xref: /openbmc/linux/net/mptcp/protocol.c (revision 1edd0337)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3  *
4  * Copyright (c) 2017 - 2019, Intel Corporation.
5  */
6 
7 #define pr_fmt(fmt) "MPTCP: " fmt
8 
9 #include <linux/kernel.h>
10 #include <linux/module.h>
11 #include <linux/netdevice.h>
12 #include <linux/sched/signal.h>
13 #include <linux/atomic.h>
14 #include <net/sock.h>
15 #include <net/inet_common.h>
16 #include <net/inet_hashtables.h>
17 #include <net/protocol.h>
18 #include <net/tcp.h>
19 #include <net/tcp_states.h>
20 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
21 #include <net/transp_v6.h>
22 #endif
23 #include <net/mptcp.h>
24 #include <net/xfrm.h>
25 #include <asm/ioctls.h>
26 #include "protocol.h"
27 #include "mib.h"
28 
29 #define CREATE_TRACE_POINTS
30 #include <trace/events/mptcp.h>
31 
32 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
33 struct mptcp6_sock {
34 	struct mptcp_sock msk;
35 	struct ipv6_pinfo np;
36 };
37 #endif
38 
39 enum {
40 	MPTCP_CMSG_TS = BIT(0),
41 	MPTCP_CMSG_INQ = BIT(1),
42 };
43 
44 static struct percpu_counter mptcp_sockets_allocated ____cacheline_aligned_in_smp;
45 
46 static void __mptcp_destroy_sock(struct sock *sk);
47 static void __mptcp_check_send_data_fin(struct sock *sk);
48 
49 DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
50 static struct net_device mptcp_napi_dev;
51 
52 /* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not
53  * completed yet or has failed, return the subflow socket.
54  * Otherwise return NULL.
55  */
56 struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
57 {
58 	if (!msk->subflow || READ_ONCE(msk->can_ack))
59 		return NULL;
60 
61 	return msk->subflow;
62 }
63 
64 /* Returns end sequence number of the receiver's advertised window */
65 static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
66 {
67 	return READ_ONCE(msk->wnd_end);
68 }
69 
70 static bool mptcp_is_tcpsk(struct sock *sk)
71 {
72 	struct socket *sock = sk->sk_socket;
73 
74 	if (unlikely(sk->sk_prot == &tcp_prot)) {
75 		/* we are being invoked after mptcp_accept() has
76 		 * accepted a non-mp-capable flow: sk is a tcp_sk,
77 		 * not an mptcp one.
78 		 *
79 		 * Hand the socket over to tcp so all further socket ops
80 		 * bypass mptcp.
81 		 */
82 		sock->ops = &inet_stream_ops;
83 		return true;
84 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
85 	} else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
86 		sock->ops = &inet6_stream_ops;
87 		return true;
88 #endif
89 	}
90 
91 	return false;
92 }
93 
94 static int __mptcp_socket_create(struct mptcp_sock *msk)
95 {
96 	struct mptcp_subflow_context *subflow;
97 	struct sock *sk = (struct sock *)msk;
98 	struct socket *ssock;
99 	int err;
100 
101 	err = mptcp_subflow_create_socket(sk, sk->sk_family, &ssock);
102 	if (err)
103 		return err;
104 
105 	msk->first = ssock->sk;
106 	msk->subflow = ssock;
107 	subflow = mptcp_subflow_ctx(ssock->sk);
108 	list_add(&subflow->node, &msk->conn_list);
109 	sock_hold(ssock->sk);
110 	subflow->request_mptcp = 1;
111 
112 	/* This is the first subflow, always with id 0 */
113 	subflow->local_id_valid = 1;
114 	mptcp_sock_graft(msk->first, sk->sk_socket);
115 
116 	return 0;
117 }
118 
119 static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
120 {
121 	sk_drops_add(sk, skb);
122 	__kfree_skb(skb);
123 }
124 
125 static void mptcp_rmem_charge(struct sock *sk, int size)
126 {
127 	mptcp_sk(sk)->rmem_fwd_alloc -= size;
128 }
129 
130 static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
131 			       struct sk_buff *from)
132 {
133 	bool fragstolen;
134 	int delta;
135 
136 	if (MPTCP_SKB_CB(from)->offset ||
137 	    !skb_try_coalesce(to, from, &fragstolen, &delta))
138 		return false;
139 
140 	pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
141 		 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
142 		 to->len, MPTCP_SKB_CB(from)->end_seq);
143 	MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
144 
145 	/* note the fwd memory can reach a negative value after accounting
146 	 * for the delta, but the later skb free will restore a non
147 	 * negative one
148 	 */
149 	atomic_add(delta, &sk->sk_rmem_alloc);
