Loading...
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#include <net/rstreason.h>
16
17#define MPTCP_SUPPORTED_VERSION 1
18
19/* MPTCP option bits */
20#define OPTION_MPTCP_MPC_SYN BIT(0)
21#define OPTION_MPTCP_MPC_SYNACK BIT(1)
22#define OPTION_MPTCP_MPC_ACK BIT(2)
23#define OPTION_MPTCP_MPJ_SYN BIT(3)
24#define OPTION_MPTCP_MPJ_SYNACK BIT(4)
25#define OPTION_MPTCP_MPJ_ACK BIT(5)
26#define OPTION_MPTCP_ADD_ADDR BIT(6)
27#define OPTION_MPTCP_RM_ADDR BIT(7)
28#define OPTION_MPTCP_FASTCLOSE BIT(8)
29#define OPTION_MPTCP_PRIO BIT(9)
30#define OPTION_MPTCP_RST BIT(10)
31#define OPTION_MPTCP_DSS BIT(11)
32#define OPTION_MPTCP_FAIL BIT(12)
33
34#define OPTION_MPTCP_CSUMREQD BIT(13)
35
36#define OPTIONS_MPTCP_MPC (OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK | \
37 OPTION_MPTCP_MPC_ACK)
38#define OPTIONS_MPTCP_MPJ (OPTION_MPTCP_MPJ_SYN | OPTION_MPTCP_MPJ_SYNACK | \
39 OPTION_MPTCP_MPJ_ACK)
40
41/* MPTCP option subtypes */
42#define MPTCPOPT_MP_CAPABLE 0
43#define MPTCPOPT_MP_JOIN 1
44#define MPTCPOPT_DSS 2
45#define MPTCPOPT_ADD_ADDR 3
46#define MPTCPOPT_RM_ADDR 4
47#define MPTCPOPT_MP_PRIO 5
48#define MPTCPOPT_MP_FAIL 6
49#define MPTCPOPT_MP_FASTCLOSE 7
50#define MPTCPOPT_RST 8
51
52/* MPTCP suboption lengths */
53#define TCPOLEN_MPTCP_MPC_SYN 4
54#define TCPOLEN_MPTCP_MPC_SYNACK 12
55#define TCPOLEN_MPTCP_MPC_ACK 20
56#define TCPOLEN_MPTCP_MPC_ACK_DATA 22
57#define TCPOLEN_MPTCP_MPJ_SYN 12
58#define TCPOLEN_MPTCP_MPJ_SYNACK 16
59#define TCPOLEN_MPTCP_MPJ_ACK 24
60#define TCPOLEN_MPTCP_DSS_BASE 4
61#define TCPOLEN_MPTCP_DSS_ACK32 4
62#define TCPOLEN_MPTCP_DSS_ACK64 8
63#define TCPOLEN_MPTCP_DSS_MAP32 10
64#define TCPOLEN_MPTCP_DSS_MAP64 14
65#define TCPOLEN_MPTCP_DSS_CHECKSUM 2
66#define TCPOLEN_MPTCP_ADD_ADDR 16
67#define TCPOLEN_MPTCP_ADD_ADDR_PORT 18
68#define TCPOLEN_MPTCP_ADD_ADDR_BASE 8
69#define TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT 10
70#define TCPOLEN_MPTCP_ADD_ADDR6 28
71#define TCPOLEN_MPTCP_ADD_ADDR6_PORT 30
72#define TCPOLEN_MPTCP_ADD_ADDR6_BASE 20
73#define TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT 22
74#define TCPOLEN_MPTCP_PORT_LEN 2
75#define TCPOLEN_MPTCP_PORT_ALIGN 2
76#define TCPOLEN_MPTCP_RM_ADDR_BASE 3
77#define TCPOLEN_MPTCP_PRIO 3
78#define TCPOLEN_MPTCP_PRIO_ALIGN 4
79#define TCPOLEN_MPTCP_FASTCLOSE 12
80#define TCPOLEN_MPTCP_RST 4
81#define TCPOLEN_MPTCP_FAIL 12
82
83#define TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM (TCPOLEN_MPTCP_DSS_CHECKSUM + TCPOLEN_MPTCP_MPC_ACK_DATA)
84
85/* MPTCP MP_JOIN flags */
86#define MPTCPOPT_BACKUP BIT(0)
87#define MPTCPOPT_THMAC_LEN 8
88
89/* MPTCP MP_CAPABLE flags */
90#define MPTCP_VERSION_MASK (0x0F)
91#define MPTCP_CAP_CHECKSUM_REQD BIT(7)
92#define MPTCP_CAP_EXTENSIBILITY BIT(6)
93#define MPTCP_CAP_DENY_JOIN_ID0 BIT(5)
94#define MPTCP_CAP_HMAC_SHA256 BIT(0)
95#define MPTCP_CAP_FLAG_MASK (0x1F)
96
97/* MPTCP DSS flags */
98#define MPTCP_DSS_DATA_FIN BIT(4)
99#define MPTCP_DSS_DSN64 BIT(3)
100#define MPTCP_DSS_HAS_MAP BIT(2)
101#define MPTCP_DSS_ACK64 BIT(1)
102#define MPTCP_DSS_HAS_ACK BIT(0)
103#define MPTCP_DSS_FLAG_MASK (0x1F)
104
105/* MPTCP ADD_ADDR flags */
106#define MPTCP_ADDR_ECHO BIT(0)
107
108/* MPTCP MP_PRIO flags */
109#define MPTCP_PRIO_BKUP BIT(0)
110
111/* MPTCP TCPRST flags */
112#define MPTCP_RST_TRANSIENT BIT(0)
113
114/* MPTCP socket atomic flags */
115#define MPTCP_WORK_RTX 1
116#define MPTCP_FALLBACK_DONE 2
117#define MPTCP_WORK_CLOSE_SUBFLOW 3
118
119/* MPTCP socket release cb flags */
120#define MPTCP_PUSH_PENDING 1
121#define MPTCP_CLEAN_UNA 2
122#define MPTCP_ERROR_REPORT 3
123#define MPTCP_RETRANSMIT 4
124#define MPTCP_FLUSH_JOIN_LIST 5
125#define MPTCP_SYNC_STATE 6
126#define MPTCP_SYNC_SNDBUF 7
127
128struct mptcp_skb_cb {
129 u64 map_seq;
130 u64 end_seq;
131 u32 offset;
132 u8 has_rxtstamp:1;
133};
134
135#define MPTCP_SKB_CB(__skb) ((struct mptcp_skb_cb *)&((__skb)->cb[0]))
136
137static inline bool before64(__u64 seq1, __u64 seq2)
138{
139 return (__s64)(seq1 - seq2) < 0;
140}
141
142#define after64(seq2, seq1) before64(seq1, seq2)
143
144struct mptcp_options_received {
145 u64 sndr_key;
146 u64 rcvr_key;
147 u64 data_ack;
148 u64 data_seq;
149 u32 subflow_seq;
150 u16 data_len;
151 __sum16 csum;
152 struct_group(status,
153 u16 suboptions;
154 u16 use_map:1,
155 dsn64:1,
156 data_fin:1,
157 use_ack:1,
158 ack64:1,
159 mpc_map:1,
160 reset_reason:4,
161 reset_transient:1,
162 echo:1,
163 backup:1,
164 deny_join_id0:1,
165 __unused:2;
166 );
167 u8 join_id;
168 u32 token;
169 u32 nonce;
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
178static 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
184enum 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
196enum 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
207enum 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
216struct 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
241struct mptcp_pm_local {
242 struct mptcp_addr_info addr;
243 u8 flags;
244 int ifindex;
245};
246
247struct mptcp_pm_addr_entry {
248 struct list_head list;
249 struct mptcp_addr_info addr;
250 u8 flags;
251 int ifindex;
252 struct socket *lsk;
253};
254
255struct mptcp_data_frag {
256 struct list_head list;
257 u64 data_seq;
258 u16 data_len;
259 u16 offset;
260 u16 overhead;
261 u16 already_sent;
262 struct page *page;
263};
264
265/* MPTCP connection sock */
266struct mptcp_sock {
267 /* inet_connection_sock must be the first member */
268 struct inet_connection_sock sk;
269 u64 local_key; /* protected by the first subflow socket lock
270 * lockless access read
271 */
272 u64 remote_key; /* same as above */
273 u64 write_seq;
274 u64 bytes_sent;
275 u64 snd_nxt;
276 u64 bytes_received;
277 u64 ack_seq;
278 atomic64_t rcv_wnd_sent;
279 u64 rcv_data_fin_seq;
280 u64 bytes_retrans;
281 u64 bytes_consumed;
282 int rmem_fwd_alloc;
283 int snd_burst;
284 int old_wspace;
285 u64 recovery_snd_nxt; /* in recovery mode accept up to this seq;
286 * recovery related fields are under data_lock
287 * protection
288 */
289 u64 bytes_acked;
290 u64 snd_una;
291 u64 wnd_end;
292 u32 last_data_sent;
293 u32 last_data_recv;
294 u32 last_ack_recv;
295 unsigned long timer_ival;
296 u32 token;
297 int rmem_released;
298 unsigned long flags;
299 unsigned long cb_flags;
300 bool recovery; /* closing subflow write queue reinjected */
301 bool can_ack;
302 bool fully_established;
303 bool rcv_data_fin;
304 bool snd_data_fin_enable;
305 bool rcv_fastclose;
306 bool use_64bit_ack; /* Set when we received a 64-bit DSN */
307 bool csum_enabled;
308 bool allow_infinite_fallback;
309 u8 pending_state; /* A subflow asked to set this sk_state,
310 * protected by the msk data lock
311 */
312 u8 mpc_endpoint_id;
313 u8 recvmsg_inq:1,
314 cork:1,
315 nodelay:1,
316 fastopening:1,
317 in_accept_queue:1,
318 free_first:1,
319 rcvspace_init:1;
320 u32 notsent_lowat;
321 int keepalive_cnt;
322 int keepalive_idle;
323 int keepalive_intvl;
324 struct work_struct work;
325 struct sk_buff *ooo_last_skb;
326 struct rb_root out_of_order_queue;
327 struct sk_buff_head receive_queue;
328 struct list_head conn_list;
329 struct list_head rtx_queue;
330 struct mptcp_data_frag *first_pending;
331 struct list_head join_list;
332 struct sock *first; /* The mptcp ops can safely dereference, using suitable
333 * ONCE annotation, the subflow outside the socket
334 * lock as such sock is freed after close().
