Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
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
14#define MPTCP_SUPPORTED_VERSION 1
15
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)
26
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
36
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
61
62/* MPTCP MP_JOIN flags */
63#define MPTCPOPT_BACKUP BIT(0)
64#define MPTCPOPT_HMAC_LEN 20
65#define MPTCPOPT_THMAC_LEN 8
66
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)
73
74/* MPTCP DSS flags */
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)
81
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
86
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
94struct mptcp_options_received {
95 u64 sndr_key;
96 u64 rcvr_key;
97 u64 data_ack;
98 u64 data_seq;
99 u32 subflow_seq;
100 u16 data_len;
101 u16 mp_capable : 1,
102 mp_join : 1,
103 dss : 1,
104 add_addr : 1,
105 rm_addr : 1,
106 family : 4,
107 echo : 1,
108 backup : 1;
109 u32 token;
110 u32 nonce;
111 u64 thmac;
112 u8 hmac[20];
113 u8 join_id;
114 u8 use_map:1,
115 dsn64:1,
116 data_fin:1,
117 use_ack:1,
118 ack64:1,
119 mpc_map:1,
120 __unused:2;
121 u8 addr_id;
122 u8 rm_id;
123 union {
124 struct in_addr addr;
125#if IS_ENABLED(CONFIG_MPTCP_IPV6)
126 struct in6_addr addr6;
127#endif
128 };
129 u64 ahmac;
130 u16 port;
131};
132
133static inline __be32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field)
134{
135 return htonl((TCPOPT_MPTCP << 24) | (len << 16) | (subopt << 12) |
136 ((nib & 0xF) << 8) | field);
137}
138
139struct mptcp_addr_info {
140 sa_family_t family;
141 __be16 port;
142 u8 id;
143 union {
144 struct in_addr addr;
145#if IS_ENABLED(CONFIG_MPTCP_IPV6)
146 struct in6_addr addr6;
147#endif
148 };
149};
150
151enum mptcp_pm_status {
152 MPTCP_PM_ADD_ADDR_RECEIVED,
153 MPTCP_PM_ESTABLISHED,
154 MPTCP_PM_SUBFLOW_ESTABLISHED,
155};
156
157struct mptcp_pm_data {
158 struct mptcp_addr_info local;
159 struct mptcp_addr_info remote;
160
161 spinlock_t lock; /*protects the whole PM data */
162
163 bool addr_signal;
164 bool server_side;
165 bool work_pending;
166 bool accept_addr;
167 bool accept_subflow;
168 u8 add_addr_signaled;
169 u8 add_addr_accepted;
170 u8 local_addr_used;
171 u8 subflows;
172 u8 add_addr_signal_max;
173 u8 add_addr_accept_max;
174 u8 local_addr_max;
175 u8 subflows_max;
176 u8 status;
177};
178
179struct mptcp_data_frag {
180 struct list_head list;
181 u64 data_seq;
182 int data_len;
183 int offset;
184 int overhead;
185 struct page *page;
186};
187
188/* MPTCP connection sock */
189struct mptcp_sock {
190 /* inet_connection_sock must be the first member */
191 struct inet_connection_sock sk;
192 u64 local_key;
193 u64 remote_key;
194 u64 write_seq;
195 u64 ack_seq;
196 u64 rcv_data_fin_seq;
197 atomic64_t snd_una;
198 unsigned long timer_ival;
199 u32 token;
200 unsigned long flags;
201 bool can_ack;
202 bool fully_established;
203 bool rcv_data_fin;
204 bool snd_data_fin_enable;
205 spinlock_t join_list_lock;
206 struct work_struct work;
207 struct list_head conn_list;
208 struct list_head rtx_queue;
209 struct list_head join_list;
210 struct skb_ext *cached_ext; /* for the next sendmsg */
211 struct socket *subflow; /* outgoing connect/listener/!mp_capable */
212 struct sock *first;
213 struct mptcp_pm_data pm;
214 struct {
215 u32 space; /* bytes copied in last measurement window */
