Quelle  receive.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0
/*
 * Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
 */

#include "queueing.h"
#include "device.h"
#include "peer.h"
#include "timers.h"
#include "messages.h"
#include "cookie.h"
#include "socket.h"

#include <linux/ip.h>
#include <linux/ipv6.h>
#include <linux/udp.h>
#include <net/ip_tunnels.h>

/* Must be called with bh disabled. */
static void update_rx_stats(struct wg_peer *peer, size_t len)
{
 dev_sw_netstats_rx_add(peer->device->dev, len);
 peer->rx_bytes += len;
}

#define SKB_TYPE_LE32(skb) (((struct message_header *)(skb)->data)->type)

static size_t validate_header_len(struct sk_buff *skb)
{
 if (unlikely(skb->len < sizeof(struct message_header)))
  return 0;
 if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_DATA) &&
     skb->len >= MESSAGE_MINIMUM_LENGTH)
  return sizeof(struct message_data);
 if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION) &&
     skb->len == sizeof(struct message_handshake_initiation))
  return sizeof(struct message_handshake_initiation);
 if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE) &&
     skb->len == sizeof(struct message_handshake_response))
  return sizeof(struct message_handshake_response);
 if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE) &&
     skb->len == sizeof(struct message_handshake_cookie))
  return sizeof(struct message_handshake_cookie);
 return 0;
}

static int prepare_skb_header(struct sk_buff *skb, struct wg_device *wg)
{
 size_t data_offset, data_len, header_len;
 struct udphdr *udp;

 if (unlikely(!wg_check_packet_protocol(skb) ||
       skb_transport_header(skb) < skb->head ||
       (skb_transport_header(skb) + sizeof(struct udphdr)) >
        skb_tail_pointer(skb)))
  return -EINVAL; /* Bogus IP header */
 udp = udp_hdr(skb);
 data_offset = (u8 *)udp - skb->data;
 if (unlikely(data_offset > U16_MAX ||
       data_offset + sizeof(struct udphdr) > skb->len))
  /* Packet has offset at impossible location or isn't big enough
 * to have UDP fields.
 */
  return -EINVAL;
 data_len = ntohs(udp->len);
 if (unlikely(data_len < sizeof(struct udphdr) ||
       data_len > skb->len - data_offset))
  /* UDP packet is reporting too small of a size or lying about
 * its size.
 */
  return -EINVAL;
 data_len -= sizeof(struct udphdr);
 data_offset = (u8 *)udp + sizeof(struct udphdr) - skb->data;
 if (unlikely(!pskb_may_pull(skb,
    data_offset + sizeof(struct message_header)) ||
       pskb_trim(skb, data_len + data_offset) < 0))
  return -EINVAL;
 skb_pull(skb, data_offset);
 if (unlikely(skb->len != data_len))
  /* Final len does not agree with calculated len */
  return -EINVAL;
 header_len = validate_header_len(skb);
 if (unlikely(!header_len))
  return -EINVAL;
 __skb_push(skb, data_offset);
 if (unlikely(!pskb_may_pull(skb, data_offset + header_len)))
  return -EINVAL;
 __skb_pull(skb, data_offset);
 return 0;
}

static void wg_receive_handshake_packet(struct wg_device *wg,
     struct sk_buff *skb)
{
 enum cookie_mac_state mac_state;
 struct wg_peer *peer = NULL;
 /* This is global, so that our load calculation applies to the whole
 * system. We don't care about races with it at all.
 */
 static u64 last_under_load;
 bool packet_needs_cookie;
 bool under_load;

 if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE)) {
  net_dbg_skb_ratelimited("%s: Receiving cookie response from %pISpfsc\n",
     wg->dev->name, skb);
  wg_cookie_message_consume(
   (struct message_handshake_cookie *)skb->data, wg);
  return;
 }

 under_load = atomic_read(&wg->handshake_queue_len) >=
   MAX_QUEUED_INCOMING_HANDSHAKES / 8;
 if (under_load) {
  last_under_load = ktime_get_coarse_boottime_ns();
 } else if (last_under_load) {
  under_load = !wg_birthdate_has_expired(last_under_load, 1);
  if (!under_load)
   last_under_load = 0;
 }
 mac_state = wg_cookie_validate_packet(&wg->cookie_checker, skb,
           under_load);
 if ((under_load && mac_state == VALID_MAC_WITH_COOKIE) ||
     (!under_load && mac_state == VALID_MAC_BUT_NO_COOKIE)) {
  packet_needs_cookie = false;
 } else if (under_load && mac_state == VALID_MAC_BUT_NO_COOKIE) {
  packet_needs_cookie = true;
 } else {
  net_dbg_skb_ratelimited("%s: Invalid MAC of handshake, dropping packet from %pISpfsc\n",
     wg->dev->name, skb);
  return;
 }

