blob: 6d39766e7828c4351a004fa8ce7d839348c3681d [file] [log] [blame]
Pravin B Shelar58264842013-08-19 11:23:34 -07001/*
Pravin B Shelar8c8b1b82014-09-15 19:28:44 -07002 * Copyright (c) 2014 Nicira, Inc.
Pravin B Shelar58264842013-08-19 11:23:34 -07003 * Copyright (c) 2013 Cisco Systems, Inc.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of version 2 of the GNU General Public
7 * License as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
17 * 02110-1301, USA
18 */
19
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22#include <linux/in.h>
23#include <linux/ip.h>
24#include <linux/net.h>
25#include <linux/rculist.h>
26#include <linux/udp.h>
Thomas Graf62b9c8d2014-10-22 17:29:06 +020027#include <linux/module.h>
Pravin B Shelar58264842013-08-19 11:23:34 -070028
29#include <net/icmp.h>
30#include <net/ip.h>
31#include <net/udp.h>
32#include <net/ip_tunnels.h>
Pravin B Shelar58264842013-08-19 11:23:34 -070033#include <net/rtnetlink.h>
34#include <net/route.h>
35#include <net/dsfield.h>
36#include <net/inet_ecn.h>
37#include <net/net_namespace.h>
38#include <net/netns/generic.h>
39#include <net/vxlan.h>
40
41#include "datapath.h"
42#include "vport.h"
Thomas Graf1dd144c2015-01-15 03:53:59 +010043#include "vport-vxlan.h"
Pravin B Shelar58264842013-08-19 11:23:34 -070044
45/**
46 * struct vxlan_port - Keeps track of open UDP ports
47 * @vs: vxlan_sock created for the port.
48 * @name: vport name.
49 */
50struct vxlan_port {
51 struct vxlan_sock *vs;
52 char name[IFNAMSIZ];
Thomas Graf1dd144c2015-01-15 03:53:59 +010053 u32 exts; /* VXLAN_F_* in <net/vxlan.h> */
Pravin B Shelar58264842013-08-19 11:23:34 -070054};
55
Thomas Graf62b9c8d2014-10-22 17:29:06 +020056static struct vport_ops ovs_vxlan_vport_ops;
57
Pravin B Shelar58264842013-08-19 11:23:34 -070058static inline struct vxlan_port *vxlan_vport(const struct vport *vport)
59{
60 return vport_priv(vport);
61}
62
63/* Called with rcu_read_lock and BH disabled. */
Thomas Graf35114942015-01-15 03:53:55 +010064static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
65 struct vxlan_metadata *md)
Pravin B Shelar58264842013-08-19 11:23:34 -070066{
Jesse Grossf0b128c2014-10-03 15:35:31 -070067 struct ovs_tunnel_info tun_info;
Thomas Graf1dd144c2015-01-15 03:53:59 +010068 struct vxlan_port *vxlan_port;
Pravin B Shelar58264842013-08-19 11:23:34 -070069 struct vport *vport = vs->data;
70 struct iphdr *iph;
Thomas Graf1dd144c2015-01-15 03:53:59 +010071 struct ovs_vxlan_opts opts = {
72 .gbp = md->gbp,
73 };
Pravin B Shelar58264842013-08-19 11:23:34 -070074 __be64 key;
Thomas Graf1dd144c2015-01-15 03:53:59 +010075 __be16 flags;
76
Jesse Grossb8693872015-01-28 16:32:46 -080077 flags = TUNNEL_KEY | (udp_hdr(skb)->check != 0 ? TUNNEL_CSUM : 0);
Thomas Graf1dd144c2015-01-15 03:53:59 +010078 vxlan_port = vxlan_vport(vport);
