[PATCH net 0/2] wireguard: memcg accounting and initiation race fix

[PATCH net 0/2] wireguard: memcg accounting and initiation race fix

[Alexander Ildiyarov]

  Two independent fixes:
  
  1/2 charges peer, allowedips, dst_cache and hashtable allocations to the
  memcg of the configuring process, which currently can allocate
  unaccounted kernel memory through WG_CMD_SET_DEVICE.
  
  2/2 rechecks the replay/flood conditions under the write lock in
  wg_noise_handshake_consume_initiation(), the initiation-side counterpart
  of 2a73c9b2c0ca. The race window is narrow and was only reproduced with
  an artificial delay; details are in the commit message.
  
  Tested in qemu (x86_64, 4 vCPUs) with reproducers for both issues and
  CONFIG_WIREGUARD_DEBUG self-tests. netns.sh was run up to the
  allowed-ips removal step, where the available wg(8) is too old; that
  part was not run.


Alexander Ildiyarov (2):
  wireguard: account peer and allowedips allocations to memcg
  wireguard: noise: recheck replay and flood under write lock on
    initiation

 drivers/net/wireguard/allowedips.c |  2 +-
 drivers/net/wireguard/noise.c      | 29 ++++++++++++++++++++++-------
 drivers/net/wireguard/peer.c       |  4 ++--
 drivers/net/wireguard/peerlookup.c |  4 ++--
 4 files changed, 27 insertions(+), 12 deletions(-)

-- 
2.43.0

[PATCH net 1/2] wireguard: account peer and allowedips allocations to memcg

[Alexander Ildiyarov]

Peers, allowed IP trie nodes, the per-peer dst_cache and the per-device
hashtables are all allocated in the context of the process that
configures the device over netlink, but none of these allocations are
charged to that process's memory cgroup: the wg_peer and allowedips_node
caches are created without SLAB_ACCOUNT, and the hashtables and
dst_cache use plain GFP_KERNEL.

The amount of memory is bounded only by MAX_PEERS_PER_DEVICE (2^20) per
device, with no limit on the number of devices or on the number of
allowed IPs per peer. Since WG_CMD_SET_DEVICE uses GENL_UNS_ADMIN_PERM,
any user who can create a user and network namespace can do this, and
thereby escape the memory limits of the cgroup it runs in.

For example, a process in a cgroup with memory.max=64M that adds 100000
peers with 20 /128 allowed IPs each grows Slab by ~790 MiB, yet the
cgroup's memory.current stays below 1 MiB and its memory limit is never
hit.

Fix this by creating the two caches with SLAB_ACCOUNT and using
GFP_KERNEL_ACCOUNT for the hashtables and the dst_cache. The keypair
allocations are left alone, as they happen from the handshake worker and
are bounded to three per peer.

Fixes: e7096c131e51 ("net: WireGuard secure network tunnel")
Cc: stable@vger.kernel.org
Assisted-by: LLM
Signed-off-by: Alexander Ildiyarov <alex@badcode.im>
---
 drivers/net/wireguard/allowedips.c | 2 +-
 drivers/net/wireguard/peer.c       | 4 ++--
 drivers/net/wireguard/peerlookup.c | 4 ++--
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/drivers/net/wireguard/allowedips.c b/drivers/net/wireguard/allowedips.c
index 5ece9acad64d..2f2b8a94c8a7 100644
--- a/drivers/net/wireguard/allowedips.c
+++ b/drivers/net/wireguard/allowedips.c
@@ -411,7 +411,7 @@ struct wg_peer *wg_allowedips_lookup_src(struct allowedips *table,
 
 int __init wg_allowedips_slab_init(void)
 {
-	node_cache = KMEM_CACHE(allowedips_node, 0);
+	node_cache = KMEM_CACHE(allowedips_node, SLAB_ACCOUNT);
 	return node_cache ? 0 : -ENOMEM;
 }
 
diff --git a/drivers/net/wireguard/peer.c b/drivers/net/wireguard/peer.c
index 1cb502a932e0..05b80521c706 100644
--- a/drivers/net/wireguard/peer.c
+++ b/drivers/net/wireguard/peer.c
@@ -33,7 +33,7 @@ struct wg_peer *wg_peer_create(struct wg_device *wg,
 	peer = kmem_cache_zalloc(peer_cache, GFP_KERNEL);
 	if (unlikely(!peer))
 		return ERR_PTR(ret);
-	if (unlikely(dst_cache_init(&peer->endpoint_cache, GFP_KERNEL)))
+	if (unlikely(dst_cache_init(&peer->endpoint_cache, GFP_KERNEL_ACCOUNT)))
 		goto err;
 
