.dst))
 893                         return -EMSGSIZE;
 894                 break;
 895         case AF_INET6:
 896                 if (!ipv6_addr_any(&output->u.ipv6.src) &&
 897                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
 898                                      &output->u.ipv6.src))
 899                         return -EMSGSIZE;
 900                 if (!ipv6_addr_any(&output->u.ipv6.dst) &&
 901                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
 902                                      &output->u.ipv6.dst))
 903                         return -EMSGSIZE;
 904                 break;
 905         }
 906         if (output->tos &&
 907             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 908                 return -EMSGSIZE;
 909         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 910                 return -EMSGSIZE;
 911         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 912             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 913                 return -EMSGSIZE;
 914         if ((output->tun_flags & TUNNEL_CSUM) &&
 915             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 916                 return -EMSGSIZE;
 917         if (output->tp_src &&
 918             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 919                 return -EMSGSIZE;
 920         if (output->tp_dst &&
 921             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 922                 return -EMSGSIZE;
 923         if ((output->tun_flags & TUNNEL_OAM) &&
 924             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 925                 return -EMSGSIZE;
 926         if (swkey_tun_opts_len) {
 927                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 928                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 929                             swkey_tun_opts_len, tun_opts))
 930                         return -EMSGSIZE;
 931                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 932                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 933                         return -EMSGSIZE;
 934                 else if (output->tun_flags & TUNNEL_ERSPAN_OPT &&
 935                          nla_put(skb, OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
 936                                  swkey_tun_opts_len, tun_opts))
 937                         return -EMSGSIZE;
 938         }
 939
 940         return 0;
 941 }
 942
 943 static int ip_tun_to_nlattr(struct sk_buff *skb,
 944                             const struct ip_tunnel_key *output,
 945                             const void *tun_opts, int swkey_tun_opts_len,
 946                             unsigned short tun_proto, u8 mode)
 947 {
 948         struct nlattr *nla;
 949         int err;
 950
 951         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 952         if (!nla)
 953                 return -EMSGSIZE;
 954
 955         err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
 956                                  tun_proto, mode);
 957         if (err)
 958                 return err;
 959
 960         nla_nest_end(skb, nla);
 961         return 0;
 962 }
 963
 964 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
 965                             struct ip_tunnel_info *tun_info)
 966 {
 967         return __ip_tun_to_nlattr(skb, &tun_info->key,
 968                                   ip_tunnel_info_opts(tun_info),
 969                                   tun_info->options_len,
 970                                   ip_tunnel_info_af(tun_info), tun_info->mode);
 971 }
 972
 973 static int encode_vlan_from_nlattrs(struct sw_flow_match *match,
 974                                     const struct nlattr *a[],
 975                                     bool is_mask, bool inner)
 976 {
 977         __be16 tci = 0;
 978         __be16 tpid = 0;
 979
 980         if (a[OVS_KEY_ATTR_VLAN])
 981                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 982
 983         if (a[OVS_KEY_ATTR_ETHERTYPE])
 984                 tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 985
 986         if (likely(!inner)) {
 987                 SW_FLOW_KEY_PUT(match, eth.vlan.tpid, tpid, is_mask);
 988                 SW_FLOW_KEY_PUT(match, eth.vlan.tci, tci, is_mask);
 989         } else {
 990                 SW_FLOW_KEY_PUT(match, eth.cvlan.tpid, tpid, is_mask);
 991                 SW_FLOW_KEY_PUT(match, eth.cvlan.tci, tci, is_mask);
 992         }
 993         return 0;
 994 }
 995
 996 static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
 997                                       u64 key_attrs, bool inner,
 998                                       const struct nlattr **a, bool log)
 999 {
1000         __be16 tci = 0;
1001
1002         if (!((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1003               (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1004                eth_type_vlan(nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE])))) {
1005                 /* Not a VLAN. */
1006                 return 0;
1007         }
1008
1009         if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1010               (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1011                 OVS_NLERR(log, "Invalid %s frame", (inner) ? "C-VLAN" : "VLAN");
1012                 return -EINVAL;
1013         }
1014
1015         if (a[OVS_KEY_ATTR_VLAN])
1016                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1017
1018         if (!(tci & htons(VLAN_CFI_MASK))) {
1019                 if (tci) {
1020                         OVS_NLERR(log, "%s TCI does not have VLAN_CFI_MASK bit set.",
1021                                   (inner) ? "C-VLAN" : "VLAN");
1022                         return -EINVAL;
1023                 } else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
1024                         /* Corner case for truncated VLAN header. */
1025                         OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
1026                                   (inner) ? "C-VLAN" : "VLAN");
1027                         return -EINVAL;
1028                 }
1029         }
1030
1031         return 1;
1032 }
1033
1034 static int validate_vlan_mask_from_nlattrs(const struct sw_flow_match *match,
1035                                            u64 key_attrs, bool inner,
1036                                            const struct nlattr **a, bool log)
1037 {
1038         __be16 tci = 0;
1039         __be16 tpid = 0;
1040         bool encap_valid = !!(match->key->eth.vlan.tci &
1041                               htons(VLAN_CFI_MASK));
1042         bool i_encap_valid = !!(match->key->eth.cvlan.tci &
1043                                 htons(VLAN_CFI_MASK));
1044
1045         if (!(key_attrs & (1 << OVS_KEY_ATTR_ENCAP))) {
1046                 /* Not a VLAN. */
1047                 return 0;
1048         }
1049
1050         if ((!inner && !encap_valid) || (inner && !i_encap_valid)) {
1051                 OVS_NLERR(log, "Encap mask attribute is set for non-%s frame.",
1052                           (inner) ? "C-VLAN" : "VLAN");
1053                 return -EINVAL;
1054         }
1055
1056         if (a[OVS_KEY_ATTR_VLAN])
1057                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1058
1059         if (a[OVS_KEY_ATTR_ETHERTYPE])
1060                 tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1061
1062         if (tpid != htons(0xffff)) {
1063                 OVS_NLERR(log, "Must have an exact match on %s TPID (mask=%x).",
1064                           (inner) ? "C-VLAN" : "VLAN", ntohs(tpid));
1065                 return -EINVAL;
1066         }
1067         if (!(tci & htons(VLAN_CFI_MASK))) {
1068                 OVS_NLERR(log, "%s TCI mask does not have exact match for VLAN_CFI_MASK bit.",
1069                           (inner) ? "C-VLAN" : "VLAN");
1070                 return -EINVAL;
1071         }
1072
1073         return 1;
1074 }
1075
1076 static int __parse_vlan_from_nlattrs(struct sw_flow_match *match,
1077                                      u64 *key_attrs, bool inner,
1078                                      const struct nlattr **a, bool is_mask,
1079                                      bool log)
1080 {
1081         int err;
1082         const struct nlattr *encap;
1083
1084         if (!is_mask)
1085                 err = validate_vlan_from_nlattrs(match, *key_attrs, inner,
1086                                                  a, log);
1087         else
1088                 err = validate_vlan_mask_from_nlattrs(match, *key_attrs, inner,
1089                                                       a, log);
1090         if (err <= 0)
1091                 return err;
1092
1093         err = encode_vlan_from_nlattrs(match, a, is_mask, inner);
1094         if (err)
1095                 return err;
1096
1097         *key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1098         *key_attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
1099         *key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1100
1101         encap = a[OVS_KEY_ATTR_ENCAP];
1102
1103         if (!is_mask)
1104                 err = parse_flow_nlattrs(encap, a, key_attrs, log);
1105         else
1106                 err = parse_flow_mask_nlattrs(encap, a, key_attrs, log);
1107
1108         return err;
1109 }
1110
1111 static int parse_vlan_from_nlattrs(struct sw_flow_match *match,
1112                                    u64 *key_attrs, const struct nlattr **a,
1113                                    bool is_mask, bool log)
1114 {
1115         int err;
1116         bool encap_valid = false;
1117
1118         err = __parse_vlan_from_nlattrs(match, key_attrs, false, a,
1119                                         is_mask, log);
1120         if (err)
1121                 return err;
1122
1123         encap_valid = !!(match->key->eth.vlan.tci & htons(VLAN_CFI_MASK));
1124         if (encap_valid) {
1125                 err = __parse_vlan_from_nlattrs(match, key_attrs, true, a,
1126                                                 is_mask, log);
1127                 if (err)
1128                         return err;
1129         }
1130
1131         return 0;
1132 }
1133
1134 static int parse_eth_type_from_nlattrs(struct sw_flow_match *match,
1135                                        u64 *attrs, const struct nlattr **a,
1136                                        bool is_mask, bool log)
1137 {
1138         __be16 eth_type;
1139
1140         eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1141         if (is_mask) {
1142                 /* Always exact match EtherType. */
1143                 eth_type = htons(0xffff);
1144         } else if (!eth_proto_is_802_3(eth_type)) {
1145                 OVS_NLERR(log, "EtherType %x is less than min %x",
1146                                 ntohs(eth_type), ETH_P_802_3_MIN);
1147                 return -EINVAL;
1148         }
1149
1150         SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1151         *attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1152         return 0;
1153 }
1154
1155 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
1156                                  u64 *attrs, const struct nlattr **a,
1157                                  bool is_mask, bool log)
1158 {
1159         u8 mac_proto = MAC_PROTO_ETHERNET;
1160
1161         if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
1162                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
1163
1164                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
1165                 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
1166         }
1167
1168         if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
1169                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
1170
1171                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
1172                 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
1173         }
1174
1175         if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1176                 SW_FLOW_KEY_PUT(match, phy.priority,
1177                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1178                 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1179         }
1180
1181         if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1182                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1183
1184                 if (is_mask) {
1185                         in_port = 0xffffffff; /* Always exact match in_port. */
1186                 } else if (in_port >= DP_MAX_PORTS) {
1187                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
1188                                   in_port, DP_MAX_PORTS);
1189                         return -EINVAL;
1190                 }
1191
1192                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1193                 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1194         } else if (!is_mask) {
1195                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1196         }
1197
1198         if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1199                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1200
1201                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1202                 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1203         }
1204         if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1205                 if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1206                                        is_mask, log) < 0)
1207                         return -EINVAL;
1208                 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1209         }
1210
1211         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
1212             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
1213                 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
1214
1215                 if (ct_state & ~CT_SUPPORTED_MASK) {
1216                         OVS_NLERR(log, "ct_state flags %08x unsupported",
1217                                   ct_state);
1218                         return -EINVAL;
1219                 }
1220
1221                 SW_FLOW_KEY_PUT(match, ct_state, ct_state, is_mask);
1222                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
1223         }
1224         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
1225             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
1226                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
1227
1228                 SW_FLOW_KEY_PUT(match, ct_zone, ct_zone, is_mask);
1229                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
1230         }
1231         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
1232             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
1233                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
1234
1235                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
1236                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
1237         }
1238         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
1239             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
1240                 const struct ovs_key_ct_labels *cl;
1241
1242                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
1243                 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
1244                                    sizeof(*cl), is_mask);
1245                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
1246         }
1247         if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)) {
1248                 const struct ovs_key_ct_tuple_ipv4 *ct;
1249
1250                 ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4]);
1251
1252                 SW_FLOW_KEY_PUT(match, ipv4.ct_orig.src, ct->ipv4_src, is_mask);
1253                 SW_FLOW_KEY_PUT(match, ipv4.ct_orig.dst, ct->ipv4_dst, is_mask);
1254                 SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1255                 SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1256                 SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv4_proto, is_mask);
1257                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4);
1258         }
1259         if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)) {
1260                 const struct ovs_key_ct_tuple_ipv6 *ct;
1261
1262                 ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6]);
1263
1264                 SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.src, &ct->ipv6_src,
1265                                    sizeof(match->key->ipv6.ct_orig.src),
1266                                    is_mask);
1267                 SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.dst, &ct->ipv6_dst,
1268                                    sizeof(match->key->ipv6.ct_orig.dst),
1269                                    is_mask);
1270                 SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1271                 SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1272                 SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv6_proto, is_mask);
1273                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
1274         }
1275
1276         /* For layer 3 packets the Ethernet type is provided
1277          * and treated as metadata but no MAC addresses are provided.
