drivers/net/wireguard/socket.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
   4  */
   5
   6 #include "device.h"
   7 #include "peer.h"
   8 #include "socket.h"
   9 #include "queueing.h"
  10 #include "messages.h"
  11
  12 #include <linux/ctype.h>
  13 #include <linux/net.h>
  14 #include <linux/if_vlan.h>
  15 #include <linux/if_ether.h>
  16 #include <linux/inetdevice.h>
  17 #include <net/udp_tunnel.h>
  18 #include <net/ipv6.h>
  19
  20 static int send4(struct wg_device *wg, struct sk_buff *skb,
  21                  struct endpoint *endpoint, u8 ds, struct dst_cache *cache)
  22 {
  23         struct flowi4 fl = {
  24                 .saddr = endpoint->src4.s_addr,
  25                 .daddr = endpoint->addr4.sin_addr.s_addr,
  26                 .fl4_dport = endpoint->addr4.sin_port,
  27                 .flowi4_mark = wg->fwmark,
  28                 .flowi4_proto = IPPROTO_UDP
  29         };
  30         struct rtable *rt = NULL;
  31         struct sock *sock;
  32         int ret = 0;
  33
  34         skb_mark_not_on_list(skb);
  35         skb->dev = wg->dev;
  36         skb->mark = wg->fwmark;
  37
  38         rcu_read_lock_bh();
  39         sock = rcu_dereference_bh(wg->sock4);
  40
  41         if (unlikely(!sock)) {
  42                 ret = -ENONET;
  43                 goto err;
  44         }
  45
  46         fl.fl4_sport = inet_sk(sock)->inet_sport;
  47
  48         if (cache)
  49                 rt = dst_cache_get_ip4(cache, &fl.saddr);
  50
  51         if (!rt) {
  52                 security_sk_classify_flow(sock, flowi4_to_flowi_common(&fl));
  53                 if (unlikely(!inet_confirm_addr(sock_net(sock), NULL, 0,
  54                                                 fl.saddr, RT_SCOPE_HOST))) {
  55                         endpoint->src4.s_addr = 0;
  56                         *(__force __be32 *)&endpoint->src_if4 = 0;
  57                         fl.saddr = 0;
  58                         if (cache)
  59                                 dst_cache_reset(cache);
  60                 }
  61                 rt = ip_route_output_flow(sock_net(sock), &fl, sock);
  62                 if (unlikely(endpoint->src_if4 && ((IS_ERR(rt) &&
  63                              PTR_ERR(rt) == -EINVAL) || (!IS_ERR(rt) &&
  64                              rt->dst.dev->ifindex != endpoint->src_if4)))) {
  65                         endpoint->src4.s_addr = 0;
  66                         *(__force __be32 *)&endpoint->src_if4 = 0;
  67                         fl.saddr = 0;
  68                         if (cache)
  69                                 dst_cache_reset(cache);
  70                         if (!IS_ERR(rt))
  71                                 ip_rt_put(rt);
  72                         rt = ip_route_output_flow(sock_net(sock), &fl, sock);
  73                 }
  74                 if (unlikely(IS_ERR(rt))) {
  75                         ret = PTR_ERR(rt);
  76                         net_dbg_ratelimited("%s: No route to %pISpfsc, error %d\n",
  77                                             wg->dev->name, &endpoint->addr, ret);
  78                         goto err;
  79                 }
  80                 if (cache)
  81                         dst_cache_set_ip4(cache, &rt->dst, fl.saddr);
  82         }
  83
  84         skb->ignore_df = 1;
  85         udp_tunnel_xmit_skb(rt, sock, skb, fl.saddr, fl.daddr, ds,
  86                             ip4_dst_hoplimit(&rt->dst), 0, fl.fl4_sport,
  87                             fl.fl4_dport, false, false);
  88         goto out;
  89
  90 err:
  91         kfree_skb(skb);
  92 out:
  93         rcu_read_unlock_bh();
  94         return ret;
  95 }
  96
  97 static int send6(struct wg_device *wg, struct sk_buff *skb,
  98                  struct endpoint *endpoint, u8 ds, struct dst_cache *cache)
  99 {
 100 #if IS_ENABLED(CONFIG_IPV6)
 101         struct flowi6 fl = {
 102                 .saddr = endpoint->src6,
 103                 .daddr = endpoint->addr6.sin6_addr,
 104                 .fl6_dport = endpoint->addr6.sin6_port,
 105                 .flowi6_mark = wg->fwmark,
 106                 .flowi6_oif = endpoint->addr6.sin6_scope_id,
 107                 .flowi6_proto = IPPROTO_UDP
 108                 /* TODO: addr->sin6_flowinfo */
 109         };
 110         struct dst_entry *dst = NULL;
 111         struct sock *sock;
 112         int ret = 0;
 113
 114         skb_mark_not_on_list(skb);
