drivers/net/wireguard/device.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 "queueing.h"
   7 #include "socket.h"
   8 #include "timers.h"
   9 #include "device.h"
  10 #include "ratelimiter.h"
  11 #include "peer.h"
  12 #include "messages.h"
  13
  14 #include <linux/module.h>
  15 #include <linux/rtnetlink.h>
  16 #include <linux/inet.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/inetdevice.h>
  19 #include <linux/if_arp.h>
  20 #include <linux/icmp.h>
  21 #include <linux/suspend.h>
  22 #include <net/dst_metadata.h>
  23 #include <net/icmp.h>
  24 #include <net/rtnetlink.h>
  25 #include <net/ip_tunnels.h>
  26 #include <net/addrconf.h>
  27
  28 static LIST_HEAD(device_list);
  29
  30 static int wg_open(struct net_device *dev)
  31 {
  32         struct in_device *dev_v4 = __in_dev_get_rtnl(dev);
  33         struct inet6_dev *dev_v6 = __in6_dev_get(dev);
  34         struct wg_device *wg = netdev_priv(dev);
  35         struct wg_peer *peer;
  36         int ret;
  37
  38         if (dev_v4) {
  39                 /* At some point we might put this check near the ip_rt_send_
  40                  * redirect call of ip_forward in net/ipv4/ip_forward.c, similar
  41                  * to the current secpath check.
  42                  */
  43                 IN_DEV_CONF_SET(dev_v4, SEND_REDIRECTS, false);
  44                 IPV4_DEVCONF_ALL(dev_net(dev), SEND_REDIRECTS) = false;
  45         }
  46         if (dev_v6)
  47                 dev_v6->cnf.addr_gen_mode = IN6_ADDR_GEN_MODE_NONE;
  48
  49         mutex_lock(&wg->device_update_lock);
  50         ret = wg_socket_init(wg, wg->incoming_port);
  51         if (ret < 0)
  52                 goto out;
  53         list_for_each_entry(peer, &wg->peer_list, peer_list) {
  54                 wg_packet_send_staged_packets(peer);
  55                 if (peer->persistent_keepalive_interval)
  56                         wg_packet_send_keepalive(peer);
  57         }
  58 out:
  59         mutex_unlock(&wg->device_update_lock);
  60         return ret;
  61 }
  62
  63 static int wg_pm_notification(struct notifier_block *nb, unsigned long action, void *data)
  64 {
  65         struct wg_device *wg;
  66         struct wg_peer *peer;
  67
  68         /* If the machine is constantly suspending and resuming, as part of
  69          * its normal operation rather than as a somewhat rare event, then we
  70          * don't actually want to clear keys.
  71          */
  72         if (IS_ENABLED(CONFIG_PM_AUTOSLEEP) || IS_ENABLED(CONFIG_ANDROID))
  73                 return 0;
  74
  75         if (action != PM_HIBERNATION_PREPARE && action != PM_SUSPEND_PREPARE)
  76                 return 0;
  77
  78         rtnl_lock();
  79         list_for_each_entry(wg, &device_list, device_list) {
  80                 mutex_lock(&wg->device_update_lock);
  81                 list_for_each_entry(peer, &wg->peer_list, peer_list) {
  82                         del_timer(&peer->timer_zero_key_material);
  83                         wg_noise_handshake_clear(&peer->handshake);
  84                         wg_noise_keypairs_clear(&peer->keypairs);
  85                 }
  86                 mutex_unlock(&wg->device_update_lock);
  87         }
  88         rtnl_unlock();
  89         rcu_barrier();
  90         return 0;
  91 }
  92
  93 static struct notifier_block pm_notifier = { .notifier_call = wg_pm_notification };
  94
  95 static int wg_vm_notification(struct notifier_block *nb, unsigned long action, void *data)
  96 {
  97         struct wg_device *wg;
  98         struct wg_peer *peer;
  99
 100         rtnl_lock();
 101         list_for_each_entry(wg, &device_list, device_list) {
 102                 mutex_lock(&wg->device_update_lock);
 103                 list_for_each_entry(peer, &wg->peer_list, peer_list)
