datapath: Move MASKED* macros to datapath.h
[cascardo/ovs.git] / datapath / flow_netlink.c
index 890a889..f95aa14 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2007-2014 Nicira, Inc.
+ * Copyright (c) 2007-2015 Nicira, Inc.
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of version 2 of the GNU General Public
 #include <net/ipv6.h>
 #include <net/ndisc.h>
 #include <net/mpls.h>
+#include <net/vxlan.h>
 
 #include "datapath.h"
 #include "flow.h"
 #include "flow_netlink.h"
-#include "vport-vxlan.h"
+#include "gso.h"
 
 struct ovs_len_tbl {
        int len;
@@ -57,6 +58,7 @@ struct ovs_len_tbl {
 };
 
 #define OVS_ATTR_NESTED -1
+#define OVS_ATTR_VARIABLE -2
 
 static void update_range(struct sw_flow_match *match,
                         size_t offset, size_t size, bool is_mask)
@@ -281,7 +283,7 @@ size_t ovs_key_attr_size(void)
        /* Whenever adding new OVS_KEY_ FIELDS, we should consider
         * updating this function.
         */
-       BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 22);
+       BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 26);
 
        return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
                + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
@@ -300,6 +302,10 @@ size_t ovs_key_attr_size(void)
                + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 }
 
+static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
+       [OVS_VXLAN_EXT_GBP]         = { .len = sizeof(u32) },
+};
+
 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
        [OVS_TUNNEL_KEY_ATTR_ID]            = { .len = sizeof(u64) },
        [OVS_TUNNEL_KEY_ATTR_IPV4_SRC]      = { .len = sizeof(u32) },
@@ -311,8 +317,9 @@ static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1]
        [OVS_TUNNEL_KEY_ATTR_TP_SRC]        = { .len = sizeof(u16) },
        [OVS_TUNNEL_KEY_ATTR_TP_DST]        = { .len = sizeof(u16) },
        [OVS_TUNNEL_KEY_ATTR_OAM]           = { .len = 0 },
-       [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_NESTED },
-       [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED },
+       [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
+       [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
+                                               .next = ovs_vxlan_ext_key_lens },
 };
 
 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
@@ -340,6 +347,14 @@ static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
                                     .next = ovs_tunnel_key_lens, },
        [OVS_KEY_ATTR_MPLS]      = { .len = sizeof(struct ovs_key_mpls) },
 };
+
+static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
+{
+       return expected_len == attr_len ||
+              expected_len == OVS_ATTR_NESTED ||
+              expected_len == OVS_ATTR_VARIABLE;
+}
+
 static bool is_all_zero(const u8 *fp, size_t size)
 {
        int i;
@@ -379,7 +394,7 @@ static int __parse_flow_nlattrs(const struct nlattr *attr,
                }
 
                expected_len = ovs_key_lens[type].len;
-               if (nla_len(nla) != expected_len && expected_len != OVS_ATTR_NESTED) {
+               if (!check_attr_len(nla_len(nla), expected_len)) {
                        OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
                                  type, nla_len(nla), expected_len);
                        return -EINVAL;
@@ -464,29 +479,50 @@ static int genev_tun_opt_from_nlattr(const struct nlattr *a,
        return 0;
 }
 
-static const struct nla_policy vxlan_opt_policy[OVS_VXLAN_EXT_MAX + 1] = {
-       [OVS_VXLAN_EXT_GBP]     = { .type = NLA_U32 },
-};
-
-static int vxlan_tun_opt_from_nlattr(const struct nlattr *a,
+static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
                                     struct sw_flow_match *match, bool is_mask,
                                     bool log)
 {
-       struct nlattr *tb[OVS_VXLAN_EXT_MAX+1];
+       struct nlattr *a;
+       int rem;
        unsigned long opt_key_offset;
-       struct ovs_vxlan_opts opts;
-       int err;
+       struct vxlan_metadata opts;
 
        BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
 
-       err = nla_parse_nested(tb, OVS_VXLAN_EXT_MAX, a, vxlan_opt_policy);
-       if (err < 0)
-               return err;
-
        memset(&opts, 0, sizeof(opts));
+       nla_for_each_nested(a, attr, rem) {
+               int type = nla_type(a);
 
-       if (tb[OVS_VXLAN_EXT_GBP])
-               opts.gbp = nla_get_u32(tb[OVS_VXLAN_EXT_GBP]);
+               if (type > OVS_VXLAN_EXT_MAX) {
+                       OVS_NLERR(log, "VXLAN extension %d out of range max %d",
+                                 type, OVS_VXLAN_EXT_MAX);
+                       return -EINVAL;
+               }
+
+               if (!check_attr_len(nla_len(a),
+                                   ovs_vxlan_ext_key_lens[type].len)) {
+                       OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
+                                 type, nla_len(a),
+                                 ovs_vxlan_ext_key_lens[type].len);
+                       return -EINVAL;
+               }
+
+               switch (type) {
+               case OVS_VXLAN_EXT_GBP:
+                       opts.gbp = nla_get_u32(a);
+                       break;
+               default:
+                       OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
+                                 type);
+                       return -EINVAL;
+               }
+       }
+       if (rem) {
+               OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
+                         rem);
+               return -EINVAL;
+       }
 
        if (!is_mask)
                SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
@@ -519,8 +555,8 @@ static int ipv4_tun_from_nlattr(const struct nlattr *attr,
                        return -EINVAL;
                }
 
-               if (ovs_tunnel_key_lens[type].len != nla_len(a) &&
-                  ovs_tunnel_key_lens[type].len != OVS_ATTR_NESTED) {
+               if (!check_attr_len(nla_len(a),
+                                   ovs_tunnel_key_lens[type].len)) {
                        OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
                                  type, nla_len(a), ovs_tunnel_key_lens[type].len);
                        return -EINVAL;
@@ -533,19 +569,19 @@ static int ipv4_tun_from_nlattr(const struct nlattr *attr,
                        tun_flags |= TUNNEL_KEY;
                        break;
                case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
-                       SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
-                                       nla_get_be32(a), is_mask);
+                       SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
+                                       nla_get_in_addr(a), is_mask);
                        break;
                case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
-                       SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
-                                       nla_get_be32(a), is_mask);
+                       SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
+                                       nla_get_in_addr(a), is_mask);
                        break;
                case OVS_TUNNEL_KEY_ATTR_TOS:
-                       SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
+                       SW_FLOW_KEY_PUT(match, tun_key.tos,
                                        nla_get_u8(a), is_mask);
                        break;
                case OVS_TUNNEL_KEY_ATTR_TTL:
-                       SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
+                       SW_FLOW_KEY_PUT(match, tun_key.ttl,
                                        nla_get_u8(a), is_mask);
                        ttl = true;
                        break;
@@ -608,7 +644,7 @@ static int ipv4_tun_from_nlattr(const struct nlattr *attr,
        }
 
        if (!is_mask) {
-               if (!match->key->tun_key.ipv4_dst) {
+               if (!match->key->tun_key.u.ipv4.dst) {
                        OVS_NLERR(log, "IPv4 tunnel dst address is zero");
                        return -EINVAL;
                }
@@ -625,7 +661,7 @@ static int ipv4_tun_from_nlattr(const struct nlattr *attr,
 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
                               const void *tun_opts, int swkey_tun_opts_len)
 {
-       const struct ovs_vxlan_opts *opts = tun_opts;
+       const struct vxlan_metadata *opts = tun_opts;
        struct nlattr *nla;
 
        nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
@@ -640,22 +676,24 @@ static int vxlan_opt_to_nlattr(struct sk_buff *skb,
 }
 