1278          */
1279         if (!(*attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1280             (*attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)))
1281                 mac_proto = MAC_PROTO_NONE;
1282
1283         /* Always exact match mac_proto */
1284         SW_FLOW_KEY_PUT(match, mac_proto, is_mask ? 0xff : mac_proto, is_mask);
1285
1286         if (mac_proto == MAC_PROTO_NONE)
1287                 return parse_eth_type_from_nlattrs(match, attrs, a, is_mask,
1288                                                    log);
1289
1290         return 0;
1291 }
1292
1293 int nsh_hdr_from_nlattr(const struct nlattr *attr,
1294                         struct nshhdr *nh, size_t size)
1295 {
1296         struct nlattr *a;
1297         int rem;
1298         u8 flags = 0;
1299         u8 ttl = 0;
1300         int mdlen = 0;
1301
1302         /* validate_nsh has check this, so we needn't do duplicate check here
1303          */
1304         if (size < NSH_BASE_HDR_LEN)
1305                 return -ENOBUFS;
1306
1307         nla_for_each_nested(a, attr, rem) {
1308                 int type = nla_type(a);
1309
1310                 switch (type) {
1311                 case OVS_NSH_KEY_ATTR_BASE: {
1312                         const struct ovs_nsh_key_base *base = nla_data(a);
1313
1314                         flags = base->flags;
1315                         ttl = base->ttl;
1316                         nh->np = base->np;
1317                         nh->mdtype = base->mdtype;
1318                         nh->path_hdr = base->path_hdr;
1319                         break;
1320                 }
1321                 case OVS_NSH_KEY_ATTR_MD1:
1322                         mdlen = nla_len(a);
1323                         if (mdlen > size - NSH_BASE_HDR_LEN)
1324                                 return -ENOBUFS;
1325                         memcpy(&nh->md1, nla_data(a), mdlen);
1326                         break;
1327
1328                 case OVS_NSH_KEY_ATTR_MD2:
1329                         mdlen = nla_len(a);
1330                         if (mdlen > size - NSH_BASE_HDR_LEN)
1331                                 return -ENOBUFS;
1332                         memcpy(&nh->md2, nla_data(a), mdlen);
1333                         break;
1334
1335                 default:
1336                         return -EINVAL;
1337                 }
1338         }
1339
1340         /* nsh header length  = NSH_BASE_HDR_LEN + mdlen */
1341         nh->ver_flags_ttl_len = 0;
1342         nsh_set_flags_ttl_len(nh, flags, ttl, NSH_BASE_HDR_LEN + mdlen);
1343
1344         return 0;
1345 }
1346
1347 int nsh_key_from_nlattr(const struct nlattr *attr,
1348                         struct ovs_key_nsh *nsh, struct ovs_key_nsh *nsh_mask)
1349 {
1350         struct nlattr *a;
1351         int rem;
1352
1353         /* validate_nsh has check this, so we needn't do duplicate check here
1354          */
1355         nla_for_each_nested(a, attr, rem) {
1356                 int type = nla_type(a);
1357
1358                 switch (type) {
1359                 case OVS_NSH_KEY_ATTR_BASE: {
1360                         const struct ovs_nsh_key_base *base = nla_data(a);
1361                         const struct ovs_nsh_key_base *base_mask = base + 1;
1362
1363                         nsh->base = *base;
1364                         nsh_mask->base = *base_mask;
1365                         break;
1366                 }
1367                 case OVS_NSH_KEY_ATTR_MD1: {
1368                         const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1369                         const struct ovs_nsh_key_md1 *md1_mask = md1 + 1;
1370
1371                         memcpy(nsh->context, md1->context, sizeof(*md1));
1372                         memcpy(nsh_mask->context, md1_mask->context,
1373                                sizeof(*md1_mask));
1374                         break;
1375                 }
1376                 case OVS_NSH_KEY_ATTR_MD2:
1377                         /* Not supported yet */
1378                         return -ENOTSUPP;
1379                 default:
1380                         return -EINVAL;
1381                 }
1382         }
1383
1384         return 0;
1385 }
1386
1387 static int nsh_key_put_from_nlattr(const struct nlattr *attr,
1388                                    struct sw_flow_match *match, bool is_mask,
1389                                    bool is_push_nsh, bool log)
1390 {
1391         struct nlattr *a;
1392         int rem;
1393         bool has_base = false;
1394         bool has_md1 = false;
1395         bool has_md2 = false;
1396         u8 mdtype = 0;
1397         int mdlen = 0;
1398
1399         if (WARN_ON(is_push_nsh && is_mask))
1400                 return -EINVAL;
1401
1402         nla_for_each_nested(a, attr, rem) {
1403                 int type = nla_type(a);
1404                 int i;
1405
1406                 if (type > OVS_NSH_KEY_ATTR_MAX) {
1407                         OVS_NLERR(log, "nsh attr %d is out of range max %d",
1408                                   type, OVS_NSH_KEY_ATTR_MAX);
1409                         return -EINVAL;
1410                 }
1411
1412                 if (!check_attr_len(nla_len(a),
1413                                     ovs_nsh_key_attr_lens[type].len)) {
1414                         OVS_NLERR(
1415                             log,
1416                             "nsh attr %d has unexpected len %d expected %d",
1417                             type,
1418                             nla_len(a),
1419                             ovs_nsh_key_attr_lens[type].len
1420                         );
1421                         return -EINVAL;
1422                 }
1423
1424                 switch (type) {
1425                 case OVS_NSH_KEY_ATTR_BASE: {
1426                         const struct ovs_nsh_key_base *base = nla_data(a);
1427
1428                         has_base = true;
1429                         mdtype = base->mdtype;
1430                         SW_FLOW_KEY_PUT(match, nsh.base.flags,
1431                                         base->flags, is_mask);
1432                         SW_FLOW_KEY_PUT(match, nsh.base.ttl,
1433                                         base->ttl, is_mask);
1434                         SW_FLOW_KEY_PUT(match, nsh.base.mdtype,
1435                                         base->mdtype, is_mask);
1436                         SW_FLOW_KEY_PUT(match, nsh.base.np,
1437                                         base->np, is_mask);
1438                         SW_FLOW_KEY_PUT(match, nsh.base.path_hdr,
1439                                         base->path_hdr, is_mask);
1440                         break;
1441                 }
1442                 case OVS_NSH_KEY_ATTR_MD1: {
1443                         const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1444
1445                         has_md1 = true;
1446                         for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++)
1447                                 SW_FLOW_KEY_PUT(match, nsh.context[i],
1448                                                 md1->context[i], is_mask);
1449                         break;
1450                 }
1451                 case OVS_NSH_KEY_ATTR_MD2:
1452                         if (!is_push_nsh) /* Not supported MD type 2 yet */
1453                                 return -ENOTSUPP;
1454
1455                         has_md2 = true;
1456                         mdlen = nla_len(a);
1457                         if (mdlen > NSH_CTX_HDRS_MAX_LEN || mdlen <= 0) {
1458                                 OVS_NLERR(
1459                                     log,
1460                                     "Invalid MD length %d for MD type %d",
1461                                     mdlen,
1462                                     mdtype
1463                                 );
1464                                 return -EINVAL;
1465                         }
1466                         break;
1467                 default:
1468                         OVS_NLERR(log, "Unknown nsh attribute %d",
1469                                   type);
1470                         return -EINVAL;
1471                 }
1472         }
1473
1474         if (rem > 0) {
1475                 OVS_NLERR(log, "nsh attribute has %d unknown bytes.", rem);
1476                 return -EINVAL;
1477         }
1478
1479         if (has_md1 && has_md2) {
1480                 OVS_NLERR(
1481                     1,
1482                     "invalid nsh attribute: md1 and md2 are exclusive."
1483                 );
1484                 return -EINVAL;
1485         }
1486
1487         if (!is_mask) {
1488                 if ((has_md1 && mdtype != NSH_M_TYPE1) ||
1489                     (has_md2 && mdtype != NSH_M_TYPE2)) {
1490                         OVS_NLERR(1, "nsh attribute has unmatched MD type %d.",
1491                                   mdtype);
1492                         return -EINVAL;
1493                 }
1494
1495                 if (is_push_nsh &&
1496                     (!has_base || (!has_md1 && !has_md2))) {
1497                         OVS_NLERR(
1498                             1,
1499                             "push_nsh: missing base or metadata attributes"
1500                         );
1501                         return -EINVAL;
1502                 }
1503         }
1504
1505         return 0;
1506 }
1507
1508 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
1509                                 u64 attrs, const struct nlattr **a,
1510                                 bool is_mask, bool log)
1511 {
1512         int err;
1513
1514         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
1515         if (err)
1516                 return err;
1517
1518         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1519                 const struct ovs_key_ethernet *eth_key;
1520
1521                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1522                 SW_FLOW_KEY_MEMCPY(match, eth.src,
1523                                 eth_key->eth_src, ETH_ALEN, is_mask);
1524                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1525                                 eth_key->eth_dst, ETH_ALEN, is_mask);
1526                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1527
1528                 if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1529                         /* VLAN attribute is always parsed before getting here since it
1530                          * may occur multiple times.
