xref: /linux/net/mptcp/protocol.h (revision 79ac11393328fb1717d17c12e3c0eef0e9fa0647)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* Multipath TCP
3  *
4  * Copyright (c) 2017 - 2019, Intel Corporation.
5  */
6 
7 #ifndef __MPTCP_PROTOCOL_H
8 #define __MPTCP_PROTOCOL_H
9 
10 #include <linux/random.h>
11 #include <net/tcp.h>
12 #include <net/inet_connection_sock.h>
13 #include <uapi/linux/mptcp.h>
14 #include <net/genetlink.h>
15 
16 #include "mptcp_pm_gen.h"
17 
18 #define MPTCP_SUPPORTED_VERSION	1
19 
20 /* MPTCP option bits */
21 #define OPTION_MPTCP_MPC_SYN	BIT(0)
22 #define OPTION_MPTCP_MPC_SYNACK	BIT(1)
23 #define OPTION_MPTCP_MPC_ACK	BIT(2)
24 #define OPTION_MPTCP_MPJ_SYN	BIT(3)
25 #define OPTION_MPTCP_MPJ_SYNACK	BIT(4)
26 #define OPTION_MPTCP_MPJ_ACK	BIT(5)
27 #define OPTION_MPTCP_ADD_ADDR	BIT(6)
28 #define OPTION_MPTCP_RM_ADDR	BIT(7)
29 #define OPTION_MPTCP_FASTCLOSE	BIT(8)
30 #define OPTION_MPTCP_PRIO	BIT(9)
31 #define OPTION_MPTCP_RST	BIT(10)
32 #define OPTION_MPTCP_DSS	BIT(11)
33 #define OPTION_MPTCP_FAIL	BIT(12)
34 
35 #define OPTION_MPTCP_CSUMREQD	BIT(13)
36 
37 #define OPTIONS_MPTCP_MPC	(OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK | \
38 				 OPTION_MPTCP_MPC_ACK)
39 #define OPTIONS_MPTCP_MPJ	(OPTION_MPTCP_MPJ_SYN | OPTION_MPTCP_MPJ_SYNACK | \
40 				 OPTION_MPTCP_MPJ_ACK)
41 
42 /* MPTCP option subtypes */
43 #define MPTCPOPT_MP_CAPABLE	0
44 #define MPTCPOPT_MP_JOIN	1
45 #define MPTCPOPT_DSS		2
46 #define MPTCPOPT_ADD_ADDR	3
47 #define MPTCPOPT_RM_ADDR	4
48 #define MPTCPOPT_MP_PRIO	5
49 #define MPTCPOPT_MP_FAIL	6
50 #define MPTCPOPT_MP_FASTCLOSE	7
51 #define MPTCPOPT_RST		8
52 
53 /* MPTCP suboption lengths */
54 #define TCPOLEN_MPTCP_MPC_SYN		4
55 #define TCPOLEN_MPTCP_MPC_SYNACK	12
56 #define TCPOLEN_MPTCP_MPC_ACK		20
57 #define TCPOLEN_MPTCP_MPC_ACK_DATA	22
58 #define TCPOLEN_MPTCP_MPJ_SYN		12
59 #define TCPOLEN_MPTCP_MPJ_SYNACK	16
60 #define TCPOLEN_MPTCP_MPJ_ACK		24
61 #define TCPOLEN_MPTCP_DSS_BASE		4
62 #define TCPOLEN_MPTCP_DSS_ACK32		4
63 #define TCPOLEN_MPTCP_DSS_ACK64		8
64 #define TCPOLEN_MPTCP_DSS_MAP32		10
65 #define TCPOLEN_MPTCP_DSS_MAP64		14
66 #define TCPOLEN_MPTCP_DSS_CHECKSUM	2
67 #define TCPOLEN_MPTCP_ADD_ADDR		16
68 #define TCPOLEN_MPTCP_ADD_ADDR_PORT	18
69 #define TCPOLEN_MPTCP_ADD_ADDR_BASE	8
70 #define TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT	10
71 #define TCPOLEN_MPTCP_ADD_ADDR6		28
72 #define TCPOLEN_MPTCP_ADD_ADDR6_PORT	30
73 #define TCPOLEN_MPTCP_ADD_ADDR6_BASE	20
74 #define TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT	22
75 #define TCPOLEN_MPTCP_PORT_LEN		2
76 #define TCPOLEN_MPTCP_PORT_ALIGN	2
77 #define TCPOLEN_MPTCP_RM_ADDR_BASE	3
78 #define TCPOLEN_MPTCP_PRIO		3
79 #define TCPOLEN_MPTCP_PRIO_ALIGN	4
80 #define TCPOLEN_MPTCP_FASTCLOSE		12
81 #define TCPOLEN_MPTCP_RST		4
82 #define TCPOLEN_MPTCP_FAIL		12
83 
84 #define TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM	(TCPOLEN_MPTCP_DSS_CHECKSUM + TCPOLEN_MPTCP_MPC_ACK_DATA)
85 
86 /* MPTCP MP_JOIN flags */
87 #define MPTCPOPT_BACKUP		BIT(0)
88 #define MPTCPOPT_THMAC_LEN	8
89 
90 /* MPTCP MP_CAPABLE flags */
91 #define MPTCP_VERSION_MASK	(0x0F)
92 #define MPTCP_CAP_CHECKSUM_REQD	BIT(7)
93 #define MPTCP_CAP_EXTENSIBILITY	BIT(6)
94 #define MPTCP_CAP_DENY_JOIN_ID0	BIT(5)
95 #define MPTCP_CAP_HMAC_SHA256	BIT(0)
96 #define MPTCP_CAP_FLAG_MASK	(0x1F)
97 
98 /* MPTCP DSS flags */
99 #define MPTCP_DSS_DATA_FIN	BIT(4)
100 #define MPTCP_DSS_DSN64		BIT(3)
101 #define MPTCP_DSS_HAS_MAP	BIT(2)
102 #define MPTCP_DSS_ACK64		BIT(1)
103 #define MPTCP_DSS_HAS_ACK	BIT(0)
104 #define MPTCP_DSS_FLAG_MASK	(0x1F)
105 
106 /* MPTCP ADD_ADDR flags */
107 #define MPTCP_ADDR_ECHO		BIT(0)
108 
109 /* MPTCP MP_PRIO flags */
110 #define MPTCP_PRIO_BKUP		BIT(0)
111 
112 /* MPTCP TCPRST flags */
113 #define MPTCP_RST_TRANSIENT	BIT(0)
114 
115 /* MPTCP socket atomic flags */
116 #define MPTCP_NOSPACE		1
117 #define MPTCP_WORK_RTX		2
118 #define MPTCP_FALLBACK_DONE	4
119 #define MPTCP_WORK_CLOSE_SUBFLOW 5
120 
121 /* MPTCP socket release cb flags */
122 #define MPTCP_PUSH_PENDING	1
123 #define MPTCP_CLEAN_UNA		2
124 #define MPTCP_ERROR_REPORT	3
125 #define MPTCP_RETRANSMIT	4
126 #define MPTCP_FLUSH_JOIN_LIST	5
127 #define MPTCP_CONNECTED		6
128 #define MPTCP_SYNC_SNDBUF	7
129 
130 struct mptcp_skb_cb {
131 	u64 map_seq;
132 	u64 end_seq;
133 	u32 offset;
134 	u8  has_rxtstamp:1;
135 };
136 
137 #define MPTCP_SKB_CB(__skb)	((struct mptcp_skb_cb *)&((__skb)->cb[0]))
138 
139 static inline bool before64(__u64 seq1, __u64 seq2)