 static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
-                               const struct ovs_key_ipv4_tunnel *output,
+                               const struct ip_tunnel_key *output,
                                const void *tun_opts, int swkey_tun_opts_len)
 {
        if (output->tun_flags & TUNNEL_KEY &&
            nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
                return -EMSGSIZE;
-       if (output->ipv4_src &&
-           nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
+       if (output->u.ipv4.src &&
+           nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
+                           output->u.ipv4.src))
                return -EMSGSIZE;
-       if (output->ipv4_dst &&
-           nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
+       if (output->u.ipv4.dst &&
+           nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
+                           output->u.ipv4.dst))
                return -EMSGSIZE;
-       if (output->ipv4_tos &&
-           nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
+       if (output->tos &&
+           nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
                return -EMSGSIZE;
-       if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
+       if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
                return -EMSGSIZE;
        if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
            nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
@@ -677,8 +715,8 @@ static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
                    nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
                            swkey_tun_opts_len, tun_opts))
                        return -EMSGSIZE;
-               else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
-                       vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
+               else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
+                        vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
                        return -EMSGSIZE;
        }
 
@@ -686,7 +724,7 @@ static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
 }
 
 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
-                             const struct ovs_key_ipv4_tunnel *output,
+                             const struct ip_tunnel_key *output,
                              const void *tun_opts, int swkey_tun_opts_len)
 {
        struct nlattr *nla;
@@ -705,10 +743,11 @@ static int ipv4_tun_to_nlattr(struct sk_buff *skb,
 }
 
 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
-                                 const struct ovs_tunnel_info *egress_tun_info)
+                                 const struct ip_tunnel_info *egress_tun_info,
+                                 const void *egress_tun_opts)
 {
-       return __ipv4_tun_to_nlattr(skb, &egress_tun_info->tunnel,
-                                   egress_tun_info->options,
+       return __ipv4_tun_to_nlattr(skb, &egress_tun_info->key,
+                                   egress_tun_opts,
                                    egress_tun_info->options_len);
 }
 
@@ -813,7 +852,7 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
                if (is_mask) {
                        /* Always exact match EtherType. */
                        eth_type = htons(0xffff);
-               } else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
+               } else if (!eth_proto_is_802_3(eth_type)) {
                        OVS_NLERR(log, "EtherType %x is less than min %x",
                                  ntohs(eth_type), ETH_P_802_3_MIN);
                        return -EINVAL;
@@ -825,7 +864,7 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
                SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
        }
 
-       if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
+       if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
                const struct ovs_key_ipv4 *ipv4_key;
 
                ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
@@ -846,7 +885,7 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
                                ipv4_key->ipv4_src, is_mask);
                SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
                                ipv4_key->ipv4_dst, is_mask);
-               attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
+               attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
        }
 
        if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
@@ -858,12 +897,14 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
                                  ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
                        return -EINVAL;
                }
+
                if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
                        OVS_NLERR(log,
                                  "Invalid IPv6 flow label value (value=%x, max=%x).",
                                  ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
                        return -EINVAL;
                }
+
                SW_FLOW_KEY_PUT(match, ipv6.label,
                                ipv6_key->ipv6_label, is_mask);
                SW_FLOW_KEY_PUT(match, ip.proto,
@@ -1008,10 +1049,13 @@ static void nlattr_set(struct nlattr *attr, u8 val,
 
        /* The nlattr stream should already have been validated */
        nla_for_each_nested(nla, attr, rem) {
-               if (tbl && tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
-                       nlattr_set(nla, val, tbl[nla_type(nla)].next);
-               else
+               if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
+                       if (tbl[nla_type(nla)].next)
+                               tbl = tbl[nla_type(nla)].next;
+                       nlattr_set(nla, val, tbl);
+               } else {
                        memset(nla_data(nla), val, nla_len(nla));
+               }
        }
 }
 
