datapath/flow_netlink.c

   1 /*
   2  * Copyright (c) 2007-2015 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/uaccess.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <net/llc_pdu.h>
  27 #include <linux/kernel.h>
  28 #include <linux/jhash.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/llc.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/rcupdate.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/sctp.h>
  38 #include <linux/tcp.h>
  39 #include <linux/udp.h>
  40 #include <linux/icmp.h>
  41 #include <linux/icmpv6.h>
  42 #include <linux/rculist.h>
  43 #include <net/geneve.h>
  44 #include <net/ip.h>
  45 #include <net/ipv6.h>
  46 #include <net/ndisc.h>
  47 #include <net/mpls.h>
  48 #include <net/vxlan.h>
  49
  50 #include "datapath.h"
  51 #include "conntrack.h"
  52 #include "flow.h"
  53 #include "flow_netlink.h"
  54 #include "gso.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 void update_range(struct sw_flow_match *match,
  65                          size_t offset, size_t size, bool is_mask)
  66 {
  67         struct sw_flow_key_range *range;
  68         size_t start = rounddown(offset, sizeof(long));
  69         size_t end = roundup(offset + size, sizeof(long));
  70
  71         if (!is_mask)
  72                 range = &match->range;
  73         else
  74                 range = &match->mask->range;
  75
  76         if (range->start == range->end) {
  77                 range->start = start;
  78                 range->end = end;
  79                 return;
  80         }
  81
  82         if (range->start > start)
  83                 range->start = start;
  84
  85         if (range->end < end)
  86                 range->end = end;
  87 }
  88
  89 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  90         do { \
  91                 update_range(match, offsetof(struct sw_flow_key, field),    \
  92                              sizeof((match)->key->field), is_mask);         \
  93                 if (is_mask)                                                \
  94                         (match)->mask->key.field = value;                   \
  95                 else                                                        \
  96                         (match)->key->field = value;                        \
  97         } while (0)
  98
  99 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
 100         do {                                                                \
 101                 update_range(match, offset, len, is_mask);                  \
 102                 if (is_mask)                                                \
 103                         memcpy((u8 *)&(match)->mask->key + offset, value_p, len);\
 104                 else                                                        \
 105                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
 106         } while (0)
 107
 108 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
 109         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 110                                   value_p, len, is_mask)
 111
 112 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 113         do {                                                                \
 114                 update_range(match, offsetof(struct sw_flow_key, field),    \
 115                              sizeof((match)->key->field), is_mask);         \
 116                 if (is_mask)                                                \
 117                         memset((u8 *)&(match)->mask->key.field, value,      \
 118                                sizeof((match)->mask->key.field));           \
 119                 else                                                        \
 120                         memset((u8 *)&(match)->key->field, value,           \
 121                                sizeof((match)->key->field));                \
 122         } while (0)
 123
 124 static bool match_validate(const struct sw_flow_match *match,
 125                            u64 key_attrs, u64 mask_attrs, bool log)
 126 {
 127         u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
 128         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 129
 130         /* The following mask attributes allowed only if they
 131          * pass the validation tests.
 132          */
 133         mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
 134                         | (1ULL << OVS_KEY_ATTR_IPV6)
 135                         | (1ULL << OVS_KEY_ATTR_TCP)
 136                         | (1ULL << OVS_KEY_ATTR_TCP_FLAGS)
 137                         | (1ULL << OVS_KEY_ATTR_UDP)
 138                         | (1ULL << OVS_KEY_ATTR_SCTP)
 139                         | (1ULL << OVS_KEY_ATTR_ICMP)
 140                         | (1ULL << OVS_KEY_ATTR_ICMPV6)
 141                         | (1ULL << OVS_KEY_ATTR_ARP)
 142                         | (1ULL << OVS_KEY_ATTR_ND)
 143                         | (1ULL << OVS_KEY_ATTR_MPLS));
 144
 145         /* Always allowed mask fields. */
 146         mask_allowed |= ((1ULL << OVS_KEY_ATTR_TUNNEL)
 147                        | (1ULL << OVS_KEY_ATTR_IN_PORT)
 148                        | (1ULL << OVS_KEY_ATTR_ETHERTYPE));
 149
 150         /* Check key attributes. */
 151         if (match->key->eth.type == htons(ETH_P_ARP)
 152                         || match->key->eth.type == htons(ETH_P_RARP)) {
 153                 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
 154                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 155                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
 156         }
 157
 158         if (eth_p_mpls(match->key->eth.type)) {
 159                 key_expected |= 1ULL << OVS_KEY_ATTR_MPLS;
 160                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 161                         mask_allowed |= 1ULL << OVS_KEY_ATTR_MPLS;
 162         }
 163
 164         if (match->key->eth.type == htons(ETH_P_IP)) {
 165                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
 166                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 167                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
 168
 169                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 170                         if (match->key->ip.proto == IPPROTO_UDP) {
 171                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 172                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 173                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 174                         }
 175
 176                         if (match->key->ip.proto == IPPROTO_SCTP) {
 177                                 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
 178                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 179                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
 180                         }
 181
 182                         if (match->key->ip.proto == IPPROTO_TCP) {
 183                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 184                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 185                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 186                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 187                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 188                                 }
 189                         }
 190
 191                         if (match->key->ip.proto == IPPROTO_ICMP) {
 192                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
 193                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 194                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
 195                         }
 196                 }
 197         }
 198
 199         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 200                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
 201                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 202                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
 203
 204                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 205                         if (match->key->ip.proto == IPPROTO_UDP) {
 206                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 207                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 208                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 209                         }
 210
 211                         if (match->key->ip.proto == IPPROTO_SCTP) {
 212                                 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
 213                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 214                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
 215                         }
 216
 217                         if (match->key->ip.proto == IPPROTO_TCP) {
 218                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 219                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 220                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 221                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 222                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
 223                                 }
 224                         }
 225
 226                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 227                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 228                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 229                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 230
 231                                 if (match->key->tp.src ==
 232                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 233                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 234                                         key_expected |= 1ULL << OVS_KEY_ATTR_ND;
 235                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 236                                                 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
 237                                 }
 238                         }
 239                 }
 240         }
 241
 242         if ((key_attrs & key_expected) != key_expected) {
 243                 /* Key attributes check failed. */
 244                 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
 245                           (unsigned long long)key_attrs,
 246                           (unsigned long long)key_expected);
 247                 return false;
 248         }
 249
 250         if ((mask_attrs & mask_allowed) != mask_attrs) {
 251                 /* Mask attributes check failed. */
 252                 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
 253                           (unsigned long long)mask_attrs,
 254                           (unsigned long long)mask_allowed);
 255                 return false;
 256         }
 257
 258         return true;
 259 }
 260
 261 size_t ovs_tun_key_attr_size(void)
 262 {
 263         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 264          * updating this function.
 265          */
 266         return    nla_total_size(8)    /* OVS_TUNNEL_KEY_ATTR_ID */
 267                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 268                 + nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 269                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
 270                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
 271                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 272                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
 273                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
 274                 + nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 275                 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
 276                  * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
 277                  */
 278                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
 279                 + nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
 280 }
 281
 282 size_t ovs_key_attr_size(void)
 283 {
 284         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 285          * updating this function.
