mirror of
https://github.com/torvalds/linux.git
synced 2024-12-26 12:52:30 +00:00
8b5553ace8
Having two ring buffers per-peer means that every peer results in two
massive ring allocations. On an 8-core x86_64 machine, this commit
reduces the per-peer allocation from 18,688 bytes to 1,856 bytes, which
is an 90% reduction. Ninety percent! With some single-machine
deployments approaching 500,000 peers, we're talking about a reduction
from 7 gigs of memory down to 700 megs of memory.
In order to get rid of these per-peer allocations, this commit switches
to using a list-based queueing approach. Currently GSO fragments are
chained together using the skb->next pointer (the skb_list_* singly
linked list approach), so we form the per-peer queue around the unused
skb->prev pointer (which sort of makes sense because the links are
pointing backwards). Use of skb_queue_* is not possible here, because
that is based on doubly linked lists and spinlocks. Multiple cores can
write into the queue at any given time, because its writes occur in the
start_xmit path or in the udp_recv path. But reads happen in a single
workqueue item per-peer, amounting to a multi-producer, single-consumer
paradigm.
The MPSC queue is implemented locklessly and never blocks. However, it
is not linearizable (though it is serializable), with a very tight and
unlikely race on writes, which, when hit (some tiny fraction of the
0.15% of partial adds on a fully loaded 16-core x86_64 system), causes
the queue reader to terminate early. However, because every packet sent
queues up the same workqueue item after it is fully added, the worker
resumes again, and stopping early isn't actually a problem, since at
that point the packet wouldn't have yet been added to the encryption
queue. These properties allow us to avoid disabling interrupts or
spinning. The design is based on Dmitry Vyukov's algorithm [1].
Performance-wise, ordinarily list-based queues aren't preferable to
ringbuffers, because of cache misses when following pointers around.
However, we *already* have to follow the adjacent pointers when working
through fragments, so there shouldn't actually be any change there. A
potential downside is that dequeueing is a bit more complicated, but the
ptr_ring structure used prior had a spinlock when dequeueing, so all and
all the difference appears to be a wash.
Actually, from profiling, the biggest performance hit, by far, of this
commit winds up being atomic_add_unless(count, 1, max) and atomic_
dec(count), which account for the majority of CPU time, according to
perf. In that sense, the previous ring buffer was superior in that it
could check if it was full by head==tail, which the list-based approach
cannot do.
But all and all, this enables us to get massive memory savings, allowing
WireGuard to scale for real world deployments, without taking much of a
performance hit.
[1] http://www.1024cores.net/home/lock-free-algorithms/queues/intrusive-mpsc-node-based-queue
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Fixes: e7096c131e
("net: WireGuard secure network tunnel")
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
414 lines
13 KiB
C
414 lines
13 KiB
C
// SPDX-License-Identifier: GPL-2.0
|
|
/*
|
|
* Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
|
|
*/
|
|
|
|
#include "queueing.h"
|
|
#include "timers.h"
|
|
#include "device.h"
|
|
#include "peer.h"
|
|
#include "socket.h"
|
|
#include "messages.h"
|
|
#include "cookie.h"
|
|
|
|
#include <linux/uio.h>
|
|
#include <linux/inetdevice.h>
|
|
#include <linux/socket.h>
|
|
#include <net/ip_tunnels.h>
|
|
#include <net/udp.h>
|
|
#include <net/sock.h>
|
|
|
|
static void wg_packet_send_handshake_initiation(struct wg_peer *peer)
|
|
{
|
|
struct message_handshake_initiation packet;
|
|
|
|
if (!wg_birthdate_has_expired(atomic64_read(&peer->last_sent_handshake),
|
|
REKEY_TIMEOUT))
|
|
return; /* This function is rate limited. */
|
|
|
|
atomic64_set(&peer->last_sent_handshake, ktime_get_coarse_boottime_ns());
|
|
net_dbg_ratelimited("%s: Sending handshake initiation to peer %llu (%pISpfsc)\n",
|
|
peer->device->dev->name, peer->internal_id,
|
|
&peer->endpoint.addr);
|
|
|
|