150 	mptcp_rmem_charge(sk, delta);
151 	kfree_skb_partial(from, fragstolen);
152 
153 	return true;
154 }
155 
156 static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
157 				   struct sk_buff *from)
158 {
159 	if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
160 		return false;
161 
162 	return mptcp_try_coalesce((struct sock *)msk, to, from);
163 }
164 
165 static void __mptcp_rmem_reclaim(struct sock *sk, int amount)
166 {
167 	amount >>= PAGE_SHIFT;
168 	mptcp_sk(sk)->rmem_fwd_alloc -= amount << PAGE_SHIFT;
169 	__sk_mem_reduce_allocated(sk, amount);
170 }
171 
172 static void mptcp_rmem_uncharge(struct sock *sk, int size)
173 {
174 	struct mptcp_sock *msk = mptcp_sk(sk);
175 	int reclaimable;
176 
177 	msk->rmem_fwd_alloc += size;
178 	reclaimable = msk->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
179 
180 	/* see sk_mem_uncharge() for the rationale behind the following schema */
181 	if (unlikely(reclaimable >= PAGE_SIZE))
182 		__mptcp_rmem_reclaim(sk, reclaimable);
183 }
184 
185 static void mptcp_rfree(struct sk_buff *skb)
186 {
187 	unsigned int len = skb->truesize;
188 	struct sock *sk = skb->sk;
189 
190 	atomic_sub(len, &sk->sk_rmem_alloc);
191 	mptcp_rmem_uncharge(sk, len);
192 }
193 
194 void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
195 {
196 	skb_orphan(skb);
197 	skb->sk = sk;
198 	skb->destructor = mptcp_rfree;
199 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
200 	mptcp_rmem_charge(sk, skb->truesize);
201 }
202 
203 /* "inspired" by tcp_data_queue_ofo(), main differences:
204  * - use mptcp seqs
205  * - don't cope with sacks
206  */
207 static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
208 {
209 	struct sock *sk = (struct sock *)msk;
210 	struct rb_node **p, *parent;
211 	u64 seq, end_seq, max_seq;
212 	struct sk_buff *skb1;
213 
214 	seq = MPTCP_SKB_CB(skb)->map_seq;
215 	end_seq = MPTCP_SKB_CB(skb)->end_seq;
216 	max_seq = atomic64_read(&msk->rcv_wnd_sent);
217 
218 	pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
219 		 RB_EMPTY_ROOT(&msk->out_of_order_queue));
220 	if (after64(end_seq, max_seq)) {
221 		/* out of window */
222 		mptcp_drop(sk, skb);
223 		pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
224 			 (unsigned long long)end_seq - (unsigned long)max_seq,
225 			 (unsigned long long)atomic64_read(&msk->rcv_wnd_sent));
226 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
227 		return;
228 	}
229 
230 	p = &msk->out_of_order_queue.rb_node;
231 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
232 	if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
233 		rb_link_node(&skb->rbnode, NULL, p);
234 		rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
235 		msk->ooo_last_skb = skb;
236 		goto end;
237 	}
238 
239 	/* with 2 subflows, adding at end of ooo queue is quite likely
240 	 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
241 	 */
242 	if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
243 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
244 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
245 		return;
246 	}
247 
248 	/* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
249 	if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
250 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
251 		parent = &msk->ooo_last_skb->rbnode;
252 		p = &parent->rb_right;
253 		goto insert;
254 	}
255 
256 	/* Find place to insert this segment. Handle overlaps on the way. */
257 	parent = NULL;
258 	while (*p) {
259 		parent = *p;
260 		skb1 = rb_to_skb(parent);
261 		if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
262 			p = &parent->rb_left;
263 			continue;
264 		}
265 		if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
266 			if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
267 				/* All the bits are present. Drop. */
268 				mptcp_drop(sk, skb);
269 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
270 				return;
271 			}
272 			if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
273 				/* partial overlap:
274 				 *     |     skb      |
275 				 *  |     skb1    |
276 				 * continue traversing
277 				 */
278 			} else {
279 				/* skb's seq == skb1's seq and skb covers skb1.
