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
  24 #include <linux/in.h>
  25 #include <linux/ip.h>
  26 #include <linux/net.h>
  27 #include <linux/rculist.h>
  28 #include <linux/udp.h>
  29
  30 #include <net/icmp.h>
  31 #include <net/ip.h>
  32 #include <net/route.h>
  33 #include <net/udp.h>
  34 #include <net/xfrm.h>
  35
  36 #include "datapath.h"
  37 #include "vport.h"
  38
  39 /*
  40  *  LISP encapsulation header:
  41  *
  42  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  43  *  |N|L|E|V|I|flags|            Nonce/Map-Version                  |
  44  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  45  *  |                 Instance ID/Locator Status Bits               |
  46  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  47  *
  48  */
  49
  50 /**
  51  * struct lisphdr - LISP header
  52  * @nonce_present: Flag indicating the presence of a 24 bit nonce value.
  53  * @locator_status_bits_present: Flag indicating the presence of Locator Status
  54  *                               Bits (LSB).
  55  * @solicit_echo_nonce: Flag indicating the use of the echo noncing mechanism.
  56  * @map_version_present: Flag indicating the use of mapping versioning.
  57  * @instance_id_present: Flag indicating the presence of a 24 bit Instance ID.
  58  * @reserved_flags: 3 bits reserved for future flags.
  59  * @nonce: 24 bit nonce value.
  60  * @map_version: 24 bit mapping version.
  61  * @locator_status_bits: Locator Status Bits: 32 bits when instance_id_present
  62  *                       is not set, 8 bits when it is.
  63  * @instance_id: 24 bit Instance ID
  64  */
  65 struct lisphdr {
  66 #ifdef __LITTLE_ENDIAN_BITFIELD
  67         __u8 reserved_flags:3;
  68         __u8 instance_id_present:1;
  69         __u8 map_version_present:1;
  70         __u8 solicit_echo_nonce:1;
  71         __u8 locator_status_bits_present:1;
  72         __u8 nonce_present:1;
  73 #else
  74         __u8 nonce_present:1;
  75         __u8 locator_status_bits_present:1;
  76         __u8 solicit_echo_nonce:1;
  77         __u8 map_version_present:1;
  78         __u8 instance_id_present:1;
  79         __u8 reserved_flags:3;
  80 #endif
  81         union {
  82                 __u8 nonce[3];
  83                 __u8 map_version[3];
  84         } u1;
  85         union {
  86                 __be32 locator_status_bits;
  87                 struct {
  88                         __u8 instance_id[3];
  89                         __u8 locator_status_bits;
  90                 } word2;
  91         } u2;
  92 };
  93
  94 #define LISP_HLEN (sizeof(struct udphdr) + sizeof(struct lisphdr))
  95
  96 /**
  97  * struct lisp_port - Keeps track of open UDP ports
  98  * @dst_port: lisp UDP port no.
  99  * @list: list element in @lisp_ports.
 100  * @lisp_rcv_socket: The socket created for this port number.
 101  * @name: vport name.
 102  */
 103 struct lisp_port {
 104         __be16 dst_port;
 105         struct list_head list;
 106         struct socket *lisp_rcv_socket;
 107         char name[IFNAMSIZ];
 108 };
 109
 110 static LIST_HEAD(lisp_ports);
 111
 112 static inline struct lisp_port *lisp_vport(const struct vport *vport)
 113 {
 114         return vport_priv(vport);
 115 }
 116
 117 static struct lisp_port *lisp_find_port(struct net *net, __be16 port)
 118 {
 119         struct lisp_port *lisp_port;
 120
 121         list_for_each_entry_rcu(lisp_port, &lisp_ports, list) {
 122                 if (lisp_port->dst_port == port &&
 123                         net_eq(sock_net(lisp_port->lisp_rcv_socket->sk), net))
 124                         return lisp_port;
 125         }
 126
 127         return NULL;
 128 }
 129
 130 static inline struct lisphdr *lisp_hdr(const struct sk_buff *skb)
 131 {
 132         return (struct lisphdr *)(udp_hdr(skb) + 1);
 133 }
 134
 135 /* Convert 64 bit tunnel ID to 24 bit Instance ID. */
 136 static void tunnel_id_to_instance_id(__be64 tun_id, __u8 *iid)
 137 {
 138
 139 #ifdef __BIG_ENDIAN
 140         iid[0] = (__force __u8)(tun_id >> 16);
 141         iid[1] = (__force __u8)(tun_id >> 8);
 142         iid[2] = (__force __u8)tun_id;
 143 #else
 144         iid[0] = (__force __u8)((__force u64)tun_id >> 40);
 145         iid[1] = (__force __u8)((__force u64)tun_id >> 48);
 146         iid[2] = (__force __u8)((__force u64)tun_id >> 56);
 147 #endif
 148 }
 149
 150 /* Convert 24 bit Instance ID to 64 bit tunnel ID. */
 151 static __be64 instance_id_to_tunnel_id(__u8 *iid)
 152 {
 153 #ifdef __BIG_ENDIAN
 154         return (iid[0] << 16) | (iid[1] << 8) | iid[2];
 155 #else
 156         return (__force __be64)(((__force u64)iid[0] << 40) |
 157                                 ((__force u64)iid[1] << 48) |
 158                                 ((__force u64)iid[2] << 56));
 159 #endif
 160 }
 161
 162 /* Compute source UDP port for outgoing packet.
