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(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
 268                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_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         [OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
 329         [OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
 330 };
 331
 332 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 333 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 334         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 335         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 336         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 337         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 338         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 339         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 340         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 341         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 342         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 343         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 344         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 345         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 346         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 347         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 348         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 349         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 350         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 351         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 352         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 353         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 354                                      .next = ovs_tunnel_key_lens, },
 355         [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 356         [OVS_KEY_ATTR_CT_STATE]  = { .len = sizeof(u32) },
 357         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 358         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 359         [OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
 360 };
 361
 362 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
 363 {
 364         return expected_len == attr_len ||
 365                expected_len == OVS_ATTR_NESTED ||
 366                expected_len == OVS_ATTR_VARIABLE;
 367 }
 368
 369 static bool is_all_zero(const u8 *fp, size_t size)
 370 {
 371         int i;
 372
 373         if (!fp)
 374                 return false;
 375
 376         for (i = 0; i < size; i++)
 377                 if (fp[i])
 378                         return false;
 379
 380         return true;
 381 }
 382
 383 static int __parse_flow_nlattrs(const struct nlattr *attr,
 384                                 const struct nlattr *a[],
 385                                 u64 *attrsp, bool log, bool nz)
 386 {
 387         const struct nlattr *nla;
 388         u64 attrs;
 389         int rem;
 390
 391         attrs = *attrsp;
 392         nla_for_each_nested(nla, attr, rem) {
 393                 u16 type = nla_type(nla);
 394                 int expected_len;
 395
 396                 if (type > OVS_KEY_ATTR_MAX) {
 397                         OVS_NLERR(log, "Key type %d is out of range max %d",
 398                                   type, OVS_KEY_ATTR_MAX);
 399                         return -EINVAL;
 400                 }
 401
 402                 if (attrs & (1ULL << type)) {
 403                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 404                         return -EINVAL;
 405                 }
 406
 407                 expected_len = ovs_key_lens[type].len;
 408                 if (!check_attr_len(nla_len(nla), expected_len)) {
 409                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 410                                   type, nla_len(nla), expected_len);
 411                         return -EINVAL;
 412                 }
 413
 414                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
 415                         attrs |= 1ULL << type;
 416                         a[type] = nla;
 417                 }
 418         }
 419         if (rem) {
 420                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 421                 return -EINVAL;
 422         }
 423
 424         *attrsp = attrs;
 425         return 0;
 426 }
 427
 428 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 429                                    const struct nlattr *a[], u64 *attrsp,
 430                                    bool log)
 431 {
 432         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 433 }
 434
 435 static int parse_flow_nlattrs(const struct nlattr *attr,
 436                               const struct nlattr *a[], u64 *attrsp,
 437                               bool log)
 438 {
 439         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 440 }
 441
 442 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 443                                      struct sw_flow_match *match, bool is_mask,
 444                                      bool log)
 445 {
 446         unsigned long opt_key_offset;
 447
 448         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 449                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 450                           nla_len(a), sizeof(match->key->tun_opts));
 451                 return -EINVAL;
 452         }
 453
 454         if (nla_len(a) % 4 != 0) {
 455                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 456                           nla_len(a));
 457                 return -EINVAL;
 458         }
 459
 460         /* We need to record the length of the options passed
 461          * down, otherwise packets with the same format but
 462          * additional options will be silently matched.
 463          */
 464         if (!is_mask) {
 465                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 466                                 false);
 467         } else {
 468                 /* This is somewhat unusual because it looks at
 469                  * both the key and mask while parsing the
 470                  * attributes (and by extension assumes the key
 471                  * is parsed first). Normally, we would verify
 472                  * that each is the correct length and that the
 473                  * attributes line up in the validate function.
 474                  * However, that is difficult because this is
 475                  * variable length and we won't have the
 476                  * information later.
 477                  */
 478                 if (match->key->tun_opts_len != nla_len(a)) {
 479                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 480                                   match->key->tun_opts_len, nla_len(a));
 481                         return -EINVAL;
 482                 }
 483
 484                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 485         }
 486
 487         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 488         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 489                                   nla_len(a), is_mask);
 490         return 0;
 491 }
 492
 493 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
 494                                      struct sw_flow_match *match, bool is_mask,
 495                                      bool log)
 496 {
 497         struct nlattr *a;
 498         int rem;
 499         unsigned long opt_key_offset;
 500         struct vxlan_metadata opts;
 501
 502         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 503
 504         memset(&opts, 0, sizeof(opts));
 505         nla_for_each_nested(a, attr, rem) {
 506                 int type = nla_type(a);
 507
 508                 if (type > OVS_VXLAN_EXT_MAX) {
 509                         OVS_NLERR(log, "VXLAN extension %d out of range max %d",
 510                                   type, OVS_VXLAN_EXT_MAX);
 511                         return -EINVAL;
 512                 }
 513
 514                 if (!check_attr_len(nla_len(a),
 515                                     ovs_vxlan_ext_key_lens[type].len)) {
 516                         OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
 517                                   type, nla_len(a),
 518                                   ovs_vxlan_ext_key_lens[type].len);
 519                         return -EINVAL;
 520                 }
 521
 522                 switch (type) {
 523                 case OVS_VXLAN_EXT_GBP:
 524                         opts.gbp = nla_get_u32(a);
 525                         break;
 526                 default:
 527                         OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
 528                                   type);
 529                         return -EINVAL;
 530                 }
 531         }
 532         if (rem) {
 533                 OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
 534                           rem);
 535                 return -EINVAL;
 536         }
 537
 538         if (!is_mask)
 539                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 540         else
 541                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 542
 543         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 544         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 545                                   is_mask);
 546         return 0;
 547 }
 548
 549 static int ip_tun_from_nlattr(const struct nlattr *attr,
 550                               struct sw_flow_match *match, bool is_mask,
 551                               bool log)
 552 {
 553         bool ttl = false, ipv4 = false, ipv6 = false;
 554         __be16 tun_flags = 0;
 555         int opts_type = 0;
 556         struct nlattr *a;
 557         int rem;
 558
 559         nla_for_each_nested(a, attr, rem) {
 560                 int type = nla_type(a);
 561                 int err;
 562
 563                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 564                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 565                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 566                         return -EINVAL;
 567                 }
 568
 569                 if (!check_attr_len(nla_len(a),
 570                                     ovs_tunnel_key_lens[type].len)) {
 571                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 572                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 573                         return -EINVAL;
 574                 }
 575
 576                 switch (type) {
 577                 case OVS_TUNNEL_KEY_ATTR_ID:
 578                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 579                                         nla_get_be64(a), is_mask);
 580                         tun_flags |= TUNNEL_KEY;
 581                         break;
 582                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 583                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
 584                                         nla_get_in_addr(a), is_mask);
 585                         ipv4 = true;
 586                         break;
 587                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 588                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
 589                                         nla_get_in_addr(a), is_mask);
 590                         ipv4 = true;
 591                         break;
 592                 case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
 593                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
 594                                         nla_get_in6_addr(a), is_mask);
 595                         ipv6 = true;
 596                         break;
 597                 case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
 598                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
 599                                         nla_get_in6_addr(a), is_mask);
 600                         ipv6 = true;
 601                         break;
 602                 case OVS_TUNNEL_KEY_ATTR_TOS:
 603                         SW_FLOW_KEY_PUT(match, tun_key.tos,
 604                                         nla_get_u8(a), is_mask);
 605                         break;
 606                 case OVS_TUNNEL_KEY_ATTR_TTL:
 607                         SW_FLOW_KEY_PUT(match, tun_key.ttl,
 608                                         nla_get_u8(a), is_mask);
 609                         ttl = true;
 610                         break;
 611                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 612                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 613                         break;
 614                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 615                         tun_flags |= TUNNEL_CSUM;
