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_64bit(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                          OVS_TUNNEL_KEY_ATTR_PAD))
 727                 return -EMSGSIZE;
 728         switch (tun_proto) {
 729         case AF_INET:
 730                 if (output->u.ipv4.src &&
 731                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
 732                                     output->u.ipv4.src))
 733                         return -EMSGSIZE;
 734                 if (output->u.ipv4.dst &&
 735                     nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
 736                                     output->u.ipv4.dst))
 737                         return -EMSGSIZE;
 738                 break;
 739         case AF_INET6:
 740                 if (!ipv6_addr_any(&output->u.ipv6.src) &&
 741                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
 742                                      &output->u.ipv6.src))
 743                         return -EMSGSIZE;
 744                 if (!ipv6_addr_any(&output->u.ipv6.dst) &&
 745                     nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
 746                                      &output->u.ipv6.dst))
 747                         return -EMSGSIZE;
 748                 break;
 749         }
 750         if (output->tos &&
 751             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
 752                 return -EMSGSIZE;
 753         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
 754                 return -EMSGSIZE;
 755         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
 756             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
 757                 return -EMSGSIZE;
 758         if ((output->tun_flags & TUNNEL_CSUM) &&
 759             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
 760                 return -EMSGSIZE;
 761         if (output->tp_src &&
 762             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
 763                 return -EMSGSIZE;
 764         if (output->tp_dst &&
 765             nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
 766                 return -EMSGSIZE;
 767         if ((output->tun_flags & TUNNEL_OAM) &&
 768             nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
 769                 return -EMSGSIZE;
 770         if (swkey_tun_opts_len) {
 771                 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
 772                     nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
 773                             swkey_tun_opts_len, tun_opts))
 774                         return -EMSGSIZE;
 775                 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
 776                          vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
 777                         return -EMSGSIZE;
 778         }
 779
 780         return 0;
 781 }
 782
 783 static int ip_tun_to_nlattr(struct sk_buff *skb,
 784                             const struct ip_tunnel_key *output,
 785                             const void *tun_opts, int swkey_tun_opts_len,
 786                             unsigned short tun_proto)
 787 {
 788         struct nlattr *nla;
 789         int err;
 790
 791         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
 792         if (!nla)
 793                 return -EMSGSIZE;
 794
 795         err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
 796                                  tun_proto);
 797         if (err)
 798                 return err;
 799
 800         nla_nest_end(skb, nla);
 801         return 0;
 802 }
 803
 804 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
 805                             struct ip_tunnel_info *tun_info)
 806 {
 807         return __ip_tun_to_nlattr(skb, &tun_info->key,
 808                                   ip_tunnel_info_opts(tun_info),
 809                                   tun_info->options_len,
 810                                   ip_tunnel_info_af(tun_info));
 811 }
 812
 813 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
 814                                   u64 *attrs, const struct nlattr **a,
 815                                  bool is_mask, bool log)
 816 {
 817         if (*attrs & (1ULL << OVS_KEY_ATTR_DP_HASH)) {
 818                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 819
 820                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 821                 *attrs &= ~(1ULL << OVS_KEY_ATTR_DP_HASH);
 822         }
 823
 824         if (*attrs & (1ULL << OVS_KEY_ATTR_RECIRC_ID)) {
 825                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 826
 827                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 828                 *attrs &= ~(1ULL << OVS_KEY_ATTR_RECIRC_ID);
 829         }
 830
 831         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
 832                 SW_FLOW_KEY_PUT(match, phy.priority,
 833                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 834                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
 835         }
 836
 837         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
 838                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 839
 840                 if (is_mask) {
 841                         in_port = 0xffffffff; /* Always exact match in_port. */
 842                 } else if (in_port >= DP_MAX_PORTS) {
 843                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
 844                                   in_port, DP_MAX_PORTS);
 845                         return -EINVAL;
 846                 }
 847
 848                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
 849                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
 850         } else if (!is_mask) {
 851                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
 852         }
 853
 854         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
 855                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
 856
 857                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
 858                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
 859         }
 860         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
 861                 if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
 862                                          is_mask, log) < 0)
 863                         return -EINVAL;
 864                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
 865         }
 866
 867         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
 868             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
 869                 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
 870
 871                 if (ct_state & ~CT_SUPPORTED_MASK) {
 872                         OVS_NLERR(log, "ct_state flags %08x unsupported",
 873                                   ct_state);
 874                         return -EINVAL;
 875                 }
 876
 877                 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
 878                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
 879         }
 880         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
 881             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
 882                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
 883
 884                 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
 885                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
 886         }
 887         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
 888             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
 889                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
 890
 891                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
 892                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
 893         }
 894         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
 895             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
 896                 const struct ovs_key_ct_labels *cl;
 897
 898                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
 899                 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
 900                                    sizeof(*cl), is_mask);
 901                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
 902         }
