1 /* SPDX-License-Identifier: GPL-2.0 */
4 * Copyright (c) 2017 - 2019, Intel Corporation.
7 #ifndef __MPTCP_PROTOCOL_H
8 #define __MPTCP_PROTOCOL_H
10 #include <linux/random.h>
12 #include <net/inet_connection_sock.h>
14 #define MPTCP_SUPPORTED_VERSION 1
16 /* MPTCP option bits */
17 #define OPTION_MPTCP_MPC_SYN BIT(0)
18 #define OPTION_MPTCP_MPC_SYNACK BIT(1)
19 #define OPTION_MPTCP_MPC_ACK BIT(2)
20 #define OPTION_MPTCP_MPJ_SYN BIT(3)
21 #define OPTION_MPTCP_MPJ_SYNACK BIT(4)
22 #define OPTION_MPTCP_MPJ_ACK BIT(5)
23 #define OPTION_MPTCP_ADD_ADDR BIT(6)
24 #define OPTION_MPTCP_ADD_ADDR6 BIT(7)
25 #define OPTION_MPTCP_RM_ADDR BIT(8)
27 /* MPTCP option subtypes */
28 #define MPTCPOPT_MP_CAPABLE 0
29 #define MPTCPOPT_MP_JOIN 1
30 #define MPTCPOPT_DSS 2
31 #define MPTCPOPT_ADD_ADDR 3
32 #define MPTCPOPT_RM_ADDR 4
33 #define MPTCPOPT_MP_PRIO 5
34 #define MPTCPOPT_MP_FAIL 6
35 #define MPTCPOPT_MP_FASTCLOSE 7
37 /* MPTCP suboption lengths */
38 #define TCPOLEN_MPTCP_MPC_SYN 4
39 #define TCPOLEN_MPTCP_MPC_SYNACK 12
40 #define TCPOLEN_MPTCP_MPC_ACK 20
41 #define TCPOLEN_MPTCP_MPC_ACK_DATA 22
42 #define TCPOLEN_MPTCP_MPJ_SYN 12
43 #define TCPOLEN_MPTCP_MPJ_SYNACK 16
44 #define TCPOLEN_MPTCP_MPJ_ACK 24
45 #define TCPOLEN_MPTCP_DSS_BASE 4
46 #define TCPOLEN_MPTCP_DSS_ACK32 4
47 #define TCPOLEN_MPTCP_DSS_ACK64 8
48 #define TCPOLEN_MPTCP_DSS_MAP32 10
49 #define TCPOLEN_MPTCP_DSS_MAP64 14
50 #define TCPOLEN_MPTCP_DSS_CHECKSUM 2
51 #define TCPOLEN_MPTCP_ADD_ADDR 16
52 #define TCPOLEN_MPTCP_ADD_ADDR_PORT 18
53 #define TCPOLEN_MPTCP_ADD_ADDR_BASE 8
54 #define TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT 10
55 #define TCPOLEN_MPTCP_ADD_ADDR6 28
56 #define TCPOLEN_MPTCP_ADD_ADDR6_PORT 30
57 #define TCPOLEN_MPTCP_ADD_ADDR6_BASE 20
58 #define TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT 22
59 #define TCPOLEN_MPTCP_PORT_LEN 2
60 #define TCPOLEN_MPTCP_RM_ADDR_BASE 4
62 /* MPTCP MP_JOIN flags */
63 #define MPTCPOPT_BACKUP BIT(0)
64 #define MPTCPOPT_HMAC_LEN 20
65 #define MPTCPOPT_THMAC_LEN 8
67 /* MPTCP MP_CAPABLE flags */
68 #define MPTCP_VERSION_MASK (0x0F)
69 #define MPTCP_CAP_CHECKSUM_REQD BIT(7)
70 #define MPTCP_CAP_EXTENSIBILITY BIT(6)
71 #define MPTCP_CAP_HMAC_SHA256 BIT(0)
72 #define MPTCP_CAP_FLAG_MASK (0x3F)
75 #define MPTCP_DSS_DATA_FIN BIT(4)
76 #define MPTCP_DSS_DSN64 BIT(3)
77 #define MPTCP_DSS_HAS_MAP BIT(2)
78 #define MPTCP_DSS_ACK64 BIT(1)
79 #define MPTCP_DSS_HAS_ACK BIT(0)
80 #define MPTCP_DSS_FLAG_MASK (0x1F)
82 /* MPTCP ADD_ADDR flags */
83 #define MPTCP_ADDR_ECHO BIT(0)
84 #define MPTCP_ADDR_IPVERSION_4 4
85 #define MPTCP_ADDR_IPVERSION_6 6
87 /* MPTCP socket flags */
88 #define MPTCP_DATA_READY 0
89 #define MPTCP_SEND_SPACE 1
90 #define MPTCP_WORK_RTX 2
91 #define MPTCP_WORK_EOF 3
92 #define MPTCP_FALLBACK_DONE 4
93 #define MPTCP_WORK_CLOSE_SUBFLOW 5
95 struct mptcp_options_received {
126 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
127 struct in6_addr addr6;
134 static inline __be32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field)
136 return htonl((TCPOPT_MPTCP << 24) | (len << 16) | (subopt << 12) |
137 ((nib & 0xF) << 8) | field);
140 struct mptcp_addr_info {
148 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
149 struct in6_addr addr6;
154 enum mptcp_pm_status {
155 MPTCP_PM_ADD_ADDR_RECEIVED,
156 MPTCP_PM_RM_ADDR_RECEIVED,
157 MPTCP_PM_ESTABLISHED,
158 MPTCP_PM_SUBFLOW_ESTABLISHED,
161 struct mptcp_pm_data {
162 struct mptcp_addr_info local;
163 struct mptcp_addr_info remote;
164 struct list_head anno_list;
166 spinlock_t lock; /*protects the whole PM data */
168 bool add_addr_signal;
175 u8 add_addr_signaled;
176 u8 add_addr_accepted;
179 u8 add_addr_signal_max;
180 u8 add_addr_accept_max;
187 struct mptcp_data_frag {
188 struct list_head list;
197 /* MPTCP connection sock */
199 /* inet_connection_sock must be the first member */
