Commit | Line | Data |
---|---|---|
aa310701 PS |
1 | /* |
2 | * Copyright (c) 2007-2013 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 | ||
aa310701 PS |
19 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
20 | ||
21 | #include <linux/if.h> | |
22 | #include <linux/skbuff.h> | |
23 | #include <linux/ip.h> | |
24 | #include <linux/if_tunnel.h> | |
25 | #include <linux/if_vlan.h> | |
26 | #include <linux/in.h> | |
aa310701 PS |
27 | #include <linux/in_route.h> |
28 | #include <linux/inetdevice.h> | |
29 | #include <linux/jhash.h> | |
30 | #include <linux/list.h> | |
31 | #include <linux/kernel.h> | |
32 | #include <linux/workqueue.h> | |
33 | #include <linux/rculist.h> | |
34 | #include <net/route.h> | |
35 | #include <net/xfrm.h> | |
36 | ||
37 | #include <net/icmp.h> | |
38 | #include <net/ip.h> | |
39 | #include <net/ip_tunnels.h> | |
40 | #include <net/gre.h> | |
41 | #include <net/net_namespace.h> | |
42 | #include <net/netns/generic.h> | |
43 | #include <net/protocol.h> | |
44 | ||
45 | #include "datapath.h" | |
46 | #include "vport.h" | |
47 | ||
48 | /* Returns the least-significant 32 bits of a __be64. */ | |
49 | static __be32 be64_get_low32(__be64 x) | |
50 | { | |
51 | #ifdef __BIG_ENDIAN | |
52 | return (__force __be32)x; | |
53 | #else | |
54 | return (__force __be32)((__force u64)x >> 32); | |
55 | #endif | |
56 | } | |
57 | ||
58 | static __be16 filter_tnl_flags(__be16 flags) | |
59 | { | |
60 | return flags & (TUNNEL_CSUM | TUNNEL_KEY); | |
61 | } | |
62 | ||
63 | static struct sk_buff *__build_header(struct sk_buff *skb, | |
64 | int tunnel_hlen) | |
65 | { | |
66 | const struct ovs_key_ipv4_tunnel *tun_key = OVS_CB(skb)->tun_key; | |
67 | struct tnl_ptk_info tpi; | |
68 | ||
69 | skb = gre_handle_offloads(skb, !!(tun_key->tun_flags & TUNNEL_CSUM)); | |
70 | if (IS_ERR(skb)) | |
71 | return NULL; | |
72 | ||
73 | tpi.flags = filter_tnl_flags(tun_key->tun_flags); | |
74 | tpi.proto = htons(ETH_P_TEB); | |
75 | tpi.key = be64_get_low32(tun_key->tun_id); | |
76 | tpi.seq = 0; | |
77 | gre_build_header(skb, &tpi, tunnel_hlen); | |
78 | ||
79 | return skb; | |
80 | } | |
81 | ||
82 | static __be64 key_to_tunnel_id(__be32 key, __be32 seq) | |
83 | { | |
84 | #ifdef __BIG_ENDIAN | |
85 | return (__force __be64)((__force u64)seq << 32 | (__force u32)key); | |
86 | #else | |
87 | return (__force __be64)((__force u64)key << 32 | (__force u32)seq); | |
88 | #endif | |
89 | } | |
90 | ||
91 | /* Called with rcu_read_lock and BH disabled. */ | |
92 | static int gre_rcv(struct sk_buff *skb, | |
93 | const struct tnl_ptk_info *tpi) | |
94 | { | |
95 | struct ovs_key_ipv4_tunnel tun_key; | |
96 | struct ovs_net *ovs_net; | |
97 | struct vport *vport; | |
98 | __be64 key; | |
99 | ||
100 | ovs_net = net_generic(dev_net(skb->dev), ovs_net_id); | |
101 | vport = rcu_dereference(ovs_net->vport_net.gre_vport); | |
102 | if (unlikely(!vport)) | |
103 | return PACKET_REJECT; | |
104 | ||
105 | key = key_to_tunnel_id(tpi->key, tpi->seq); | |
106 | ovs_flow_tun_key_init(&tun_key, ip_hdr(skb), key, | |
107 | filter_tnl_flags(tpi->flags)); | |
108 | ||
109 | ovs_vport_receive(vport, skb, &tun_key); | |
110 | return PACKET_RCVD; | |
111 | } | |
112 | ||
113 | static int gre_tnl_send(struct vport *vport, struct sk_buff *skb) | |
114 | { | |
115 | struct net *net = ovs_dp_get_net(vport->dp); | |
116 | struct flowi4 fl; | |
117 | struct rtable *rt; | |
118 | int min_headroom; | |
119 | int tunnel_hlen; | |
120 | __be16 df; | |
121 | int err; | |
122 | ||
123 | if (unlikely(!OVS_CB(skb)->tun_key)) { | |
124 | err = -EINVAL; | |
125 | goto error; | |
126 | } | |
127 | ||
128 | /* Route lookup */ | |
129 | memset(&fl, 0, sizeof(fl)); | |
130 | fl.daddr = OVS_CB(skb)->tun_key->ipv4_dst; | |
131 | fl.saddr = OVS_CB(skb)->tun_key->ipv4_src; | |
132 | fl.flowi4_tos = RT_TOS(OVS_CB(skb)->tun_key->ipv4_tos); | |
133 | fl.flowi4_mark = skb->mark; | |
134 | fl.flowi4_proto = IPPROTO_GRE; | |
135 | ||
136 | rt = ip_route_output_key(net, &fl); | |