 903         return 0;
 904 }
 905
 906 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
 907                                 u64 attrs, const struct nlattr **a,
 908                                 bool is_mask, bool log)
 909 {
 910         int err;
 911
 912         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
 913         if (err)
 914                 return err;
 915
 916         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
 917                 const struct ovs_key_ethernet *eth_key;
 918
 919                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
 920                 SW_FLOW_KEY_MEMCPY(match, eth.src,
 921                                 eth_key->eth_src, ETH_ALEN, is_mask);
 922                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
 923                                 eth_key->eth_dst, ETH_ALEN, is_mask);
 924                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
 925         }
 926
 927         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
 928                 __be16 tci;
 929
 930                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 931                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
 932                         if (is_mask)
 933                                 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
 934                         else
 935                                 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
 936
 937                         return -EINVAL;
 938                 }
 939
 940                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
 941                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
 942         }
 943
 944         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
 945                 __be16 eth_type;
 946
 947                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 948                 if (is_mask) {
 949                         /* Always exact match EtherType. */
 950                         eth_type = htons(0xffff);
 951                 } else if (!eth_proto_is_802_3(eth_type)) {
 952                         OVS_NLERR(log, "EtherType %x is less than min %x",
 953                                   ntohs(eth_type), ETH_P_802_3_MIN);
 954                         return -EINVAL;
 955                 }
 956
 957                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
 958                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
 959         } else if (!is_mask) {
 960                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
 961         }
 962
 963         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
 964                 const struct ovs_key_ipv4 *ipv4_key;
 965
 966                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
 967                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
 968                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
 969                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
 970                         return -EINVAL;
 971                 }
 972                 SW_FLOW_KEY_PUT(match, ip.proto,
 973                                 ipv4_key->ipv4_proto, is_mask);
 974                 SW_FLOW_KEY_PUT(match, ip.tos,
 975                                 ipv4_key->ipv4_tos, is_mask);
 976                 SW_FLOW_KEY_PUT(match, ip.ttl,
 977                                 ipv4_key->ipv4_ttl, is_mask);
 978                 SW_FLOW_KEY_PUT(match, ip.frag,
 979                                 ipv4_key->ipv4_frag, is_mask);
 980                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
 981                                 ipv4_key->ipv4_src, is_mask);
 982                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
 983                                 ipv4_key->ipv4_dst, is_mask);
 984                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
 985         }
 986
 987         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
 988                 const struct ovs_key_ipv6 *ipv6_key;
 989
 990                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
 991                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
 992                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
 993                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
 994                         return -EINVAL;
 995                 }
 996
 997                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
 998                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
 999                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1000                         return -EINVAL;
1001                 }
1002
1003                 SW_FLOW_KEY_PUT(match, ipv6.label,
1004                                 ipv6_key->ipv6_label, is_mask);
1005                 SW_FLOW_KEY_PUT(match, ip.proto,
1006                                 ipv6_key->ipv6_proto, is_mask);
1007                 SW_FLOW_KEY_PUT(match, ip.tos,
1008                                 ipv6_key->ipv6_tclass, is_mask);
1009                 SW_FLOW_KEY_PUT(match, ip.ttl,
1010                                 ipv6_key->ipv6_hlimit, is_mask);
1011                 SW_FLOW_KEY_PUT(match, ip.frag,
1012                                 ipv6_key->ipv6_frag, is_mask);
1013                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1014                                 ipv6_key->ipv6_src,
1015                                 sizeof(match->key->ipv6.addr.src),
1016                                 is_mask);
1017                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1018                                 ipv6_key->ipv6_dst,
1019                                 sizeof(match->key->ipv6.addr.dst),
1020                                 is_mask);
1021
1022                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1023         }
1024
1025         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1026                 const struct ovs_key_arp *arp_key;
1027
1028                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1029                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1030                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1031                                   arp_key->arp_op);
1032                         return -EINVAL;
1033                 }
1034
1035                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1036                                 arp_key->arp_sip, is_mask);
1037                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1038                         arp_key->arp_tip, is_mask);
1039                 SW_FLOW_KEY_PUT(match, ip.proto,
1040                                 ntohs(arp_key->arp_op), is_mask);
1041                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1042                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1043                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1044                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1045
1046                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1047         }
1048
1049         if (attrs & (1ULL << OVS_KEY_ATTR_MPLS)) {
1050                 const struct ovs_key_mpls *mpls_key;
1051
1052                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1053                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1054                                 mpls_key->mpls_lse, is_mask);
1055
1056                 attrs &= ~(1ULL << OVS_KEY_ATTR_MPLS);
1057         }
1058
1059         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1060                 const struct ovs_key_tcp *tcp_key;
1061
1062                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1063                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1064                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1065                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1066         }
1067
1068         if (attrs & (1ULL << OVS_KEY_ATTR_TCP_FLAGS)) {
1069                 SW_FLOW_KEY_PUT(match, tp.flags,
1070                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1071                                 is_mask);
1072                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP_FLAGS);
1073         }
1074
1075         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1076                 const struct ovs_key_udp *udp_key;