200 struct inet_connection_sock sk;
205 u64 rcv_data_fin_seq;
206 struct sock *last_snd;
209 unsigned long timer_ival;
213 bool fully_established;
215 bool snd_data_fin_enable;
216 bool use_64bit_ack; /* Set when we received a 64-bit DSN */
217 spinlock_t join_list_lock;
218 struct work_struct work;
219 struct sk_buff *ooo_last_skb;
220 struct rb_root out_of_order_queue;
221 struct list_head conn_list;
222 struct list_head rtx_queue;
223 struct mptcp_data_frag *first_pending;
224 struct list_head join_list;
225 struct skb_ext *cached_ext; /* for the next sendmsg */
226 struct socket *subflow; /* outgoing connect/listener/!mp_capable */
228 struct mptcp_pm_data pm;
230 u32 space; /* bytes copied in last measurement window */
231 u32 copied; /* bytes copied in this measurement window */
232 u64 time; /* start time of measurement window */
233 u64 rtt_us; /* last maximum rtt of subflows */
237 #define mptcp_for_each_subflow(__msk, __subflow) \
238 list_for_each_entry(__subflow, &((__msk)->conn_list), node)
240 static inline struct mptcp_sock *mptcp_sk(const struct sock *sk)
242 return (struct mptcp_sock *)sk;
245 static inline struct mptcp_data_frag *mptcp_send_head(const struct sock *sk)
247 const struct mptcp_sock *msk = mptcp_sk(sk);
249 return READ_ONCE(msk->first_pending);
252 static inline struct mptcp_data_frag *mptcp_send_next(struct sock *sk)
254 struct mptcp_sock *msk = mptcp_sk(sk);
255 struct mptcp_data_frag *cur;
257 cur = msk->first_pending;
258 return list_is_last(&cur->list, &msk->rtx_queue) ? NULL :
259 list_next_entry(cur, list);
262 static inline struct mptcp_data_frag *mptcp_pending_tail(const struct sock *sk)
264 struct mptcp_sock *msk = mptcp_sk(sk);
266 if (!msk->first_pending)
269 if (WARN_ON_ONCE(list_empty(&msk->rtx_queue)))
272 return list_last_entry(&msk->rtx_queue, struct mptcp_data_frag, list);
275 static inline struct mptcp_data_frag *mptcp_rtx_tail(const struct sock *sk)
277 struct mptcp_sock *msk = mptcp_sk(sk);
279 if (list_empty(&msk->rtx_queue))
282 return list_last_entry(&msk->rtx_queue, struct mptcp_data_frag, list);
285 static inline struct mptcp_data_frag *mptcp_rtx_head(const struct sock *sk)
287 struct mptcp_sock *msk = mptcp_sk(sk);
289 return list_first_entry_or_null(&msk->rtx_queue, struct mptcp_data_frag, list);
292 struct mptcp_subflow_request_sock {
293 struct tcp_request_sock sk;
306 struct mptcp_sock *msk;
307 struct hlist_nulls_node token_node;
310 static inline struct mptcp_subflow_request_sock *
311 mptcp_subflow_rsk(const struct request_sock *rsk)
313 return (struct mptcp_subflow_request_sock *)rsk;
316 enum mptcp_data_avail {
317 MPTCP_SUBFLOW_NODATA,
318 MPTCP_SUBFLOW_DATA_AVAIL,
319 MPTCP_SUBFLOW_OOO_DATA
322 /* MPTCP subflow context */
323 struct mptcp_subflow_context {
324 struct list_head node;/* conn_list of subflows */
335 u32 request_mptcp : 1, /* send MP_CAPABLE */
336 request_join : 1, /* send MP_JOIN */
338 mp_capable : 1, /* remote is MPTCP capable */
339 mp_join : 1, /* remote is JOINing */
340 fully_established : 1, /* path validated */
341 pm_notified : 1, /* PM hook called for established status */
347 can_ack : 1; /* only after processing the remote a key */
348 enum mptcp_data_avail data_avail;
353 u8 hmac[MPTCPOPT_HMAC_LEN];
357 struct sock *tcp_sock; /* tcp sk backpointer */
358 struct sock *conn; /* parent mptcp_sock */
359 const struct inet_connection_sock_af_ops *icsk_af_ops;
360 void (*tcp_data_ready)(struct sock *sk);
361 void (*tcp_state_change)(struct sock *sk);
362 void (*tcp_write_space)(struct sock *sk);
367 static inline struct mptcp_subflow_context *
368 mptcp_subflow_ctx(const struct sock *sk)
370 struct inet_connection_sock *icsk = inet_csk(sk);
372 /* Use RCU on icsk_ulp_data only for sock diag code */
373 return (__force struct mptcp_subflow_context *)icsk->icsk_ulp_data;