Thomas Graffd3137c2015-02-09 16:56:37 +010079 if (vxlan_port->exts & VXLAN_F_GBP && md->gbp)
Thomas Graf1dd144c2015-01-15 03:53:59 +010080 flags |= TUNNEL_VXLAN_OPT;
Pravin B Shelar58264842013-08-19 11:23:34 -070081
82 /* Save outer tunnel values */
83 iph = ip_hdr(skb);
Thomas Graf35114942015-01-15 03:53:55 +010084 key = cpu_to_be64(ntohl(md->vni) >> 8);
Wenyu Zhang8f0aad62014-11-06 06:51:24 -080085 ovs_flow_tun_info_init(&tun_info, iph,
86 udp_hdr(skb)->source, udp_hdr(skb)->dest,
Thomas Graf1dd144c2015-01-15 03:53:59 +010087 key, flags, &opts, sizeof(opts));
Pravin B Shelar58264842013-08-19 11:23:34 -070088
Jesse Grossf0b128c2014-10-03 15:35:31 -070089 ovs_vport_receive(vport, skb, &tun_info);
Pravin B Shelar58264842013-08-19 11:23:34 -070090}
91
92static int vxlan_get_options(const struct vport *vport, struct sk_buff *skb)
93{
94 struct vxlan_port *vxlan_port = vxlan_vport(vport);
95 __be16 dst_port = inet_sk(vxlan_port->vs->sock->sk)->inet_sport;
96
97 if (nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT, ntohs(dst_port)))
98 return -EMSGSIZE;
Thomas Graf1dd144c2015-01-15 03:53:59 +010099
100 if (vxlan_port->exts) {
101 struct nlattr *exts;
102
103 exts = nla_nest_start(skb, OVS_TUNNEL_ATTR_EXTENSION);
104 if (!exts)
105 return -EMSGSIZE;
106
107 if (vxlan_port->exts & VXLAN_F_GBP &&
108 nla_put_flag(skb, OVS_VXLAN_EXT_GBP))
109 return -EMSGSIZE;
110
111 nla_nest_end(skb, exts);
112 }
113
Pravin B Shelar58264842013-08-19 11:23:34 -0700114 return 0;
115}
116
117static void vxlan_tnl_destroy(struct vport *vport)
118{
119 struct vxlan_port *vxlan_port = vxlan_vport(vport);
120
121 vxlan_sock_release(vxlan_port->vs);
122
123 ovs_vport_deferred_free(vport);
124}
125
Thomas Graf1dd144c2015-01-15 03:53:59 +0100126static const struct nla_policy exts_policy[OVS_VXLAN_EXT_MAX+1] = {
127 [OVS_VXLAN_EXT_GBP] = { .type = NLA_FLAG, },
128};
129
130static int vxlan_configure_exts(struct vport *vport, struct nlattr *attr)
131{
132 struct nlattr *exts[OVS_VXLAN_EXT_MAX+1];
133 struct vxlan_port *vxlan_port;
134 int err;
135
136 if (nla_len(attr) < sizeof(struct nlattr))
137 return -EINVAL;
138
139 err = nla_parse_nested(exts, OVS_VXLAN_EXT_MAX, attr, exts_policy);
140 if (err < 0)
141 return err;
142
143 vxlan_port = vxlan_vport(vport);
144
145 if (exts[OVS_VXLAN_EXT_GBP])
146 vxlan_port->exts |= VXLAN_F_GBP;
147
148 return 0;
149}
150
Pravin B Shelar58264842013-08-19 11:23:34 -0700151static struct vport *vxlan_tnl_create(const struct vport_parms *parms)
152{
153 struct net *net = ovs_dp_get_net(parms->dp);
154 struct nlattr *options = parms->options;
155 struct vxlan_port *vxlan_port;
156 struct vxlan_sock *vs;
157 struct vport *vport;
158 struct nlattr *a;
159 u16 dst_port;
160 int err;
161
162 if (!options) {
163 err = -EINVAL;
164 goto error;
165 }
166 a = nla_find_nested(options, OVS_TUNNEL_ATTR_DST_PORT);