if (wg_noise_handshake_create_initiation(&packet, &peer->handshake)) {
|
|
wg_cookie_add_mac_to_packet(&packet, sizeof(packet), peer);
|
|
wg_timers_any_authenticated_packet_traversal(peer);
|
|
wg_timers_any_authenticated_packet_sent(peer);
|
|
atomic64_set(&peer->last_sent_handshake,
|
|
ktime_get_coarse_boottime_ns());
|
|
wg_socket_send_buffer_to_peer(peer, &packet, sizeof(packet),
|
|
HANDSHAKE_DSCP);
|
|
wg_timers_handshake_initiated(peer);
|
|
}
|
|
}
|
|
|
|
void wg_packet_handshake_send_worker(struct work_struct *work)
|
|
{
|
|
struct wg_peer *peer = container_of(work, struct wg_peer,
|
|
transmit_handshake_work);
|
|
|
|
wg_packet_send_handshake_initiation(peer);
|
|
wg_peer_put(peer);
|
|
}
|
|
|
|
void wg_packet_send_queued_handshake_initiation(struct wg_peer *peer,
|
|
bool is_retry)
|
|
{
|
|
if (!is_retry)
|
|
peer->timer_handshake_attempts = 0;
|
|
|
|
rcu_read_lock_bh();
|
|
/* We check last_sent_handshake here in addition to the actual function
|
|
* we're queueing up, so that we don't queue things if not strictly
|
|
* necessary:
|
|
*/
|
|
if (!wg_birthdate_has_expired(atomic64_read(&peer->last_sent_handshake),
|
|
REKEY_TIMEOUT) ||
|
|
unlikely(READ_ONCE(peer->is_dead)))
|
|
goto out;
|
|
|
|
wg_peer_get(peer);
|
|
/* Queues up calling packet_send_queued_handshakes(peer), where we do a
|
|
* peer_put(peer) after:
|
|
*/
|
|
if (!queue_work(peer->device->handshake_send_wq,
|
|
&peer->transmit_handshake_work))
|
|
/* If the work was already queued, we want to drop the
|
|
* extra reference:
|
|
*/
|
|
wg_peer_put(peer);
|
|
out:
|
|
rcu_read_unlock_bh();
|
|
}
|
|
|
|
void wg_packet_send_handshake_response(struct wg_peer *peer)
|
|
{
|
|
struct message_handshake_response packet;
|
|
|
|
atomic64_set(&peer->last_sent_handshake, ktime_get_coarse_boottime_ns());
|
|
net_dbg_ratelimited("%s: Sending handshake response to peer %llu (%pISpfsc)\n",
|
|
peer->device->dev->name, peer->internal_id,
|
|
&peer->endpoint.addr);
|
|
|
|
if (wg_noise_handshake_create_response(&packet, &peer->handshake)) {
|
|
wg_cookie_add_mac_to_packet(&packet, sizeof(packet), peer);
|
|
if (wg_noise_handshake_begin_session(&peer->handshake,
|
|
&peer->keypairs)) {
|
|
wg_timers_session_derived(peer);
|
|
wg_timers_any_authenticated_packet_traversal(peer);
|
|
wg_timers_any_authenticated_packet_sent(peer);
|
|
atomic64_set(&peer->last_sent_handshake,
|
|
ktime_get_coarse_boottime_ns());
|
|
wg_socket_send_buffer_to_peer(peer, &packet,
|
|
sizeof(packet),
|
|
HANDSHAKE_DSCP);
|
|
}
|
|
}
|
|
}
|
|
|
|
void wg_packet_send_handshake_cookie(struct wg_device *wg,
|
|
struct sk_buff *initiating_skb,
|
|
__le32 sender_index)
|
|
{
|
|
struct message_handshake_cookie packet;
|
|
|
|
net_dbg_skb_ratelimited("%s: Sending cookie response for denied handshake message for %pISpfsc\n",
|
|
wg->dev->name, initiating_skb);
|
|
wg_cookie_message_create(&packet, initiating_skb, sender_index,
|
|
&wg->cookie_checker);
|
|
wg_socket_send_buffer_as_reply_to_skb(wg, initiating_skb, &packet,
|
|
sizeof(packet));
|
|
}
|
|
|
|
static void keep_key_fresh(struct wg_peer *peer)
|
|
{
|
|
struct noise_keypair *keypair;
|
|
bool send;
|
|
|
|
rcu_read_lock_bh();
|
|
keypair = rcu_dereference_bh(peer->keypairs.current_keypair);
|
|
send = keypair && READ_ONCE(keypair->sending.is_valid) &&
|
|
(atomic64_read(&keypair->sending_counter) > REKEY_AFTER_MESSAGES ||
|
|
(keypair->i_am_the_initiator &&
|
|
wg_birthdate_has_expired(keypair->sending.birthdate, REKEY_AFTER_TIME)));
|
|
rcu_read_unlock_bh();
|
|
|
|
if (unlikely(send))
|
|
wg_packet_send_queued_handshake_initiation(peer, false);
|
|
}
|
|
|
|
static unsigned int calculate_skb_padding(struct sk_buff *skb)
|
|
{
|
|
unsigned int padded_size, last_unit = skb->len;
|
|
|
|
if (unlikely(!PACKET_CB(skb)->mtu))
|
|
return ALIGN(last_unit, MESSAGE_PADDING_MULTIPLE) - last_unit;
|
|
|
|
/* We do this modulo business with the MTU, just in case the networking
|
|
* layer gives us a packet that's bigger than the MTU. In that case, we
|
|
* wouldn't want the final subtraction to overflow in the case of the
|
|
* padded_size being clamped. Fortunately, that's very rarely the case,
|
|
* so we optimize for that not happening.