1531                          */
1532                         OVS_NLERR(log, "VLAN attribute unexpected.");
1533                         return -EINVAL;
1534                 }
1535
1536                 if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1537                         err = parse_eth_type_from_nlattrs(match, &attrs, a, is_mask,
1538                                                           log);
1539                         if (err)
1540                                 return err;
1541                 } else if (!is_mask) {
1542                         SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1543                 }
1544         } else if (!match->key->eth.type) {
1545                 OVS_NLERR(log, "Either Ethernet header or EtherType is required.");
1546                 return -EINVAL;
1547         }
1548
1549         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1550                 const struct ovs_key_ipv4 *ipv4_key;
1551
1552                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1553                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1554                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
1555                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1556                         return -EINVAL;
1557                 }
1558                 SW_FLOW_KEY_PUT(match, ip.proto,
1559                                 ipv4_key->ipv4_proto, is_mask);
1560                 SW_FLOW_KEY_PUT(match, ip.tos,
1561                                 ipv4_key->ipv4_tos, is_mask);
1562                 SW_FLOW_KEY_PUT(match, ip.ttl,
1563                                 ipv4_key->ipv4_ttl, is_mask);
1564                 SW_FLOW_KEY_PUT(match, ip.frag,
1565                                 ipv4_key->ipv4_frag, is_mask);
1566                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1567                                 ipv4_key->ipv4_src, is_mask);
1568                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1569                                 ipv4_key->ipv4_dst, is_mask);
1570                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1571         }
1572
1573         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1574                 const struct ovs_key_ipv6 *ipv6_key;
1575
1576                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1577                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1578                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
1579                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1580                         return -EINVAL;
1581                 }
1582
1583                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
1584                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x)",
1585                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1586                         return -EINVAL;
1587                 }
1588
1589                 SW_FLOW_KEY_PUT(match, ipv6.label,
1590                                 ipv6_key->ipv6_label, is_mask);
1591                 SW_FLOW_KEY_PUT(match, ip.proto,
1592                                 ipv6_key->ipv6_proto, is_mask);
1593                 SW_FLOW_KEY_PUT(match, ip.tos,
1594                                 ipv6_key->ipv6_tclass, is_mask);
1595                 SW_FLOW_KEY_PUT(match, ip.ttl,
1596                                 ipv6_key->ipv6_hlimit, is_mask);
1597                 SW_FLOW_KEY_PUT(match, ip.frag,
1598                                 ipv6_key->ipv6_frag, is_mask);
1599                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1600                                 ipv6_key->ipv6_src,
1601                                 sizeof(match->key->ipv6.addr.src),
1602                                 is_mask);
1603                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1604                                 ipv6_key->ipv6_dst,
1605                                 sizeof(match->key->ipv6.addr.dst),
1606                                 is_mask);
1607
1608                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1609         }
1610
1611         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1612                 const struct ovs_key_arp *arp_key;
1613
1614                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1615                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1616                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1617                                   arp_key->arp_op);
1618                         return -EINVAL;
1619                 }
1620
1621                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1622                                 arp_key->arp_sip, is_mask);
1623                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1624                         arp_key->arp_tip, is_mask);
1625                 SW_FLOW_KEY_PUT(match, ip.proto,
1626                                 ntohs(arp_key->arp_op), is_mask);
1627                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1628                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1629                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1630                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1631
1632                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1633         }
1634
1635         if (attrs & (1 << OVS_KEY_ATTR_NSH)) {
1636                 if (nsh_key_put_from_nlattr(a[OVS_KEY_ATTR_NSH], match,
1637                                             is_mask, false, log) < 0)
1638                         return -EINVAL;
1639                 attrs &= ~(1 << OVS_KEY_ATTR_NSH);
1640         }
1641
1642         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1643                 const struct ovs_key_mpls *mpls_key;
1644
1645                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1646                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1647                                 mpls_key->mpls_lse, is_mask);
1648
1649                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1650          }
1651
1652         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1653                 const struct ovs_key_tcp *tcp_key;
1654
1655                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1656                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1657                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1658                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1659         }
1660
1661         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1662                 SW_FLOW_KEY_PUT(match, tp.flags,
1663                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1664                                 is_mask);
1665                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1666         }
1667
1668         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1669                 const struct ovs_key_udp *udp_key;
1670
1671                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1672                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1673                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1674                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1675         }
1676
1677         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1678                 const struct ovs_key_sctp *sctp_key;
1679
1680                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1681                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1682                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1683                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1684         }
1685
1686         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1687                 const struct ovs_key_icmp *icmp_key;
1688
1689                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1690                 SW_FLOW_KEY_PUT(match, tp.src,
1691                                 htons(icmp_key->icmp_type), is_mask);
1692                 SW_FLOW_KEY_PUT(match, tp.dst,
1693                                 htons(icmp_key->icmp_code), is_mask);
1694                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1695         }
1696
1697         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1698                 const struct ovs_key_icmpv6 *icmpv6_key;
1699
1700                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1701                 SW_FLOW_KEY_PUT(match, tp.src,
1702                                 htons(icmpv6_key->icmpv6_type), is_mask);
1703                 SW_FLOW_KEY_PUT(match, tp.dst,
1704                                 htons(icmpv6_key->icmpv6_code), is_mask);
1705                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1706         }
1707
1708         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1709                 const struct ovs_key_nd *nd_key;
1710
1711                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1712                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1713                         nd_key->nd_target,
1714                         sizeof(match->key->ipv6.nd.target),
1715                         is_mask);
1716                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1717                         nd_key->nd_sll, ETH_ALEN, is_mask);
1718                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1719                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1720                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1721         }
1722
1723         if (attrs != 0) {
1724                 OVS_NLERR(log, "Unknown key attributes %llx",
1725                           (unsigned long long)attrs);
1726                 return -EINVAL;
1727         }
1728
1729         return 0;
1730 }
1731
1732 static void nlattr_set(struct nlattr *attr, u8 val,
1733                        const struct ovs_len_tbl *tbl)
1734 {
1735         struct nlattr *nla;
1736         int rem;
1737
1738         /* The nlattr stream should already have been validated */
1739         nla_for_each_nested(nla, attr, rem) {
1740                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1741                         nlattr_set(nla, val, tbl[nla_type(nla)].next ? : tbl);
1742                 else
1743                         memset(nla_data(nla), val, nla_len(nla));
1744
1745                 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1746                         *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1747         }
1748 }
1749
1750 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1751 {
1752         nlattr_set(attr, val, ovs_key_lens);
1753 }
1754
1755 /**
1756  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1757  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1758  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1759  * does not include any don't care bit.
1760  * @net: Used to determine per-namespace field support.
1761  * @match: receives the extracted flow match information.
1762  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1763  * sequence. The fields should of the packet that triggered the creation
1764  * of this flow.
1765  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1766  * attribute specifies the mask field of the wildcarded flow.
1767  * @log: Boolean to allow kernel error logging.  Normally true, but when
1768  * probing for feature compatibility this should be passed in as false to
1769  * suppress unnecessary error logging.
1770  */
1771 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1772                       const struct nlattr *nla_key,
1773                       const struct nlattr *nla_mask,
1774                       bool log)
1775 {
1776         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1777         struct nlattr *newmask = NULL;
1778         u64 key_attrs = 0;
1779         u64 mask_attrs = 0;
1780         int err;
1781
1782         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1783         if (err)
1784                 return err;
1785
1786         err = parse_vlan_from_nlattrs(match, &key_attrs, a, false, log);
1787         if (err)
1788                 return err;
1789
1790         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1791         if (err)
1792                 return err;
1793
1794         if (match->mask) {
1795                 if (!nla_mask) {
1796                         /* Create an exact match mask. We need to set to 0xff
1797                          * all the 'match->mask' fields that have been touched
1798                          * in 'match->key'. We cannot simply memset
1799                          * 'match->mask', because padding bytes and fields not
1800                          * specified in 'match->key' should be left to 0.
1801                          * Instead, we use a stream of netlink attributes,
1802                          * copied from 'key' and set to 0xff.
1803                          * ovs_key_from_nlattrs() will take care of filling
1804                          * 'match->mask' appropriately.
1805                          */
1806                         newmask = kmemdup(nla_key,
1807                                           nla_total_size(nla_len(nla_key)),
1808                                           GFP_KERNEL);
1809                         if (!newmask)
1810                                 return -ENOMEM;
1811
1812                         mask_set_nlattr(newmask, 0xff);
1813
1814                         /* The userspace does not send tunnel attributes that
1815                          * are 0, but we should not wildcard them nonetheless.
1816                          */
1817                         if (match->key->tun_proto)
1818                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1819                                                          0xff, true);
1820
1821                         nla_mask = newmask;
1822                 }
1823
1824                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1825                 if (err)
1826                         goto free_newmask;
1827
1828                 /* Always match on tci. */
1829                 SW_FLOW_KEY_PUT(match, eth.vlan.tci, htons(0xffff), true);
1830                 SW_FLOW_KEY_PUT(match, eth.cvlan.tci, htons(0xffff), true);
1831
1832                 err = parse_vlan_from_nlattrs(match, &mask_attrs, a, true, log);
1833                 if (err)
1834                         goto free_newmask;
1835
1836                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1837                                            log);
1838                 if (err)
1839                         goto free_newmask;
1840         }
1841
1842         if (!match_validate(match, key_attrs, mask_attrs, log))
1843                 err = -EINVAL;
1844
1845 free_newmask:
1846         kfree(newmask);
1847         return err;
1848 }
1849
1850 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1851 {
1852         size_t len;
1853
1854         if (!attr)
1855                 return 0;
1856
1857         len = nla_len(attr);
1858         if (len < 1 || len > MAX_UFID_LENGTH) {
1859                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1860                           nla_len(attr), MAX_UFID_LENGTH);
1861                 return 0;
1862         }
1863
1864         return len;
1865 }
1866
1867 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1868  * or false otherwise.
1869  */
1870 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1871                       bool log)
1872 {
1873         sfid->ufid_len = get_ufid_len(attr, log);
1874         if (sfid->ufid_len)
1875                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1876
1877         return sfid->ufid_len;
1878 }
1879
1880 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1881                            const struct sw_flow_key *key, bool log)
1882 {
1883         struct sw_flow_key *new_key;
1884
1885         if (ovs_nla_get_ufid(sfid, ufid, log))
1886                 return 0;
1887
1888         /* If UFID was not provided, use unmasked key. */
1889         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1890         if (!new_key)
1891                 return -ENOMEM;
1892         memcpy(new_key, key, sizeof(*key));
1893         sfid->unmasked_key = new_key;
1894
1895         return 0;
1896 }
1897
1898 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1899 {
1900         return attr ? nla_get_u32(attr) : 0;
1901 }
1902
1903 /**
1904  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1905  * @net: Network namespace.