 switch (SKB_TYPE_LE32(skb)) {
 case cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION): {
  struct message_handshake_initiation *message =
   (struct message_handshake_initiation *)skb->data;

  if (packet_needs_cookie) {
   wg_packet_send_handshake_cookie(wg, skb,
       message->sender_index);
   return;
  }
  peer = wg_noise_handshake_consume_initiation(message, wg);
  if (unlikely(!peer)) {
   net_dbg_skb_ratelimited("%s: Invalid handshake initiation from %pISpfsc\n",
      wg->dev->name, skb);
   return;
  }
  wg_socket_set_peer_endpoint_from_skb(peer, skb);
  net_dbg_ratelimited("%s: Receiving handshake initiation from peer %llu (%pISpfsc)\n",
        wg->dev->name, peer->internal_id,
        &peer->endpoint.addr);
  wg_packet_send_handshake_response(peer);
  break;
 }
 case cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE): {
  struct message_handshake_response *message =
   (struct message_handshake_response *)skb->data;

  if (packet_needs_cookie) {
   wg_packet_send_handshake_cookie(wg, skb,
       message->sender_index);
   return;
  }
  peer = wg_noise_handshake_consume_response(message, wg);
  if (unlikely(!peer)) {
   net_dbg_skb_ratelimited("%s: Invalid handshake response from %pISpfsc\n",
      wg->dev->name, skb);
   return;
  }
  wg_socket_set_peer_endpoint_from_skb(peer, skb);
  net_dbg_ratelimited("%s: Receiving handshake response from peer %llu (%pISpfsc)\n",
        wg->dev->name, peer->internal_id,
        &peer->endpoint.addr);
  if (wg_noise_handshake_begin_session(&peer->handshake,
           &peer->keypairs)) {
   wg_timers_session_derived(peer);
   wg_timers_handshake_complete(peer);
   /* Calling this function will either send any existing
 * packets in the queue and not send a keepalive, which
 * is the best case, Or, if there's nothing in the
 * queue, it will send a keepalive, in order to give
 * immediate confirmation of the session.
 */
   wg_packet_send_keepalive(peer);
  }
  break;
 }
 }

 if (unlikely(!peer)) {
  WARN(1, "Somehow a wrong type of packet wound up in the handshake queue!\n");
  return;
 }

 local_bh_disable();
 update_rx_stats(peer, skb->len);
 local_bh_enable();

 wg_timers_any_authenticated_packet_received(peer);
 wg_timers_any_authenticated_packet_traversal(peer);
 wg_peer_put(peer);
}

void wg_packet_handshake_receive_worker(struct work_struct *work)
{
 struct crypt_queue *queue = container_of(work, struct multicore_worker, work)->ptr;
 struct wg_device *wg = container_of(queue, struct wg_device, handshake_queue);
 struct sk_buff *skb;

 while ((skb = ptr_ring_consume_bh(&queue->ring)) != NULL) {
  wg_receive_handshake_packet(wg, skb);
  dev_kfree_skb(skb);
  atomic_dec(&wg->handshake_queue_len);
  cond_resched();
 }
}

static void keep_key_fresh(struct wg_peer *peer)
{
 struct noise_keypair *keypair;
 bool send;

 if (peer->sent_lastminute_handshake)
  return;

 rcu_read_lock_bh();
 keypair = rcu_dereference_bh(peer->keypairs.current_keypair);
 send = keypair && READ_ONCE(keypair->sending.is_valid) &&
        keypair->i_am_the_initiator &&
        wg_birthdate_has_expired(keypair->sending.birthdate,
   REJECT_AFTER_TIME - KEEPALIVE_TIMEOUT - REKEY_TIMEOUT);
 rcu_read_unlock_bh();

 if (unlikely(send)) {
  peer->sent_lastminute_handshake = true;
  wg_packet_send_queued_handshake_initiation(peer, false);
 }
}

static bool decrypt_packet(struct sk_buff *skb, struct noise_keypair *keypair)
{
 struct scatterlist sg[MAX_SKB_FRAGS + 8];
 struct sk_buff *trailer;
 unsigned int offset;
 int num_frags;

 if (unlikely(!keypair))
  return false;

 if (unlikely(!READ_ONCE(keypair->receiving.is_valid) ||
    wg_birthdate_has_expired(keypair->receiving.birthdate, REJECT_AFTER_TIME) ||
    READ_ONCE(keypair->receiving_counter.counter) >= REJECT_AFTER_MESSAGES)) {
  WRITE_ONCE(keypair->receiving.is_valid, false);
  return false;
 }

