net/openvswitch/flow_netlink.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include "flow.h"
  22 #include "datapath.h"
  23 #include <linux/uaccess.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/etherdevice.h>
  26 #include <linux/if_ether.h>
  27 #include <linux/if_vlan.h>
  28 #include <net/llc_pdu.h>
  29 #include <linux/kernel.h>
  30 #include <linux/jhash.h>
  31 #include <linux/jiffies.h>
  32 #include <linux/llc.h>
  33 #include <linux/module.h>
  34 #include <linux/in.h>
  35 #include <linux/rcupdate.h>
  36 #include <linux/if_arp.h>
  37 #include <linux/ip.h>
  38 #include <linux/ipv6.h>
  39 #include <linux/sctp.h>
  40 #include <linux/tcp.h>
  41 #include <linux/udp.h>
  42 #include <linux/icmp.h>
  43 #include <linux/icmpv6.h>
  44 #include <linux/rculist.h>
  45 #include <net/geneve.h>
  46 #include <net/ip.h>
  47 #include <net/ipv6.h>
  48 #include <net/ndisc.h>
  49 #include <net/mpls.h>
  50 #include <net/vxlan.h>
  51
  52 #include "flow_netlink.h"
  53
  54 struct ovs_len_tbl {
  55         int len;
  56         const struct ovs_len_tbl *next;
  57 };
  58
  59 #define OVS_ATTR_NESTED -1
  60 #define OVS_ATTR_VARIABLE -2
  61
  62 static void update_range(struct sw_flow_match *match,
  63                          size_t offset, size_t size, bool is_mask)
  64 {
  65         struct sw_flow_key_range *range;
  66         size_t start = rounddown(offset, sizeof(long));
  67         size_t end = roundup(offset + size, sizeof(long));
  68
  69         if (!is_mask)
  70                 range = &match->range;
  71         else
  72                 range = &match->mask->range;
  73
  74         if (range->start == range->end) {
  75                 range->start = start;
  76                 range->end = end;
  77                 return;
  78         }
  79
  80         if (range->start > start)
  81                 range->start = start;
  82
  83         if (range->end < end)
  84                 range->end = end;
  85 }
  86
  87 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  88         do { \
  89                 update_range(match, offsetof(struct sw_flow_key, field),    \
  90                              sizeof((match)->key->field), is_mask);         \
  91                 if (is_mask)                                                \
  92                         (match)->mask->key.field = value;                   \
  93                 else                                                        \
  94                         (match)->key->field = value;                        \
  95         } while (0)
  96
  97 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)     \
  98         do {                                                                \
  99                 update_range(match, offset, len, is_mask);                  \
 100                 if (is_mask)                                                \
 101                         memcpy((u8 *)&(match)->mask->key + offset, value_p, \
 102                                len);                                       \
 103                 else                                                        \
 104                         memcpy((u8 *)(match)->key + offset, value_p, len);  \
 105         } while (0)
 106
 107 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)               \
 108         SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
 109                                   value_p, len, is_mask)
 110
 111 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)              \
 112         do {                                                                \
 113                 update_range(match, offsetof(struct sw_flow_key, field),    \
 114                              sizeof((match)->key->field), is_mask);         \
 115                 if (is_mask)                                                \
 116                         memset((u8 *)&(match)->mask->key.field, value,      \
 117                                sizeof((match)->mask->key.field));           \
 118                 else                                                        \
 119                         memset((u8 *)&(match)->key->field, value,           \
 120                                sizeof((match)->key->field));                \
 121         } while (0)
 122
 123 static bool match_validate(const struct sw_flow_match *match,
 124                            u64 key_attrs, u64 mask_attrs, bool log)
 125 {
 126         u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
 127         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 128
 129         /* The following mask attributes allowed only if they
 130          * pass the validation tests. */
 131         mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
 132                         | (1 << OVS_KEY_ATTR_IPV6)
 133                         | (1 << OVS_KEY_ATTR_TCP)
 134                         | (1 << OVS_KEY_ATTR_TCP_FLAGS)
 135                         | (1 << OVS_KEY_ATTR_UDP)
 136                         | (1 << OVS_KEY_ATTR_SCTP)
 137                         | (1 << OVS_KEY_ATTR_ICMP)
 138                         | (1 << OVS_KEY_ATTR_ICMPV6)
 139                         | (1 << OVS_KEY_ATTR_ARP)
 140                         | (1 << OVS_KEY_ATTR_ND)
 141                         | (1 << OVS_KEY_ATTR_MPLS));
 142
 143         /* Always allowed mask fields. */
 144         mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
 145                        | (1 << OVS_KEY_ATTR_IN_PORT)
 146                        | (1 << OVS_KEY_ATTR_ETHERTYPE));
 147
 148         /* Check key attributes. */
 149         if (match->key->eth.type == htons(ETH_P_ARP)
 150                         || match->key->eth.type == htons(ETH_P_RARP)) {
 151                 key_expected |= 1 << OVS_KEY_ATTR_ARP;
 152                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 153                         mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
 154         }
 155
 156         if (eth_p_mpls(match->key->eth.type)) {
 157                 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
 158                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 159                         mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
 160         }
 161
 162         if (match->key->eth.type == htons(ETH_P_IP)) {
 163                 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
 164                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 165                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
 166
 167                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 168                         if (match->key->ip.proto == IPPROTO_UDP) {
 169                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 170                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 171                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 172                         }
 173
 174                         if (match->key->ip.proto == IPPROTO_SCTP) {
 175                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 176                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 177                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 178                         }
 179
 180                         if (match->key->ip.proto == IPPROTO_TCP) {
 181                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 182                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 183                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 184                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 185                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 186                                 }
 187                         }
 188
 189                         if (match->key->ip.proto == IPPROTO_ICMP) {
 190                                 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
 191                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 192                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
 193                         }
 194                 }
 195         }
 196
 197         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 198                 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
 199                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 200                         mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
 201
 202                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 203                         if (match->key->ip.proto == IPPROTO_UDP) {
 204                                 key_expected |= 1 << OVS_KEY_ATTR_UDP;
 205                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 206                                         mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
 207                         }
 208
 209                         if (match->key->ip.proto == IPPROTO_SCTP) {
 210                                 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
 211                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 212                                         mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
 213                         }
 214
 215                         if (match->key->ip.proto == IPPROTO_TCP) {
 216                                 key_expected |= 1 << OVS_KEY_ATTR_TCP;
 217                                 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 218                                 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
 219                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
 220                                         mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
 221                                 }
 222                         }
 223
 224                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 225                                 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
 226                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 227                                         mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
 228
 229                                 if (match->key->tp.src ==
 230                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 231                                     match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 232                                         key_expected |= 1 << OVS_KEY_ATTR_ND;
 233                                         if (match->mask && (match->mask->key.tp.src == htons(0xff)))
 234                                                 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
 235                                 }
 236                         }
 237                 }
 238         }
 239
 240         if ((key_attrs & key_expected) != key_expected) {
 241                 /* Key attributes check failed. */
 242                 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
 243                           (unsigned long long)key_attrs,
 244                           (unsigned long long)key_expected);
 245                 return false;
 246         }
 247
 248         if ((mask_attrs & mask_allowed) != mask_attrs) {
 249                 /* Mask attributes check failed. */
 250                 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
 251                           (unsigned long long)mask_attrs,
 252                           (unsigned long long)mask_allowed);
 253                 return false;
 254         }
 255
 256         return true;
 257 }
 258
 259 size_t ovs_tun_key_attr_size(void)
 260 {
 261         /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
 262          * updating this function.
 263          */
 264         return    nla_total_size_64bit(8) /* OVS_TUNNEL_KEY_ATTR_ID */
 265                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
 266                 + nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
 267                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
 268                 + nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
 269                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 270                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
 271                 + nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
 272                 + nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 273                 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
 274                  * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
 275                  */
 276                 + nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
 277                 + nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
 278 }
 279
 280 size_t ovs_key_attr_size(void)
 281 {
 282         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 283          * updating this function.