 616                         break;
 617                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 618                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 619                                         nla_get_be16(a), is_mask);
 620                         break;
 621                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 622                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 623                                         nla_get_be16(a), is_mask);
 624                         break;
 625                 case OVS_TUNNEL_KEY_ATTR_OAM:
 626                         tun_flags |= TUNNEL_OAM;
 627                         break;
 628                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 629                         if (opts_type) {
 630                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 631                                 return -EINVAL;
 632                         }
 633
 634                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 635                         if (err)
 636                                 return err;
 637
 638                         tun_flags |= TUNNEL_GENEVE_OPT;
 639                         opts_type = type;
 640                         break;
 641                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 642                         if (opts_type) {
 643                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 644                                 return -EINVAL;
 645                         }
 646
 647                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 648                         if (err)
 649                                 return err;
 650
 651                         tun_flags |= TUNNEL_VXLAN_OPT;
 652                         opts_type = type;
 653                         break;
 654                 default:
 655                         OVS_NLERR(log, "Unknown IP tunnel attribute %d",
 656                                   type);
 657                         return -EINVAL;
 658                 }
 659         }
 660
 661         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
 662         if (is_mask)
 663                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
 664         else
 665                 SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
 666                                 false);
 667
 668         if (rem > 0) {
 669                 OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
 670                           rem);
 671                 return -EINVAL;
 672         }
 673
 674         if (ipv4 && ipv6) {
 675                 OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
 676                 return -EINVAL;
 677         }
 678
 679         if (!is_mask) {
 680                 if (!ipv4 && !ipv6) {
 681                         OVS_NLERR(log, "IP tunnel dst address not specified");
 682                         return -EINVAL;
 683                 }
 684                 if (ipv4 && !match->key->tun_key.u.ipv4.dst) {
 685                         OVS_NLERR(log, "IPv4 tunnel dst address is zero");
 686                         return -EINVAL;
 687                 }
 688                 if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
 689                         OVS_NLERR(log, "IPv6 tunnel dst address is zero");
 690                         return -EINVAL;
 691                 }
 692
 693                 if (!ttl) {
 694                         OVS_NLERR(log, "IP tunnel TTL not specified.");
 695                         return -EINVAL;
 696                 }
 697         }
 698
 699         return opts_type;
 700 }
 701
 702 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 703                                const void *tun_opts, int swkey_tun_opts_len)
 704 {
 705         const struct vxlan_metadata *opts = tun_opts;
 706         struct nlattr *nla;
 707
 708         nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
 709         if (!nla)
 710                 return -EMSGSIZE;
 711
 712         if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
 713                 return -EMSGSIZE;
 714
 715         nla_nest_end(skb, nla);
 716         return 0;
 717 }
 718
 719 static int __ip_tun_to_nlattr(struct sk_buff *skb,
 720                               const struct ip_tunnel_key *output,
 721                               const void *tun_opts, int swkey_tun_opts_len,
 722                               unsigned short tun_proto)
 723 {
 724         if (output->tun_flags & TUNNEL_KEY &&
 725             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
 726                 return -EMSGSIZE;
 727         switch (tun_proto) {
 728         case AF_INET:
 729                 if (output->u.ipv4.src &&
 730                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 731                                     output->u.ipv4.src))
 732                         return -EMSGSIZE;
 733                 if (output->u.ipv4.dst &&
 734                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 735                                     output->u.ipv4.dst))
 736                         return -EMSGSIZE;
 737                 break;
 738         case AF_INET6:
 739                 if (!ipv6_addr_any(&output->u.ipv6.src) &&
 740                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
 741                                      &output->u.ipv6.src))
 742                         return -EMSGSIZE;
 743                 if (!ipv6_addr_any(&output->u.ipv6.dst) &&
 744                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
 745                                      &output->u.ipv6.dst))
 746                         return -EMSGSIZE;
 747                 break;
 748         }
 749         if (output->tos &&
 750             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 751                 return -EMSGSIZE;
 752         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 753                 return -EMSGSIZE;
 754         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 755             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 756                 return -EMSGSIZE;
 757         if ((output->tun_flags & TUNNEL_CSUM) &&
 758             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 759                 return -EMSGSIZE;
 760         if (output->tp_src &&
 761             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 762                 return -EMSGSIZE;
 763         if (output->tp_dst &&
 764             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 765                 return -EMSGSIZE;
 766         if ((output->tun_flags & TUNNEL_OAM) &&
 767             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 768                 return -EMSGSIZE;
 769         if (swkey_tun_opts_len) {
 770                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 771                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 772                             swkey_tun_opts_len, tun_opts))
 773                         return -EMSGSIZE;
 774                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 775                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 776                         return -EMSGSIZE;
 777         }
 778
 779         return 0;
 780 }
 781
 782 static int ip_tun_to_nlattr(struct sk_buff *skb,
 783                             const struct ip_tunnel_key *output,
 784                             const void *tun_opts, int swkey_tun_opts_len,
 785                             unsigned short tun_proto)
 786 {
 787         struct nlattr *nla;
 788         int err;
 789
 790         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 791         if (!nla)
 792                 return -EMSGSIZE;
 793
 794         err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
 795                                  tun_proto);
 796         if (err)
 797                 return err;
 798
 799         nla_nest_end(skb, nla);
 800         return 0;
 801 }
 802
 803 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
 804                             struct ip_tunnel_info *tun_info)
 805 {
 806         return __ip_tun_to_nlattr(skb, &tun_info->key,
 807                                   ip_tunnel_info_opts(tun_info),
 808                                   tun_info->options_len,
 809                                   ip_tunnel_info_af(tun_info));
 810 }
 811
 812 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
 813                                   u64 *attrs, const struct nlattr **a,
 814                                  bool is_mask, bool log)
 815 {
 816         if (*attrs & (1ULL << OVS_KEY_ATTR_DP_HASH)) {
 817                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 818
 819                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 820                 *attrs &= ~(1ULL << OVS_KEY_ATTR_DP_HASH);
 821         }
 822
 823         if (*attrs & (1ULL << OVS_KEY_ATTR_RECIRC_ID)) {
 824                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 825
 826                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 827                 *attrs &= ~(1ULL << OVS_KEY_ATTR_RECIRC_ID);
 828         }
 829
 830         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
 831                 SW_FLOW_KEY_PUT(match, phy.priority,
 832                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 833                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
 834         }
 835
 836         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
 837                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 838
 839                 if (is_mask) {
 840                         in_port = 0xffffffff; /* Always exact match in_port. */
 841                 } else if (in_port >= DP_MAX_PORTS) {
 842                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 843                                   in_port, DP_MAX_PORTS);
 844                         return -EINVAL;
 845                 }
 846
 847                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 848                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
 849         } else if (!is_mask) {
 850                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 851         }
 852
 853         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
 854                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 855
 856                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 857                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
 858         }
 859         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
 860                 if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 861                                          is_mask, log) < 0)
 862                         return -EINVAL;
 863                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
 864         }
 865
 866         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
 867             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
 868                 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
 869
 870                 if (ct_state & ~CT_SUPPORTED_MASK) {
 871                         OVS_NLERR(log, "ct_state flags %08x unsupported",
 872                                   ct_state);
 873                         return -EINVAL;
 874                 }
 875
 876                 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
 877                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
 878         }
 879         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
 880             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
 881                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
 882
 883                 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
 884                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
 885         }
 886         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
 887             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
 888                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
 889
 890                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
 891                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
 892         }
 893         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
 894             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
 895                 const struct ovs_key_ct_labels *cl;
 896
 897                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
 898                 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
 899                                    sizeof(*cl), is_mask);
 900                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
 901         }
 902         return 0;
 903 }
 904
 905 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
 906                                 u64 attrs, const struct nlattr **a,