 PACKET_CB(skb)->nonce =
  le64_to_cpu(((struct message_data *)skb->data)->counter);

 /* We ensure that the network header is part of the packet before we
 * call skb_cow_data, so that there's no chance that data is removed
 * from the skb, so that later we can extract the original endpoint.
 */
 offset = -skb_network_offset(skb);
 skb_push(skb, offset);
 num_frags = skb_cow_data(skb, 0, &trailer);
 offset += sizeof(struct message_data);
 skb_pull(skb, offset);
 if (unlikely(num_frags < 0 || num_frags > ARRAY_SIZE(sg)))
  return false;

 sg_init_table(sg, num_frags);
 if (skb_to_sgvec(skb, sg, 0, skb->len) <= 0)
  return false;

 if (!chacha20poly1305_decrypt_sg_inplace(sg, skb->len, NULL, 0,
              PACKET_CB(skb)->nonce,
       keypair->receiving.key))
  return false;

 /* Another ugly situation of pushing and pulling the header so as to
 * keep endpoint information intact.
 */
 skb_push(skb, offset);
 if (pskb_trim(skb, skb->len - noise_encrypted_len(0)))
  return false;
 skb_pull(skb, offset);

 return true;
}

/* This is RFC6479, a replay detection bitmap algorithm that avoids bitshifts */
static bool counter_validate(struct noise_replay_counter *counter, u64 their_counter)
{
 unsigned long index, index_current, top, i;
 bool ret = false;

 spin_lock_bh(&counter->lock);

 if (unlikely(counter->counter >= REJECT_AFTER_MESSAGES + 1 ||
       their_counter >= REJECT_AFTER_MESSAGES))
  goto out;

 ++their_counter;

 if (unlikely((COUNTER_WINDOW_SIZE + their_counter) <
       counter->counter))
  goto out;

 index = their_counter >> ilog2(BITS_PER_LONG);

 if (likely(their_counter > counter->counter)) {
  index_current = counter->counter >> ilog2(BITS_PER_LONG);
  top = min_t(unsigned long, index - index_current,
       COUNTER_BITS_TOTAL / BITS_PER_LONG);
  for (i = 1; i <= top; ++i)
   counter->backtrack[(i + index_current) &
    ((COUNTER_BITS_TOTAL / BITS_PER_LONG) - 1)] = 0;
  WRITE_ONCE(counter->counter, their_counter);
 }

 index &= (COUNTER_BITS_TOTAL / BITS_PER_LONG) - 1;
 ret = !test_and_set_bit(their_counter & (BITS_PER_LONG - 1),
    &counter->backtrack[index]);

out:
 spin_unlock_bh(&counter->lock);
 return ret;
}

#include "selftest/counter.c"

static void wg_packet_consume_data_done(struct wg_peer *peer,
     struct sk_buff *skb,
     struct endpoint *endpoint)
{
 struct net_device *dev = peer->device->dev;
 unsigned int len, len_before_trim;
 struct wg_peer *routed_peer;

 wg_socket_set_peer_endpoint(peer, endpoint);

 if (unlikely(wg_noise_received_with_keypair(&peer->keypairs,
          PACKET_CB(skb)->keypair))) {
  wg_timers_handshake_complete(peer);
  wg_packet_send_staged_packets(peer);
 }

 keep_key_fresh(peer);

 wg_timers_any_authenticated_packet_received(peer);
 wg_timers_any_authenticated_packet_traversal(peer);