 284          */
 285         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 26);
 286
 287         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 288                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 289                   + ovs_tun_key_attr_size()
 290                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 291                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 292                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 293                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 294                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
 295                 + nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
 296                 + nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
 297                 + nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
 298                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 299                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 300                 + nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
 301                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 302                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 303                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 304                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 305                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 306 }
 307
 308 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
 309         [OVS_VXLAN_EXT_GBP]         = { .len = sizeof(u32) },
 310 };
 311
 312 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 313         [OVS_TUNNEL_KEY_ATTR_ID]            = { .len = sizeof(u64) },
 314         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC]      = { .len = sizeof(u32) },
 315         [OVS_TUNNEL_KEY_ATTR_IPV4_DST]      = { .len = sizeof(u32) },
 316         [OVS_TUNNEL_KEY_ATTR_TOS]           = { .len = 1 },
 317         [OVS_TUNNEL_KEY_ATTR_TTL]           = { .len = 1 },
 318         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
 319         [OVS_TUNNEL_KEY_ATTR_CSUM]          = { .len = 0 },
 320         [OVS_TUNNEL_KEY_ATTR_TP_SRC]        = { .len = sizeof(u16) },
 321         [OVS_TUNNEL_KEY_ATTR_TP_DST]        = { .len = sizeof(u16) },
 322         [OVS_TUNNEL_KEY_ATTR_OAM]           = { .len = 0 },
 323         [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
 324         [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
 325                                                 .next = ovs_vxlan_ext_key_lens },
 326         [OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
 327         [OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
 328 };
 329
 330 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 331 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 332         [OVS_KEY_ATTR_ENCAP]     = { .len = OVS_ATTR_NESTED },
 333         [OVS_KEY_ATTR_PRIORITY]  = { .len = sizeof(u32) },
 334         [OVS_KEY_ATTR_IN_PORT]   = { .len = sizeof(u32) },
 335         [OVS_KEY_ATTR_SKB_MARK]  = { .len = sizeof(u32) },
 336         [OVS_KEY_ATTR_ETHERNET]  = { .len = sizeof(struct ovs_key_ethernet) },
 337         [OVS_KEY_ATTR_VLAN]      = { .len = sizeof(__be16) },
 338         [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
 339         [OVS_KEY_ATTR_IPV4]      = { .len = sizeof(struct ovs_key_ipv4) },
 340         [OVS_KEY_ATTR_IPV6]      = { .len = sizeof(struct ovs_key_ipv6) },
 341         [OVS_KEY_ATTR_TCP]       = { .len = sizeof(struct ovs_key_tcp) },
 342         [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
 343         [OVS_KEY_ATTR_UDP]       = { .len = sizeof(struct ovs_key_udp) },
 344         [OVS_KEY_ATTR_SCTP]      = { .len = sizeof(struct ovs_key_sctp) },
 345         [OVS_KEY_ATTR_ICMP]      = { .len = sizeof(struct ovs_key_icmp) },
 346         [OVS_KEY_ATTR_ICMPV6]    = { .len = sizeof(struct ovs_key_icmpv6) },
 347         [OVS_KEY_ATTR_ARP]       = { .len = sizeof(struct ovs_key_arp) },
 348         [OVS_KEY_ATTR_ND]        = { .len = sizeof(struct ovs_key_nd) },
 349         [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
 350         [OVS_KEY_ATTR_DP_HASH]   = { .len = sizeof(u32) },
 351         [OVS_KEY_ATTR_TUNNEL]    = { .len = OVS_ATTR_NESTED,
 352                                      .next = ovs_tunnel_key_lens, },
 353         [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 354         [OVS_KEY_ATTR_CT_STATE]  = { .len = sizeof(u32) },
 355         [OVS_KEY_ATTR_CT_ZONE]   = { .len = sizeof(u16) },
 356         [OVS_KEY_ATTR_CT_MARK]   = { .len = sizeof(u32) },
 357         [OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
 358 };
 359
 360 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
 361 {
 362         return expected_len == attr_len ||
 363                expected_len == OVS_ATTR_NESTED ||
 364                expected_len == OVS_ATTR_VARIABLE;
 365 }
 366
 367 static bool is_all_zero(const u8 *fp, size_t size)
 368 {
 369         int i;
 370
 371         if (!fp)
 372                 return false;
 373
 374         for (i = 0; i < size; i++)
 375                 if (fp[i])
 376                         return false;
 377
 378         return true;
 379 }
 380
 381 static int __parse_flow_nlattrs(const struct nlattr *attr,
 382                                 const struct nlattr *a[],
 383                                 u64 *attrsp, bool log, bool nz)
 384 {
 385         const struct nlattr *nla;
 386         u64 attrs;
 387         int rem;
 388
 389         attrs = *attrsp;
 390         nla_for_each_nested(nla, attr, rem) {
 391                 u16 type = nla_type(nla);
 392                 int expected_len;
 393
 394                 if (type > OVS_KEY_ATTR_MAX) {
 395                         OVS_NLERR(log, "Key type %d is out of range max %d",
 396                                   type, OVS_KEY_ATTR_MAX);
 397                         return -EINVAL;
 398                 }
 399
 400                 if (attrs & (1 << type)) {
 401                         OVS_NLERR(log, "Duplicate key (type %d).", type);
 402                         return -EINVAL;
 403                 }
 404
 405                 expected_len = ovs_key_lens[type].len;
 406                 if (!check_attr_len(nla_len(nla), expected_len)) {
 407                         OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
 408                                   type, nla_len(nla), expected_len);
 409                         return -EINVAL;
 410                 }
 411
 412                 if (!nz || !is_all_zero(nla_data(nla), nla_len(nla))) {
 413                         attrs |= 1 << type;
 414                         a[type] = nla;
 415                 }
 416         }
 417         if (rem) {
 418                 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
 419                 return -EINVAL;
 420         }
 421
 422         *attrsp = attrs;
 423         return 0;
 424 }
 425
 426 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
 427                                    const struct nlattr *a[], u64 *attrsp,
 428                                    bool log)
 429 {
 430         return __parse_flow_nlattrs(attr, a, attrsp, log, true);
 431 }
 432
 433 static int parse_flow_nlattrs(const struct nlattr *attr,
 434                               const struct nlattr *a[], u64 *attrsp,
 435                               bool log)
 436 {
 437         return __parse_flow_nlattrs(attr, a, attrsp, log, false);
 438 }
 439
 440 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
 441                                      struct sw_flow_match *match, bool is_mask,
 442                                      bool log)
 443 {
 444         unsigned long opt_key_offset;
 445
 446         if (nla_len(a) > sizeof(match->key->tun_opts)) {
 447                 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
 448                           nla_len(a), sizeof(match->key->tun_opts));
 449                 return -EINVAL;
 450         }
 451
 452         if (nla_len(a) % 4 != 0) {
 453                 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
 454                           nla_len(a));
 455                 return -EINVAL;
 456         }
 457
 458         /* We need to record the length of the options passed
 459          * down, otherwise packets with the same format but
 460          * additional options will be silently matched.
 461          */
 462         if (!is_mask) {
 463                 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
 464                                 false);
 465         } else {
 466                 /* This is somewhat unusual because it looks at
 467                  * both the key and mask while parsing the
 468                  * attributes (and by extension assumes the key
 469                  * is parsed first). Normally, we would verify
 470                  * that each is the correct length and that the
 471                  * attributes line up in the validate function.
 472                  * However, that is difficult because this is
 473                  * variable length and we won't have the
 474                  * information later.
 475                  */
 476                 if (match->key->tun_opts_len != nla_len(a)) {
 477                         OVS_NLERR(log, "Geneve option len %d != mask len %d",
 478                                   match->key->tun_opts_len, nla_len(a));
 479                         return -EINVAL;
 480                 }
 481
 482                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 483         }
 484
 485         opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
 486         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
 487                                   nla_len(a), is_mask);
 488         return 0;
 489 }
 490
 491 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
 492                                      struct sw_flow_match *match, bool is_mask,
 493                                      bool log)
 494 {
 495         struct nlattr *a;
 496         int rem;
 497         unsigned long opt_key_offset;
 498         struct vxlan_metadata opts;
 499
 500         BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 501
 502         memset(&opts, 0, sizeof(opts));
 503         nla_for_each_nested(a, attr, rem) {
 504                 int type = nla_type(a);
 505
 506                 if (type > OVS_VXLAN_EXT_MAX) {
 507                         OVS_NLERR(log, "VXLAN extension %d out of range max %d",
 508                                   type, OVS_VXLAN_EXT_MAX);
 509                         return -EINVAL;
 510                 }
 511
 512                 if (!check_attr_len(nla_len(a),
 513                                     ovs_vxlan_ext_key_lens[type].len)) {
 514                         OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
 515                                   type, nla_len(a),
 516                                   ovs_vxlan_ext_key_lens[type].len);
 517                         return -EINVAL;
 518                 }
 519
 520                 switch (type) {
 521                 case OVS_VXLAN_EXT_GBP:
 522                         opts.gbp = nla_get_u32(a);
 523                         break;
 524                 default:
 525                         OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
 526                                   type);
 527                         return -EINVAL;
 528                 }
 529         }
 530         if (rem) {
 531                 OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
 532                           rem);
 533                 return -EINVAL;
 534         }
 535
 536         if (!is_mask)
 537                 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
 538         else
 539                 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
 540
 541         opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
 542         SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
 543                                   is_mask);
 544         return 0;
 545 }
 546
 547 static int ip_tun_from_nlattr(const struct nlattr *attr,
 548                               struct sw_flow_match *match, bool is_mask,
 549                               bool log)
 550 {
 551         bool ttl = false, ipv4 = false, ipv6 = false;
 552         __be16 tun_flags = 0;
 553         int opts_type = 0;
 554         struct nlattr *a;
 555         int rem;
 556
 557         nla_for_each_nested(a, attr, rem) {
 558                 int type = nla_type(a);
 559                 int err;
 560
 561                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
 562                         OVS_NLERR(log, "Tunnel attr %d out of range max %d",
 563                                   type, OVS_TUNNEL_KEY_ATTR_MAX);
 564                         return -EINVAL;
 565                 }
 566
 567                 if (!check_attr_len(nla_len(a),
 568                                     ovs_tunnel_key_lens[type].len)) {
 569                         OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
 570                                   type, nla_len(a), ovs_tunnel_key_lens[type].len);
 571                         return -EINVAL;
 572                 }
 573
 574                 switch (type) {
 575                 case OVS_TUNNEL_KEY_ATTR_ID:
 576                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
 577                                         nla_get_be64(a), is_mask);
 578                         tun_flags |= TUNNEL_KEY;
 579                         break;
 580                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 581                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
 582                                         nla_get_in_addr(a), is_mask);
 583                         ipv4 = true;
 584                         break;
 585                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 586                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
 587                                         nla_get_in_addr(a), is_mask);
 588                         ipv4 = true;
 589                         break;
 590                 case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
 591                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.src,
 592                                         nla_get_in6_addr(a), is_mask);
 593                         ipv6 = true;
 594                         break;
 595                 case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
 596                         SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
 597                                         nla_get_in6_addr(a), is_mask);
 598                         ipv6 = true;
 599                         break;
 600                 case OVS_TUNNEL_KEY_ATTR_TOS:
 601                         SW_FLOW_KEY_PUT(match, tun_key.tos,
