]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blobdiff - net/openvswitch/flow_netlink.c
openvswitch: Add support for Geneve tunneling.
[mirror_ubuntu-bionic-kernel.git] / net / openvswitch / flow_netlink.c
index d757848da89ca734ac95cef806c2dd314f881bf6..368f23307911cf14d4efc3082f379b58d4e4ae3d 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2007-2013 Nicira, Inc.
+ * Copyright (c) 2007-2014 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
@@ -42,6 +42,7 @@
 #include <linux/icmp.h>
 #include <linux/icmpv6.h>
 #include <linux/rculist.h>
+#include <net/geneve.h>
 #include <net/ip.h>
 #include <net/ipv6.h>
 #include <net/ndisc.h>
@@ -88,18 +89,20 @@ static void update_range__(struct sw_flow_match *match,
                }                                                           \
        } while (0)
 
-#define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
-       do { \
-               update_range__(match, offsetof(struct sw_flow_key, field),  \
-                               len, is_mask);                              \
-               if (is_mask) {                                              \
-                       if ((match)->mask)                                  \
-                               memcpy(&(match)->mask->key.field, value_p, len);\
-               } else {                                                    \
-                       memcpy(&(match)->key->field, value_p, len);         \
-               }                                                           \
+#define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)            \
+       do {                                                                \
+               update_range__(match, offset, len, is_mask);                \
+               if (is_mask)                                                \
+                       memcpy((u8 *)&(match)->mask->key + offset, value_p, \
+                              len);                                        \
+               else                                                        \
+                       memcpy((u8 *)(match)->key + offset, value_p, len);  \
        } while (0)
 
+#define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)                      \
+       SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
+                                 value_p, len, is_mask)
+
 static u16 range_n_bytes(const struct sw_flow_key_range *range)
 {
        return range->end - range->start;
@@ -251,6 +254,8 @@ static const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
        [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
        [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
        [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
+       [OVS_KEY_ATTR_RECIRC_ID] = sizeof(u32),
+       [OVS_KEY_ATTR_DP_HASH] = sizeof(u32),
        [OVS_KEY_ATTR_TUNNEL] = -1,
 };
 
@@ -333,6 +338,7 @@ static int ipv4_tun_from_nlattr(const struct nlattr *attr,
        int rem;
        bool ttl = false;
        __be16 tun_flags = 0;
+       unsigned long opt_key_offset;
 
        nla_for_each_nested(a, attr, rem) {
                int type = nla_type(a);
@@ -344,6 +350,8 @@ static int ipv4_tun_from_nlattr(const struct nlattr *attr,
                        [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
                        [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
                        [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
+                       [OVS_TUNNEL_KEY_ATTR_OAM] = 0,
+                       [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS] = -1,
                };
 
                if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
@@ -352,7 +360,8 @@ static int ipv4_tun_from_nlattr(const struct nlattr *attr,
                        return -EINVAL;
                }
 
-               if (ovs_tunnel_key_lens[type] != nla_len(a)) {
+               if (ovs_tunnel_key_lens[type] != nla_len(a) &&
+                   ovs_tunnel_key_lens[type] != -1) {
                        OVS_NLERR("IPv4 tunnel attribute type has unexpected "
                                  " length (type=%d, length=%d, expected=%d).\n",
                                  type, nla_len(a), ovs_tunnel_key_lens[type]);
@@ -388,7 +397,63 @@ static int ipv4_tun_from_nlattr(const struct nlattr *attr,
                case OVS_TUNNEL_KEY_ATTR_CSUM:
                        tun_flags |= TUNNEL_CSUM;
                        break;
+               case OVS_TUNNEL_KEY_ATTR_OAM:
+                       tun_flags |= TUNNEL_OAM;
+                       break;
+               case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
+                       tun_flags |= TUNNEL_OPTIONS_PRESENT;
+                       if (nla_len(a) > sizeof(match->key->tun_opts)) {
+                               OVS_NLERR("Geneve option length exceeds maximum size (len %d, max %zu).\n",
+                                         nla_len(a),
+                                         sizeof(match->key->tun_opts));
+                               return -EINVAL;
+                       }
+
+                       if (nla_len(a) % 4 != 0) {
+                               OVS_NLERR("Geneve option length is not a multiple of 4 (len %d).\n",
+                                         nla_len(a));
+                               return -EINVAL;
+                       }
+
+                       /* We need to record the length of the options passed
+                        * down, otherwise packets with the same format but
+                        * additional options will be silently matched.
+                        */
+                       if (!is_mask) {
+                               SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
+                                               false);
+                       } else {
+                               /* This is somewhat unusual because it looks at
+                                * both the key and mask while parsing the
+                                * attributes (and by extension assumes the key
+                                * is parsed first). Normally, we would verify
+                                * that each is the correct length and that the
+                                * attributes line up in the validate function.
+                                * However, that is difficult because this is
+                                * variable length and we won't have the
+                                * information later.
+                                */
+                               if (match->key->tun_opts_len != nla_len(a)) {
+                                       OVS_NLERR("Geneve option key length (%d) is different from mask length (%d).",
+                                                 match->key->tun_opts_len,
+                                                 nla_len(a));
+                                       return -EINVAL;
+                               }
+
+                               SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff,
+                                               true);
+                       }
+
+                       opt_key_offset = (unsigned long)GENEVE_OPTS(
+                                         (struct sw_flow_key *)0,
+                                         nla_len(a));
+                       SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset,
+                                                 nla_data(a), nla_len(a),
+                                                 is_mask);
+                       break;
                default:
+                       OVS_NLERR("Unknown IPv4 tunnel attribute (%d).\n",
+                                 type);
                        return -EINVAL;
                }
        }
@@ -415,45 +480,80 @@ static int ipv4_tun_from_nlattr(const struct nlattr *attr,
        return 0;
 }
 
