Merge branch 'wireguard-siphash-patches-for-5-16-rc6'

Jason A. Donenfeld says:

====================
wireguard/siphash patches for 5.16-rc

Here's quite a largeish set of stable patches I've had queued up and
testing for a number of months now:

  - Patch (1) squelches a sparse warning by fixing an annotation.
  - Patches (2), (3), and (5) are minor improvements and fixes to the
    test suite.
  - Patch (4) is part of a tree-wide cleanup to have module-specific
    init and exit functions.
  - Patch (6) fixes a an issue with dangling dst references, by having a
    function to release references immediately rather than deferring,
    and adds an associated test case to prevent this from regressing.
  - Patches (7) and (8) help mitigate somewhat a potential DoS on the
    ingress path due to the use of skb_list's locking hitting contention
    on multiple cores by switching to using a ring buffer and dropping
    packets on contention rather than locking up another core spinning.
  - Patch (9) switches kvzalloc to kvcalloc for better form.
  - Patch (10) fixes alignment traps in siphash with clang-13 (and maybe
    other compilers) on armv6, by switching to using the unaligned
    functions by default instead of the aligned functions by default.
====================

Link: https://lore.kernel.org/r/20211129153929.3457-1-Jason@zx2c4.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

1 files changed, 23 insertions, 16 deletions

diff --git a/drivers/net/wireguard/receive.c b/drivers/net/wireguard/receive.c
index 7dc84bc..7b8df40 100644
--- a/drivers/net/wireguard/receive.c
+++ b/drivers/net/wireguard/receive.c
@@ -116,8 +116,8 @@ static void wg_receive_handshake_packet(struct wg_device *wg,
 		return;
 	}
 
-	under_load = skb_queue_len(&wg->incoming_handshakes) >=
-		     MAX_QUEUED_INCOMING_HANDSHAKES / 8;
+	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) {
@@ -212,13 +212,14 @@ static void wg_receive_handshake_packet(struct wg_device *wg,
 
 void wg_packet_handshake_receive_worker(struct work_struct *work)
 {
-	struct wg_device *wg = container_of(work, struct multicore_worker,
-					    work)->ptr;
+	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 = skb_dequeue(&wg->incoming_handshakes)) != NULL) {
+	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();
 	}
 }
@@ -553,22 +554,28 @@ void wg_packet_receive(struct wg_device *wg, struct sk_buff *skb)
 	case cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION):
 	case cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE):
 	case cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE): {
-		int cpu;
-
-		if (skb_queue_len(&wg->incoming_handshakes) >
-			    MAX_QUEUED_INCOMING_HANDSHAKES ||
-		    unlikely(!rng_is_initialized())) {
+		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;
 		}
-		skb_queue_tail(&wg->incoming_handshakes, skb);
-		/* Queues up a call to packet_process_queued_handshake_
-		 * packets(skb):
-		 */
-		cpu = wg_cpumask_next_online(&wg->incoming_handshake_cpu);
+		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->incoming_handshakes_worker, cpu)->work);
+			      &per_cpu_ptr(wg->handshake_queue.worker, cpu)->work);
 		break;
 	}
 	case cpu_to_le32(MESSAGE_DATA):