140 {
141 	return (__s64)(seq1 - seq2) < 0;
142 }
143 
144 #define after64(seq2, seq1)	before64(seq1, seq2)
145 
146 struct mptcp_options_received {
147 	u64	sndr_key;
148 	u64	rcvr_key;
149 	u64	data_ack;
150 	u64	data_seq;
151 	u32	subflow_seq;
152 	u16	data_len;
153 	__sum16	csum;
154 	u16	suboptions;
155 	u32	token;
156 	u32	nonce;
157 	u16	use_map:1,
158 		dsn64:1,
159 		data_fin:1,
160 		use_ack:1,
161 		ack64:1,
162 		mpc_map:1,
163 		reset_reason:4,
164 		reset_transient:1,
165 		echo:1,
166 		backup:1,
167 		deny_join_id0:1,
168 		__unused:2;
169 	u8	join_id;
170 	u64	thmac;
171 	u8	hmac[MPTCPOPT_HMAC_LEN];
172 	struct mptcp_addr_info addr;
173 	struct mptcp_rm_list rm_list;
174 	u64	ahmac;
175 	u64	fail_seq;
176 };
177 
178 static inline __be32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field)
179 {
180 	return htonl((TCPOPT_MPTCP << 24) | (len << 16) | (subopt << 12) |
181 		     ((nib & 0xF) << 8) | field);
182 }
183 
184 enum mptcp_pm_status {
185 	MPTCP_PM_ADD_ADDR_RECEIVED,
186 	MPTCP_PM_ADD_ADDR_SEND_ACK,
187 	MPTCP_PM_RM_ADDR_RECEIVED,
188 	MPTCP_PM_ESTABLISHED,
189 	MPTCP_PM_SUBFLOW_ESTABLISHED,
190 	MPTCP_PM_ALREADY_ESTABLISHED,	/* persistent status, set after ESTABLISHED event */
191 	MPTCP_PM_MPC_ENDPOINT_ACCOUNTED /* persistent status, set after MPC local address is
192 					 * accounted int id_avail_bitmap
193 					 */
194 };
195 
196 enum mptcp_pm_type {
197 	MPTCP_PM_TYPE_KERNEL = 0,
198 	MPTCP_PM_TYPE_USERSPACE,
199 
200 	__MPTCP_PM_TYPE_NR,
201 	__MPTCP_PM_TYPE_MAX = __MPTCP_PM_TYPE_NR - 1,
202 };
203 
204 /* Status bits below MPTCP_PM_ALREADY_ESTABLISHED need pm worker actions */
205 #define MPTCP_PM_WORK_MASK ((1 << MPTCP_PM_ALREADY_ESTABLISHED) - 1)
206 
207 enum mptcp_addr_signal_status {
208 	MPTCP_ADD_ADDR_SIGNAL,
209 	MPTCP_ADD_ADDR_ECHO,
210 	MPTCP_RM_ADDR_SIGNAL,
211 };
212 
213 /* max value of mptcp_addr_info.id */
214 #define MPTCP_PM_MAX_ADDR_ID		U8_MAX
215 
216 struct mptcp_pm_data {
217 	struct mptcp_addr_info local;
218 	struct mptcp_addr_info remote;
219 	struct list_head anno_list;
220 	struct list_head userspace_pm_local_addr_list;
221 
222 	spinlock_t	lock;		/*protects the whole PM data */
223 
224 	u8		addr_signal;
225 	bool		server_side;
226 	bool		work_pending;
227 	bool		accept_addr;
228 	bool		accept_subflow;
229 	bool		remote_deny_join_id0;
230 	u8		add_addr_signaled;
231 	u8		add_addr_accepted;
232 	u8		local_addr_used;
233 	u8		pm_type;
234 	u8		subflows;
235 	u8		status;
236 	DECLARE_BITMAP(id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
237 	struct mptcp_rm_list rm_list_tx;
238 	struct mptcp_rm_list rm_list_rx;
239 };
240 
241 struct mptcp_pm_addr_entry {
242 	struct list_head	list;
243 	struct mptcp_addr_info	addr;
244 	u8			flags;
245 	int			ifindex;
246 	struct socket		*lsk;
247 };
248 
249 struct mptcp_data_frag {
250 	struct list_head list;
251 	u64 data_seq;
252 	u16 data_len;
253 	u16 offset;
254 	u16 overhead;
255 	u16 already_sent;
256 	struct page *page;
257 };
258 
259 /* MPTCP connection sock */
260 struct mptcp_sock {
261 	/* inet_connection_sock must be the first member */
262 	struct inet_connection_sock sk;
263 	u64		local_key;
264 	u64		remote_key;
265 	u64		write_seq;
266 	u64		bytes_sent;
267 	u64		snd_nxt;
268 	u64		bytes_received;
269 	u64		ack_seq;
270 	atomic64_t	rcv_wnd_sent;
271 	u64		rcv_data_fin_seq;
272 	u64		bytes_retrans;
273 	u64		bytes_consumed;
274 	int		rmem_fwd_alloc;
275 	int		snd_burst;
276 	int		old_wspace;
277 	u64		recovery_snd_nxt;	/* in recovery mode accept up to this seq;
278 						 * recovery related fields are under data_lock
279 						 * protection
280 						 */
281 	u64		bytes_acked;
282 	u64		snd_una;
283 	u64		wnd_end;
284 	unsigned long	timer_ival;
285 	u32		token;
286 	int		rmem_released;
287 	unsigned long	flags;
288 	unsigned long	cb_flags;
289 	unsigned long	push_pending;
290 	bool		recovery;		/* closing subflow write queue reinjected */
291 	bool		can_ack;
292 	bool		fully_established;
293 	bool		rcv_data_fin;
294 	bool		snd_data_fin_enable;
295 	bool		rcv_fastclose;
296 	bool		use_64bit_ack; /* Set when we received a 64-bit DSN */
297 	bool		csum_enabled;
298 	bool		allow_infinite_fallback;
299 	u8		mpc_endpoint_id;
300 	u8		recvmsg_inq:1,
301 			cork:1,
302 			nodelay:1,
303 			fastopening:1,
304 			in_accept_queue:1,
305 			free_first:1;
306 	struct work_struct work;
307 	struct sk_buff  *ooo_last_skb;
308 	struct rb_root  out_of_order_queue;
309 	struct sk_buff_head receive_queue;