 	peer->device = wg;
@@ -229,7 +229,7 @@ void wg_peer_put(struct wg_peer *peer)
 
 int __init wg_peer_init(void)
 {
-	peer_cache = KMEM_CACHE(wg_peer, 0);
+	peer_cache = KMEM_CACHE(wg_peer, SLAB_ACCOUNT);
 	return peer_cache ? 0 : -ENOMEM;
 }
 
diff --git a/drivers/net/wireguard/peerlookup.c b/drivers/net/wireguard/peerlookup.c
index 99651a86ac95..f333e906014a 100644
--- a/drivers/net/wireguard/peerlookup.c
+++ b/drivers/net/wireguard/peerlookup.c
@@ -21,7 +21,7 @@ static struct hlist_head *pubkey_bucket(struct pubkey_hashtable *table,
 
 struct pubkey_hashtable *wg_pubkey_hashtable_alloc(void)
 {
-	struct pubkey_hashtable *table = kvmalloc_obj(*table);
+	struct pubkey_hashtable *table = kvmalloc_obj(*table, GFP_KERNEL_ACCOUNT);
 
 	if (!table)
 		return NULL;
@@ -82,7 +82,7 @@ static struct hlist_head *index_bucket(struct index_hashtable *table,
 
 struct index_hashtable *wg_index_hashtable_alloc(void)
 {
-	struct index_hashtable *table = kvmalloc_obj(*table);
+	struct index_hashtable *table = kvmalloc_obj(*table, GFP_KERNEL_ACCOUNT);
 
 	if (!table)
 		return NULL;
-- 
2.43.0

[PATCH net 2/2] wireguard: noise: recheck replay and flood under write lock on initiation

[Alexander Ildiyarov]

wg_noise_handshake_consume_initiation() checks the initiation's
timestamp against latest_timestamp, and last_initiation_consumption
against the current time, while holding handshake->lock for reading.
It then drops the lock and takes it again for writing to store the new
state, without repeating the checks.

Incoming handshake packets are spread round-robin over all online CPUs,
so two copies of the same initiation, received back to back, are
processed in parallel. Both copies can pass the checks before either of
them updates latest_timestamp and last_initiation_consumption, so the
replay and flood protection is bypassed. The responder then creates two
responses, each with a new ephemeral key, and installs the keypair of
the last one, while the initiator accepts whichever response arrives
first. When those differ, the session is unusable until the initiator
gives up on it and starts a new handshake, and an attacker who can see
and duplicate initiations can repeat this for every handshake.

The window is narrow, since the expensive DH computations happen before
the check, so it is between up_read() and down_write() only. With netem
duplicating every initiation between two namespaces, no double response
was observed in 200 handshakes on a 4-CPU VM. With a usleep_range(1000,
2000) inserted before down_write() to widen the window, 30 of 50
duplicated initiations produced two responses, and in 12 of 50 cases
the session was unusable after the handshake. With this patch applied,
and the same delay, every duplicate was rejected and all 50 sessions
worked.

Fix this by doing the checks again once the write lock is held, keeping
the read-locked check as a fast path for rejecting replays.

This is the initiation-side counterpart of the response-side race fixed
by:
commit 2a73c9b2c0ca ("wireguard: noise: reject response consumption after intermediate initiation")

Fixes: e7096c131e51 ("net: WireGuard secure network tunnel")
Cc: stable@vger.kernel.org
Assisted-by: LLM
Signed-off-by: Alexander Ildiyarov <alex@badcode.im>
---
 drivers/net/wireguard/noise.c | 29 ++++++++++++++++++++++-------
 1 file changed, 22 insertions(+), 7 deletions(-)

diff --git a/drivers/net/wireguard/noise.c b/drivers/net/wireguard/noise.c
index 428d3afe3ce4..209127654e65 100644
--- a/drivers/net/wireguard/noise.c
+++ b/drivers/net/wireguard/noise.c
@@ -580,13 +580,23 @@ wg_noise_handshake_create_initiation(struct message_handshake_initiation *dst,
 	return ret;
 }
 
