datapath/vport-lisp.c

   1 /*
   2  * Copyright (c) 2011 Nicira, Inc.
   3  * 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/version.h>
  23 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,26)
  24
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/net.h>
  28 #include <linux/rculist.h>
  29 #include <linux/udp.h>
  30
  31 #include <net/icmp.h>
  32 #include <net/ip.h>
  33 #include <net/udp.h>
  34
  35 #include "datapath.h"
  36 #include "tunnel.h"
  37 #include "vport.h"
  38
  39
  40 /*
  41  *  LISP encapsulation header:
  42  *
  43  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  44  *  |N|L|E|V|I|flags|            Nonce/Map-Version                  |
  45  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  46  *  |                 Instance ID/Locator Status Bits               |
  47  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  48  *
  49  */
  50
  51 /**
  52  * struct lisphdr - LISP header
  53  * @nonce_present: Flag indicating the presence of a 24 bit nonce value.
  54  * @locator_status_bits_present: Flag indicating the presence of Locator Status
  55  *                               Bits (LSB).
  56  * @solicit_echo_nonce: Flag indicating the use of the echo noncing mechanism.
  57  * @map_version_present: Flag indicating the use of mapping versioning.
  58  * @instance_id_present: Flag indicating the presence of a 24 bit Instance ID.
  59  * @reserved_flags: 3 bits reserved for future flags.
  60  * @nonce: 24 bit nonce value.
  61  * @map_version: 24 bit mapping version.
  62  * @locator_status_bits: Locator Status Bits: 32 bits when instance_id_present
  63  *                       is not set, 8 bits when it is.
  64  * @instance_id: 24 bit Instance ID
  65  */
  66 struct lisphdr {
  67 #ifdef __LITTLE_ENDIAN_BITFIELD
  68         __u8 reserved_flags:3;
  69         __u8 instance_id_present:1;
  70         __u8 map_version_present:1;
  71         __u8 solicit_echo_nonce:1;
  72         __u8 locator_status_bits_present:1;
  73         __u8 nonce_present:1;
  74 #else
  75         __u8 nonce_present:1;
  76         __u8 locator_status_bits_present:1;
  77         __u8 solicit_echo_nonce:1;
  78         __u8 map_version_present:1;
  79         __u8 instance_id_present:1;
  80         __u8 reserved_flags:3;
  81 #endif
  82         union {
  83                 __u8 nonce[3];
  84                 __u8 map_version[3];
  85         } u1;
  86         union {
  87                 __be32 locator_status_bits;
  88                 struct {
  89                         __u8 instance_id[3];
  90                         __u8 locator_status_bits;
  91                 } word2;
  92         } u2;
  93 };
  94
  95 #define LISP_HLEN (sizeof(struct udphdr) + sizeof(struct lisphdr))
  96
  97 /**
  98  * struct lisp_port - Keeps track of open UDP ports
  99  * @dst_port: lisp UDP port no.
 100  * @list: list element in @lisp_ports.
 101  * @lisp_rcv_socket: The socket created for this port number.
 102  * @name: vport name.
 103  */
 104 struct lisp_port {
 105         __be16 dst_port;
 106         struct list_head list;
 107         struct socket *lisp_rcv_socket;
 108         char name[IFNAMSIZ];
 109 };
 110
 111 static LIST_HEAD(lisp_ports);
 112
 113 static inline struct lisp_port *lisp_vport(const struct vport *vport)
 114 {
 115         return vport_priv(vport);
 116 }
 117
 118 static struct lisp_port *lisp_find_port(struct net *net, __be16 port)
 119 {
 120         struct lisp_port *lisp_port;
 121
 122         list_for_each_entry_rcu(lisp_port, &lisp_ports, list) {
 123                 if (lisp_port->dst_port == port &&
 124                         net_eq(sock_net(lisp_port->lisp_rcv_socket->sk), net))
 125                         return lisp_port;
 126         }
 127
 128         return NULL;
 129 }
 130
 131 static inline struct lisphdr *lisp_hdr(const struct sk_buff *skb)
 132 {
 133         return (struct lisphdr *)(udp_hdr(skb) + 1);
 134 }
 135
 136 /* Convert 64 bit tunnel ID to 24 bit Instance ID. */
 137 static void tunnel_id_to_instance_id(__be64 tun_id, __u8 *iid)
 138 {
 139
 140 #ifdef __BIG_ENDIAN
 141         iid[0] = (__force __u8)(tun_id >> 16);
 142         iid[1] = (__force __u8)(tun_id >> 8);
 143         iid[2] = (__force __u8)tun_id;
 144 #else
 145         iid[0] = (__force __u8)((__force u64)tun_id >> 40);
 146         iid[1] = (__force __u8)((__force u64)tun_id >> 48);
 147         iid[2] = (__force __u8)((__force u64)tun_id >> 56);
 148 #endif
 149 }
 150
 151 /* Convert 24 bit Instance ID to 64 bit tunnel ID. */