167 if (a && nla_len(a) == sizeof(u16)) {
168 dst_port = nla_get_u16(a);
169 } else {
170 /* Require destination port from userspace. */
171 err = -EINVAL;
172 goto error;
173 }
174
175 vport = ovs_vport_alloc(sizeof(struct vxlan_port),
176 &ovs_vxlan_vport_ops, parms);
177 if (IS_ERR(vport))
178 return vport;
179
180 vxlan_port = vxlan_vport(vport);
181 strncpy(vxlan_port->name, parms->name, IFNAMSIZ);
182
Thomas Graf1dd144c2015-01-15 03:53:59 +0100183 a = nla_find_nested(options, OVS_TUNNEL_ATTR_EXTENSION);
184 if (a) {
185 err = vxlan_configure_exts(vport, a);
186 if (err) {
187 ovs_vport_free(vport);
188 goto error;
189 }
190 }
191
192 vs = vxlan_sock_add(net, htons(dst_port), vxlan_rcv, vport, true,
193 vxlan_port->exts);
Pravin B Shelar58264842013-08-19 11:23:34 -0700194 if (IS_ERR(vs)) {
195 ovs_vport_free(vport);
196 return (void *)vs;
197 }
198 vxlan_port->vs = vs;
199
200 return vport;
201
202error:
203 return ERR_PTR(err);
204}
205
Thomas Graf1dd144c2015-01-15 03:53:59 +0100206static int vxlan_ext_gbp(struct sk_buff *skb)
207{
208 const struct ovs_tunnel_info *tun_info;
209 const struct ovs_vxlan_opts *opts;
210
211 tun_info = OVS_CB(skb)->egress_tun_info;
212 opts = tun_info->options;
213
214 if (tun_info->tunnel.tun_flags & TUNNEL_VXLAN_OPT &&
215 tun_info->options_len >= sizeof(*opts))
216 return opts->gbp;
217 else
218 return 0;
219}
220
Pravin B Shelar58264842013-08-19 11:23:34 -0700221static int vxlan_tnl_send(struct vport *vport, struct sk_buff *skb)
222{
223 struct net *net = ovs_dp_get_net(vport->dp);
224 struct vxlan_port *vxlan_port = vxlan_vport(vport);
David Miller79b16aa2015-04-05 22:19:09 -0400225 struct sock *sk = vxlan_port->vs->sock->sk;
226 __be16 dst_port = inet_sk(sk)->inet_sport;
Fan Du3f4c1d82015-01-14 13:10:35 +0800227 const struct ovs_key_ipv4_tunnel *tun_key;
Thomas Graf35114942015-01-15 03:53:55 +0100228 struct vxlan_metadata md = {0};
Pravin B Shelar58264842013-08-19 11:23:34 -0700229 struct rtable *rt;
230 struct flowi4 fl;
231 __be16 src_port;
Pravin B Shelar58264842013-08-19 11:23:34 -0700232 __be16 df;
233 int err;
Jesse Grossb8693872015-01-28 16:32:46 -0800234 u32 vxflags;
Pravin B Shelar58264842013-08-19 11:23:34 -0700235
Jesse Grossf0b128c2014-10-03 15:35:31 -0700236 if (unlikely(!OVS_CB(skb)->egress_tun_info)) {
Pravin B Shelar58264842013-08-19 11:23:34 -0700237 err = -EINVAL;
238 goto error;
239 }
240
Jesse Grossf0b128c2014-10-03 15:35:31 -0700241 tun_key = &OVS_CB(skb)->egress_tun_info->tunnel;
Fan Du3f4c1d82015-01-14 13:10:35 +0800242 rt = ovs_tunnel_route_lookup(net, tun_key, skb->mark, &fl, IPPROTO_UDP);
Pravin B Shelar58264842013-08-19 11:23:34 -0700243 if (IS_ERR(rt)) {
244 err = PTR_ERR(rt);
245 goto error;
246 }
247
Pravin B Shelar8c8b1b82014-09-15 19:28:44 -0700248 df = tun_key->tun_flags & TUNNEL_DONT_FRAGMENT ?