 907                                 bool is_mask, bool log)
 908 {
 909         int err;
 910
 911         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
 912         if (err)
 913                 return err;
 914
 915         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
 916                 const struct ovs_key_ethernet *eth_key;
 917
 918                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 919                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 920                                 eth_key->eth_src, ETH_ALEN, is_mask);
 921                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 922                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 923                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
 924         }
 925
 926         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
 927                 __be16 tci;
 928
 929                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 930                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 931                         if (is_mask)
 932                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 933                         else
 934                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 935
 936                         return -EINVAL;
 937                 }
 938
 939                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 940                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
 941         }
 942
 943         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
 944                 __be16 eth_type;
 945
 946                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 947                 if (is_mask) {
 948                         /* Always exact match EtherType. */
 949                         eth_type = htons(0xffff);
 950                 } else if (!eth_proto_is_802_3(eth_type)) {
 951                         OVS_NLERR(log, "EtherType %x is less than min %x",
 952                                   ntohs(eth_type), ETH_P_802_3_MIN);
 953                         return -EINVAL;
 954                 }
 955
 956                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 957                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
 958         } else if (!is_mask) {
 959                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 960         }
 961
 962         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
 963                 const struct ovs_key_ipv4 *ipv4_key;
 964
 965                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 966                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 967                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 968                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 969                         return -EINVAL;
 970                 }
 971                 SW_FLOW_KEY_PUT(match, ip.proto,
 972                                 ipv4_key->ipv4_proto, is_mask);
 973                 SW_FLOW_KEY_PUT(match, ip.tos,
 974                                 ipv4_key->ipv4_tos, is_mask);
 975                 SW_FLOW_KEY_PUT(match, ip.ttl,
 976                                 ipv4_key->ipv4_ttl, is_mask);
 977                 SW_FLOW_KEY_PUT(match, ip.frag,
 978                                 ipv4_key->ipv4_frag, is_mask);
 979                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 980                                 ipv4_key->ipv4_src, is_mask);
 981                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 982                                 ipv4_key->ipv4_dst, is_mask);
 983                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
 984         }
 985
 986         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
 987                 const struct ovs_key_ipv6 *ipv6_key;
 988
 989                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 990                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 991                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 992                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 993                         return -EINVAL;
 994                 }
 995
 996                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 997                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
 998                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
 999                         return -EINVAL;
1000                 }
1001
1002                 SW_FLOW_KEY_PUT(match, ipv6.label,
1003                                 ipv6_key->ipv6_label, is_mask);
1004                 SW_FLOW_KEY_PUT(match, ip.proto,
1005                                 ipv6_key->ipv6_proto, is_mask);
1006                 SW_FLOW_KEY_PUT(match, ip.tos,
1007                                 ipv6_key->ipv6_tclass, is_mask);
1008                 SW_FLOW_KEY_PUT(match, ip.ttl,
1009                                 ipv6_key->ipv6_hlimit, is_mask);
1010                 SW_FLOW_KEY_PUT(match, ip.frag,
1011                                 ipv6_key->ipv6_frag, is_mask);
1012                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1013                                 ipv6_key->ipv6_src,
1014                                 sizeof(match->key->ipv6.addr.src),
1015                                 is_mask);
1016                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1017                                 ipv6_key->ipv6_dst,
1018                                 sizeof(match->key->ipv6.addr.dst),
1019                                 is_mask);
1020
1021                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1022         }
1023
1024         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1025                 const struct ovs_key_arp *arp_key;
1026
1027                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1028                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1029                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1030                                   arp_key->arp_op);
1031                         return -EINVAL;
1032                 }
1033
1034                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1035                                 arp_key->arp_sip, is_mask);
1036                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1037                         arp_key->arp_tip, is_mask);
1038                 SW_FLOW_KEY_PUT(match, ip.proto,
1039                                 ntohs(arp_key->arp_op), is_mask);
1040                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1041                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1042                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1043                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1044
1045                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1046         }
1047
1048         if (attrs & (1ULL << OVS_KEY_ATTR_MPLS)) {
1049                 const struct ovs_key_mpls *mpls_key;
1050
1051                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1052                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1053                                 mpls_key->mpls_lse, is_mask);
1054
1055                 attrs &= ~(1ULL << OVS_KEY_ATTR_MPLS);
1056         }
1057
1058         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1059                 const struct ovs_key_tcp *tcp_key;
1060
1061                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1062                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1063                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1064                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1065         }
1066
1067         if (attrs & (1ULL << OVS_KEY_ATTR_TCP_FLAGS)) {
1068                 SW_FLOW_KEY_PUT(match, tp.flags,
1069                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1070                                 is_mask);
1071                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP_FLAGS);
1072         }
1073
1074         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1075                 const struct ovs_key_udp *udp_key;
1076
1077                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1078                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1079                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1080                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1081         }
1082
1083         if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
1084                 const struct ovs_key_sctp *sctp_key;
1085
1086                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1087                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1088                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1089                 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
1090         }
1091
1092         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1093                 const struct ovs_key_icmp *icmp_key;
1094
1095                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1096                 SW_FLOW_KEY_PUT(match, tp.src,
1097                                 htons(icmp_key->icmp_type), is_mask);
1098                 SW_FLOW_KEY_PUT(match, tp.dst,
1099                                 htons(icmp_key->icmp_code), is_mask);
1100                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1101         }
1102
1103         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1104                 const struct ovs_key_icmpv6 *icmpv6_key;
1105
1106                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1107                 SW_FLOW_KEY_PUT(match, tp.src,
1108                                 htons(icmpv6_key->icmpv6_type), is_mask);
1109                 SW_FLOW_KEY_PUT(match, tp.dst,
1110                                 htons(icmpv6_key->icmpv6_code), is_mask);
1111                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1112         }
1113
1114         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1115                 const struct ovs_key_nd *nd_key;
1116
1117                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1118                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1119                         nd_key->nd_target,
1120                         sizeof(match->key->ipv6.nd.target),
1121                         is_mask);
1122                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1123                         nd_key->nd_sll, ETH_ALEN, is_mask);
1124                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1125                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1126                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1127         }
1128
1129         if (attrs != 0) {
1130                 OVS_NLERR(log, "Unknown key attributes %llx",
1131                           (unsigned long long)attrs);
1132                 return -EINVAL;
1133         }
1134
1135         return 0;
1136 }
1137
1138 static void nlattr_set(struct nlattr *attr, u8 val,
1139                        const struct ovs_len_tbl *tbl)
1140 {
1141         struct nlattr *nla;
1142         int rem;
1143
1144         /* The nlattr stream should already have been validated */
1145         nla_for_each_nested(nla, attr, rem) {
1146                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1147                         if (tbl[nla_type(nla)].next)
1148                                 tbl = tbl[nla_type(nla)].next;
1149                         nlattr_set(nla, val, tbl);
1150                 } else {
1151                         memset(nla_data(nla), val, nla_len(nla));
1152                 }
1153
1154                 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1155                         *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1156         }
1157 }
1158
1159 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1160 {
1161         nlattr_set(attr, val, ovs_key_lens);
1162 }
1163
1164 /**
1165  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1166  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1167  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1168  * does not include any don't care bit.
1169  * @net: Used to determine per-namespace field support.
1170  * @match: receives the extracted flow match information.
1171  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1172  * sequence. The fields should of the packet that triggered the creation
1173  * of this flow.
1174  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1175  * attribute specifies the mask field of the wildcarded flow.
1176  * @log: Boolean to allow kernel error logging.  Normally true, but when
1177  * probing for feature compatibility this should be passed in as false to
1178  * suppress unnecessary error logging.