 115         skb->dev = wg->dev;
 116         skb->mark = wg->fwmark;
 117
 118         rcu_read_lock_bh();
 119         sock = rcu_dereference_bh(wg->sock6);
 120
 121         if (unlikely(!sock)) {
 122                 ret = -ENONET;
 123                 goto err;
 124         }
 125
 126         fl.fl6_sport = inet_sk(sock)->inet_sport;
 127
 128         if (cache)
 129                 dst = dst_cache_get_ip6(cache, &fl.saddr);
 130
 131         if (!dst) {
 132                 security_sk_classify_flow(sock, flowi6_to_flowi_common(&fl));
 133                 if (unlikely(!ipv6_addr_any(&fl.saddr) &&
 134                              !ipv6_chk_addr(sock_net(sock), &fl.saddr, NULL, 0))) {
 135                         endpoint->src6 = fl.saddr = in6addr_any;
 136                         if (cache)
 137                                 dst_cache_reset(cache);
 138                 }
 139                 dst = ipv6_stub->ipv6_dst_lookup_flow(sock_net(sock), sock, &fl,
 140                                                       NULL);
 141                 if (unlikely(IS_ERR(dst))) {
 142                         ret = PTR_ERR(dst);
 143                         net_dbg_ratelimited("%s: No route to %pISpfsc, error %d\n",
 144                                             wg->dev->name, &endpoint->addr, ret);
 145                         goto err;
 146                 }
 147                 if (cache)
 148                         dst_cache_set_ip6(cache, dst, &fl.saddr);
 149         }
 150
 151         skb->ignore_df = 1;
 152         udp_tunnel6_xmit_skb(dst, sock, skb, skb->dev, &fl.saddr, &fl.daddr, ds,
 153                              ip6_dst_hoplimit(dst), 0, fl.fl6_sport,
 154                              fl.fl6_dport, false);
 155         goto out;
 156
 157 err:
 158         kfree_skb(skb);
 159 out:
 160         rcu_read_unlock_bh();
 161         return ret;
 162 #else
 163         return -EAFNOSUPPORT;
 164 #endif
 165 }
 166
 167 int wg_socket_send_skb_to_peer(struct wg_peer *peer, struct sk_buff *skb, u8 ds)
 168 {
 169         size_t skb_len = skb->len;
 170         int ret = -EAFNOSUPPORT;
 171
 172         read_lock_bh(&peer->endpoint_lock);
 173         if (peer->endpoint.addr.sa_family == AF_INET)
 174                 ret = send4(peer->device, skb, &peer->endpoint, ds,
 175                             &peer->endpoint_cache);
 176         else if (peer->endpoint.addr.sa_family == AF_INET6)
 177                 ret = send6(peer->device, skb, &peer->endpoint, ds,
 178                             &peer->endpoint_cache);
 179         else
 180                 dev_kfree_skb(skb);
 181         if (likely(!ret))
 182                 peer->tx_bytes += skb_len;
 183         read_unlock_bh(&peer->endpoint_lock);
 184
 185         return ret;
 186 }
 187
 188 int wg_socket_send_buffer_to_peer(struct wg_peer *peer, void *buffer,
 189                                   size_t len, u8 ds)
 190 {
 191         struct sk_buff *skb = alloc_skb(len + SKB_HEADER_LEN, GFP_ATOMIC);
 192
 193         if (unlikely(!skb))
 194                 return -ENOMEM;
 195
 196         skb_reserve(skb, SKB_HEADER_LEN);
 197         skb_set_inner_network_header(skb, 0);
 198         skb_put_data(skb, buffer, len);
 199         return wg_socket_send_skb_to_peer(peer, skb, ds);
 200 }
 201
 202 int wg_socket_send_buffer_as_reply_to_skb(struct wg_device *wg,
 203                                           struct sk_buff *in_skb, void *buffer,
 204                                           size_t len)
 205 {
 206         int ret = 0;
 207         struct sk_buff *skb;
 208         struct endpoint endpoint;
 209
 210         if (unlikely(!in_skb))
 211                 return -EINVAL;
 212         ret = wg_socket_endpoint_from_skb(&endpoint, in_skb);
 213         if (unlikely(ret < 0))
 214                 return ret;
 215
 216         skb = alloc_skb(len + SKB_HEADER_LEN, GFP_ATOMIC);
 217         if (unlikely(!skb))
 218                 return -ENOMEM;
 219         skb_reserve(skb, SKB_HEADER_LEN);
 220         skb_set_inner_network_header(skb, 0);
 221         skb_put_data(skb, buffer, len);
 222
 223         if (endpoint.addr.sa_family == AF_INET)
 224                 ret = send4(wg, skb, &endpoint, 0, NULL);
 225         else if (endpoint.addr.sa_family == AF_INET6)
 226                 ret = send6(wg, skb, &endpoint, 0, NULL);
 227         /* No other possibilities if the endpoint is valid, which it is,
 228          * as we checked above.