 163  * Currently we use the flow hash.
 164  */
 165 static u16 ovs_tnl_get_src_port(struct sk_buff *skb)
 166 {
 167         int low;
 168         int high;
 169         unsigned int range;
 170         struct sw_flow_key *pkt_key = OVS_CB(skb)->pkt_key;
 171         u32 hash = jhash2((const u32 *)pkt_key,
 172                           sizeof(*pkt_key) / sizeof(u32), 0);
 173
 174         inet_get_local_port_range(&low, &high);
 175         range = (high - low) + 1;
 176         return (((u64) hash * range) >> 32) + low;
 177 }
 178
 179 static void lisp_build_header(const struct vport *vport,
 180                               struct sk_buff *skb,
 181                               int tunnel_hlen)
 182 {
 183         struct lisp_port *lisp_port = lisp_vport(vport);
 184         struct udphdr *udph = udp_hdr(skb);
 185         struct lisphdr *lisph = (struct lisphdr *)(udph + 1);
 186         const struct ovs_key_ipv4_tunnel *tun_key = OVS_CB(skb)->tun_key;
 187
 188         udph->dest = lisp_port->dst_port;
 189         udph->source = htons(ovs_tnl_get_src_port(skb));
 190         udph->check = 0;
 191         udph->len = htons(skb->len - skb_transport_offset(skb));
 192
 193         lisph->nonce_present = 0;       /* We don't support echo nonce algorithm */
 194         lisph->locator_status_bits_present = 1; /* Set LSB */
 195         lisph->solicit_echo_nonce = 0;  /* No echo noncing */
 196         lisph->map_version_present = 0; /* No mapping versioning, nonce instead */
 197         lisph->instance_id_present = 1; /* Store the tun_id as Instance ID  */
 198         lisph->reserved_flags = 0;      /* Reserved flags, set to 0  */
 199
 200         lisph->u1.nonce[0] = 0;
 201         lisph->u1.nonce[1] = 0;
 202         lisph->u1.nonce[2] = 0;
 203
 204         tunnel_id_to_instance_id(tun_key->tun_id, &lisph->u2.word2.instance_id[0]);
 205         lisph->u2.word2.locator_status_bits = 1;
 206 }
 207
 208 /* Called with rcu_read_lock and BH disabled. */
 209 static int lisp_rcv(struct sock *sk, struct sk_buff *skb)
 210 {
 211         struct lisp_port *lisp_port;
 212         struct lisphdr *lisph;
 213         struct iphdr *iph, *inner_iph;
 214         struct ovs_key_ipv4_tunnel tun_key;
 215         __be64 key;
 216         struct ethhdr *ethh;
 217         __be16 protocol;
 218
 219         lisp_port = lisp_find_port(dev_net(skb->dev), udp_hdr(skb)->dest);
 220         if (unlikely(!lisp_port))
 221                 goto error;
 222
 223         if (iptunnel_pull_header(skb, LISP_HLEN, 0))
 224                 goto error;
 225
 226         lisph = lisp_hdr(skb);
 227
 228         if (lisph->instance_id_present != 1)
 229                 key = 0;
 230         else
 231                 key = instance_id_to_tunnel_id(&lisph->u2.word2.instance_id[0]);
 232
 233         /* Save outer tunnel values */
 234         iph = ip_hdr(skb);
 235         ovs_flow_tun_key_init(&tun_key, iph, key, TUNNEL_KEY);
 236
 237         /* Drop non-IP inner packets */
 238         inner_iph = (struct iphdr *)(lisph + 1);
 239         switch (inner_iph->version) {
 240         case 4:
 241                 protocol = htons(ETH_P_IP);
 242                 break;
 243         case 6:
 244                 protocol = htons(ETH_P_IPV6);
 245                 break;
 246         default:
 247                 goto error;
 248         }
 249         skb->protocol = protocol;
 250
 251         /* Add Ethernet header */
 252         ethh = (struct ethhdr *)skb_push(skb, ETH_HLEN);
 253         memset(ethh, 0, ETH_HLEN);
 254         ethh->h_dest[0] = 0x02;
 255         ethh->h_source[0] = 0x02;
 256         ethh->h_proto = protocol;
 257
 258         ovs_skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
 259
 260         ovs_vport_receive(vport_from_priv(lisp_port), skb, &tun_key);
 261         goto out;
 262
 263 error:
 264         kfree_skb(skb);
 265 out:
 266         return 0;
 267 }
 268