1906  * @key: Receives extracted in_port, priority, tun_key, skb_mark and conntrack
1907  * metadata.
1908  * @a: Array of netlink attributes holding parsed %OVS_KEY_ATTR_* Netlink
1909  * attributes.
1910  * @attrs: Bit mask for the netlink attributes included in @a.
1911  * @log: Boolean to allow kernel error logging.  Normally true, but when
1912  * probing for feature compatibility this should be passed in as false to
1913  * suppress unnecessary error logging.
1914  *
1915  * This parses a series of Netlink attributes that form a flow key, which must
1916  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1917  * get the metadata, that is, the parts of the flow key that cannot be
1918  * extracted from the packet itself.
1919  *
1920  * This must be called before the packet key fields are filled in 'key'.
1921  */
1922
1923 int ovs_nla_get_flow_metadata(struct net *net,
1924                               const struct nlattr *a[OVS_KEY_ATTR_MAX + 1],
1925                               u64 attrs, struct sw_flow_key *key, bool log)
1926 {
1927         struct sw_flow_match match;
1928
1929         memset(&match, 0, sizeof(match));
1930         match.key = key;
1931
1932         key->ct_state = 0;
1933         key->ct_zone = 0;
1934         key->ct_orig_proto = 0;
1935         memset(&key->ct, 0, sizeof(key->ct));
1936         memset(&key->ipv4.ct_orig, 0, sizeof(key->ipv4.ct_orig));
1937         memset(&key->ipv6.ct_orig, 0, sizeof(key->ipv6.ct_orig));
1938
1939         key->phy.in_port = DP_MAX_PORTS;
1940
1941         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1942 }
1943
1944 static int ovs_nla_put_vlan(struct sk_buff *skb, const struct vlan_head *vh,
1945                             bool is_mask)
1946 {
1947         __be16 eth_type = !is_mask ? vh->tpid : htons(0xffff);
1948
1949         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1950             nla_put_be16(skb, OVS_KEY_ATTR_VLAN, vh->tci))
1951                 return -EMSGSIZE;
1952         return 0;
1953 }
1954
1955 static int nsh_key_to_nlattr(const struct ovs_key_nsh *nsh, bool is_mask,
1956                              struct sk_buff *skb)
1957 {
1958         struct nlattr *start;
1959
1960         start = nla_nest_start(skb, OVS_KEY_ATTR_NSH);
1961         if (!start)
1962                 return -EMSGSIZE;
1963
1964         if (nla_put(skb, OVS_NSH_KEY_ATTR_BASE, sizeof(nsh->base), &nsh->base))
1965                 goto nla_put_failure;
1966
1967         if (is_mask || nsh->base.mdtype == NSH_M_TYPE1) {
1968                 if (nla_put(skb, OVS_NSH_KEY_ATTR_MD1,
1969                             sizeof(nsh->context), nsh->context))
1970                         goto nla_put_failure;
1971         }
1972
1973         /* Don't support MD type 2 yet */
1974
1975         nla_nest_end(skb, start);
1976
1977         return 0;
1978
1979 nla_put_failure:
1980         return -EMSGSIZE;
1981 }
1982
1983 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1984                              const struct sw_flow_key *output, bool is_mask,
1985                              struct sk_buff *skb)
1986 {
1987         struct ovs_key_ethernet *eth_key;
1988         struct nlattr *nla;
1989         struct nlattr *encap = NULL;
1990         struct nlattr *in_encap = NULL;
1991
1992         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1993                 goto nla_put_failure;
1994
1995         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1996                 goto nla_put_failure;
1997
1998         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1999                 goto nla_put_failure;
2000
2001         if ((swkey->tun_proto || is_mask)) {
2002                 const void *opts = NULL;
2003
2004                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
2005                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
2006
2007                 if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
2008                                      swkey->tun_opts_len, swkey->tun_proto, 0))
2009                         goto nla_put_failure;
2010         }
2011
2012         if (swkey->phy.in_port == DP_MAX_PORTS) {
2013                 if (is_mask && (output->phy.in_port == 0xffff))
2014                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
2015                                 goto nla_put_failure;
2016         } else {
2017                 u16 upper_u16;
2018                 upper_u16 = !is_mask ? 0 : 0xffff;
2019
2020                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
2021                                 (upper_u16 << 16) | output->phy.in_port))
2022                         goto nla_put_failure;
2023         }
2024
2025         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
2026                 goto nla_put_failure;
2027
2028         if (ovs_ct_put_key(swkey, output, skb))
2029                 goto nla_put_failure;
2030
2031         if (ovs_key_mac_proto(swkey) == MAC_PROTO_ETHERNET) {
2032                 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
2033                 if (!nla)
2034                         goto nla_put_failure;
2035
2036                 eth_key = nla_data(nla);
2037                 ether_addr_copy(eth_key->eth_src, output->eth.src);
2038                 ether_addr_copy(eth_key->eth_dst, output->eth.dst);
2039
2040                 if (swkey->eth.vlan.tci || eth_type_vlan(swkey->eth.type)) {
2041                         if (ovs_nla_put_vlan(skb, &output->eth.vlan, is_mask))
2042                                 goto nla_put_failure;
2043                         encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
2044                         if (!swkey->eth.vlan.tci)
2045                                 goto unencap;
2046
2047                         if (swkey->eth.cvlan.tci || eth_type_vlan(swkey->eth.type)) {
2048                                 if (ovs_nla_put_vlan(skb, &output->eth.cvlan, is_mask))
2049                                         goto nla_put_failure;
2050                                 in_encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
2051                                 if (!swkey->eth.cvlan.tci)
2052                                         goto unencap;
2053                         }
2054                 }
2055
2056                 if (swkey->eth.type == htons(ETH_P_802_2)) {
2057                         /*
2058                         * Ethertype 802.2 is represented in the netlink with omitted
2059                         * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
2060                         * 0xffff in the mask attribute.  Ethertype can also
2061                         * be wildcarded.
2062                         */
2063                         if (is_mask && output->eth.type)
2064                                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
2065                                                         output->eth.type))
2066                                         goto nla_put_failure;
2067                         goto unencap;
2068                 }
2069         }
2070
2071         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
2072                 goto nla_put_failure;
2073
2074         if (eth_type_vlan(swkey->eth.type)) {
2075                 /* There are 3 VLAN tags, we don't know anything about the rest
2076                  * of the packet, so truncate here.
2077                  */
2078                 WARN_ON_ONCE(!(encap && in_encap));
2079                 goto unencap;
2080         }
2081
2082         if (swkey->eth.type == htons(ETH_P_IP)) {
2083                 struct ovs_key_ipv4 *ipv4_key;
2084
2085                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
2086                 if (!nla)
2087                         goto nla_put_failure;
2088                 ipv4_key = nla_data(nla);
2089                 ipv4_key->ipv4_src = output->ipv4.addr.src;
2090                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
2091                 ipv4_key->ipv4_proto = output->ip.proto;
2092                 ipv4_key->ipv4_tos = output->ip.tos;
2093                 ipv4_key->ipv4_ttl = output->ip.ttl;
2094                 ipv4_key->ipv4_frag = output->ip.frag;
2095         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
2096                 struct ovs_key_ipv6 *ipv6_key;
2097
2098                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
2099                 if (!nla)
2100                         goto nla_put_failure;
2101                 ipv6_key = nla_data(nla);
2102                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
2103                                 sizeof(ipv6_key->ipv6_src));
2104                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
2105                                 sizeof(ipv6_key->ipv6_dst));
2106                 ipv6_key->ipv6_label = output->ipv6.label;
2107                 ipv6_key->ipv6_proto = output->ip.proto;
2108                 ipv6_key->ipv6_tclass = output->ip.tos;
2109                 ipv6_key->ipv6_hlimit = output->ip.ttl;
2110                 ipv6_key->ipv6_frag = output->ip.frag;
2111         } else if (swkey->eth.type == htons(ETH_P_NSH)) {
2112                 if (nsh_key_to_nlattr(&output->nsh, is_mask, skb))
2113                         goto nla_put_failure;
2114         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
2115                    swkey->eth.type == htons(ETH_P_RARP)) {
2116                 struct ovs_key_arp *arp_key;
2117
2118                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
2119                 if (!nla)
2120                         goto nla_put_failure;
2121                 arp_key = nla_data(nla);
2122                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
2123                 arp_key->arp_sip = output->ipv4.addr.src;
2124                 arp_key->arp_tip = output->ipv4.addr.dst;
2125                 arp_key->arp_op = htons(output->ip.proto);
2126                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
2127                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
2128         } else if (eth_p_mpls(swkey->eth.type)) {
2129                 struct ovs_key_mpls *mpls_key;
2130
2131                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
2132                 if (!nla)
2133                         goto nla_put_failure;
2134                 mpls_key = nla_data(nla);
2135                 mpls_key->mpls_lse = output->mpls.top_lse;
2136         }
2137
2138         if ((swkey->eth.type == htons(ETH_P_IP) ||
2139              swkey->eth.type == htons(ETH_P_IPV6)) &&
2140              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
2141
2142                 if (swkey->ip.proto == IPPROTO_TCP) {
2143                         struct ovs_key_tcp *tcp_key;
2144
2145                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
2146                         if (!nla)
2147                                 goto nla_put_failure;
2148                         tcp_key = nla_data(nla);
2149                         tcp_key->tcp_src = output->tp.src;
2150                         tcp_key->tcp_dst = output->tp.dst;
2151                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
2152                                          output->tp.flags))
2153                                 goto nla_put_failure;
2154                 } else if (swkey->ip.proto == IPPROTO_UDP) {
2155                         struct ovs_key_udp *udp_key;
2156
2157                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
2158                         if (!nla)
2159                                 goto nla_put_failure;
2160                         udp_key = nla_data(nla);
2161                         udp_key->udp_src = output->tp.src;