 286          */
 287         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 26);
 288
 289         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 290                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 291                   + ovs_tun_key_attr_size()
 292                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 293                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 294                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 295                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 296                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
 297                 + nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
 298                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
 299                 + nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
 300                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 301                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 302                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 303                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 304                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 305                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 306                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 307                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 308 }
 309
 310 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
 311         [OVS_VXLAN_EXT_GBP]         = { .len = sizeof(u32) },
 312 };
 313
 314 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 315         [OVS_TUNNEL_KEY_ATTR_ID]            = { .len = sizeof(u64) },
 316         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC]      = { .len = sizeof(u32) },
 317         [OVS_TUNNEL_KEY_ATTR_IPV4_DST]      = { .len = sizeof(u32) },
 318         [OVS_TUNNEL_KEY_ATTR_TOS]           = { .len = 1 },
 319         [OVS_TUNNEL_KEY_ATTR_TTL]           = { .len = 1 },
 320         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
 321         [OVS_TUNNEL_KEY_ATTR_CSUM]          = { .len = 0 },
 322         [OVS_TUNNEL_KEY_ATTR_TP_SRC]        = { .len = sizeof(u16) },
 323         [OVS_TUNNEL_KEY_ATTR_TP_DST]        = { .len = sizeof(u16) },
 324         [OVS_TUNNEL_KEY_ATTR_OAM]           = { .len = 0 },
 325         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
 326         [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
 327                                                 .next = ovs_vxlan_ext_key_lens },
 328 };
 329
 330 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 331 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 332         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 333         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 334         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 335         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 336         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 337         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 338         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 339         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 340         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 341         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 342         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 343         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 344         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 345         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 346         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 347         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 348         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 349         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 350         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 351         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 352                                      .next = ovs_tunnel_key_lens, },
 353         [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 354         [OVS_KEY_ATTR_CT_STATE]  = { .len = sizeof(u32) },
 355         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 356         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 357         [OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
 358 };
 359
 360 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
 361 {
 362         return expected_len == attr_len ||
 363                expected_len == OVS_ATTR_NESTED ||
 364                expected_len == OVS_ATTR_VARIABLE;
 365 }
 366
 367 static bool is_all_zero(const u8 *fp, size_t size)
 368 {
 369         int i;
 370
 371         if (!fp)
 372                 return false;
 373
 374         for (i = 0; i < size; i++)
 375                 if (fp[i])
 376                         return false;
 377
 378         return true;
 379 }
 380
 381 static int __parse_flow_nlattrs(const struct nlattr *attr,
 382                                 const struct nlattr *a[],
 383                                 u64 *attrsp, bool log, bool nz)
 384 {
 385         const struct nlattr *nla;
 386         u64 attrs;
 387         int rem;
 388
 389         attrs = *attrsp;
 390         nla_for_each_nested(nla, attr, rem) {
 391                 u16 type = nla_type(nla);
 392                 int expected_len;
 393
 394                 if (type > OVS_KEY_ATTR_MAX) {
 395                         OVS_NLERR(log, "Key type %d is out of range max %d",
 396                                   type, OVS_KEY_ATTR_MAX);
 397                         return -EINVAL;
 398                 }
 399
 400                 if (attrs & (1ULL << type)) {
 401                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 402                         return -EINVAL;
 403                 }
 404
 405                 expected_len = ovs_key_lens[type].len;
 406                 if (!check_attr_len(nla_len(nla), expected_len)) {
 407                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 408                                   type, nla_len(nla), expected_len);
 409                         return -EINVAL;
 410                 }
 411
 412                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 413                         attrs |= 1ULL << type;
 414                         a[type] = nla;
 415                 }
 416         }
 417         if (rem) {
 418                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 419                 return -EINVAL;
 420         }
 421
 422         *attrsp = attrs;
 423         return 0;
 424 }
 425
 426 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 427                                    const struct nlattr *a[], u64 *attrsp,
 428                                    bool log)
 429 {
 430         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 431 }
 432
 433 static int parse_flow_nlattrs(const struct nlattr *attr,
 434                               const struct nlattr *a[], u64 *attrsp,
 435                               bool log)
 436 {
 437         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 438 }
 439
 440 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 441                                      struct sw_flow_match *match, bool is_mask,
 442                                      bool log)
 443 {
 444         unsigned long opt_key_offset;
 445
 446         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 447                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 448                           nla_len(a), sizeof(match->key->tun_opts));
 449                 return -EINVAL;
 450         }
 451
 452         if (nla_len(a) % 4 != 0) {
 453                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 454                           nla_len(a));
 455                 return -EINVAL;
 456         }
 457
 458         /* We need to record the length of the options passed
 459          * down, otherwise packets with the same format but
 460          * additional options will be silently matched.
 461          */
 462         if (!is_mask) {
 463                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 464                                 false);
 465         } else {
 466                 /* This is somewhat unusual because it looks at
 467                  * both the key and mask while parsing the
 468                  * attributes (and by extension assumes the key
 469                  * is parsed first). Normally, we would verify
 470                  * that each is the correct length and that the
 471                  * attributes line up in the validate function.
 472                  * However, that is difficult because this is
 473                  * variable length and we won't have the
 474                  * information later.
 475                  */
 476                 if (match->key->tun_opts_len != nla_len(a)) {
 477                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 478                                   match->key->tun_opts_len, nla_len(a));
 479                         return -EINVAL;
 480                 }
 481
 482                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 483         }
 484
 485         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 486         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 487                                   nla_len(a), is_mask);
 488         return 0;
 489 }
 490
 491 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
 492                                      struct sw_flow_match *match, bool is_mask,
 493                                      bool log)
 494 {
 495         struct nlattr *a;
 496         int rem;
 497         unsigned long opt_key_offset;
 498         struct vxlan_metadata opts;
 499
 500         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 501
 502         memset(&opts, 0, sizeof(opts));
 503         nla_for_each_nested(a, attr, rem) {
 504                 int type = nla_type(a);
 505
 506                 if (type > OVS_VXLAN_EXT_MAX) {
 507                         OVS_NLERR(log, "VXLAN extension %d out of range max %d",
 508                                   type, OVS_VXLAN_EXT_MAX);
 509                         return -EINVAL;
 510                 }
 511
 512                 if (!check_attr_len(nla_len(a),
 513                                     ovs_vxlan_ext_key_lens[type].len)) {
 514                         OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
 515                                   type, nla_len(a),
 516                                   ovs_vxlan_ext_key_lens[type].len);
 517                         return -EINVAL;
 518                 }
 519
 520                 switch (type) {
 521                 case OVS_VXLAN_EXT_GBP:
 522                         opts.gbp = nla_get_u32(a);
 523                         break;
 524                 default:
 525                         OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
 526                                   type);
 527                         return -EINVAL;
 528                 }
 529         }
 530         if (rem) {
 531                 OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
 532                           rem);
 533                 return -EINVAL;
 534         }
 535
 536         if (!is_mask)
 537                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 538         else
 539                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 540
 541         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 542         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 543                                   is_mask);
 544         return 0;
 545 }
 546
 547 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
 548                                 struct sw_flow_match *match, bool is_mask,
 549                                 bool log)
 550 {
 551         struct nlattr *a;
 552         int rem;
 553         bool ttl = false;
 554         __be16 tun_flags = 0;
 555         int opts_type = 0;
 556
 557         nla_for_each_nested(a, attr, rem) {
 558                 int type = nla_type(a);
 559                 int err;
 560
 561                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 562                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 563                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 564                         return -EINVAL;
 565                 }
 566
 567                 if (!check_attr_len(nla_len(a),
 568                                     ovs_tunnel_key_lens[type].len)) {
 569                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 570                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 571                         return -EINVAL;
 572                 }
 573
 574                 switch (type) {
 575                 case OVS_TUNNEL_KEY_ATTR_ID:
 576                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 577                                         nla_get_be64(a), is_mask);
 578                         tun_flags |= TUNNEL_KEY;
 579                         break;
 580                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 581                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
 582                                         nla_get_in_addr(a), is_mask);
 583                         break;
 584                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 585                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
 586                                         nla_get_in_addr(a), is_mask);
 587                         break;
 588                 case OVS_TUNNEL_KEY_ATTR_TOS:
 589                         SW_FLOW_KEY_PUT(match, tun_key.tos,
 590                                         nla_get_u8(a), is_mask);
 591                         break;
 592                 case OVS_TUNNEL_KEY_ATTR_TTL:
 593                         SW_FLOW_KEY_PUT(match, tun_key.ttl,
 594                                         nla_get_u8(a), is_mask);
 595                         ttl = true;
 596                         break;
 597                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 598                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 599                         break;
 600                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 601                         tun_flags |= TUNNEL_CSUM;
 602                         break;
 603                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 604                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 605                                         nla_get_be16(a), is_mask);
 606                         break;
 607                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 608                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 609                                         nla_get_be16(a), is_mask);
 610                         break;
 611                 case OVS_TUNNEL_KEY_ATTR_OAM:
 612                         tun_flags |= TUNNEL_OAM;
 613                         break;
 614                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 615                         if (opts_type) {
 616                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 617                                 return -EINVAL;