 269 /* Arbitrary value.  Irrelevant as long as it's not 0 since we set the handler. */
 270 #define UDP_ENCAP_LISP 1
 271 static int lisp_socket_init(struct lisp_port *lisp_port, struct net *net)
 272 {
 273         struct sockaddr_in sin;
 274         int err;
 275
 276         err = sock_create_kern(AF_INET, SOCK_DGRAM, 0,
 277                                &lisp_port->lisp_rcv_socket);
 278         if (err)
 279                 goto error;
 280
 281         /* release net ref. */
 282         sk_change_net(lisp_port->lisp_rcv_socket->sk, net);
 283
 284         sin.sin_family = AF_INET;
 285         sin.sin_addr.s_addr = htonl(INADDR_ANY);
 286         sin.sin_port = lisp_port->dst_port;
 287
 288         err = kernel_bind(lisp_port->lisp_rcv_socket, (struct sockaddr *)&sin,
 289                           sizeof(struct sockaddr_in));
 290         if (err)
 291                 goto error_sock;
 292
 293         udp_sk(lisp_port->lisp_rcv_socket->sk)->encap_type = UDP_ENCAP_LISP;
 294         udp_sk(lisp_port->lisp_rcv_socket->sk)->encap_rcv = lisp_rcv;
 295
 296         udp_encap_enable();
 297
 298         return 0;
 299
 300 error_sock:
 301         sk_release_kernel(lisp_port->lisp_rcv_socket->sk);
 302 error:
 303         pr_warn("cannot register lisp protocol handler: %d\n", err);
 304         return err;
 305 }
 306
 307 static int lisp_get_options(const struct vport *vport, struct sk_buff *skb)
 308 {
 309         struct lisp_port *lisp_port = lisp_vport(vport);
 310
 311         if (nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT, ntohs(lisp_port->dst_port)))
 312                 return -EMSGSIZE;
 313         return 0;
 314 }
 315
 316 static void lisp_tnl_destroy(struct vport *vport)
 317 {
 318         struct lisp_port *lisp_port = lisp_vport(vport);
 319
 320         list_del_rcu(&lisp_port->list);
 321         /* Release socket */
 322         sk_release_kernel(lisp_port->lisp_rcv_socket->sk);
 323
 324         ovs_vport_deferred_free(vport);
 325 }
 326
 327 static struct vport *lisp_tnl_create(const struct vport_parms *parms)
 328 {
 329         struct net *net = ovs_dp_get_net(parms->dp);
 330         struct nlattr *options = parms->options;
 331         struct lisp_port *lisp_port;
 332         struct vport *vport;
 333         struct nlattr *a;
 334         int err;
 335         u16 dst_port;
 336
 337         if (!options) {
 338                 err = -EINVAL;
 339                 goto error;
 340         }
 341
 342         a = nla_find_nested(options, OVS_TUNNEL_ATTR_DST_PORT);
 343         if (a && nla_len(a) == sizeof(u16)) {
 344                 dst_port = nla_get_u16(a);
 345         } else {
 346                 /* Require destination port from userspace. */
 347                 err = -EINVAL;
 348                 goto error;
 349         }
 350
 351         /* Verify if we already have a socket created for this port */
 352         if (lisp_find_port(net, htons(dst_port))) {
 353                 err = -EEXIST;
 354                 goto error;
 355         }
 356
 357         vport = ovs_vport_alloc(sizeof(struct lisp_port),
 358                                 &ovs_lisp_vport_ops, parms);
 359         if (IS_ERR(vport))
 360                 return vport;
 361
 362         lisp_port = lisp_vport(vport);
 363         lisp_port->dst_port = htons(dst_port);
 364         strncpy(lisp_port->name, parms->name, IFNAMSIZ);
 365
 366         err = lisp_socket_init(lisp_port, net);
 367         if (err)
 368                 goto error_free;
 369
 370         list_add_tail_rcu(&lisp_port->list, &lisp_ports);
 371         return vport;
 372
 373 error_free:
 374         ovs_vport_free(vport);
 375 error:
 376         return ERR_PTR(err);
 377 }
 378
 379 static bool need_linearize(const struct sk_buff *skb)
 380 {
 381         int i;
 382
 383         if (unlikely(skb_shinfo(skb)->frag_list))
 384                 return true;
 385
 386         /*
 387          * Generally speaking we should linearize if there are paged frags.