|
|
*/
|
|
if (unlikely(last_unit > PACKET_CB(skb)->mtu))
|
|
last_unit %= PACKET_CB(skb)->mtu;
|
|
|
|
padded_size = min(PACKET_CB(skb)->mtu,
|
|
ALIGN(last_unit, MESSAGE_PADDING_MULTIPLE));
|
|
return padded_size - last_unit;
|
|
}
|
|
|
|
static bool encrypt_packet(struct sk_buff *skb, struct noise_keypair *keypair)
|
|
{
|
|
unsigned int padding_len, plaintext_len, trailer_len;
|
|
struct scatterlist sg[MAX_SKB_FRAGS + 8];
|
|
struct message_data *header;
|
|
struct sk_buff *trailer;
|
|
int num_frags;
|
|
|
|
/* Force hash calculation before encryption so that flow analysis is
|
|
* consistent over the inner packet.
|
|
*/
|
|
skb_get_hash(skb);
|
|
|
|
/* Calculate lengths. */
|
|
padding_len = calculate_skb_padding(skb);
|
|
trailer_len = padding_len + noise_encrypted_len(0);
|
|
plaintext_len = skb->len + padding_len;
|
|
|
|
/* Expand data section to have room for padding and auth tag. */
|
|
num_frags = skb_cow_data(skb, trailer_len, &trailer);
|
|
if (unlikely(num_frags < 0 || num_frags > ARRAY_SIZE(sg)))
|
|
return false;
|
|
|
|
/* Set the padding to zeros, and make sure it and the auth tag are part
|
|
* of the skb.
|
|
*/
|
|
memset(skb_tail_pointer(trailer), 0, padding_len);
|
|
|
|
/* Expand head section to have room for our header and the network
|
|
* stack's headers.
|
|
*/
|
|
if (unlikely(skb_cow_head(skb, DATA_PACKET_HEAD_ROOM) < 0))
|
|
return false;
|
|
|
|
/* Finalize checksum calculation for the inner packet, if required. */
|
|
if (unlikely(skb->ip_summed == CHECKSUM_PARTIAL &&
|
|
skb_checksum_help(skb)))
|
|
return false;
|
|
|
|
/* Only after checksumming can we safely add on the padding at the end
|
|
* and the header.