 618                         }
 619
 620                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 621                         if (err)
 622                                 return err;
 623
 624                         tun_flags |= TUNNEL_GENEVE_OPT;
 625                         opts_type = type;
 626                         break;
 627                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 628                         if (opts_type) {
 629                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 630                                 return -EINVAL;
 631                         }
 632
 633                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 634                         if (err)
 635                                 return err;
 636
 637                         tun_flags |= TUNNEL_VXLAN_OPT;
 638                         opts_type = type;
 639                         break;
 640                 default:
 641                         OVS_NLERR(log, "Unknown IPv4 tunnel attribute %d",
 642                                   type);
 643                         return -EINVAL;
 644                 }
 645         }
 646
 647         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 648
 649         if (rem > 0) {
 650                 OVS_NLERR(log, "IPv4 tunnel attribute has %d unknown bytes.",
 651                           rem);
 652                 return -EINVAL;
 653         }
 654
 655         if (!is_mask) {
 656                 if (!match->key->tun_key.u.ipv4.dst) {
 657                         OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 658                         return -EINVAL;
 659                 }
 660
 661                 if (!ttl) {
 662                         OVS_NLERR(log, "IPv4 tunnel TTL not specified.");
 663                         return -EINVAL;
 664                 }
 665         }
 666
 667         return opts_type;
 668 }
 669
 670 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 671                                const void *tun_opts, int swkey_tun_opts_len)
 672 {
 673         const struct vxlan_metadata *opts = tun_opts;
 674         struct nlattr *nla;
 675
 676         nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 677         if (!nla)
 678                 return -EMSGSIZE;
 679
 680         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 681                 return -EMSGSIZE;
 682
 683         nla_nest_end(skb, nla);
 684         return 0;
 685 }
 686
 687 static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
 688                                 const struct ip_tunnel_key *output,
 689                                 const void *tun_opts, int swkey_tun_opts_len)
 690 {
 691         if (output->tun_flags & TUNNEL_KEY &&
 692             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 693                 return -EMSGSIZE;
 694         if (output->u.ipv4.src &&
 695             nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 696                             output->u.ipv4.src))
 697                 return -EMSGSIZE;
 698         if (output->u.ipv4.dst &&
 699             nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 700                             output->u.ipv4.dst))
 701                 return -EMSGSIZE;
 702         if (output->tos &&
 703             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 704                 return -EMSGSIZE;
 705         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 706                 return -EMSGSIZE;
 707         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 708             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 709                 return -EMSGSIZE;
 710         if ((output->tun_flags & TUNNEL_CSUM) &&
 711             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 712                 return -EMSGSIZE;
 713         if (output->tp_src &&
 714             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 715                 return -EMSGSIZE;
 716         if (output->tp_dst &&
 717             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 718                 return -EMSGSIZE;
 719         if ((output->tun_flags & TUNNEL_OAM) &&
 720             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 721                 return -EMSGSIZE;
 722         if (tun_opts) {
 723                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 724                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 725                             swkey_tun_opts_len, tun_opts))
 726                         return -EMSGSIZE;
 727                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 728                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 729                         return -EMSGSIZE;
 730         }
 731
 732         return 0;
 733 }
 734
 735 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
 736                               const struct ip_tunnel_key *output,
 737                               const void *tun_opts, int swkey_tun_opts_len)
 738 {
 739         struct nlattr *nla;
 740         int err;
 741
 742         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 743         if (!nla)
 744                 return -EMSGSIZE;
 745
 746         err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
 747         if (err)
 748                 return err;
 749
 750         nla_nest_end(skb, nla);
 751         return 0;
 752 }
 753
 754 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
 755                                   const struct ip_tunnel_info *egress_tun_info,
 756                                   const void *egress_tun_opts)
 757 {
 758         return __ipv4_tun_to_nlattr(skb, &egress_tun_info->key,
 759                                     egress_tun_opts,
 760                                     egress_tun_info->options_len);
 761 }
 762
 763 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
 764                                   u64 *attrs, const struct nlattr **a,
 765                                  bool is_mask, bool log)
 766 {
 767         if (*attrs & (1ULL << OVS_KEY_ATTR_DP_HASH)) {
 768                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 769
 770                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 771                 *attrs &= ~(1ULL << OVS_KEY_ATTR_DP_HASH);
 772         }
 773
 774         if (*attrs & (1ULL << OVS_KEY_ATTR_RECIRC_ID)) {
 775                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 776
 777                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 778                 *attrs &= ~(1ULL << OVS_KEY_ATTR_RECIRC_ID);
 779         }
 780
 781         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
 782                 SW_FLOW_KEY_PUT(match, phy.priority,
 783                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 784                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
 785         }
 786
 787         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
 788                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 789
 790                 if (is_mask) {
 791                         in_port = 0xffffffff; /* Always exact match in_port. */
 792                 } else if (in_port >= DP_MAX_PORTS) {
 793                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 794                                   in_port, DP_MAX_PORTS);
 795                         return -EINVAL;
 796                 }
 797
 798                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 799                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
 800         } else if (!is_mask) {
 801                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 802         }
 803
 804         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
 805                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 806
 807                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 808                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
 809         }
 810         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
 811                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 812                                          is_mask, log) < 0)
 813                         return -EINVAL;
 814                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
 815         }
 816
 817         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
 818             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
 819                 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
 820
 821                 if (ct_state & ~CT_SUPPORTED_MASK) {
 822                         OVS_NLERR(log, "ct_state flags %08x unsupported",
 823                                   ct_state);
 824                         return -EINVAL;
 825                 }
 826
 827                 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
 828                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
 829         }
 830         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
 831             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
 832                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
 833
 834                 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
 835                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
 836         }
 837         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
 838             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
 839                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
 840
 841                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
 842                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
 843         }
 844         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
 845             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
 846                 const struct ovs_key_ct_labels *cl;
 847
 848                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
 849                 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
 850                                    sizeof(*cl), is_mask);
 851                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
 852         }
 853         return 0;
 854 }
 855
 856 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
 857                                 u64 attrs, const struct nlattr **a,
 858                                 bool is_mask, bool log)
 859 {
 860         int err;
 861
 862         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
 863         if (err)
 864                 return err;
 865
 866         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
 867                 const struct ovs_key_ethernet *eth_key;
 868
 869                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 870                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 871                                 eth_key->eth_src, ETH_ALEN, is_mask);
 872                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 873                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 874                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
 875         }
 876
 877         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
 878                 __be16 tci;
 879
 880                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 881                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 882                         if (is_mask)
 883                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 884                         else
 885                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 886
 887                         return -EINVAL;
 888                 }
 889
 890                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 891                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
 892         }
 893
 894         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
 895                 __be16 eth_type;
 896
 897                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 898                 if (is_mask) {
 899                         /* Always exact match EtherType. */
 900                         eth_type = htons(0xffff);
 901                 } else if (!eth_proto_is_802_3(eth_type)) {
 902                         OVS_NLERR(log, "EtherType %x is less than min %x",
 903                                   ntohs(eth_type), ETH_P_802_3_MIN);
 904                         return -EINVAL;
 905                 }
 906
 907                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 908                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
 909         } else if (!is_mask) {
 910                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 911         }
 912
 913         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
 914                 const struct ovs_key_ipv4 *ipv4_key;
 915
 916                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 917                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 918                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 919                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 920                         return -EINVAL;
 921                 }
 922                 SW_FLOW_KEY_PUT(match, ip.proto,
 923                                 ipv4_key->ipv4_proto, is_mask);
 924                 SW_FLOW_KEY_PUT(match, ip.tos,
 925                                 ipv4_key->ipv4_tos, is_mask);
 926                 SW_FLOW_KEY_PUT(match, ip.ttl,
 927                                 ipv4_key->ipv4_ttl, is_mask);
 928                 SW_FLOW_KEY_PUT(match, ip.frag,
 929                                 ipv4_key->ipv4_frag, is_mask);
 930                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 931                                 ipv4_key->ipv4_src, is_mask);
 932                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 933                                 ipv4_key->ipv4_dst, is_mask);
 934                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
 935         }
 936
 937         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
 938                 const struct ovs_key_ipv6 *ipv6_key;
 939
 940                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 941                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 942                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 943                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 944                         return -EINVAL;
 945                 }
 946
 947                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 948                         OVS_NLERR(log,
 949                                   "Invalid IPv6 flow label value (value=%x, max=%x).",
 950                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 951                         return -EINVAL;
 952                 }
 953
 954                 SW_FLOW_KEY_PUT(match, ipv6.label,
 955                                 ipv6_key->ipv6_label, is_mask);
 956                 SW_FLOW_KEY_PUT(match, ip.proto,
 957                                 ipv6_key->ipv6_proto, is_mask);
 958                 SW_FLOW_KEY_PUT(match, ip.tos,