310 	struct list_head conn_list;
311 	struct list_head rtx_queue;
312 	struct mptcp_data_frag *first_pending;
313 	struct list_head join_list;
314 	struct sock	*first; /* The mptcp ops can safely dereference, using suitable
315 				 * ONCE annotation, the subflow outside the socket
316 				 * lock as such sock is freed after close().
317 				 */
318 	struct mptcp_pm_data	pm;
319 	struct mptcp_sched_ops	*sched;
320 	struct {
321 		u32	space;	/* bytes copied in last measurement window */
322 		u32	copied; /* bytes copied in this measurement window */
323 		u64	time;	/* start time of measurement window */
324 		u64	rtt_us; /* last maximum rtt of subflows */
325 	} rcvq_space;
326 	u8		scaling_ratio;
327 
328 	u32		subflow_id;
329 	u32		setsockopt_seq;
330 	char		ca_name[TCP_CA_NAME_MAX];
331 };
332 
333 #define mptcp_data_lock(sk) spin_lock_bh(&(sk)->sk_lock.slock)
334 #define mptcp_data_unlock(sk) spin_unlock_bh(&(sk)->sk_lock.slock)
335 
336 #define mptcp_for_each_subflow(__msk, __subflow)			\
337 	list_for_each_entry(__subflow, &((__msk)->conn_list), node)
338 #define mptcp_for_each_subflow_safe(__msk, __subflow, __tmp)			\
339 	list_for_each_entry_safe(__subflow, __tmp, &((__msk)->conn_list), node)
340 
341 static inline void msk_owned_by_me(const struct mptcp_sock *msk)
342 {
343 	sock_owned_by_me((const struct sock *)msk);
344 }
345 
346 #define mptcp_sk(ptr) container_of_const(ptr, struct mptcp_sock, sk.icsk_inet.sk)
347 
348 /* the msk socket don't use the backlog, also account for the bulk
349  * free memory
350  */
351 static inline int __mptcp_rmem(const struct sock *sk)
352 {
353 	return atomic_read(&sk->sk_rmem_alloc) - READ_ONCE(mptcp_sk(sk)->rmem_released);
354 }
355 
356 static inline int mptcp_win_from_space(const struct sock *sk, int space)
357 {
358 	return __tcp_win_from_space(mptcp_sk(sk)->scaling_ratio, space);
359 }
360 
361 static inline int __mptcp_space(const struct sock *sk)
362 {
363 	return mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - __mptcp_rmem(sk));
364 }
365 
366 static inline struct mptcp_data_frag *mptcp_send_head(const struct sock *sk)
367 {
368 	const struct mptcp_sock *msk = mptcp_sk(sk);
369 
370 	return READ_ONCE(msk->first_pending);
371 }
372 
373 static inline struct mptcp_data_frag *mptcp_send_next(struct sock *sk)
374 {
375 	struct mptcp_sock *msk = mptcp_sk(sk);
376 	struct mptcp_data_frag *cur;
377 
378 	cur = msk->first_pending;
379 	return list_is_last(&cur->list, &msk->rtx_queue) ? NULL :
380 						     list_next_entry(cur, list);
381 }
382 
383 static inline struct mptcp_data_frag *mptcp_pending_tail(const struct sock *sk)
384 {
385 	const struct mptcp_sock *msk = mptcp_sk(sk);
386 
387 	if (!msk->first_pending)
388 		return NULL;
389 
390 	if (WARN_ON_ONCE(list_empty(&msk->rtx_queue)))
391 		return NULL;
392 
393 	return list_last_entry(&msk->rtx_queue, struct mptcp_data_frag, list);
394 }
395 
396 static inline struct mptcp_data_frag *mptcp_rtx_head(struct sock *sk)
397 {
398 	struct mptcp_sock *msk = mptcp_sk(sk);
399 
400 	if (msk->snd_una == READ_ONCE(msk->snd_nxt))
401 		return NULL;
402 
403 	return list_first_entry_or_null(&msk->rtx_queue, struct mptcp_data_frag, list);
404 }
405 
406 struct csum_pseudo_header {
407 	__be64 data_seq;
408 	__be32 subflow_seq;
409 	__be16 data_len;
410 	__sum16 csum;
411 };
412 
413 struct mptcp_subflow_request_sock {
414 	struct	tcp_request_sock sk;
415 	u16	mp_capable : 1,
416 		mp_join : 1,
417 		backup : 1,
418 		csum_reqd : 1,
419 		allow_join_id0 : 1;
420 	u8	local_id;
421 	u8	remote_id;
422 	u64	local_key;
423 	u64	idsn;
424 	u32	token;
425 	u32	ssn_offset;
426 	u64	thmac;
427 	u32	local_nonce;
428 	u32	remote_nonce;
429 	struct mptcp_sock	*msk;
430 	struct hlist_nulls_node token_node;
431 };
432 
433 static inline struct mptcp_subflow_request_sock *
434 mptcp_subflow_rsk(const struct request_sock *rsk)
435 {
436 	return (struct mptcp_subflow_request_sock *)rsk;
437 }
438 
439 struct mptcp_delegated_action {
440 	struct napi_struct napi;
441 	struct list_head head;
442 };
443 
444 DECLARE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
445 
446 #define MPTCP_DELEGATE_SCHEDULED	0
447 #define MPTCP_DELEGATE_SEND		1
448 #define MPTCP_DELEGATE_ACK		2
449 #define MPTCP_DELEGATE_SNDBUF		3
450 
451 #define MPTCP_DELEGATE_ACTIONS_MASK	(~BIT(MPTCP_DELEGATE_SCHEDULED))
452 /* MPTCP subflow context */
453 struct mptcp_subflow_context {
454 	struct	list_head node;/* conn_list of subflows */
455 
456 	struct_group(reset,
457 