376 static inline struct sock *
377 mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow)
379 return subflow->tcp_sock;
383 mptcp_subflow_get_map_offset(const struct mptcp_subflow_context *subflow)
385 return tcp_sk(mptcp_subflow_tcp_sock(subflow))->copied_seq -
386 subflow->ssn_offset -
387 subflow->map_subflow_seq;
391 mptcp_subflow_get_mapped_dsn(const struct mptcp_subflow_context *subflow)
393 return subflow->map_seq + mptcp_subflow_get_map_offset(subflow);
396 int mptcp_is_enabled(struct net *net);
397 unsigned int mptcp_get_add_addr_timeout(struct net *net);
398 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
399 struct mptcp_options_received *mp_opt);
400 bool mptcp_subflow_data_available(struct sock *sk);
401 void __init mptcp_subflow_init(void);
402 void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how);
403 void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
404 struct mptcp_subflow_context *subflow,
406 void mptcp_subflow_reset(struct sock *ssk);
408 /* called with sk socket lock held */
409 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
410 const struct mptcp_addr_info *remote);
411 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock);
413 static inline void mptcp_subflow_tcp_fallback(struct sock *sk,
414 struct mptcp_subflow_context *ctx)
416 sk->sk_data_ready = ctx->tcp_data_ready;
417 sk->sk_state_change = ctx->tcp_state_change;
418 sk->sk_write_space = ctx->tcp_write_space;
420 inet_csk(sk)->icsk_af_ops = ctx->icsk_af_ops;
423 void __init mptcp_proto_init(void);
424 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
425 int __init mptcp_proto_v6_init(void);
428 struct sock *mptcp_sk_clone(const struct sock *sk,
429 const struct mptcp_options_received *mp_opt,
430 struct request_sock *req);
431 void mptcp_get_options(const struct sk_buff *skb,
432 struct mptcp_options_received *mp_opt);
434 void mptcp_finish_connect(struct sock *sk);
435 static inline bool mptcp_is_fully_established(struct sock *sk)
437 return inet_sk_state_load(sk) == TCP_ESTABLISHED &&
438 READ_ONCE(mptcp_sk(sk)->fully_established);
440 void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk);
441 void mptcp_data_ready(struct sock *sk, struct sock *ssk);
442 bool mptcp_finish_join(struct sock *sk);
443 bool mptcp_schedule_work(struct sock *sk);
444 void mptcp_data_acked(struct sock *sk);
445 void mptcp_subflow_eof(struct sock *sk);
446 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit);
447 void mptcp_destroy_common(struct mptcp_sock *msk);
449 void __init mptcp_token_init(void);
450 static inline void mptcp_token_init_request(struct request_sock *req)
452 mptcp_subflow_rsk(req)->token_node.pprev = NULL;
455 int mptcp_token_new_request(struct request_sock *req);
456 void mptcp_token_destroy_request(struct request_sock *req);
457 int mptcp_token_new_connect(struct sock *sk);
458 void mptcp_token_accept(struct mptcp_subflow_request_sock *r,
459 struct mptcp_sock *msk);
460 bool mptcp_token_exists(u32 token);
461 struct mptcp_sock *mptcp_token_get_sock(u32 token);
462 struct mptcp_sock *mptcp_token_iter_next(const struct net *net, long *s_slot,
464 void mptcp_token_destroy(struct mptcp_sock *msk);
466 void mptcp_crypto_key_sha(u64 key, u32 *token, u64 *idsn);
468 void mptcp_crypto_hmac_sha(u64 key1, u64 key2, u8 *msg, int len, void *hmac);
470 void __init mptcp_pm_init(void);
471 void mptcp_pm_data_init(struct mptcp_sock *msk);
472 void mptcp_pm_new_connection(struct mptcp_sock *msk, int server_side);
473 void mptcp_pm_fully_established(struct mptcp_sock *msk);
474 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk);
475 void mptcp_pm_connection_closed(struct mptcp_sock *msk);
476 void mptcp_pm_subflow_established(struct mptcp_sock *msk,
477 struct mptcp_subflow_context *subflow);
478 void mptcp_pm_subflow_closed(struct mptcp_sock *msk, u8 id);