335 */
336 struct mptcp_pm_data pm;
337 struct mptcp_sched_ops *sched;
338 struct {
339 u32 space; /* bytes copied in last measurement window */
340 u32 copied; /* bytes copied in this measurement window */
341 u64 time; /* start time of measurement window */
342 u64 rtt_us; /* last maximum rtt of subflows */
343 } rcvq_space;
344 u8 scaling_ratio;
345
346 u32 subflow_id;
347 u32 setsockopt_seq;
348 char ca_name[TCP_CA_NAME_MAX];
349};
350
351#define mptcp_data_lock(sk) spin_lock_bh(&(sk)->sk_lock.slock)
352#define mptcp_data_unlock(sk) spin_unlock_bh(&(sk)->sk_lock.slock)
353
354#define mptcp_for_each_subflow(__msk, __subflow) \
355 list_for_each_entry(__subflow, &((__msk)->conn_list), node)
356#define mptcp_for_each_subflow_safe(__msk, __subflow, __tmp) \
357 list_for_each_entry_safe(__subflow, __tmp, &((__msk)->conn_list), node)
358
359extern struct genl_family mptcp_genl_family;
360
361static inline void msk_owned_by_me(const struct mptcp_sock *msk)
362{
363 sock_owned_by_me((const struct sock *)msk);
364}
365
366#ifdef CONFIG_DEBUG_NET
367/* MPTCP-specific: we might (indirectly) call this helper with the wrong sk */
368#undef tcp_sk
369#define tcp_sk(ptr) ({ \
370 typeof(ptr) _ptr = (ptr); \
371 WARN_ON(_ptr->sk_protocol != IPPROTO_TCP); \
372 container_of_const(_ptr, struct tcp_sock, inet_conn.icsk_inet.sk); \
373})
374#define mptcp_sk(ptr) ({ \
375 typeof(ptr) _ptr = (ptr); \
376 WARN_ON(_ptr->sk_protocol != IPPROTO_MPTCP); \
377 container_of_const(_ptr, struct mptcp_sock, sk.icsk_inet.sk); \
378})
379
380#else /* !CONFIG_DEBUG_NET */
381#define mptcp_sk(ptr) container_of_const(ptr, struct mptcp_sock, sk.icsk_inet.sk)
382#endif
383
384/* the msk socket don't use the backlog, also account for the bulk
385 * free memory
386 */
387static inline int __mptcp_rmem(const struct sock *sk)
388{
389 return atomic_read(&sk->sk_rmem_alloc) - READ_ONCE(mptcp_sk(sk)->rmem_released);
390}
391
392static inline int mptcp_win_from_space(const struct sock *sk, int space)
393{
394 return __tcp_win_from_space(mptcp_sk(sk)->scaling_ratio, space);
395}
396
397static inline int mptcp_space_from_win(const struct sock *sk, int win)
398{
399 return __tcp_space_from_win(mptcp_sk(sk)->scaling_ratio, win);
400}
401
402static inline int __mptcp_space(const struct sock *sk)
403{
404 return mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - __mptcp_rmem(sk));
405}
406
407static inline struct mptcp_data_frag *mptcp_send_head(const struct sock *sk)
408{
409 const struct mptcp_sock *msk = mptcp_sk(sk);
410
411 return READ_ONCE(msk->first_pending);
412}
413
414static inline struct mptcp_data_frag *mptcp_send_next(struct sock *sk)
415{
416 struct mptcp_sock *msk = mptcp_sk(sk);
417 struct mptcp_data_frag *cur;
418
419 cur = msk->first_pending;
420 return list_is_last(&cur->list, &msk->rtx_queue) ? NULL :
421 list_next_entry(cur, list);
422}
423
424static inline struct mptcp_data_frag *mptcp_pending_tail(const struct sock *sk)
425{
426 const struct mptcp_sock *msk = mptcp_sk(sk);
427
428 if (!msk->first_pending)
429 return NULL;
430
431 if (WARN_ON_ONCE(list_empty(&msk->rtx_queue)))
432 return NULL;
433
434 return list_last_entry(&msk->rtx_queue, struct mptcp_data_frag, list);
435}
436
437static inline struct mptcp_data_frag *mptcp_rtx_head(struct sock *sk)
438{
439 struct mptcp_sock *msk = mptcp_sk(sk);
440
441 if (msk->snd_una == msk->snd_nxt)
442 return NULL;
443
444 return list_first_entry_or_null(&msk->rtx_queue, struct mptcp_data_frag, list);
445}
446
447struct csum_pseudo_header {
448 __be64 data_seq;
449 __be32 subflow_seq;
450 __be16 data_len;
451 __sum16 csum;
452};
453
454struct mptcp_subflow_request_sock {
455 struct tcp_request_sock sk;
456 u16 mp_capable : 1,
457 mp_join : 1,
458 backup : 1,
459 request_bkup : 1,
460 csum_reqd : 1,
461 allow_join_id0 : 1;
462 u8 local_id;
463 u8 remote_id;
464 u64 local_key;
465 u64 idsn;
466 u32 token;
467 u32 ssn_offset;
468 u64 thmac;
469 u32 local_nonce;
470 u32 remote_nonce;
471 struct mptcp_sock *msk;
472 struct hlist_nulls_node token_node;
473};
474
475static inline struct mptcp_subflow_request_sock *
476mptcp_subflow_rsk(const struct request_sock *rsk)
477{
478 return (struct mptcp_subflow_request_sock *)rsk;
479}
480
481struct mptcp_delegated_action {
482 struct napi_struct napi;
483 struct list_head head;
484};
485
486DECLARE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
487
488#define MPTCP_DELEGATE_SCHEDULED 0
489#define MPTCP_DELEGATE_SEND 1
490#define MPTCP_DELEGATE_ACK 2
491#define MPTCP_DELEGATE_SNDBUF 3
492
493#define MPTCP_DELEGATE_ACTIONS_MASK (~BIT(MPTCP_DELEGATE_SCHEDULED))
494/* MPTCP subflow context */
495struct mptcp_subflow_context {
496 struct list_head node;/* conn_list of subflows */
497
498 struct_group(reset,
499
500 unsigned long avg_pacing_rate; /* protected by msk socket lock */
501 u64 local_key;
502 u64 remote_key;
503 u64 idsn;
504 u64 map_seq;
505 u32 snd_isn;
506 u32 token;
507 u32 rel_write_seq;
508 u32 map_subflow_seq;
509 u32 ssn_offset;
510 u32 map_data_len;
511 __wsum map_data_csum;
512 u32 map_csum_len;
513 u32 request_mptcp : 1, /* send MP_CAPABLE */
514 request_join : 1, /* send MP_JOIN */
515 request_bkup : 1,
516 mp_capable : 1, /* remote is MPTCP capable */
517 mp_join : 1, /* remote is JOINing */
518 pm_notified : 1, /* PM hook called for established status */
519 conn_finished : 1,
520 map_valid : 1,
521 map_csum_reqd : 1,
522 map_data_fin : 1,
523 mpc_map : 1,
524 backup : 1,
525 send_mp_prio : 1,
526 send_mp_fail : 1,
527 send_fastclose : 1,
528 send_infinite_map : 1,
529 remote_key_valid : 1, /* received the peer key from */
530 disposable : 1, /* ctx can be free at ulp release time */
531 stale : 1, /* unable to snd/rcv data, do not use for xmit */
532 valid_csum_seen : 1, /* at least one csum validated */
533 is_mptfo : 1, /* subflow is doing TFO */
534 close_event_done : 1, /* has done the post-closed part */
535 mpc_drop : 1, /* the MPC option has been dropped in a rtx */
536 __unused : 9;
537 bool data_avail;
538 bool scheduled;
539 bool pm_listener; /* a listener managed by the kernel PM? */
540 bool fully_established; /* path validated */
541 u32 remote_nonce;
542 u64 thmac;
543 u32 local_nonce;
544 u32 remote_token;
545 union {
546 u8 hmac[MPTCPOPT_HMAC_LEN]; /* MPJ subflow only */
547 u64 iasn; /* initial ack sequence number, MPC subflows only */
548 };
549 s16 local_id; /* if negative not initialized yet */
550 u8 remote_id;
551 u8 reset_seen:1;
552 u8 reset_transient:1;
553 u8 reset_reason:4;
554 u8 stale_count;
555
556 u32 subflow_id;
557
558 long delegated_status;
559 unsigned long fail_tout;
560
561 );
562
563 struct list_head delegated_node; /* link into delegated_action, protected by local BH */
564
565 u32 setsockopt_seq;
566 u32 stale_rcv_tstamp;
567 int cached_sndbuf; /* sndbuf size when last synced with the msk sndbuf,
568 * protected by the msk socket lock
569 */
570
571 struct sock *tcp_sock; /* tcp sk backpointer */
572 struct sock *conn; /* parent mptcp_sock */
573 const struct inet_connection_sock_af_ops *icsk_af_ops;
574 void (*tcp_state_change)(struct sock *sk);
575 void (*tcp_error_report)(struct sock *sk);
576
577 struct rcu_head rcu;
578};
579
580static inline struct mptcp_subflow_context *
581mptcp_subflow_ctx(const struct sock *sk)
582{
583 const struct inet_connection_sock *icsk = inet_csk(sk);
584
585 /* Use RCU on icsk_ulp_data only for sock diag code */
586 return (__force struct mptcp_subflow_context *)icsk->icsk_ulp_data;
587}
588
589static inline struct sock *
590mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow)
591{
592 return subflow->tcp_sock;
593}
594
595static inline void
596mptcp_subflow_ctx_reset(struct mptcp_subflow_context *subflow)
597{
598 memset(&subflow->reset, 0, sizeof(subflow->reset));
599 subflow->request_mptcp = 1;
600 WRITE_ONCE(subflow->local_id, -1);
601}
602
603/* Convert reset reasons in MPTCP to enum sk_rst_reason type */
604static inline enum sk_rst_reason
605sk_rst_convert_mptcp_reason(u32 reason)
606{
607 switch (reason) {
608 case MPTCP_RST_EUNSPEC:
609 return SK_RST_REASON_MPTCP_RST_EUNSPEC;
610 case MPTCP_RST_EMPTCP:
611 return SK_RST_REASON_MPTCP_RST_EMPTCP;
612 case MPTCP_RST_ERESOURCE:
613 return SK_RST_REASON_MPTCP_RST_ERESOURCE;
614 case MPTCP_RST_EPROHIBIT:
615 return SK_RST_REASON_MPTCP_RST_EPROHIBIT;
616 case MPTCP_RST_EWQ2BIG:
617 return SK_RST_REASON_MPTCP_RST_EWQ2BIG;
618 case MPTCP_RST_EBADPERF:
619 return SK_RST_REASON_MPTCP_RST_EBADPERF;
620 case MPTCP_RST_EMIDDLEBOX:
621 return SK_RST_REASON_MPTCP_RST_EMIDDLEBOX;
622 default:
623 /* It should not happen, or else errors may occur
624 * in MPTCP layer
625 */
626 return SK_RST_REASON_ERROR;
627 }
628}
629
630static inline void
631mptcp_send_active_reset_reason(struct sock *sk)
632{
633 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
634 enum sk_rst_reason reason;
635
636 reason = sk_rst_convert_mptcp_reason(subflow->reset_reason);
637 tcp_send_active_reset(sk, GFP_ATOMIC, reason);
638}
639
640static inline u64
641mptcp_subflow_get_map_offset(const struct mptcp_subflow_context *subflow)
642{
643 return tcp_sk(mptcp_subflow_tcp_sock(subflow))->copied_seq -
644 subflow->ssn_offset -
645 subflow->map_subflow_seq;
646}
647
648static inline u64
649mptcp_subflow_get_mapped_dsn(const struct mptcp_subflow_context *subflow)
650{
651 return subflow->map_seq + mptcp_subflow_get_map_offset(subflow);
652}
653
654void mptcp_subflow_process_delegated(struct sock *ssk, long actions);
655
656static inline void mptcp_subflow_delegate(struct mptcp_subflow_context *subflow, int action)
657{
658 long old, set_bits = BIT(MPTCP_DELEGATE_SCHEDULED) | BIT(action);
659 struct mptcp_delegated_action *delegated;
660 bool schedule;
661
662 /* the caller held the subflow bh socket lock */
663 lockdep_assert_in_softirq();
664
665 /* The implied barrier pairs with tcp_release_cb_override()
666 * mptcp_napi_poll(), and ensures the below list check sees list
667 * updates done prior to delegated status bits changes
668 */
669 old = set_mask_bits(&subflow->delegated_status, 0, set_bits);
670 if (!(old & BIT(MPTCP_DELEGATE_SCHEDULED))) {
671 if (WARN_ON_ONCE(!list_empty(&subflow->delegated_node)))
672 return;
673
674 delegated = this_cpu_ptr(&mptcp_delegated_actions);