2162                         udp_key->udp_dst = output->tp.dst;
2163                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
2164                         struct ovs_key_sctp *sctp_key;
2165
2166                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
2167                         if (!nla)
2168                                 goto nla_put_failure;
2169                         sctp_key = nla_data(nla);
2170                         sctp_key->sctp_src = output->tp.src;
2171                         sctp_key->sctp_dst = output->tp.dst;
2172                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
2173                            swkey->ip.proto == IPPROTO_ICMP) {
2174                         struct ovs_key_icmp *icmp_key;
2175
2176                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
2177                         if (!nla)
2178                                 goto nla_put_failure;
2179                         icmp_key = nla_data(nla);
2180                         icmp_key->icmp_type = ntohs(output->tp.src);
2181                         icmp_key->icmp_code = ntohs(output->tp.dst);
2182                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
2183                            swkey->ip.proto == IPPROTO_ICMPV6) {
2184                         struct ovs_key_icmpv6 *icmpv6_key;
2185
2186                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
2187                                                 sizeof(*icmpv6_key));
2188                         if (!nla)
2189                                 goto nla_put_failure;
2190                         icmpv6_key = nla_data(nla);
2191                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
2192                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
2193
2194                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
2195                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
2196                                 struct ovs_key_nd *nd_key;
2197
2198                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
2199                                 if (!nla)
2200                                         goto nla_put_failure;
2201                                 nd_key = nla_data(nla);
2202                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
2203                                                         sizeof(nd_key->nd_target));
2204                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
2205                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
2206                         }
2207                 }
2208         }
2209
2210 unencap:
2211         if (in_encap)
2212                 nla_nest_end(skb, in_encap);
2213         if (encap)
2214                 nla_nest_end(skb, encap);
2215
2216         return 0;
2217
2218 nla_put_failure:
2219         return -EMSGSIZE;
2220 }
2221
2222 int ovs_nla_put_key(const struct sw_flow_key *swkey,
2223                     const struct sw_flow_key *output, int attr, bool is_mask,
2224                     struct sk_buff *skb)
2225 {
2226         int err;
2227         struct nlattr *nla;
2228
2229         nla = nla_nest_start(skb, attr);
2230         if (!nla)
2231                 return -EMSGSIZE;
2232         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
2233         if (err)
2234                 return err;
2235         nla_nest_end(skb, nla);
2236
2237         return 0;
2238 }
2239
2240 /* Called with ovs_mutex or RCU read lock. */
2241 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
2242 {
2243         if (ovs_identifier_is_ufid(&flow->id))
2244                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
2245                                flow->id.ufid);
2246
2247         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
2248                                OVS_FLOW_ATTR_KEY, false, skb);
2249 }
2250
2251 /* Called with ovs_mutex or RCU read lock. */
2252 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
2253 {
2254         return ovs_nla_put_key(&flow->key, &flow->key,
2255                                 OVS_FLOW_ATTR_KEY, false, skb);
2256 }
2257
2258 /* Called with ovs_mutex or RCU read lock. */
2259 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
2260 {
2261         return ovs_nla_put_key(&flow->key, &flow->mask->key,
2262                                 OVS_FLOW_ATTR_MASK, true, skb);
2263 }
2264
2265 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
2266
2267 static struct sw_flow_actions *nla_alloc_flow_actions(int size)
2268 {
2269         struct sw_flow_actions *sfa;
2270
2271         WARN_ON_ONCE(size > MAX_ACTIONS_BUFSIZE);
2272
2273         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
2274         if (!sfa)
2275                 return ERR_PTR(-ENOMEM);
2276
2277         sfa->actions_len = 0;
2278         return sfa;
2279 }
2280
2281 static void ovs_nla_free_set_action(const struct nlattr *a)
2282 {
2283         const struct nlattr *ovs_key = nla_data(a);
2284         struct ovs_tunnel_info *ovs_tun;
2285
2286         switch (nla_type(ovs_key)) {
2287         case OVS_KEY_ATTR_TUNNEL_INFO:
2288                 ovs_tun = nla_data(ovs_key);
2289                 dst_release((struct dst_entry *)ovs_tun->tun_dst);
2290                 break;
2291         }
2292 }
2293
2294 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
2295 {
2296         const struct nlattr *a;
2297         int rem;
2298
2299         if (!sf_acts)
2300                 return;
2301
2302         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
2303                 switch (nla_type(a)) {
2304                 case OVS_ACTION_ATTR_SET:
2305                         ovs_nla_free_set_action(a);
2306                         break;
2307                 case OVS_ACTION_ATTR_CT:
2308                         ovs_ct_free_action(a);
2309                         break;
2310                 }
2311         }
2312
2313         kfree(sf_acts);
2314 }
2315
2316 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
2317 {
2318         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
2319 }
2320
2321 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
2322  * The caller must hold rcu_read_lock for this to be sensible. */
2323 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
2324 {
2325         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
2326 }
2327
2328 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
2329                                        int attr_len, bool log)
2330 {
2331
2332         struct sw_flow_actions *acts;
2333         int new_acts_size;
2334         size_t req_size = NLA_ALIGN(attr_len);
2335         int next_offset = offsetof(struct sw_flow_actions, actions) +
2336                                         (*sfa)->actions_len;
2337
2338         if (req_size <= (ksize(*sfa) - next_offset))
2339                 goto out;
2340
2341         new_acts_size = max(next_offset + req_size, ksize(*sfa) * 2);
2342
2343         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
2344                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size) {
2345                         OVS_NLERR(log, "Flow action size exceeds max %u",
2346                                   MAX_ACTIONS_BUFSIZE);
2347                         return ERR_PTR(-EMSGSIZE);
2348                 }
2349                 new_acts_size = MAX_ACTIONS_BUFSIZE;
2350         }
2351
2352         acts = nla_alloc_flow_actions(new_acts_size);
2353         if (IS_ERR(acts))
2354                 return (void *)acts;
2355
2356         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
2357         acts->actions_len = (*sfa)->actions_len;
2358         acts->orig_len = (*sfa)->orig_len;
2359         kfree(*sfa);
2360         *sfa = acts;
2361
2362 out:
2363         (*sfa)->actions_len += req_size;
2364         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
2365 }
2366
2367 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
2368                                    int attrtype, void *data, int len, bool log)
2369 {
2370         struct nlattr *a;
2371
2372         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
2373         if (IS_ERR(a))
2374                 return a;
2375
2376         a->nla_type = attrtype;
2377         a->nla_len = nla_attr_size(len);
2378
2379         if (data)
2380                 memcpy(nla_data(a), data, len);
2381         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
2382
2383         return a;
2384 }
2385
2386 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
2387                        int len, bool log)
2388 {
2389         struct nlattr *a;
2390
2391         a = __add_action(sfa, attrtype, data, len, log);
2392
2393         return PTR_ERR_OR_ZERO(a);
2394 }
2395
2396 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
2397                                           int attrtype, bool log)
2398 {
2399         int used = (*sfa)->actions_len;
2400         int err;
2401
2402         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
2403         if (err)
2404                 return err;
2405
2406         return used;
2407 }
2408
2409 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
2410                                          int st_offset)
2411 {
2412         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
2413                                                                st_offset);
2414
2415         a->nla_len = sfa->actions_len - st_offset;
2416 }
2417
2418 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2419                                   const struct sw_flow_key *key,
2420                                   struct sw_flow_actions **sfa,
2421                                   __be16 eth_type, __be16 vlan_tci, bool log);
2422
2423 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
2424                                     const struct sw_flow_key *key,
2425                                     struct sw_flow_actions **sfa,
2426                                     __be16 eth_type, __be16 vlan_tci,
2427                                     bool log, bool last)
2428 {
2429         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
2430         const struct nlattr *probability, *actions;
2431         const struct nlattr *a;
2432         int rem, start, err;
2433         struct sample_arg arg;
2434
2435         memset(attrs, 0, sizeof(attrs));
2436         nla_for_each_nested(a, attr, rem) {
2437                 int type = nla_type(a);
2438                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
2439                         return -EINVAL;
2440                 attrs[type] = a;
2441         }
2442         if (rem)
2443                 return -EINVAL;
2444
2445         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
2446         if (!probability || nla_len(probability) != sizeof(u32))
2447                 return -EINVAL;
2448
2449         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
2450         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
2451                 return -EINVAL;
2452
2453         /* validation done, copy sample action. */
2454         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
2455         if (start < 0)
2456                 return start;
2457
2458         /* When both skb and flow may be changed, put the sample
2459          * into a deferred fifo. On the other hand, if only skb
2460          * may be modified, the actions can be executed in place.
2461          *
2462          * Do this analysis at the flow installation time.
2463          * Set 'clone_action->exec' to true if the actions can be
2464          * executed without being deferred.
2465          *
2466          * If the sample is the last action, it can always be excuted
2467          * rather than deferred.