 388          * However, if all of the refcounts are 1 we know nobody else can
 389          * change them from underneath us and we can skip the linearization.
 390          */
 391         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 392                 if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
 393                         return true;
 394
 395         return false;
 396 }
 397
 398 static struct sk_buff *handle_offloads(struct sk_buff *skb)
 399 {
 400         int err;
 401
 402         if (skb_is_gso(skb)) {
 403                 struct sk_buff *nskb;
 404                 char cb[sizeof(skb->cb)];
 405
 406                 memcpy(cb, skb->cb, sizeof(cb));
 407
 408                 nskb = __skb_gso_segment(skb, 0, false);
 409                 if (IS_ERR(nskb)) {
 410                         err = PTR_ERR(nskb);
 411                         goto error;
 412                 }
 413
 414                 consume_skb(skb);
 415                 skb = nskb;
 416                 while (nskb) {
 417                         memcpy(nskb->cb, cb, sizeof(cb));
 418                         nskb = nskb->next;
 419                 }
 420         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 421                 /* Pages aren't locked and could change at any time.
 422                  * If this happens after we compute the checksum, the
 423                  * checksum will be wrong.  We linearize now to avoid
 424                  * this problem.
 425                  */
 426                 if (unlikely(need_linearize(skb))) {
 427                         err = __skb_linearize(skb);
 428                         if (unlikely(err))
 429                                 goto error;
 430                 }
 431
 432                 err = skb_checksum_help(skb);
 433                 if (unlikely(err))
 434                         goto error;
 435         }
 436
 437         skb->ip_summed = CHECKSUM_NONE;
 438         return skb;
 439
 440 error:
 441         return ERR_PTR(err);
 442 }
 443
 444 static int ovs_tnl_send(struct vport *vport, struct sk_buff *skb,
 445                         u8 ipproto, int tunnel_hlen,
 446                         void (*build_header)(const struct vport *,
 447                                              struct sk_buff *,
 448                                              int tunnel_hlen))
 449 {
 450         int min_headroom;
 451         struct rtable *rt;
 452         __be32 saddr;
 453         int sent_len = 0;
 454         int err;
 455         struct sk_buff *nskb;
 456
 457         /* Route lookup */
 458         saddr = OVS_CB(skb)->tun_key->ipv4_src;
 459         rt = find_route(ovs_dp_get_net(vport->dp),
 460                         &saddr,
 461                         OVS_CB(skb)->tun_key->ipv4_dst,
 462                         ipproto,
 463                         OVS_CB(skb)->tun_key->ipv4_tos,
 464                         skb->mark);
 465         if (IS_ERR(rt)) {
 466                 err = PTR_ERR(rt);
 467                 goto error;
 468         }
 469
 470         tunnel_hlen += sizeof(struct iphdr);
 471
 472         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
 473                         + tunnel_hlen
 474                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
 475
 476         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 477                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 478                                                 skb_headroom(skb) +
 479                                                 16);
 480
 481                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 482                                         0, GFP_ATOMIC);
 483                 if (unlikely(err))
 484                         goto err_free_rt;
 485         }
 486
 487         /* Offloading */
 488         nskb = handle_offloads(skb);
 489         if (IS_ERR(nskb)) {
 490                 err = PTR_ERR(nskb);
 491                 goto err_free_rt;
 492         }
 493         skb = nskb;
 494
 495         /* Reset SKB */
 496         nf_reset(skb);
 497         secpath_reset(skb);
 498         skb_dst_drop(skb);
 499         skb_clear_rxhash(skb);
 500