1077
1078                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1079                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1080                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1081                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1082         }
1083
1084         if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
1085                 const struct ovs_key_sctp *sctp_key;
1086
1087                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1088                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1089                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1090                 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
1091         }
1092
1093         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1094                 const struct ovs_key_icmp *icmp_key;
1095
1096                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1097                 SW_FLOW_KEY_PUT(match, tp.src,
1098                                 htons(icmp_key->icmp_type), is_mask);
1099                 SW_FLOW_KEY_PUT(match, tp.dst,
1100                                 htons(icmp_key->icmp_code), is_mask);
1101                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1102         }
1103
1104         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1105                 const struct ovs_key_icmpv6 *icmpv6_key;
1106
1107                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1108                 SW_FLOW_KEY_PUT(match, tp.src,
1109                                 htons(icmpv6_key->icmpv6_type), is_mask);
1110                 SW_FLOW_KEY_PUT(match, tp.dst,
1111                                 htons(icmpv6_key->icmpv6_code), is_mask);
1112                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1113         }
1114
1115         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1116                 const struct ovs_key_nd *nd_key;
1117
1118                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1119                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1120                         nd_key->nd_target,
1121                         sizeof(match->key->ipv6.nd.target),
1122                         is_mask);
1123                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1124                         nd_key->nd_sll, ETH_ALEN, is_mask);
1125                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1126                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1127                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1128         }
1129
1130         if (attrs != 0) {
1131                 OVS_NLERR(log, "Unknown key attributes %llx",
1132                           (unsigned long long)attrs);
1133                 return -EINVAL;
1134         }
1135
1136         return 0;
1137 }
1138
1139 static void nlattr_set(struct nlattr *attr, u8 val,
1140                        const struct ovs_len_tbl *tbl)
1141 {
1142         struct nlattr *nla;
1143         int rem;
1144
1145         /* The nlattr stream should already have been validated */
1146         nla_for_each_nested(nla, attr, rem) {
1147                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1148                         if (tbl[nla_type(nla)].next)
1149                                 tbl = tbl[nla_type(nla)].next;
1150                         nlattr_set(nla, val, tbl);
1151                 } else {
1152                         memset(nla_data(nla), val, nla_len(nla));
1153                 }
1154
1155                 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1156                         *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1157         }
1158 }
1159
1160 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1161 {
1162         nlattr_set(attr, val, ovs_key_lens);
1163 }
1164
1165 /**
1166  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1167  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1168  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1169  * does not include any don't care bit.
1170  * @net: Used to determine per-namespace field support.
1171  * @match: receives the extracted flow match information.
1172  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1173  * sequence. The fields should of the packet that triggered the creation
1174  * of this flow.
1175  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1176  * attribute specifies the mask field of the wildcarded flow.
1177  * @log: Boolean to allow kernel error logging.  Normally true, but when
1178  * probing for feature compatibility this should be passed in as false to
1179  * suppress unnecessary error logging.
1180  */
1181 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1182                       const struct nlattr *nla_key,
1183                       const struct nlattr *nla_mask,
1184                       bool log)
1185 {
1186         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1187         const struct nlattr *encap;
1188         struct nlattr *newmask = NULL;
1189         u64 key_attrs = 0;
1190         u64 mask_attrs = 0;
1191         bool encap_valid = false;
1192         int err;
1193
1194         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1195         if (err)
1196                 return err;
1197
1198         if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1199             (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
1200             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1201                 __be16 tci;
1202
1203                 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
1204                       (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
1205                         OVS_NLERR(log, "Invalid Vlan frame.");
1206                         return -EINVAL;
1207                 }
1208
1209                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1210                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1211                 encap = a[OVS_KEY_ATTR_ENCAP];
1212                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1213                 encap_valid = true;
1214
1215                 if (tci & htons(VLAN_TAG_PRESENT)) {
1216                         err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1217                         if (err)
1218                                 return err;
1219                 } else if (!tci) {
1220                         /* Corner case for truncated 802.1Q header. */
1221                         if (nla_len(encap)) {
1222                                 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1223                                 return -EINVAL;
1224                         }
1225                 } else {
1226                         OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1227                         return  -EINVAL;
1228                 }
1229         }
1230
1231         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1232         if (err)
1233                 return err;
1234
1235         if (match->mask) {
1236                 if (!nla_mask) {
1237                         /* Create an exact match mask. We need to set to 0xff
1238                          * all the 'match->mask' fields that have been touched
1239                          * in 'match->key'. We cannot simply memset
1240                          * 'match->mask', because padding bytes and fields not
1241                          * specified in 'match->key' should be left to 0.
1242                          * Instead, we use a stream of netlink attributes,
1243                          * copied from 'key' and set to 0xff.
1244                          * ovs_key_from_nlattrs() will take care of filling
1245                          * 'match->mask' appropriately.
1246                          */
1247                         newmask = kmemdup(nla_key,
1248                                           nla_total_size(nla_len(nla_key)),
1249                                           GFP_KERNEL);
1250                         if (!newmask)
1251                                 return -ENOMEM;
1252
1253                         mask_set_nlattr(newmask, 0xff);
1254
1255                         /* The userspace does not send tunnel attributes that
1256                          * are 0, but we should not wildcard them nonetheless.