 229          */
 230
 231         return ret;
 232 }
 233
 234 int wg_socket_endpoint_from_skb(struct endpoint *endpoint,
 235                                 const struct sk_buff *skb)
 236 {
 237         memset(endpoint, 0, sizeof(*endpoint));
 238         if (skb->protocol == htons(ETH_P_IP)) {
 239                 endpoint->addr4.sin_family = AF_INET;
 240                 endpoint->addr4.sin_port = udp_hdr(skb)->source;
 241                 endpoint->addr4.sin_addr.s_addr = ip_hdr(skb)->saddr;
 242                 endpoint->src4.s_addr = ip_hdr(skb)->daddr;
 243                 endpoint->src_if4 = skb->skb_iif;
 244         } else if (skb->protocol == htons(ETH_P_IPV6)) {
 245                 endpoint->addr6.sin6_family = AF_INET6;
 246                 endpoint->addr6.sin6_port = udp_hdr(skb)->source;
 247                 endpoint->addr6.sin6_addr = ipv6_hdr(skb)->saddr;
 248                 endpoint->addr6.sin6_scope_id = ipv6_iface_scope_id(
 249                         &ipv6_hdr(skb)->saddr, skb->skb_iif);
 250                 endpoint->src6 = ipv6_hdr(skb)->daddr;
 251         } else {
 252                 return -EINVAL;
 253         }
 254         return 0;
 255 }
 256
 257 static bool endpoint_eq(const struct endpoint *a, const struct endpoint *b)
 258 {
 259         return (a->addr.sa_family == AF_INET && b->addr.sa_family == AF_INET &&
 260                 a->addr4.sin_port == b->addr4.sin_port &&
 261                 a->addr4.sin_addr.s_addr == b->addr4.sin_addr.s_addr &&
 262                 a->src4.s_addr == b->src4.s_addr && a->src_if4 == b->src_if4) ||
 263                (a->addr.sa_family == AF_INET6 &&
 264                 b->addr.sa_family == AF_INET6 &&
 265                 a->addr6.sin6_port == b->addr6.sin6_port &&
 266                 ipv6_addr_equal(&a->addr6.sin6_addr, &b->addr6.sin6_addr) &&
 267                 a->addr6.sin6_scope_id == b->addr6.sin6_scope_id &&
 268                 ipv6_addr_equal(&a->src6, &b->src6)) ||
 269                unlikely(!a->addr.sa_family && !b->addr.sa_family);
 270 }
 271
 272 void wg_socket_set_peer_endpoint(struct wg_peer *peer,
 273                                  const struct endpoint *endpoint)
 274 {
 275         /* First we check unlocked, in order to optimize, since it's pretty rare
 276          * that an endpoint will change. If we happen to be mid-write, and two
 277          * CPUs wind up writing the same thing or something slightly different,
 278          * it doesn't really matter much either.
 279          */
 280         if (endpoint_eq(endpoint, &peer->endpoint))
 281                 return;
 282         write_lock_bh(&peer->endpoint_lock);
 283         if (endpoint->addr.sa_family == AF_INET) {
 284                 peer->endpoint.addr4 = endpoint->addr4;
 285                 peer->endpoint.src4 = endpoint->src4;
 286                 peer->endpoint.src_if4 = endpoint->src_if4;
 287         } else if (endpoint->addr.sa_family == AF_INET6) {
 288                 peer->endpoint.addr6 = endpoint->addr6;
 289                 peer->endpoint.src6 = endpoint->src6;
 290         } else {
 291                 goto out;
 292         }
 293         dst_cache_reset(&peer->endpoint_cache);
 294 out:
 295         write_unlock_bh(&peer->endpoint_lock);
 296 }
 297
 298 void wg_socket_set_peer_endpoint_from_skb(struct wg_peer *peer,
 299                                           const struct sk_buff *skb)
 300 {
 301         struct endpoint endpoint;
 302
 303         if (!wg_socket_endpoint_from_skb(&endpoint, skb))
 304                 wg_socket_set_peer_endpoint(peer, &endpoint);
 305 }
 306
 307 void wg_socket_clear_peer_endpoint_src(struct wg_peer *peer)
 308 {
 309         write_lock_bh(&peer->endpoint_lock);
 310         memset(&peer->endpoint.src6, 0, sizeof(peer->endpoint.src6));
 311         dst_cache_reset(&peer->endpoint_cache);
 312         write_unlock_bh(&peer->endpoint_lock);
 313 }
 314
 315 static int wg_receive(struct sock *sk, struct sk_buff *skb)
 316 {
 317         struct wg_device *wg;
 318
 319         if (unlikely(!sk))