 602                                         nla_get_u8(a), is_mask);
 603                         break;
 604                 case OVS_TUNNEL_KEY_ATTR_TTL:
 605                         SW_FLOW_KEY_PUT(match, tun_key.ttl,
 606                                         nla_get_u8(a), is_mask);
 607                         ttl = true;
 608                         break;
 609                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
 610                         tun_flags |= TUNNEL_DONT_FRAGMENT;
 611                         break;
 612                 case OVS_TUNNEL_KEY_ATTR_CSUM:
 613                         tun_flags |= TUNNEL_CSUM;
 614                         break;
 615                 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
 616                         SW_FLOW_KEY_PUT(match, tun_key.tp_src,
 617                                         nla_get_be16(a), is_mask);
 618                         break;
 619                 case OVS_TUNNEL_KEY_ATTR_TP_DST:
 620                         SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
 621                                         nla_get_be16(a), is_mask);
 622                         break;
 623                 case OVS_TUNNEL_KEY_ATTR_OAM:
 624                         tun_flags |= TUNNEL_OAM;
 625                         break;
 626                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
 627                         if (opts_type) {
 628                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 629                                 return -EINVAL;
 630                         }
 631
 632                         err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
 633                         if (err)
 634                                 return err;
 635
 636                         tun_flags |= TUNNEL_GENEVE_OPT;
 637                         opts_type = type;
 638                         break;
 639                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
 640                         if (opts_type) {
 641                                 OVS_NLERR(log, "Multiple metadata blocks provided");
 642                                 return -EINVAL;
 643                         }
 644
 645                         err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
 646                         if (err)
 647                                 return err;
 648
 649                         tun_flags |= TUNNEL_VXLAN_OPT;
 650                         opts_type = type;
 651                         break;
 652                 case OVS_TUNNEL_KEY_ATTR_PAD:
 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 encode_vlan_from_nlattrs(struct sw_flow_match *match,
 814                                     const struct nlattr *a[],
 815                                     bool is_mask, bool inner)
 816 {
 817         __be16 tci = 0;
 818         __be16 tpid = 0;
 819
 820         if (a[OVS_KEY_ATTR_VLAN])
 821                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 822
 823         if (a[OVS_KEY_ATTR_ETHERTYPE])
 824                 tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 825
 826         if (likely(!inner)) {
 827                 SW_FLOW_KEY_PUT(match, eth.vlan.tpid, tpid, is_mask);
 828                 SW_FLOW_KEY_PUT(match, eth.vlan.tci, tci, is_mask);
 829         } else {
 830                 SW_FLOW_KEY_PUT(match, eth.cvlan.tpid, tpid, is_mask);
 831                 SW_FLOW_KEY_PUT(match, eth.cvlan.tci, tci, is_mask);
 832         }
 833         return 0;
 834 }
 835
 836 static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
 837                                       u64 key_attrs, bool inner,
 838                                       const struct nlattr **a, bool log)
 839 {
 840         __be16 tci = 0;
 841
 842         if (!((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
 843               (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
 844                eth_type_vlan(nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE])))) {
 845                 /* Not a VLAN. */
 846                 return 0;
 847         }
 848
 849         if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
 850               (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
 851                 OVS_NLERR(log, "Invalid %s frame", (inner) ? "C-VLAN" : "VLAN");
 852                 return -EINVAL;
 853         }
 854
 855         if (a[OVS_KEY_ATTR_VLAN])
 856                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 857
 858         if (!(tci & htons(VLAN_TAG_PRESENT))) {
 859                 if (tci) {
 860                         OVS_NLERR(log, "%s TCI does not have VLAN_TAG_PRESENT bit set.",
 861                                   (inner) ? "C-VLAN" : "VLAN");
 862                         return -EINVAL;
 863                 } else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
 864                         /* Corner case for truncated VLAN header. */
 865                         OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
 866                                   (inner) ? "C-VLAN" : "VLAN");
 867                         return -EINVAL;
 868                 }
 869         }
 870
 871         return 1;
 872 }
 873
 874 static int validate_vlan_mask_from_nlattrs(const struct sw_flow_match *match,
 875                                            u64 key_attrs, bool inner,
 876                                            const struct nlattr **a, bool log)
 877 {
 878         __be16 tci = 0;
 879         __be16 tpid = 0;
 880         bool encap_valid = !!(match->key->eth.vlan.tci &
 881                               htons(VLAN_TAG_PRESENT));
 882         bool i_encap_valid = !!(match->key->eth.cvlan.tci &
 883                                 htons(VLAN_TAG_PRESENT));
 884
 885         if (!(key_attrs & (1 << OVS_KEY_ATTR_ENCAP))) {
 886                 /* Not a VLAN. */
 887                 return 0;
 888         }
 889
 890         if ((!inner && !encap_valid) || (inner && !i_encap_valid)) {
 891                 OVS_NLERR(log, "Encap mask attribute is set for non-%s frame.",
 892                           (inner) ? "C-VLAN" : "VLAN");
 893                 return -EINVAL;
 894         }
 895
 896         if (a[OVS_KEY_ATTR_VLAN])
 897                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
 898
 899         if (a[OVS_KEY_ATTR_ETHERTYPE])
 900                 tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
 901
 902         if (tpid != htons(0xffff)) {
 903                 OVS_NLERR(log, "Must have an exact match on %s TPID (mask=%x).",
 904                           (inner) ? "C-VLAN" : "VLAN", ntohs(tpid));
 905                 return -EINVAL;
 906         }
 907         if (!(tci & htons(VLAN_TAG_PRESENT))) {
 908                 OVS_NLERR(log, "%s TCI mask does not have exact match for VLAN_TAG_PRESENT bit.",
 909                           (inner) ? "C-VLAN" : "VLAN");
 910                 return -EINVAL;
 911         }
 912
 913         return 1;
 914 }
 915
 916 static int __parse_vlan_from_nlattrs(struct sw_flow_match *match,
 917                                      u64 *key_attrs, bool inner,
 918                                      const struct nlattr **a, bool is_mask,
 919                                      bool log)
 920 {
 921         int err;
 922         const struct nlattr *encap;
 923
 924         if (!is_mask)
 925                 err = validate_vlan_from_nlattrs(match, *key_attrs, inner,
 926                                                  a, log);
 927         else
 928                 err = validate_vlan_mask_from_nlattrs(match, *key_attrs, inner,
 929                                                       a, log);
 930         if (err <= 0)
 931                 return err;
 932
 933         err = encode_vlan_from_nlattrs(match, a, is_mask, inner);
 934         if (err)
 935                 return err;
 936
 937         *key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
 938         *key_attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
 939         *key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
 940
 941         encap = a[OVS_KEY_ATTR_ENCAP];
 942
 943         if (!is_mask)
 944                 err = parse_flow_nlattrs(encap, a, key_attrs, log);
 945         else
 946                 err = parse_flow_mask_nlattrs(encap, a, key_attrs, log);
 947
 948         return err;
 949 }
 950
 951 static int parse_vlan_from_nlattrs(struct sw_flow_match *match,
 952                                    u64 *key_attrs, const struct nlattr **a,
 953                                    bool is_mask, bool log)
 954 {
 955         int err;
 956         bool encap_valid = false;
 957
 958         err = __parse_vlan_from_nlattrs(match, key_attrs, false, a,
 959                                         is_mask, log);
 960         if (err)
 961                 return err;
 962
 963         encap_valid = !!(match->key->eth.vlan.tci & htons(VLAN_TAG_PRESENT));
 964         if (encap_valid) {
 965                 err = __parse_vlan_from_nlattrs(match, key_attrs, true, a,
 966                                                 is_mask, log);
 967                 if (err)
 968                         return err;
 969         }
 970
 971         return 0;
 972 }
 973
 974 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
 975                                  u64 *attrs, const struct nlattr **a,
 976                                  bool is_mask, bool log)
 977 {
 978         if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
 979                 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
 980
 981                 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
 982                 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
 983         }
 984
 985         if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
 986                 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
 987
 988                 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
 989                 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
 990         }
 991
 992         if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 993                 SW_FLOW_KEY_PUT(match, phy.priority,
 994                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
 995                 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 996         }
 997
 998         if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 999                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1000
1001                 if (is_mask) {
1002                         in_port = 0xffffffff; /* Always exact match in_port. */
1003                 } else if (in_port >= DP_MAX_PORTS) {
1004                         OVS_NLERR(log, "Port %d exceeds max allowable %d",
1005                                   in_port, DP_MAX_PORTS);
1006                         return -EINVAL;
1007                 }
1008
1009                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1010                 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1011         } else if (!is_mask) {
1012                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1013         }
1014
1015         if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1016                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1017
1018                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1019                 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1020         }
1021         if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1022                 if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1023                                        is_mask, log) < 0)
1024                         return -EINVAL;
1025                 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1026         }
1027
1028         if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
1029             ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
1030                 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
1031
1032                 if (ct_state & ~CT_SUPPORTED_MASK) {
1033                         OVS_NLERR(log, "ct_state flags %08x unsupported",
1034                                   ct_state);
1035                         return -EINVAL;
1036                 }
1037
1038                 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
1039                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
1040         }
1041         if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
1042             ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
1043                 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
1044
1045                 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
1046                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
1047         }
1048         if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
1049             ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
1050                 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
1051
1052                 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
1053                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
1054         }
1055         if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
1056             ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
1057                 const struct ovs_key_ct_labels *cl;
1058
1059                 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
1060                 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
1061                                    sizeof(*cl), is_mask);
1062                 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
1063         }
1064         return 0;
1065 }
1066
1067 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
1068                                 u64 attrs, const struct nlattr **a,
1069                                 bool is_mask, bool log)
1070 {
1071         int err;
1072
1073         err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
1074         if (err)
1075                 return err;
1076
1077         if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1078                 const struct ovs_key_ethernet *eth_key;
1079
1080                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1081                 SW_FLOW_KEY_MEMCPY(match, eth.src,
1082                                 eth_key->eth_src, ETH_ALEN, is_mask);
1083                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1084                                 eth_key->eth_dst, ETH_ALEN, is_mask);
1085                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1086         }
1087
1088         if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1089                 /* VLAN attribute is always parsed before getting here since it
1090                  * may occur multiple times.