137 | if (IS_ERR(rt)) | |
138 | return PTR_ERR(rt); | |
139 | ||
140 | tunnel_hlen = ip_gre_calc_hlen(OVS_CB(skb)->tun_key->tun_flags); | |
141 | ||
142 | min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len | |
143 | + tunnel_hlen + sizeof(struct iphdr) | |
144 | + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0); | |
145 | if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) { | |
146 | int head_delta = SKB_DATA_ALIGN(min_headroom - | |
147 | skb_headroom(skb) + | |
148 | 16); | |
149 | err = pskb_expand_head(skb, max_t(int, head_delta, 0), | |
150 | 0, GFP_ATOMIC); | |
151 | if (unlikely(err)) | |
152 | goto err_free_rt; | |
153 | } | |
154 | ||
155 | if (vlan_tx_tag_present(skb)) { | |
156 | if (unlikely(!__vlan_put_tag(skb, | |
157 | skb->vlan_proto, | |
158 | vlan_tx_tag_get(skb)))) { | |
159 | err = -ENOMEM; | |
160 | goto err_free_rt; | |
161 | } | |
162 | skb->vlan_tci = 0; | |
163 | } | |
164 | ||
165 | /* Push Tunnel header. */ | |
166 | skb = __build_header(skb, tunnel_hlen); | |
167 | if (unlikely(!skb)) { | |
168 | err = 0; | |
169 | goto err_free_rt; | |
170 | } | |
171 | ||
172 | df = OVS_CB(skb)->tun_key->tun_flags & TUNNEL_DONT_FRAGMENT ? | |
173 | htons(IP_DF) : 0; | |
174 | ||
60ff7467 | 175 | skb->ignore_df = 1; |
aa310701 | 176 | |
aad88724 | 177 | return iptunnel_xmit(skb->sk, rt, skb, fl.saddr, |
aa310701 PS |
178 | OVS_CB(skb)->tun_key->ipv4_dst, IPPROTO_GRE, |
179 | OVS_CB(skb)->tun_key->ipv4_tos, | |
963a88b3 | 180 | OVS_CB(skb)->tun_key->ipv4_ttl, df, false); |
aa310701 PS |
181 | err_free_rt: |
182 | ip_rt_put(rt); | |
183 | error: | |
184 | return err; | |
185 | } | |
186 | ||
187 | static struct gre_cisco_protocol gre_protocol = { | |
188 | .handler = gre_rcv, | |
189 | .priority = 1, | |
190 | }; | |
191 | ||
192 | static int gre_ports; | |
193 | static int gre_init(void) | |
194 | { | |
195 | int err; | |
196 | ||
197 | gre_ports++; | |
198 | if (gre_ports > 1) | |
199 | return 0; | |
200 | ||
201 | err = gre_cisco_register(&gre_protocol); | |
202 | if (err) | |
203 | pr_warn("cannot register gre protocol handler\n"); | |
204 | ||
205 | return err; | |
206 | } | |
207 | ||
208 | static void gre_exit(void) | |
209 | { | |
210 | gre_ports--; | |
211 | if (gre_ports > 0) | |
212 | return; | |
213 | ||
214 | gre_cisco_unregister(&gre_protocol); | |
215 | } | |
216 | ||
217 | static const char *gre_get_name(const struct vport *vport) | |
218 | { | |
219 | return vport_priv(vport); | |
220 | } | |
221 | ||
222 | static struct vport *gre_create(const struct vport_parms *parms) | |
223 | { | |
224 | struct net *net = ovs_dp_get_net(parms->dp); | |
225 | struct ovs_net *ovs_net; | |
226 | struct vport *vport; | |
227 | int err; | |
228 | ||
229 | err = gre_init(); | |
230 | if (err) | |
231 | return ERR_PTR(err); | |
232 | ||
233 | ovs_net = net_generic(net, ovs_net_id); | |
234 | if (ovsl_dereference(ovs_net->vport_net.gre_vport)) { | |
235 | vport = ERR_PTR(-EEXIST); | |
236 | goto error; | |
237 | } | |
238 | ||
239 | vport = ovs_vport_alloc(IFNAMSIZ, &ovs_gre_vport_ops, parms); | |
240 | if (IS_ERR(vport)) | |
241 | goto error; | |
242 | ||
243 | strncpy(vport_priv(vport), parms->name, IFNAMSIZ); | |
244 | rcu_assign_pointer(ovs_net->vport_net.gre_vport, vport); | |
245 | return vport; | |
246 | ||
247 | error: | |
248 | gre_exit(); | |
249 | return vport; | |
250 | } | |
251 | ||
252 | static void gre_tnl_destroy(struct vport *vport) | |
253 | { | |
254 | struct net *net = ovs_dp_get_net(vport->dp); | |
255 | struct ovs_net *ovs_net; | |
256 | ||
257 | ovs_net = net_generic(net, ovs_net_id); | |
258 | ||
944df8ae | 259 | RCU_INIT_POINTER(ovs_net->vport_net.gre_vport, NULL); |
aa310701 PS |
260 | ovs_vport_deferred_free(vport); |
261 | gre_exit(); | |
262 | } | |
263 | ||
264 | const struct vport_ops ovs_gre_vport_ops = { | |
265 | .type = OVS_VPORT_TYPE_GRE, | |
266 | .create = gre_create, | |
267 | .destroy = gre_tnl_destroy, | |
268 | .get_name = gre_get_name, | |
269 | .send = gre_tnl_send, | |
270 | }; |