 104                         wg_noise_expire_current_peer_keypairs(peer);
 105                 mutex_unlock(&wg->device_update_lock);
 106         }
 107         rtnl_unlock();
 108         return 0;
 109 }
 110
 111 static struct notifier_block vm_notifier = { .notifier_call = wg_vm_notification };
 112
 113 static int wg_stop(struct net_device *dev)
 114 {
 115         struct wg_device *wg = netdev_priv(dev);
 116         struct wg_peer *peer;
 117         struct sk_buff *skb;
 118
 119         mutex_lock(&wg->device_update_lock);
 120         list_for_each_entry(peer, &wg->peer_list, peer_list) {
 121                 wg_packet_purge_staged_packets(peer);
 122                 wg_timers_stop(peer);
 123                 wg_noise_handshake_clear(&peer->handshake);
 124                 wg_noise_keypairs_clear(&peer->keypairs);
 125                 wg_noise_reset_last_sent_handshake(&peer->last_sent_handshake);
 126         }
 127         mutex_unlock(&wg->device_update_lock);
 128         while ((skb = ptr_ring_consume(&wg->handshake_queue.ring)) != NULL)
 129                 kfree_skb(skb);
 130         atomic_set(&wg->handshake_queue_len, 0);
 131         wg_socket_reinit(wg, NULL, NULL);
 132         return 0;
 133 }
 134
 135 static netdev_tx_t wg_xmit(struct sk_buff *skb, struct net_device *dev)
 136 {
 137         struct wg_device *wg = netdev_priv(dev);
 138         struct sk_buff_head packets;
 139         struct wg_peer *peer;
 140         struct sk_buff *next;
 141         sa_family_t family;
 142         u32 mtu;
 143         int ret;
 144
 145         if (unlikely(!wg_check_packet_protocol(skb))) {
 146                 ret = -EPROTONOSUPPORT;
 147                 net_dbg_ratelimited("%s: Invalid IP packet\n", dev->name);
 148                 goto err;
 149         }
 150
 151         peer = wg_allowedips_lookup_dst(&wg->peer_allowedips, skb);
 152         if (unlikely(!peer)) {
 153                 ret = -ENOKEY;
 154                 if (skb->protocol == htons(ETH_P_IP))
 155                         net_dbg_ratelimited("%s: No peer has allowed IPs matching %pI4\n",
 156                                             dev->name, &ip_hdr(skb)->daddr);
 157                 else if (skb->protocol == htons(ETH_P_IPV6))
 158                         net_dbg_ratelimited("%s: No peer has allowed IPs matching %pI6\n",
 159                                             dev->name, &ipv6_hdr(skb)->daddr);
 160                 goto err_icmp;
 161         }
 162
 163         family = READ_ONCE(peer->endpoint.addr.sa_family);
 164         if (unlikely(family != AF_INET && family != AF_INET6)) {
 165                 ret = -EDESTADDRREQ;
 166                 net_dbg_ratelimited("%s: No valid endpoint has been configured or discovered for peer %llu\n",
 167                                     dev->name, peer->internal_id);
 168                 goto err_peer;
 169         }
 170
 171         mtu = skb_valid_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
 172
 173         __skb_queue_head_init(&packets);
 174         if (!skb_is_gso(skb)) {
 175                 skb_mark_not_on_list(skb);
 176         } else {
 177                 struct sk_buff *segs = skb_gso_segment(skb, 0);
 178
 179                 if (IS_ERR(segs)) {
 180                         ret = PTR_ERR(segs);
 181                         goto err_peer;
 182                 }
 183                 dev_kfree_skb(skb);
 184                 skb = segs;
 185         }
 186
 187         skb_list_walk_safe(skb, skb, next) {
 188                 skb_mark_not_on_list(skb);
 189
 190                 skb = skb_share_check(skb, GFP_ATOMIC);
 191                 if (unlikely(!skb))
 192                         continue;
 193
 194                 /* We only need to keep the original dst around for icmp,
 195                  * so at this point we're in a position to drop it.