1091                  */
1092                 OVS_NLERR(log, "VLAN attribute unexpected.");
1093                 return -EINVAL;
1094         }
1095
1096         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1097                 __be16 eth_type;
1098
1099                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1100                 if (is_mask) {
1101                         /* Always exact match EtherType. */
1102                         eth_type = htons(0xffff);
1103                 } else if (!eth_proto_is_802_3(eth_type)) {
1104                         OVS_NLERR(log, "EtherType %x is less than min %x",
1105                                   ntohs(eth_type), ETH_P_802_3_MIN);
1106                         return -EINVAL;
1107                 }
1108
1109                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1110                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1111         } else if (!is_mask) {
1112                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1113         }
1114
1115         if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1116                 const struct ovs_key_ipv4 *ipv4_key;
1117
1118                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1119                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1120                         OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
1121                                   ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1122                         return -EINVAL;
1123                 }
1124                 SW_FLOW_KEY_PUT(match, ip.proto,
1125                                 ipv4_key->ipv4_proto, is_mask);
1126                 SW_FLOW_KEY_PUT(match, ip.tos,
1127                                 ipv4_key->ipv4_tos, is_mask);
1128                 SW_FLOW_KEY_PUT(match, ip.ttl,
1129                                 ipv4_key->ipv4_ttl, is_mask);
1130                 SW_FLOW_KEY_PUT(match, ip.frag,
1131                                 ipv4_key->ipv4_frag, is_mask);
1132                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1133                                 ipv4_key->ipv4_src, is_mask);
1134                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1135                                 ipv4_key->ipv4_dst, is_mask);
1136                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1137         }
1138
1139         if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1140                 const struct ovs_key_ipv6 *ipv6_key;
1141
1142                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1143                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1144                         OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
1145                                   ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1146                         return -EINVAL;
1147                 }
1148
1149                 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
1150                         OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
1151                                   ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1152                         return -EINVAL;
1153                 }
1154
1155                 SW_FLOW_KEY_PUT(match, ipv6.label,
1156                                 ipv6_key->ipv6_label, is_mask);
1157                 SW_FLOW_KEY_PUT(match, ip.proto,
1158                                 ipv6_key->ipv6_proto, is_mask);
1159                 SW_FLOW_KEY_PUT(match, ip.tos,
1160                                 ipv6_key->ipv6_tclass, is_mask);
1161                 SW_FLOW_KEY_PUT(match, ip.ttl,
1162                                 ipv6_key->ipv6_hlimit, is_mask);
1163                 SW_FLOW_KEY_PUT(match, ip.frag,
1164                                 ipv6_key->ipv6_frag, is_mask);
1165                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1166                                 ipv6_key->ipv6_src,
1167                                 sizeof(match->key->ipv6.addr.src),
1168                                 is_mask);
1169                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1170                                 ipv6_key->ipv6_dst,
1171                                 sizeof(match->key->ipv6.addr.dst),
1172                                 is_mask);
1173
1174                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1175         }
1176
1177         if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1178                 const struct ovs_key_arp *arp_key;
1179
1180                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1181                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1182                         OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1183                                   arp_key->arp_op);
1184                         return -EINVAL;
1185                 }
1186
1187                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1188                                 arp_key->arp_sip, is_mask);
1189                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1190                         arp_key->arp_tip, is_mask);
1191                 SW_FLOW_KEY_PUT(match, ip.proto,
1192                                 ntohs(arp_key->arp_op), is_mask);
1193                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1194                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1195                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1196                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1197
1198                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1199         }
1200
1201         if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1202                 const struct ovs_key_mpls *mpls_key;
1203
1204                 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1205                 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1206                                 mpls_key->mpls_lse, is_mask);
1207
1208                 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1209          }
1210
1211         if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1212                 const struct ovs_key_tcp *tcp_key;
1213
1214                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1215                 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1216                 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1217                 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1218         }
1219
1220         if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1221                 SW_FLOW_KEY_PUT(match, tp.flags,
1222                                 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1223                                 is_mask);
1224                 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1225         }
1226
1227         if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1228                 const struct ovs_key_udp *udp_key;
1229
1230                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1231                 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1232                 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1233                 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1234         }
1235
1236         if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1237                 const struct ovs_key_sctp *sctp_key;
1238
1239                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1240                 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1241                 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1242                 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1243         }
1244
1245         if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1246                 const struct ovs_key_icmp *icmp_key;
1247
1248                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1249                 SW_FLOW_KEY_PUT(match, tp.src,
1250                                 htons(icmp_key->icmp_type), is_mask);
1251                 SW_FLOW_KEY_PUT(match, tp.dst,
1252                                 htons(icmp_key->icmp_code), is_mask);
1253                 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1254         }
1255
1256         if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1257                 const struct ovs_key_icmpv6 *icmpv6_key;
1258
1259                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1260                 SW_FLOW_KEY_PUT(match, tp.src,
1261                                 htons(icmpv6_key->icmpv6_type), is_mask);
1262                 SW_FLOW_KEY_PUT(match, tp.dst,
1263                                 htons(icmpv6_key->icmpv6_code), is_mask);
1264                 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1265         }
1266
1267         if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1268                 const struct ovs_key_nd *nd_key;
1269
1270                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1271                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1272                         nd_key->nd_target,
1273                         sizeof(match->key->ipv6.nd.target),
1274                         is_mask);
1275                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1276                         nd_key->nd_sll, ETH_ALEN, is_mask);
1277                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1278                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1279                 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1280         }
1281
1282         if (attrs != 0) {
1283                 OVS_NLERR(log, "Unknown key attributes %llx",
1284                           (unsigned long long)attrs);
1285                 return -EINVAL;
1286         }
1287
1288         return 0;
1289 }
1290
1291 static void nlattr_set(struct nlattr *attr, u8 val,
1292                        const struct ovs_len_tbl *tbl)
1293 {
1294         struct nlattr *nla;
1295         int rem;
1296
1297         /* The nlattr stream should already have been validated */
1298         nla_for_each_nested(nla, attr, rem) {
1299                 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1300                         nlattr_set(nla, val, tbl[nla_type(nla)].next ? : tbl);
1301                 else
1302                         memset(nla_data(nla), val, nla_len(nla));
1303
1304                 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1305                         *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1306         }
1307 }
1308
1309 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1310 {
1311         nlattr_set(attr, val, ovs_key_lens);
1312 }
1313
1314 /**
1315  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1316  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1317  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1318  * does not include any don't care bit.
1319  * @net: Used to determine per-namespace field support.
1320  * @match: receives the extracted flow match information.
1321  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1322  * sequence. The fields should of the packet that triggered the creation
1323  * of this flow.
1324  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1325  * attribute specifies the mask field of the wildcarded flow.
1326  * @log: Boolean to allow kernel error logging.  Normally true, but when
1327  * probing for feature compatibility this should be passed in as false to
1328  * suppress unnecessary error logging.
1329  */
1330 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1331                       const struct nlattr *nla_key,
1332                       const struct nlattr *nla_mask,
1333                       bool log)
1334 {
1335         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1336         struct nlattr *newmask = NULL;
1337         u64 key_attrs = 0;
1338         u64 mask_attrs = 0;
1339         int err;
1340
1341         err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1342         if (err)
1343                 return err;
1344
1345         err = parse_vlan_from_nlattrs(match, &key_attrs, a, false, log);
1346         if (err)
1347                 return err;
1348
1349         err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1350         if (err)
1351                 return err;
1352
1353         if (match->mask) {
1354                 if (!nla_mask) {
1355                         /* Create an exact match mask. We need to set to 0xff
1356                          * all the 'match->mask' fields that have been touched
1357                          * in 'match->key'. We cannot simply memset
1358                          * 'match->mask', because padding bytes and fields not
1359                          * specified in 'match->key' should be left to 0.
1360                          * Instead, we use a stream of netlink attributes,
1361                          * copied from 'key' and set to 0xff.
1362                          * ovs_key_from_nlattrs() will take care of filling
1363                          * 'match->mask' appropriately.
1364                          */
1365                         newmask = kmemdup(nla_key,
1366                                           nla_total_size(nla_len(nla_key)),
1367                                           GFP_KERNEL);
1368                         if (!newmask)
1369                                 return -ENOMEM;
1370
1371                         mask_set_nlattr(newmask, 0xff);
1372
1373                         /* The userspace does not send tunnel attributes that
1374                          * are 0, but we should not wildcard them nonetheless.