@@ -1112,7 +1156,7 @@ int ovs_nla_get_match(struct sw_flow_match *match,
                        /* The userspace does not send tunnel attributes that
                         * are 0, but we should not wildcard them nonetheless.
                         */
-                       if (match->key->tun_key.ipv4_dst)
+                       if (match->key->tun_key.u.ipv4.dst)
                                SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
                                                         0xff, true);
 
@@ -1284,7 +1328,7 @@ static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
        if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
                goto nla_put_failure;
 
-       if ((swkey->tun_key.ipv4_dst || is_mask)) {
+       if ((swkey->tun_key.u.ipv4.dst || is_mask)) {
                const void *opts = NULL;
 
                if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
@@ -1545,20 +1589,49 @@ static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
        return sfa;
 }
 
-/* RCU callback used by ovs_nla_free_flow_actions. */
-static void rcu_free_acts_callback(struct rcu_head *rcu)
+static void ovs_nla_free_set_action(const struct nlattr *a)
+{
+       const struct nlattr *ovs_key = nla_data(a);
+       struct ovs_tunnel_info *ovs_tun;
+
+       switch (nla_type(ovs_key)) {
+       case OVS_KEY_ATTR_TUNNEL_INFO:
+               ovs_tun = nla_data(ovs_key);
+               ovs_dst_release((struct dst_entry *)ovs_tun->tun_dst);
+               break;
+       }
+}
+
+void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
 {
-       struct sw_flow_actions *sf_acts = container_of(rcu,
-                       struct sw_flow_actions, rcu);
+       const struct nlattr *a;
+       int rem;
+
+       if (!sf_acts)
+               return;
+
+       nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
+               switch (nla_type(a)) {
+               case OVS_ACTION_ATTR_SET:
+                       ovs_nla_free_set_action(a);
+                       break;
+               }
+       }
+
        kfree(sf_acts);
 }
 
+static void __ovs_nla_free_flow_actions(struct rcu_head *head)
+{
+       ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
+}
+
 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
  * The caller must hold rcu_read_lock for this to be sensible.
  */
-void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
+void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
 {
-       call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
+       call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
 }
 
 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
@@ -1705,16 +1778,6 @@ static int validate_and_copy_sample(const struct nlattr *attr,
        return 0;
 }
 
-static int validate_tp_port(const struct sw_flow_key *flow_key,
-                           __be16 eth_type)
-{
-       if ((eth_type == htons(ETH_P_IP) || eth_type == htons(ETH_P_IPV6)) &&
-           (flow_key->tp.src || flow_key->tp.dst))
-               return 0;
-
-       return -EINVAL;
-}
-
 void ovs_match_init(struct sw_flow_match *match,
                    struct sw_flow_key *key,
                    struct sw_flow_mask *mask)
@@ -1764,10 +1827,11 @@ static int validate_and_copy_set_tun(const struct nlattr *attr,
 {
        struct sw_flow_match match;
        struct sw_flow_key key;
-       struct ovs_tunnel_info *tun_info;
+       struct metadata_dst *tun_dst;
+       struct ip_tunnel_info *tun_info;
+       struct ovs_tunnel_info *ovs_tun;
        struct nlattr *a;
-       int start, opts_type;
-       int err = 0;
+       int err = 0, start, opts_type;
 
        ovs_match_init(&match, &key, NULL);
        opts_type = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
@@ -1790,48 +1854,72 @@ static int validate_and_copy_set_tun(const struct nlattr *attr,
        if (start < 0)
                return start;
 
+       tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
+       if (!tun_dst)
+               return -ENOMEM;
+
        a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
-                        sizeof(*tun_info) + key.tun_opts_len, log);
-       if (IS_ERR(a))
+                        sizeof(*ovs_tun), log);
+       if (IS_ERR(a)) {
+               ovs_dst_release((struct dst_entry *)tun_dst);
                return PTR_ERR(a);
+       }
 