675 schedule = list_empty(&delegated->head);
676 list_add_tail(&subflow->delegated_node, &delegated->head);
677 sock_hold(mptcp_subflow_tcp_sock(subflow));
678 if (schedule)
679 napi_schedule(&delegated->napi);
680 }
681}
682
683static inline struct mptcp_subflow_context *
684mptcp_subflow_delegated_next(struct mptcp_delegated_action *delegated)
685{
686 struct mptcp_subflow_context *ret;
687
688 if (list_empty(&delegated->head))
689 return NULL;
690
691 ret = list_first_entry(&delegated->head, struct mptcp_subflow_context, delegated_node);
692 list_del_init(&ret->delegated_node);
693 return ret;
694}
695
696int mptcp_is_enabled(const struct net *net);
697unsigned int mptcp_get_add_addr_timeout(const struct net *net);
698int mptcp_is_checksum_enabled(const struct net *net);
699int mptcp_allow_join_id0(const struct net *net);
700unsigned int mptcp_stale_loss_cnt(const struct net *net);
701unsigned int mptcp_close_timeout(const struct sock *sk);
702int mptcp_get_pm_type(const struct net *net);
703const char *mptcp_get_scheduler(const struct net *net);
704
705void mptcp_active_disable(struct sock *sk);
706bool mptcp_active_should_disable(struct sock *ssk);
707void mptcp_active_enable(struct sock *sk);
708
709void mptcp_get_available_schedulers(char *buf, size_t maxlen);
710void __mptcp_subflow_fully_established(struct mptcp_sock *msk,
711 struct mptcp_subflow_context *subflow,
712 const struct mptcp_options_received *mp_opt);
713bool __mptcp_retransmit_pending_data(struct sock *sk);
714void mptcp_check_and_set_pending(struct sock *sk);
715void __mptcp_push_pending(struct sock *sk, unsigned int flags);
716bool mptcp_subflow_data_available(struct sock *sk);
717void __init mptcp_subflow_init(void);
718void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how);
719void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
720 struct mptcp_subflow_context *subflow);
721void __mptcp_subflow_send_ack(struct sock *ssk);
722void mptcp_subflow_reset(struct sock *ssk);
723void mptcp_subflow_queue_clean(struct sock *sk, struct sock *ssk);
724void mptcp_sock_graft(struct sock *sk, struct socket *parent);
725struct sock *__mptcp_nmpc_sk(struct mptcp_sock *msk);
726bool __mptcp_close(struct sock *sk, long timeout);
727void mptcp_cancel_work(struct sock *sk);
728void __mptcp_unaccepted_force_close(struct sock *sk);
729void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk);
730void mptcp_set_state(struct sock *sk, int state);
731
732bool mptcp_addresses_equal(const struct mptcp_addr_info *a,
733 const struct mptcp_addr_info *b, bool use_port);
734void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr);
735
736/* called with sk socket lock held */
737int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_pm_local *local,
738 const struct mptcp_addr_info *remote);
739int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
740 struct socket **new_sock);
741void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
742 struct sockaddr_storage *addr,
743 unsigned short family);
744struct mptcp_sched_ops *mptcp_sched_find(const char *name);
745int mptcp_register_scheduler(struct mptcp_sched_ops *sched);
746void mptcp_unregister_scheduler(struct mptcp_sched_ops *sched);
747void mptcp_sched_init(void);
748int mptcp_init_sched(struct mptcp_sock *msk,
749 struct mptcp_sched_ops *sched);
750void mptcp_release_sched(struct mptcp_sock *msk);
751void mptcp_subflow_set_scheduled(struct mptcp_subflow_context *subflow,
752 bool scheduled);
753struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk);
754struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk);
755int mptcp_sched_get_send(struct mptcp_sock *msk);
756int mptcp_sched_get_retrans(struct mptcp_sock *msk);
757
758static inline u64 mptcp_data_avail(const struct mptcp_sock *msk)
759{
760 return READ_ONCE(msk->bytes_received) - READ_ONCE(msk->bytes_consumed);
761}
762
763static inline bool mptcp_epollin_ready(const struct sock *sk)
764{
765 u64 data_avail = mptcp_data_avail(mptcp_sk(sk));
766
767 if (!data_avail)
768 return false;
769
770 /* mptcp doesn't have to deal with small skbs in the receive queue,
771 * as it can always coalesce them
772 */
773 return (data_avail >= sk->sk_rcvlowat) ||
774 (mem_cgroup_sockets_enabled && sk->sk_memcg &&
775 mem_cgroup_under_socket_pressure(sk->sk_memcg)) ||
776 READ_ONCE(tcp_memory_pressure);
777}
778
779int mptcp_set_rcvlowat(struct sock *sk, int val);
780
781static inline bool __tcp_can_send(const struct sock *ssk)
782{
783 /* only send if our side has not closed yet */
784 return ((1 << inet_sk_state_load(ssk)) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT));
785}
786
787static inline bool __mptcp_subflow_active(struct mptcp_subflow_context *subflow)
788{
789 /* can't send if JOIN hasn't completed yet (i.e. is usable for mptcp) */
790 if (subflow->request_join && !READ_ONCE(subflow->fully_established))
791 return false;
792
793 return __tcp_can_send(mptcp_subflow_tcp_sock(subflow));
794}
795
796void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow);
797
798bool mptcp_subflow_active(struct mptcp_subflow_context *subflow);
799
800void mptcp_subflow_drop_ctx(struct sock *ssk);
801
802static inline void mptcp_subflow_tcp_fallback(struct sock *sk,
803 struct mptcp_subflow_context *ctx)
804{
805 sk->sk_data_ready = sock_def_readable;
806 sk->sk_state_change = ctx->tcp_state_change;
807 sk->sk_write_space = sk_stream_write_space;
808 sk->sk_error_report = ctx->tcp_error_report;
809
810 inet_csk(sk)->icsk_af_ops = ctx->icsk_af_ops;
811}
812
813void __init mptcp_proto_init(void);
814#if IS_ENABLED(CONFIG_MPTCP_IPV6)
815int __init mptcp_proto_v6_init(void);
816#endif
817
818struct sock *mptcp_sk_clone_init(const struct sock *sk,
819 const struct mptcp_options_received *mp_opt,
820 struct sock *ssk,
821 struct request_sock *req);
822void mptcp_get_options(const struct sk_buff *skb,
823 struct mptcp_options_received *mp_opt);
824
825void mptcp_finish_connect(struct sock *sk);
826void __mptcp_sync_state(struct sock *sk, int state);
827void mptcp_reset_tout_timer(struct mptcp_sock *msk, unsigned long fail_tout);
828
829static inline void mptcp_stop_tout_timer(struct sock *sk)
830{
831 if (!inet_csk(sk)->icsk_mtup.probe_timestamp)
832 return;
833
834 sk_stop_timer(sk, &sk->sk_timer);
835 inet_csk(sk)->icsk_mtup.probe_timestamp = 0;
836}
837
838static inline void mptcp_set_close_tout(struct sock *sk, unsigned long tout)
839{
840 /* avoid 0 timestamp, as that means no close timeout */
841 inet_csk(sk)->icsk_mtup.probe_timestamp = tout ? : 1;
842}
843
844static inline void mptcp_start_tout_timer(struct sock *sk)
845{
846 mptcp_set_close_tout(sk, tcp_jiffies32);
847 mptcp_reset_tout_timer(mptcp_sk(sk), 0);
848}
849
850static inline bool mptcp_is_fully_established(struct sock *sk)
851{
852 return inet_sk_state_load(sk) == TCP_ESTABLISHED &&
853 READ_ONCE(mptcp_sk(sk)->fully_established);
854}
855
856void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk);
857void mptcp_data_ready(struct sock *sk, struct sock *ssk);
858bool mptcp_finish_join(struct sock *sk);
859bool mptcp_schedule_work(struct sock *sk);
860int mptcp_setsockopt(struct sock *sk, int level, int optname,
861 sockptr_t optval, unsigned int optlen);
862int mptcp_getsockopt(struct sock *sk, int level, int optname,
863 char __user *optval, int __user *option);
864
865u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq);
866static inline u64 mptcp_expand_seq(u64 old_seq, u64 cur_seq, bool use_64bit)
867{
868 if (use_64bit)
869 return cur_seq;
870
871 return __mptcp_expand_seq(old_seq, cur_seq);
872}
873void __mptcp_check_push(struct sock *sk, struct sock *ssk);
874void __mptcp_data_acked(struct sock *sk);
875void __mptcp_error_report(struct sock *sk);
876bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit);
877static inline bool mptcp_data_fin_enabled(const struct mptcp_sock *msk)
878{
879 return READ_ONCE(msk->snd_data_fin_enable) &&
880 READ_ONCE(msk->write_seq) == READ_ONCE(msk->snd_nxt);
881}
882
883static inline u32 mptcp_notsent_lowat(const struct sock *sk)
884{
885 struct net *net = sock_net(sk);
886 u32 val;
887
888 val = READ_ONCE(mptcp_sk(sk)->notsent_lowat);
889 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
890}
891
892static inline bool mptcp_stream_memory_free(const struct sock *sk, int wake)
893{
894 const struct mptcp_sock *msk = mptcp_sk(sk);
895 u32 notsent_bytes;
896
897 notsent_bytes = READ_ONCE(msk->write_seq) - READ_ONCE(msk->snd_nxt);
898 return (notsent_bytes << wake) < mptcp_notsent_lowat(sk);
899}
900
901static inline bool __mptcp_stream_is_writeable(const struct sock *sk, int wake)
902{
903 return mptcp_stream_memory_free(sk, wake) &&
904 __sk_stream_is_writeable(sk, wake);
905}
906
907static inline void mptcp_write_space(struct sock *sk)
908{
909 /* pairs with memory barrier in mptcp_poll */
910 smp_mb();
911 if (mptcp_stream_memory_free(sk, 1))
912 sk_stream_write_space(sk);
913}
914
915static inline void __mptcp_sync_sndbuf(struct sock *sk)
916{
917 struct mptcp_subflow_context *subflow;
918 int ssk_sndbuf, new_sndbuf;
919
920 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
921 return;
922
923 new_sndbuf = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_wmem[0]);
924 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
925 ssk_sndbuf = READ_ONCE(mptcp_subflow_tcp_sock(subflow)->sk_sndbuf);
926
927 subflow->cached_sndbuf = ssk_sndbuf;
928 new_sndbuf += ssk_sndbuf;
929 }
930
931 /* the msk max wmem limit is <nr_subflows> * tcp wmem[2] */
932 WRITE_ONCE(sk->sk_sndbuf, new_sndbuf);
933 mptcp_write_space(sk);
934}
935
936/* The called held both the msk socket and the subflow socket locks,
937 * possibly under BH
938 */
939static inline void __mptcp_propagate_sndbuf(struct sock *sk, struct sock *ssk)
940{
941 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
942
943 if (READ_ONCE(ssk->sk_sndbuf) != subflow->cached_sndbuf)
944 __mptcp_sync_sndbuf(sk);
945}
946
947/* the caller held only the subflow socket lock, either in process or
948 * BH context. Additionally this can be called under the msk data lock,
949 * so we can't acquire such lock here: let the delegate action acquires
950 * the needed locks in suitable order.