216 u32 copied; /* bytes copied in this measurement window */
217 u64 time; /* start time of measurement window */
218 u64 rtt_us; /* last maximum rtt of subflows */
219 } rcvq_space;
220};
221
222#define mptcp_for_each_subflow(__msk, __subflow) \
223 list_for_each_entry(__subflow, &((__msk)->conn_list), node)
224
225static inline struct mptcp_sock *mptcp_sk(const struct sock *sk)
226{
227 return (struct mptcp_sock *)sk;
228}
229
230static inline struct mptcp_data_frag *mptcp_rtx_tail(const struct sock *sk)
231{
232 struct mptcp_sock *msk = mptcp_sk(sk);
233
234 if (list_empty(&msk->rtx_queue))
235 return NULL;
236
237 return list_last_entry(&msk->rtx_queue, struct mptcp_data_frag, list);
238}
239
240static inline struct mptcp_data_frag *mptcp_rtx_head(const struct sock *sk)
241{
242 struct mptcp_sock *msk = mptcp_sk(sk);
243
244 return list_first_entry_or_null(&msk->rtx_queue, struct mptcp_data_frag, list);
245}
246
247struct mptcp_subflow_request_sock {
248 struct tcp_request_sock sk;
249 u16 mp_capable : 1,
250 mp_join : 1,
251 backup : 1;
252 u8 local_id;
253 u8 remote_id;
254 u64 local_key;
255 u64 idsn;
256 u32 token;
257 u32 ssn_offset;
258 u64 thmac;
259 u32 local_nonce;
260 u32 remote_nonce;
261 struct mptcp_sock *msk;
262 struct hlist_nulls_node token_node;
263};
264
265static inline struct mptcp_subflow_request_sock *
266mptcp_subflow_rsk(const struct request_sock *rsk)
267{
268 return (struct mptcp_subflow_request_sock *)rsk;
269}
270
271/* MPTCP subflow context */
272struct mptcp_subflow_context {
273 struct list_head node;/* conn_list of subflows */
274 u64 local_key;
275 u64 remote_key;
276 u64 idsn;
277 u64 map_seq;
278 u32 snd_isn;
279 u32 token;
280 u32 rel_write_seq;
281 u32 map_subflow_seq;
282 u32 ssn_offset;
283 u32 map_data_len;
284 u32 request_mptcp : 1, /* send MP_CAPABLE */
285 request_join : 1, /* send MP_JOIN */
286 request_bkup : 1,
287 mp_capable : 1, /* remote is MPTCP capable */
288 mp_join : 1, /* remote is JOINing */
289 fully_established : 1, /* path validated */
290 pm_notified : 1, /* PM hook called for established status */
291 conn_finished : 1,
292 map_valid : 1,
293 mpc_map : 1,
294 backup : 1,
295 data_avail : 1,
296 rx_eof : 1,
297 use_64bit_ack : 1, /* Set when we received a 64-bit DSN */
298 can_ack : 1; /* only after processing the remote a key */
299 u32 remote_nonce;
300 u64 thmac;
301 u32 local_nonce;
302 u32 remote_token;
303 u8 hmac[MPTCPOPT_HMAC_LEN];
304 u8 local_id;
305 u8 remote_id;
306
307 struct sock *tcp_sock; /* tcp sk backpointer */
308 struct sock *conn; /* parent mptcp_sock */
309 const struct inet_connection_sock_af_ops *icsk_af_ops;
310 void (*tcp_data_ready)(struct sock *sk);
311 void (*tcp_state_change)(struct sock *sk);
312 void (*tcp_write_space)(struct sock *sk);
313
314 struct rcu_head rcu;
315};
316
317static inline struct mptcp_subflow_context *
318mptcp_subflow_ctx(const struct sock *sk)
319{
320 struct inet_connection_sock *icsk = inet_csk(sk);
321
322 /* Use RCU on icsk_ulp_data only for sock diag code */
323 return (__force struct mptcp_subflow_context *)icsk->icsk_ulp_data;
324}
325
326static inline struct sock *
327mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow)
328{
329 return subflow->tcp_sock;
330}
331
332static inline u64
333mptcp_subflow_get_map_offset(const struct mptcp_subflow_context *subflow)
334{
335 return tcp_sk(mptcp_subflow_tcp_sock(subflow))->copied_seq -