-       tun_info = nla_data(a);
-       tun_info->tunnel = key.tun_key;
-       tun_info->options_len = key.tun_opts_len;
+       ovs_tun = nla_data(a);
+       ovs_tun->tun_dst = tun_dst;
 
-       if (tun_info->options_len) {
-               /* We need to store the options in the action itself since
-                * everything else will go away after flow setup. We can append
-                * it to tun_info and then point there.
-                */
-               memcpy((tun_info + 1),
-                      TUN_METADATA_OPTS(&key, key.tun_opts_len), key.tun_opts_len);
-               tun_info->options = (tun_info + 1);
-
-       } else {
-               tun_info->options = NULL;
-       }
+       tun_info = &tun_dst->u.tun_info;
+       tun_info->mode = IP_TUNNEL_INFO_TX;
+       tun_info->key = key.tun_key;
 
+       /* We need to store the options in the action itself since
+        * everything else will go away after flow setup. We can append
+        * it to tun_info and then point there.
+        */
+       ip_tunnel_info_opts_set(tun_info,
+                               TUN_METADATA_OPTS(&key, key.tun_opts_len),
+                               key.tun_opts_len);
        add_nested_action_end(*sfa, start);
 
        return err;
 }
 
+/* Return false if there are any non-masked bits set.
+ * Mask follows data immediately, before any netlink padding.
+ */
+static bool validate_masked(u8 *data, int len)
+{
+       u8 *mask = data + len;
+
+       while (len--)
+               if (*data++ & ~*mask++)
+                       return false;
+
+       return true;
+}
+
 static int validate_set(const struct nlattr *a,
                        const struct sw_flow_key *flow_key,
                        struct sw_flow_actions **sfa,
-                       bool *set_tun, __be16 eth_type, bool log)
+                       bool *skip_copy, __be16 eth_type, bool masked, bool log)
 {
        const struct nlattr *ovs_key = nla_data(a);
        int key_type = nla_type(ovs_key);
+       size_t key_len;
 
        /* There can be only one key in a action */
        if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
                return -EINVAL;
 
+       key_len = nla_len(ovs_key);
+       if (masked)
+               key_len /= 2;
+
        if (key_type > OVS_KEY_ATTR_MAX ||
-           (ovs_key_lens[key_type].len != nla_len(ovs_key) &&
-            ovs_key_lens[key_type].len != OVS_ATTR_NESTED))
+           !check_attr_len(key_len, ovs_key_lens[key_type].len))
+               return -EINVAL;
+
+       if (masked && !validate_masked(nla_data(ovs_key), key_len))
                return -EINVAL;
 
        switch (key_type) {
@@ -1848,7 +1936,10 @@ static int validate_set(const struct nlattr *a,
                if (eth_p_mpls(eth_type))
                        return -EINVAL;
 
-               *set_tun = true;
+               if (masked)
+                       return -EINVAL; /* Masked tunnel set not supported. */
+
+               *skip_copy = true;
                err = validate_and_copy_set_tun(a, sfa, log);
                if (err)
                        return err;
@@ -1858,48 +1949,66 @@ static int validate_set(const struct nlattr *a,
                if (eth_type != htons(ETH_P_IP))
                        return -EINVAL;
 
-               if (!flow_key->ip.proto)
-                       return -EINVAL;
-
                ipv4_key = nla_data(ovs_key);
-               if (ipv4_key->ipv4_proto != flow_key->ip.proto)
-                       return -EINVAL;
 
-               if (ipv4_key->ipv4_frag != flow_key->ip.frag)
-                       return -EINVAL;
+               if (masked) {
+                       const struct ovs_key_ipv4 *mask = ipv4_key + 1;
 
+                       /* Non-writeable fields. */
+                       if (mask->ipv4_proto || mask->ipv4_frag)
+                               return -EINVAL;
+               } else {
+                       if (ipv4_key->ipv4_proto != flow_key->ip.proto)
+                               return -EINVAL;
+
+                       if (ipv4_key->ipv4_frag != flow_key->ip.frag)
+                               return -EINVAL;
+               }
                break;
 
        case OVS_KEY_ATTR_IPV6:
                if (eth_type != htons(ETH_P_IPV6))
                        return -EINVAL;
 