951 */
952static inline void mptcp_propagate_sndbuf(struct sock *sk, struct sock *ssk)
953{
954 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
955
956 if (likely(READ_ONCE(ssk->sk_sndbuf) == subflow->cached_sndbuf))
957 return;
958
959 local_bh_disable();
960 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_SNDBUF);
961 local_bh_enable();
962}
963
964void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags);
965
966#define MPTCP_TOKEN_MAX_RETRIES 4
967
968void __init mptcp_token_init(void);
969static inline void mptcp_token_init_request(struct request_sock *req)
970{
971 mptcp_subflow_rsk(req)->token_node.pprev = NULL;
972}
973
974int mptcp_token_new_request(struct request_sock *req);
975void mptcp_token_destroy_request(struct request_sock *req);
976int mptcp_token_new_connect(struct sock *ssk);
977void mptcp_token_accept(struct mptcp_subflow_request_sock *r,
978 struct mptcp_sock *msk);
979bool mptcp_token_exists(u32 token);
980struct mptcp_sock *mptcp_token_get_sock(struct net *net, u32 token);
981struct mptcp_sock *mptcp_token_iter_next(const struct net *net, long *s_slot,
982 long *s_num);
983void mptcp_token_destroy(struct mptcp_sock *msk);
984
985void mptcp_crypto_key_sha(u64 key, u32 *token, u64 *idsn);
986
987void mptcp_crypto_hmac_sha(u64 key1, u64 key2, u8 *msg, int len, void *hmac);
988__sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum);
989
990void __init mptcp_pm_init(void);
991void mptcp_pm_data_init(struct mptcp_sock *msk);
992void mptcp_pm_data_reset(struct mptcp_sock *msk);
993int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
994 struct mptcp_addr_info *addr);
995int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
996 bool require_family,
997 struct mptcp_pm_addr_entry *entry);
998bool mptcp_pm_addr_families_match(const struct sock *sk,
999 const struct mptcp_addr_info *loc,
1000 const struct mptcp_addr_info *rem);
1001void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk);
1002void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk);
1003void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side);
1004void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk);
1005bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk);
1006void mptcp_pm_connection_closed(struct mptcp_sock *msk);
1007void mptcp_pm_subflow_established(struct mptcp_sock *msk);
1008bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk);
1009void mptcp_pm_subflow_check_next(struct mptcp_sock *msk,
1010 const struct mptcp_subflow_context *subflow);
1011void mptcp_pm_add_addr_received(const struct sock *ssk,
1012 const struct mptcp_addr_info *addr);
1013void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,
1014 const struct mptcp_addr_info *addr);
1015void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk);
1016bool mptcp_pm_nl_is_init_remote_addr(struct mptcp_sock *msk,
1017 const struct mptcp_addr_info *remote);
1018void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk);
1019void mptcp_pm_rm_addr_received(struct mptcp_sock *msk,
1020 const struct mptcp_rm_list *rm_list);
1021void mptcp_pm_mp_prio_received(struct sock *sk, u8 bkup);
1022void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq);
1023int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
1024 struct mptcp_addr_info *addr,
1025 struct mptcp_addr_info *rem,
1026 u8 bkup);
1027bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
1028 const struct mptcp_addr_info *addr);
1029void mptcp_pm_free_anno_list(struct mptcp_sock *msk);
1030bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk);
1031struct mptcp_pm_add_entry *
1032mptcp_pm_del_add_timer(struct mptcp_sock *msk,
1033 const struct mptcp_addr_info *addr, bool check_id);
1034struct mptcp_pm_add_entry *
1035mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk,
1036 const struct mptcp_addr_info *addr);
1037int mptcp_pm_set_flags(struct sk_buff *skb, struct genl_info *info);
1038int mptcp_pm_nl_set_flags(struct sk_buff *skb, struct genl_info *info);
1039int mptcp_userspace_pm_set_flags(struct sk_buff *skb, struct genl_info *info);
1040int mptcp_pm_announce_addr(struct mptcp_sock *msk,
1041 const struct mptcp_addr_info *addr,
1042 bool echo);
1043int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list);
1044void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list);
1045
1046void mptcp_free_local_addr_list(struct mptcp_sock *msk);
1047
1048void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
1049 const struct sock *ssk, gfp_t gfp);
1050void mptcp_event_addr_announced(const struct sock *ssk, const struct mptcp_addr_info *info);
1051void mptcp_event_addr_removed(const struct mptcp_sock *msk, u8 id);
1052void mptcp_event_pm_listener(const struct sock *ssk,
1053 enum mptcp_event_type event);
1054bool mptcp_userspace_pm_active(const struct mptcp_sock *msk);
1055
1056void __mptcp_fastopen_gen_msk_ackseq(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
1057 const struct mptcp_options_received *mp_opt);
1058void mptcp_fastopen_subflow_synack_set_params(struct mptcp_subflow_context *subflow,
1059 struct request_sock *req);
1060int mptcp_nl_fill_addr(struct sk_buff *skb,
1061 struct mptcp_pm_addr_entry *entry);
1062
1063static inline bool mptcp_pm_should_add_signal(struct mptcp_sock *msk)
1064{
1065 return READ_ONCE(msk->pm.addr_signal) &
1066 (BIT(MPTCP_ADD_ADDR_SIGNAL) | BIT(MPTCP_ADD_ADDR_ECHO));
1067}
1068
1069static inline bool mptcp_pm_should_add_signal_addr(struct mptcp_sock *msk)
1070{
1071 return READ_ONCE(msk->pm.addr_signal) & BIT(MPTCP_ADD_ADDR_SIGNAL);
1072}
1073
1074static inline bool mptcp_pm_should_add_signal_echo(struct mptcp_sock *msk)
1075{
1076 return READ_ONCE(msk->pm.addr_signal) & BIT(MPTCP_ADD_ADDR_ECHO);
1077}
1078
1079static inline bool mptcp_pm_should_rm_signal(struct mptcp_sock *msk)
1080{
1081 return READ_ONCE(msk->pm.addr_signal) & BIT(MPTCP_RM_ADDR_SIGNAL);
1082}
1083
1084static inline bool mptcp_pm_is_userspace(const struct mptcp_sock *msk)
1085{
1086 return READ_ONCE(msk->pm.pm_type) == MPTCP_PM_TYPE_USERSPACE;
1087}
1088
1089static inline bool mptcp_pm_is_kernel(const struct mptcp_sock *msk)
1090{
1091 return READ_ONCE(msk->pm.pm_type) == MPTCP_PM_TYPE_KERNEL;
1092}
1093
1094static inline unsigned int mptcp_add_addr_len(int family, bool echo, bool port)
1095{
1096 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1097
1098 if (family == AF_INET6)
1099 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1100 if (!echo)
1101 len += MPTCPOPT_THMAC_LEN;
1102 /* account for 2 trailing 'nop' options */
1103 if (port)
1104 len += TCPOLEN_MPTCP_PORT_LEN + TCPOLEN_MPTCP_PORT_ALIGN;
1105
1106 return len;
1107}
1108
1109static inline int mptcp_rm_addr_len(const struct mptcp_rm_list *rm_list)
1110{
1111 if (rm_list->nr == 0 || rm_list->nr > MPTCP_RM_IDS_MAX)
1112 return -EINVAL;
1113
1114 return TCPOLEN_MPTCP_RM_ADDR_BASE + roundup(rm_list->nr - 1, 4) + 1;
1115}
1116
1117bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, const struct sk_buff *skb,
1118 unsigned int opt_size, unsigned int remaining,
1119 struct mptcp_addr_info *addr, bool *echo,
1120 bool *drop_other_suboptions);
1121bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
1122 struct mptcp_rm_list *rm_list);
1123int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
1124int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc);
1125int mptcp_userspace_pm_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc);
1126bool mptcp_pm_is_backup(struct mptcp_sock *msk, struct sock_common *skc);
1127bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc);
1128bool mptcp_userspace_pm_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc);
1129int mptcp_pm_dump_addr(struct sk_buff *msg, struct netlink_callback *cb);
1130int mptcp_pm_nl_dump_addr(struct sk_buff *msg,
1131 struct netlink_callback *cb);
1132int mptcp_userspace_pm_dump_addr(struct sk_buff *msg,
1133 struct netlink_callback *cb);
1134int mptcp_pm_get_addr(struct sk_buff *skb, struct genl_info *info);
1135int mptcp_pm_nl_get_addr(struct sk_buff *skb, struct genl_info *info);
1136int mptcp_userspace_pm_get_addr(struct sk_buff *skb,
1137 struct genl_info *info);
1138
1139static inline u8 subflow_get_local_id(const struct mptcp_subflow_context *subflow)
1140{
1141 int local_id = READ_ONCE(subflow->local_id);
1142
1143 if (local_id < 0)
1144 return 0;
1145 return local_id;
1146}
1147
1148void __init mptcp_pm_nl_init(void);
1149void mptcp_pm_nl_work(struct mptcp_sock *msk);
1150unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk);
1151unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk);
1152unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk);
1153unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk);
1154
1155/* called under PM lock */
1156static inline void __mptcp_pm_close_subflow(struct mptcp_sock *msk)
1157{
1158 if (--msk->pm.subflows < mptcp_pm_get_subflows_max(msk))
1159 WRITE_ONCE(msk->pm.accept_subflow, true);
1160}
1161
1162static inline void mptcp_pm_close_subflow(struct mptcp_sock *msk)