1179  */
1180 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1181                       const struct nlattr *nla_key,
1182                       const struct nlattr *nla_mask,
1183                       bool log)
1184 {
1185         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1186         const struct nlattr *encap;
1187         struct nlattr *newmask = NULL;
1188         u64 key_attrs = 0;
1189         u64 mask_attrs = 0;
1190         bool encap_valid = false;
1191         int err;
1192
1193         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1194         if (err)
1195                 return err;
1196
1197         if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1198             (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
1199             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1200                 __be16 tci;
1201
1202                 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
1203                       (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
1204                         OVS_NLERR(log, "Invalid Vlan frame.");
1205                         return -EINVAL;
1206                 }
1207
1208                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1209                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1210                 encap = a[OVS_KEY_ATTR_ENCAP];
1211                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1212                 encap_valid = true;
1213
1214                 if (tci & htons(VLAN_TAG_PRESENT)) {
1215                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1216                         if (err)
1217                                 return err;
1218                 } else if (!tci) {
1219                         /* Corner case for truncated 802.1Q header. */
1220                         if (nla_len(encap)) {
1221                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1222                                 return -EINVAL;
1223                         }
1224                 } else {
1225                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1226                         return  -EINVAL;
1227                 }
1228         }
1229
1230         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1231         if (err)
1232                 return err;
1233
1234         if (match->mask) {
1235                 if (!nla_mask) {
1236                         /* Create an exact match mask. We need to set to 0xff
1237                          * all the 'match->mask' fields that have been touched
1238                          * in 'match->key'. We cannot simply memset
1239                          * 'match->mask', because padding bytes and fields not
1240                          * specified in 'match->key' should be left to 0.
1241                          * Instead, we use a stream of netlink attributes,
1242                          * copied from 'key' and set to 0xff.
1243                          * ovs_key_from_nlattrs() will take care of filling
1244                          * 'match->mask' appropriately.
1245                          */
1246                         newmask = kmemdup(nla_key,
1247                                           nla_total_size(nla_len(nla_key)),
1248                                           GFP_KERNEL);
1249                         if (!newmask)
1250                                 return -ENOMEM;
1251
1252                         mask_set_nlattr(newmask, 0xff);
1253
1254                         /* The userspace does not send tunnel attributes that
1255                          * are 0, but we should not wildcard them nonetheless.
1256                          */
1257                         if (match->key->tun_key.u.ipv4.dst)
1258                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1259                                                          0xff, true);
1260
1261                         nla_mask = newmask;
1262                 }
1263
1264                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1265                 if (err)
1266                         goto free_newmask;
1267
1268                 /* Always match on tci. */
1269                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1270
1271                 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1272                         __be16 eth_type = 0;
1273                         __be16 tci = 0;
1274
1275                         if (!encap_valid) {
1276                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1277                                 err = -EINVAL;
1278                                 goto free_newmask;
1279                         }
1280
1281                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1282                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1283                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1284
1285                         if (eth_type == htons(0xffff)) {
1286                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1287                                 encap = a[OVS_KEY_ATTR_ENCAP];
1288                                 err = parse_flow_mask_nlattrs(encap, a,
1289                                                               &mask_attrs, log);
1290                                 if (err)
1291                                         goto free_newmask;
1292                         } else {
1293                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1294                                           ntohs(eth_type));
1295                                 err = -EINVAL;
1296                                 goto free_newmask;
1297                         }
1298
1299                         if (a[OVS_KEY_ATTR_VLAN])
1300                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1301
1302                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1303                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1304                                           ntohs(tci));
1305                                 err = -EINVAL;
1306                                 goto free_newmask;
1307                         }
1308                 }
1309
1310                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1311                                            log);
1312                 if (err)
1313                         goto free_newmask;
1314         }
1315
1316         if (!match_validate(match, key_attrs, mask_attrs, log))
1317                 err = -EINVAL;
1318
1319 free_newmask:
1320         kfree(newmask);
1321         return err;
1322 }
1323
1324 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1325 {
1326         size_t len;
1327
1328         if (!attr)
1329                 return 0;
1330
1331         len = nla_len(attr);
1332         if (len < 1 || len > MAX_UFID_LENGTH) {
1333                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1334                           nla_len(attr), MAX_UFID_LENGTH);
1335                 return 0;
1336         }
1337
1338         return len;
1339 }
1340
1341 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1342  * or false otherwise.
1343  */
1344 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1345                       bool log)
1346 {
1347         sfid->ufid_len = get_ufid_len(attr, log);
1348         if (sfid->ufid_len)
1349                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1350
1351         return sfid->ufid_len;
1352 }
1353
1354 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1355                            const struct sw_flow_key *key, bool log)
1356 {
1357         struct sw_flow_key *new_key;
1358
1359         if (ovs_nla_get_ufid(sfid, ufid, log))
1360                 return 0;
1361
1362         /* If UFID was not provided, use unmasked key. */
1363         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1364         if (!new_key)
1365                 return -ENOMEM;
1366         memcpy(new_key, key, sizeof(*key));
1367         sfid->unmasked_key = new_key;
1368
1369         return 0;
1370 }
1371
1372 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1373 {
1374         return attr ? nla_get_u32(attr) : 0;
1375 }
1376
1377 /**
1378  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1379  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1380  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1381  * sequence.
1382  * @log: Boolean to allow kernel error logging.  Normally true, but when
1383  * probing for feature compatibility this should be passed in as false to
1384  * suppress unnecessary error logging.
1385  *
1386  * This parses a series of Netlink attributes that form a flow key, which must
1387  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1388  * get the metadata, that is, the parts of the flow key that cannot be
1389  * extracted from the packet itself.
1390  */
1391
1392 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1393                               struct sw_flow_key *key,
1394                               bool log)
1395 {
1396         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1397         struct sw_flow_match match;
1398         u64 attrs = 0;
1399         int err;
1400
1401         err = parse_flow_nlattrs(attr, a, &attrs, log);
1402         if (err)
1403                 return -EINVAL;
1404
1405         memset(&match, 0, sizeof(match));
1406         match.key = key;
1407
1408         memset(key, 0, OVS_SW_FLOW_KEY_METADATA_SIZE);
1409         memset(&key->ct, 0, sizeof(key->ct));
1410         key->phy.in_port = DP_MAX_PORTS;
1411
1412         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1413 }
1414
1415 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1416                              const struct sw_flow_key *output, bool is_mask,
1417                              struct sk_buff *skb)
1418 {
1419         struct ovs_key_ethernet *eth_key;
1420         struct nlattr *nla, *encap;
1421
1422         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1423                 goto nla_put_failure;
1424
1425         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1426                 goto nla_put_failure;
1427
1428         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1429                 goto nla_put_failure;
1430
1431         if ((swkey->tun_proto || is_mask)) {
1432                 const void *opts = NULL;
1433
1434                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1435                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1436
1437                 if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
1438                                      swkey->tun_opts_len, swkey->tun_proto))
1439                         goto nla_put_failure;
1440         }
1441
1442         if (swkey->phy.in_port == DP_MAX_PORTS) {
1443                 if (is_mask && (output->phy.in_port == 0xffff))
1444                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1445                                 goto nla_put_failure;
1446         } else {
1447                 u16 upper_u16;
1448                 upper_u16 = !is_mask ? 0 : 0xffff;
1449
1450                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1451                                 (upper_u16 << 16) | output->phy.in_port))
1452                         goto nla_put_failure;
1453         }
1454
1455         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1456                 goto nla_put_failure;
1457
1458         if (ovs_ct_put_key(output, skb))
1459                 goto nla_put_failure;
1460
1461         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1462         if (!nla)
1463                 goto nla_put_failure;
1464
1465         eth_key = nla_data(nla);
1466         ether_addr_copy(eth_key->eth_src, output->eth.src);
1467         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1468
1469         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1470                 __be16 eth_type;
1471                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1472                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1473                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1474                         goto nla_put_failure;
1475                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1476                 if (!swkey->eth.tci)
1477                         goto unencap;
1478         } else
1479                 encap = NULL;
1480
1481         if (swkey->eth.type == htons(ETH_P_802_2)) {
1482                 /*
1483                  * Ethertype 802.2 is represented in the netlink with omitted
1484                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1485                  * 0xffff in the mask attribute.  Ethertype can also
1486                  * be wildcarded.