-               if (!flow_key->ip.proto)
-                       return -EINVAL;
-
                ipv6_key = nla_data(ovs_key);
-               if (ipv6_key->ipv6_proto != flow_key->ip.proto)
-                       return -EINVAL;
 
-               if (ipv6_key->ipv6_frag != flow_key->ip.frag)
-                       return -EINVAL;
+               if (masked) {
+                       const struct ovs_key_ipv6 *mask = ipv6_key + 1;
+
+                       /* Non-writeable fields. */
+                       if (mask->ipv6_proto || mask->ipv6_frag)
+                               return -EINVAL;
+
+                       /* Invalid bits in the flow label mask? */
+                       if (ntohl(mask->ipv6_label) & 0xFFF00000)
+                               return -EINVAL;
+               } else {
+                       if (ipv6_key->ipv6_proto != flow_key->ip.proto)
+                               return -EINVAL;
 
+                       if (ipv6_key->ipv6_frag != flow_key->ip.frag)
+                               return -EINVAL;
+               }
                if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
                        return -EINVAL;
 
                break;
 
        case OVS_KEY_ATTR_TCP:
-               if (flow_key->ip.proto != IPPROTO_TCP)
+               if ((eth_type != htons(ETH_P_IP) &&
+                    eth_type != htons(ETH_P_IPV6)) ||
+                   flow_key->ip.proto != IPPROTO_TCP)
                        return -EINVAL;
 
-               return validate_tp_port(flow_key, eth_type);
+               break;
 
        case OVS_KEY_ATTR_UDP:
-               if (flow_key->ip.proto != IPPROTO_UDP)
+               if ((eth_type != htons(ETH_P_IP) &&
+                    eth_type != htons(ETH_P_IPV6)) ||
+                   flow_key->ip.proto != IPPROTO_UDP)
                        return -EINVAL;
 
-               return validate_tp_port(flow_key, eth_type);
+               break;
 
        case OVS_KEY_ATTR_MPLS:
                if (!eth_p_mpls(eth_type))
@@ -1907,15 +2016,45 @@ static int validate_set(const struct nlattr *a,
                break;
 
        case OVS_KEY_ATTR_SCTP:
-               if (flow_key->ip.proto != IPPROTO_SCTP)
+               if ((eth_type != htons(ETH_P_IP) &&
+                    eth_type != htons(ETH_P_IPV6)) ||
+                   flow_key->ip.proto != IPPROTO_SCTP)
                        return -EINVAL;
 
-               return validate_tp_port(flow_key, eth_type);
+               break;
 
        default:
                return -EINVAL;
        }
 
+       /* Convert non-masked non-tunnel set actions to masked set actions. */
+       if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
+               int start, len = key_len * 2;
+               struct nlattr *at;
+
+               *skip_copy = true;
+
+               start = add_nested_action_start(sfa,
+                                               OVS_ACTION_ATTR_SET_TO_MASKED,
+                                               log);
+               if (start < 0)
+                       return start;
+
+               at = __add_action(sfa, key_type, NULL, len, log);
+               if (IS_ERR(at))
+                       return PTR_ERR(at);
+
+               memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
+               memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
+               /* Clear non-writeable bits from otherwise writeable fields. */
+               if (key_type == OVS_KEY_ATTR_IPV6) {
+                       struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
+
+                       mask->ipv6_label &= htonl(0x000FFFFF);
+               }
+               add_nested_action_end(*sfa, start);
+       }
+
        return 0;
 }
 