 196                  */
 197                 skb_dst_drop(skb);
 198
 199                 PACKET_CB(skb)->mtu = mtu;
 200
 201                 __skb_queue_tail(&packets, skb);
 202         }
 203
 204         spin_lock_bh(&peer->staged_packet_queue.lock);
 205         /* If the queue is getting too big, we start removing the oldest packets
 206          * until it's small again. We do this before adding the new packet, so
 207          * we don't remove GSO segments that are in excess.
 208          */
 209         while (skb_queue_len(&peer->staged_packet_queue) > MAX_STAGED_PACKETS) {
 210                 dev_kfree_skb(__skb_dequeue(&peer->staged_packet_queue));
 211                 ++dev->stats.tx_dropped;
 212         }
 213         skb_queue_splice_tail(&packets, &peer->staged_packet_queue);
 214         spin_unlock_bh(&peer->staged_packet_queue.lock);
 215
 216         wg_packet_send_staged_packets(peer);
 217
 218         wg_peer_put(peer);
 219         return NETDEV_TX_OK;
 220
 221 err_peer:
 222         wg_peer_put(peer);
 223 err_icmp:
 224         if (skb->protocol == htons(ETH_P_IP))
 225                 icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
 226         else if (skb->protocol == htons(ETH_P_IPV6))
 227                 icmpv6_ndo_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
 228 err:
 229         ++dev->stats.tx_errors;
 230         kfree_skb(skb);
 231         return ret;
 232 }
 233
 234 static const struct net_device_ops netdev_ops = {
 235         .ndo_open               = wg_open,
 236         .ndo_stop               = wg_stop,
 237         .ndo_start_xmit         = wg_xmit,
 238         .ndo_get_stats64        = dev_get_tstats64
 239 };
 240
 241 static void wg_destruct(struct net_device *dev)
 242 {
 243         struct wg_device *wg = netdev_priv(dev);
 244
 245         rtnl_lock();
 246         list_del(&wg->device_list);
 247         rtnl_unlock();
 248         mutex_lock(&wg->device_update_lock);
 249         rcu_assign_pointer(wg->creating_net, NULL);
 250         wg->incoming_port = 0;
 251         wg_socket_reinit(wg, NULL, NULL);
 252         /* The final references are cleared in the below calls to destroy_workqueue. */
 253         wg_peer_remove_all(wg);
 254         destroy_workqueue(wg->handshake_receive_wq);
 255         destroy_workqueue(wg->handshake_send_wq);
 256         destroy_workqueue(wg->packet_crypt_wq);
 257         wg_packet_queue_free(&wg->handshake_queue, true);
 258         wg_packet_queue_free(&wg->decrypt_queue, false);
 259         wg_packet_queue_free(&wg->encrypt_queue, false);
 260         rcu_barrier(); /* Wait for all the peers to be actually freed. */
 261         wg_ratelimiter_uninit();
 262         memzero_explicit(&wg->static_identity, sizeof(wg->static_identity));
 263         free_percpu(dev->tstats);
 264         kvfree(wg->index_hashtable);
 265         kvfree(wg->peer_hashtable);
 266         mutex_unlock(&wg->device_update_lock);
 267
 268         pr_debug("%s: Interface destroyed\n", dev->name);
 269         free_netdev(dev);
 270 }
 271
 272 static const struct device_type device_type = { .name = KBUILD_MODNAME };
 273
 274 static void wg_setup(struct net_device *dev)
 275 {
 276         struct wg_device *wg = netdev_priv(dev);
 277         enum { WG_NETDEV_FEATURES = NETIF_F_HW_CSUM | NETIF_F_RXCSUM |
 278                                     NETIF_F_SG | NETIF_F_GSO |
 279                                     NETIF_F_GSO_SOFTWARE | NETIF_F_HIGHDMA };
 280         const int overhead = MESSAGE_MINIMUM_LENGTH + sizeof(struct udphdr) +
 281                              max(sizeof(struct ipv6hdr), sizeof(struct iphdr));
 282
 283         dev->netdev_ops = &netdev_ops;
 284         dev->header_ops = &ip_tunnel_header_ops;
 285         dev->hard_header_len = 0;
 286         dev->addr_len = 0;
 287         dev->needed_headroom = DATA_PACKET_HEAD_ROOM;