 959                                 ipv6_key->ipv6_tclass, is_mask);
 960                 SW_FLOW_KEY_PUT(match, ip.ttl,
 961                                 ipv6_key->ipv6_hlimit, is_mask);
 962                 SW_FLOW_KEY_PUT(match, ip.frag,
 963                                 ipv6_key->ipv6_frag, is_mask);
 964                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
 965                                 ipv6_key->ipv6_src,
 966                                 sizeof(match->key->ipv6.addr.src),
 967                                 is_mask);
 968                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
 969                                 ipv6_key->ipv6_dst,
 970                                 sizeof(match->key->ipv6.addr.dst),
 971                                 is_mask);
 972
 973                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
 974         }
 975
 976         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
 977                 const struct ovs_key_arp *arp_key;
 978
 979                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
 980                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
 981                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
 982                                   arp_key->arp_op);
 983                         return -EINVAL;
 984                 }
 985
 986                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 987                                 arp_key->arp_sip, is_mask);
 988                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 989                         arp_key->arp_tip, is_mask);
 990                 SW_FLOW_KEY_PUT(match, ip.proto,
 991                                 ntohs(arp_key->arp_op), is_mask);
 992                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
 993                                 arp_key->arp_sha, ETH_ALEN, is_mask);
 994                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
 995                                 arp_key->arp_tha, ETH_ALEN, is_mask);
 996
 997                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
 998         }
 999
1000         if (attrs & (1ULL << OVS_KEY_ATTR_MPLS)) {
1001                 const struct ovs_key_mpls *mpls_key;
1002
1003                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1004                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1005                                 mpls_key->mpls_lse, is_mask);
1006
1007                 attrs &= ~(1ULL << OVS_KEY_ATTR_MPLS);
1008         }
1009
1010         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1011                 const struct ovs_key_tcp *tcp_key;
1012
1013                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1014                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1015                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1016                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1017         }
1018
1019         if (attrs & (1ULL << OVS_KEY_ATTR_TCP_FLAGS)) {
1020                 SW_FLOW_KEY_PUT(match, tp.flags,
1021                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1022                                 is_mask);
1023                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP_FLAGS);
1024         }
1025
1026         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1027                 const struct ovs_key_udp *udp_key;
1028
1029                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1030                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1031                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1032                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1033         }
1034
1035         if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
1036                 const struct ovs_key_sctp *sctp_key;
1037
1038                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1039                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1040                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1041                 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
1042         }
1043
1044         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1045                 const struct ovs_key_icmp *icmp_key;
1046
1047                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1048                 SW_FLOW_KEY_PUT(match, tp.src,
1049                                 htons(icmp_key->icmp_type), is_mask);
1050                 SW_FLOW_KEY_PUT(match, tp.dst,
1051                                 htons(icmp_key->icmp_code), is_mask);
1052                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1053         }
1054
1055         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1056                 const struct ovs_key_icmpv6 *icmpv6_key;
1057
1058                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1059                 SW_FLOW_KEY_PUT(match, tp.src,
1060                                 htons(icmpv6_key->icmpv6_type), is_mask);
1061                 SW_FLOW_KEY_PUT(match, tp.dst,
1062                                 htons(icmpv6_key->icmpv6_code), is_mask);
1063                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1064         }
1065
1066         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1067                 const struct ovs_key_nd *nd_key;
1068
1069                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1070                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1071                         nd_key->nd_target,
1072                         sizeof(match->key->ipv6.nd.target),
1073                         is_mask);
1074                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1075                         nd_key->nd_sll, ETH_ALEN, is_mask);
1076                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1077                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1078                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1079         }
1080
1081         if (attrs != 0) {
1082                 OVS_NLERR(log, "Unknown key attributes %llx",
1083                           (unsigned long long)attrs);
1084                 return -EINVAL;
1085         }
1086
1087         return 0;
1088 }
1089
1090 static void nlattr_set(struct nlattr *attr, u8 val,
1091                        const struct ovs_len_tbl *tbl)
1092 {
1093         struct nlattr *nla;
1094         int rem;
1095
1096         /* The nlattr stream should already have been validated */
1097         nla_for_each_nested(nla, attr, rem) {
1098                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1099                         if (tbl[nla_type(nla)].next)
1100                                 tbl = tbl[nla_type(nla)].next;
1101                         nlattr_set(nla, val, tbl);
1102                 } else {
1103                         memset(nla_data(nla), val, nla_len(nla));
1104                 }
1105
1106                 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1107                         *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1108         }
1109 }
1110
1111 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1112 {
1113         nlattr_set(attr, val, ovs_key_lens);
1114 }
1115
1116 /**
1117  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1118  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1119  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1120  * does not include any don't care bit.
1121  * @net: Used to determine per-namespace field support.
1122  * @match: receives the extracted flow match information.
1123  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1124  * sequence. The fields should of the packet that triggered the creation
1125  * of this flow.
1126  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1127  * attribute specifies the mask field of the wildcarded flow.
1128  * @log: Boolean to allow kernel error logging.  Normally true, but when
1129  * probing for feature compatibility this should be passed in as false to
1130  * suppress unnecessary error logging.
1131  */
1132 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1133                       const struct nlattr *nla_key,
1134                       const struct nlattr *nla_mask,
1135                       bool log)
1136 {
1137         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1138         const struct nlattr *encap;
1139         struct nlattr *newmask = NULL;
1140         u64 key_attrs = 0;
1141         u64 mask_attrs = 0;
1142         bool encap_valid = false;
1143         int err;
1144
1145         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1146         if (err)
1147                 return err;
1148
1149         if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1150             (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
1151             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1152                 __be16 tci;
1153
1154                 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
1155                       (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
1156                         OVS_NLERR(log, "Invalid Vlan frame.");
1157                         return -EINVAL;
1158                 }
1159
1160                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1161                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1162                 encap = a[OVS_KEY_ATTR_ENCAP];
1163                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1164                 encap_valid = true;
1165
1166                 if (tci & htons(VLAN_TAG_PRESENT)) {
1167                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1168                         if (err)
1169                                 return err;
1170                 } else if (!tci) {
1171                         /* Corner case for truncated 802.1Q header. */
1172                         if (nla_len(encap)) {
1173                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1174                                 return -EINVAL;
1175                         }
1176                 } else {
1177                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1178                         return  -EINVAL;
1179                 }
1180         }
1181
1182         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1183         if (err)
1184                 return err;
1185
1186         if (match->mask) {
1187                 if (!nla_mask) {
1188                         /* Create an exact match mask. We need to set to 0xff
1189                          * all the 'match->mask' fields that have been touched
1190                          * in 'match->key'. We cannot simply memset
1191                          * 'match->mask', because padding bytes and fields not
1192                          * specified in 'match->key' should be left to 0.
1193                          * Instead, we use a stream of netlink attributes,
1194                          * copied from 'key' and set to 0xff.
1195                          * ovs_key_from_nlattrs() will take care of filling
1196                          * 'match->mask' appropriately.
1197                          */
1198                         newmask = kmemdup(nla_key,
1199                                           nla_total_size(nla_len(nla_key)),
1200                                           GFP_KERNEL);
1201                         if (!newmask)
1202                                 return -ENOMEM;
1203
1204                         mask_set_nlattr(newmask, 0xff);
1205
1206                         /* The userspace does not send tunnel attributes that
1207                          * are 0, but we should not wildcard them nonetheless.
1208                          */
1209                         if (match->key->tun_key.u.ipv4.dst)
1210                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1211                                                          0xff, true);
1212
1213                         nla_mask = newmask;
1214                 }
1215
1216                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1217                 if (err)
1218                         goto free_newmask;
1219
1220                 /* Always match on tci. */
1221                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1222
1223                 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1224                         __be16 eth_type = 0;
1225                         __be16 tci = 0;
1226
1227                         if (!encap_valid) {
1228                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1229                                 err = -EINVAL;
1230                                 goto free_newmask;
1231                         }
1232
1233                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1234                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1235                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1236
1237                         if (eth_type == htons(0xffff)) {
1238                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1239                                 encap = a[OVS_KEY_ATTR_ENCAP];
1240                                 err = parse_flow_mask_nlattrs(encap, a,
1241                                                               &mask_attrs, log);
1242                                 if (err)
1243                                         goto free_newmask;
1244                         } else {
1245                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1246                                           ntohs(eth_type));
1247                                 err = -EINVAL;
1248                                 goto free_newmask;
1249                         }
1250
1251                         if (a[OVS_KEY_ATTR_VLAN])
1252                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1253
1254                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1255                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1256                                           ntohs(tci));
1257                                 err = -EINVAL;
1258                                 goto free_newmask;
1259                         }
1260                 }
1261
1262                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1263                                            log);
1264                 if (err)
1265                         goto free_newmask;
1266         }
1267
1268         if (!match_validate(match, key_attrs, mask_attrs, log))
1269                 err = -EINVAL;
1270
1271 free_newmask:
1272         kfree(newmask);
1273         return err;
1274 }
1275
1276 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1277 {
1278         size_t len;
1279
1280         if (!attr)
1281                 return 0;
1282
1283         len = nla_len(attr);
1284         if (len < 1 || len > MAX_UFID_LENGTH) {
1285                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1286                           nla_len(attr), MAX_UFID_LENGTH);
1287                 return 0;
1288         }
1289
1290         return len;
1291 }
1292
1293 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1294  * or false otherwise.