 /* A packet with length 0 is a keepalive packet */
 if (unlikely(!skb->len)) {
  update_rx_stats(peer, message_data_len(0));
  net_dbg_ratelimited("%s: Receiving keepalive packet from peer %llu (%pISpfsc)\n",
        dev->name, peer->internal_id,
        &peer->endpoint.addr);
  goto packet_processed;
 }

 wg_timers_data_received(peer);

 if (unlikely(skb_network_header(skb) < skb->head))
  goto dishonest_packet_size;
 if (unlikely(!(pskb_network_may_pull(skb, sizeof(struct iphdr)) &&
         (ip_hdr(skb)->version == 4 ||
   (ip_hdr(skb)->version == 6 &&
    pskb_network_may_pull(skb, sizeof(struct ipv6hdr)))))))
  goto dishonest_packet_type;

 skb->dev = dev;
 /* We've already verified the Poly1305 auth tag, which means this packet
 * was not modified in transit. We can therefore tell the networking
 * stack that all checksums of every layer of encapsulation have already
 * been checked "by the hardware" and therefore is unnecessary to check
 * again in software.
 */
 skb->ip_summed = CHECKSUM_UNNECESSARY;
 skb->csum_level = ~0; /* All levels */
 skb->protocol = ip_tunnel_parse_protocol(skb);
 if (skb->protocol == htons(ETH_P_IP)) {
  len = ntohs(ip_hdr(skb)->tot_len);
  if (unlikely(len < sizeof(struct iphdr)))
   goto dishonest_packet_size;
  INET_ECN_decapsulate(skb, PACKET_CB(skb)->ds, ip_hdr(skb)->tos);
 } else if (skb->protocol == htons(ETH_P_IPV6)) {
  len = ntohs(ipv6_hdr(skb)->payload_len) +
        sizeof(struct ipv6hdr);
  INET_ECN_decapsulate(skb, PACKET_CB(skb)->ds, ipv6_get_dsfield(ipv6_hdr(skb)));
 } else {
  goto dishonest_packet_type;
 }

 if (unlikely(len > skb->len))
  goto dishonest_packet_size;
 len_before_trim = skb->len;
 if (unlikely(pskb_trim(skb, len)))
  goto packet_processed;

 routed_peer = wg_allowedips_lookup_src(&peer->device->peer_allowedips,
            skb);
 wg_peer_put(routed_peer); /* We don't need the extra reference. */

 if (unlikely(routed_peer != peer))
  goto dishonest_packet_peer;

 napi_gro_receive(&peer->napi, skb);
 update_rx_stats(peer, message_data_len(len_before_trim));
 return;

dishonest_packet_peer:
 net_dbg_skb_ratelimited("%s: Packet has unallowed src IP (%pISc) from peer %llu (%pISpfsc)\n",
    dev->name, skb, peer->internal_id,
    &peer->endpoint.addr);
 DEV_STATS_INC(dev, rx_errors);
 DEV_STATS_INC(dev, rx_frame_errors);
 goto packet_processed;
dishonest_packet_type:
 net_dbg_ratelimited("%s: Packet is neither ipv4 nor ipv6 from peer %llu (%pISpfsc)\n",
       dev->name, peer->internal_id, &peer->endpoint.addr);
 DEV_STATS_INC(dev, rx_errors);
 DEV_STATS_INC(dev, rx_frame_errors);
 goto packet_processed;
dishonest_packet_size:
 net_dbg_ratelimited("%s: Packet has incorrect size from peer %llu (%pISpfsc)\n",
       dev->name, peer->internal_id, &peer->endpoint.addr);
 DEV_STATS_INC(dev, rx_errors);
 DEV_STATS_INC(dev, rx_length_errors);
 goto packet_processed;
packet_processed:
 dev_kfree_skb(skb);
}

int wg_packet_rx_poll(struct napi_struct *napi, int budget)
{
 struct wg_peer *peer = container_of(napi, struct wg_peer, napi);
 struct noise_keypair *keypair;
 struct endpoint endpoint;
 enum packet_state state;
 struct sk_buff *skb;
 int work_done = 0;
 bool free;

 if (unlikely(budget <= 0))
  return 0;

 while ((skb = wg_prev_queue_peek(&peer->rx_queue)) != NULL &&
        (state = atomic_read_acquire(&PACKET_CB(skb)->state)) !=
         PACKET_STATE_UNCRYPTED) {
  wg_prev_queue_drop_peeked(&peer->rx_queue);
  keypair = PACKET_CB(skb)->keypair;
  free = true;

  if (unlikely(state != PACKET_STATE_CRYPTED))
   goto next;

  if (unlikely(!counter_validate(&keypair->receiving_counter,
            PACKET_CB(skb)->nonce))) {
   net_dbg_ratelimited("%s: Packet has invalid nonce %llu (max %llu)\n",
         peer->device->dev->name,
         PACKET_CB(skb)->nonce,
         READ_ONCE(keypair->receiving_counter.counter));
   goto next;
  }