Pravin B Shelar58264842013-08-19 11:23:34 -0700249 htons(IP_DF) : 0;
250
WANG Cong60ff7462014-05-04 16:39:18 -0700251 skb->ignore_df = 1;
Pravin B Shelar58264842013-08-19 11:23:34 -0700252
Tom Herbert535fb8d2014-07-01 21:32:49 -0700253 src_port = udp_flow_src_port(net, skb, 0, 0, true);
Thomas Graf35114942015-01-15 03:53:55 +0100254 md.vni = htonl(be64_to_cpu(tun_key->tun_id) << 8);
Thomas Graf1dd144c2015-01-15 03:53:59 +0100255 md.gbp = vxlan_ext_gbp(skb);
Jesse Grossb8693872015-01-28 16:32:46 -0800256 vxflags = vxlan_port->exts |
257 (tun_key->tun_flags & TUNNEL_CSUM ? VXLAN_F_UDP_CSUM : 0);
Pravin B Shelar58264842013-08-19 11:23:34 -0700258
David Miller79b16aa2015-04-05 22:19:09 -0400259 err = vxlan_xmit_skb(rt, sk, skb, fl.saddr, tun_key->ipv4_dst,
Pravin B Shelar8c8b1b82014-09-15 19:28:44 -0700260 tun_key->ipv4_tos, tun_key->ipv4_ttl, df,
Pravin B Shelar58264842013-08-19 11:23:34 -0700261 src_port, dst_port,
Jesse Grossb8693872015-01-28 16:32:46 -0800262 &md, false, vxflags);
Pravin B Shelar58264842013-08-19 11:23:34 -0700263 if (err < 0)
264 ip_rt_put(rt);
Pravin B Shelar997e0682014-12-23 16:20:32 -0800265 return err;
Pravin B Shelar58264842013-08-19 11:23:34 -0700266error:
Pravin B Shelar997e0682014-12-23 16:20:32 -0800267 kfree_skb(skb);
Pravin B Shelar58264842013-08-19 11:23:34 -0700268 return err;
269}
270
Wenyu Zhang8f0aad62014-11-06 06:51:24 -0800271static int vxlan_get_egress_tun_info(struct vport *vport, struct sk_buff *skb,
272 struct ovs_tunnel_info *egress_tun_info)
273{
274 struct net *net = ovs_dp_get_net(vport->dp);
275 struct vxlan_port *vxlan_port = vxlan_vport(vport);
276 __be16 dst_port = inet_sk(vxlan_port->vs->sock->sk)->inet_sport;
277 __be16 src_port;
278 int port_min;
279 int port_max;
280
281 inet_get_local_port_range(net, &port_min, &port_max);
282 src_port = udp_flow_src_port(net, skb, 0, 0, true);
283
284 return ovs_tunnel_get_egress_info(egress_tun_info, net,
285 OVS_CB(skb)->egress_tun_info,
286 IPPROTO_UDP, skb->mark,
287 src_port, dst_port);
288}
289
Pravin B Shelar58264842013-08-19 11:23:34 -0700290static const char *vxlan_get_name(const struct vport *vport)
291{
292 struct vxlan_port *vxlan_port = vxlan_vport(vport);
293 return vxlan_port->name;
294}
295
Thomas Graf62b9c8d2014-10-22 17:29:06 +0200296static struct vport_ops ovs_vxlan_vport_ops = {
Pravin B Shelar58264842013-08-19 11:23:34 -0700297 .type = OVS_VPORT_TYPE_VXLAN,
298 .create = vxlan_tnl_create,
299 .destroy = vxlan_tnl_destroy,
300 .get_name = vxlan_get_name,
301 .get_options = vxlan_get_options,
302 .send = vxlan_tnl_send,
Wenyu Zhang8f0aad62014-11-06 06:51:24 -0800303 .get_egress_tun_info = vxlan_get_egress_tun_info,
Thomas Graf62b9c8d2014-10-22 17:29:06 +0200304 .owner = THIS_MODULE,
Pravin B Shelar58264842013-08-19 11:23:34 -0700305};
Thomas Graf62b9c8d2014-10-22 17:29:06 +0200306
307static int __init ovs_vxlan_tnl_init(void)
308{
309 return ovs_vport_ops_register(&ovs_vxlan_vport_ops);
310}
311
312static void __exit ovs_vxlan_tnl_exit(void)
313{
314 ovs_vport_ops_unregister(&ovs_vxlan_vport_ops);
315}
316
317module_init(ovs_vxlan_tnl_init);
318module_exit(ovs_vxlan_tnl_exit);
319
320MODULE_DESCRIPTION("OVS: VXLAN switching port");
321MODULE_LICENSE("GPL");
322MODULE_ALIAS("vport-type-4");