336 subflow->ssn_offset -
337 subflow->map_subflow_seq;
338}
339
340static inline u64
341mptcp_subflow_get_mapped_dsn(const struct mptcp_subflow_context *subflow)
342{
343 return subflow->map_seq + mptcp_subflow_get_map_offset(subflow);
344}
345
346int mptcp_is_enabled(struct net *net);
347void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
348 struct mptcp_options_received *mp_opt);
349bool mptcp_subflow_data_available(struct sock *sk);
350void __init mptcp_subflow_init(void);
351
352/* called with sk socket lock held */
353int __mptcp_subflow_connect(struct sock *sk, int ifindex,
354 const struct mptcp_addr_info *loc,
355 const struct mptcp_addr_info *remote);
356int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock);
357
358static inline void mptcp_subflow_tcp_fallback(struct sock *sk,
359 struct mptcp_subflow_context *ctx)
360{
361 sk->sk_data_ready = ctx->tcp_data_ready;
362 sk->sk_state_change = ctx->tcp_state_change;
363 sk->sk_write_space = ctx->tcp_write_space;
364
365 inet_csk(sk)->icsk_af_ops = ctx->icsk_af_ops;
366}
367
368void __init mptcp_proto_init(void);
369#if IS_ENABLED(CONFIG_MPTCP_IPV6)
370int __init mptcp_proto_v6_init(void);
371#endif
372
373struct sock *mptcp_sk_clone(const struct sock *sk,
374 const struct mptcp_options_received *mp_opt,
375 struct request_sock *req);
376void mptcp_get_options(const struct sk_buff *skb,
377 struct mptcp_options_received *mp_opt);
378
379void mptcp_finish_connect(struct sock *sk);
380static inline bool mptcp_is_fully_established(struct sock *sk)
381{
382 return inet_sk_state_load(sk) == TCP_ESTABLISHED &&
383 READ_ONCE(mptcp_sk(sk)->fully_established);
384}
385void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk);
386void mptcp_data_ready(struct sock *sk, struct sock *ssk);
387bool mptcp_finish_join(struct sock *sk);
388void mptcp_data_acked(struct sock *sk);
389void mptcp_subflow_eof(struct sock *sk);
390bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit);
391
392void __init mptcp_token_init(void);
393static inline void mptcp_token_init_request(struct request_sock *req)
394{
395 mptcp_subflow_rsk(req)->token_node.pprev = NULL;
396}
397
398int mptcp_token_new_request(struct request_sock *req);
399void mptcp_token_destroy_request(struct request_sock *req);
400int mptcp_token_new_connect(struct sock *sk);
401void mptcp_token_accept(struct mptcp_subflow_request_sock *r,
402 struct mptcp_sock *msk);
403bool mptcp_token_exists(u32 token);
404struct mptcp_sock *mptcp_token_get_sock(u32 token);
405struct mptcp_sock *mptcp_token_iter_next(const struct net *net, long *s_slot,
406 long *s_num);
407void mptcp_token_destroy(struct mptcp_sock *msk);
408
409void mptcp_crypto_key_sha(u64 key, u32 *token, u64 *idsn);
410
411void mptcp_crypto_hmac_sha(u64 key1, u64 key2, u8 *msg, int len, void *hmac);
412
413void __init mptcp_pm_init(void);
414void mptcp_pm_data_init(struct mptcp_sock *msk);
415void mptcp_pm_new_connection(struct mptcp_sock *msk, int server_side);
416void mptcp_pm_fully_established(struct mptcp_sock *msk);
417bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk);
418void mptcp_pm_connection_closed(struct mptcp_sock *msk);
419void mptcp_pm_subflow_established(struct mptcp_sock *msk,
420 struct mptcp_subflow_context *subflow);
421void mptcp_pm_subflow_closed(struct mptcp_sock *msk, u8 id);
422void mptcp_pm_add_addr_received(struct mptcp_sock *msk,