-static int ipv4_tun_to_nlattr(struct sk_buff *skb,
-                             const struct ovs_key_ipv4_tunnel *tun_key,
-                             const struct ovs_key_ipv4_tunnel *output)
+static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
+                               const struct ovs_key_ipv4_tunnel *output,
+                               const struct geneve_opt *tun_opts,
+                               int swkey_tun_opts_len)
 {
-       struct nlattr *nla;
-
-       nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
-       if (!nla)
-               return -EMSGSIZE;
-
        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))
+           nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
                return -EMSGSIZE;
        if (output->ipv4_dst &&
-               nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
+           nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
                return -EMSGSIZE;
        if (output->ipv4_tos &&
-               nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
+           nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
                return -EMSGSIZE;
        if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
                return -EMSGSIZE;
        if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
-               nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
+           nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
                return -EMSGSIZE;
        if ((output->tun_flags & TUNNEL_CSUM) &&
-               nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
+           nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
+               return -EMSGSIZE;
+       if ((output->tun_flags & TUNNEL_OAM) &&
+           nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
+               return -EMSGSIZE;
+       if (tun_opts &&
+           nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
+                   swkey_tun_opts_len, tun_opts))
                return -EMSGSIZE;
 
-       nla_nest_end(skb, nla);
        return 0;
 }
 
 
+static int ipv4_tun_to_nlattr(struct sk_buff *skb,
+                             const struct ovs_key_ipv4_tunnel *output,
+                             const struct geneve_opt *tun_opts,
+                             int swkey_tun_opts_len)
+{
+       struct nlattr *nla;
+       int err;
+
+       nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
+       if (!nla)
+               return -EMSGSIZE;
+
+       err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
+       if (err)
+               return err;
+
+       nla_nest_end(skb, nla);
+       return 0;
+}
+
 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
                                 const struct nlattr **a, bool is_mask)
 {
+       if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
+               u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
+
+               SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
+               *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
+       }
+
+       if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
+               u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
+
+               SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
+               *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
+       }
+
        if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
                SW_FLOW_KEY_PUT(match, phy.priority,
                          nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
@@ -836,7 +936,7 @@ int ovs_nla_get_match(struct sw_flow_match *match,
 
 /**
  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
- * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
+ * @key: Receives extracted in_port, priority, tun_key and skb_mark.
  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
  * sequence.
  *
@@ -846,32 +946,24 @@ int ovs_nla_get_match(struct sw_flow_match *match,
  * extracted from the packet itself.
  */
 
-int ovs_nla_get_flow_metadata(struct sw_flow *flow,
-                             const struct nlattr *attr)
+int ovs_nla_get_flow_metadata(const struct nlattr *attr,
+                             struct sw_flow_key *key)
 {
-       struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
        const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
+       struct sw_flow_match match;
        u64 attrs = 0;
        int err;
-       struct sw_flow_match match;
-
-       flow->key.phy.in_port = DP_MAX_PORTS;
-       flow->key.phy.priority = 0;
-       flow->key.phy.skb_mark = 0;
-       memset(tun_key, 0, sizeof(flow->key.tun_key));
 
        err = parse_flow_nlattrs(attr, a, &attrs);
        if (err)
                return -EINVAL;
 
        memset(&match, 0, sizeof(match));
-       match.key = &flow->key;
+       match.key = key;
 
-       err = metadata_from_nlattrs(&match, &attrs, a, false);
-       if (err)
-               return err;
+       key->phy.in_port = DP_MAX_PORTS;
 
-       return 0;
+       return metadata_from_nlattrs(&match, &attrs, a, false);
 }
 
 int ovs_nla_put_flow(const struct sw_flow_key *swkey,
@@ -881,13 +973,26 @@ int ovs_nla_put_flow(const struct sw_flow_key *swkey,
        struct nlattr *nla, *encap;
        bool is_mask = (swkey != output);
 
-       if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
+       if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
+               goto nla_put_failure;
+
+       if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
                goto nla_put_failure;
 
-       if ((swkey->tun_key.ipv4_dst || is_mask) &&
-           ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
+       if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
                goto nla_put_failure;
 