479 void mptcp_pm_add_addr_received(struct mptcp_sock *msk,
480 const struct mptcp_addr_info *addr);
481 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk, u8 rm_id);
482 void mptcp_pm_free_anno_list(struct mptcp_sock *msk);
483 struct mptcp_pm_add_entry *
484 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
485 struct mptcp_addr_info *addr);
487 int mptcp_pm_announce_addr(struct mptcp_sock *msk,
488 const struct mptcp_addr_info *addr,
490 int mptcp_pm_remove_addr(struct mptcp_sock *msk, u8 local_id);
491 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, u8 local_id);
493 static inline bool mptcp_pm_should_add_signal(struct mptcp_sock *msk)
495 return READ_ONCE(msk->pm.add_addr_signal);
498 static inline bool mptcp_pm_should_rm_signal(struct mptcp_sock *msk)
500 return READ_ONCE(msk->pm.rm_addr_signal);
503 static inline unsigned int mptcp_add_addr_len(int family, bool echo)
505 if (family == AF_INET)
506 return echo ? TCPOLEN_MPTCP_ADD_ADDR_BASE
507 : TCPOLEN_MPTCP_ADD_ADDR;
508 return echo ? TCPOLEN_MPTCP_ADD_ADDR6_BASE : TCPOLEN_MPTCP_ADD_ADDR6;
511 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
512 struct mptcp_addr_info *saddr, bool *echo);
513 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
515 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
517 void __init mptcp_pm_nl_init(void);
518 void mptcp_pm_nl_data_init(struct mptcp_sock *msk);
519 void mptcp_pm_nl_fully_established(struct mptcp_sock *msk);
520 void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk);
521 void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk);
522 void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk);
523 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, u8 rm_id);
524 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
526 static inline struct mptcp_ext *mptcp_get_ext(struct sk_buff *skb)
528 return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP);
531 static inline bool before64(__u64 seq1, __u64 seq2)
533 return (__s64)(seq1 - seq2) < 0;
536 #define after64(seq2, seq1) before64(seq1, seq2)
538 void mptcp_diag_subflow_init(struct tcp_ulp_ops *ops);
540 static inline bool __mptcp_check_fallback(const struct mptcp_sock *msk)
542 return test_bit(MPTCP_FALLBACK_DONE, &msk->flags);
545 static inline bool mptcp_check_fallback(const struct sock *sk)
547 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
548 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
550 return __mptcp_check_fallback(msk);
553 static inline void __mptcp_do_fallback(struct mptcp_sock *msk)
555 if (test_bit(MPTCP_FALLBACK_DONE, &msk->flags)) {
556 pr_debug("TCP fallback already done (msk=%p)", msk);
559 set_bit(MPTCP_FALLBACK_DONE, &msk->flags);
562 static inline void mptcp_do_fallback(struct sock *sk)
564 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
565 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
567 __mptcp_do_fallback(msk);
570 #define pr_fallback(a) pr_debug("%s:fallback to TCP (msk=%p)", __func__, a)
572 static inline bool subflow_simultaneous_connect(struct sock *sk)
574 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
575 struct sock *parent = subflow->conn;
577 return sk->sk_state == TCP_ESTABLISHED &&
578 !mptcp_sk(parent)->pm.server_side &&
579 !subflow->conn_finished;
582 #ifdef CONFIG_SYN_COOKIES
583 void subflow_init_req_cookie_join_save(const struct mptcp_subflow_request_sock *subflow_req,
584 struct sk_buff *skb);
585 bool mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subflow_req,
586 struct sk_buff *skb);
587 void __init mptcp_join_cookie_init(void);
590 subflow_init_req_cookie_join_save(const struct mptcp_subflow_request_sock *subflow_req,
591 struct sk_buff *skb) {}
593 mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subflow_req,
599 static inline void mptcp_join_cookie_init(void) {}
602 #endif /* __MPTCP_PROTOCOL_H */