423 const struct mptcp_addr_info *addr);
424
425int mptcp_pm_announce_addr(struct mptcp_sock *msk,
426 const struct mptcp_addr_info *addr);
427int mptcp_pm_remove_addr(struct mptcp_sock *msk, u8 local_id);
428int mptcp_pm_remove_subflow(struct mptcp_sock *msk, u8 remote_id);
429
430static inline bool mptcp_pm_should_signal(struct mptcp_sock *msk)
431{
432 return READ_ONCE(msk->pm.addr_signal);
433}
434
435static inline unsigned int mptcp_add_addr_len(int family)
436{
437 if (family == AF_INET)
438 return TCPOLEN_MPTCP_ADD_ADDR;
439 return TCPOLEN_MPTCP_ADD_ADDR6;
440}
441
442bool mptcp_pm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
443 struct mptcp_addr_info *saddr);
444int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
445
446void __init mptcp_pm_nl_init(void);
447void mptcp_pm_nl_data_init(struct mptcp_sock *msk);
448void mptcp_pm_nl_fully_established(struct mptcp_sock *msk);
449void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk);
450void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk);
451int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
452
453static inline struct mptcp_ext *mptcp_get_ext(struct sk_buff *skb)
454{
455 return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP);
456}
457
458static inline bool before64(__u64 seq1, __u64 seq2)
459{
460 return (__s64)(seq1 - seq2) < 0;
461}
462
463#define after64(seq2, seq1) before64(seq1, seq2)
464
465void mptcp_diag_subflow_init(struct tcp_ulp_ops *ops);
466
467static inline bool __mptcp_check_fallback(struct mptcp_sock *msk)
468{
469 return test_bit(MPTCP_FALLBACK_DONE, &msk->flags);
470}
471
472static inline bool mptcp_check_fallback(struct sock *sk)
473{
474 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
475 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
476
477 return __mptcp_check_fallback(msk);
478}
479
480static inline void __mptcp_do_fallback(struct mptcp_sock *msk)
481{
482 if (test_bit(MPTCP_FALLBACK_DONE, &msk->flags)) {
483 pr_debug("TCP fallback already done (msk=%p)", msk);
484 return;
485 }
486 set_bit(MPTCP_FALLBACK_DONE, &msk->flags);
487}
488
489static inline void mptcp_do_fallback(struct sock *sk)
490{
491 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
492 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
493
494 __mptcp_do_fallback(msk);
495}
496
497#define pr_fallback(a) pr_debug("%s:fallback to TCP (msk=%p)", __func__, a)
498
499static inline bool subflow_simultaneous_connect(struct sock *sk)
500{
501 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
502 struct sock *parent = subflow->conn;
503
504 return sk->sk_state == TCP_ESTABLISHED &&
505 !mptcp_sk(parent)->pm.server_side &&
506 !subflow->conn_finished;
507}
508
509#ifdef CONFIG_SYN_COOKIES
510void subflow_init_req_cookie_join_save(const struct mptcp_subflow_request_sock *subflow_req,
511 struct sk_buff *skb);
512bool mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subflow_req,
513 struct sk_buff *skb);
514void __init mptcp_join_cookie_init(void);
515#else
516static inline void
517subflow_init_req_cookie_join_save(const struct mptcp_subflow_request_sock *subflow_req,
518 struct sk_buff *skb) {}
519static inline bool
520mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subflow_req,
521 struct sk_buff *skb)
522{
523 return false;
524}
525
526static inline void mptcp_join_cookie_init(void) {}
527#endif
528
529#endif /* __MPTCP_PROTOCOL_H */