 501         while (skb) {
 502                 struct sk_buff *next_skb = skb->next;
 503                 struct iphdr *iph;
 504                 int frag_len;
 505
 506                 skb->next = NULL;
 507
 508                 if (vlan_tx_tag_present(skb)) {
 509                         if (unlikely(!__vlan_put_tag(skb,
 510                                                         skb->vlan_proto,
 511                                                         vlan_tx_tag_get(skb))))
 512                                 goto next;
 513
 514                         vlan_set_tci(skb, 0);
 515                 }
 516
 517                 frag_len = skb->len;
 518                 skb_push(skb, tunnel_hlen);
 519                 skb_reset_network_header(skb);
 520                 skb_set_transport_header(skb, sizeof(struct iphdr));
 521
 522                 if (next_skb)
 523                         skb_dst_set(skb, dst_clone(&rt_dst(rt)));
 524                 else
 525                         skb_dst_set(skb, &rt_dst(rt));
 526
 527                 /* Push Tunnel header. */
 528                 build_header(vport, skb, tunnel_hlen);
 529
 530                 /* Push IP header. */
 531                 iph = ip_hdr(skb);
 532                 iph->version    = 4;
 533                 iph->ihl        = sizeof(struct iphdr) >> 2;
 534                 iph->protocol   = ipproto;
 535                 iph->daddr      = OVS_CB(skb)->tun_key->ipv4_dst;
 536                 iph->saddr      = saddr;
 537                 iph->tos        = OVS_CB(skb)->tun_key->ipv4_tos;
 538                 iph->ttl        = OVS_CB(skb)->tun_key->ipv4_ttl;
 539                 iph->frag_off   = OVS_CB(skb)->tun_key->tun_flags &
 540                                   TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 541                 /*
 542                  * Allow our local IP stack to fragment the outer packet even
 543                  * if the DF bit is set as a last resort.  We also need to
 544                  * force selection of an IP ID here with __ip_select_ident(),
 545                  * as ip_select_ident() assumes a proper ID is not needed when
 546                  * when the DF bit is set.
 547                  */
 548                 skb->local_df = 1;
 549                 __ip_select_ident(iph, skb_dst(skb), 0);
 550
 551                 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 552
 553                 err = ip_local_out(skb);
 554                 if (unlikely(net_xmit_eval(err)))
 555                         goto next;
 556
 557                 sent_len += frag_len;
 558
 559 next:
 560                 skb = next_skb;
 561         }
 562
 563         return sent_len;
 564
 565 err_free_rt:
 566         ip_rt_put(rt);
 567 error:
 568         return err;
 569 }
 570
 571 static int lisp_tnl_send(struct vport *vport, struct sk_buff *skb)
 572 {
 573         int tnl_len;
 574         int network_offset = skb_network_offset(skb);
 575
 576         if (unlikely(!OVS_CB(skb)->tun_key))
 577                 return -EINVAL;
 578
 579         /* We only encapsulate IPv4 and IPv6 packets */
 580         switch (skb->protocol) {
 581         case htons(ETH_P_IP):
 582         case htons(ETH_P_IPV6):
 583                 /* Pop off "inner" Ethernet header */
 584                 skb_pull(skb, network_offset);
 585                 tnl_len = ovs_tnl_send(vport, skb, IPPROTO_UDP,
 586                                 LISP_HLEN, lisp_build_header);
 587                 return tnl_len > 0 ? tnl_len + network_offset : tnl_len;
 588         default:
 589                 kfree_skb(skb);
 590                 return 0;
 591         }
 592 }
 593
 594 static const char *lisp_get_name(const struct vport *vport)
 595 {
 596         struct lisp_port *lisp_port = lisp_vport(vport);
 597         return lisp_port->name;
 598 }
 599
 600 const struct vport_ops ovs_lisp_vport_ops = {
 601         .type           = OVS_VPORT_TYPE_LISP,
 602         .create         = lisp_tnl_create,
 603         .destroy        = lisp_tnl_destroy,
 604         .get_name       = lisp_get_name,
 605         .get_options    = lisp_get_options,
 606         .send           = lisp_tnl_send,
 607 };