458 	unsigned long avg_pacing_rate; /* protected by msk socket lock */
459 	u64	local_key;
460 	u64	remote_key;
461 	u64	idsn;
462 	u64	map_seq;
463 	u32	snd_isn;
464 	u32	token;
465 	u32	rel_write_seq;
466 	u32	map_subflow_seq;
467 	u32	ssn_offset;
468 	u32	map_data_len;
469 	__wsum	map_data_csum;
470 	u32	map_csum_len;
471 	u32	request_mptcp : 1,  /* send MP_CAPABLE */
472 		request_join : 1,   /* send MP_JOIN */
473 		request_bkup : 1,
474 		mp_capable : 1,	    /* remote is MPTCP capable */
475 		mp_join : 1,	    /* remote is JOINing */
476 		fully_established : 1,	    /* path validated */
477 		pm_notified : 1,    /* PM hook called for established status */
478 		conn_finished : 1,
479 		map_valid : 1,
480 		map_csum_reqd : 1,
481 		map_data_fin : 1,
482 		mpc_map : 1,
483 		backup : 1,
484 		send_mp_prio : 1,
485 		send_mp_fail : 1,
486 		send_fastclose : 1,
487 		send_infinite_map : 1,
488 		remote_key_valid : 1,        /* received the peer key from */
489 		disposable : 1,	    /* ctx can be free at ulp release time */
490 		stale : 1,	    /* unable to snd/rcv data, do not use for xmit */
491 		local_id_valid : 1, /* local_id is correctly initialized */
492 		valid_csum_seen : 1,        /* at least one csum validated */
493 		is_mptfo : 1,	    /* subflow is doing TFO */
494 		__unused : 9;
495 	bool	data_avail;
496 	bool	scheduled;
497 	u32	remote_nonce;
498 	u64	thmac;
499 	u32	local_nonce;
500 	u32	remote_token;
501 	union {
502 		u8	hmac[MPTCPOPT_HMAC_LEN]; /* MPJ subflow only */
503 		u64	iasn;	    /* initial ack sequence number, MPC subflows only */
504 	};
505 	u8	local_id;
506 	u8	remote_id;
507 	u8	reset_seen:1;
508 	u8	reset_transient:1;
509 	u8	reset_reason:4;
510 	u8	stale_count;
511 
512 	u32	subflow_id;
513 
514 	long	delegated_status;
515 	unsigned long	fail_tout;
516 
517 	);
518 
519 	struct	list_head delegated_node;   /* link into delegated_action, protected by local BH */
520 
521 	u32	setsockopt_seq;
522 	u32	stale_rcv_tstamp;
523 	int     cached_sndbuf;	    /* sndbuf size when last synced with the msk sndbuf,
524 				     * protected by the msk socket lock
525 				     */
526 
527 	struct	sock *tcp_sock;	    /* tcp sk backpointer */
528 	struct	sock *conn;	    /* parent mptcp_sock */
529 	const	struct inet_connection_sock_af_ops *icsk_af_ops;
530 	void	(*tcp_state_change)(struct sock *sk);
531 	void	(*tcp_error_report)(struct sock *sk);
532 
533 	struct	rcu_head rcu;
534 };
535 
536 static inline struct mptcp_subflow_context *
537 mptcp_subflow_ctx(const struct sock *sk)
538 {
539 	struct inet_connection_sock *icsk = inet_csk(sk);
540 
541 	/* Use RCU on icsk_ulp_data only for sock diag code */
542 	return (__force struct mptcp_subflow_context *)icsk->icsk_ulp_data;
543 }
544 
545 static inline struct sock *
546 mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow)
547 {
548 	return subflow->tcp_sock;
549 }
550 
551 static inline void
552 mptcp_subflow_ctx_reset(struct mptcp_subflow_context *subflow)
553 {
554 	memset(&subflow->reset, 0, sizeof(subflow->reset));
555 	subflow->request_mptcp = 1;
556 }
557 
558 static inline u64
559 mptcp_subflow_get_map_offset(const struct mptcp_subflow_context *subflow)
560 {
561 	return tcp_sk(mptcp_subflow_tcp_sock(subflow))->copied_seq -
562 		      subflow->ssn_offset -
563 		      subflow->map_subflow_seq;
564 }
565 
566 static inline u64
567 mptcp_subflow_get_mapped_dsn(const struct mptcp_subflow_context *subflow)
568 {
569 	return subflow->map_seq + mptcp_subflow_get_map_offset(subflow);
570 }
571 
572 void mptcp_subflow_process_delegated(struct sock *ssk, long actions);
573 
574 static inline void mptcp_subflow_delegate(struct mptcp_subflow_context *subflow, int action)
575 {
576 	long old, set_bits = BIT(MPTCP_DELEGATE_SCHEDULED) | BIT(action);
577 	struct mptcp_delegated_action *delegated;
578 	bool schedule;
579 
580 	/* the caller held the subflow bh socket lock */
581 	lockdep_assert_in_softirq();
582 
583 	/* The implied barrier pairs with tcp_release_cb_override()
584 	 * mptcp_napi_poll(), and ensures the below list check sees list
585 	 * updates done prior to delegated status bits changes
586 	 */
587 	old = set_mask_bits(&subflow->delegated_status, 0, set_bits);
588 	if (!(old & BIT(MPTCP_DELEGATE_SCHEDULED))) {
589 		if (WARN_ON_ONCE(!list_empty(&subflow->delegated_node)))
590 			return;
591 
592 		delegated = this_cpu_ptr(&mptcp_delegated_actions);
593 		schedule = list_empty(&delegated->head);
594 		list_add_tail(&subflow->delegated_node, &delegated->head);