1487                  */
1488                 if (is_mask && output->eth.type)
1489                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1490                                                 output->eth.type))
1491                                 goto nla_put_failure;
1492                 goto unencap;
1493         }
1494
1495         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1496                 goto nla_put_failure;
1497
1498         if (swkey->eth.type == htons(ETH_P_IP)) {
1499                 struct ovs_key_ipv4 *ipv4_key;
1500
1501                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1502                 if (!nla)
1503                         goto nla_put_failure;
1504                 ipv4_key = nla_data(nla);
1505                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1506                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1507                 ipv4_key->ipv4_proto = output->ip.proto;
1508                 ipv4_key->ipv4_tos = output->ip.tos;
1509                 ipv4_key->ipv4_ttl = output->ip.ttl;
1510                 ipv4_key->ipv4_frag = output->ip.frag;
1511         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1512                 struct ovs_key_ipv6 *ipv6_key;
1513
1514                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1515                 if (!nla)
1516                         goto nla_put_failure;
1517                 ipv6_key = nla_data(nla);
1518                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1519                                 sizeof(ipv6_key->ipv6_src));
1520                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1521                                 sizeof(ipv6_key->ipv6_dst));
1522                 ipv6_key->ipv6_label = output->ipv6.label;
1523                 ipv6_key->ipv6_proto = output->ip.proto;
1524                 ipv6_key->ipv6_tclass = output->ip.tos;
1525                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1526                 ipv6_key->ipv6_frag = output->ip.frag;
1527         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1528                    swkey->eth.type == htons(ETH_P_RARP)) {
1529                 struct ovs_key_arp *arp_key;
1530
1531                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1532                 if (!nla)
1533                         goto nla_put_failure;
1534                 arp_key = nla_data(nla);
1535                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1536                 arp_key->arp_sip = output->ipv4.addr.src;
1537                 arp_key->arp_tip = output->ipv4.addr.dst;
1538                 arp_key->arp_op = htons(output->ip.proto);
1539                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1540                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1541         } else if (eth_p_mpls(swkey->eth.type)) {
1542                 struct ovs_key_mpls *mpls_key;
1543
1544                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1545                 if (!nla)
1546                         goto nla_put_failure;
1547                 mpls_key = nla_data(nla);
1548                 mpls_key->mpls_lse = output->mpls.top_lse;
1549         }
1550
1551         if ((swkey->eth.type == htons(ETH_P_IP) ||
1552              swkey->eth.type == htons(ETH_P_IPV6)) &&
1553              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1554
1555                 if (swkey->ip.proto == IPPROTO_TCP) {
1556                         struct ovs_key_tcp *tcp_key;
1557
1558                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1559                         if (!nla)
1560                                 goto nla_put_failure;
1561                         tcp_key = nla_data(nla);
1562                         tcp_key->tcp_src = output->tp.src;
1563                         tcp_key->tcp_dst = output->tp.dst;
1564                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1565                                          output->tp.flags))
1566                                 goto nla_put_failure;
1567                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1568                         struct ovs_key_udp *udp_key;
1569
1570                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1571                         if (!nla)
1572                                 goto nla_put_failure;
1573                         udp_key = nla_data(nla);
1574                         udp_key->udp_src = output->tp.src;
1575                         udp_key->udp_dst = output->tp.dst;
1576                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1577                         struct ovs_key_sctp *sctp_key;
1578
1579                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1580                         if (!nla)
1581                                 goto nla_put_failure;
1582                         sctp_key = nla_data(nla);
1583                         sctp_key->sctp_src = output->tp.src;
1584                         sctp_key->sctp_dst = output->tp.dst;
1585                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1586                            swkey->ip.proto == IPPROTO_ICMP) {
1587                         struct ovs_key_icmp *icmp_key;
1588
1589                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1590                         if (!nla)
1591                                 goto nla_put_failure;
1592                         icmp_key = nla_data(nla);
1593                         icmp_key->icmp_type = ntohs(output->tp.src);
1594                         icmp_key->icmp_code = ntohs(output->tp.dst);
1595                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1596                            swkey->ip.proto == IPPROTO_ICMPV6) {
1597                         struct ovs_key_icmpv6 *icmpv6_key;
1598
1599                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1600                                                 sizeof(*icmpv6_key));
1601                         if (!nla)
1602                                 goto nla_put_failure;
1603                         icmpv6_key = nla_data(nla);
1604                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1605                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1606
1607                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1608                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1609                                 struct ovs_key_nd *nd_key;
1610
1611                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1612                                 if (!nla)
1613                                         goto nla_put_failure;
1614                                 nd_key = nla_data(nla);
1615                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1616                                                         sizeof(nd_key->nd_target));
1617                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1618                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1619                         }
1620                 }
1621         }
1622
1623 unencap:
1624         if (encap)
1625                 nla_nest_end(skb, encap);
1626
1627         return 0;
1628
1629 nla_put_failure:
1630         return -EMSGSIZE;
1631 }
1632
1633 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1634                     const struct sw_flow_key *output, int attr, bool is_mask,
1635                     struct sk_buff *skb)
1636 {
1637         int err;
1638         struct nlattr *nla;
1639
1640         nla = nla_nest_start(skb, attr);
1641         if (!nla)
1642                 return -EMSGSIZE;
1643         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1644         if (err)
1645                 return err;
1646         nla_nest_end(skb, nla);
1647
1648         return 0;
1649 }
1650
1651 /* Called with ovs_mutex or RCU read lock. */
1652 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1653 {
1654         if (ovs_identifier_is_ufid(&flow->id))
1655                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1656                                flow->id.ufid);
1657
1658         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1659                                OVS_FLOW_ATTR_KEY, false, skb);
1660 }
1661
1662 /* Called with ovs_mutex or RCU read lock. */
1663 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1664 {
1665         return ovs_nla_put_key(&flow->key, &flow->key,
1666                                 OVS_FLOW_ATTR_KEY, false, skb);
1667 }
1668
1669 /* Called with ovs_mutex or RCU read lock. */
1670 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1671 {
1672         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1673                                 OVS_FLOW_ATTR_MASK, true, skb);
1674 }
1675
1676 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1677
1678 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1679 {
1680         struct sw_flow_actions *sfa;
1681
1682         if (size > MAX_ACTIONS_BUFSIZE) {
1683                 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1684                 return ERR_PTR(-EINVAL);
1685         }
1686
1687         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1688         if (!sfa)
1689                 return ERR_PTR(-ENOMEM);
1690
1691         sfa->actions_len = 0;
1692         return sfa;
1693 }
1694
1695 static void ovs_nla_free_set_action(const struct nlattr *a)
1696 {
1697         const struct nlattr *ovs_key = nla_data(a);
1698         struct ovs_tunnel_info *ovs_tun;
1699
1700         switch (nla_type(ovs_key)) {
1701         case OVS_KEY_ATTR_TUNNEL_INFO:
1702                 ovs_tun = nla_data(ovs_key);
1703                 ovs_dst_release((struct dst_entry *)ovs_tun->tun_dst);
1704                 break;
1705         }
1706 }
1707
1708 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1709 {
1710         const struct nlattr *a;
1711         int rem;
1712
1713         if (!sf_acts)
1714                 return;
1715
1716         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1717                 switch (nla_type(a)) {
1718                 case OVS_ACTION_ATTR_SET:
1719                         ovs_nla_free_set_action(a);
1720                         break;
1721                 case OVS_ACTION_ATTR_CT:
1722                         ovs_ct_free_action(a);
1723                         break;
1724                 }
1725         }
1726
1727         kfree(sf_acts);
1728 }
1729
1730 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1731 {
1732         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1733 }
1734
1735 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1736  * The caller must hold rcu_read_lock for this to be sensible. */
1737 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1738 {
1739         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1740 }
1741
1742 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1743                                        int attr_len, bool log)
1744 {
1745
1746         struct sw_flow_actions *acts;
1747         int new_acts_size;
1748         int req_size = NLA_ALIGN(attr_len);
1749         int next_offset = offsetof(struct sw_flow_actions, actions) +
1750                                         (*sfa)->actions_len;
1751
1752         if (req_size <= (ksize(*sfa) - next_offset))
1753                 goto out;
1754
1755         new_acts_size = ksize(*sfa) * 2;
1756
1757         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1758                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1759                         return ERR_PTR(-EMSGSIZE);
1760                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1761         }
1762
1763         acts = nla_alloc_flow_actions(new_acts_size, log);
1764         if (IS_ERR(acts))
1765                 return (void *)acts;
1766
1767         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1768         acts->actions_len = (*sfa)->actions_len;
1769         acts->orig_len = (*sfa)->orig_len;
1770         kfree(*sfa);
1771         *sfa = acts;
1772
1773 out:
1774         (*sfa)->actions_len += req_size;
1775         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1776 }
1777
1778 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1779                                    int attrtype, void *data, int len, bool log)
1780 {
1781         struct nlattr *a;
1782
1783         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1784         if (IS_ERR(a))
1785                 return a;
1786
1787         a->nla_type = attrtype;
1788         a->nla_len = nla_attr_size(len);
1789
1790         if (data)
1791                 memcpy(nla_data(a), data, len);
1792         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1793
1794         return a;
1795 }
1796
1797 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1798                        int len, bool log)
1799 {
1800         struct nlattr *a;
1801
1802         a = __add_action(sfa, attrtype, data, len, log);
1803         if (IS_ERR(a))
1804                 return PTR_ERR(a);
1805
1806         return 0;
1807 }
1808
1809 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1810                                           int attrtype, bool log)
1811 {
1812         int used = (*sfa)->actions_len;
1813         int err;
1814
1815         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1816         if (err)
1817                 return err;
1818
1819         return used;
1820 }
1821
1822 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1823                                          int st_offset)
1824 {
1825         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1826                                                                st_offset);
1827
1828         a->nla_len = sfa->actions_len - st_offset;
1829 }
1830
1831 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1832                                   const struct sw_flow_key *key,
1833                                   int depth, struct sw_flow_actions **sfa,
1834                                   __be16 eth_type, __be16 vlan_tci, bool log);
1835
1836 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1837                                     const struct sw_flow_key *key, int depth,
1838                                     struct sw_flow_actions **sfa,
1839                                     __be16 eth_type, __be16 vlan_tci, bool log)
1840 {
1841         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1842         const struct nlattr *probability, *actions;
1843         const struct nlattr *a;
1844         int rem, start, err, st_acts;
1845
1846         memset(attrs, 0, sizeof(attrs));
1847         nla_for_each_nested(a, attr, rem) {
1848                 int type = nla_type(a);
1849                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1850                         return -EINVAL;
1851                 attrs[type] = a;
1852         }
1853         if (rem)
1854                 return -EINVAL;
1855
1856         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1857         if (!probability || nla_len(probability) != sizeof(u32))
1858                 return -EINVAL;
1859
1860         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1861         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1862                 return -EINVAL;
1863
1864         /* validation done, copy sample action. */
1865         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1866         if (start < 0)
1867                 return start;
1868         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1869                                  nla_data(probability), sizeof(u32), log);
1870         if (err)
1871                 return err;
1872         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1873         if (st_acts < 0)
1874                 return st_acts;
1875
1876         err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1877                                      eth_type, vlan_tci, log);
1878         if (err)
1879                 return err;
1880
1881         add_nested_action_end(*sfa, st_acts);
1882         add_nested_action_end(*sfa, start);
1883
1884         return 0;
1885 }
1886
1887 void ovs_match_init(struct sw_flow_match *match,
1888                     struct sw_flow_key *key,
1889                     struct sw_flow_mask *mask)
1890 {
1891         memset(match, 0, sizeof(*match));
1892         match->key = key;
1893         match->mask = mask;
1894
1895         memset(key, 0, sizeof(*key));
1896
1897         if (mask) {
1898                 memset(&mask->key, 0, sizeof(mask->key));
1899                 mask->range.start = mask->range.end = 0;
1900         }
1901 }
1902
1903 static int validate_geneve_opts(struct sw_flow_key *key)
1904 {
1905         struct geneve_opt *option;
1906         int opts_len = key->tun_opts_len;
1907         bool crit_opt = false;
1908
1909         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1910         while (opts_len > 0) {
1911                 int len;
1912
1913                 if (opts_len < sizeof(*option))
1914                         return -EINVAL;
1915
1916                 len = sizeof(*option) + option->length * 4;
1917                 if (len > opts_len)
1918                         return -EINVAL;
1919
1920                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1921
1922                 option = (struct geneve_opt *)((u8 *)option + len);
1923                 opts_len -= len;
1924         };
1925
1926         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1927
1928         return 0;
1929 }
1930
1931 static int validate_and_copy_set_tun(const struct nlattr *attr,
1932                                      struct sw_flow_actions **sfa, bool log)
1933 {
1934         struct sw_flow_match match;
1935         struct sw_flow_key key;
1936         struct metadata_dst *tun_dst;
1937         struct ip_tunnel_info *tun_info;
1938         struct ovs_tunnel_info *ovs_tun;
1939         struct nlattr *a;
1940         int err = 0, start, opts_type;
1941
1942         ovs_match_init(&match, &key, NULL);
1943         opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
1944         if (opts_type < 0)
1945                 return opts_type;
1946
1947         if (key.tun_opts_len) {
1948                 switch (opts_type) {
1949                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1950                         err = validate_geneve_opts(&key);
1951                         if (err < 0)
1952                                 return err;
1953                         break;
1954                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1955                         break;
1956                 }
1957         };
1958
1959         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1960         if (start < 0)
1961                 return start;
1962
1963         tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1964         if (!tun_dst)
1965                 return -ENOMEM;
1966
1967         err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
1968         if (err) {
1969                 dst_release((struct dst_entry *)tun_dst);
1970                 return err;
1971         }
1972         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1973                          sizeof(*ovs_tun), log);
1974         if (IS_ERR(a)) {
1975                 ovs_dst_release((struct dst_entry *)tun_dst);
1976                 return PTR_ERR(a);
1977         }
1978
1979         ovs_tun = nla_data(a);
1980         ovs_tun->tun_dst = tun_dst;
1981
1982         tun_info = &tun_dst->u.tun_info;
1983         tun_info->mode = IP_TUNNEL_INFO_TX;
1984         if (key.tun_proto == AF_INET6)
1985                 tun_info->mode |= IP_TUNNEL_INFO_IPV6;
1986         tun_info->key = key.tun_key;
1987
1988         /* We need to store the options in the action itself since
1989          * everything else will go away after flow setup. We can append
1990          * it to tun_info and then point there.
1991          */
1992         ip_tunnel_info_opts_set(tun_info,
1993                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
1994                                 key.tun_opts_len);
1995         add_nested_action_end(*sfa, start);
1996
1997         return err;
1998 }
1999
2000 /* Return false if there are any non-masked bits set.
2001  * Mask follows data immediately, before any netlink padding.