2468          */
2469         arg.exec = last || !actions_may_change_flow(actions);
2470         arg.probability = nla_get_u32(probability);
2471
2472         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_ARG, &arg, sizeof(arg),
2473                                  log);
2474         if (err)
2475                 return err;
2476
2477         err = __ovs_nla_copy_actions(net, actions, key, sfa,
2478                                      eth_type, vlan_tci, log);
2479
2480         if (err)
2481                 return err;
2482
2483         add_nested_action_end(*sfa, start);
2484
2485         return 0;
2486 }
2487
2488 static int validate_and_copy_clone(struct net *net,
2489                                    const struct nlattr *attr,
2490                                    const struct sw_flow_key *key,
2491                                    struct sw_flow_actions **sfa,
2492                                    __be16 eth_type, __be16 vlan_tci,
2493                                    bool log, bool last)
2494 {
2495         int start, err;
2496         u32 exec;
2497
2498         if (nla_len(attr) && nla_len(attr) < NLA_HDRLEN)
2499                 return -EINVAL;
2500
2501         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CLONE, log);
2502         if (start < 0)
2503                 return start;
2504
2505         exec = last || !actions_may_change_flow(attr);
2506
2507         err = ovs_nla_add_action(sfa, OVS_CLONE_ATTR_EXEC, &exec,
2508                                  sizeof(exec), log);
2509         if (err)
2510                 return err;
2511
2512         err = __ovs_nla_copy_actions(net, attr, key, sfa,
2513                                      eth_type, vlan_tci, log);
2514         if (err)
2515                 return err;
2516
2517         add_nested_action_end(*sfa, start);
2518
2519         return 0;
2520 }
2521
2522 void ovs_match_init(struct sw_flow_match *match,
2523                     struct sw_flow_key *key,
2524                     bool reset_key,
2525                     struct sw_flow_mask *mask)
2526 {
2527         memset(match, 0, sizeof(*match));
2528         match->key = key;
2529         match->mask = mask;
2530
2531         if (reset_key)
2532                 memset(key, 0, sizeof(*key));
2533
2534         if (mask) {
2535                 memset(&mask->key, 0, sizeof(mask->key));
2536                 mask->range.start = mask->range.end = 0;
2537         }
2538 }
2539
2540 static int validate_geneve_opts(struct sw_flow_key *key)
2541 {
2542         struct geneve_opt *option;
2543         int opts_len = key->tun_opts_len;
2544         bool crit_opt = false;
2545
2546         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
2547         while (opts_len > 0) {
2548                 int len;
2549
2550                 if (opts_len < sizeof(*option))
2551                         return -EINVAL;
2552
2553                 len = sizeof(*option) + option->length * 4;
2554                 if (len > opts_len)
2555                         return -EINVAL;
2556
2557                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
2558
2559                 option = (struct geneve_opt *)((u8 *)option + len);
2560                 opts_len -= len;
2561         }
2562
2563         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
2564
2565         return 0;
2566 }
2567
2568 static int validate_and_copy_set_tun(const struct nlattr *attr,
2569                                      struct sw_flow_actions **sfa, bool log)
2570 {
2571         struct sw_flow_match match;
2572         struct sw_flow_key key;
2573         struct metadata_dst *tun_dst;
2574         struct ip_tunnel_info *tun_info;
2575         struct ovs_tunnel_info *ovs_tun;
2576         struct nlattr *a;
2577         int err = 0, start, opts_type;
2578         __be16 dst_opt_type;
2579
2580         dst_opt_type = 0;
2581         ovs_match_init(&match, &key, true, NULL);
2582         opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
2583         if (opts_type < 0)
2584                 return opts_type;
2585
2586         if (key.tun_opts_len) {
2587                 switch (opts_type) {
2588                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2589                         err = validate_geneve_opts(&key);
2590                         if (err < 0)
2591                                 return err;
2592                         dst_opt_type = TUNNEL_GENEVE_OPT;
2593                         break;
2594                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2595                         dst_opt_type = TUNNEL_VXLAN_OPT;
2596                         break;
2597                 case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
2598                         dst_opt_type = TUNNEL_ERSPAN_OPT;
2599                         break;
2600                 }
2601         }
2602
2603         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
2604         if (start < 0)
2605                 return start;
2606
2607         tun_dst = metadata_dst_alloc(key.tun_opts_len, METADATA_IP_TUNNEL,
2608                                      GFP_KERNEL);
2609
2610         if (!tun_dst)
2611                 return -ENOMEM;
2612
2613         err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
2614         if (err) {
2615                 dst_release((struct dst_entry *)tun_dst);
2616                 return err;
2617         }
2618
2619         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
2620                          sizeof(*ovs_tun), log);
2621         if (IS_ERR(a)) {
2622                 dst_release((struct dst_entry *)tun_dst);
2623                 return PTR_ERR(a);
2624         }
2625
2626         ovs_tun = nla_data(a);
2627         ovs_tun->tun_dst = tun_dst;
2628
2629         tun_info = &tun_dst->u.tun_info;
2630         tun_info->mode = IP_TUNNEL_INFO_TX;
2631         if (key.tun_proto == AF_INET6)
2632                 tun_info->mode |= IP_TUNNEL_INFO_IPV6;
2633         else if (key.tun_proto == AF_INET && key.tun_key.u.ipv4.dst == 0)
2634                 tun_info->mode |= IP_TUNNEL_INFO_BRIDGE;
2635         tun_info->key = key.tun_key;
2636
2637         /* We need to store the options in the action itself since
2638          * everything else will go away after flow setup. We can append
2639          * it to tun_info and then point there.
2640          */
2641         ip_tunnel_info_opts_set(tun_info,
2642                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
2643                                 key.tun_opts_len, dst_opt_type);
2644         add_nested_action_end(*sfa, start);
2645
2646         return err;
2647 }
2648
2649 static bool validate_nsh(const struct nlattr *attr, bool is_mask,
2650                          bool is_push_nsh, bool log)
2651 {
2652         struct sw_flow_match match;
2653         struct sw_flow_key key;
2654         int ret = 0;
2655
2656         ovs_match_init(&match, &key, true, NULL);
2657         ret = nsh_key_put_from_nlattr(attr, &match, is_mask,
2658                                       is_push_nsh, log);
2659         return !ret;
2660 }
2661
2662 /* Return false if there are any non-masked bits set.
2663  * Mask follows data immediately, before any netlink padding.
2664  */
2665 static bool validate_masked(u8 *data, int len)
2666 {
2667         u8 *mask = data + len;
2668
2669         while (len--)
2670                 if (*data++ & ~*mask++)
2671                         return false;
2672
2673         return true;
2674 }
2675
2676 static int validate_set(const struct nlattr *a,
2677                         const struct sw_flow_key *flow_key,
2678                         struct sw_flow_actions **sfa, bool *skip_copy,
2679                         u8 mac_proto, __be16 eth_type, bool masked, bool log)
2680 {
2681         const struct nlattr *ovs_key = nla_data(a);
2682         int key_type = nla_type(ovs_key);
2683         size_t key_len;
2684
2685         /* There can be only one key in a action */
2686         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2687                 return -EINVAL;
2688
2689         key_len = nla_len(ovs_key);
2690         if (masked)
2691                 key_len /= 2;
2692
2693         if (key_type > OVS_KEY_ATTR_MAX ||
2694             !check_attr_len(key_len, ovs_key_lens[key_type].len))
2695                 return -EINVAL;
2696
2697         if (masked && !validate_masked(nla_data(ovs_key), key_len))
2698                 return -EINVAL;
2699
2700         switch (key_type) {
2701         const struct ovs_key_ipv4 *ipv4_key;
2702         const struct ovs_key_ipv6 *ipv6_key;
2703         int err;
2704
2705         case OVS_KEY_ATTR_PRIORITY:
2706         case OVS_KEY_ATTR_SKB_MARK:
2707         case OVS_KEY_ATTR_CT_MARK:
2708         case OVS_KEY_ATTR_CT_LABELS:
2709                 break;
2710
2711         case OVS_KEY_ATTR_ETHERNET:
2712                 if (mac_proto != MAC_PROTO_ETHERNET)
2713                         return -EINVAL;
2714                 break;
2715
2716         case OVS_KEY_ATTR_TUNNEL:
2717                 if (masked)
2718                         return -EINVAL; /* Masked tunnel set not supported. */
2719
2720                 *skip_copy = true;
2721                 err = validate_and_copy_set_tun(a, sfa, log);
2722                 if (err)
2723                         return err;
2724                 break;
2725
2726         case OVS_KEY_ATTR_IPV4:
2727                 if (eth_type != htons(ETH_P_IP))
2728                         return -EINVAL;
2729
2730                 ipv4_key = nla_data(ovs_key);
2731
2732                 if (masked) {
2733                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2734
2735                         /* Non-writeable fields. */
2736                         if (mask->ipv4_proto || mask->ipv4_frag)
2737                                 return -EINVAL;
2738                 } else {
2739                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2740                                 return -EINVAL;
2741
2742                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2743                                 return -EINVAL;
2744                 }
2745                 break;
2746
2747         case OVS_KEY_ATTR_IPV6:
2748                 if (eth_type != htons(ETH_P_IPV6))
2749                         return -EINVAL;
2750
2751                 ipv6_key = nla_data(ovs_key);
2752
2753                 if (masked) {
2754                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2755
2756                         /* Non-writeable fields. */
2757                         if (mask->ipv6_proto || mask->ipv6_frag)
2758                                 return -EINVAL;
2759
2760                         /* Invalid bits in the flow label mask? */
2761                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2762                                 return -EINVAL;
2763                 } else {
2764                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2765                                 return -EINVAL;
2766
2767                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2768                                 return -EINVAL;
2769                 }
2770                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2771                         return -EINVAL;
2772
2773                 break;
2774
2775         case OVS_KEY_ATTR_TCP:
2776                 if ((eth_type != htons(ETH_P_IP) &&
2777                      eth_type != htons(ETH_P_IPV6)) ||
2778                     flow_key->ip.proto != IPPROTO_TCP)
2779                         return -EINVAL;
2780
2781                 break;
2782
2783         case OVS_KEY_ATTR_UDP:
2784                 if ((eth_type != htons(ETH_P_IP) &&
2785                      eth_type != htons(ETH_P_IPV6)) ||
2786                     flow_key->ip.proto != IPPROTO_UDP)
2787                         return -EINVAL;
2788
2789                 break;
2790
2791         case OVS_KEY_ATTR_MPLS:
2792                 if (!eth_p_mpls(eth_type))
2793                         return -EINVAL;
2794                 break;
2795
2796         case OVS_KEY_ATTR_SCTP:
2797                 if ((eth_type != htons(ETH_P_IP) &&
2798                      eth_type != htons(ETH_P_IPV6)) ||
2799                     flow_key->ip.proto != IPPROTO_SCTP)
2800                         return -EINVAL;
2801
2802                 break;
2803
2804         case OVS_KEY_ATTR_NSH:
2805                 if (eth_type != htons(ETH_P_NSH))
2806                         return -EINVAL;
2807                 if (!validate_nsh(nla_data(a), masked, false, log))
2808                         return -EINVAL;
2809                 break;
2810
2811         default:
2812                 return -EINVAL;
2813         }
2814
2815         /* Convert non-masked non-tunnel set actions to masked set actions. */