1295  */
1296 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1297                       bool log)
1298 {
1299         sfid->ufid_len = get_ufid_len(attr, log);
1300         if (sfid->ufid_len)
1301                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1302
1303         return sfid->ufid_len;
1304 }
1305
1306 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1307                            const struct sw_flow_key *key, bool log)
1308 {
1309         struct sw_flow_key *new_key;
1310
1311         if (ovs_nla_get_ufid(sfid, ufid, log))
1312                 return 0;
1313
1314         /* If UFID was not provided, use unmasked key. */
1315         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1316         if (!new_key)
1317                 return -ENOMEM;
1318         memcpy(new_key, key, sizeof(*key));
1319         sfid->unmasked_key = new_key;
1320
1321         return 0;
1322 }
1323
1324 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1325 {
1326         return attr ? nla_get_u32(attr) : 0;
1327 }
1328
1329 /**
1330  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1331  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1332  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1333  * sequence.
1334  * @log: Boolean to allow kernel error logging.  Normally true, but when
1335  * probing for feature compatibility this should be passed in as false to
1336  * suppress unnecessary error logging.
1337  *
1338  * This parses a series of Netlink attributes that form a flow key, which must
1339  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1340  * get the metadata, that is, the parts of the flow key that cannot be
1341  * extracted from the packet itself.
1342  */
1343
1344 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1345                               struct sw_flow_key *key,
1346                               bool log)
1347 {
1348         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1349         struct sw_flow_match match;
1350         u64 attrs = 0;
1351         int err;
1352
1353         err = parse_flow_nlattrs(attr, a, &attrs, log);
1354         if (err)
1355                 return -EINVAL;
1356
1357         memset(&match, 0, sizeof(match));
1358         match.key = key;
1359
1360         memset(key, 0, OVS_SW_FLOW_KEY_METADATA_SIZE);
1361         memset(&key->ct, 0, sizeof(key->ct));
1362         key->phy.in_port = DP_MAX_PORTS;
1363
1364         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1365 }
1366
1367 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1368                              const struct sw_flow_key *output, bool is_mask,
1369                              struct sk_buff *skb)
1370 {
1371         struct ovs_key_ethernet *eth_key;
1372         struct nlattr *nla, *encap;
1373
1374         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1375                 goto nla_put_failure;
1376
1377         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1378                 goto nla_put_failure;
1379
1380         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1381                 goto nla_put_failure;
1382
1383         if ((swkey->tun_key.u.ipv4.dst || is_mask)) {
1384                 const void *opts = NULL;
1385
1386                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1387                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1388
1389                 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1390                                        swkey->tun_opts_len))
1391                         goto nla_put_failure;
1392         }
1393
1394         if (swkey->phy.in_port == DP_MAX_PORTS) {
1395                 if (is_mask && (output->phy.in_port == 0xffff))
1396                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1397                                 goto nla_put_failure;
1398         } else {
1399                 u16 upper_u16;
1400                 upper_u16 = !is_mask ? 0 : 0xffff;
1401
1402                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1403                                 (upper_u16 << 16) | output->phy.in_port))
1404                         goto nla_put_failure;
1405         }
1406
1407         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1408                 goto nla_put_failure;
1409
1410         if (ovs_ct_put_key(output, skb))
1411                 goto nla_put_failure;
1412
1413         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1414         if (!nla)
1415                 goto nla_put_failure;
1416
1417         eth_key = nla_data(nla);
1418         ether_addr_copy(eth_key->eth_src, output->eth.src);
1419         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1420
1421         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1422                 __be16 eth_type;
1423                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1424                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1425                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1426                         goto nla_put_failure;
1427                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1428                 if (!swkey->eth.tci)
1429                         goto unencap;
1430         } else
1431                 encap = NULL;
1432
1433         if (swkey->eth.type == htons(ETH_P_802_2)) {
1434                 /*
1435                  * Ethertype 802.2 is represented in the netlink with omitted
1436                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1437                  * 0xffff in the mask attribute.  Ethertype can also
1438                  * be wildcarded.
1439                  */
1440                 if (is_mask && output->eth.type)
1441                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1442                                                 output->eth.type))
1443                                 goto nla_put_failure;
1444                 goto unencap;
1445         }
1446
1447         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1448                 goto nla_put_failure;
1449
1450         if (swkey->eth.type == htons(ETH_P_IP)) {
1451                 struct ovs_key_ipv4 *ipv4_key;
1452
1453                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1454                 if (!nla)
1455                         goto nla_put_failure;
1456                 ipv4_key = nla_data(nla);
1457                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1458                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1459                 ipv4_key->ipv4_proto = output->ip.proto;
1460                 ipv4_key->ipv4_tos = output->ip.tos;
1461                 ipv4_key->ipv4_ttl = output->ip.ttl;
1462                 ipv4_key->ipv4_frag = output->ip.frag;
1463         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1464                 struct ovs_key_ipv6 *ipv6_key;
1465
1466                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1467                 if (!nla)
1468                         goto nla_put_failure;
1469                 ipv6_key = nla_data(nla);
1470                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1471                                 sizeof(ipv6_key->ipv6_src));
1472                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1473                                 sizeof(ipv6_key->ipv6_dst));
1474                 ipv6_key->ipv6_label = output->ipv6.label;
1475                 ipv6_key->ipv6_proto = output->ip.proto;
1476                 ipv6_key->ipv6_tclass = output->ip.tos;
1477                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1478                 ipv6_key->ipv6_frag = output->ip.frag;
1479         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1480                    swkey->eth.type == htons(ETH_P_RARP)) {
1481                 struct ovs_key_arp *arp_key;
1482
1483                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1484                 if (!nla)
1485                         goto nla_put_failure;
1486                 arp_key = nla_data(nla);
1487                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1488                 arp_key->arp_sip = output->ipv4.addr.src;
1489                 arp_key->arp_tip = output->ipv4.addr.dst;
1490                 arp_key->arp_op = htons(output->ip.proto);
1491                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1492                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1493         } else if (eth_p_mpls(swkey->eth.type)) {
1494                 struct ovs_key_mpls *mpls_key;
1495
1496                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1497                 if (!nla)
1498                         goto nla_put_failure;
1499                 mpls_key = nla_data(nla);
1500                 mpls_key->mpls_lse = output->mpls.top_lse;
1501         }
1502
1503         if ((swkey->eth.type == htons(ETH_P_IP) ||
1504              swkey->eth.type == htons(ETH_P_IPV6)) &&
1505              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1506
1507                 if (swkey->ip.proto == IPPROTO_TCP) {
1508                         struct ovs_key_tcp *tcp_key;
1509
1510                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1511                         if (!nla)
1512                                 goto nla_put_failure;
1513                         tcp_key = nla_data(nla);
1514                         tcp_key->tcp_src = output->tp.src;
1515                         tcp_key->tcp_dst = output->tp.dst;
1516                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1517                                          output->tp.flags))
1518                                 goto nla_put_failure;
1519                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1520                         struct ovs_key_udp *udp_key;
1521
1522                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1523                         if (!nla)
1524                                 goto nla_put_failure;
1525                         udp_key = nla_data(nla);
1526                         udp_key->udp_src = output->tp.src;
1527                         udp_key->udp_dst = output->tp.dst;
1528                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1529                         struct ovs_key_sctp *sctp_key;
1530
1531                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1532                         if (!nla)
1533                                 goto nla_put_failure;
1534                         sctp_key = nla_data(nla);
1535                         sctp_key->sctp_src = output->tp.src;
1536                         sctp_key->sctp_dst = output->tp.dst;
1537                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1538                            swkey->ip.proto == IPPROTO_ICMP) {
1539                         struct ovs_key_icmp *icmp_key;
1540
1541                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1542                         if (!nla)
1543                                 goto nla_put_failure;
1544                         icmp_key = nla_data(nla);
1545                         icmp_key->icmp_type = ntohs(output->tp.src);
1546                         icmp_key->icmp_code = ntohs(output->tp.dst);
1547                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1548                            swkey->ip.proto == IPPROTO_ICMPV6) {
1549                         struct ovs_key_icmpv6 *icmpv6_key;
1550
1551                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1552                                                 sizeof(*icmpv6_key));
1553                         if (!nla)
1554                                 goto nla_put_failure;
1555                         icmpv6_key = nla_data(nla);
1556                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1557                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1558
1559                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1560                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1561                                 struct ovs_key_nd *nd_key;
1562
1563                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1564                                 if (!nla)
1565                                         goto nla_put_failure;
1566                                 nd_key = nla_data(nla);
1567                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1568                                                         sizeof(nd_key->nd_target));
1569                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1570                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1571                         }
