1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
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
;
22 /* queueing.c APIs: */
23 int wg_packet_queue_init(struct crypt_queue
*queue
, work_func_t function
,
24 bool multicore
, unsigned int len
);
25 void wg_packet_queue_free(struct crypt_queue
*queue
, bool multicore
);
26 struct multicore_worker __percpu
*
27 wg_packet_percpu_multicore_worker_alloc(work_func_t function
, void *ptr
);
30 void wg_packet_receive(struct wg_device
*wg
, struct sk_buff
*skb
);
31 void wg_packet_handshake_receive_worker(struct work_struct
*work
);
32 /* NAPI poll function: */
33 int wg_packet_rx_poll(struct napi_struct
*napi
, int budget
);
34 /* Workqueue worker: */
35 void wg_packet_decrypt_worker(struct work_struct
*work
);
38 void wg_packet_send_queued_handshake_initiation(struct wg_peer
*peer
,
40 void wg_packet_send_handshake_response(struct wg_peer
*peer
);
41 void wg_packet_send_handshake_cookie(struct wg_device
*wg
,
42 struct sk_buff
*initiating_skb
,
44 void wg_packet_send_keepalive(struct wg_peer
*peer
);
45 void wg_packet_purge_staged_packets(struct wg_peer
*peer
);
46 void wg_packet_send_staged_packets(struct wg_peer
*peer
);
47 /* Workqueue workers: */
48 void wg_packet_handshake_send_worker(struct work_struct
*work
);
49 void wg_packet_tx_worker(struct work_struct
*work
);
50 void wg_packet_encrypt_worker(struct work_struct
*work
);
53 PACKET_STATE_UNCRYPTED
,
60 struct noise_keypair
*keypair
;
66 #define PACKET_CB(skb) ((struct packet_cb *)((skb)->cb))
67 #define PACKET_PEER(skb) (PACKET_CB(skb)->keypair->entry.peer)
69 static inline bool wg_check_packet_protocol(struct sk_buff
*skb
)
71 __be16 real_protocol
= ip_tunnel_parse_protocol(skb
);
72 return real_protocol
&& skb
->protocol
== real_protocol
;
75 static inline void wg_reset_packet(struct sk_buff
*skb
, bool encapsulating
)
77 u8 l4_hash
= skb
->l4_hash
;
78 u8 sw_hash
= skb
->sw_hash
;
80 skb_scrub_packet(skb
, true);
81 memset(&skb
->headers_start
, 0,
82 offsetof(struct sk_buff
, headers_end
) -
83 offsetof(struct sk_buff
, headers_start
));
85 skb
->l4_hash
= l4_hash
;
86 skb
->sw_hash
= sw_hash
;
89 skb
->queue_mapping
= 0;
94 #ifdef CONFIG_NET_SCHED
97 skb_reset_redirect(skb
);
98 skb
->hdr_len
= skb_headroom(skb
);
99 skb_reset_mac_header(skb
);
100 skb_reset_network_header(skb
);
101 skb_reset_transport_header(skb
);
102 skb_probe_transport_header(skb
);
103 skb_reset_inner_headers(skb
);
106 static inline int wg_cpumask_choose_online(int *stored_cpu
, unsigned int id
)
108 unsigned int cpu
= *stored_cpu
, cpu_index
, i
;
110 if (unlikely(cpu
== nr_cpumask_bits
||
111 !cpumask_test_cpu(cpu
, cpu_online_mask
))) {
112 cpu_index
= id
% cpumask_weight(cpu_online_mask
);
113 cpu
= cpumask_first(cpu_online_mask
);
114 for (i
= 0; i
< cpu_index
; ++i
)
115 cpu
= cpumask_next(cpu
, cpu_online_mask
);
121 /* This function is racy, in the sense that next is unlocked, so it could return
122 * the same CPU twice. A race-free version of this would be to instead store an
123 * atomic sequence number, do an increment-and-return, and then iterate through
124 * every possible CPU until we get to that index -- choose_cpu. However that's
125 * a bit slower, and it doesn't seem like this potential race actually
126 * introduces any performance loss, so we live with it.
128 static inline int wg_cpumask_next_online(int *next
)
132 while (unlikely(!cpumask_test_cpu(cpu
, cpu_online_mask
)))
133 cpu
= cpumask_next(cpu
, cpu_online_mask
) % nr_cpumask_bits
;
134 *next
= cpumask_next(cpu
, cpu_online_mask
) % nr_cpumask_bits
;
138 static inline int wg_queue_enqueue_per_device_and_peer(
139 struct crypt_queue
*device_queue
, struct crypt_queue
*peer_queue
,
140 struct sk_buff
*skb
, struct workqueue_struct
*wq
, int *next_cpu
)
144 atomic_set_release(&PACKET_CB(skb
)->state
, PACKET_STATE_UNCRYPTED
);
145 /* We first queue this up for the peer ingestion, but the consumer
146 * will wait for the state to change to CRYPTED or DEAD before.
148 if (unlikely(ptr_ring_produce_bh(&peer_queue
->ring
, skb
)))
150 /* Then we queue it up in the device queue, which consumes the
151 * packet as soon as it can.
153 cpu
= wg_cpumask_next_online(next_cpu
);
154 if (unlikely(ptr_ring_produce_bh(&device_queue
->ring
, skb
)))
156 queue_work_on(cpu
, wq
, &per_cpu_ptr(device_queue
->worker
, cpu
)->work
);
160 static inline void wg_queue_enqueue_per_peer(struct crypt_queue
*queue
,
162 enum packet_state state
)
164 /* We take a reference, because as soon as we call atomic_set, the
165 * peer can be freed from below us.
167 struct wg_peer
*peer
= wg_peer_get(PACKET_PEER(skb
));
169 atomic_set_release(&PACKET_CB(skb
)->state
, state
);
170 queue_work_on(wg_cpumask_choose_online(&peer
->serial_work_cpu
,
172 peer
->device
->packet_crypt_wq
, &queue
->work
);
176 static inline void wg_queue_enqueue_per_peer_napi(struct sk_buff
*skb
,
177 enum packet_state state
)
179 /* We take a reference, because as soon as we call atomic_set, the
180 * peer can be freed from below us.
182 struct wg_peer
*peer
= wg_peer_get(PACKET_PEER(skb
));
184 atomic_set_release(&PACKET_CB(skb
)->state
, state
);
185 napi_schedule(&peer
->napi
);
190 bool wg_packet_counter_selftest(void);
193 #endif /* _WG_QUEUEING_H */