 288         dev->needed_tailroom = noise_encrypted_len(MESSAGE_PADDING_MULTIPLE);
 289         dev->type = ARPHRD_NONE;
 290         dev->flags = IFF_POINTOPOINT | IFF_NOARP;
 291         dev->priv_flags |= IFF_NO_QUEUE;
 292         dev->features |= NETIF_F_LLTX;
 293         dev->features |= WG_NETDEV_FEATURES;
 294         dev->hw_features |= WG_NETDEV_FEATURES;
 295         dev->hw_enc_features |= WG_NETDEV_FEATURES;
 296         dev->mtu = ETH_DATA_LEN - overhead;
 297         dev->max_mtu = round_down(INT_MAX, MESSAGE_PADDING_MULTIPLE) - overhead;
 298
 299         SET_NETDEV_DEVTYPE(dev, &device_type);
 300
 301         /* We need to keep the dst around in case of icmp replies. */
 302         netif_keep_dst(dev);
 303
 304         memset(wg, 0, sizeof(*wg));
 305         wg->dev = dev;
 306 }
 307
 308 static int wg_newlink(struct net *src_net, struct net_device *dev,
 309                       struct nlattr *tb[], struct nlattr *data[],
 310                       struct netlink_ext_ack *extack)
 311 {
 312         struct wg_device *wg = netdev_priv(dev);
 313         int ret = -ENOMEM;
 314
 315         rcu_assign_pointer(wg->creating_net, src_net);
 316         init_rwsem(&wg->static_identity.lock);
 317         mutex_init(&wg->socket_update_lock);
 318         mutex_init(&wg->device_update_lock);
 319         wg_allowedips_init(&wg->peer_allowedips);
 320         wg_cookie_checker_init(&wg->cookie_checker, wg);
 321         INIT_LIST_HEAD(&wg->peer_list);
 322         wg->device_update_gen = 1;
 323
 324         wg->peer_hashtable = wg_pubkey_hashtable_alloc();
 325         if (!wg->peer_hashtable)
 326                 return ret;
 327
 328         wg->index_hashtable = wg_index_hashtable_alloc();
 329         if (!wg->index_hashtable)
 330                 goto err_free_peer_hashtable;
 331
 332         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 333         if (!dev->tstats)
 334                 goto err_free_index_hashtable;
 335
 336         wg->handshake_receive_wq = alloc_workqueue("wg-kex-%s",
 337                         WQ_CPU_INTENSIVE | WQ_FREEZABLE, 0, dev->name);
 338         if (!wg->handshake_receive_wq)
 339                 goto err_free_tstats;
 340
 341         wg->handshake_send_wq = alloc_workqueue("wg-kex-%s",
 342                         WQ_UNBOUND | WQ_FREEZABLE, 0, dev->name);
 343         if (!wg->handshake_send_wq)
 344                 goto err_destroy_handshake_receive;
 345
 346         wg->packet_crypt_wq = alloc_workqueue("wg-crypt-%s",
 347                         WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 0, dev->name);
 348         if (!wg->packet_crypt_wq)
 349                 goto err_destroy_handshake_send;
 350
 351         ret = wg_packet_queue_init(&wg->encrypt_queue, wg_packet_encrypt_worker,
 352                                    MAX_QUEUED_PACKETS);
 353         if (ret < 0)
 354                 goto err_destroy_packet_crypt;
 355
 356         ret = wg_packet_queue_init(&wg->decrypt_queue, wg_packet_decrypt_worker,
 357                                    MAX_QUEUED_PACKETS);
 358         if (ret < 0)
 359                 goto err_free_encrypt_queue;
 360
 361         ret = wg_packet_queue_init(&wg->handshake_queue, wg_packet_handshake_receive_worker,
 362                                    MAX_QUEUED_INCOMING_HANDSHAKES);
 363         if (ret < 0)
 364                 goto err_free_decrypt_queue;
 365
 366         ret = wg_ratelimiter_init();
 367         if (ret < 0)
 368                 goto err_free_handshake_queue;
 369
 370         ret = register_netdevice(dev);
 371         if (ret < 0)
 372                 goto err_uninit_ratelimiter;
 373
 374         list_add(&wg->device_list, &device_list);
 375
 376         /* We wait until the end to assign priv_destructor, so that
 377          * register_netdevice doesn't call it for us if it fails.