1163{
1164 spin_lock_bh(&msk->pm.lock);
1165 __mptcp_pm_close_subflow(msk);
1166 spin_unlock_bh(&msk->pm.lock);
1167}
1168
1169void mptcp_sockopt_sync_locked(struct mptcp_sock *msk, struct sock *ssk);
1170
1171static inline struct mptcp_ext *mptcp_get_ext(const struct sk_buff *skb)
1172{
1173 return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP);
1174}
1175
1176void mptcp_diag_subflow_init(struct tcp_ulp_ops *ops);
1177
1178static inline bool __mptcp_check_fallback(const struct mptcp_sock *msk)
1179{
1180 return test_bit(MPTCP_FALLBACK_DONE, &msk->flags);
1181}
1182
1183static inline bool mptcp_check_fallback(const struct sock *sk)
1184{
1185 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1186 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1187
1188 return __mptcp_check_fallback(msk);
1189}
1190
1191static inline void __mptcp_do_fallback(struct mptcp_sock *msk)
1192{
1193 if (__mptcp_check_fallback(msk)) {
1194 pr_debug("TCP fallback already done (msk=%p)\n", msk);
1195 return;
1196 }
1197 set_bit(MPTCP_FALLBACK_DONE, &msk->flags);
1198}
1199
1200static inline bool __mptcp_has_initial_subflow(const struct mptcp_sock *msk)
1201{
1202 struct sock *ssk = READ_ONCE(msk->first);
1203
1204 return ssk && ((1 << inet_sk_state_load(ssk)) &
1205 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
1206 TCPF_SYN_RECV | TCPF_LISTEN));
1207}
1208
1209static inline void mptcp_do_fallback(struct sock *ssk)
1210{
1211 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1212 struct sock *sk = subflow->conn;
1213 struct mptcp_sock *msk;
1214
1215 msk = mptcp_sk(sk);
1216 __mptcp_do_fallback(msk);
1217 if (READ_ONCE(msk->snd_data_fin_enable) && !(ssk->sk_shutdown & SEND_SHUTDOWN)) {
1218 gfp_t saved_allocation = ssk->sk_allocation;
1219
1220 /* we are in a atomic (BH) scope, override ssk default for data
1221 * fin allocation
1222 */
1223 ssk->sk_allocation = GFP_ATOMIC;
1224 ssk->sk_shutdown |= SEND_SHUTDOWN;
1225 tcp_shutdown(ssk, SEND_SHUTDOWN);
1226 ssk->sk_allocation = saved_allocation;
1227 }
1228}
1229
1230#define pr_fallback(a) pr_debug("%s:fallback to TCP (msk=%p)\n", __func__, a)
1231
1232static inline void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
1233 struct mptcp_subflow_context *subflow)
1234{
1235 pr_fallback(msk);
1236 subflow->request_mptcp = 0;
1237 __mptcp_do_fallback(msk);
1238}
1239
1240static inline bool mptcp_check_infinite_map(struct sk_buff *skb)
1241{
1242 struct mptcp_ext *mpext;
1243
1244 mpext = skb ? mptcp_get_ext(skb) : NULL;
1245 if (mpext && mpext->infinite_map)
1246 return true;
1247
1248 return false;
1249}
1250
1251static inline bool is_active_ssk(struct mptcp_subflow_context *subflow)
1252{
1253 return (subflow->request_mptcp || subflow->request_join);
1254}
1255
1256static inline bool subflow_simultaneous_connect(struct sock *sk)
1257{
1258 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1259
1260 return (1 << sk->sk_state) &
1261 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING) &&
1262 is_active_ssk(subflow) &&
1263 !subflow->conn_finished;
1264}
1265
1266#ifdef CONFIG_SYN_COOKIES
1267void subflow_init_req_cookie_join_save(const struct mptcp_subflow_request_sock *subflow_req,
1268 struct sk_buff *skb);
1269bool mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subflow_req,
1270 struct sk_buff *skb);
1271void __init mptcp_join_cookie_init(void);
1272#else
1273static inline void
1274subflow_init_req_cookie_join_save(const struct mptcp_subflow_request_sock *subflow_req,
1275 struct sk_buff *skb) {}
1276static inline bool
1277mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subflow_req,
1278 struct sk_buff *skb)
1279{
1280 return false;
1281}
1282
1283static inline void mptcp_join_cookie_init(void) {}
1284#endif
1285
1286#endif /* __MPTCP_PROTOCOL_H */
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_WORK_RTX 1
117#define MPTCP_FALLBACK_DONE 2
118#define MPTCP_WORK_CLOSE_SUBFLOW 3
119
120/* MPTCP socket release cb flags */
121#define MPTCP_PUSH_PENDING 1
122#define MPTCP_CLEAN_UNA 2
123#define MPTCP_ERROR_REPORT 3
124#define MPTCP_RETRANSMIT 4
125#define MPTCP_FLUSH_JOIN_LIST 5
126#define MPTCP_SYNC_STATE 6
127#define MPTCP_SYNC_SNDBUF 7
128
129struct mptcp_skb_cb {
130 u64 map_seq;
131 u64 end_seq;
132 u32 offset;
133 u8 has_rxtstamp:1;
134};
135
136#define MPTCP_SKB_CB(__skb) ((struct mptcp_skb_cb *)&((__skb)->cb[0]))
137
138static inline bool before64(__u64 seq1, __u64 seq2)
139{
140 return (__s64)(seq1 - seq2) < 0;
141}
142
143#define after64(seq2, seq1) before64(seq1, seq2)
144
145struct mptcp_options_received {
146 u64 sndr_key;
147 u64 rcvr_key;
148 u64 data_ack;
149 u64 data_seq;
150 u32 subflow_seq;
151 u16 data_len;
152 __sum16 csum;
153 u16 suboptions;
154 u32 token;
155 u32 nonce;
156 u16 use_map:1,
157 dsn64:1,
158 data_fin:1,
159 use_ack:1,
160 ack64:1,
161 mpc_map:1,
162 reset_reason:4,
163 reset_transient:1,
164 echo:1,
165 backup:1,
166 deny_join_id0:1,
167 __unused:2;
168 u8 join_id;
169 u64 thmac;
170 u8 hmac[MPTCPOPT_HMAC_LEN];
171 struct mptcp_addr_info addr;
172 struct mptcp_rm_list rm_list;
173 u64 ahmac;
174 u64 fail_seq;
175};
176
177static inline __be32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field)
178{
179 return htonl((TCPOPT_MPTCP << 24) | (len << 16) | (subopt << 12) |
180 ((nib & 0xF) << 8) | field);
181}
182
183enum mptcp_pm_status {
184 MPTCP_PM_ADD_ADDR_RECEIVED,
185 MPTCP_PM_ADD_ADDR_SEND_ACK,
186 MPTCP_PM_RM_ADDR_RECEIVED,
187 MPTCP_PM_ESTABLISHED,
188 MPTCP_PM_SUBFLOW_ESTABLISHED,
189 MPTCP_PM_ALREADY_ESTABLISHED, /* persistent status, set after ESTABLISHED event */
190 MPTCP_PM_MPC_ENDPOINT_ACCOUNTED /* persistent status, set after MPC local address is
191 * accounted int id_avail_bitmap
192 */
193};
194
195enum mptcp_pm_type {
196 MPTCP_PM_TYPE_KERNEL = 0,
197 MPTCP_PM_TYPE_USERSPACE,
198
199 __MPTCP_PM_TYPE_NR,
200 __MPTCP_PM_TYPE_MAX = __MPTCP_PM_TYPE_NR - 1,
201};
202
203/* Status bits below MPTCP_PM_ALREADY_ESTABLISHED need pm worker actions */
204#define MPTCP_PM_WORK_MASK ((1 << MPTCP_PM_ALREADY_ESTABLISHED) - 1)
205
206enum mptcp_addr_signal_status {
207 MPTCP_ADD_ADDR_SIGNAL,
208 MPTCP_ADD_ADDR_ECHO,
209 MPTCP_RM_ADDR_SIGNAL,
210};
211
212/* max value of mptcp_addr_info.id */
213#define MPTCP_PM_MAX_ADDR_ID U8_MAX
214
215struct mptcp_pm_data {
216 struct mptcp_addr_info local;
217 struct mptcp_addr_info remote;
218 struct list_head anno_list;
219 struct list_head userspace_pm_local_addr_list;
220
221 spinlock_t lock; /*protects the whole PM data */
222
223 u8 addr_signal;
224 bool server_side;
225 bool work_pending;
226 bool accept_addr;
227 bool accept_subflow;
228 bool remote_deny_join_id0;
229 u8 add_addr_signaled;
230 u8 add_addr_accepted;
231 u8 local_addr_used;
232 u8 pm_type;
233 u8 subflows;
234 u8 status;
235 DECLARE_BITMAP(id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
236 struct mptcp_rm_list rm_list_tx;
237 struct mptcp_rm_list rm_list_rx;
238};
239
240struct mptcp_pm_addr_entry {
241 struct list_head list;
242 struct mptcp_addr_info addr;
243 u8 flags;
244 int ifindex;
245 struct socket *lsk;
246};
247
248struct mptcp_data_frag {
249 struct list_head list;
250 u64 data_seq;
251 u16 data_len;
252 u16 offset;
253 u16 overhead;
254 u16 already_sent;
255 struct page *page;
256};
257
258/* MPTCP connection sock */
259struct mptcp_sock {
260 /* inet_connection_sock must be the first member */
261 struct inet_connection_sock sk;
262 u64 local_key; /* protected by the first subflow socket lock
263 * lockless access read
264 */
265 u64 remote_key; /* same as above */
266 u64 write_seq;
267 u64 bytes_sent;
268 u64 snd_nxt;
269 u64 bytes_received;
270 u64 ack_seq;
271 atomic64_t rcv_wnd_sent;
272 u64 rcv_data_fin_seq;
273 u64 bytes_retrans;
274 u64 bytes_consumed;
275 int rmem_fwd_alloc;
276 int snd_burst;
277 int old_wspace;
278 u64 recovery_snd_nxt; /* in recovery mode accept up to this seq;
279 * recovery related fields are under data_lock
280 * protection
281 */
282 u64 bytes_acked;
283 u64 snd_una;
284 u64 wnd_end;
285 unsigned long timer_ival;
286 u32 token;
287 int rmem_released;
288 unsigned long flags;
289 unsigned long cb_flags;
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 pending_state; /* A subflow asked to set this sk_state,
300 * protected by the msk data lock
301 */
302 u8 mpc_endpoint_id;
303 u8 recvmsg_inq:1,
304 cork:1,
305 nodelay:1,
306 fastopening:1,
307 in_accept_queue:1,
308 free_first:1,
309 rcvspace_init:1;
310 u32 notsent_lowat;
311 int keepalive_cnt;
312 int keepalive_idle;
313 int keepalive_intvl;
314 struct work_struct work;
315 struct sk_buff *ooo_last_skb;
316 struct rb_root out_of_order_queue;
317 struct sk_buff_head receive_queue;
318 struct list_head conn_list;
319 struct list_head rtx_queue;
320 struct mptcp_data_frag *first_pending;
321 struct list_head join_list;
322 struct sock *first; /* The mptcp ops can safely dereference, using suitable
323 * ONCE annotation, the subflow outside the socket
324 * lock as such sock is freed after close().