2816         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2817                 int start, len = key_len * 2;
2818                 struct nlattr *at;
2819
2820                 *skip_copy = true;
2821
2822                 start = add_nested_action_start(sfa,
2823                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2824                                                 log);
2825                 if (start < 0)
2826                         return start;
2827
2828                 at = __add_action(sfa, key_type, NULL, len, log);
2829                 if (IS_ERR(at))
2830                         return PTR_ERR(at);
2831
2832                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2833                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2834                 /* Clear non-writeable bits from otherwise writeable fields. */
2835                 if (key_type == OVS_KEY_ATTR_IPV6) {
2836                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2837
2838                         mask->ipv6_label &= htonl(0x000FFFFF);
2839                 }
2840                 add_nested_action_end(*sfa, start);
2841         }
2842
2843         return 0;
2844 }
2845
2846 static int validate_userspace(const struct nlattr *attr)
2847 {
2848         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2849                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2850                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2851                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2852         };
2853         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2854         int error;
2855
2856         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, attr,
2857                                  userspace_policy, NULL);
2858         if (error)
2859                 return error;
2860
2861         if (!a[OVS_USERSPACE_ATTR_PID] ||
2862             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2863                 return -EINVAL;
2864
2865         return 0;
2866 }
2867
2868 static const struct nla_policy cpl_policy[OVS_CHECK_PKT_LEN_ATTR_MAX + 1] = {
2869         [OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] = {.type = NLA_U16 },
2870         [OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER] = {.type = NLA_NESTED },
2871         [OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL] = {.type = NLA_NESTED },
2872 };
2873
2874 static int validate_and_copy_check_pkt_len(struct net *net,
2875                                            const struct nlattr *attr,
2876                                            const struct sw_flow_key *key,
2877                                            struct sw_flow_actions **sfa,
2878                                            __be16 eth_type, __be16 vlan_tci,
2879                                            bool log, bool last)
2880 {
2881         const struct nlattr *acts_if_greater, *acts_if_lesser_eq;
2882         struct nlattr *a[OVS_CHECK_PKT_LEN_ATTR_MAX + 1];
2883         struct check_pkt_len_arg arg;
2884         int nested_acts_start;
2885         int start, err;
2886
2887         err = nla_parse_strict(a, OVS_CHECK_PKT_LEN_ATTR_MAX, nla_data(attr),
2888                                nla_len(attr), cpl_policy, NULL);
2889         if (err)
2890                 return err;
2891
2892         if (!a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] ||
2893             !nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]))
2894                 return -EINVAL;
2895
2896         acts_if_lesser_eq = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL];
2897         acts_if_greater = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER];
2898
2899         /* Both the nested action should be present. */
2900         if (!acts_if_greater || !acts_if_lesser_eq)
2901                 return -EINVAL;
2902
2903         /* validation done, copy the nested actions. */
2904         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CHECK_PKT_LEN,
2905                                         log);
2906         if (start < 0)
2907                 return start;
2908
2909         arg.pkt_len = nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]);
2910         arg.exec_for_lesser_equal =
2911                 last || !actions_may_change_flow(acts_if_lesser_eq);
2912         arg.exec_for_greater =
2913                 last || !actions_may_change_flow(acts_if_greater);
2914
2915         err = ovs_nla_add_action(sfa, OVS_CHECK_PKT_LEN_ATTR_ARG, &arg,
2916                                  sizeof(arg), log);
2917         if (err)
2918                 return err;
2919
2920         nested_acts_start = add_nested_action_start(sfa,
2921                 OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL, log);
2922         if (nested_acts_start < 0)
2923                 return nested_acts_start;
2924
2925         err = __ovs_nla_copy_actions(net, acts_if_lesser_eq, key, sfa,
2926                                      eth_type, vlan_tci, log);
2927
2928         if (err)
2929                 return err;
2930
2931         add_nested_action_end(*sfa, nested_acts_start);
2932
2933         nested_acts_start = add_nested_action_start(sfa,
2934                 OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER, log);
2935         if (nested_acts_start < 0)
2936                 return nested_acts_start;
2937
2938         err = __ovs_nla_copy_actions(net, acts_if_greater, key, sfa,
2939                                      eth_type, vlan_tci, log);
2940
2941         if (err)
2942                 return err;
2943
2944         add_nested_action_end(*sfa, nested_acts_start);
2945         add_nested_action_end(*sfa, start);
2946         return 0;
2947 }
2948
2949 static int copy_action(const struct nlattr *from,
2950                        struct sw_flow_actions **sfa, bool log)
2951 {
2952         int totlen = NLA_ALIGN(from->nla_len);
2953         struct nlattr *to;
2954
2955         to = reserve_sfa_size(sfa, from->nla_len, log);
2956         if (IS_ERR(to))
2957                 return PTR_ERR(to);
2958
2959         memcpy(to, from, totlen);
2960         return 0;
2961 }
2962
2963 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2964                                   const struct sw_flow_key *key,
2965                                   struct sw_flow_actions **sfa,
2966                                   __be16 eth_type, __be16 vlan_tci, bool log)
2967 {
2968         u8 mac_proto = ovs_key_mac_proto(key);
2969         const struct nlattr *a;
2970         int rem, err;
2971
2972         nla_for_each_nested(a, attr, rem) {
2973                 /* Expected argument lengths, (u32)-1 for variable length. */
2974                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2975                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2976                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2977                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2978                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2979                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2980                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2981                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2982                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2983                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2984                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2985                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2986                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2987                         [OVS_ACTION_ATTR_CT_CLEAR] = 0,
2988                         [OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
2989                         [OVS_ACTION_ATTR_PUSH_ETH] = sizeof(struct ovs_action_push_eth),
2990                         [OVS_ACTION_ATTR_POP_ETH] = 0,
2991                         [OVS_ACTION_ATTR_PUSH_NSH] = (u32)-1,
2992                         [OVS_ACTION_ATTR_POP_NSH] = 0,
2993                         [OVS_ACTION_ATTR_METER] = sizeof(u32),
2994                         [OVS_ACTION_ATTR_CLONE] = (u32)-1,
2995                         [OVS_ACTION_ATTR_CHECK_PKT_LEN] = (u32)-1,
2996                 };
2997                 const struct ovs_action_push_vlan *vlan;
2998                 int type = nla_type(a);
2999                 bool skip_copy;
3000
3001                 if (type > OVS_ACTION_ATTR_MAX ||
3002                     (action_lens[type] != nla_len(a) &&
3003                      action_lens[type] != (u32)-1))
3004                         return -EINVAL;
3005
3006                 skip_copy = false;
3007                 switch (type) {
3008                 case OVS_ACTION_ATTR_UNSPEC:
3009                         return -EINVAL;
3010
3011                 case OVS_ACTION_ATTR_USERSPACE:
3012                         err = validate_userspace(a);
3013                         if (err)
3014                                 return err;
3015                         break;
3016
3017                 case OVS_ACTION_ATTR_OUTPUT:
3018                         if (nla_get_u32(a) >= DP_MAX_PORTS)
3019                                 return -EINVAL;
3020                         break;
3021
3022                 case OVS_ACTION_ATTR_TRUNC: {
3023                         const struct ovs_action_trunc *trunc = nla_data(a);
3024
3025                         if (trunc->max_len < ETH_HLEN)
3026                                 return -EINVAL;
3027                         break;
3028                 }
3029
3030                 case OVS_ACTION_ATTR_HASH: {
3031                         const struct ovs_action_hash *act_hash = nla_data(a);
3032
3033                         switch (act_hash->hash_alg) {
3034                         case OVS_HASH_ALG_L4:
3035                                 break;
3036                         default:
3037                                 return  -EINVAL;
3038                         }
3039
3040                         break;
3041                 }
3042
3043                 case OVS_ACTION_ATTR_POP_VLAN:
3044                         if (mac_proto != MAC_PROTO_ETHERNET)
3045                                 return -EINVAL;
3046                         vlan_tci = htons(0);
3047                         break;
3048
3049                 case OVS_ACTION_ATTR_PUSH_VLAN:
3050                         if (mac_proto != MAC_PROTO_ETHERNET)
3051                                 return -EINVAL;
3052                         vlan = nla_data(a);
3053                         if (!eth_type_vlan(vlan->vlan_tpid))
3054                                 return -EINVAL;
3055                         if (!(vlan->vlan_tci & htons(VLAN_CFI_MASK)))
3056                                 return -EINVAL;
3057                         vlan_tci = vlan->vlan_tci;
3058                         break;
3059
3060                 case OVS_ACTION_ATTR_RECIRC:
3061                         break;
3062
3063                 case OVS_ACTION_ATTR_PUSH_MPLS: {
3064                         const struct ovs_action_push_mpls *mpls = nla_data(a);
3065
3066                         if (!eth_p_mpls(mpls->mpls_ethertype))
3067                                 return -EINVAL;
3068                         /* Prohibit push MPLS other than to a white list
3069                          * for packets that have a known tag order.
3070                          */
3071                         if (vlan_tci & htons(VLAN_CFI_MASK) ||
3072                             (eth_type != htons(ETH_P_IP) &&
3073                              eth_type != htons(ETH_P_IPV6) &&
3074                              eth_type != htons(ETH_P_ARP) &&
3075                              eth_type != htons(ETH_P_RARP) &&
3076                              !eth_p_mpls(eth_type)))
3077                                 return -EINVAL;
3078                         eth_type = mpls->mpls_ethertype;
3079                         break;
3080                 }
3081
3082                 case OVS_ACTION_ATTR_POP_MPLS:
3083                         if (vlan_tci & htons(VLAN_CFI_MASK) ||
3084                             !eth_p_mpls(eth_type))
3085                                 return -EINVAL;
3086
3087                         /* Disallow subsequent L2.5+ set and mpls_pop actions
3088                          * as there is no check here to ensure that the new
3089                          * eth_type is valid and thus set actions could
3090                          * write off the end of the packet or otherwise
3091                          * corrupt it.
3092                          *
3093                          * Support for these actions is planned using packet
3094                          * recirculation.