595 		sock_hold(mptcp_subflow_tcp_sock(subflow));
596 		if (schedule)
597 			napi_schedule(&delegated->napi);
598 	}
599 }
600 
601 static inline struct mptcp_subflow_context *
602 mptcp_subflow_delegated_next(struct mptcp_delegated_action *delegated)
603 {
604 	struct mptcp_subflow_context *ret;
605 
606 	if (list_empty(&delegated->head))
607 		return NULL;
608 
609 	ret = list_first_entry(&delegated->head, struct mptcp_subflow_context, delegated_node);
610 	list_del_init(&ret->delegated_node);
611 	return ret;
612 }
613 
614 int mptcp_is_enabled(const struct net *net);
615 unsigned int mptcp_get_add_addr_timeout(const struct net *net);
616 int mptcp_is_checksum_enabled(const struct net *net);
617 int mptcp_allow_join_id0(const struct net *net);
618 unsigned int mptcp_stale_loss_cnt(const struct net *net);
619 unsigned int mptcp_close_timeout(const struct sock *sk);
620 int mptcp_get_pm_type(const struct net *net);
621 const char *mptcp_get_scheduler(const struct net *net);
622 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
623 				     const struct mptcp_options_received *mp_opt);
624 bool __mptcp_retransmit_pending_data(struct sock *sk);
625 void mptcp_check_and_set_pending(struct sock *sk);
626 void __mptcp_push_pending(struct sock *sk, unsigned int flags);
627 bool mptcp_subflow_data_available(struct sock *sk);
628 void __init mptcp_subflow_init(void);
629 void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how);
630 void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
631 		     struct mptcp_subflow_context *subflow);
632 void __mptcp_subflow_send_ack(struct sock *ssk);
633 void mptcp_subflow_reset(struct sock *ssk);
634 void mptcp_subflow_queue_clean(struct sock *sk, struct sock *ssk);
635 void mptcp_sock_graft(struct sock *sk, struct socket *parent);
636 struct sock *__mptcp_nmpc_sk(struct mptcp_sock *msk);
637 bool __mptcp_close(struct sock *sk, long timeout);
638 void mptcp_cancel_work(struct sock *sk);
639 void __mptcp_unaccepted_force_close(struct sock *sk);
640 void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk);
641 
642 bool mptcp_addresses_equal(const struct mptcp_addr_info *a,
643 			   const struct mptcp_addr_info *b, bool use_port);
644 void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr);
645 
646 /* called with sk socket lock held */
647 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
648 			    const struct mptcp_addr_info *remote);
649 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
650 				struct socket **new_sock);
651 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
652 			 struct sockaddr_storage *addr,
653 			 unsigned short family);
654 struct mptcp_sched_ops *mptcp_sched_find(const char *name);
655 int mptcp_register_scheduler(struct mptcp_sched_ops *sched);
656 void mptcp_unregister_scheduler(struct mptcp_sched_ops *sched);
657 void mptcp_sched_init(void);
658 int mptcp_init_sched(struct mptcp_sock *msk,
659 		     struct mptcp_sched_ops *sched);
660 void mptcp_release_sched(struct mptcp_sock *msk);
661 void mptcp_subflow_set_scheduled(struct mptcp_subflow_context *subflow,
662 				 bool scheduled);
663 struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk);
664 struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk);
665 int mptcp_sched_get_send(struct mptcp_sock *msk);
666 int mptcp_sched_get_retrans(struct mptcp_sock *msk);
667 
668 static inline u64 mptcp_data_avail(const struct mptcp_sock *msk)
669 {
670 	return READ_ONCE(msk->bytes_received) - READ_ONCE(msk->bytes_consumed);
671 }
672 
673 static inline bool mptcp_epollin_ready(const struct sock *sk)
674 {
675 	/* mptcp doesn't have to deal with small skbs in the receive queue,
676 	 * at it can always coalesce them
677 	 */
678 	return (mptcp_data_avail(mptcp_sk(sk)) >= sk->sk_rcvlowat) ||
679 	       (mem_cgroup_sockets_enabled && sk->sk_memcg &&
680 		mem_cgroup_under_socket_pressure(sk->sk_memcg)) ||
681 	       READ_ONCE(tcp_memory_pressure);
682 }
683 
684 int mptcp_set_rcvlowat(struct sock *sk, int val);
685 
686 static inline bool __tcp_can_send(const struct sock *ssk)
687 {
688 	/* only send if our side has not closed yet */
689 	return ((1 << inet_sk_state_load(ssk)) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT));
690 }
691 
692 static inline bool __mptcp_subflow_active(struct mptcp_subflow_context *subflow)
693 {
694 	/* can't send if JOIN hasn't completed yet (i.e. is usable for mptcp) */
695 	if (subflow->request_join && !subflow->fully_established)
696 		return false;
697 