325 */
326 struct mptcp_pm_data pm;
327 struct mptcp_sched_ops *sched;
328 struct {
329 u32 space; /* bytes copied in last measurement window */
330 u32 copied; /* bytes copied in this measurement window */
331 u64 time; /* start time of measurement window */
332 u64 rtt_us; /* last maximum rtt of subflows */
333 } rcvq_space;
334 u8 scaling_ratio;
335
336 u32 subflow_id;
337 u32 setsockopt_seq;
338 char ca_name[TCP_CA_NAME_MAX];
339};
340
341#define mptcp_data_lock(sk) spin_lock_bh(&(sk)->sk_lock.slock)
342#define mptcp_data_unlock(sk) spin_unlock_bh(&(sk)->sk_lock.slock)
343
344#define mptcp_for_each_subflow(__msk, __subflow) \
345 list_for_each_entry(__subflow, &((__msk)->conn_list), node)
346#define mptcp_for_each_subflow_safe(__msk, __subflow, __tmp) \
347 list_for_each_entry_safe(__subflow, __tmp, &((__msk)->conn_list), node)
348
349extern struct genl_family mptcp_genl_family;
350
351static inline void msk_owned_by_me(const struct mptcp_sock *msk)
352{
353 sock_owned_by_me((const struct sock *)msk);
354}
355
356#ifdef CONFIG_DEBUG_NET
357/* MPTCP-specific: we might (indirectly) call this helper with the wrong sk */
358#undef tcp_sk
359#define tcp_sk(ptr) ({ \
360 typeof(ptr) _ptr = (ptr); \
361 WARN_ON(_ptr->sk_protocol != IPPROTO_TCP); \
362 container_of_const(_ptr, struct tcp_sock, inet_conn.icsk_inet.sk); \
363})
364#define mptcp_sk(ptr) ({ \
365 typeof(ptr) _ptr = (ptr); \
366 WARN_ON(_ptr->sk_protocol != IPPROTO_MPTCP); \
367 container_of_const(_ptr, struct mptcp_sock, sk.icsk_inet.sk); \
368})
369
370#else /* !CONFIG_DEBUG_NET */
371#define mptcp_sk(ptr) container_of_const(ptr, struct mptcp_sock, sk.icsk_inet.sk)
372#endif
373
374/* the msk socket don't use the backlog, also account for the bulk
375 * free memory
376 */
377static inline int __mptcp_rmem(const struct sock *sk)
378{
379 return atomic_read(&sk->sk_rmem_alloc) - READ_ONCE(mptcp_sk(sk)->rmem_released);
380}
381
382static inline int mptcp_win_from_space(const struct sock *sk, int space)
383{
384 return __tcp_win_from_space(mptcp_sk(sk)->scaling_ratio, space);
385}
386
387static inline int __mptcp_space(const struct sock *sk)
388{
389 return mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - __mptcp_rmem(sk));
390}
391
392static inline struct mptcp_data_frag *mptcp_send_head(const struct sock *sk)
393{
394 const struct mptcp_sock *msk = mptcp_sk(sk);
395
396 return READ_ONCE(msk->first_pending);
397}
398
399static inline struct mptcp_data_frag *mptcp_send_next(struct sock *sk)
400{
401 struct mptcp_sock *msk = mptcp_sk(sk);
402 struct mptcp_data_frag *cur;
403
404 cur = msk->first_pending;
405 return list_is_last(&cur->list, &msk->rtx_queue) ? NULL :
406 list_next_entry(cur, list);
407}
408
409static inline struct mptcp_data_frag *mptcp_pending_tail(const struct sock *sk)
410{
411 const struct mptcp_sock *msk = mptcp_sk(sk);
412
413 if (!msk->first_pending)
414 return NULL;
415
416 if (WARN_ON_ONCE(list_empty(&msk->rtx_queue)))
417 return NULL;
418
419 return list_last_entry(&msk->rtx_queue, struct mptcp_data_frag, list);
420}
421
422static inline struct mptcp_data_frag *mptcp_rtx_head(struct sock *sk)
423{
424 struct mptcp_sock *msk = mptcp_sk(sk);
425
426 if (msk->snd_una == msk->snd_nxt)
427 return NULL;
428
429 return list_first_entry_or_null(&msk->rtx_queue, struct mptcp_data_frag, list);
430}
431
432struct csum_pseudo_header {
433 __be64 data_seq;
434 __be32 subflow_seq;
435 __be16 data_len;
436 __sum16 csum;
437};
438
439struct mptcp_subflow_request_sock {
440 struct tcp_request_sock sk;
441 u16 mp_capable : 1,
442 mp_join : 1,
443 backup : 1,
444 csum_reqd : 1,
445 allow_join_id0 : 1;
446 u8 local_id;
447 u8 remote_id;
448 u64 local_key;
449 u64 idsn;
450 u32 token;
451 u32 ssn_offset;
452 u64 thmac;
453 u32 local_nonce;
454 u32 remote_nonce;
455 struct mptcp_sock *msk;
456 struct hlist_nulls_node token_node;
457};
458
459static inline struct mptcp_subflow_request_sock *
460mptcp_subflow_rsk(const struct request_sock *rsk)
461{
462 return (struct mptcp_subflow_request_sock *)rsk;
463}
464
465struct mptcp_delegated_action {
466 struct napi_struct napi;
467 struct list_head head;
468};
469
470DECLARE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
471
472#define MPTCP_DELEGATE_SCHEDULED 0
473#define MPTCP_DELEGATE_SEND 1
474#define MPTCP_DELEGATE_ACK 2
475#define MPTCP_DELEGATE_SNDBUF 3
476
477#define MPTCP_DELEGATE_ACTIONS_MASK (~BIT(MPTCP_DELEGATE_SCHEDULED))
478/* MPTCP subflow context */
479struct mptcp_subflow_context {
480 struct list_head node;/* conn_list of subflows */
481
482 struct_group(reset,
483
484 unsigned long avg_pacing_rate; /* protected by msk socket lock */
485 u64 local_key;
486 u64 remote_key;
487 u64 idsn;
488 u64 map_seq;
489 u32 snd_isn;
490 u32 token;
491 u32 rel_write_seq;
492 u32 map_subflow_seq;
493 u32 ssn_offset;
494 u32 map_data_len;
495 __wsum map_data_csum;
496 u32 map_csum_len;
497 u32 request_mptcp : 1, /* send MP_CAPABLE */
498 request_join : 1, /* send MP_JOIN */
499 request_bkup : 1,
500 mp_capable : 1, /* remote is MPTCP capable */
501 mp_join : 1, /* remote is JOINing */
502 fully_established : 1, /* path validated */
503 pm_notified : 1, /* PM hook called for established status */
504 conn_finished : 1,
505 map_valid : 1,
506 map_csum_reqd : 1,
507 map_data_fin : 1,
508 mpc_map : 1,
509 backup : 1,
510 send_mp_prio : 1,
511 send_mp_fail : 1,
512 send_fastclose : 1,
513 send_infinite_map : 1,
514 remote_key_valid : 1, /* received the peer key from */
515 disposable : 1, /* ctx can be free at ulp release time */
516 stale : 1, /* unable to snd/rcv data, do not use for xmit */
517 valid_csum_seen : 1, /* at least one csum validated */
518 is_mptfo : 1, /* subflow is doing TFO */
519 __unused : 10;
520 bool data_avail;
521 bool scheduled;
522 u32 remote_nonce;
523 u64 thmac;
524 u32 local_nonce;
525 u32 remote_token;
526 union {
527 u8 hmac[MPTCPOPT_HMAC_LEN]; /* MPJ subflow only */
528 u64 iasn; /* initial ack sequence number, MPC subflows only */
529 };
530 s16 local_id; /* if negative not initialized yet */
531 u8 remote_id;
532 u8 reset_seen:1;
533 u8 reset_transient:1;
534 u8 reset_reason:4;
535 u8 stale_count;
536
537 u32 subflow_id;
538
539 long delegated_status;
540 unsigned long fail_tout;
541
542 );
543
544 struct list_head delegated_node; /* link into delegated_action, protected by local BH */
545
546 u32 setsockopt_seq;
547 u32 stale_rcv_tstamp;
548 int cached_sndbuf; /* sndbuf size when last synced with the msk sndbuf,
549 * protected by the msk socket lock
550 */
551
552 struct sock *tcp_sock; /* tcp sk backpointer */
553 struct sock *conn; /* parent mptcp_sock */
554 const struct inet_connection_sock_af_ops *icsk_af_ops;
555 void (*tcp_state_change)(struct sock *sk);
556 void (*tcp_error_report)(struct sock *sk);
557
558 struct rcu_head rcu;
559};
560
561static inline struct mptcp_subflow_context *
562mptcp_subflow_ctx(const struct sock *sk)
563{
564 struct inet_connection_sock *icsk = inet_csk(sk);
565
566 /* Use RCU on icsk_ulp_data only for sock diag code */
567 return (__force struct mptcp_subflow_context *)icsk->icsk_ulp_data;
568}
569
570static inline struct sock *
571mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow)
572{
573 return subflow->tcp_sock;
574}
575
576static inline void
577mptcp_subflow_ctx_reset(struct mptcp_subflow_context *subflow)
578{
579 memset(&subflow->reset, 0, sizeof(subflow->reset));
580 subflow->request_mptcp = 1;
581 WRITE_ONCE(subflow->local_id, -1);
582}
583
584static inline u64
585mptcp_subflow_get_map_offset(const struct mptcp_subflow_context *subflow)
586{
587 return tcp_sk(mptcp_subflow_tcp_sock(subflow))->copied_seq -
588 subflow->ssn_offset -
589 subflow->map_subflow_seq;
590}
591
592static inline u64
593mptcp_subflow_get_mapped_dsn(const struct mptcp_subflow_context *subflow)
594{
595 return subflow->map_seq + mptcp_subflow_get_map_offset(subflow);
596}
597
598void mptcp_subflow_process_delegated(struct sock *ssk, long actions);
599
600static inline void mptcp_subflow_delegate(struct mptcp_subflow_context *subflow, int action)
601{
602 long old, set_bits = BIT(MPTCP_DELEGATE_SCHEDULED) | BIT(action);
603 struct mptcp_delegated_action *delegated;
604 bool schedule;
605
606 /* the caller held the subflow bh socket lock */
607 lockdep_assert_in_softirq();
608
609 /* The implied barrier pairs with tcp_release_cb_override()
610 * mptcp_napi_poll(), and ensures the below list check sees list
611 * updates done prior to delegated status bits changes
612 */
613 old = set_mask_bits(&subflow->delegated_status, 0, set_bits);
614 if (!(old & BIT(MPTCP_DELEGATE_SCHEDULED))) {
615 if (WARN_ON_ONCE(!list_empty(&subflow->delegated_node)))
616 return;
617
618 delegated = this_cpu_ptr(&mptcp_delegated_actions);
619 schedule = list_empty(&delegated->head);
620 list_add_tail(&subflow->delegated_node, &delegated->head);
621 sock_hold(mptcp_subflow_tcp_sock(subflow));
622 if (schedule)
623 napi_schedule(&delegated->napi);
624 }
625}
626
627static inline struct mptcp_subflow_context *
628mptcp_subflow_delegated_next(struct mptcp_delegated_action *delegated)
629{
630 struct mptcp_subflow_context *ret;
631
632 if (list_empty(&delegated->head))
633 return NULL;
634
635 ret = list_first_entry(&delegated->head, struct mptcp_subflow_context, delegated_node);