  if (unlikely(wg_socket_endpoint_from_skb(&endpoint, skb)))
   goto next;

  wg_reset_packet(skb, false);
  wg_packet_consume_data_done(peer, skb, &endpoint);
  free = false;

next:
  wg_noise_keypair_put(keypair, false);
  wg_peer_put(peer);
  if (unlikely(free))
   dev_kfree_skb(skb);

  if (++work_done >= budget)
   break;
 }

 if (work_done < budget)
  napi_complete_done(napi, work_done);

 return work_done;
}

void wg_packet_decrypt_worker(struct work_struct *work)
{
 struct crypt_queue *queue = container_of(work, struct multicore_worker,
       work)->ptr;
 struct sk_buff *skb;

 while ((skb = ptr_ring_consume_bh(&queue->ring)) != NULL) {
  enum packet_state state =
   likely(decrypt_packet(skb, PACKET_CB(skb)->keypair)) ?
    PACKET_STATE_CRYPTED : PACKET_STATE_DEAD;
  wg_queue_enqueue_per_peer_rx(skb, state);
  if (need_resched())
   cond_resched();
 }
}

static void wg_packet_consume_data(struct wg_device *wg, struct sk_buff *skb)
{
 __le32 idx = ((struct message_data *)skb->data)->key_idx;
 struct wg_peer *peer = NULL;
 int ret;

 rcu_read_lock_bh();
 PACKET_CB(skb)->keypair =
  (struct noise_keypair *)wg_index_hashtable_lookup(
   wg->index_hashtable, INDEX_HASHTABLE_KEYPAIR, idx,
   &peer);
 if (unlikely(!wg_noise_keypair_get(PACKET_CB(skb)->keypair)))
  goto err_keypair;

 if (unlikely(READ_ONCE(peer->is_dead)))
  goto err;

 ret = wg_queue_enqueue_per_device_and_peer(&wg->decrypt_queue, &peer->rx_queue, skb,
         wg->packet_crypt_wq);
 if (unlikely(ret == -EPIPE))
  wg_queue_enqueue_per_peer_rx(skb, PACKET_STATE_DEAD);
 if (likely(!ret || ret == -EPIPE)) {
  rcu_read_unlock_bh();
  return;
 }
err:
 wg_noise_keypair_put(PACKET_CB(skb)->keypair, false);
err_keypair:
 rcu_read_unlock_bh();
 wg_peer_put(peer);
 dev_kfree_skb(skb);
}

void wg_packet_receive(struct wg_device *wg, struct sk_buff *skb)
{
 if (unlikely(prepare_skb_header(skb, wg) < 0))
  goto err;
 switch (SKB_TYPE_LE32(skb)) {
 case cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION):
 case cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE):
 case cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE): {
  int cpu, ret = -EBUSY;

  if (unlikely(!rng_is_initialized()))
   goto drop;
  if (atomic_read(&wg->handshake_queue_len) > MAX_QUEUED_INCOMING_HANDSHAKES / 2) {
   if (spin_trylock_bh(&wg->handshake_queue.ring.producer_lock)) {
    ret = __ptr_ring_produce(&wg->handshake_queue.ring, skb);
    spin_unlock_bh(&wg->handshake_queue.ring.producer_lock);
   }
  } else
   ret = ptr_ring_produce_bh(&wg->handshake_queue.ring, skb);
  if (ret) {
 drop:
   net_dbg_skb_ratelimited("%s: Dropping handshake packet from %pISpfsc\n",
      wg->dev->name, skb);
   goto err;
  }
  atomic_inc(&wg->handshake_queue_len);
  cpu = wg_cpumask_next_online(&wg->handshake_queue.last_cpu);
  /* Queues up a call to packet_process_queued_handshake_packets(skb): */
  queue_work_on(cpu, wg->handshake_receive_wq,
         &per_cpu_ptr(wg->handshake_queue.worker, cpu)->work);
  break;
 }
 case cpu_to_le32(MESSAGE_DATA):
  PACKET_CB(skb)->ds = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
  wg_packet_consume_data(wg, skb);
  break;
 default:
  WARN(1, "Non-exhaustive parsing of packet header lead to unknown packet type!\n");
  goto err;
 }
 return;

err:
 dev_kfree_skb(skb);
}