698 	return __tcp_can_send(mptcp_subflow_tcp_sock(subflow));
699 }
700 
701 void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow);
702 
703 bool mptcp_subflow_active(struct mptcp_subflow_context *subflow);
704 
705 void mptcp_subflow_drop_ctx(struct sock *ssk);
706 
707 static inline void mptcp_subflow_tcp_fallback(struct sock *sk,
708 					      struct mptcp_subflow_context *ctx)
709 {
710 	sk->sk_data_ready = sock_def_readable;
711 	sk->sk_state_change = ctx->tcp_state_change;
712 	sk->sk_write_space = sk_stream_write_space;
713 	sk->sk_error_report = ctx->tcp_error_report;
714 
715 	inet_csk(sk)->icsk_af_ops = ctx->icsk_af_ops;
716 }
717 
718 void __init mptcp_proto_init(void);
719 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
720 int __init mptcp_proto_v6_init(void);
721 #endif
722 
723 struct sock *mptcp_sk_clone_init(const struct sock *sk,
724 				 const struct mptcp_options_received *mp_opt,
725 				 struct sock *ssk,
726 				 struct request_sock *req);
727 void mptcp_get_options(const struct sk_buff *skb,
728 		       struct mptcp_options_received *mp_opt);
729 
730 void mptcp_finish_connect(struct sock *sk);
731 void __mptcp_set_connected(struct sock *sk);
732 void mptcp_reset_tout_timer(struct mptcp_sock *msk, unsigned long fail_tout);
733 
734 static inline void mptcp_stop_tout_timer(struct sock *sk)
735 {
736 	if (!inet_csk(sk)->icsk_mtup.probe_timestamp)
737 		return;
738 
739 	sk_stop_timer(sk, &sk->sk_timer);
740 	inet_csk(sk)->icsk_mtup.probe_timestamp = 0;
741 }
742 
743 static inline void mptcp_set_close_tout(struct sock *sk, unsigned long tout)
744 {
745 	/* avoid 0 timestamp, as that means no close timeout */
746 	inet_csk(sk)->icsk_mtup.probe_timestamp = tout ? : 1;
747 }
748 
749 static inline void mptcp_start_tout_timer(struct sock *sk)
750 {
751 	mptcp_set_close_tout(sk, tcp_jiffies32);
752 	mptcp_reset_tout_timer(mptcp_sk(sk), 0);
753 }
754 
755 static inline bool mptcp_is_fully_established(struct sock *sk)
756 {
757 	return inet_sk_state_load(sk) == TCP_ESTABLISHED &&
758 	       READ_ONCE(mptcp_sk(sk)->fully_established);
759 }
760 
761 void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk);
762 void mptcp_data_ready(struct sock *sk, struct sock *ssk);
763 bool mptcp_finish_join(struct sock *sk);
764 bool mptcp_schedule_work(struct sock *sk);
765 int mptcp_setsockopt(struct sock *sk, int level, int optname,
766 		     sockptr_t optval, unsigned int optlen);
767 int mptcp_getsockopt(struct sock *sk, int level, int optname,
768 		     char __user *optval, int __user *option);
769 
770 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq);
771 static inline u64 mptcp_expand_seq(u64 old_seq, u64 cur_seq, bool use_64bit)
772 {
773 	if (use_64bit)
774 		return cur_seq;
775 
776 	return __mptcp_expand_seq(old_seq, cur_seq);
777 }
778 void __mptcp_check_push(struct sock *sk, struct sock *ssk);
779 void __mptcp_data_acked(struct sock *sk);
780 void __mptcp_error_report(struct sock *sk);
781 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit);
782 static inline bool mptcp_data_fin_enabled(const struct mptcp_sock *msk)
783 {
784 	return READ_ONCE(msk->snd_data_fin_enable) &&
785 	       READ_ONCE(msk->write_seq) == READ_ONCE(msk->snd_nxt);
786 }
787 
788 static inline void __mptcp_sync_sndbuf(struct sock *sk)
789 {
790 	struct mptcp_subflow_context *subflow;
791 	int ssk_sndbuf, new_sndbuf;
792 
793 	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
794 		return;
795 
796 	new_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[0];
797 	mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
798 		ssk_sndbuf =  READ_ONCE(mptcp_subflow_tcp_sock(subflow)->sk_sndbuf);
799 
800 		subflow->cached_sndbuf = ssk_sndbuf;
801 		new_sndbuf += ssk_sndbuf;
802 	}
803 
804 	/* the msk max wmem limit is <nr_subflows> * tcp wmem[2] */
805 	WRITE_ONCE(sk->sk_sndbuf, new_sndbuf);
806 }
807 
808 /* The called held both the msk socket and the subflow socket locks,
809  * possibly under BH
810  */
811 static inline void __mptcp_propagate_sndbuf(struct sock *sk, struct sock *ssk)
812 {
813 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
814 
815 	if (READ_ONCE(ssk->sk_sndbuf) != subflow->cached_sndbuf)
816 		__mptcp_sync_sndbuf(sk);
817 }
818 
819 /* the caller held only the subflow socket lock, either in process or
820  * BH context. Additionally this can be called under the msk data lock,
821  * so we can't acquire such lock here: let the delegate action acquires
822  * the needed locks in suitable order.