1375                          */
1376                         if (match->key->tun_proto)
1377                                 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1378                                                          0xff, true);
1379
1380                         nla_mask = newmask;
1381                 }
1382
1383                 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1384                 if (err)
1385                         goto free_newmask;
1386
1387                 /* Always match on tci. */
1388                 SW_FLOW_KEY_PUT(match, eth.vlan.tci, htons(0xffff), true);
1389                 SW_FLOW_KEY_PUT(match, eth.cvlan.tci, htons(0xffff), true);
1390
1391                 err = parse_vlan_from_nlattrs(match, &mask_attrs, a, true, log);
1392                 if (err)
1393                         goto free_newmask;
1394
1395                 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1396                                            log);
1397                 if (err)
1398                         goto free_newmask;
1399         }
1400
1401         if (!match_validate(match, key_attrs, mask_attrs, log))
1402                 err = -EINVAL;
1403
1404 free_newmask:
1405         kfree(newmask);
1406         return err;
1407 }
1408
1409 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1410 {
1411         size_t len;
1412
1413         if (!attr)
1414                 return 0;
1415
1416         len = nla_len(attr);
1417         if (len < 1 || len > MAX_UFID_LENGTH) {
1418                 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1419                           nla_len(attr), MAX_UFID_LENGTH);
1420                 return 0;
1421         }
1422
1423         return len;
1424 }
1425
1426 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1427  * or false otherwise.
1428  */
1429 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1430                       bool log)
1431 {
1432         sfid->ufid_len = get_ufid_len(attr, log);
1433         if (sfid->ufid_len)
1434                 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1435
1436         return sfid->ufid_len;
1437 }
1438
1439 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1440                            const struct sw_flow_key *key, bool log)
1441 {
1442         struct sw_flow_key *new_key;
1443
1444         if (ovs_nla_get_ufid(sfid, ufid, log))
1445                 return 0;
1446
1447         /* If UFID was not provided, use unmasked key. */
1448         new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1449         if (!new_key)
1450                 return -ENOMEM;
1451         memcpy(new_key, key, sizeof(*key));
1452         sfid->unmasked_key = new_key;
1453
1454         return 0;
1455 }
1456
1457 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1458 {
1459         return attr ? nla_get_u32(attr) : 0;
1460 }
1461
1462 /**
1463  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1464  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1465  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1466  * sequence.
1467  * @log: Boolean to allow kernel error logging.  Normally true, but when
1468  * probing for feature compatibility this should be passed in as false to
1469  * suppress unnecessary error logging.
1470  *
1471  * This parses a series of Netlink attributes that form a flow key, which must
1472  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1473  * get the metadata, that is, the parts of the flow key that cannot be
1474  * extracted from the packet itself.
1475  */
1476
1477 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1478                               struct sw_flow_key *key,
1479                               bool log)
1480 {
1481         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1482         struct sw_flow_match match;
1483         u64 attrs = 0;
1484         int err;
1485
1486         err = parse_flow_nlattrs(attr, a, &attrs, log);
1487         if (err)
1488                 return -EINVAL;
1489
1490         memset(&match, 0, sizeof(match));
1491         match.key = key;
1492
1493         memset(&key->ct, 0, sizeof(key->ct));
1494         key->phy.in_port = DP_MAX_PORTS;
1495
1496         return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1497 }
1498
1499 static int ovs_nla_put_vlan(struct sk_buff *skb, const struct vlan_head *vh,
1500                             bool is_mask)
1501 {
1502         __be16 eth_type = !is_mask ? vh->tpid : htons(0xffff);
1503
1504         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1505             nla_put_be16(skb, OVS_KEY_ATTR_VLAN, vh->tci))
1506                 return -EMSGSIZE;
1507         return 0;
1508 }
1509
1510 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1511                              const struct sw_flow_key *output, bool is_mask,
1512                              struct sk_buff *skb)
1513 {
1514         struct ovs_key_ethernet *eth_key;
1515         struct nlattr *nla;
1516         struct nlattr *encap = NULL;
1517         struct nlattr *in_encap = NULL;
1518
1519         if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1520                 goto nla_put_failure;
1521
1522         if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1523                 goto nla_put_failure;
1524
1525         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1526                 goto nla_put_failure;
1527
1528         if ((swkey->tun_proto || is_mask)) {
1529                 const void *opts = NULL;
1530
1531                 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1532                         opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1533
1534                 if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
1535                                      swkey->tun_opts_len, swkey->tun_proto))
1536                         goto nla_put_failure;
1537         }
1538
1539         if (swkey->phy.in_port == DP_MAX_PORTS) {
1540                 if (is_mask && (output->phy.in_port == 0xffff))
1541                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1542                                 goto nla_put_failure;
1543         } else {
1544                 u16 upper_u16;
1545                 upper_u16 = !is_mask ? 0 : 0xffff;
1546
1547                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1548                                 (upper_u16 << 16) | output->phy.in_port))
1549                         goto nla_put_failure;
1550         }
1551
1552         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1553                 goto nla_put_failure;
1554
1555         if (ovs_ct_put_key(output, skb))
1556                 goto nla_put_failure;
1557
1558         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1559         if (!nla)
1560                 goto nla_put_failure;
1561
1562         eth_key = nla_data(nla);
1563         ether_addr_copy(eth_key->eth_src, output->eth.src);
1564         ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1565
1566         if (swkey->eth.vlan.tci || eth_type_vlan(swkey->eth.type)) {
1567                 if (ovs_nla_put_vlan(skb, &output->eth.vlan, is_mask))
1568                         goto nla_put_failure;
1569                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1570                 if (!swkey->eth.vlan.tci)
1571                         goto unencap;
1572
1573                 if (swkey->eth.cvlan.tci || eth_type_vlan(swkey->eth.type)) {
1574                         if (ovs_nla_put_vlan(skb, &output->eth.cvlan, is_mask))
1575                                 goto nla_put_failure;
1576                         in_encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1577                         if (!swkey->eth.cvlan.tci)
1578                                 goto unencap;
1579                 }
1580         }
1581
1582         if (swkey->eth.type == htons(ETH_P_802_2)) {
1583                 /*
1584                  * Ethertype 802.2 is represented in the netlink with omitted
1585                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1586                  * 0xffff in the mask attribute.  Ethertype can also
1587                  * be wildcarded.
1588                  */
1589                 if (is_mask && output->eth.type)
1590                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1591                                                 output->eth.type))
1592                                 goto nla_put_failure;
1593                 goto unencap;
1594         }
1595
1596         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1597                 goto nla_put_failure;
1598
1599         if (eth_type_vlan(swkey->eth.type)) {
1600                 /* There are 3 VLAN tags, we don't know anything about the rest
1601                  * of the packet, so truncate here.
1602                  */
1603                 WARN_ON_ONCE(!(encap && in_encap));
1604                 goto unencap;
1605         }
1606
1607         if (swkey->eth.type == htons(ETH_P_IP)) {
1608                 struct ovs_key_ipv4 *ipv4_key;
1609
1610                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1611                 if (!nla)
1612                         goto nla_put_failure;
1613                 ipv4_key = nla_data(nla);
1614                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1615                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1616                 ipv4_key->ipv4_proto = output->ip.proto;
1617                 ipv4_key->ipv4_tos = output->ip.tos;
1618                 ipv4_key->ipv4_ttl = output->ip.ttl;
1619                 ipv4_key->ipv4_frag = output->ip.frag;
1620         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1621                 struct ovs_key_ipv6 *ipv6_key;
1622
1623                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1624                 if (!nla)
1625                         goto nla_put_failure;
1626                 ipv6_key = nla_data(nla);
1627                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1628                                 sizeof(ipv6_key->ipv6_src));
1629                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1630                                 sizeof(ipv6_key->ipv6_dst));
1631                 ipv6_key->ipv6_label = output->ipv6.label;
1632                 ipv6_key->ipv6_proto = output->ip.proto;
1633                 ipv6_key->ipv6_tclass = output->ip.tos;
1634                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1635                 ipv6_key->ipv6_frag = output->ip.frag;
1636         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1637                    swkey->eth.type == htons(ETH_P_RARP)) {
1638                 struct ovs_key_arp *arp_key;
1639
1640                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1641                 if (!nla)
1642                         goto nla_put_failure;
1643                 arp_key = nla_data(nla);
1644                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1645                 arp_key->arp_sip = output->ipv4.addr.src;
1646                 arp_key->arp_tip = output->ipv4.addr.dst;
1647                 arp_key->arp_op = htons(output->ip.proto);
1648                 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1649                 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1650         } else if (eth_p_mpls(swkey->eth.type)) {
1651                 struct ovs_key_mpls *mpls_key;
1652
1653                 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1654                 if (!nla)
1655                         goto nla_put_failure;
1656                 mpls_key = nla_data(nla);
1657                 mpls_key->mpls_lse = output->mpls.top_lse;
1658         }
1659
1660         if ((swkey->eth.type == htons(ETH_P_IP) ||
1661              swkey->eth.type == htons(ETH_P_IPV6)) &&
1662              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1663
1664                 if (swkey->ip.proto == IPPROTO_TCP) {
1665                         struct ovs_key_tcp *tcp_key;
1666
1667                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1668                         if (!nla)
1669                                 goto nla_put_failure;
1670                         tcp_key = nla_data(nla);
1671                         tcp_key->tcp_src = output->tp.src;
1672                         tcp_key->tcp_dst = output->tp.dst;
1673                         if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1674                                          output->tp.flags))
1675                                 goto nla_put_failure;
1676                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1677                         struct ovs_key_udp *udp_key;
1678
1679                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1680                         if (!nla)
1681                                 goto nla_put_failure;
1682                         udp_key = nla_data(nla);
1683                         udp_key->udp_src = output->tp.src;
1684                         udp_key->udp_dst = output->tp.dst;
1685                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1686                         struct ovs_key_sctp *sctp_key;
1687
1688                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1689                         if (!nla)
1690                                 goto nla_put_failure;
1691                         sctp_key = nla_data(nla);
1692                         sctp_key->sctp_src = output->tp.src;
1693                         sctp_key->sctp_dst = output->tp.dst;
1694                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1695                            swkey->ip.proto == IPPROTO_ICMP) {
1696                         struct ovs_key_icmp *icmp_key;
1697
1698                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1699                         if (!nla)
1700                                 goto nla_put_failure;
1701                         icmp_key = nla_data(nla);
1702                         icmp_key->icmp_type = ntohs(output->tp.src);
1703                         icmp_key->icmp_code = ntohs(output->tp.dst);
1704                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1705                            swkey->ip.proto == IPPROTO_ICMPV6) {
1706                         struct ovs_key_icmpv6 *icmpv6_key;
1707