3095                          */
3096                         eth_type = htons(0);
3097                         break;
3098
3099                 case OVS_ACTION_ATTR_SET:
3100                         err = validate_set(a, key, sfa,
3101                                            &skip_copy, mac_proto, eth_type,
3102                                            false, log);
3103                         if (err)
3104                                 return err;
3105                         break;
3106
3107                 case OVS_ACTION_ATTR_SET_MASKED:
3108                         err = validate_set(a, key, sfa,
3109                                            &skip_copy, mac_proto, eth_type,
3110                                            true, log);
3111                         if (err)
3112                                 return err;
3113                         break;
3114
3115                 case OVS_ACTION_ATTR_SAMPLE: {
3116                         bool last = nla_is_last(a, rem);
3117
3118                         err = validate_and_copy_sample(net, a, key, sfa,
3119                                                        eth_type, vlan_tci,
3120                                                        log, last);
3121                         if (err)
3122                                 return err;
3123                         skip_copy = true;
3124                         break;
3125                 }
3126
3127                 case OVS_ACTION_ATTR_CT:
3128                         err = ovs_ct_copy_action(net, a, key, sfa, log);
3129                         if (err)
3130                                 return err;
3131                         skip_copy = true;
3132                         break;
3133
3134                 case OVS_ACTION_ATTR_CT_CLEAR:
3135                         break;
3136
3137                 case OVS_ACTION_ATTR_PUSH_ETH:
3138                         /* Disallow pushing an Ethernet header if one
3139                          * is already present */
3140                         if (mac_proto != MAC_PROTO_NONE)
3141                                 return -EINVAL;
3142                         mac_proto = MAC_PROTO_ETHERNET;
3143                         break;
3144
3145                 case OVS_ACTION_ATTR_POP_ETH:
3146                         if (mac_proto != MAC_PROTO_ETHERNET)
3147                                 return -EINVAL;
3148                         if (vlan_tci & htons(VLAN_CFI_MASK))
3149                                 return -EINVAL;
3150                         mac_proto = MAC_PROTO_NONE;
3151                         break;
3152
3153                 case OVS_ACTION_ATTR_PUSH_NSH:
3154                         if (mac_proto != MAC_PROTO_ETHERNET) {
3155                                 u8 next_proto;
3156
3157                                 next_proto = tun_p_from_eth_p(eth_type);
3158                                 if (!next_proto)
3159                                         return -EINVAL;
3160                         }
3161                         mac_proto = MAC_PROTO_NONE;
3162                         if (!validate_nsh(nla_data(a), false, true, true))
3163                                 return -EINVAL;
3164                         break;
3165
3166                 case OVS_ACTION_ATTR_POP_NSH: {
3167                         __be16 inner_proto;
3168
3169                         if (eth_type != htons(ETH_P_NSH))
3170                                 return -EINVAL;
3171                         inner_proto = tun_p_to_eth_p(key->nsh.base.np);
3172                         if (!inner_proto)
3173                                 return -EINVAL;
3174                         if (key->nsh.base.np == TUN_P_ETHERNET)
3175                                 mac_proto = MAC_PROTO_ETHERNET;
3176                         else
3177                                 mac_proto = MAC_PROTO_NONE;
3178                         break;
3179                 }
3180
3181                 case OVS_ACTION_ATTR_METER:
3182                         /* Non-existent meters are simply ignored.  */
3183                         break;
3184
3185                 case OVS_ACTION_ATTR_CLONE: {
3186                         bool last = nla_is_last(a, rem);
3187
3188                         err = validate_and_copy_clone(net, a, key, sfa,
3189                                                       eth_type, vlan_tci,
3190                                                       log, last);
3191                         if (err)
3192                                 return err;
3193                         skip_copy = true;
3194                         break;
3195                 }
3196
3197                 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
3198                         bool last = nla_is_last(a, rem);
3199
3200                         err = validate_and_copy_check_pkt_len(net, a, key, sfa,
3201                                                               eth_type,
3202                                                               vlan_tci, log,
3203                                                               last);
3204                         if (err)
3205                                 return err;
3206                         skip_copy = true;
3207                         break;
3208                 }
3209
3210                 default:
3211                         OVS_NLERR(log, "Unknown Action type %d", type);
3212                         return -EINVAL;
3213                 }
3214                 if (!skip_copy) {
3215                         err = copy_action(a, sfa, log);
3216                         if (err)
3217                                 return err;
3218                 }
3219         }
3220
3221         if (rem > 0)
3222                 return -EINVAL;
3223
3224         return 0;
3225 }
3226
3227 /* 'key' must be the masked key. */
3228 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3229                          const struct sw_flow_key *key,
3230                          struct sw_flow_actions **sfa, bool log)
3231 {
3232         int err;
3233
3234         *sfa = nla_alloc_flow_actions(min(nla_len(attr), MAX_ACTIONS_BUFSIZE));
3235         if (IS_ERR(*sfa))
3236                 return PTR_ERR(*sfa);
3237
3238         (*sfa)->orig_len = nla_len(attr);
3239         err = __ovs_nla_copy_actions(net, attr, key, sfa, key->eth.type,
3240                                      key->eth.vlan.tci, log);
3241         if (err)
3242                 ovs_nla_free_flow_actions(*sfa);
3243
3244         return err;
3245 }
3246
3247 static int sample_action_to_attr(const struct nlattr *attr,
3248                                  struct sk_buff *skb)
3249 {
3250         struct nlattr *start, *ac_start = NULL, *sample_arg;
3251         int err = 0, rem = nla_len(attr);
3252         const struct sample_arg *arg;
3253         struct nlattr *actions;
3254
3255         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
3256         if (!start)
3257                 return -EMSGSIZE;
3258
3259         sample_arg = nla_data(attr);
3260         arg = nla_data(sample_arg);
3261         actions = nla_next(sample_arg, &rem);
3262
3263         if (nla_put_u32(skb, OVS_SAMPLE_ATTR_PROBABILITY, arg->probability)) {
3264                 err = -EMSGSIZE;
3265                 goto out;
3266         }
3267
3268         ac_start = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
3269         if (!ac_start) {
3270                 err = -EMSGSIZE;
3271                 goto out;
3272         }
3273
3274         err = ovs_nla_put_actions(actions, rem, skb);
3275
3276 out:
3277         if (err) {
3278                 nla_nest_cancel(skb, ac_start);
3279                 nla_nest_cancel(skb, start);
3280         } else {
3281                 nla_nest_end(skb, ac_start);
3282                 nla_nest_end(skb, start);
3283         }
3284
3285         return err;
3286 }
3287
3288 static int clone_action_to_attr(const struct nlattr *attr,
3289                                 struct sk_buff *skb)
3290 {
3291         struct nlattr *start;
3292         int err = 0, rem = nla_len(attr);
3293
3294         start = nla_nest_start(skb, OVS_ACTION_ATTR_CLONE);
3295         if (!start)
3296                 return -EMSGSIZE;
3297
3298         err = ovs_nla_put_actions(nla_data(attr), rem, skb);
3299
3300         if (err)
3301                 nla_nest_cancel(skb, start);
3302         else
3303                 nla_nest_end(skb, start);
3304
3305         return err;
3306 }
3307
3308 static int check_pkt_len_action_to_attr(const struct nlattr *attr,
3309                                         struct sk_buff *skb)
3310 {
3311         struct nlattr *start, *ac_start = NULL;
3312         const struct check_pkt_len_arg *arg;
3313         const struct nlattr *a, *cpl_arg;
3314         int err = 0, rem = nla_len(attr);
3315
3316         start = nla_nest_start(skb, OVS_ACTION_ATTR_CHECK_PKT_LEN);
3317         if (!start)
3318                 return -EMSGSIZE;
3319
3320         /* The first nested attribute in 'attr' is always
3321          * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
3322          */
3323         cpl_arg = nla_data(attr);
3324         arg = nla_data(cpl_arg);
3325
3326         if (nla_put_u16(skb, OVS_CHECK_PKT_LEN_ATTR_PKT_LEN, arg->pkt_len)) {
3327                 err = -EMSGSIZE;
3328                 goto out;
3329         }
3330
3331         /* Second nested attribute in 'attr' is always
3332          * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
3333          */
3334         a = nla_next(cpl_arg, &rem);
3335         ac_start =  nla_nest_start(skb,
3336                 OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL);
3337         if (!ac_start) {
3338                 err = -EMSGSIZE;
3339                 goto out;
3340         }
3341
3342         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3343         if (err) {
3344                 nla_nest_cancel(skb, ac_start);
3345                 goto out;
3346         } else {
3347                 nla_nest_end(skb, ac_start);
3348         }
3349
3350         /* Third nested attribute in 'attr' is always
3351          * OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER.
3352          */
3353         a = nla_next(a, &rem);
3354         ac_start =  nla_nest_start(skb,
3355                                    OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER);
3356         if (!ac_start) {
3357                 err = -EMSGSIZE;
3358                 goto out;
3359         }
3360
3361         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3362         if (err) {
3363                 nla_nest_cancel(skb, ac_start);
3364                 goto out;
3365         } else {
3366                 nla_nest_end(skb, ac_start);
3367         }
3368
3369         nla_nest_end(skb, start);
3370         return 0;
3371
3372 out:
3373         nla_nest_cancel(skb, start);
3374         return err;
3375 }
3376
3377 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
3378 {
3379         const struct nlattr *ovs_key = nla_data(a);
3380         int key_type = nla_type(ovs_key);
3381         struct nlattr *start;
3382         int err;
3383
3384         switch (key_type) {
3385         case OVS_KEY_ATTR_TUNNEL_INFO: {
3386                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
3387                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
3388
3389                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
3390                 if (!start)
3391                         return -EMSGSIZE;
3392
3393                 err =  ip_tun_to_nlattr(skb, &tun_info->key,
3394                                         ip_tunnel_info_opts(tun_info),
3395                                         tun_info->options_len,
3396                                         ip_tunnel_info_af(tun_info), tun_info->mode);
3397                 if (err)
3398                         return err;
3399                 nla_nest_end(skb, start);
3400                 break;
3401         }
3402         default:
3403                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
3404                         return -EMSGSIZE;
3405                 break;
3406         }
3407
3408         return 0;
3409 }
3410
3411 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
3412                                                 struct sk_buff *skb)
3413 {
3414         const struct nlattr *ovs_key = nla_data(a);
3415         struct nlattr *nla;
3416         size_t key_len = nla_len(ovs_key) / 2;
3417
3418         /* Revert the conversion we did from a non-masked set action to
3419          * masked set action.
3420          */
3421         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
3422         if (!nla)
3423                 return -EMSGSIZE;
3424
3425         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
3426                 return -EMSGSIZE;
3427
3428         nla_nest_end(skb, nla);
3429         return 0;
3430 }
3431
3432 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
3433 {
3434         const struct nlattr *a;
3435         int rem, err;
3436
3437         nla_for_each_attr(a, attr, len, rem) {
3438                 int type = nla_type(a);
3439
3440                 switch (type) {
3441                 case OVS_ACTION_ATTR_SET:
3442                         err = set_action_to_attr(a, skb);
3443                         if (err)
3444                                 return err;
3445                         break;
3446
3447                 case OVS_ACTION_ATTR_SET_TO_MASKED:
3448                         err = masked_set_action_to_set_action_attr(a, skb);
3449                         if (err)
3450                                 return err;
3451                         break;
3452
3453                 case OVS_ACTION_ATTR_SAMPLE:
3454                         err = sample_action_to_attr(a, skb);
3455                         if (err)
3456                                 return err;
3457                         break;
3458
3459                 case OVS_ACTION_ATTR_CT:
3460                         err = ovs_ct_action_to_attr(nla_data(a), skb);
3461                         if (err)
3462                                 return err;
3463                         break;
3464
3465                 case OVS_ACTION_ATTR_CLONE:
3466                         err = clone_action_to_attr(a, skb);
3467                         if (err)
3468                                 return err;
3469                         break;
3470
3471                 case OVS_ACTION_ATTR_CHECK_PKT_LEN:
3472                         err = check_pkt_len_action_to_attr(a, skb);
3473                         if (err)
3474                                 return err;
3475                         break;
3476
3477                 default:
3478                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
3479                                 return -EMSGSIZE;
3480                         break;
3481                 }
3482         }
3483
3484         return 0;
3485 }