2002  */
2003 static bool validate_masked(u8 *data, int len)
2004 {
2005         u8 *mask = data + len;
2006
2007         while (len--)
2008                 if (*data++ & ~*mask++)
2009                         return false;
2010
2011         return true;
2012 }
2013
2014 static int validate_set(const struct nlattr *a,
2015                         const struct sw_flow_key *flow_key,
2016                         struct sw_flow_actions **sfa,
2017                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
2018 {
2019         const struct nlattr *ovs_key = nla_data(a);
2020         int key_type = nla_type(ovs_key);
2021         size_t key_len;
2022
2023         /* There can be only one key in a action */
2024         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2025                 return -EINVAL;
2026
2027         key_len = nla_len(ovs_key);
2028         if (masked)
2029                 key_len /= 2;
2030
2031         if (key_type > OVS_KEY_ATTR_MAX ||
2032             !check_attr_len(key_len, ovs_key_lens[key_type].len))
2033                 return -EINVAL;
2034
2035         if (masked && !validate_masked(nla_data(ovs_key), key_len))
2036                 return -EINVAL;
2037
2038         switch (key_type) {
2039         const struct ovs_key_ipv4 *ipv4_key;
2040         const struct ovs_key_ipv6 *ipv6_key;
2041         int err;
2042
2043         case OVS_KEY_ATTR_PRIORITY:
2044         case OVS_KEY_ATTR_SKB_MARK:
2045         case OVS_KEY_ATTR_CT_MARK:
2046         case OVS_KEY_ATTR_CT_LABELS:
2047         case OVS_KEY_ATTR_ETHERNET:
2048                 break;
2049
2050         case OVS_KEY_ATTR_TUNNEL:
2051                 if (eth_p_mpls(eth_type))
2052                         return -EINVAL;
2053
2054                 if (masked)
2055                         return -EINVAL; /* Masked tunnel set not supported. */
2056
2057                 *skip_copy = true;
2058                 err = validate_and_copy_set_tun(a, sfa, log);
2059                 if (err)
2060                         return err;
2061                 break;
2062
2063         case OVS_KEY_ATTR_IPV4:
2064                 if (eth_type != htons(ETH_P_IP))
2065                         return -EINVAL;
2066
2067                 ipv4_key = nla_data(ovs_key);
2068
2069                 if (masked) {
2070                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2071
2072                         /* Non-writeable fields. */
2073                         if (mask->ipv4_proto || mask->ipv4_frag)
2074                                 return -EINVAL;
2075                 } else {
2076                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2077                                 return -EINVAL;
2078
2079                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2080                                 return -EINVAL;
2081                 }
2082                 break;
2083
2084         case OVS_KEY_ATTR_IPV6:
2085                 if (eth_type != htons(ETH_P_IPV6))
2086                         return -EINVAL;
2087
2088                 ipv6_key = nla_data(ovs_key);
2089
2090                 if (masked) {
2091                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2092
2093                         /* Non-writeable fields. */
2094                         if (mask->ipv6_proto || mask->ipv6_frag)
2095                                 return -EINVAL;
2096
2097                         /* Invalid bits in the flow label mask? */
2098                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2099                                 return -EINVAL;
2100                 } else {
2101                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2102                                 return -EINVAL;
2103
2104                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2105                                 return -EINVAL;
2106                 }
2107                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2108                         return -EINVAL;
2109
2110                 break;
2111
2112         case OVS_KEY_ATTR_TCP:
2113                 if ((eth_type != htons(ETH_P_IP) &&
2114                      eth_type != htons(ETH_P_IPV6)) ||
2115                     flow_key->ip.proto != IPPROTO_TCP)
2116                         return -EINVAL;
2117
2118                 break;
2119
2120         case OVS_KEY_ATTR_UDP:
2121                 if ((eth_type != htons(ETH_P_IP) &&
2122                      eth_type != htons(ETH_P_IPV6)) ||
2123                     flow_key->ip.proto != IPPROTO_UDP)
2124                         return -EINVAL;
2125
2126                 break;
2127
2128         case OVS_KEY_ATTR_MPLS:
2129                 if (!eth_p_mpls(eth_type))
2130                         return -EINVAL;
2131                 break;
2132
2133         case OVS_KEY_ATTR_SCTP:
2134                 if ((eth_type != htons(ETH_P_IP) &&
2135                      eth_type != htons(ETH_P_IPV6)) ||
2136                     flow_key->ip.proto != IPPROTO_SCTP)
2137                         return -EINVAL;
2138
2139                 break;
2140
2141         default:
2142                 return -EINVAL;
2143         }
2144
2145         /* Convert non-masked non-tunnel set actions to masked set actions. */
2146         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2147                 int start, len = key_len * 2;
2148                 struct nlattr *at;
2149
2150                 *skip_copy = true;
2151
2152                 start = add_nested_action_start(sfa,
2153                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2154                                                 log);
2155                 if (start < 0)
2156                         return start;
2157
2158                 at = __add_action(sfa, key_type, NULL, len, log);
2159                 if (IS_ERR(at))
2160                         return PTR_ERR(at);
2161
2162                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2163                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2164                 /* Clear non-writeable bits from otherwise writeable fields. */
2165                 if (key_type == OVS_KEY_ATTR_IPV6) {
2166                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2167
2168                         mask->ipv6_label &= htonl(0x000FFFFF);
2169                 }
2170                 add_nested_action_end(*sfa, start);
2171         }
2172
2173         return 0;
2174 }
2175
2176 static int validate_userspace(const struct nlattr *attr)
2177 {
2178         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2179                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2180                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2181                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2182         };
2183         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2184         int error;
2185
2186         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2187                                  attr, userspace_policy);
2188         if (error)
2189                 return error;
2190
2191         if (!a[OVS_USERSPACE_ATTR_PID] ||
2192             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2193                 return -EINVAL;
2194
2195         return 0;
2196 }
2197
2198 static int copy_action(const struct nlattr *from,
2199                        struct sw_flow_actions **sfa, bool log)
2200 {
2201         int totlen = NLA_ALIGN(from->nla_len);
2202         struct nlattr *to;
2203
2204         to = reserve_sfa_size(sfa, from->nla_len, log);
2205         if (IS_ERR(to))
2206                 return PTR_ERR(to);
2207
2208         memcpy(to, from, totlen);
2209         return 0;
2210 }
2211
2212 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2213                                   const struct sw_flow_key *key,
2214                                   int depth, struct sw_flow_actions **sfa,
2215                                   __be16 eth_type, __be16 vlan_tci, bool log)
2216 {
2217         const struct nlattr *a;
2218         int rem, err;
2219
2220         if (depth >= SAMPLE_ACTION_DEPTH)
2221                 return -EOVERFLOW;
2222
2223         nla_for_each_nested(a, attr, rem) {
2224                 /* Expected argument lengths, (u32)-1 for variable length. */
2225                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2226                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2227                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2228                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2229                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2230                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2231                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2232                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2233                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2234                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2235                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2236                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2237                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2238                         [OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
2239                 };
2240                 const struct ovs_action_push_vlan *vlan;
2241                 int type = nla_type(a);
2242                 bool skip_copy;
2243
2244                 if (type > OVS_ACTION_ATTR_MAX ||
2245                     (action_lens[type] != nla_len(a) &&
2246                      action_lens[type] != (u32)-1))
2247                         return -EINVAL;
2248
2249                 skip_copy = false;
2250                 switch (type) {
2251                 case OVS_ACTION_ATTR_UNSPEC:
2252                         return -EINVAL;
2253
2254                 case OVS_ACTION_ATTR_USERSPACE:
2255                         err = validate_userspace(a);
2256                         if (err)
2257                                 return err;
2258                         break;
2259
2260                 case OVS_ACTION_ATTR_OUTPUT:
2261                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2262                                 return -EINVAL;
2263                         break;
2264
2265                 case OVS_ACTION_ATTR_TRUNC: {
2266                         const struct ovs_action_trunc *trunc = nla_data(a);
2267
2268                         if (trunc->max_len < ETH_HLEN)
2269                                 return -EINVAL;
2270                         break;
2271                 }
2272
2273                 case OVS_ACTION_ATTR_HASH: {
2274                         const struct ovs_action_hash *act_hash = nla_data(a);
2275
2276                         switch (act_hash->hash_alg) {
2277                         case OVS_HASH_ALG_L4:
2278                                 break;
2279                         default:
2280                                 return  -EINVAL;
2281                         }
2282
2283                         break;
2284                 }
2285
2286                 case OVS_ACTION_ATTR_POP_VLAN:
2287                         vlan_tci = htons(0);
2288                         break;
2289
2290                 case OVS_ACTION_ATTR_PUSH_VLAN:
2291                         vlan = nla_data(a);
2292                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2293                                 return -EINVAL;
2294                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2295                                 return -EINVAL;
2296                         vlan_tci = vlan->vlan_tci;
2297                         break;
2298
2299                 case OVS_ACTION_ATTR_RECIRC:
2300                         break;
2301
2302                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2303                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2304
2305                         if (!eth_p_mpls(mpls->mpls_ethertype))
2306                                 return -EINVAL;
2307                         /* Prohibit push MPLS other than to a white list
2308                          * for packets that have a known tag order.