636 list_del_init(&ret->delegated_node);
637 return ret;
638}
639
640int mptcp_is_enabled(const struct net *net);
641unsigned int mptcp_get_add_addr_timeout(const struct net *net);
642int mptcp_is_checksum_enabled(const struct net *net);
643int mptcp_allow_join_id0(const struct net *net);
644unsigned int mptcp_stale_loss_cnt(const struct net *net);
645unsigned int mptcp_close_timeout(const struct sock *sk);
646int mptcp_get_pm_type(const struct net *net);
647const char *mptcp_get_scheduler(const struct net *net);
648void __mptcp_subflow_fully_established(struct mptcp_sock *msk,
649 struct mptcp_subflow_context *subflow,
650 const struct mptcp_options_received *mp_opt);
651bool __mptcp_retransmit_pending_data(struct sock *sk);
652void mptcp_check_and_set_pending(struct sock *sk);
653void __mptcp_push_pending(struct sock *sk, unsigned int flags);
654bool mptcp_subflow_data_available(struct sock *sk);
655void __init mptcp_subflow_init(void);
656void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how);
657void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
658 struct mptcp_subflow_context *subflow);
659void __mptcp_subflow_send_ack(struct sock *ssk);
660void mptcp_subflow_reset(struct sock *ssk);
661void mptcp_subflow_queue_clean(struct sock *sk, struct sock *ssk);
662void mptcp_sock_graft(struct sock *sk, struct socket *parent);
663struct sock *__mptcp_nmpc_sk(struct mptcp_sock *msk);
664bool __mptcp_close(struct sock *sk, long timeout);
665void mptcp_cancel_work(struct sock *sk);
666void __mptcp_unaccepted_force_close(struct sock *sk);
667void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk);
668void mptcp_set_state(struct sock *sk, int state);
669
670bool mptcp_addresses_equal(const struct mptcp_addr_info *a,
671 const struct mptcp_addr_info *b, bool use_port);
672void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr);
673
674/* called with sk socket lock held */
675int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
676 const struct mptcp_addr_info *remote);
677int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
678 struct socket **new_sock);
679void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
680 struct sockaddr_storage *addr,
681 unsigned short family);
682struct mptcp_sched_ops *mptcp_sched_find(const char *name);
683int mptcp_register_scheduler(struct mptcp_sched_ops *sched);
684void mptcp_unregister_scheduler(struct mptcp_sched_ops *sched);
685void mptcp_sched_init(void);
686int mptcp_init_sched(struct mptcp_sock *msk,
687 struct mptcp_sched_ops *sched);
688void mptcp_release_sched(struct mptcp_sock *msk);
689void mptcp_subflow_set_scheduled(struct mptcp_subflow_context *subflow,
690 bool scheduled);
691struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk);
692struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk);
693int mptcp_sched_get_send(struct mptcp_sock *msk);
694int mptcp_sched_get_retrans(struct mptcp_sock *msk);
695
696static inline u64 mptcp_data_avail(const struct mptcp_sock *msk)
697{
698 return READ_ONCE(msk->bytes_received) - READ_ONCE(msk->bytes_consumed);
699}
700
701static inline bool mptcp_epollin_ready(const struct sock *sk)
702{
703 /* mptcp doesn't have to deal with small skbs in the receive queue,
704 * at it can always coalesce them
705 */
706 return (mptcp_data_avail(mptcp_sk(sk)) >= sk->sk_rcvlowat) ||
707 (mem_cgroup_sockets_enabled && sk->sk_memcg &&
708 mem_cgroup_under_socket_pressure(sk->sk_memcg)) ||
709 READ_ONCE(tcp_memory_pressure);
710}
711
712int mptcp_set_rcvlowat(struct sock *sk, int val);
713
714static inline bool __tcp_can_send(const struct sock *ssk)
715{
716 /* only send if our side has not closed yet */
717 return ((1 << inet_sk_state_load(ssk)) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT));
718}
719
720static inline bool __mptcp_subflow_active(struct mptcp_subflow_context *subflow)
721{
722 /* can't send if JOIN hasn't completed yet (i.e. is usable for mptcp) */
723 if (subflow->request_join && !subflow->fully_established)
724 return false;
725
726 return __tcp_can_send(mptcp_subflow_tcp_sock(subflow));
727}
728
729void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow);
730
731bool mptcp_subflow_active(struct mptcp_subflow_context *subflow);
732
733void mptcp_subflow_drop_ctx(struct sock *ssk);
734
735static inline void mptcp_subflow_tcp_fallback(struct sock *sk,
736 struct mptcp_subflow_context *ctx)
737{
738 sk->sk_data_ready = sock_def_readable;
739 sk->sk_state_change = ctx->tcp_state_change;
740 sk->sk_write_space = sk_stream_write_space;
741 sk->sk_error_report = ctx->tcp_error_report;
742
743 inet_csk(sk)->icsk_af_ops = ctx->icsk_af_ops;
744}
745
746void __init mptcp_proto_init(void);
747#if IS_ENABLED(CONFIG_MPTCP_IPV6)
748int __init mptcp_proto_v6_init(void);
749#endif
750
751struct sock *mptcp_sk_clone_init(const struct sock *sk,
752 const struct mptcp_options_received *mp_opt,
753 struct sock *ssk,
754 struct request_sock *req);
755void mptcp_get_options(const struct sk_buff *skb,
756 struct mptcp_options_received *mp_opt);
757
758void mptcp_finish_connect(struct sock *sk);
759void __mptcp_sync_state(struct sock *sk, int state);
760void mptcp_reset_tout_timer(struct mptcp_sock *msk, unsigned long fail_tout);
761
762static inline void mptcp_stop_tout_timer(struct sock *sk)
763{
764 if (!inet_csk(sk)->icsk_mtup.probe_timestamp)
765 return;
766
767 sk_stop_timer(sk, &sk->sk_timer);
768 inet_csk(sk)->icsk_mtup.probe_timestamp = 0;
769}
770
771static inline void mptcp_set_close_tout(struct sock *sk, unsigned long tout)
772{
773 /* avoid 0 timestamp, as that means no close timeout */
774 inet_csk(sk)->icsk_mtup.probe_timestamp = tout ? : 1;
775}
776
777static inline void mptcp_start_tout_timer(struct sock *sk)
778{
779 mptcp_set_close_tout(sk, tcp_jiffies32);
780 mptcp_reset_tout_timer(mptcp_sk(sk), 0);
781}
782
783static inline bool mptcp_is_fully_established(struct sock *sk)
784{
785 return inet_sk_state_load(sk) == TCP_ESTABLISHED &&
786 READ_ONCE(mptcp_sk(sk)->fully_established);
787}
788
789void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk);
790void mptcp_data_ready(struct sock *sk, struct sock *ssk);
791bool mptcp_finish_join(struct sock *sk);
792bool mptcp_schedule_work(struct sock *sk);
793int mptcp_setsockopt(struct sock *sk, int level, int optname,
794 sockptr_t optval, unsigned int optlen);
795int mptcp_getsockopt(struct sock *sk, int level, int optname,
796 char __user *optval, int __user *option);
797
798u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq);
799static inline u64 mptcp_expand_seq(u64 old_seq, u64 cur_seq, bool use_64bit)
800{
801 if (use_64bit)
802 return cur_seq;
803
804 return __mptcp_expand_seq(old_seq, cur_seq);
805}
806void __mptcp_check_push(struct sock *sk, struct sock *ssk);
807void __mptcp_data_acked(struct sock *sk);
808void __mptcp_error_report(struct sock *sk);
809bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit);
810static inline bool mptcp_data_fin_enabled(const struct mptcp_sock *msk)
811{
812 return READ_ONCE(msk->snd_data_fin_enable) &&
813 READ_ONCE(msk->write_seq) == READ_ONCE(msk->snd_nxt);
814}
815
816static inline u32 mptcp_notsent_lowat(const struct sock *sk)
817{
818 struct net *net = sock_net(sk);
819 u32 val;
820
821 val = READ_ONCE(mptcp_sk(sk)->notsent_lowat);
822 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
823}
824
825static inline bool mptcp_stream_memory_free(const struct sock *sk, int wake)
826{
827 const struct mptcp_sock *msk = mptcp_sk(sk);
828 u32 notsent_bytes;
829
830 notsent_bytes = READ_ONCE(msk->write_seq) - READ_ONCE(msk->snd_nxt);
831 return (notsent_bytes << wake) < mptcp_notsent_lowat(sk);
832}
833
834static inline bool __mptcp_stream_is_writeable(const struct sock *sk, int wake)
835{
836 return mptcp_stream_memory_free(sk, wake) &&
837 __sk_stream_is_writeable(sk, wake);
838}
839
840static inline void mptcp_write_space(struct sock *sk)
841{
842 /* pairs with memory barrier in mptcp_poll */
843 smp_mb();
844 if (mptcp_stream_memory_free(sk, 1))
845 sk_stream_write_space(sk);
846}
847
848static inline void __mptcp_sync_sndbuf(struct sock *sk)
849{
850 struct mptcp_subflow_context *subflow;
851 int ssk_sndbuf, new_sndbuf;
852
853 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
854 return;
855
856 new_sndbuf = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_wmem[0]);
857 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
858 ssk_sndbuf = READ_ONCE(mptcp_subflow_tcp_sock(subflow)->sk_sndbuf);
859
860 subflow->cached_sndbuf = ssk_sndbuf;
861 new_sndbuf += ssk_sndbuf;
862 }
863
864 /* the msk max wmem limit is <nr_subflows> * tcp wmem[2] */
865 WRITE_ONCE(sk->sk_sndbuf, new_sndbuf);
866 mptcp_write_space(sk);
867}
868
869/* The called held both the msk socket and the subflow socket locks,
870 * possibly under BH
871 */
872static inline void __mptcp_propagate_sndbuf(struct sock *sk, struct sock *ssk)
873{
874 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
875
876 if (READ_ONCE(ssk->sk_sndbuf) != subflow->cached_sndbuf)
877 __mptcp_sync_sndbuf(sk);
878}
879
880/* the caller held only the subflow socket lock, either in process or
881 * BH context. Additionally this can be called under the msk data lock,
882 * so we can't acquire such lock here: let the delegate action acquires
883 * the needed locks in suitable order.