 320                 goto err;
 321         wg = sk->sk_user_data;
 322         if (unlikely(!wg))
 323                 goto err;
 324         skb_mark_not_on_list(skb);
 325         wg_packet_receive(wg, skb);
 326         return 0;
 327
 328 err:
 329         kfree_skb(skb);
 330         return 0;
 331 }
 332
 333 static void sock_free(struct sock *sock)
 334 {
 335         if (unlikely(!sock))
 336                 return;
 337         sk_clear_memalloc(sock);
 338         udp_tunnel_sock_release(sock->sk_socket);
 339 }
 340
 341 static void set_sock_opts(struct socket *sock)
 342 {
 343         sock->sk->sk_allocation = GFP_ATOMIC;
 344         sock->sk->sk_sndbuf = INT_MAX;
 345         sk_set_memalloc(sock->sk);
 346 }
 347
 348 int wg_socket_init(struct wg_device *wg, u16 port)
 349 {
 350         struct net *net;
 351         int ret;
 352         struct udp_tunnel_sock_cfg cfg = {
 353                 .sk_user_data = wg,
 354                 .encap_type = 1,
 355                 .encap_rcv = wg_receive
 356         };
 357         struct socket *new4 = NULL, *new6 = NULL;
 358         struct udp_port_cfg port4 = {
 359                 .family = AF_INET,
 360                 .local_ip.s_addr = htonl(INADDR_ANY),
 361                 .local_udp_port = htons(port),
 362                 .use_udp_checksums = true
 363         };
 364 #if IS_ENABLED(CONFIG_IPV6)
 365         int retries = 0;
 366         struct udp_port_cfg port6 = {
 367                 .family = AF_INET6,
 368                 .local_ip6 = IN6ADDR_ANY_INIT,
 369                 .use_udp6_tx_checksums = true,
 370                 .use_udp6_rx_checksums = true,
 371                 .ipv6_v6only = true
 372         };
 373 #endif
 374
 375         rcu_read_lock();
 376         net = rcu_dereference(wg->creating_net);
 377         net = net ? maybe_get_net(net) : NULL;
 378         rcu_read_unlock();
 379         if (unlikely(!net))
 380                 return -ENONET;
 381
 382 #if IS_ENABLED(CONFIG_IPV6)
 383 retry:
 384 #endif
 385
 386         ret = udp_sock_create(net, &port4, &new4);
 387         if (ret < 0) {
 388                 pr_err("%s: Could not create IPv4 socket\n", wg->dev->name);
 389                 goto out;
 390         }
 391         set_sock_opts(new4);
 392         setup_udp_tunnel_sock(net, new4, &cfg);
 393
 394 #if IS_ENABLED(CONFIG_IPV6)
 395         if (ipv6_mod_enabled()) {
 396                 port6.local_udp_port = inet_sk(new4->sk)->inet_sport;
 397                 ret = udp_sock_create(net, &port6, &new6);
 398                 if (ret < 0) {
 399                         udp_tunnel_sock_release(new4);
 400                         if (ret == -EADDRINUSE && !port && retries++ < 100)
 401                                 goto retry;
 402                         pr_err("%s: Could not create IPv6 socket\n",
 403                                wg->dev->name);
 404                         goto out;
 405                 }
 406                 set_sock_opts(new6);
 407                 setup_udp_tunnel_sock(net, new6, &cfg);
 408         }
 409 #endif
 410
 411         wg_socket_reinit(wg, new4->sk, new6 ? new6->sk : NULL);
 412         ret = 0;
 413 out:
 414         put_net(net);
 415         return ret;
 416 }
 417
 418 void wg_socket_reinit(struct wg_device *wg, struct sock *new4,
 419                       struct sock *new6)
 420 {
 421         struct sock *old4, *old6;
 422
 423         mutex_lock(&wg->socket_update_lock);
 424         old4 = rcu_dereference_protected(wg->sock4,
 425                                 lockdep_is_held(&wg->socket_update_lock));
 426         old6 = rcu_dereference_protected(wg->sock6,
 427                                 lockdep_is_held(&wg->socket_update_lock));
 428         rcu_assign_pointer(wg->sock4, new4);
 429         rcu_assign_pointer(wg->sock6, new6);
 430         if (new4)
 431                 wg->incoming_port = ntohs(inet_sk(new4)->inet_sport);
 432         mutex_unlock(&wg->socket_update_lock);
 433         synchronize_rcu();
 434         sock_free(old4);
 435         sock_free(old6);
 436 }