+       if ((swkey->tun_key.ipv4_dst || is_mask)) {
+               const struct geneve_opt *opts = NULL;
+
+               if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
+                       opts = GENEVE_OPTS(output, swkey->tun_opts_len);
+
+               if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
+                                      swkey->tun_opts_len))
+                       goto nla_put_failure;
+       }
+
        if (swkey->phy.in_port == DP_MAX_PORTS) {
                if (is_mask && (output->phy.in_port == 0xffff))
                        if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
@@ -1127,13 +1232,14 @@ out:
        return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
 }
 
-static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
+static struct nlattr *__add_action(struct sw_flow_actions **sfa,
+                                  int attrtype, void *data, int len)
 {
        struct nlattr *a;
 
        a = reserve_sfa_size(sfa, nla_attr_size(len));
        if (IS_ERR(a))
-               return PTR_ERR(a);
+               return a;
 
        a->nla_type = attrtype;
        a->nla_len = nla_attr_size(len);
@@ -1142,6 +1248,18 @@ static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, in
                memcpy(nla_data(a), data, len);
        memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
 
+       return a;
+}
+
+static int add_action(struct sw_flow_actions **sfa, int attrtype,
+                     void *data, int len)
+{
+       struct nlattr *a;
+
+       a = __add_action(sfa, attrtype, data, len);
+       if (IS_ERR(a))
+               return PTR_ERR(a);
+
        return 0;
 }
 
@@ -1247,6 +1365,8 @@ 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 nlattr *a;
        int err, start;
 
        ovs_match_init(&match, &key, NULL);
@@ -1254,12 +1374,56 @@ static int validate_and_copy_set_tun(const struct nlattr *attr,
        if (err)
                return err;
 
+       if (key.tun_opts_len) {
+               struct geneve_opt *option = GENEVE_OPTS(&key,
+                                                       key.tun_opts_len);
+               int opts_len = key.tun_opts_len;
+               bool crit_opt = false;
+
+               while (opts_len > 0) {
+                       int len;
+
+                       if (opts_len < sizeof(*option))
+                               return -EINVAL;
+
+                       len = sizeof(*option) + option->length * 4;
+                       if (len > opts_len)
+                               return -EINVAL;
+
+                       crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
+
+                       option = (struct geneve_opt *)((u8 *)option + len);
+                       opts_len -= len;
+               };
+
+               key.tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
+       };
+
        start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
        if (start < 0)
                return start;
 
-       err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &match.key->tun_key,
-                       sizeof(match.key->tun_key));
+       a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
+                        sizeof(*tun_info) + key.tun_opts_len);
+       if (IS_ERR(a))
+               return PTR_ERR(a);
+
+       tun_info = nla_data(a);
+       tun_info->tunnel = key.tun_key;
+       tun_info->options_len = key.tun_opts_len;
+
+       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), GENEVE_OPTS(&key, key.tun_opts_len),
+                      key.tun_opts_len);
+               tun_info->options = (struct geneve_opt *)(tun_info + 1);
+       } else {
+               tun_info->options = NULL;
+       }
+
        add_nested_action_end(*sfa, start);
 
        return err;
@@ -1409,11 +1573,13 @@ int ovs_nla_copy_actions(const struct nlattr *attr,
                /* Expected argument lengths, (u32)-1 for variable length. */
                static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
                        [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
+                       [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
                        [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
                        [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_SAMPLE] = (u32)-1
+                       [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
+                       [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash)
                };
                const struct ovs_action_push_vlan *vlan;
                int type = nla_type(a);
@@ -1440,6 +1606,18 @@ int ovs_nla_copy_actions(const struct nlattr *attr,
                                return -EINVAL;
                        break;
 
+               case OVS_ACTION_ATTR_HASH: {
+                       const struct ovs_action_hash *act_hash = nla_data(a);
+
+                       switch (act_hash->hash_alg) {
+                       case OVS_HASH_ALG_L4:
+                               break;
+                       default:
+                               return  -EINVAL;
+                       }
+
+                       break;
+               }
 
                case OVS_ACTION_ATTR_POP_VLAN:
                        break;
@@ -1452,6 +1630,9 @@ int ovs_nla_copy_actions(const struct nlattr *attr,
                                return -EINVAL;
                        break;
 
+               case OVS_ACTION_ATTR_RECIRC:
+                       break;
+
                case OVS_ACTION_ATTR_SET:
                        err = validate_set(a, key, sfa, &skip_copy);
                        if (err)
@@ -1525,17 +1706,22 @@ static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
        int err;
 
        switch (key_type) {
-       case OVS_KEY_ATTR_IPV4_TUNNEL:
+       case OVS_KEY_ATTR_TUNNEL_INFO: {
+               struct ovs_tunnel_info *tun_info = nla_data(ovs_key);
+
                start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
                if (!start)
                        return -EMSGSIZE;
 
-               err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key),
-                                            nla_data(ovs_key));
+               err = ipv4_tun_to_nlattr(skb, &tun_info->tunnel,
+                                        tun_info->options_len ?
+                                               tun_info->options : NULL,
+                                        tun_info->options_len);
                if (err)
                        return err;
                nla_nest_end(skb, start);
                break;
+       }
        default:
                if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
                        return -EMSGSIZE;