@@ -1977,6 +2116,7 @@ static int __ovs_nla_copy_actions(const struct nlattr *attr,
                        [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
                        [OVS_ACTION_ATTR_POP_VLAN] = 0,
                        [OVS_ACTION_ATTR_SET] = (u32)-1,
+                       [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
                        [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
                        [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash)
                };
@@ -2003,7 +2143,6 @@ static int __ovs_nla_copy_actions(const struct nlattr *attr,
                case OVS_ACTION_ATTR_OUTPUT:
                        if (nla_get_u32(a) >= DP_MAX_PORTS)
                                return -EINVAL;
-
                        break;
 
                case OVS_ACTION_ATTR_HASH: {
@@ -2040,7 +2179,6 @@ static int __ovs_nla_copy_actions(const struct nlattr *attr,
 
                        if (!eth_p_mpls(mpls->mpls_ethertype))
                                return -EINVAL;
-
                        /* Prohibit push MPLS other than to a white list
                         * for packets that have a known tag order.
                         */
@@ -2074,7 +2212,14 @@ static int __ovs_nla_copy_actions(const struct nlattr *attr,
 
                case OVS_ACTION_ATTR_SET:
                        err = validate_set(a, key, sfa,
-                                          &skip_copy, eth_type, log);
+                                          &skip_copy, eth_type, false, log);
+                       if (err)
+                               return err;
+                       break;
+
+               case OVS_ACTION_ATTR_SET_MASKED:
+                       err = validate_set(a, key, sfa,
+                                          &skip_copy, eth_type, true, log);
                        if (err)
                                return err;
                        break;
@@ -2104,6 +2249,7 @@ static int __ovs_nla_copy_actions(const struct nlattr *attr,
        return 0;
 }
 
+/* 'key' must be the masked key. */
 int ovs_nla_copy_actions(const struct nlattr *attr,
                         const struct sw_flow_key *key,
                         struct sw_flow_actions **sfa, bool log)
@@ -2117,7 +2263,7 @@ int ovs_nla_copy_actions(const struct nlattr *attr,
        err = __ovs_nla_copy_actions(attr, key, 0, sfa, key->eth.type,
                                     key->eth.tci, log);
        if (err)
-               kfree(*sfa);
+               ovs_nla_free_flow_actions(*sfa);
 
        return err;
 }
@@ -2167,15 +2313,16 @@ static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
 
        switch (key_type) {
        case OVS_KEY_ATTR_TUNNEL_INFO: {
-               struct ovs_tunnel_info *tun_info = nla_data(ovs_key);
+               struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
+               struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
 
                start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
                if (!start)
                        return -EMSGSIZE;
 
-               err = ipv4_tun_to_nlattr(skb, &tun_info->tunnel,
+               err = ipv4_tun_to_nlattr(skb, &tun_info->key,
                                         tun_info->options_len ?
-                                               tun_info->options : NULL,
+                                            ip_tunnel_info_opts(tun_info) : NULL,
                                         tun_info->options_len);
                if (err)
                        return err;
@@ -2191,6 +2338,21 @@ static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
        return 0;
 }
 
+static int masked_set_action_to_set_action_attr(const struct nlattr *a,
+                                               struct sk_buff *skb)
+{
+       const struct nlattr *ovs_key = nla_data(a);
+       size_t key_len = nla_len(ovs_key) / 2;
+
+       /* Revert the conversion we did from a non-masked set action to
+        * masked set action.
+        */
+       if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a) - key_len, ovs_key))
+               return -EMSGSIZE;
+
+       return 0;
+}
+
 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
 {
        const struct nlattr *a;
@@ -2206,6 +2368,12 @@ int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
                                return err;
                        break;
 
+               case OVS_ACTION_ATTR_SET_TO_MASKED:
+                       err = masked_set_action_to_set_action_attr(a, skb);
+                       if (err)
+                               return err;
+                       break;
+
                case OVS_ACTION_ATTR_SAMPLE:
                        err = sample_action_to_attr(a, skb);
                        if (err)