1572                 }
1573         }
1574
1575 unencap:
1576         if (encap)
1577                 nla_nest_end(skb, encap);
1578
1579         return 0;
1580
1581 nla_put_failure:
1582         return -EMSGSIZE;
1583 }
1584
1585 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1586                     const struct sw_flow_key *output, int attr, bool is_mask,
1587                     struct sk_buff *skb)
1588 {
1589         int err;
1590         struct nlattr *nla;
1591
1592         nla = nla_nest_start(skb, attr);
1593         if (!nla)
1594                 return -EMSGSIZE;
1595         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1596         if (err)
1597                 return err;
1598         nla_nest_end(skb, nla);
1599
1600         return 0;
1601 }
1602
1603 /* Called with ovs_mutex or RCU read lock. */
1604 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1605 {
1606         if (ovs_identifier_is_ufid(&flow->id))
1607                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1608                                flow->id.ufid);
1609
1610         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1611                                OVS_FLOW_ATTR_KEY, false, skb);
1612 }
1613
1614 /* Called with ovs_mutex or RCU read lock. */
1615 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1616 {
1617         return ovs_nla_put_key(&flow->key, &flow->key,
1618                                 OVS_FLOW_ATTR_KEY, false, skb);
1619 }
1620
1621 /* Called with ovs_mutex or RCU read lock. */
1622 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1623 {
1624         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1625                                 OVS_FLOW_ATTR_MASK, true, skb);
1626 }
1627
1628 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1629
1630 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1631 {
1632         struct sw_flow_actions *sfa;
1633
1634         if (size > MAX_ACTIONS_BUFSIZE) {
1635                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1636                 return ERR_PTR(-EINVAL);
1637         }
1638
1639         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1640         if (!sfa)
1641                 return ERR_PTR(-ENOMEM);
1642
1643         sfa->actions_len = 0;
1644         return sfa;
1645 }
1646
1647 static void ovs_nla_free_set_action(const struct nlattr *a)
1648 {
1649         const struct nlattr *ovs_key = nla_data(a);
1650         struct ovs_tunnel_info *ovs_tun;
1651
1652         switch (nla_type(ovs_key)) {
1653         case OVS_KEY_ATTR_TUNNEL_INFO:
1654                 ovs_tun = nla_data(ovs_key);
1655                 ovs_dst_release((struct dst_entry *)ovs_tun->tun_dst);
1656                 break;
1657         }
1658 }
1659
1660 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1661 {
1662         const struct nlattr *a;
1663         int rem;
1664
1665         if (!sf_acts)
1666                 return;
1667
1668         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1669                 switch (nla_type(a)) {
1670                 case OVS_ACTION_ATTR_SET:
1671                         ovs_nla_free_set_action(a);
1672                         break;
1673                 case OVS_ACTION_ATTR_CT:
1674                         ovs_ct_free_action(a);
1675                         break;
1676                 }
1677         }
1678
1679         kfree(sf_acts);
1680 }
1681
1682 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1683 {
1684         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1685 }
1686
1687 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1688  * The caller must hold rcu_read_lock for this to be sensible.
1689  */
1690 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1691 {
1692         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1693 }
1694
1695 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1696                                        int attr_len, bool log)
1697 {
1698
1699         struct sw_flow_actions *acts;
1700         int new_acts_size;
1701         int req_size = NLA_ALIGN(attr_len);
1702         int next_offset = offsetof(struct sw_flow_actions, actions) +
1703                                         (*sfa)->actions_len;
1704
1705         if (req_size <= (ksize(*sfa) - next_offset))
1706                 goto out;
1707
1708         new_acts_size = ksize(*sfa) * 2;
1709
1710         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1711                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1712                         return ERR_PTR(-EMSGSIZE);
1713                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1714         }
1715
1716         acts = nla_alloc_flow_actions(new_acts_size, log);
1717         if (IS_ERR(acts))
1718                 return (void *)acts;
1719
1720         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1721         acts->actions_len = (*sfa)->actions_len;
1722         acts->orig_len = (*sfa)->orig_len;
1723         kfree(*sfa);
1724         *sfa = acts;
1725
1726 out:
1727         (*sfa)->actions_len += req_size;
1728         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1729 }
1730
1731 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1732                                    int attrtype, void *data, int len, bool log)
1733 {
1734         struct nlattr *a;
1735
1736         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1737         if (IS_ERR(a))
1738                 return a;
1739
1740         a->nla_type = attrtype;
1741         a->nla_len = nla_attr_size(len);
1742
1743         if (data)
1744                 memcpy(nla_data(a), data, len);
1745         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1746
1747         return a;
1748 }
1749
1750 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1751                        int len, bool log)
1752 {
1753         struct nlattr *a;
1754
1755         a = __add_action(sfa, attrtype, data, len, log);
1756         if (IS_ERR(a))
1757                 return PTR_ERR(a);
1758
1759         return 0;
1760 }
1761
1762 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1763                                           int attrtype, bool log)
1764 {
1765         int used = (*sfa)->actions_len;
1766         int err;
1767
1768         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1769         if (err)
1770                 return err;
1771
1772         return used;
1773 }
1774
1775 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1776                                          int st_offset)
1777 {
1778         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1779                                                                st_offset);
1780
1781         a->nla_len = sfa->actions_len - st_offset;
1782 }
1783
1784 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1785                                   const struct sw_flow_key *key,
1786                                   int depth, struct sw_flow_actions **sfa,
1787                                   __be16 eth_type, __be16 vlan_tci, bool log);
1788
1789 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1790                                     const struct sw_flow_key *key, int depth,
1791                                     struct sw_flow_actions **sfa,
1792                                     __be16 eth_type, __be16 vlan_tci, bool log)
1793 {
1794         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1795         const struct nlattr *probability, *actions;
1796         const struct nlattr *a;
1797         int rem, start, err, st_acts;
1798
1799         memset(attrs, 0, sizeof(attrs));
1800         nla_for_each_nested(a, attr, rem) {
1801                 int type = nla_type(a);
1802                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1803                         return -EINVAL;
1804                 attrs[type] = a;
1805         }
1806         if (rem)
1807                 return -EINVAL;
1808
1809         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1810         if (!probability || nla_len(probability) != sizeof(u32))
1811                 return -EINVAL;
1812
1813         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1814         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1815                 return -EINVAL;
1816
1817         /* validation done, copy sample action. */
1818         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1819         if (start < 0)
1820                 return start;
1821         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1822                                  nla_data(probability), sizeof(u32), log);
1823         if (err)
1824                 return err;
1825         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1826         if (st_acts < 0)
1827                 return st_acts;
1828
1829         err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1830                                      eth_type, vlan_tci, log);
1831         if (err)
1832                 return err;
1833
1834         add_nested_action_end(*sfa, st_acts);
1835         add_nested_action_end(*sfa, start);
1836
1837         return 0;
1838 }
1839
1840 void ovs_match_init(struct sw_flow_match *match,
1841                     struct sw_flow_key *key,
1842                     struct sw_flow_mask *mask)
1843 {
1844         memset(match, 0, sizeof(*match));
1845         match->key = key;
1846         match->mask = mask;
1847
1848         memset(key, 0, sizeof(*key));
1849
1850         if (mask) {
1851                 memset(&mask->key, 0, sizeof(mask->key));
1852                 mask->range.start = mask->range.end = 0;
1853         }
1854 }
1855
1856 static int validate_geneve_opts(struct sw_flow_key *key)
1857 {
1858         struct geneve_opt *option;
1859         int opts_len = key->tun_opts_len;
1860         bool crit_opt = false;
1861
1862         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1863         while (opts_len > 0) {
1864                 int len;
1865
1866                 if (opts_len < sizeof(*option))
1867                         return -EINVAL;
1868
1869                 len = sizeof(*option) + option->length * 4;
1870                 if (len > opts_len)
1871                         return -EINVAL;
1872
1873                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1874
1875                 option = (struct geneve_opt *)((u8 *)option + len);
1876                 opts_len -= len;
1877         };
1878
1879         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1880
1881         return 0;
1882 }
1883
1884 static int validate_and_copy_set_tun(const struct nlattr *attr,
1885                                      struct sw_flow_actions **sfa, bool log)
1886 {
1887         struct sw_flow_match match;
1888         struct sw_flow_key key;
1889         struct metadata_dst *tun_dst;
1890         struct ip_tunnel_info *tun_info;
1891         struct ovs_tunnel_info *ovs_tun;
1892         struct nlattr *a;
1893         int err = 0, start, opts_type;
1894
1895         ovs_match_init(&match, &key, NULL);
1896         opts_type = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1897         if (opts_type < 0)
1898                 return opts_type;
1899
1900         if (key.tun_opts_len) {
1901                 switch (opts_type) {
1902                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1903                         err = validate_geneve_opts(&key);
1904                         if (err < 0)
1905                                 return err;
1906                         break;
1907                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1908                         break;
1909                 }
1910         };
1911
1912         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1913         if (start < 0)
1914                 return start;
1915
1916         tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1917         if (!tun_dst)
1918                 return -ENOMEM;
1919
1920         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1921                          sizeof(*ovs_tun), log);
1922         if (IS_ERR(a)) {
1923                 ovs_dst_release((struct dst_entry *)tun_dst);
1924                 return PTR_ERR(a);
1925         }
1926
1927         ovs_tun = nla_data(a);
1928         ovs_tun->tun_dst = tun_dst;
1929
1930         tun_info = &tun_dst->u.tun_info;
1931         tun_info->mode = IP_TUNNEL_INFO_TX;
1932         tun_info->key = key.tun_key;
1933
1934         /* We need to store the options in the action itself since
1935          * everything else will go away after flow setup. We can append
1936          * it to tun_info and then point there.