+/* Must hold handshake->lock, for reading or writing */
+static bool initiation_is_replay_or_flood(const struct noise_handshake *handshake,
+					  const u8 timestamp[NOISE_TIMESTAMP_LEN])
+{
+	return memcmp(timestamp, handshake->latest_timestamp,
+		      NOISE_TIMESTAMP_LEN) <= 0 ||
+	       (s64)handshake->last_initiation_consumption +
+			       NSEC_PER_SEC / INITIATIONS_PER_SECOND >
+		       (s64)ktime_get_coarse_boottime_ns();
+}
+
 struct wg_peer *
 wg_noise_handshake_consume_initiation(struct message_handshake_initiation *src,
 				      struct wg_device *wg)
 {
 	struct wg_peer *peer = NULL, *ret_peer = NULL;
 	struct noise_handshake *handshake;
-	bool replay_attack, flood_attack;
 	u8 key[NOISE_SYMMETRIC_KEY_LEN];
 	u8 chaining_key[NOISE_HASH_LEN];
 	u8 hash[NOISE_HASH_LEN];
@@ -594,6 +604,7 @@ wg_noise_handshake_consume_initiation(struct message_handshake_initiation *src,
 	u8 e[NOISE_PUBLIC_KEY_LEN];
 	u8 t[NOISE_TIMESTAMP_LEN];
 	u64 initiation_consumption;
+	bool reject;
 
 	down_read(&wg->static_identity.lock);
 	if (unlikely(!wg->static_identity.has_identity))
@@ -630,17 +641,21 @@ wg_noise_handshake_consume_initiation(struct message_handshake_initiation *src,
 		goto out;
 
 	down_read(&handshake->lock);
-	replay_attack = memcmp(t, handshake->latest_timestamp,
-			       NOISE_TIMESTAMP_LEN) <= 0;
-	flood_attack = (s64)handshake->last_initiation_consumption +
-			       NSEC_PER_SEC / INITIATIONS_PER_SECOND >
-		       (s64)ktime_get_coarse_boottime_ns();
+	reject = initiation_is_replay_or_flood(handshake, t);
 	up_read(&handshake->lock);
-	if (replay_attack || flood_attack)
+	if (reject)
 		goto out;
 
 	/* Success! Copy everything to peer */
 	down_write(&handshake->lock);
+	/* Check again, now that we have an exclusive lock, so that two threads
+	 * processing the same initiation in parallel can't both pass the check
+	 * above before either of them updates the timestamp.
+	 */
+	if (initiation_is_replay_or_flood(handshake, t)) {
+		up_write(&handshake->lock);
+		goto out;
+	}
 	memcpy(handshake->remote_ephemeral, e, NOISE_PUBLIC_KEY_LEN);
 	if (memcmp(t, handshake->latest_timestamp, NOISE_TIMESTAMP_LEN) > 0)
 		memcpy(handshake->latest_timestamp, t, NOISE_TIMESTAMP_LEN);
-- 
2.43.0

Re: [PATCH net 0/2] wireguard: memcg accounting and initiation race fix

[netdev-bot+sinfo]

Hi!

This is an automated message. This series looks like a fix, but its
commit messages seem to be missing some information:

 - How the issue was discovered, e.g. hit in production, hit during
   development, syzbot report, manual code inspection, LLM or static
   analysis tool scan.

Please do not repost the series just to address the above. Instead,
reply to this email with the missing information, so that reviewers
can take it into account. If the series needs another revision for
other reasons, please include the information in the commit messages
then.

The evaluation is done by an LLM so it may be wrong, if you think
that is the case please reply and explain.

Re: [PATCH net 1/2] wireguard: account peer and allowedips allocations to memcg

[netdev-bot+sashiko]

Thank you for your contribution! Sashiko AI review found 3 potential
issue(s) to consider.

Critical: 0 · High: 1 · Medium: 2 · Low: 0

If you believe none of the issues are legitimate, please reply to each
issue of Medium severity or higher and explain why the AI is wrong.
If any of them are legitimate and you plan to respin, please let patchwork
know by including "pw-bot: cr" as a separate line at the end of your reply
(one such reply per series is enough).