1708                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1709                                                 sizeof(*icmpv6_key));
1710                         if (!nla)
1711                                 goto nla_put_failure;
1712                         icmpv6_key = nla_data(nla);
1713                         icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1714                         icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1715
1716                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1717                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1718                                 struct ovs_key_nd *nd_key;
1719
1720                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1721                                 if (!nla)
1722                                         goto nla_put_failure;
1723                                 nd_key = nla_data(nla);
1724                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1725                                                         sizeof(nd_key->nd_target));
1726                                 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1727                                 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1728                         }
1729                 }
1730         }
1731
1732 unencap:
1733         if (in_encap)
1734                 nla_nest_end(skb, in_encap);
1735         if (encap)
1736                 nla_nest_end(skb, encap);
1737
1738         return 0;
1739
1740 nla_put_failure:
1741         return -EMSGSIZE;
1742 }
1743
1744 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1745                     const struct sw_flow_key *output, int attr, bool is_mask,
1746                     struct sk_buff *skb)
1747 {
1748         int err;
1749         struct nlattr *nla;
1750
1751         nla = nla_nest_start(skb, attr);
1752         if (!nla)
1753                 return -EMSGSIZE;
1754         err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1755         if (err)
1756                 return err;
1757         nla_nest_end(skb, nla);
1758
1759         return 0;
1760 }
1761
1762 /* Called with ovs_mutex or RCU read lock. */
1763 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1764 {
1765         if (ovs_identifier_is_ufid(&flow->id))
1766                 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1767                                flow->id.ufid);
1768
1769         return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1770                                OVS_FLOW_ATTR_KEY, false, skb);
1771 }
1772
1773 /* Called with ovs_mutex or RCU read lock. */
1774 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1775 {
1776         return ovs_nla_put_key(&flow->key, &flow->key,
1777                                 OVS_FLOW_ATTR_KEY, false, skb);
1778 }
1779
1780 /* Called with ovs_mutex or RCU read lock. */
1781 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1782 {
1783         return ovs_nla_put_key(&flow->key, &flow->mask->key,
1784                                 OVS_FLOW_ATTR_MASK, true, skb);
1785 }
1786
1787 #define MAX_ACTIONS_BUFSIZE     (32 * 1024)
1788
1789 static struct sw_flow_actions *nla_alloc_flow_actions(int size)
1790 {
1791         struct sw_flow_actions *sfa;
1792
1793         WARN_ON_ONCE(size > MAX_ACTIONS_BUFSIZE);
1794
1795         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1796         if (!sfa)
1797                 return ERR_PTR(-ENOMEM);
1798
1799         sfa->actions_len = 0;
1800         return sfa;
1801 }
1802
1803 static void ovs_nla_free_set_action(const struct nlattr *a)
1804 {
1805         const struct nlattr *ovs_key = nla_data(a);
1806         struct ovs_tunnel_info *ovs_tun;
1807
1808         switch (nla_type(ovs_key)) {
1809         case OVS_KEY_ATTR_TUNNEL_INFO:
1810                 ovs_tun = nla_data(ovs_key);
1811                 dst_release((struct dst_entry *)ovs_tun->tun_dst);
1812                 break;
1813         }
1814 }
1815
1816 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1817 {
1818         const struct nlattr *a;
1819         int rem;
1820
1821         if (!sf_acts)
1822                 return;
1823
1824         nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1825                 switch (nla_type(a)) {
1826                 case OVS_ACTION_ATTR_SET:
1827                         ovs_nla_free_set_action(a);
1828                         break;
1829                 case OVS_ACTION_ATTR_CT:
1830                         ovs_ct_free_action(a);
1831                         break;
1832                 }
1833         }
1834
1835         kfree(sf_acts);
1836 }
1837
1838 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1839 {
1840         ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1841 }
1842
1843 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1844  * The caller must hold rcu_read_lock for this to be sensible. */
1845 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1846 {
1847         call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1848 }
1849
1850 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1851                                        int attr_len, bool log)
1852 {
1853
1854         struct sw_flow_actions *acts;
1855         int new_acts_size;
1856         size_t req_size = NLA_ALIGN(attr_len);
1857         int next_offset = offsetof(struct sw_flow_actions, actions) +
1858                                         (*sfa)->actions_len;
1859
1860         if (req_size <= (ksize(*sfa) - next_offset))
1861                 goto out;
1862
1863         new_acts_size = max(next_offset + req_size, ksize(*sfa) * 2);
1864
1865         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1866                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size) {
1867                         OVS_NLERR(log, "Flow action size exceeds max %u",
1868                                   MAX_ACTIONS_BUFSIZE);
1869                         return ERR_PTR(-EMSGSIZE);
1870                 }
1871                 new_acts_size = MAX_ACTIONS_BUFSIZE;
1872         }
1873
1874         acts = nla_alloc_flow_actions(new_acts_size);
1875         if (IS_ERR(acts))
1876                 return (void *)acts;
1877
1878         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1879         acts->actions_len = (*sfa)->actions_len;
1880         acts->orig_len = (*sfa)->orig_len;
1881         kfree(*sfa);
1882         *sfa = acts;
1883
1884 out:
1885         (*sfa)->actions_len += req_size;
1886         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1887 }
1888
1889 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1890                                    int attrtype, void *data, int len, bool log)
1891 {
1892         struct nlattr *a;
1893
1894         a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1895         if (IS_ERR(a))
1896                 return a;
1897
1898         a->nla_type = attrtype;
1899         a->nla_len = nla_attr_size(len);
1900
1901         if (data)
1902                 memcpy(nla_data(a), data, len);
1903         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1904
1905         return a;
1906 }
1907
1908 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1909                        int len, bool log)
1910 {
1911         struct nlattr *a;
1912
1913         a = __add_action(sfa, attrtype, data, len, log);
1914
1915         return PTR_ERR_OR_ZERO(a);
1916 }
1917
1918 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1919                                           int attrtype, bool log)
1920 {
1921         int used = (*sfa)->actions_len;
1922         int err;
1923
1924         err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1925         if (err)
1926                 return err;
1927
1928         return used;
1929 }
1930
1931 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1932                                          int st_offset)
1933 {
1934         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1935                                                                st_offset);
1936
1937         a->nla_len = sfa->actions_len - st_offset;
1938 }
1939
1940 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1941                                   const struct sw_flow_key *key,
1942                                   int depth, struct sw_flow_actions **sfa,
1943                                   __be16 eth_type, __be16 vlan_tci, bool log);
1944
1945 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1946                                     const struct sw_flow_key *key, int depth,
1947                                     struct sw_flow_actions **sfa,
1948                                     __be16 eth_type, __be16 vlan_tci, bool log)
1949 {
1950         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1951         const struct nlattr *probability, *actions;
1952         const struct nlattr *a;
1953         int rem, start, err, st_acts;
1954
1955         memset(attrs, 0, sizeof(attrs));
1956         nla_for_each_nested(a, attr, rem) {
1957                 int type = nla_type(a);
1958                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1959                         return -EINVAL;
1960                 attrs[type] = a;
1961         }
1962         if (rem)
1963                 return -EINVAL;
1964
1965         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1966         if (!probability || nla_len(probability) != sizeof(u32))
1967                 return -EINVAL;
1968
1969         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1970         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1971                 return -EINVAL;
1972
1973         /* validation done, copy sample action. */
1974         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1975         if (start < 0)
1976                 return start;
1977         err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1978                                  nla_data(probability), sizeof(u32), log);
1979         if (err)
1980                 return err;
1981         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1982         if (st_acts < 0)
1983                 return st_acts;
1984
1985         err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1986                                      eth_type, vlan_tci, log);
1987         if (err)
1988                 return err;
1989
1990         add_nested_action_end(*sfa, st_acts);
1991         add_nested_action_end(*sfa, start);
1992
1993         return 0;
1994 }
1995
1996 void ovs_match_init(struct sw_flow_match *match,
1997                     struct sw_flow_key *key,
1998                     bool reset_key,
1999                     struct sw_flow_mask *mask)
2000 {
2001         memset(match, 0, sizeof(*match));
2002         match->key = key;
2003         match->mask = mask;
2004
2005         if (reset_key)
2006                 memset(key, 0, sizeof(*key));
2007
2008         if (mask) {
2009                 memset(&mask->key, 0, sizeof(mask->key));
2010                 mask->range.start = mask->range.end = 0;
2011         }
2012 }
2013
2014 static int validate_geneve_opts(struct sw_flow_key *key)
2015 {
2016         struct geneve_opt *option;
2017         int opts_len = key->tun_opts_len;
2018         bool crit_opt = false;
2019
2020         option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
2021         while (opts_len > 0) {
2022                 int len;
2023
2024                 if (opts_len < sizeof(*option))
2025                         return -EINVAL;
2026
2027                 len = sizeof(*option) + option->length * 4;
2028                 if (len > opts_len)
2029                         return -EINVAL;
2030
2031                 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
2032
2033                 option = (struct geneve_opt *)((u8 *)option + len);
2034                 opts_len -= len;
2035         };
2036
2037         key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
2038
2039         return 0;
2040 }
2041
2042 static int validate_and_copy_set_tun(const struct nlattr *attr,
2043                                      struct sw_flow_actions **sfa, bool log)
2044 {
2045         struct sw_flow_match match;
2046         struct sw_flow_key key;
2047         struct metadata_dst *tun_dst;
2048         struct ip_tunnel_info *tun_info;
2049         struct ovs_tunnel_info *ovs_tun;
2050         struct nlattr *a;
2051         int err = 0, start, opts_type;
2052
2053         ovs_match_init(&match, &key, true, NULL);
2054         opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
2055         if (opts_type < 0)
2056                 return opts_type;
2057
2058         if (key.tun_opts_len) {
2059                 switch (opts_type) {
2060                 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2061                         err = validate_geneve_opts(&key);
2062                         if (err < 0)
2063                                 return err;
2064                         break;
2065                 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2066                         break;
2067                 }
2068         };
2069
2070         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
2071         if (start < 0)
2072                 return start;
2073
2074         tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
2075         if (!tun_dst)
2076                 return -ENOMEM;
2077
2078         err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
2079         if (err) {
2080                 dst_release((struct dst_entry *)tun_dst);
2081                 return err;
2082         }
2083
2084         a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
2085                          sizeof(*ovs_tun), log);
2086         if (IS_ERR(a)) {
2087                 dst_release((struct dst_entry *)tun_dst);
2088                 return PTR_ERR(a);
2089         }
2090
2091         ovs_tun = nla_data(a);
2092         ovs_tun->tun_dst = tun_dst;
2093
2094         tun_info = &tun_dst->u.tun_info;
2095         tun_info->mode = IP_TUNNEL_INFO_TX;
2096         if (key.tun_proto == AF_INET6)
2097                 tun_info->mode |= IP_TUNNEL_INFO_IPV6;
2098         tun_info->key = key.tun_key;
2099
2100         /* We need to store the options in the action itself since
2101          * everything else will go away after flow setup. We can append
2102          * it to tun_info and then point there.