1937          */
1938         ip_tunnel_info_opts_set(tun_info,
1939                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
1940                                 key.tun_opts_len);
1941         add_nested_action_end(*sfa, start);
1942
1943         return err;
1944 }
1945
1946 /* Return false if there are any non-masked bits set.
1947  * Mask follows data immediately, before any netlink padding.
1948  */
1949 static bool validate_masked(u8 *data, int len)
1950 {
1951         u8 *mask = data + len;
1952
1953         while (len--)
1954                 if (*data++ & ~*mask++)
1955                         return false;
1956
1957         return true;
1958 }
1959
1960 static int validate_set(const struct nlattr *a,
1961                         const struct sw_flow_key *flow_key,
1962                         struct sw_flow_actions **sfa,
1963                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
1964 {
1965         const struct nlattr *ovs_key = nla_data(a);
1966         int key_type = nla_type(ovs_key);
1967         size_t key_len;
1968
1969         /* There can be only one key in a action */
1970         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1971                 return -EINVAL;
1972
1973         key_len = nla_len(ovs_key);
1974         if (masked)
1975                 key_len /= 2;
1976
1977         if (key_type > OVS_KEY_ATTR_MAX ||
1978             !check_attr_len(key_len, ovs_key_lens[key_type].len))
1979                 return -EINVAL;
1980
1981         if (masked && !validate_masked(nla_data(ovs_key), key_len))
1982                 return -EINVAL;
1983
1984         switch (key_type) {
1985         const struct ovs_key_ipv4 *ipv4_key;
1986         const struct ovs_key_ipv6 *ipv6_key;
1987         int err;
1988
1989         case OVS_KEY_ATTR_PRIORITY:
1990         case OVS_KEY_ATTR_SKB_MARK:
1991         case OVS_KEY_ATTR_CT_MARK:
1992         case OVS_KEY_ATTR_CT_LABELS:
1993         case OVS_KEY_ATTR_ETHERNET:
1994                 break;
1995
1996         case OVS_KEY_ATTR_TUNNEL:
1997                 if (eth_p_mpls(eth_type))
1998                         return -EINVAL;
1999
2000                 if (masked)
2001                         return -EINVAL; /* Masked tunnel set not supported. */
2002
2003                 *skip_copy = true;
2004                 err = validate_and_copy_set_tun(a, sfa, log);
2005                 if (err)
2006                         return err;
2007                 break;
2008
2009         case OVS_KEY_ATTR_IPV4:
2010                 if (eth_type != htons(ETH_P_IP))
2011                         return -EINVAL;
2012
2013                 ipv4_key = nla_data(ovs_key);
2014
2015                 if (masked) {
2016                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2017
2018                         /* Non-writeable fields. */
2019                         if (mask->ipv4_proto || mask->ipv4_frag)
2020                                 return -EINVAL;
2021                 } else {
2022                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2023                                 return -EINVAL;
2024
2025                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2026                                 return -EINVAL;
2027                 }
2028                 break;
2029
2030         case OVS_KEY_ATTR_IPV6:
2031                 if (eth_type != htons(ETH_P_IPV6))
2032                         return -EINVAL;
2033
2034                 ipv6_key = nla_data(ovs_key);
2035
2036                 if (masked) {
2037                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2038
2039                         /* Non-writeable fields. */
2040                         if (mask->ipv6_proto || mask->ipv6_frag)
2041                                 return -EINVAL;
2042
2043                         /* Invalid bits in the flow label mask? */
2044                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2045                                 return -EINVAL;
2046                 } else {
2047                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2048                                 return -EINVAL;
2049
2050                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2051                                 return -EINVAL;
2052                 }
2053                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2054                         return -EINVAL;
2055
2056                 break;
2057
2058         case OVS_KEY_ATTR_TCP:
2059                 if ((eth_type != htons(ETH_P_IP) &&
2060                      eth_type != htons(ETH_P_IPV6)) ||
2061                     flow_key->ip.proto != IPPROTO_TCP)
2062                         return -EINVAL;
2063
2064                 break;
2065
2066         case OVS_KEY_ATTR_UDP:
2067                 if ((eth_type != htons(ETH_P_IP) &&
2068                      eth_type != htons(ETH_P_IPV6)) ||
2069                     flow_key->ip.proto != IPPROTO_UDP)
2070                         return -EINVAL;
2071
2072                 break;
2073
2074         case OVS_KEY_ATTR_MPLS:
2075                 if (!eth_p_mpls(eth_type))
2076                         return -EINVAL;
2077                 break;
2078
2079         case OVS_KEY_ATTR_SCTP:
2080                 if ((eth_type != htons(ETH_P_IP) &&
2081                      eth_type != htons(ETH_P_IPV6)) ||
2082                     flow_key->ip.proto != IPPROTO_SCTP)
2083                         return -EINVAL;
2084
2085                 break;
2086
2087         default:
2088                 return -EINVAL;
2089         }
2090
2091         /* Convert non-masked non-tunnel set actions to masked set actions. */
2092         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2093                 int start, len = key_len * 2;
2094                 struct nlattr *at;
2095
2096                 *skip_copy = true;
2097
2098                 start = add_nested_action_start(sfa,
2099                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2100                                                 log);
2101                 if (start < 0)
2102                         return start;
2103
2104                 at = __add_action(sfa, key_type, NULL, len, log);
2105                 if (IS_ERR(at))
2106                         return PTR_ERR(at);
2107
2108                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2109                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2110                 /* Clear non-writeable bits from otherwise writeable fields. */
2111                 if (key_type == OVS_KEY_ATTR_IPV6) {
2112                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2113
2114                         mask->ipv6_label &= htonl(0x000FFFFF);
2115                 }
2116                 add_nested_action_end(*sfa, start);
2117         }
2118
2119         return 0;
2120 }
2121
2122 static int validate_userspace(const struct nlattr *attr)
2123 {
2124         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2125                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2126                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2127                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2128         };
2129         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2130         int error;
2131
2132         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2133                                  attr, userspace_policy);
2134         if (error)
2135                 return error;
2136
2137         if (!a[OVS_USERSPACE_ATTR_PID] ||
2138             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2139                 return -EINVAL;
2140
2141         return 0;
2142 }
2143
2144 static int copy_action(const struct nlattr *from,
2145                        struct sw_flow_actions **sfa, bool log)
2146 {
2147         int totlen = NLA_ALIGN(from->nla_len);
2148         struct nlattr *to;
2149
2150         to = reserve_sfa_size(sfa, from->nla_len, log);
2151         if (IS_ERR(to))
2152                 return PTR_ERR(to);
2153
2154         memcpy(to, from, totlen);
2155         return 0;
2156 }
2157
2158 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2159                                   const struct sw_flow_key *key,
2160                                   int depth, struct sw_flow_actions **sfa,
2161                                   __be16 eth_type, __be16 vlan_tci, bool log)
2162 {
2163         const struct nlattr *a;
2164         int rem, err;
2165
2166         if (depth >= SAMPLE_ACTION_DEPTH)
2167                 return -EOVERFLOW;
2168
2169         nla_for_each_nested(a, attr, rem) {
2170                 /* Expected argument lengths, (u32)-1 for variable length. */
2171                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2172                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2173                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2174                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2175                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2176                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2177                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2178                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2179                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2180                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2181                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2182                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2183                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2184                 };
2185                 const struct ovs_action_push_vlan *vlan;
2186                 int type = nla_type(a);
2187                 bool skip_copy;
2188
2189                 if (type > OVS_ACTION_ATTR_MAX ||
2190                     (action_lens[type] != nla_len(a) &&
2191                      action_lens[type] != (u32)-1))
2192                         return -EINVAL;
2193
2194                 skip_copy = false;
2195                 switch (type) {
2196                 case OVS_ACTION_ATTR_UNSPEC:
2197                         return -EINVAL;
2198
2199                 case OVS_ACTION_ATTR_USERSPACE:
2200                         err = validate_userspace(a);
2201                         if (err)
2202                                 return err;
2203                         break;
2204
2205                 case OVS_ACTION_ATTR_OUTPUT:
2206                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2207                                 return -EINVAL;
2208                         break;
2209
2210                 case OVS_ACTION_ATTR_HASH: {
2211                         const struct ovs_action_hash *act_hash = nla_data(a);
2212
2213                         switch (act_hash->hash_alg) {
2214                         case OVS_HASH_ALG_L4:
2215                                 break;
2216                         default:
2217                                 return  -EINVAL;
2218                         }
2219
2220                         break;
2221                 }
2222
2223                 case OVS_ACTION_ATTR_POP_VLAN:
2224                         vlan_tci = htons(0);
2225                         break;
2226
2227                 case OVS_ACTION_ATTR_PUSH_VLAN:
2228                         vlan = nla_data(a);
2229                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2230                                 return -EINVAL;
2231                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2232                                 return -EINVAL;
2233                         vlan_tci = vlan->vlan_tci;
2234                         break;
2235
2236                 case OVS_ACTION_ATTR_RECIRC:
2237                         break;
2238
2239                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2240                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2241
2242                         if (!eth_p_mpls(mpls->mpls_ethertype))
2243                                 return -EINVAL;
2244                         /* Prohibit push MPLS other than to a white list
2245                          * for packets that have a known tag order.