1257                          */
1258                         if (match->key->tun_key.u.ipv4.dst)
1259                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1260                                                          0xff, true);
1261
1262                         nla_mask = newmask;
1263                 }
1264
1265                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1266                 if (err)
1267                         goto free_newmask;
1268
1269                 /* Always match on tci. */
1270                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1271
1272                 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1273                         __be16 eth_type = 0;
1274                         __be16 tci = 0;
1275
1276                         if (!encap_valid) {
1277                                 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1278                                 err = -EINVAL;
1279                                 goto free_newmask;
1280                         }
1281
1282                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1283                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1284                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1285
1286                         if (eth_type == htons(0xffff)) {
1287                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1288                                 encap = a[OVS_KEY_ATTR_ENCAP];
1289                                 err = parse_flow_mask_nlattrs(encap, a,
1290                                                               &mask_attrs, log);
1291                                 if (err)
1292                                         goto free_newmask;
1293                         } else {
1294                                 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1295                                           ntohs(eth_type));
1296                                 err = -EINVAL;
1297                                 goto free_newmask;
1298                         }
1299
1300                         if (a[OVS_KEY_ATTR_VLAN])
1301                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1302
1303                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1304                                 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1305                                           ntohs(tci));
1306                                 err = -EINVAL;
1307                                 goto free_newmask;
1308                         }
1309                 }
1310
1311                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1312                                            log);
1313                 if (err)
1314                         goto free_newmask;
1315         }
1316
1317         if (!match_validate(match, key_attrs, mask_attrs, log))
1318                 err = -EINVAL;
1319
1320 free_newmask:
1321         kfree(newmask);
1322         return err;
1323 }
1324
1325 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1326 {
1327         size_t len;
1328
1329         if (!attr)
1330                 return 0;
1331
1332         len = nla_len(attr);
1333         if (len < 1 || len > MAX_UFID_LENGTH) {
1334                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1335                           nla_len(attr), MAX_UFID_LENGTH);
1336                 return 0;
1337         }
1338
1339         return len;
1340 }
1341
1342 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1343  * or false otherwise.
1344  */
1345 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1346                       bool log)
1347 {
1348         sfid->ufid_len = get_ufid_len(attr, log);
1349         if (sfid->ufid_len)
1350                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1351
1352         return sfid->ufid_len;
1353 }
1354
1355 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1356                            const struct sw_flow_key *key, bool log)
1357 {
1358         struct sw_flow_key *new_key;
1359
1360         if (ovs_nla_get_ufid(sfid, ufid, log))
1361                 return 0;
1362
1363         /* If UFID was not provided, use unmasked key. */
1364         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1365         if (!new_key)
1366                 return -ENOMEM;
1367         memcpy(new_key, key, sizeof(*key));
1368         sfid->unmasked_key = new_key;
1369
1370         return 0;
1371 }
1372
1373 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1374 {
1375         return attr ? nla_get_u32(attr) : 0;
1376 }
1377
1378 /**
1379  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1380  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1381  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1382  * sequence.
1383  * @log: Boolean to allow kernel error logging.  Normally true, but when
1384  * probing for feature compatibility this should be passed in as false to
1385  * suppress unnecessary error logging.
1386  *
1387  * This parses a series of Netlink attributes that form a flow key, which must
1388  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1389  * get the metadata, that is, the parts of the flow key that cannot be
1390  * extracted from the packet itself.
1391  */
1392
1393 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1394                               struct sw_flow_key *key,
1395                               bool log)
1396 {
1397         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1398         struct sw_flow_match match;
1399         u64 attrs = 0;
1400         int err;
1401
1402         err = parse_flow_nlattrs(attr, a, &attrs, log);
1403         if (err)
1404                 return -EINVAL;
1405
1406         memset(&match, 0, sizeof(match));
1407         match.key = key;
1408
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
1804         return PTR_ERR_OR_ZERO(a);
1805 }
1806
1807 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1808                                           int attrtype, bool log)
1809 {
1810         int used = (*sfa)->actions_len;
1811         int err;
1812
1813         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1814         if (err)
1815                 return err;
1816
1817         return used;
1818 }
1819
1820 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1821                                          int st_offset)
1822 {
1823         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1824                                                                st_offset);
1825
1826         a->nla_len = sfa->actions_len - st_offset;