2103          */
2104         ip_tunnel_info_opts_set(tun_info,
2105                                 TUN_METADATA_OPTS(&key, key.tun_opts_len),
2106                                 key.tun_opts_len);
2107         add_nested_action_end(*sfa, start);
2108
2109         return err;
2110 }
2111
2112 /* Return false if there are any non-masked bits set.
2113  * Mask follows data immediately, before any netlink padding.
2114  */
2115 static bool validate_masked(u8 *data, int len)
2116 {
2117         u8 *mask = data + len;
2118
2119         while (len--)
2120                 if (*data++ & ~*mask++)
2121                         return false;
2122
2123         return true;
2124 }
2125
2126 static int validate_set(const struct nlattr *a,
2127                         const struct sw_flow_key *flow_key,
2128                         struct sw_flow_actions **sfa,
2129                         bool *skip_copy, __be16 eth_type, bool masked, bool log)
2130 {
2131         const struct nlattr *ovs_key = nla_data(a);
2132         int key_type = nla_type(ovs_key);
2133         size_t key_len;
2134
2135         /* There can be only one key in a action */
2136         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2137                 return -EINVAL;
2138
2139         key_len = nla_len(ovs_key);
2140         if (masked)
2141                 key_len /= 2;
2142
2143         if (key_type > OVS_KEY_ATTR_MAX ||
2144             !check_attr_len(key_len, ovs_key_lens[key_type].len))
2145                 return -EINVAL;
2146
2147         if (masked && !validate_masked(nla_data(ovs_key), key_len))
2148                 return -EINVAL;
2149
2150         switch (key_type) {
2151         const struct ovs_key_ipv4 *ipv4_key;
2152         const struct ovs_key_ipv6 *ipv6_key;
2153         int err;
2154
2155         case OVS_KEY_ATTR_PRIORITY:
2156         case OVS_KEY_ATTR_SKB_MARK:
2157         case OVS_KEY_ATTR_CT_MARK:
2158         case OVS_KEY_ATTR_CT_LABELS:
2159         case OVS_KEY_ATTR_ETHERNET:
2160                 break;
2161
2162         case OVS_KEY_ATTR_TUNNEL:
2163                 if (masked)
2164                         return -EINVAL; /* Masked tunnel set not supported. */
2165
2166                 *skip_copy = true;
2167                 err = validate_and_copy_set_tun(a, sfa, log);
2168                 if (err)
2169                         return err;
2170                 break;
2171
2172         case OVS_KEY_ATTR_IPV4:
2173                 if (eth_type != htons(ETH_P_IP))
2174                         return -EINVAL;
2175
2176                 ipv4_key = nla_data(ovs_key);
2177
2178                 if (masked) {
2179                         const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2180
2181                         /* Non-writeable fields. */
2182                         if (mask->ipv4_proto || mask->ipv4_frag)
2183                                 return -EINVAL;
2184                 } else {
2185                         if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2186                                 return -EINVAL;
2187
2188                         if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2189                                 return -EINVAL;
2190                 }
2191                 break;
2192
2193         case OVS_KEY_ATTR_IPV6:
2194                 if (eth_type != htons(ETH_P_IPV6))
2195                         return -EINVAL;
2196
2197                 ipv6_key = nla_data(ovs_key);
2198
2199                 if (masked) {
2200                         const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2201
2202                         /* Non-writeable fields. */
2203                         if (mask->ipv6_proto || mask->ipv6_frag)
2204                                 return -EINVAL;
2205
2206                         /* Invalid bits in the flow label mask? */
2207                         if (ntohl(mask->ipv6_label) & 0xFFF00000)
2208                                 return -EINVAL;
2209                 } else {
2210                         if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2211                                 return -EINVAL;
2212
2213                         if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2214                                 return -EINVAL;
2215                 }
2216                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2217                         return -EINVAL;
2218
2219                 break;
2220
2221         case OVS_KEY_ATTR_TCP:
2222                 if ((eth_type != htons(ETH_P_IP) &&
2223                      eth_type != htons(ETH_P_IPV6)) ||
2224                     flow_key->ip.proto != IPPROTO_TCP)
2225                         return -EINVAL;
2226
2227                 break;
2228
2229         case OVS_KEY_ATTR_UDP:
2230                 if ((eth_type != htons(ETH_P_IP) &&
2231                      eth_type != htons(ETH_P_IPV6)) ||
2232                     flow_key->ip.proto != IPPROTO_UDP)
2233                         return -EINVAL;
2234
2235                 break;
2236
2237         case OVS_KEY_ATTR_MPLS:
2238                 if (!eth_p_mpls(eth_type))
2239                         return -EINVAL;
2240                 break;
2241
2242         case OVS_KEY_ATTR_SCTP:
2243                 if ((eth_type != htons(ETH_P_IP) &&
2244                      eth_type != htons(ETH_P_IPV6)) ||
2245                     flow_key->ip.proto != IPPROTO_SCTP)
2246                         return -EINVAL;
2247
2248                 break;
2249
2250         default:
2251                 return -EINVAL;
2252         }
2253
2254         /* Convert non-masked non-tunnel set actions to masked set actions. */
2255         if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2256                 int start, len = key_len * 2;
2257                 struct nlattr *at;
2258
2259                 *skip_copy = true;
2260
2261                 start = add_nested_action_start(sfa,
2262                                                 OVS_ACTION_ATTR_SET_TO_MASKED,
2263                                                 log);
2264                 if (start < 0)
2265                         return start;
2266
2267                 at = __add_action(sfa, key_type, NULL, len, log);
2268                 if (IS_ERR(at))
2269                         return PTR_ERR(at);
2270
2271                 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2272                 memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2273                 /* Clear non-writeable bits from otherwise writeable fields. */
2274                 if (key_type == OVS_KEY_ATTR_IPV6) {
2275                         struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2276
2277                         mask->ipv6_label &= htonl(0x000FFFFF);
2278                 }
2279                 add_nested_action_end(*sfa, start);
2280         }
2281
2282         return 0;
2283 }
2284
2285 static int validate_userspace(const struct nlattr *attr)
2286 {
2287         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2288                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2289                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2290                 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2291         };
2292         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2293         int error;
2294
2295         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2296                                  attr, userspace_policy);
2297         if (error)
2298                 return error;
2299
2300         if (!a[OVS_USERSPACE_ATTR_PID] ||
2301             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2302                 return -EINVAL;
2303
2304         return 0;
2305 }
2306
2307 static int copy_action(const struct nlattr *from,
2308                        struct sw_flow_actions **sfa, bool log)
2309 {
2310         int totlen = NLA_ALIGN(from->nla_len);
2311         struct nlattr *to;
2312
2313         to = reserve_sfa_size(sfa, from->nla_len, log);
2314         if (IS_ERR(to))
2315                 return PTR_ERR(to);
2316
2317         memcpy(to, from, totlen);
2318         return 0;
2319 }
2320
2321 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2322                                   const struct sw_flow_key *key,
2323                                   int depth, struct sw_flow_actions **sfa,
2324                                   __be16 eth_type, __be16 vlan_tci, bool log)
2325 {
2326         const struct nlattr *a;
2327         int rem, err;
2328
2329         if (depth >= SAMPLE_ACTION_DEPTH)
2330                 return -EOVERFLOW;
2331
2332         nla_for_each_nested(a, attr, rem) {
2333                 /* Expected argument lengths, (u32)-1 for variable length. */
2334                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2335                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2336                         [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2337                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2338                         [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2339                         [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2340                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2341                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
2342                         [OVS_ACTION_ATTR_SET] = (u32)-1,
2343                         [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2344                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2345                         [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2346                         [OVS_ACTION_ATTR_CT] = (u32)-1,
2347                         [OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
2348                 };
2349                 const struct ovs_action_push_vlan *vlan;
2350                 int type = nla_type(a);
2351                 bool skip_copy;
2352
2353                 if (type > OVS_ACTION_ATTR_MAX ||
2354                     (action_lens[type] != nla_len(a) &&
2355                      action_lens[type] != (u32)-1))
2356                         return -EINVAL;
2357
2358                 skip_copy = false;
2359                 switch (type) {
2360                 case OVS_ACTION_ATTR_UNSPEC:
2361                         return -EINVAL;
2362
2363                 case OVS_ACTION_ATTR_USERSPACE:
2364                         err = validate_userspace(a);
2365                         if (err)
2366                                 return err;
2367                         break;
2368
2369                 case OVS_ACTION_ATTR_OUTPUT:
2370                         if (nla_get_u32(a) >= DP_MAX_PORTS)
2371                                 return -EINVAL;
2372                         break;
2373
2374                 case OVS_ACTION_ATTR_TRUNC: {
2375                         const struct ovs_action_trunc *trunc = nla_data(a);
2376
2377                         if (trunc->max_len < ETH_HLEN)
2378                                 return -EINVAL;
2379                         break;
2380                 }
2381
2382                 case OVS_ACTION_ATTR_HASH: {
2383                         const struct ovs_action_hash *act_hash = nla_data(a);
2384
2385                         switch (act_hash->hash_alg) {
2386                         case OVS_HASH_ALG_L4:
2387                                 break;
2388                         default:
2389                                 return  -EINVAL;
2390                         }
2391
2392                         break;
2393                 }
2394
2395                 case OVS_ACTION_ATTR_POP_VLAN:
2396                         vlan_tci = htons(0);
2397                         break;
2398
2399                 case OVS_ACTION_ATTR_PUSH_VLAN:
2400                         vlan = nla_data(a);
2401                         if (!eth_type_vlan(vlan->vlan_tpid))
2402                                 return -EINVAL;
2403                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2404                                 return -EINVAL;
2405                         vlan_tci = vlan->vlan_tci;
2406                         break;
2407
2408                 case OVS_ACTION_ATTR_RECIRC:
2409                         break;
2410
2411                 case OVS_ACTION_ATTR_PUSH_MPLS: {
2412                         const struct ovs_action_push_mpls *mpls = nla_data(a);
2413
2414                         if (!eth_p_mpls(mpls->mpls_ethertype))
2415                                 return -EINVAL;
2416                         /* Prohibit push MPLS other than to a white list
2417                          * for packets that have a known tag order.