2246                          */
2247                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2248                             (eth_type != htons(ETH_P_IP) &&
2249                              eth_type != htons(ETH_P_IPV6) &&
2250                              eth_type != htons(ETH_P_ARP) &&
2251                              eth_type != htons(ETH_P_RARP) &&
2252                              !eth_p_mpls(eth_type)))
2253                                 return -EINVAL;
2254                         eth_type = mpls->mpls_ethertype;
2255                         break;
2256                 }
2257
2258                 case OVS_ACTION_ATTR_POP_MPLS:
2259                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2260                             !eth_p_mpls(eth_type))
2261                                 return -EINVAL;
2262
2263                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2264                          * as there is no check here to ensure that the new
2265                          * eth_type is valid and thus set actions could
2266                          * write off the end of the packet or otherwise
2267                          * corrupt it.
2268                          *
2269                          * Support for these actions is planned using packet
2270                          * recirculation.
2271                          */
2272                         eth_type = htons(0);
2273                         break;
2274
2275                 case OVS_ACTION_ATTR_SET:
2276                         err = validate_set(a, key, sfa,
2277                                            &skip_copy, eth_type, false, log);
2278                         if (err)
2279                                 return err;
2280                         break;
2281
2282                 case OVS_ACTION_ATTR_SET_MASKED:
2283                         err = validate_set(a, key, sfa,
2284                                            &skip_copy, eth_type, true, log);
2285                         if (err)
2286                                 return err;
2287                         break;
2288
2289                 case OVS_ACTION_ATTR_SAMPLE:
2290                         err = validate_and_copy_sample(net, a, key, depth, sfa,
2291                                                        eth_type, vlan_tci, log);
2292                         if (err)
2293                                 return err;
2294                         skip_copy = true;
2295                         break;
2296
2297                 case OVS_ACTION_ATTR_CT:
2298                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2299                         if (err)
2300                                 return err;
2301                         skip_copy = true;
2302                         break;
2303
2304                 default:
2305                         OVS_NLERR(log, "Unknown Action type %d", type);
2306                         return -EINVAL;
2307                 }
2308                 if (!skip_copy) {
2309                         err = copy_action(a, sfa, log);
2310                         if (err)
2311                                 return err;
2312                 }
2313         }
2314
2315         if (rem > 0)
2316                 return -EINVAL;
2317
2318         return 0;
2319 }
2320
2321 /* 'key' must be the masked key. */
2322 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2323                          const struct sw_flow_key *key,
2324                          struct sw_flow_actions **sfa, bool log)
2325 {
2326         int err;
2327
2328         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2329         if (IS_ERR(*sfa))
2330                 return PTR_ERR(*sfa);
2331
2332         (*sfa)->orig_len = nla_len(attr);
2333         err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2334                                      key->eth.tci, log);
2335         if (err)
2336                 ovs_nla_free_flow_actions(*sfa);
2337
2338         return err;
2339 }
2340
2341 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2342 {
2343         const struct nlattr *a;
2344         struct nlattr *start;
2345         int err = 0, rem;
2346
2347         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2348         if (!start)
2349                 return -EMSGSIZE;
2350
2351         nla_for_each_nested(a, attr, rem) {
2352                 int type = nla_type(a);
2353                 struct nlattr *st_sample;
2354
2355                 switch (type) {
2356                 case OVS_SAMPLE_ATTR_PROBABILITY:
2357                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2358                                     sizeof(u32), nla_data(a)))
2359                                 return -EMSGSIZE;
2360                         break;
2361                 case OVS_SAMPLE_ATTR_ACTIONS:
2362                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2363                         if (!st_sample)
2364                                 return -EMSGSIZE;
2365                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2366                         if (err)
2367                                 return err;
2368                         nla_nest_end(skb, st_sample);
2369                         break;
2370                 }
2371         }
2372
2373         nla_nest_end(skb, start);
2374         return err;
2375 }
2376
2377 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2378 {
2379         const struct nlattr *ovs_key = nla_data(a);
2380         int key_type = nla_type(ovs_key);
2381         struct nlattr *start;
2382         int err;
2383
2384         switch (key_type) {
2385         case OVS_KEY_ATTR_TUNNEL_INFO: {
2386                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2387                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2388
2389                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2390                 if (!start)
2391                         return -EMSGSIZE;
2392
2393                 err = ipv4_tun_to_nlattr(skb, &tun_info->key,
2394                                          tun_info->options_len ?
2395                                              ip_tunnel_info_opts(tun_info) : NULL,
2396                                          tun_info->options_len);
2397                 if (err)
2398                         return err;
2399                 nla_nest_end(skb, start);
2400                 break;
2401         }
2402         default:
2403                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2404                         return -EMSGSIZE;
2405                 break;
2406         }
2407
2408         return 0;
2409 }
2410
2411 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2412                                                 struct sk_buff *skb)
2413 {
2414         const struct nlattr *ovs_key = nla_data(a);
2415         struct nlattr *nla;
2416         size_t key_len = nla_len(ovs_key) / 2;
2417
2418         /* Revert the conversion we did from a non-masked set action to
2419          * masked set action.
2420          */
2421         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2422         if (!nla)
2423                 return -EMSGSIZE;
2424
2425         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2426                 return -EMSGSIZE;
2427
2428         nla_nest_end(skb, nla);
2429         return 0;
2430 }
2431
2432 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2433 {
2434         const struct nlattr *a;
2435         int rem, err;
2436
2437         nla_for_each_attr(a, attr, len, rem) {
2438                 int type = nla_type(a);
2439
2440                 switch (type) {
2441                 case OVS_ACTION_ATTR_SET:
2442                         err = set_action_to_attr(a, skb);
2443                         if (err)
2444                                 return err;
2445                         break;
2446
2447                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2448                         err = masked_set_action_to_set_action_attr(a, skb);
2449                         if (err)
2450                                 return err;
2451                         break;
2452
2453                 case OVS_ACTION_ATTR_SAMPLE:
2454                         err = sample_action_to_attr(a, skb);
2455                         if (err)
2456                                 return err;
2457                         break;
2458
2459                 case OVS_ACTION_ATTR_CT:
2460                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2461                         if (err)
2462                                 return err;
2463                         break;
2464
2465                 default:
2466                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2467                                 return -EMSGSIZE;
2468                         break;
2469                 }
2470         }
2471
2472         return 0;
2473 }