1827 }
1828
1829 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1830                                   const struct sw_flow_key *key,
1831                                   int depth, struct sw_flow_actions **sfa,
1832                                   __be16 eth_type, __be16 vlan_tci, bool log);
1833
1834 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1835                                     const struct sw_flow_key *key, int depth,
1836                                     struct sw_flow_actions **sfa,
1837                                     __be16 eth_type, __be16 vlan_tci, bool log)
1838 {
1839         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1840         const struct nlattr *probability, *actions;
1841         const struct nlattr *a;
1842         int rem, start, err, st_acts;
1843
1844         memset(attrs, 0, sizeof(attrs));
1845         nla_for_each_nested(a, attr, rem) {
1846                 int type = nla_type(a);
1847                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1848                         return -EINVAL;
1849                 attrs[type] = a;
1850         }
1851         if (rem)
1852                 return -EINVAL;
1853
1854         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1855         if (!probability || nla_len(probability) != sizeof(u32))
1856                 return -EINVAL;
1857
1858         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1859         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1860                 return -EINVAL;
1861
1862         /* validation done, copy sample action. */
1863         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1864         if (start < 0)
1865                 return start;
1866         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1867                                  nla_data(probability), sizeof(u32), log);
1868         if (err)
1869                 return err;
1870         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1871         if (st_acts < 0)
1872                 return st_acts;
1873
1874         err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1875                                      eth_type, vlan_tci, log);
1876         if (err)
1877                 return err;
1878
1879         add_nested_action_end(*sfa, st_acts);
1880         add_nested_action_end(*sfa, start);
1881
1882         return 0;
1883 }
1884
1885 void ovs_match_init(struct sw_flow_match *match,
1886                     struct sw_flow_key *key,
1887                     struct sw_flow_mask *mask)
1888 {
1889         memset(match, 0, sizeof(*match));
1890         match->key = key;
1891         match->mask = mask;
1892
1893         memset(key, 0, sizeof(*key));
1894
1895         if (mask) {
1896                 memset(&mask->key, 0, sizeof(mask->key));
1897                 mask->range.start = mask->range.end = 0;
1898         }
1899 }
1900
1901 static int validate_geneve_opts(struct sw_flow_key *key)
1902 {
1903         struct geneve_opt *option;
1904         int opts_len = key->tun_opts_len;
1905         bool crit_opt = false;
1906
1907         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1908         while (opts_len > 0) {
1909                 int len;
1910
1911                 if (opts_len < sizeof(*option))
1912                         return -EINVAL;
1913
1914                 len = sizeof(*option) + option->length * 4;
1915                 if (len > opts_len)
1916                         return -EINVAL;
1917
1918                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1919
1920                 option = (struct geneve_opt *)((u8 *)option + len);
1921                 opts_len -= len;
1922         };
1923
1924         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1925
1926         return 0;
1927 }
1928
1929 static int validate_and_copy_set_tun(const struct nlattr *attr,
1930                                      struct sw_flow_actions **sfa, bool log)
1931 {
1932         struct sw_flow_match match;
1933         struct sw_flow_key key;
1934         struct metadata_dst *tun_dst;
1935         struct ip_tunnel_info *tun_info;
1936         struct ovs_tunnel_info *ovs_tun;
1937         struct nlattr *a;
1938         int err = 0, start, opts_type;
1939
1940         ovs_match_init(&match, &key, NULL);
1941         opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
1942         if (opts_type < 0)
1943                 return opts_type;
1944
1945         if (key.tun_opts_len) {
1946                 switch (opts_type) {
1947                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1948                         err = validate_geneve_opts(&key);
1949                         if (err < 0)
1950                                 return err;
1951                         break;
1952                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1953                         break;
1954                 }
1955         };
1956
1957         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1958         if (start < 0)
1959                 return start;
1960
1961         tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1962         if (!tun_dst)
1963                 return -ENOMEM;
1964
1965         err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
1966         if (err) {
1967                 dst_release((struct dst_entry *)tun_dst);
1968                 return err;
1969         }
1970         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1971                          sizeof(*ovs_tun), log);
1972         if (IS_ERR(a)) {
1973                 ovs_dst_release((struct dst_entry *)tun_dst);
1974                 return PTR_ERR(a);
1975         }
1976
1977         ovs_tun = nla_data(a);
1978         ovs_tun->tun_dst = tun_dst;
1979
1980         tun_info = &tun_dst->u.tun_info;
1981         tun_info->mode = IP_TUNNEL_INFO_TX;
1982         if (key.tun_proto == AF_INET6)
1983                 tun_info->mode |= IP_TUNNEL_INFO_IPV6;
1984         tun_info->key = key.tun_key;
1985
1986         /* We need to store the options in the action itself since
1987          * everything else will go away after flow setup. We can append
1988          * it to tun_info and then point there.
1989          */
1990         ip_tunnel_info_opts_set(tun_info,
1991                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
1992                                 key.tun_opts_len);
1993         add_nested_action_end(*sfa, start);
1994
1995         return err;
1996 }
1997
1998 /* Return false if there are any non-masked bits set.
1999  * Mask follows data immediately, before any netlink padding.