2309                          */
2310                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2311                             (eth_type != htons(ETH_P_IP) &&
2312                              eth_type != htons(ETH_P_IPV6) &&
2313                              eth_type != htons(ETH_P_ARP) &&
2314                              eth_type != htons(ETH_P_RARP) &&
2315                              !eth_p_mpls(eth_type)))
2316                                 return -EINVAL;
2317                         eth_type = mpls->mpls_ethertype;
2318                         break;
2319                 }
2320
2321                 case OVS_ACTION_ATTR_POP_MPLS:
2322                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2323                             !eth_p_mpls(eth_type))
2324                                 return -EINVAL;
2325
2326                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2327                          * as there is no check here to ensure that the new
2328                          * eth_type is valid and thus set actions could
2329                          * write off the end of the packet or otherwise
2330                          * corrupt it.
2331                          *
2332                          * Support for these actions is planned using packet
2333                          * recirculation.
2334                          */
2335                         eth_type = htons(0);
2336                         break;
2337
2338                 case OVS_ACTION_ATTR_SET:
2339                         err = validate_set(a, key, sfa,
2340                                            &skip_copy, eth_type, false, log);
2341                         if (err)
2342                                 return err;
2343                         break;
2344
2345                 case OVS_ACTION_ATTR_SET_MASKED:
2346                         err = validate_set(a, key, sfa,
2347                                            &skip_copy, eth_type, true, log);
2348                         if (err)
2349                                 return err;
2350                         break;
2351
2352                 case OVS_ACTION_ATTR_SAMPLE:
2353                         err = validate_and_copy_sample(net, a, key, depth, sfa,
2354                                                        eth_type, vlan_tci, log);
2355                         if (err)
2356                                 return err;
2357                         skip_copy = true;
2358                         break;
2359
2360                 case OVS_ACTION_ATTR_CT:
2361                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2362                         if (err)
2363                                 return err;
2364                         skip_copy = true;
2365                         break;
2366
2367                 default:
2368                         OVS_NLERR(log, "Unknown Action type %d", type);
2369                         return -EINVAL;
2370                 }
2371                 if (!skip_copy) {
2372                         err = copy_action(a, sfa, log);
2373                         if (err)
2374                                 return err;
2375                 }
2376         }
2377
2378         if (rem > 0)
2379                 return -EINVAL;
2380
2381         return 0;
2382 }
2383
2384 /* 'key' must be the masked key. */
2385 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2386                          const struct sw_flow_key *key,
2387                          struct sw_flow_actions **sfa, bool log)
2388 {
2389         int err;
2390
2391         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2392         if (IS_ERR(*sfa))
2393                 return PTR_ERR(*sfa);
2394
2395         (*sfa)->orig_len = nla_len(attr);
2396         err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2397                                      key->eth.tci, log);
2398         if (err)
2399                 ovs_nla_free_flow_actions(*sfa);
2400
2401         return err;
2402 }
2403
2404 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2405 {
2406         const struct nlattr *a;
2407         struct nlattr *start;
2408         int err = 0, rem;
2409
2410         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2411         if (!start)
2412                 return -EMSGSIZE;
2413
2414         nla_for_each_nested(a, attr, rem) {
2415                 int type = nla_type(a);
2416                 struct nlattr *st_sample;
2417
2418                 switch (type) {
2419                 case OVS_SAMPLE_ATTR_PROBABILITY:
2420                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2421                                     sizeof(u32), nla_data(a)))
2422                                 return -EMSGSIZE;
2423                         break;
2424                 case OVS_SAMPLE_ATTR_ACTIONS:
2425                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2426                         if (!st_sample)
2427                                 return -EMSGSIZE;
2428                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2429                         if (err)
2430                                 return err;
2431                         nla_nest_end(skb, st_sample);
2432                         break;
2433                 }
2434         }
2435
2436         nla_nest_end(skb, start);
2437         return err;
2438 }
2439
2440 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2441 {
2442         const struct nlattr *ovs_key = nla_data(a);
2443         int key_type = nla_type(ovs_key);
2444         struct nlattr *start;
2445         int err;
2446
2447         switch (key_type) {
2448         case OVS_KEY_ATTR_TUNNEL_INFO: {
2449                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2450                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2451
2452                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2453                 if (!start)
2454                         return -EMSGSIZE;
2455
2456                 err =  ip_tun_to_nlattr(skb, &tun_info->key,
2457                                         ip_tunnel_info_opts(tun_info),
2458                                         tun_info->options_len,
2459                                         ip_tunnel_info_af(tun_info));
2460                 if (err)
2461                         return err;
2462                 nla_nest_end(skb, start);
2463                 break;
2464         }
2465         default:
2466                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2467                         return -EMSGSIZE;
2468                 break;
2469         }
2470
2471         return 0;
2472 }
2473
2474 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2475                                                 struct sk_buff *skb)
2476 {
2477         const struct nlattr *ovs_key = nla_data(a);
2478         struct nlattr *nla;
2479         size_t key_len = nla_len(ovs_key) / 2;
2480
2481         /* Revert the conversion we did from a non-masked set action to
2482          * masked set action.
2483          */
2484         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2485         if (!nla)
2486                 return -EMSGSIZE;
2487
2488         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2489                 return -EMSGSIZE;
2490
2491         nla_nest_end(skb, nla);
2492         return 0;
2493 }
2494
2495 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2496 {
2497         const struct nlattr *a;
2498         int rem, err;
2499
2500         nla_for_each_attr(a, attr, len, rem) {
2501                 int type = nla_type(a);
2502
2503                 switch (type) {
2504                 case OVS_ACTION_ATTR_SET:
2505                         err = set_action_to_attr(a, skb);
2506                         if (err)
2507                                 return err;
2508                         break;
2509
2510                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2511                         err = masked_set_action_to_set_action_attr(a, skb);
2512                         if (err)
2513                                 return err;
2514                         break;
2515
2516                 case OVS_ACTION_ATTR_SAMPLE:
2517                         err = sample_action_to_attr(a, skb);
2518                         if (err)
2519                                 return err;
2520                         break;
2521
2522                 case OVS_ACTION_ATTR_CT:
2523                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2524                         if (err)
2525                                 return err;
2526                         break;
2527
2528                 default:
2529                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2530                                 return -EMSGSIZE;
2531                         break;
2532                 }
2533         }
2534
2535         return 0;
2536 }