|
|
*/
|
|
skb_set_inner_network_header(skb, 0);
|
|
header = (struct message_data *)skb_push(skb, sizeof(*header));
|
|
header->header.type = cpu_to_le32(MESSAGE_DATA);
|
|
header->key_idx = keypair->remote_index;
|
|
header->counter = cpu_to_le64(PACKET_CB(skb)->nonce);
|
|
pskb_put(skb, trailer, trailer_len);
|
|
|
|
/* Now we can encrypt the scattergather segments */
|
|
sg_init_table(sg, num_frags);
|
|
if (skb_to_sgvec(skb, sg, sizeof(struct message_data),
|
|
noise_encrypted_len(plaintext_len)) <= 0)
|
|
return false;
|
|
return chacha20poly1305_encrypt_sg_inplace(sg, plaintext_len, NULL, 0,
|
|
PACKET_CB(skb)->nonce,
|
|
keypair->sending.key);
|
|
}
|
|
|
|
void wg_packet_send_keepalive(struct wg_peer *peer)
|
|
{
|
|
struct sk_buff *skb;
|
|
|
|
if (skb_queue_empty(&peer->staged_packet_queue)) {
|
|
skb = alloc_skb(DATA_PACKET_HEAD_ROOM + MESSAGE_MINIMUM_LENGTH,
|
|
GFP_ATOMIC);
|
|
if (unlikely(!skb))
|
|
return;
|
|
skb_reserve(skb, DATA_PACKET_HEAD_ROOM);
|
|
skb->dev = peer->device->dev;
|
|
PACKET_CB(skb)->mtu = skb->dev->mtu;
|
|
skb_queue_tail(&peer->staged_packet_queue, skb);
|
|
net_dbg_ratelimited("%s: Sending keepalive packet to peer %llu (%pISpfsc)\n",
|
|
peer->device->dev->name, peer->internal_id,
|
|
&peer->endpoint.addr);
|
|
}
|
|
|
|
wg_packet_send_staged_packets(peer);
|
|
}
|
|
|
|
static void wg_packet_create_data_done(struct wg_peer *peer, struct sk_buff *first)
|
|
{
|
|
struct sk_buff *skb, *next;
|
|
bool is_keepalive, data_sent = false;
|
|
|
|
wg_timers_any_authenticated_packet_traversal(peer);
|
|
wg_timers_any_authenticated_packet_sent(peer);
|
|
skb_list_walk_safe(first, skb, next) {
|
|
is_keepalive = skb->len == message_data_len(0);
|
|
if (likely(!wg_socket_send_skb_to_peer(peer, skb,
|
|
PACKET_CB(skb)->ds) && !is_keepalive))
|
|
data_sent = true;
|
|
}
|
|
|
|
if (likely(data_sent))
|
|
wg_timers_data_sent(peer);
|
|
|
|
keep_key_fresh(peer);
|
|
}
|
|
|
|
void wg_packet_tx_worker(struct work_struct *work)
|
|
{
|
|
struct wg_peer *peer = container_of(work, struct wg_peer, transmit_packet_work);
|
|
struct noise_keypair *keypair;
|
|
enum packet_state state;
|
|
struct sk_buff *first;
|
|
|
|
while ((first = wg_prev_queue_peek(&peer->tx_queue)) != NULL &&
|
|
(state = atomic_read_acquire(&PACKET_CB(first)->state)) !=
|
|
PACKET_STATE_UNCRYPTED) {
|
|
wg_prev_queue_drop_peeked(&peer->tx_queue);
|
|
keypair = PACKET_CB(first)->keypair;
|
|
|
|
if (likely(state == PACKET_STATE_CRYPTED))
|
|
wg_packet_create_data_done(peer, first);
|
|
else
|
|
kfree_skb_list(first);
|
|
|
|
wg_noise_keypair_put(keypair, false);
|
|
wg_peer_put(peer);
|
|
if (need_resched())
|
|
cond_resched();
|
|
}
|
|
}
|
|
|
|
void wg_packet_encrypt_worker(struct work_struct *work)
|
|
{
|
|
struct crypt_queue *queue = container_of(work, struct multicore_worker,
|
|
work)->ptr;
|
|
struct sk_buff *first, *skb, *next;
|
|
|
|
while ((first = ptr_ring_consume_bh(&queue->ring)) != NULL) {
|
|
enum packet_state state = PACKET_STATE_CRYPTED;
|
|
|
|
skb_list_walk_safe(first, skb, next) {
|
|
if (likely(encrypt_packet(skb,
|
|
PACKET_CB(first)->keypair))) {
|
|
wg_reset_packet(skb, true);
|
|
} else {
|
|
state = PACKET_STATE_DEAD;
|
|
break;
|
|
}
|
|
}
|
|
wg_queue_enqueue_per_peer_tx(first, state);
|
|
if (need_resched())
|
|
cond_resched();
|
|
}
|
|
}
|
|
|
|
static void wg_packet_create_data(struct wg_peer *peer, struct sk_buff *first)
|
|
{
|
|
struct wg_device *wg = peer->device;
|
|
int ret = -EINVAL;
|
|
|
|
rcu_read_lock_bh();
|
|