2000  */
2001 static bool validate_masked(u8 *data, int len)
2002 {
2003         u8 *mask = data + len;
2004
2005         while (len--)
2006                 if (*data++ & ~*mask++)
2007                         return false;
2008
2009         return true;
2010 }
2011
2012 static int validate_set(const struct nlattr *a,
2013                         const struct sw_flow_key *flow_key,
2014                         struct sw_flow_actions **sfa,
2015                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
2016 {
2017         const struct nlattr *ovs_key = nla_data(a);
2018         int key_type = nla_type(ovs_key);
2019         size_t key_len;
2020
2021         /* There can be only one key in a action */
2022         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2023                 return -EINVAL;
2024
2025         key_len = nla_len(ovs_key);
2026         if (masked)
2027                 key_len /= 2;
2028
2029         if (key_type > OVS_KEY_ATTR_MAX ||
2030             !check_attr_len(key_len, ovs_key_lens[key_type].len))
2031                 return -EINVAL;
2032
2033         if (masked && !validate_masked(nla_data(ovs_key), key_len))
2034                 return -EINVAL;
2035
2036         switch (key_type) {
2037         const struct ovs_key_ipv4 *ipv4_key;
2038         const struct ovs_key_ipv6 *ipv6_key;
2039         int err;
2040
2041         case OVS_KEY_ATTR_PRIORITY:
2042         case OVS_KEY_ATTR_SKB_MARK:
2043         case OVS_KEY_ATTR_CT_MARK:
2044         case OVS_KEY_ATTR_CT_LABELS:
2045         case OVS_KEY_ATTR_ETHERNET:
2046                 break;
2047
2048         case OVS_KEY_ATTR_TUNNEL:
2049                 if (eth_p_mpls(eth_type))
2050                         return -EINVAL;
2051
2052                 if (masked)
2053                         return -EINVAL; /* Masked tunnel set not supported. */
2054
2055                 *skip_copy = true;
2056                 err = validate_and_copy_set_tun(a, sfa, log);
2057                 if (err)
2058                         return err;
2059                 break;
2060
2061         case OVS_KEY_ATTR_IPV4:
2062                 if (eth_type != htons(ETH_P_IP))
2063                         return -EINVAL;
2064
2065                 ipv4_key = nla_data(ovs_key);
2066
2067                 if (masked) {
2068                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2069
2070                         /* Non-writeable fields. */
2071                         if (mask->ipv4_proto || mask->ipv4_frag)
2072                                 return -EINVAL;
2073                 } else {
2074                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2075                                 return -EINVAL;
2076
2077                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2078                                 return -EINVAL;
2079                 }
2080                 break;
2081
2082         case OVS_KEY_ATTR_IPV6:
2083                 if (eth_type != htons(ETH_P_IPV6))
2084                         return -EINVAL;
2085
2086                 ipv6_key = nla_data(ovs_key);
2087
2088                 if (masked) {
2089                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2090
2091                         /* Non-writeable fields. */
2092                         if (mask->ipv6_proto || mask->ipv6_frag)
2093                                 return -EINVAL;
2094
2095                         /* Invalid bits in the flow label mask? */
2096                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2097                                 return -EINVAL;
2098                 } else {
2099                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2100                                 return -EINVAL;
2101
2102                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2103                                 return -EINVAL;
2104                 }
2105                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2106                         return -EINVAL;
2107
2108                 break;
2109
2110         case OVS_KEY_ATTR_TCP:
2111                 if ((eth_type != htons(ETH_P_IP) &&
2112                      eth_type != htons(ETH_P_IPV6)) ||
2113                     flow_key->ip.proto != IPPROTO_TCP)
2114                         return -EINVAL;
2115
2116                 break;
2117
2118         case OVS_KEY_ATTR_UDP:
2119                 if ((eth_type != htons(ETH_P_IP) &&
2120                      eth_type != htons(ETH_P_IPV6)) ||
2121                     flow_key->ip.proto != IPPROTO_UDP)
2122                         return -EINVAL;
2123
2124                 break;
2125
2126         case OVS_KEY_ATTR_MPLS:
2127                 if (!eth_p_mpls(eth_type))
2128                         return -EINVAL;
2129                 break;
2130
2131         case OVS_KEY_ATTR_SCTP:
2132                 if ((eth_type != htons(ETH_P_IP) &&
2133                      eth_type != htons(ETH_P_IPV6)) ||
2134                     flow_key->ip.proto != IPPROTO_SCTP)
2135                         return -EINVAL;
2136
2137                 break;
2138
2139         default:
2140                 return -EINVAL;
2141         }
2142
2143         /* Convert non-masked non-tunnel set actions to masked set actions. */
2144         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2145                 int start, len = key_len * 2;
2146                 struct nlattr *at;
2147
2148                 *skip_copy = true;
2149
2150                 start = add_nested_action_start(sfa,
2151                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2152                                                 log);
2153                 if (start < 0)
2154                         return start;
2155
2156                 at = __add_action(sfa, key_type, NULL, len, log);
2157                 if (IS_ERR(at))
2158                         return PTR_ERR(at);
2159
2160                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2161                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2162                 /* Clear non-writeable bits from otherwise writeable fields. */
2163                 if (key_type == OVS_KEY_ATTR_IPV6) {
2164                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2165
2166                         mask->ipv6_label &= htonl(0x000FFFFF);
2167                 }
2168                 add_nested_action_end(*sfa, start);
2169         }
2170
2171         return 0;
2172 }
2173
2174 static int validate_userspace(const struct nlattr *attr)
2175 {
2176         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2177                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2178                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2179                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2180         };