 378          */
 379         dev->priv_destructor = wg_destruct;
 380
 381         pr_debug("%s: Interface created\n", dev->name);
 382         return ret;
 383
 384 err_uninit_ratelimiter:
 385         wg_ratelimiter_uninit();
 386 err_free_handshake_queue:
 387         wg_packet_queue_free(&wg->handshake_queue, false);
 388 err_free_decrypt_queue:
 389         wg_packet_queue_free(&wg->decrypt_queue, false);
 390 err_free_encrypt_queue:
 391         wg_packet_queue_free(&wg->encrypt_queue, false);
 392 err_destroy_packet_crypt:
 393         destroy_workqueue(wg->packet_crypt_wq);
 394 err_destroy_handshake_send:
 395         destroy_workqueue(wg->handshake_send_wq);
 396 err_destroy_handshake_receive:
 397         destroy_workqueue(wg->handshake_receive_wq);
 398 err_free_tstats:
 399         free_percpu(dev->tstats);
 400 err_free_index_hashtable:
 401         kvfree(wg->index_hashtable);
 402 err_free_peer_hashtable:
 403         kvfree(wg->peer_hashtable);
 404         return ret;
 405 }
 406
 407 static struct rtnl_link_ops link_ops __read_mostly = {
 408         .kind                   = KBUILD_MODNAME,
 409         .priv_size              = sizeof(struct wg_device),
 410         .setup                  = wg_setup,
 411         .newlink                = wg_newlink,
 412 };
 413
 414 static void wg_netns_pre_exit(struct net *net)
 415 {
 416         struct wg_device *wg;
 417         struct wg_peer *peer;
 418
 419         rtnl_lock();
 420         list_for_each_entry(wg, &device_list, device_list) {
 421                 if (rcu_access_pointer(wg->creating_net) == net) {
 422                         pr_debug("%s: Creating namespace exiting\n", wg->dev->name);
 423                         netif_carrier_off(wg->dev);
 424                         mutex_lock(&wg->device_update_lock);
 425                         rcu_assign_pointer(wg->creating_net, NULL);
 426                         wg_socket_reinit(wg, NULL, NULL);
 427                         list_for_each_entry(peer, &wg->peer_list, peer_list)
 428                                 wg_socket_clear_peer_endpoint_src(peer);
 429                         mutex_unlock(&wg->device_update_lock);
 430                 }
 431         }
 432         rtnl_unlock();
 433 }
 434
 435 static struct pernet_operations pernet_ops = {
 436         .pre_exit = wg_netns_pre_exit
 437 };
 438
 439 int __init wg_device_init(void)
 440 {
 441         int ret;
 442
 443         ret = register_pm_notifier(&pm_notifier);
 444         if (ret)
 445                 return ret;
 446
 447         ret = register_random_vmfork_notifier(&vm_notifier);
 448         if (ret)
 449                 goto error_pm;
 450
 451         ret = register_pernet_device(&pernet_ops);
 452         if (ret)
 453                 goto error_vm;
 454
 455         ret = rtnl_link_register(&link_ops);
 456         if (ret)
 457                 goto error_pernet;
 458
 459         return 0;
 460
 461 error_pernet:
 462         unregister_pernet_device(&pernet_ops);
 463 error_vm:
 464         unregister_random_vmfork_notifier(&vm_notifier);
 465 error_pm:
 466         unregister_pm_notifier(&pm_notifier);
 467         return ret;
 468 }
 469
 470 void wg_device_uninit(void)
 471 {
 472         rtnl_link_unregister(&link_ops);
 473         unregister_pernet_device(&pernet_ops);
 474         unregister_random_vmfork_notifier(&vm_notifier);
 475         unregister_pm_notifier(&pm_notifier);
 476         rcu_barrier();
 477 }