1 /* SPDX-License-Identifier: GPL-2.0 */
10 #include <linux/types.h>
11 #include <linux/skbuff.h>
13 #include <linux/ipv6.h>
14 #include <net/ip_tunnels.h>
18 struct multicore_worker;
23 /* queueing.c APIs: */
24 int wg_packet_queue_init(struct crypt_queue *queue, work_func_t function,
26 void wg_packet_queue_free(struct crypt_queue *queue, bool purge);
27 struct multicore_worker __percpu *
28 wg_packet_percpu_multicore_worker_alloc(work_func_t function, void *ptr);
31 void wg_packet_receive(struct wg_device *wg, struct sk_buff *skb);
32 void wg_packet_handshake_receive_worker(struct work_struct *work);
33 /* NAPI poll function: */
34 int wg_packet_rx_poll(struct napi_struct *napi, int budget);
35 /* Workqueue worker: */
36 void wg_packet_decrypt_worker(struct work_struct *work);
39 void wg_packet_send_queued_handshake_initiation(struct wg_peer *peer,
41 void wg_packet_send_handshake_response(struct wg_peer *peer);
42 void wg_packet_send_handshake_cookie(struct wg_device *wg,
43 struct sk_buff *initiating_skb,
45 void wg_packet_send_keepalive(struct wg_peer *peer);
46 void wg_packet_purge_staged_packets(struct wg_peer *peer);
47 void wg_packet_send_staged_packets(struct wg_peer *peer);
48 /* Workqueue workers: */
49 void wg_packet_handshake_send_worker(struct work_struct *work);
50 void wg_packet_tx_worker(struct work_struct *work);
51 void wg_packet_encrypt_worker(struct work_struct *work);
54 PACKET_STATE_UNCRYPTED,
61 struct noise_keypair *keypair;
67 #define PACKET_CB(skb) ((struct packet_cb *)((skb)->cb))
68 #define PACKET_PEER(skb) (PACKET_CB(skb)->keypair->entry.peer)
70 static inline bool wg_check_packet_protocol(struct sk_buff *skb)
72 __be16 real_protocol = ip_tunnel_parse_protocol(skb);
73 return real_protocol && skb->protocol == real_protocol;
76 static inline void wg_reset_packet(struct sk_buff *skb, bool encapsulating)
78 u8 l4_hash = skb->l4_hash;
79 u8 sw_hash = skb->sw_hash;
81 skb_scrub_packet(skb, true);
82 memset(&skb->headers, 0, sizeof(skb->headers));
84 skb->l4_hash = l4_hash;
85 skb->sw_hash = sw_hash;
88 skb->queue_mapping = 0;
93 #ifdef CONFIG_NET_SCHED
96 skb_reset_redirect(skb);
97 skb->hdr_len = skb_headroom(skb);
98 skb_reset_mac_header(skb);
99 skb_reset_network_header(skb);
100 skb_reset_transport_header(skb);
101 skb_probe_transport_header(skb);
102 skb_reset_inner_headers(skb);
105 static inline int wg_cpumask_choose_online(int *stored_cpu, unsigned int id)
107 unsigned int cpu = *stored_cpu, cpu_index, i;
109 if (unlikely(cpu == nr_cpumask_bits ||
110 !cpumask_test_cpu(cpu, cpu_online_mask))) {
111 cpu_index = id % cpumask_weight(cpu_online_mask);
112 cpu = cpumask_first(cpu_online_mask);
113 for (i = 0; i < cpu_index; ++i)
114 cpu = cpumask_next(cpu, cpu_online_mask);
120 /* This function is racy, in the sense that next is unlocked, so it could return
121 * the same CPU twice. A race-free version of this would be to instead store an
122 * atomic sequence number, do an increment-and-return, and then iterate through
123 * every possible CPU until we get to that index -- choose_cpu. However that's
124 * a bit slower, and it doesn't seem like this potential race actually
125 * introduces any performance loss, so we live with it.
127 static inline int wg_cpumask_next_online(int *next)
131 while (unlikely(!cpumask_test_cpu(cpu, cpu_online_mask)))
132 cpu = cpumask_next(cpu, cpu_online_mask) % nr_cpumask_bits;
133 *next = cpumask_next(cpu, cpu_online_mask) % nr_cpumask_bits;
137 void wg_prev_queue_init(struct prev_queue *queue);
140 bool wg_prev_queue_enqueue(struct prev_queue *queue, struct sk_buff *skb);
142 /* Single consumer */
143 struct sk_buff *wg_prev_queue_dequeue(struct prev_queue *queue);
145 /* Single consumer */
146 static inline struct sk_buff *wg_prev_queue_peek(struct prev_queue *queue)
149 return queue->peeked;
150 queue->peeked = wg_prev_queue_dequeue(queue);
151 return queue->peeked;
154 /* Single consumer */
155 static inline void wg_prev_queue_drop_peeked(struct prev_queue *queue)
157 queue->peeked = NULL;
160 static inline int wg_queue_enqueue_per_device_and_peer(
161 struct crypt_queue *device_queue, struct prev_queue *peer_queue,
162 struct sk_buff *skb, struct workqueue_struct *wq, int *next_cpu)
166 atomic_set_release(&PACKET_CB(skb)->state, PACKET_STATE_UNCRYPTED);
167 /* We first queue this up for the peer ingestion, but the consumer
168 * will wait for the state to change to CRYPTED or DEAD before.
170 if (unlikely(!wg_prev_queue_enqueue(peer_queue, skb)))
173 /* Then we queue it up in the device queue, which consumes the
174 * packet as soon as it can.
176 cpu = wg_cpumask_next_online(next_cpu);
177 if (unlikely(ptr_ring_produce_bh(&device_queue->ring, skb)))
179 queue_work_on(cpu, wq, &per_cpu_ptr(device_queue->worker, cpu)->work);
183 static inline void wg_queue_enqueue_per_peer_tx(struct sk_buff *skb, enum packet_state state)
185 /* We take a reference, because as soon as we call atomic_set, the
186 * peer can be freed from below us.
188 struct wg_peer *peer = wg_peer_get(PACKET_PEER(skb));
190 atomic_set_release(&PACKET_CB(skb)->state, state);
191 queue_work_on(wg_cpumask_choose_online(&peer->serial_work_cpu, peer->internal_id),
192 peer->device->packet_crypt_wq, &peer->transmit_packet_work);
196 static inline void wg_queue_enqueue_per_peer_rx(struct sk_buff *skb, enum packet_state state)
198 /* We take a reference, because as soon as we call atomic_set, the
199 * peer can be freed from below us.
201 struct wg_peer *peer = wg_peer_get(PACKET_PEER(skb));
203 atomic_set_release(&PACKET_CB(skb)->state, state);
204 napi_schedule(&peer->napi);
209 bool wg_packet_counter_selftest(void);
212 #endif /* _WG_QUEUEING_H */