2181         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2182         int error;
2183
2184         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2185                                  attr, userspace_policy);
2186         if (error)
2187                 return error;
2188
2189         if (!a[OVS_USERSPACE_ATTR_PID] ||
2190             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2191                 return -EINVAL;
2192
2193         return 0;
2194 }
2195
2196 static int copy_action(const struct nlattr *from,
2197                        struct sw_flow_actions **sfa, bool log)
2198 {
2199         int totlen = NLA_ALIGN(from->nla_len);
2200         struct nlattr *to;
2201
2202         to = reserve_sfa_size(sfa, from->nla_len, log);
2203         if (IS_ERR(to))
2204                 return PTR_ERR(to);
2205
2206         memcpy(to, from, totlen);
2207         return 0;
2208 }
2209
2210 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2211                                   const struct sw_flow_key *key,
2212                                   int depth, struct sw_flow_actions **sfa,
2213                                   __be16 eth_type, __be16 vlan_tci, bool log)
2214 {
2215         const struct nlattr *a;
2216         int rem, err;
2217
2218         if (depth >= SAMPLE_ACTION_DEPTH)
2219                 return -EOVERFLOW;
2220
2221         nla_for_each_nested(a, attr, rem) {
2222                 /* Expected argument lengths, (u32)-1 for variable length. */
2223                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2224                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2225                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2226                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2227                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2228                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2229                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2230                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2231                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2232                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2233                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2234                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2235                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2236                         [OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
2237                 };
2238                 const struct ovs_action_push_vlan *vlan;
2239                 int type = nla_type(a);
2240                 bool skip_copy;
2241
2242                 if (type > OVS_ACTION_ATTR_MAX ||
2243                     (action_lens[type] != nla_len(a) &&
2244                      action_lens[type] != (u32)-1))
2245                         return -EINVAL;
2246
2247                 skip_copy = false;
2248                 switch (type) {
2249                 case OVS_ACTION_ATTR_UNSPEC:
2250                         return -EINVAL;
2251
2252                 case OVS_ACTION_ATTR_USERSPACE:
2253                         err = validate_userspace(a);
2254                         if (err)
2255                                 return err;
2256                         break;
2257
2258                 case OVS_ACTION_ATTR_OUTPUT:
2259                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2260                                 return -EINVAL;
2261                         break;
2262
2263                 case OVS_ACTION_ATTR_TRUNC: {
2264                         const struct ovs_action_trunc *trunc = nla_data(a);
2265
2266                         if (trunc->max_len < ETH_HLEN)
2267                                 return -EINVAL;
2268                         break;
2269                 }
2270
2271                 case OVS_ACTION_ATTR_HASH: {
2272                         const struct ovs_action_hash *act_hash = nla_data(a);
2273
2274                         switch (act_hash->hash_alg) {
2275                         case OVS_HASH_ALG_L4:
2276                                 break;
2277                         default:
2278                                 return  -EINVAL;
2279                         }
2280
2281                         break;
2282                 }
2283
2284                 case OVS_ACTION_ATTR_POP_VLAN:
2285                         vlan_tci = htons(0);
2286                         break;
2287
2288                 case OVS_ACTION_ATTR_PUSH_VLAN:
2289                         vlan = nla_data(a);
2290                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2291                                 return -EINVAL;
2292                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2293                                 return -EINVAL;
2294                         vlan_tci = vlan->vlan_tci;
2295                         break;
2296
2297                 case OVS_ACTION_ATTR_RECIRC:
2298                         break;
2299
2300                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2301                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2302
2303                         if (!eth_p_mpls(mpls->mpls_ethertype))
2304                                 return -EINVAL;
2305                         /* Prohibit push MPLS other than to a white list
2306                          * for packets that have a known tag order.
2307                          */
2308                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2309                             (eth_type != htons(ETH_P_IP) &&
2310                              eth_type != htons(ETH_P_IPV6) &&
2311                              eth_type != htons(ETH_P_ARP) &&
2312                              eth_type != htons(ETH_P_RARP) &&
2313                              !eth_p_mpls(eth_type)))
2314                                 return -EINVAL;
2315                         eth_type = mpls->mpls_ethertype;
2316                         break;
2317                 }
2318
2319                 case OVS_ACTION_ATTR_POP_MPLS:
2320                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2321                             !eth_p_mpls(eth_type))
2322                                 return -EINVAL;
2323
2324                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2325                          * as there is no check here to ensure that the new
2326                          * eth_type is valid and thus set actions could
2327                          * write off the end of the packet or otherwise
2328                          * corrupt it.
2329                          *
2330                          * Support for these actions is planned using packet
2331                          * recirculation.