823  */
824 static inline void mptcp_propagate_sndbuf(struct sock *sk, struct sock *ssk)
825 {
826 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
827 
828 	if (likely(READ_ONCE(ssk->sk_sndbuf) == subflow->cached_sndbuf))
829 		return;
830 
831 	local_bh_disable();
832 	mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_SNDBUF);
833 	local_bh_enable();
834 }
835 
836 static inline void mptcp_write_space(struct sock *sk)
837 {
838 	if (sk_stream_is_writeable(sk)) {
839 		/* pairs with memory barrier in mptcp_poll */
840 		smp_mb();
841 		if (test_and_clear_bit(MPTCP_NOSPACE, &mptcp_sk(sk)->flags))
842 			sk_stream_write_space(sk);
843 	}
844 }
845 
846 void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags);
847 
848 #define MPTCP_TOKEN_MAX_RETRIES	4
849 
850 void __init mptcp_token_init(void);
851 static inline void mptcp_token_init_request(struct request_sock *req)
852 {
853 	mptcp_subflow_rsk(req)->token_node.pprev = NULL;
854 }
855 
856 int mptcp_token_new_request(struct request_sock *req);
857 void mptcp_token_destroy_request(struct request_sock *req);
858 int mptcp_token_new_connect(struct sock *ssk);
859 void mptcp_token_accept(struct mptcp_subflow_request_sock *r,
860 			struct mptcp_sock *msk);
861 bool mptcp_token_exists(u32 token);
862 struct mptcp_sock *mptcp_token_get_sock(struct net *net, u32 token);
863 struct mptcp_sock *mptcp_token_iter_next(const struct net *net, long *s_slot,
864 					 long *s_num);
865 void mptcp_token_destroy(struct mptcp_sock *msk);
866 
867 void mptcp_crypto_key_sha(u64 key, u32 *token, u64 *idsn);
868 
869 void mptcp_crypto_hmac_sha(u64 key1, u64 key2, u8 *msg, int len, void *hmac);
870 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum);
871 
872 void __init mptcp_pm_init(void);
873 void mptcp_pm_data_init(struct mptcp_sock *msk);
874 void mptcp_pm_data_reset(struct mptcp_sock *msk);
875 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
876 			struct mptcp_addr_info *addr);
877 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
878 			 bool require_family,
879 			 struct mptcp_pm_addr_entry *entry);
880 bool mptcp_pm_addr_families_match(const struct sock *sk,
881 				  const struct mptcp_addr_info *loc,
882 				  const struct mptcp_addr_info *rem);
883 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk);
884 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk);
885 void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side);
886 void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk);
887 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk);
888 void mptcp_pm_connection_closed(struct mptcp_sock *msk);
889 void mptcp_pm_subflow_established(struct mptcp_sock *msk);
890 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk);
891 void mptcp_pm_subflow_check_next(struct mptcp_sock *msk,
892 				 const struct mptcp_subflow_context *subflow);
893 void mptcp_pm_add_addr_received(const struct sock *ssk,
894 				const struct mptcp_addr_info *addr);
895 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,
896 			      const struct mptcp_addr_info *addr);
897 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk);
898 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk);
899 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk,
900 			       const struct mptcp_rm_list *rm_list);
901 void mptcp_pm_mp_prio_received(struct sock *sk, u8 bkup);
902 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq);
903 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
904 				 struct mptcp_addr_info *addr,
905 				 struct mptcp_addr_info *rem,
906 				 u8 bkup);
907 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
908 			      const struct mptcp_addr_info *addr);
909 void mptcp_pm_free_anno_list(struct mptcp_sock *msk);
910 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk);
911 struct mptcp_pm_add_entry *
912 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
913 		       const struct mptcp_addr_info *addr, bool check_id);
914 struct mptcp_pm_add_entry *
915 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk,
916 				const struct mptcp_addr_info *addr);
917 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk,
918 					 unsigned int id,
919 					 u8 *flags, int *ifindex);
920 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
921 					    u8 *flags, int *ifindex);
922 int mptcp_userspace_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk,
923 						   unsigned int id,
924 						   u8 *flags, int *ifindex);
925 int mptcp_pm_set_flags(struct net *net, struct nlattr *token,
926 		       struct mptcp_pm_addr_entry *loc,
927 		       struct mptcp_pm_addr_entry *rem, u8 bkup);
928 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup);
929 int mptcp_userspace_pm_set_flags(struct net *net, struct nlattr *token,
930 				 struct mptcp_pm_addr_entry *loc,
931 				 struct mptcp_pm_addr_entry *rem, u8 bkup);
932 int mptcp_pm_announce_addr(struct mptcp_sock *msk,
933 			   const struct mptcp_addr_info *addr,
934 			   bool echo);
935 int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list);
936 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list);
937 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list);
938 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
939 					struct list_head *rm_list);
940 
941 void mptcp_free_local_addr_list(struct mptcp_sock *msk);
942 
943 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
944 		 const struct sock *ssk, gfp_t gfp);
945 void mptcp_event_addr_announced(const struct sock *ssk, const struct mptcp_addr_info *info);
946 void mptcp_event_addr_removed(const struct mptcp_sock *msk, u8 id);
947 void mptcp_event_pm_listener(const struct sock *ssk,
948 			     enum mptcp_event_type event);
949 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk);
950 
951 void mptcp_fastopen_gen_msk_ackseq(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
952 				   const struct mptcp_options_received *mp_opt);
953 void mptcp_fastopen_subflow_synack_set_params(struct mptcp_subflow_context *subflow,
954 					      struct request_sock *req);
955 
956 static inline bool mptcp_pm_should_add_signal(struct mptcp_sock *msk)
957 {
958 	return READ_ONCE(msk->pm.addr_signal) &
959 		(BIT(MPTCP_ADD_ADDR_SIGNAL) | BIT(MPTCP_ADD_ADDR_ECHO));
960 }
961 
962 static inline bool mptcp_pm_should_add_signal_addr(struct mptcp_sock *msk)
963 {
964 	return READ_ONCE(msk->pm.addr_signal) & BIT(MPTCP_ADD_ADDR_SIGNAL);