884 */
885static inline void mptcp_propagate_sndbuf(struct sock *sk, struct sock *ssk)
886{
887 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
888
889 if (likely(READ_ONCE(ssk->sk_sndbuf) == subflow->cached_sndbuf))
890 return;
891
892 local_bh_disable();
893 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_SNDBUF);
894 local_bh_enable();
895}
896
897void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags);
898
899#define MPTCP_TOKEN_MAX_RETRIES 4
900
901void __init mptcp_token_init(void);
902static inline void mptcp_token_init_request(struct request_sock *req)
903{
904 mptcp_subflow_rsk(req)->token_node.pprev = NULL;
905}
906
907int mptcp_token_new_request(struct request_sock *req);
908void mptcp_token_destroy_request(struct request_sock *req);
909int mptcp_token_new_connect(struct sock *ssk);
910void mptcp_token_accept(struct mptcp_subflow_request_sock *r,
911 struct mptcp_sock *msk);
912bool mptcp_token_exists(u32 token);
913struct mptcp_sock *mptcp_token_get_sock(struct net *net, u32 token);
914struct mptcp_sock *mptcp_token_iter_next(const struct net *net, long *s_slot,
915 long *s_num);
916void mptcp_token_destroy(struct mptcp_sock *msk);
917
918void mptcp_crypto_key_sha(u64 key, u32 *token, u64 *idsn);
919
920void mptcp_crypto_hmac_sha(u64 key1, u64 key2, u8 *msg, int len, void *hmac);
921__sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum);
922
923void __init mptcp_pm_init(void);
924void mptcp_pm_data_init(struct mptcp_sock *msk);
925void mptcp_pm_data_reset(struct mptcp_sock *msk);
926int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
927 struct mptcp_addr_info *addr);
928int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
929 bool require_family,
930 struct mptcp_pm_addr_entry *entry);
931bool mptcp_pm_addr_families_match(const struct sock *sk,
932 const struct mptcp_addr_info *loc,
933 const struct mptcp_addr_info *rem);
934void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk);
935void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk);
936void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side);
937void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk);
938bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk);
939void mptcp_pm_connection_closed(struct mptcp_sock *msk);
940void mptcp_pm_subflow_established(struct mptcp_sock *msk);
941bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk);
942void mptcp_pm_subflow_check_next(struct mptcp_sock *msk,
943 const struct mptcp_subflow_context *subflow);
944void mptcp_pm_add_addr_received(const struct sock *ssk,
945 const struct mptcp_addr_info *addr);
946void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,
947 const struct mptcp_addr_info *addr);
948void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk);
949void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk);
950void mptcp_pm_rm_addr_received(struct mptcp_sock *msk,
951 const struct mptcp_rm_list *rm_list);
952void mptcp_pm_mp_prio_received(struct sock *sk, u8 bkup);
953void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq);
954int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
955 struct mptcp_addr_info *addr,
956 struct mptcp_addr_info *rem,
957 u8 bkup);
958bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
959 const struct mptcp_addr_info *addr);
960void mptcp_pm_free_anno_list(struct mptcp_sock *msk);
961bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk);
962struct mptcp_pm_add_entry *
963mptcp_pm_del_add_timer(struct mptcp_sock *msk,
964 const struct mptcp_addr_info *addr, bool check_id);
965struct mptcp_pm_add_entry *
966mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk,
967 const struct mptcp_addr_info *addr);
968int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk,
969 unsigned int id,
970 u8 *flags, int *ifindex);
971int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
972 u8 *flags, int *ifindex);
973int mptcp_userspace_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk,
974 unsigned int id,
975 u8 *flags, int *ifindex);
976int mptcp_pm_set_flags(struct sk_buff *skb, struct genl_info *info);
977int mptcp_pm_nl_set_flags(struct sk_buff *skb, struct genl_info *info);
978int mptcp_userspace_pm_set_flags(struct sk_buff *skb, struct genl_info *info);
979int mptcp_pm_announce_addr(struct mptcp_sock *msk,
980 const struct mptcp_addr_info *addr,
981 bool echo);
982int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list);
983int mptcp_pm_remove_subflow(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list);
984void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list);
985
986void mptcp_free_local_addr_list(struct mptcp_sock *msk);
987
988void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
989 const struct sock *ssk, gfp_t gfp);
990void mptcp_event_addr_announced(const struct sock *ssk, const struct mptcp_addr_info *info);
991void mptcp_event_addr_removed(const struct mptcp_sock *msk, u8 id);
992void mptcp_event_pm_listener(const struct sock *ssk,
993 enum mptcp_event_type event);
994bool mptcp_userspace_pm_active(const struct mptcp_sock *msk);
995
996void __mptcp_fastopen_gen_msk_ackseq(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
997 const struct mptcp_options_received *mp_opt);
998void mptcp_fastopen_subflow_synack_set_params(struct mptcp_subflow_context *subflow,
999 struct request_sock *req);
1000int mptcp_nl_fill_addr(struct sk_buff *skb,
1001 struct mptcp_pm_addr_entry *entry);
1002
1003static inline bool mptcp_pm_should_add_signal(struct mptcp_sock *msk)
1004{
1005 return READ_ONCE(msk->pm.addr_signal) &
1006 (BIT(MPTCP_ADD_ADDR_SIGNAL) | BIT(MPTCP_ADD_ADDR_ECHO));
1007}
1008
1009static inline bool mptcp_pm_should_add_signal_addr(struct mptcp_sock *msk)
1010{
1011 return READ_ONCE(msk->pm.addr_signal) & BIT(MPTCP_ADD_ADDR_SIGNAL);
1012}
1013
1014static inline bool mptcp_pm_should_add_signal_echo(struct mptcp_sock *msk)
1015{
1016 return READ_ONCE(msk->pm.addr_signal) & BIT(MPTCP_ADD_ADDR_ECHO);
1017}
1018
1019static inline bool mptcp_pm_should_rm_signal(struct mptcp_sock *msk)
1020{
1021 return READ_ONCE(msk->pm.addr_signal) & BIT(MPTCP_RM_ADDR_SIGNAL);
1022}
1023
1024static inline bool mptcp_pm_is_userspace(const struct mptcp_sock *msk)
1025{
1026 return READ_ONCE(msk->pm.pm_type) == MPTCP_PM_TYPE_USERSPACE;
1027}
1028
1029static inline bool mptcp_pm_is_kernel(const struct mptcp_sock *msk)
1030{
1031 return READ_ONCE(msk->pm.pm_type) == MPTCP_PM_TYPE_KERNEL;
1032}
1033
1034static inline unsigned int mptcp_add_addr_len(int family, bool echo, bool port)
1035{
1036 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1037
1038 if (family == AF_INET6)
1039 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1040 if (!echo)
1041 len += MPTCPOPT_THMAC_LEN;
1042 /* account for 2 trailing 'nop' options */
1043 if (port)
1044 len += TCPOLEN_MPTCP_PORT_LEN + TCPOLEN_MPTCP_PORT_ALIGN;
1045
1046 return len;
1047}
1048
1049static inline int mptcp_rm_addr_len(const struct mptcp_rm_list *rm_list)
1050{
1051 if (rm_list->nr == 0 || rm_list->nr > MPTCP_RM_IDS_MAX)
1052 return -EINVAL;
1053
1054 return TCPOLEN_MPTCP_RM_ADDR_BASE + roundup(rm_list->nr - 1, 4) + 1;
1055}
1056
1057bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, const struct sk_buff *skb,
1058 unsigned int opt_size, unsigned int remaining,
1059 struct mptcp_addr_info *addr, bool *echo,
1060 bool *drop_other_suboptions);
1061bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
1062 struct mptcp_rm_list *rm_list);
1063int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
1064int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc);
1065int mptcp_userspace_pm_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc);
1066int mptcp_pm_dump_addr(struct sk_buff *msg, struct netlink_callback *cb);
1067int mptcp_pm_nl_dump_addr(struct sk_buff *msg,
1068 struct netlink_callback *cb);
1069int mptcp_userspace_pm_dump_addr(struct sk_buff *msg,
1070 struct netlink_callback *cb);
1071int mptcp_pm_get_addr(struct sk_buff *skb, struct genl_info *info);
1072int mptcp_pm_nl_get_addr(struct sk_buff *skb, struct genl_info *info);
1073int mptcp_userspace_pm_get_addr(struct sk_buff *skb,
1074 struct genl_info *info);
1075
1076static inline u8 subflow_get_local_id(const struct mptcp_subflow_context *subflow)
1077{
1078 int local_id = READ_ONCE(subflow->local_id);
1079
1080 if (local_id < 0)
1081 return 0;
1082 return local_id;
1083}
1084
1085void __init mptcp_pm_nl_init(void);
1086void mptcp_pm_nl_work(struct mptcp_sock *msk);
1087void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
1088 const struct mptcp_rm_list *rm_list);
1089unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk);
1090unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk);
1091unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk);
1092unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk);
1093
1094/* called under PM lock */
1095static inline void __mptcp_pm_close_subflow(struct mptcp_sock *msk)
1096{
1097 if (--msk->pm.subflows < mptcp_pm_get_subflows_max(msk))
1098 WRITE_ONCE(msk->pm.accept_subflow, true);
1099}
1100
1101static inline void mptcp_pm_close_subflow(struct mptcp_sock *msk)
1102{
1103 spin_lock_bh(&msk->pm.lock);
1104 __mptcp_pm_close_subflow(msk);
1105 spin_unlock_bh(&msk->pm.lock);
1106}
1107
1108void mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk);
1109void mptcp_sockopt_sync_locked(struct mptcp_sock *msk, struct sock *ssk);
1110
1111static inline struct mptcp_ext *mptcp_get_ext(const struct sk_buff *skb)
1112{
1113 return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP);
1114}
1115
1116void mptcp_diag_subflow_init(struct tcp_ulp_ops *ops);
1117
1118static inline bool __mptcp_check_fallback(const struct mptcp_sock *msk)
1119{
1120 return test_bit(MPTCP_FALLBACK_DONE, &msk->flags);
1121}
1122
1123static inline bool mptcp_check_fallback(const struct sock *sk)
1124{
1125 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1126 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1127
1128 return __mptcp_check_fallback(msk);
1129}
1130
1131static inline void __mptcp_do_fallback(struct mptcp_sock *msk)
1132{
1133 if (__mptcp_check_fallback(msk)) {
1134 pr_debug("TCP fallback already done (msk=%p)", msk);
1135 return;
1136 }
1137 set_bit(MPTCP_FALLBACK_DONE, &msk->flags);
1138}
1139
1140static inline bool __mptcp_has_initial_subflow(const struct mptcp_sock *msk)
1141{
1142 struct sock *ssk = READ_ONCE(msk->first);
1143
1144 return ssk && ((1 << inet_sk_state_load(ssk)) &
1145 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
1146 TCPF_SYN_RECV | TCPF_LISTEN));
1147}
1148
1149static inline void mptcp_do_fallback(struct sock *ssk)
1150{
1151 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1152 struct sock *sk = subflow->conn;
1153 struct mptcp_sock *msk;
1154
1155 msk = mptcp_sk(sk);
1156 __mptcp_do_fallback(msk);
1157 if (READ_ONCE(msk->snd_data_fin_enable) && !(ssk->sk_shutdown & SEND_SHUTDOWN)) {
1158 gfp_t saved_allocation = ssk->sk_allocation;
1159
1160 /* we are in a atomic (BH) scope, override ssk default for data
1161 * fin allocation
1162 */
1163 ssk->sk_allocation = GFP_ATOMIC;
1164 ssk->sk_shutdown |= SEND_SHUTDOWN;
1165 tcp_shutdown(ssk, SEND_SHUTDOWN);
1166 ssk->sk_allocation = saved_allocation;
1167 }
1168}
1169
1170#define pr_fallback(a) pr_debug("%s:fallback to TCP (msk=%p)", __func__, a)
1171
1172static inline bool mptcp_check_infinite_map(struct sk_buff *skb)
1173{
1174 struct mptcp_ext *mpext;
1175
1176 mpext = skb ? mptcp_get_ext(skb) : NULL;
1177 if (mpext && mpext->infinite_map)
1178 return true;
1179
1180 return false;
1181}
1182
1183static inline bool is_active_ssk(struct mptcp_subflow_context *subflow)
1184{
1185 return (subflow->request_mptcp || subflow->request_join);
1186}
1187
1188static inline bool subflow_simultaneous_connect(struct sock *sk)
1189{
1190 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1191
1192 return (1 << sk->sk_state) &
1193 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING) &&
1194 is_active_ssk(subflow) &&
1195 !subflow->conn_finished;
1196}
1197
1198#ifdef CONFIG_SYN_COOKIES
1199void subflow_init_req_cookie_join_save(const struct mptcp_subflow_request_sock *subflow_req,
1200 struct sk_buff *skb);
1201bool mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subflow_req,
1202 struct sk_buff *skb);
1203void __init mptcp_join_cookie_init(void);
1204#else
1205static inline void
1206subflow_init_req_cookie_join_save(const struct mptcp_subflow_request_sock *subflow_req,
1207 struct sk_buff *skb) {}
1208static inline bool
1209mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subflow_req,
1210 struct sk_buff *skb)
1211{
1212 return false;
1213}
1214
1215static inline void mptcp_join_cookie_init(void) {}
1216#endif
1217
1218#endif /* __MPTCP_PROTOCOL_H */