- [High] Incomplete fix: every peer still creates uncharged memory and a
  PID.
- [Medium] Incomplete fix: per-device allocations in wg_newlink() are
  still not charged.
- [Medium] The commit leaves noise_keypair allocations uncharged, saying
  they 'happen from the handshake worker and are bounded to three per…

--

wireguard: account peer and allowedips allocations to memcg

This patch creates the wg_peer and allowedips_node slab caches with
SLAB_ACCOUNT. It also switches the per-peer dst_cache and the per-device
pubkey and index hashtables to GFP_KERNEL_ACCOUNT. The goal is to stop a
user with a user and network namespace from escaping its memory cgroup
limit by adding peers over WG_CMD_SET_DEVICE.

> The keypair allocations are left alone, as they happen from the handshake
> worker and are bounded to three per peer.

[Severity: Medium]
Does the per-peer bound make the keypair memory small enough to ignore?

struct noise_keypair embeds a noise_replay_counter with:

    unsigned long backtrack[COUNTER_BITS_TOTAL / BITS_PER_LONG];

COUNTER_BITS_TOTAL is 8192, so that is a 1KB bitmap. The struct ends up
around 1.2KB, which means keypair_create() is served from kmalloc-2k:

drivers/net/wireguard/noise.c:keypair_create() {
	struct noise_keypair *keypair = kzalloc_obj(*keypair);
    ...
}

add_new_keypair() keeps two keypairs in slots per peer, plus any that are
in flight or waiting on RCU. That is about 4KB or more of uncharged memory
per peer. The now-charged wg_peer object plus the per-CPU dst_cache come
to about 2KB.

The configuring user picks every peer public key, so it knows the matching
private keys. A UDP program in the same netns can send valid handshake
initiations for every peer to the listen port. The responder path then
allocates keypairs from the kworker:

wg_packet_handshake_receive_worker()
  wg_packet_send_handshake_response()
    wg_noise_handshake_begin_session()
      keypair_create()

Once the peer count is limited only by charged memory, could the
uncharged keypair memory still reach about 1-2x memory.max?

This runs in a worker, so GFP_KERNEL_ACCOUNT alone would charge root.
Would this need the configuring memcg to be recorded, with
set_active_memcg() used in the handshake path?

> diff --git a/drivers/net/wireguard/peer.c b/drivers/net/wireguard/peer.c
> index 1cb502a932e07..05b80521c7061 100644
> --- a/drivers/net/wireguard/peer.c
> +++ b/drivers/net/wireguard/peer.c
> @@ -33,7 +33,7 @@ struct wg_peer *wg_peer_create(struct wg_device *wg,
>  	peer = kmem_cache_zalloc(peer_cache, GFP_KERNEL);
>  	if (unlikely(!peer))
>  		return ERR_PTR(ret);
> -	if (unlikely(dst_cache_init(&peer->endpoint_cache, GFP_KERNEL)))
> +	if (unlikely(dst_cache_init(&peer->endpoint_cache, GFP_KERNEL_ACCOUNT)))
>  		goto err;
>  
>  	peer->device = wg;

[Severity: High]
The peer slab and dst_cache are now charged. Does each peer still pin a
larger amount of uncharged memory through its NAPI kthread?

wg_newlink() turns on threaded NAPI for every wireguard device:

	netif_threaded_enable(dev);
	ret = register_netdevice(dev);

Later in wg_peer_create(), each peer does:

	netif_napi_add(wg->dev, &peer->napi, wg_packet_rx_poll);

Peers have no napi->config, so napi_get_threaded_config() returns
dev->threaded. That spawns one kthread per peer:

wg_peer_create()
  netif_napi_add()
    netif_napi_add_weight_locked()
      napi_kthread_create()
        kthread_run(napi_threaded_poll, n, "napi/%s-%d", ...)

kthreadd forks the thread. Its task_struct and kernel stack (about
20-30KB) are therefore charged to the root memcg. The PID is also taken
outside the caller's pids cgroup and RLIMIT_NPROC.

Each peer charges only about 2KB for wg_peer plus dst_cache. Could a
cgroup with memory.max=64M still create tens of thousands of peers and pin
hundreds of MiB of uncharged memory? With a larger limit, could this also
use up pid_max for the whole system?