 152 static __be64 instance_id_to_tunnel_id(__u8 *iid)
 153 {
 154 #ifdef __BIG_ENDIAN
 155         return (iid[0] << 16) | (iid[1] << 8) | iid[2];
 156 #else
 157         return (__force __be64)(((__force u64)iid[0] << 40) |
 158                                 ((__force u64)iid[1] << 48) |
 159                                 ((__force u64)iid[2] << 56));
 160 #endif
 161 }
 162
 163 static void lisp_build_header(const struct vport *vport,
 164                               struct sk_buff *skb,
 165                               int tunnel_hlen)
 166 {
 167         struct lisp_port *lisp_port = lisp_vport(vport);
 168         struct udphdr *udph = udp_hdr(skb);
 169         struct lisphdr *lisph = (struct lisphdr *)(udph + 1);
 170         const struct ovs_key_ipv4_tunnel *tun_key = OVS_CB(skb)->tun_key;
 171
 172         udph->dest = lisp_port->dst_port;
 173         udph->source = htons(ovs_tnl_get_src_port(skb));
 174         udph->check = 0;
 175         udph->len = htons(skb->len - skb_transport_offset(skb));
 176
 177         lisph->nonce_present = 0;       /* We don't support echo nonce algorithm */
 178         lisph->locator_status_bits_present = 1; /* Set LSB */
 179         lisph->solicit_echo_nonce = 0;  /* No echo noncing */
 180         lisph->map_version_present = 0; /* No mapping versioning, nonce instead */
 181         lisph->instance_id_present = 1; /* Store the tun_id as Instance ID  */
 182         lisph->reserved_flags = 0;      /* Reserved flags, set to 0  */
 183
 184         lisph->u1.nonce[0] = 0;
 185         lisph->u1.nonce[1] = 0;
 186         lisph->u1.nonce[2] = 0;
 187
 188         tunnel_id_to_instance_id(tun_key->tun_id, &lisph->u2.word2.instance_id[0]);
 189         lisph->u2.word2.locator_status_bits = 1;
 190 }
 191
 192 /* Called with rcu_read_lock and BH disabled. */
 193 static int lisp_rcv(struct sock *sk, struct sk_buff *skb)
 194 {
 195         struct lisp_port *lisp_port;
 196         struct lisphdr *lisph;
 197         struct iphdr *iph, *inner_iph;
 198         struct ovs_key_ipv4_tunnel tun_key;
 199         __be64 key;
 200         struct ethhdr *ethh;
 201         __be16 protocol;
 202
 203         lisp_port = lisp_find_port(dev_net(skb->dev), udp_hdr(skb)->dest);
 204         if (unlikely(!lisp_port))
 205                 goto error;
 206
 207         if (unlikely(!pskb_may_pull(skb, LISP_HLEN)))
 208                 goto error;
 209
 210         lisph = lisp_hdr(skb);
 211
 212         skb_pull_rcsum(skb, LISP_HLEN);
 213
 214         if (lisph->instance_id_present != 1)
 215                 key = 0;
 216         else
 217                 key = instance_id_to_tunnel_id(&lisph->u2.word2.instance_id[0]);
 218
 219         /* Save outer tunnel values */
 220         iph = ip_hdr(skb);
 221         ovs_flow_tun_key_init(&tun_key, iph, key, TUNNEL_KEY);
 222
 223         /* Drop non-IP inner packets */
 224         inner_iph = (struct iphdr *)(lisph + 1);
 225         switch (inner_iph->version) {
 226         case 4:
 227                 protocol = htons(ETH_P_IP);
 228                 break;
 229         case 6:
 230                 protocol = htons(ETH_P_IPV6);
 231                 break;
 232         default:
 233                 goto error;
 234         }
 235
 236         /* Add Ethernet header */
 237         ethh = (struct ethhdr *)skb_push(skb, ETH_HLEN);
 238         memset(ethh, 0, ETH_HLEN);
 239         ethh->h_dest[0] = 0x02;
 240         ethh->h_source[0] = 0x02;
 241         ethh->h_proto = protocol;
 242
 243         ovs_skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
 244
 245         ovs_tnl_rcv(vport_from_priv(lisp_port), skb, &tun_key);
 246         goto out;
 247
 248 error:
 249         kfree_skb(skb);
 250 out:
 251         return 0;
 252 }
 253
 254 /* Arbitrary value.  Irrelevant as long as it's not 0 since we set the handler. */
 255 #define UDP_ENCAP_LISP 1
 256 static int lisp_socket_init(struct lisp_port *lisp_port, struct net *net)
 257 {
 258         struct sockaddr_in sin;
 259         int err;
 260
 261         err = sock_create_kern(AF_INET, SOCK_DGRAM, 0,
 262                                &lisp_port->lisp_rcv_socket);
 263         if (err)
 264                 goto error;
 265
 266         /* release net ref. */
 267         sk_change_net(lisp_port->lisp_rcv_socket->sk, net);
 268
 269         sin.sin_family = AF_INET;
 270         sin.sin_addr.s_addr = htonl(INADDR_ANY);
 271         sin.sin_port = lisp_port->dst_port;
 272