965 }
966 
967 static inline bool mptcp_pm_should_add_signal_echo(struct mptcp_sock *msk)
968 {
969 	return READ_ONCE(msk->pm.addr_signal) & BIT(MPTCP_ADD_ADDR_ECHO);
970 }
971 
972 static inline bool mptcp_pm_should_rm_signal(struct mptcp_sock *msk)
973 {
974 	return READ_ONCE(msk->pm.addr_signal) & BIT(MPTCP_RM_ADDR_SIGNAL);
975 }
976 
977 static inline bool mptcp_pm_is_userspace(const struct mptcp_sock *msk)
978 {
979 	return READ_ONCE(msk->pm.pm_type) == MPTCP_PM_TYPE_USERSPACE;
980 }
981 
982 static inline bool mptcp_pm_is_kernel(const struct mptcp_sock *msk)
983 {
984 	return READ_ONCE(msk->pm.pm_type) == MPTCP_PM_TYPE_KERNEL;
985 }
986 
987 static inline unsigned int mptcp_add_addr_len(int family, bool echo, bool port)
988 {
989 	u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
990 
991 	if (family == AF_INET6)
992 		len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
993 	if (!echo)
994 		len += MPTCPOPT_THMAC_LEN;
995 	/* account for 2 trailing 'nop' options */
996 	if (port)
997 		len += TCPOLEN_MPTCP_PORT_LEN + TCPOLEN_MPTCP_PORT_ALIGN;
998 
999 	return len;
1000 }
1001 
1002 static inline int mptcp_rm_addr_len(const struct mptcp_rm_list *rm_list)
1003 {
1004 	if (rm_list->nr == 0 || rm_list->nr > MPTCP_RM_IDS_MAX)
1005 		return -EINVAL;
1006 
1007 	return TCPOLEN_MPTCP_RM_ADDR_BASE + roundup(rm_list->nr - 1, 4) + 1;
1008 }
1009 
1010 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, const struct sk_buff *skb,
1011 			      unsigned int opt_size, unsigned int remaining,
1012 			      struct mptcp_addr_info *addr, bool *echo,
1013 			      bool *drop_other_suboptions);
1014 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
1015 			     struct mptcp_rm_list *rm_list);
1016 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
1017 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc);
1018 int mptcp_userspace_pm_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc);
1019 
1020 void __init mptcp_pm_nl_init(void);
1021 void mptcp_pm_nl_work(struct mptcp_sock *msk);
1022 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
1023 				     const struct mptcp_rm_list *rm_list);
1024 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk);
1025 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk);
1026 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk);
1027 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk);
1028 
1029 /* called under PM lock */
1030 static inline void __mptcp_pm_close_subflow(struct mptcp_sock *msk)
1031 {
1032 	if (--msk->pm.subflows < mptcp_pm_get_subflows_max(msk))
1033 		WRITE_ONCE(msk->pm.accept_subflow, true);
1034 }
1035 
1036 static inline void mptcp_pm_close_subflow(struct mptcp_sock *msk)
1037 {
1038 	spin_lock_bh(&msk->pm.lock);
1039 	__mptcp_pm_close_subflow(msk);
1040 	spin_unlock_bh(&msk->pm.lock);
1041 }
1042 
1043 void mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk);
1044 void mptcp_sockopt_sync_locked(struct mptcp_sock *msk, struct sock *ssk);
1045 
1046 static inline struct mptcp_ext *mptcp_get_ext(const struct sk_buff *skb)
1047 {
1048 	return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP);
1049 }
1050 
1051 void mptcp_diag_subflow_init(struct tcp_ulp_ops *ops);
1052 
1053 static inline bool __mptcp_check_fallback(const struct mptcp_sock *msk)
1054 {
1055 	return test_bit(MPTCP_FALLBACK_DONE, &msk->flags);
1056 }
1057 
1058 static inline bool mptcp_check_fallback(const struct sock *sk)
1059 {
1060 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1061 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1062 
1063 	return __mptcp_check_fallback(msk);
1064 }
1065 
1066 static inline void __mptcp_do_fallback(struct mptcp_sock *msk)
1067 {
1068 	if (__mptcp_check_fallback(msk)) {
1069 		pr_debug("TCP fallback already done (msk=%p)", msk);
1070 		return;
1071 	}
1072 	set_bit(MPTCP_FALLBACK_DONE, &msk->flags);
1073 }
1074 
1075 static inline bool __mptcp_has_initial_subflow(const struct mptcp_sock *msk)
1076 {
1077 	struct sock *ssk = READ_ONCE(msk->first);
1078 
1079 	return ssk && ((1 << inet_sk_state_load(ssk)) &
1080 		       (TCPF_ESTABLISHED | TCPF_SYN_SENT |
1081 			TCPF_SYN_RECV | TCPF_LISTEN));
1082 }
1083 
1084 static inline void mptcp_do_fallback(struct sock *ssk)
1085 {
1086 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1087 	struct sock *sk = subflow->conn;
1088 	struct mptcp_sock *msk;
1089 
1090 	msk = mptcp_sk(sk);
1091 	__mptcp_do_fallback(msk);
1092 	if (READ_ONCE(msk->snd_data_fin_enable) && !(ssk->sk_shutdown & SEND_SHUTDOWN)) {
1093 		gfp_t saved_allocation = ssk->sk_allocation;
1094 
1095 		/* we are in a atomic (BH) scope, override ssk default for data
1096 		 * fin allocation
1097 		 */
1098 		ssk->sk_allocation = GFP_ATOMIC;
1099 		ssk->sk_shutdown |= SEND_SHUTDOWN;
1100 		tcp_shutdown(ssk, SEND_SHUTDOWN);
1101 		ssk->sk_allocation = saved_allocation;
1102 	}
1103 }
1104 
1105 #define pr_fallback(a) pr_debug("%s:fallback to TCP (msk=%p)", __func__, a)
1106 
1107 static inline bool mptcp_check_infinite_map(struct sk_buff *skb)
1108 {
1109 	struct mptcp_ext *mpext;
1110 
1111 	mpext = skb ? mptcp_get_ext(skb) : NULL;
1112 	if (mpext && mpext->infinite_map)
1113 		return true;
1114 
1115 	return false;
1116 }
1117 
1118 static inline bool is_active_ssk(struct mptcp_subflow_context *subflow)
1119 {
1120 	return (subflow->request_mptcp || subflow->request_join);
1121 }
1122 
1123 static inline bool subflow_simultaneous_connect(struct sock *sk)
1124 {
1125 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1126 
1127 	return sk->sk_state == TCP_ESTABLISHED &&
1128 	       is_active_ssk(subflow) &&
1129 	       !subflow->conn_finished;
1130 }
1131 
1132 #ifdef CONFIG_SYN_COOKIES
1133 void subflow_init_req_cookie_join_save(const struct mptcp_subflow_request_sock *subflow_req,
1134 				       struct sk_buff *skb);
1135 bool mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subflow_req,
1136 					struct sk_buff *skb);
1137 void __init mptcp_join_cookie_init(void);
1138 #else
1139 static inline void
1140 subflow_init_req_cookie_join_save(const struct mptcp_subflow_request_sock *subflow_req,
1141 				  struct sk_buff *skb) {}
1142 static inline bool
1143 mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subflow_req,
1144 				   struct sk_buff *skb)
1145 {
1146 	return false;
1147 }
1148 
1149 static inline void mptcp_join_cookie_init(void) {}
1150 #endif
1151 
1152 #endif /* __MPTCP_PROTOCOL_H */
1153