More GFP_KERNEL_ACCOUNT flags would not cover this, because kthreadd does
the allocation. Would this need a per-device or per-user cap on peers or
NAPI threads, or a design that does not spawn a kthread per peer?

The next patch in the series ("wireguard: noise: recheck replay and flood
under write lock on initiation") only changes the replay checks in
noise.c and does not affect this.

[ ... ]

> diff --git a/drivers/net/wireguard/peerlookup.c b/drivers/net/wireguard/peerlookup.c
> index 99651a86ac953..f333e906014ab 100644
> --- a/drivers/net/wireguard/peerlookup.c
> +++ b/drivers/net/wireguard/peerlookup.c

[ ... ]

> @@ -82,7 +82,7 @@ static struct hlist_head *index_bucket(struct index_hashtable *table,
>  
>  struct index_hashtable *wg_index_hashtable_alloc(void)
>  {
> -	struct index_hashtable *table = kvmalloc_obj(*table);
> +	struct index_hashtable *table = kvmalloc_obj(*table, GFP_KERNEL_ACCOUNT);
>  
>  	if (!table)
>  		return NULL;

[Severity: Medium]
The two per-device hashtables are now charged, and the commit message
says the number of devices has no limit. Are the other per-device
allocations in wg_newlink() still uncharged?

wg_newlink() calls wg_packet_queue_init() three times, and each call
does:

drivers/net/wireguard/queueing.c:wg_packet_queue_init() {
    ...
	ret = ptr_ring_init(&queue->ring, len, GFP_KERNEL);
    ...
	queue->worker = wg_packet_percpu_multicore_worker_alloc(function, queue);
    ...
}

ptr_ring_init() ends up in kvmalloc_array() with GFP_KERNEL. On 64-bit
that is 8KB + 8KB + 32KB per device (MAX_QUEUED_PACKETS and
MAX_QUEUED_INCOMING_HANDSHAKES).

The per-CPU worker arrays come from alloc_percpu() without
__GFP_ACCOUNT:

	struct multicore_worker __percpu *worker = alloc_percpu(struct multicore_worker);

wg_newlink() also creates three workqueues. packet_crypt_wq uses
WQ_MEM_RECLAIM, which adds a rescuer kthread forked by kthreadd:

	wg->packet_crypt_wq = alloc_workqueue("wg-crypt-%s",
			WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_PERCPU, 0,
			dev->name);

Could a user with user and net namespaces call RTM_NEWLINK in a loop to
create wireguard devices? That would still allocate per-device memory
that never shows up in memory.current.

Should the ptr_ring and per-CPU worker allocations be charged as well?

-- 
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261105141006.95083-1-alex%40badcode.im

Re: [PATCH net 2/2] wireguard: noise: recheck replay and flood under write lock on initiation

[netdev-bot+sashiko]

Thank you for your contribution! Sashiko AI review found more severe
issue(s) in other patches of this series. This mail is only an FYI, in case
you want to address the nit picks below when you respin.

Low priority / nit picks for your consideration:

- [Low] The commit message names the response-side fix as `commit
  2a73c9b2c0ca ("wireguard: noise: reject response consumption after…

--

wireguard: noise: recheck replay and flood under write lock on initiation

This commit makes wg_noise_handshake_consume_initiation() run the replay
and flood checks a second time after taking handshake->lock for writing.
Without the second check, two copies of the same initiation processed in
parallel could both get past the read-locked check. The read-locked check
stays as a fast path for rejecting replays.

> This is the initiation-side counterpart of the response-side race fixed
> by:
> commit 2a73c9b2c0ca ("wireguard: noise: reject response consumption after intermediate initiation")

[Severity: Low]
Is this the right hash for the referenced commit? 2a73c9b2c0ca doesn't
resolve here. git log reports it as an unknown revision.

The commit with this subject appears to be:

commit d0305f8d9002 ("wireguard: noise: reject response consumption after intermediate initiation")

Running git blame on the response-side recheck under down_write() in
wg_noise_handshake_consume_response() also points to d0305f8d9002:

    if (handshake->state != state ||
        crypto_memneq(handshake->ephemeral_private, ephemeral_private, NOISE_PUBLIC_KEY_LEN)) {

Could the reference be changed to d0305f8d9002? This patch is tagged for
stable, so readers who follow the cited hash to the related fix would
find the right commit.

-- 
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261105141006.95083-1-alex%40badcode.im