2418                          */
2419                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2420                             (eth_type != htons(ETH_P_IP) &&
2421                              eth_type != htons(ETH_P_IPV6) &&
2422                              eth_type != htons(ETH_P_ARP) &&
2423                              eth_type != htons(ETH_P_RARP) &&
2424                              !eth_p_mpls(eth_type)))
2425                                 return -EINVAL;
2426                         eth_type = mpls->mpls_ethertype;
2427                         break;
2428                 }
2429
2430                 case OVS_ACTION_ATTR_POP_MPLS:
2431                         if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2432                             !eth_p_mpls(eth_type))
2433                                 return -EINVAL;
2434
2435                         /* Disallow subsequent L2.5+ set and mpls_pop actions
2436                          * as there is no check here to ensure that the new
2437                          * eth_type is valid and thus set actions could
2438                          * write off the end of the packet or otherwise
2439                          * corrupt it.
2440                          *
2441                          * Support for these actions is planned using packet
2442                          * recirculation.
2443                          */
2444                         eth_type = htons(0);
2445                         break;
2446
2447                 case OVS_ACTION_ATTR_SET:
2448                         err = validate_set(a, key, sfa,
2449                                            &skip_copy, eth_type, false, log);
2450                         if (err)
2451                                 return err;
2452                         break;
2453
2454                 case OVS_ACTION_ATTR_SET_MASKED:
2455                         err = validate_set(a, key, sfa,
2456                                            &skip_copy, eth_type, true, log);
2457                         if (err)
2458                                 return err;
2459                         break;
2460
2461                 case OVS_ACTION_ATTR_SAMPLE:
2462                         err = validate_and_copy_sample(net, a, key, depth, sfa,
2463                                                        eth_type, vlan_tci, log);
2464                         if (err)
2465                                 return err;
2466                         skip_copy = true;
2467                         break;
2468
2469                 case OVS_ACTION_ATTR_CT:
2470                         err = ovs_ct_copy_action(net, a, key, sfa, log);
2471                         if (err)
2472                                 return err;
2473                         skip_copy = true;
2474                         break;
2475
2476                 default:
2477                         OVS_NLERR(log, "Unknown Action type %d", type);
2478                         return -EINVAL;
2479                 }
2480                 if (!skip_copy) {
2481                         err = copy_action(a, sfa, log);
2482                         if (err)
2483                                 return err;
2484                 }
2485         }
2486
2487         if (rem > 0)
2488                 return -EINVAL;
2489
2490         return 0;
2491 }
2492
2493 /* 'key' must be the masked key. */
2494 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2495                          const struct sw_flow_key *key,
2496                          struct sw_flow_actions **sfa, bool log)
2497 {
2498         int err;
2499
2500         *sfa = nla_alloc_flow_actions(min(nla_len(attr), MAX_ACTIONS_BUFSIZE));
2501         if (IS_ERR(*sfa))
2502                 return PTR_ERR(*sfa);
2503
2504         (*sfa)->orig_len = nla_len(attr);
2505         err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2506                                      key->eth.vlan.tci, log);
2507         if (err)
2508                 ovs_nla_free_flow_actions(*sfa);
2509
2510         return err;
2511 }
2512
2513 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2514 {
2515         const struct nlattr *a;
2516         struct nlattr *start;
2517         int err = 0, rem;
2518
2519         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2520         if (!start)
2521                 return -EMSGSIZE;
2522
2523         nla_for_each_nested(a, attr, rem) {
2524                 int type = nla_type(a);
2525                 struct nlattr *st_sample;
2526
2527                 switch (type) {
2528                 case OVS_SAMPLE_ATTR_PROBABILITY:
2529                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2530                                     sizeof(u32), nla_data(a)))
2531                                 return -EMSGSIZE;
2532                         break;
2533                 case OVS_SAMPLE_ATTR_ACTIONS:
2534                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2535                         if (!st_sample)
2536                                 return -EMSGSIZE;
2537                         err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2538                         if (err)
2539                                 return err;
2540                         nla_nest_end(skb, st_sample);
2541                         break;
2542                 }
2543         }
2544
2545         nla_nest_end(skb, start);
2546         return err;
2547 }
2548
2549 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2550 {
2551         const struct nlattr *ovs_key = nla_data(a);
2552         int key_type = nla_type(ovs_key);
2553         struct nlattr *start;
2554         int err;
2555
2556         switch (key_type) {
2557         case OVS_KEY_ATTR_TUNNEL_INFO: {
2558                 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2559                 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2560
2561                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2562                 if (!start)
2563                         return -EMSGSIZE;
2564
2565                 err =  ip_tun_to_nlattr(skb, &tun_info->key,
2566                                         ip_tunnel_info_opts(tun_info),
2567                                         tun_info->options_len,
2568                                         ip_tunnel_info_af(tun_info));
2569                 if (err)
2570                         return err;
2571                 nla_nest_end(skb, start);
2572                 break;
2573         }
2574         default:
2575                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2576                         return -EMSGSIZE;
2577                 break;
2578         }
2579
2580         return 0;
2581 }
2582
2583 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2584                                                 struct sk_buff *skb)
2585 {
2586         const struct nlattr *ovs_key = nla_data(a);
2587         struct nlattr *nla;
2588         size_t key_len = nla_len(ovs_key) / 2;
2589
2590         /* Revert the conversion we did from a non-masked set action to
2591          * masked set action.
2592          */
2593         nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2594         if (!nla)
2595                 return -EMSGSIZE;
2596
2597         if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2598                 return -EMSGSIZE;
2599
2600         nla_nest_end(skb, nla);
2601         return 0;
2602 }
2603
2604 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2605 {
2606         const struct nlattr *a;
2607         int rem, err;
2608
2609         nla_for_each_attr(a, attr, len, rem) {
2610                 int type = nla_type(a);
2611
2612                 switch (type) {
2613                 case OVS_ACTION_ATTR_SET:
2614                         err = set_action_to_attr(a, skb);
2615                         if (err)
2616                                 return err;
2617                         break;
2618
2619                 case OVS_ACTION_ATTR_SET_TO_MASKED:
2620                         err = masked_set_action_to_set_action_attr(a, skb);
2621                         if (err)
2622                                 return err;
2623                         break;
2624
2625                 case OVS_ACTION_ATTR_SAMPLE:
2626                         err = sample_action_to_attr(a, skb);
2627                         if (err)
2628                                 return err;
2629                         break;
2630
2631                 case OVS_ACTION_ATTR_CT:
2632                         err = ovs_ct_action_to_attr(nla_data(a), skb);
2633                         if (err)
2634                                 return err;
2635                         break;
2636
2637                 default:
2638                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
2639                                 return -EMSGSIZE;
2640                         break;
2641                 }
2642         }
2643
2644         return 0;
2645 }