if (unlikely(READ_ONCE(peer->is_dead)))
|
|
goto err;
|
|
|
|
ret = wg_queue_enqueue_per_device_and_peer(&wg->encrypt_queue, &peer->tx_queue, first,
|
|
wg->packet_crypt_wq, &wg->encrypt_queue.last_cpu);
|
|
if (unlikely(ret == -EPIPE))
|
|
wg_queue_enqueue_per_peer_tx(first, PACKET_STATE_DEAD);
|
|
err:
|
|
rcu_read_unlock_bh();
|
|
if (likely(!ret || ret == -EPIPE))
|
|
return;
|
|
wg_noise_keypair_put(PACKET_CB(first)->keypair, false);
|
|
wg_peer_put(peer);
|
|
kfree_skb_list(first);
|
|
}
|
|
|
|
void wg_packet_purge_staged_packets(struct wg_peer *peer)
|
|
{
|
|
spin_lock_bh(&peer->staged_packet_queue.lock);
|
|
peer->device->dev->stats.tx_dropped += peer->staged_packet_queue.qlen;
|
|
__skb_queue_purge(&peer->staged_packet_queue);
|
|
spin_unlock_bh(&peer->staged_packet_queue.lock);
|
|
}
|
|
|
|
void wg_packet_send_staged_packets(struct wg_peer *peer)
|
|
{
|
|
struct noise_keypair *keypair;
|
|
struct sk_buff_head packets;
|
|
struct sk_buff *skb;
|
|
|
|
/* Steal the current queue into our local one. */
|
|
__skb_queue_head_init(&packets);
|
|
spin_lock_bh(&peer->staged_packet_queue.lock);
|
|
skb_queue_splice_init(&peer->staged_packet_queue, &packets);
|
|
spin_unlock_bh(&peer->staged_packet_queue.lock);
|
|
if (unlikely(skb_queue_empty(&packets)))
|
|
return;
|
|
|
|
/* First we make sure we have a valid reference to a valid key. */
|
|
rcu_read_lock_bh();
|
|
keypair = wg_noise_keypair_get(
|
|
rcu_dereference_bh(peer->keypairs.current_keypair));
|
|
rcu_read_unlock_bh();
|
|
if (unlikely(!keypair))
|
|
goto out_nokey;
|
|
if (unlikely(!READ_ONCE(keypair->sending.is_valid)))
|
|
goto out_nokey;
|
|
if (unlikely(wg_birthdate_has_expired(keypair->sending.birthdate,
|
|
REJECT_AFTER_TIME)))
|
|
goto out_invalid;
|
|
|
|
/* After we know we have a somewhat valid key, we now try to assign
|
|
* nonces to all of the packets in the queue. If we can't assign nonces
|
|
* for all of them, we just consider it a failure and wait for the next
|
|
* handshake.
|
|
*/
|
|
skb_queue_walk(&packets, skb) {
|
|
/* 0 for no outer TOS: no leak. TODO: at some later point, we
|
|
* might consider using flowi->tos as outer instead.
|
|
*/
|
|
PACKET_CB(skb)->ds = ip_tunnel_ecn_encap(0, ip_hdr(skb), skb);
|
|
PACKET_CB(skb)->nonce =
|
|
atomic64_inc_return(&keypair->sending_counter) - 1;
|
|
if (unlikely(PACKET_CB(skb)->nonce >= REJECT_AFTER_MESSAGES))
|
|
goto out_invalid;
|
|
}
|
|
|
|
packets.prev->next = NULL;
|
|
wg_peer_get(keypair->entry.peer);
|
|
PACKET_CB(packets.next)->keypair = keypair;
|
|
wg_packet_create_data(peer, packets.next);
|
|
return;
|
|
|
|
out_invalid:
|
|
WRITE_ONCE(keypair->sending.is_valid, false);
|
|
out_nokey:
|
|
wg_noise_keypair_put(keypair, false);
|
|
|
|
/* We orphan the packets if we're waiting on a handshake, so that they
|
|
* don't block a socket's pool.
|
|
*/
|
|
skb_queue_walk(&packets, skb)
|
|
skb_orphan(skb);
|
|
/* Then we put them back on the top of the queue. We're not too
|
|
* concerned about accidentally getting things a little out of order if
|
|
* packets are being added really fast, because this queue is for before
|
|
* packets can even be sent and it's small anyway.
|
|
*/
|
|
spin_lock_bh(&peer->staged_packet_queue.lock);
|
|
skb_queue_splice(&packets, &peer->staged_packet_queue);
|
|
spin_unlock_bh(&peer->staged_packet_queue.lock);
|
|
|
|
/* If we're exiting because there's something wrong with the key, it
|
|
* means we should initiate a new handshake.
|
|
*/
|
|
wg_packet_send_queued_handshake_initiation(peer, false);
|
|
}
|