 273         err = kernel_bind(lisp_port->lisp_rcv_socket, (struct sockaddr *)&sin,
 274                           sizeof(struct sockaddr_in));
 275         if (err)
 276                 goto error_sock;
 277
 278         udp_sk(lisp_port->lisp_rcv_socket->sk)->encap_type = UDP_ENCAP_LISP;
 279         udp_sk(lisp_port->lisp_rcv_socket->sk)->encap_rcv = lisp_rcv;
 280
 281         udp_encap_enable();
 282
 283         return 0;
 284
 285 error_sock:
 286         sk_release_kernel(lisp_port->lisp_rcv_socket->sk);
 287 error:
 288         pr_warn("cannot register lisp protocol handler: %d\n", err);
 289         return err;
 290 }
 291
 292 static int lisp_get_options(const struct vport *vport, struct sk_buff *skb)
 293 {
 294         struct lisp_port *lisp_port = lisp_vport(vport);
 295
 296         if (nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT, ntohs(lisp_port->dst_port)))
 297                 return -EMSGSIZE;
 298         return 0;
 299 }
 300
 301 static void lisp_tnl_destroy(struct vport *vport)
 302 {
 303         struct lisp_port *lisp_port = lisp_vport(vport);
 304
 305         list_del_rcu(&lisp_port->list);
 306         /* Release socket */
 307         sk_release_kernel(lisp_port->lisp_rcv_socket->sk);
 308
 309         ovs_vport_deferred_free(vport);
 310 }
 311
 312 static struct vport *lisp_tnl_create(const struct vport_parms *parms)
 313 {
 314         struct net *net = ovs_dp_get_net(parms->dp);
 315         struct nlattr *options = parms->options;
 316         struct lisp_port *lisp_port;
 317         struct vport *vport;
 318         struct nlattr *a;
 319         int err;
 320         u16 dst_port;
 321
 322         if (!options) {
 323                 err = -EINVAL;
 324                 goto error;
 325         }
 326
 327         a = nla_find_nested(options, OVS_TUNNEL_ATTR_DST_PORT);
 328         if (a && nla_len(a) == sizeof(u16)) {
 329                 dst_port = nla_get_u16(a);
 330         } else {
 331                 /* Require destination port from userspace. */
 332                 err = -EINVAL;
 333                 goto error;
 334         }
 335
 336         /* Verify if we already have a socket created for this port */
 337         if (lisp_find_port(net, htons(dst_port))) {
 338                 err = -EEXIST;
 339                 goto error;
 340         }
 341
 342         vport = ovs_vport_alloc(sizeof(struct lisp_port),
 343                                 &ovs_lisp_vport_ops, parms);
 344         if (IS_ERR(vport))
 345                 return vport;
 346
 347         lisp_port = lisp_vport(vport);
 348         lisp_port->dst_port = htons(dst_port);
 349         strncpy(lisp_port->name, parms->name, IFNAMSIZ);
 350
 351         err = lisp_socket_init(lisp_port, net);
 352         if (err)
 353                 goto error_free;
 354
 355         list_add_tail_rcu(&lisp_port->list, &lisp_ports);
 356         return vport;
 357
 358 error_free:
 359         ovs_vport_free(vport);
 360 error:
 361         return ERR_PTR(err);
 362 }
 363
 364 static int lisp_tnl_send(struct vport *vport, struct sk_buff *skb)
 365 {
 366         int tnl_len;
 367         int network_offset = skb_network_offset(skb);
 368
 369         if (unlikely(!OVS_CB(skb)->tun_key))
 370                 return -EINVAL;
 371
 372         /* We only encapsulate IPv4 and IPv6 packets */
 373         switch (skb->protocol) {
 374         case htons(ETH_P_IP):
 375         case htons(ETH_P_IPV6):
 376                 /* Pop off "inner" Ethernet header */
 377                 skb_pull(skb, network_offset);
 378                 tnl_len = ovs_tnl_send(vport, skb, IPPROTO_UDP,
 379                                 LISP_HLEN, lisp_build_header);
 380                 return tnl_len > 0 ? tnl_len + network_offset : tnl_len;
 381         default:
 382                 kfree_skb(skb);
 383                 return 0;
 384         }
 385 }
 386
 387 static const char *lisp_get_name(const struct vport *vport)
 388 {
 389         struct lisp_port *lisp_port = lisp_vport(vport);
 390         return lisp_port->name;
 391 }
 392
 393 const struct vport_ops ovs_lisp_vport_ops = {
 394         .type           = OVS_VPORT_TYPE_LISP,
 395         .create         = lisp_tnl_create,
 396         .destroy        = lisp_tnl_destroy,
 397         .get_name       = lisp_get_name,
 398         .get_options    = lisp_get_options,
 399         .send           = lisp_tnl_send,
 400 };
 401 #else
 402 #warning LISP tunneling will not be available on kernels before 2.6.26
 403 #endif /* Linux kernel < 2.6.26 */