2332                          */
2333                         eth_type = htons(0);
2334                         break;
2335
2336                 case OVS_ACTION_ATTR_SET:
2337                         err = validate_set(a, key, sfa,
2338                                            &skip_copy, eth_type, false, log);
2339                         if (err)
2340                                 return err;
2341                         break;
2342
2343                 case OVS_ACTION_ATTR_SET_MASKED:
2344                         err = validate_set(a, key, sfa,
2345                                            &skip_copy, eth_type, true, log);
2346                         if (err)
2347                                 return err;
2348                         break;
2349
2350                 case OVS_ACTION_ATTR_SAMPLE:
2351                         err = validate_and_copy_sample(net, a, key, depth, sfa,
2352                                                        eth_type, vlan_tci, log);
2353                         if (err)
2354                                 return err;
2355                         skip_copy = true;
2356                         break;
2357
2358                 case OVS_ACTION_ATTR_CT:
2359                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2360                         if (err)
2361                                 return err;
2362                         skip_copy = true;
2363                         break;
2364
2365                 default:
2366                         OVS_NLERR(log, "Unknown Action type %d", type);
2367                         return -EINVAL;
2368                 }
2369                 if (!skip_copy) {
2370                         err = copy_action(a, sfa, log);
2371                         if (err)
2372                                 return err;
2373                 }
2374         }
2375
2376         if (rem > 0)
2377                 return -EINVAL;
2378
2379         return 0;
2380 }
2381
2382 /* 'key' must be the masked key. */
2383 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2384                          const struct sw_flow_key *key,
2385                          struct sw_flow_actions **sfa, bool log)
2386 {
2387         int err;
2388
2389         *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2390         if (IS_ERR(*sfa))
2391                 return PTR_ERR(*sfa);
2392
2393         (*sfa)->orig_len = nla_len(attr);
2394         err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2395                                      key->eth.tci, log);
2396         if (err)
2397                 ovs_nla_free_flow_actions(*sfa);
2398
2399         return err;
2400 }
2401
2402 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2403 {
2404         const struct nlattr *a;
2405         struct nlattr *start;
2406         int err = 0, rem;
2407
2408         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2409         if (!start)
2410                 return -EMSGSIZE;
2411
2412         nla_for_each_nested(a, attr, rem) {
2413                 int type = nla_type(a);
2414                 struct nlattr *st_sample;
2415
2416                 switch (type) {
2417                 case OVS_SAMPLE_ATTR_PROBABILITY:
2418                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2419                                     sizeof(u32), nla_data(a)))
2420                                 return -EMSGSIZE;
2421                         break;
2422                 case OVS_SAMPLE_ATTR_ACTIONS:
2423                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2424                         if (!st_sample)
2425                                 return -EMSGSIZE;
2426                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2427                         if (err)
2428                                 return err;
2429                         nla_nest_end(skb, st_sample);
2430                         break;
2431                 }
2432         }
2433
2434         nla_nest_end(skb, start);
2435         return err;
2436 }
2437
2438 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2439 {
2440         const struct nlattr *ovs_key = nla_data(a);
2441         int key_type = nla_type(ovs_key);
2442         struct nlattr *start;
2443         int err;
2444
2445         switch (key_type) {
2446         case OVS_KEY_ATTR_TUNNEL_INFO: {
2447                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2448                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2449
2450                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2451                 if (!start)
2452                         return -EMSGSIZE;
2453
2454                 err =  ip_tun_to_nlattr(skb, &tun_info->key,
2455                                         ip_tunnel_info_opts(tun_info),
2456                                         tun_info->options_len,
2457                                         ip_tunnel_info_af(tun_info));
2458                 if (err)
2459                         return err;
2460                 nla_nest_end(skb, start);
2461                 break;
2462         }
2463         default:
2464                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2465                         return -EMSGSIZE;
2466                 break;
2467         }
2468
2469         return 0;
2470 }
2471
2472 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2473                                                 struct sk_buff *skb)
2474 {
2475         const struct nlattr *ovs_key = nla_data(a);
2476         struct nlattr *nla;
2477         size_t key_len = nla_len(ovs_key) / 2;
2478
2479         /* Revert the conversion we did from a non-masked set action to
2480          * masked set action.
2481          */
2482         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2483         if (!nla)
2484                 return -EMSGSIZE;
2485
2486         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2487                 return -EMSGSIZE;
2488
2489         nla_nest_end(skb, nla);
2490         return 0;
2491 }
2492
2493 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2494 {
2495         const struct nlattr *a;
2496         int rem, err;
2497
2498         nla_for_each_attr(a, attr, len, rem) {
2499                 int type = nla_type(a);
2500
2501                 switch (type) {
2502                 case OVS_ACTION_ATTR_SET:
2503                         err = set_action_to_attr(a, skb);
2504                         if (err)
2505                                 return err;
2506                         break;
2507
2508                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2509                         err = masked_set_action_to_set_action_attr(a, skb);
2510                         if (err)
2511                                 return err;
2512                         break;
2513
2514                 case OVS_ACTION_ATTR_SAMPLE:
2515                         err = sample_action_to_attr(a, skb);
2516                         if (err)
2517                                 return err;
2518                         break;
2519
2520                 case OVS_ACTION_ATTR_CT:
2521                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2522                         if (err)
2523                                 return err;
2524                         break;
2525
2526                 default:
2527                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2528                                 return -EMSGSIZE;
2529                         break;
2530                 }
2531         }
2532
2533         return 0;
2534 }