280 				 * Replace skb1 with skb.
281 				 */
282 				rb_replace_node(&skb1->rbnode, &skb->rbnode,
283 						&msk->out_of_order_queue);
284 				mptcp_drop(sk, skb1);
285 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
286 				goto merge_right;
287 			}
288 		} else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
289 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
290 			return;
291 		}
292 		p = &parent->rb_right;
293 	}
294 
295 insert:
296 	/* Insert segment into RB tree. */
297 	rb_link_node(&skb->rbnode, parent, p);
298 	rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
299 
300 merge_right:
301 	/* Remove other segments covered by skb. */
302 	while ((skb1 = skb_rb_next(skb)) != NULL) {
303 		if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
304 			break;
305 		rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
306 		mptcp_drop(sk, skb1);
307 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
308 	}
309 	/* If there is no skb after us, we are the last_skb ! */
310 	if (!skb1)
311 		msk->ooo_last_skb = skb;
312 
313 end:
314 	skb_condense(skb);
315 	mptcp_set_owner_r(skb, sk);
316 }
317 
318 static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
319 {
320 	struct mptcp_sock *msk = mptcp_sk(sk);
321 	int amt, amount;
322 
323 	if (size <= msk->rmem_fwd_alloc)
324 		return true;
325 
326 	size -= msk->rmem_fwd_alloc;
327 	amt = sk_mem_pages(size);
328 	amount = amt << PAGE_SHIFT;
329 	if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV))
330 		return false;
331 
332 	msk->rmem_fwd_alloc += amount;
333 	return true;
334 }
335 
336 static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
337 			     struct sk_buff *skb, unsigned int offset,
338 			     size_t copy_len)
339 {
340 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
341 	struct sock *sk = (struct sock *)msk;
342 	struct sk_buff *tail;
343 	bool has_rxtstamp;
344 
345 	__skb_unlink(skb, &ssk->sk_receive_queue);
346 
347 	skb_ext_reset(skb);
348 	skb_orphan(skb);
349 
350 	/* try to fetch required memory from subflow */
351 	if (!mptcp_rmem_schedule(sk, ssk, skb->truesize))
352 		goto drop;
353 
354 	has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
355 
356 	/* the skb map_seq accounts for the skb offset:
357 	 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
358 	 * value
359 	 */
360 	MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
361 	MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
362 	MPTCP_SKB_CB(skb)->offset = offset;
363 	MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
364 
365 	if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
366 		/* in sequence */
367 		WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
368 		tail = skb_peek_tail(&sk->sk_receive_queue);
369 		if (tail && mptcp_try_coalesce(sk, tail, skb))
370 			return true;
371 
372 		mptcp_set_owner_r(skb, sk);
373 		__skb_queue_tail(&sk->sk_receive_queue, skb);
374 		return true;
375 	} else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
376 		mptcp_data_queue_ofo(msk, skb);
377 		return false;
378 	}
379 
380 	/* old data, keep it simple and drop the whole pkt, sender
381 	 * will retransmit as needed, if needed.
382 	 */
383 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
384 drop:
385 	mptcp_drop(sk, skb);
386 	return false;
387 }
388 
389 static void mptcp_stop_timer(struct sock *sk)
390 {
391 	struct inet_connection_sock *icsk = inet_csk(sk);
392 
393 	sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
394 	mptcp_sk(sk)->timer_ival = 0;
395 }
396 
397 static void mptcp_close_wake_up(struct sock *sk)
398 {
399 	if (sock_flag(sk, SOCK_DEAD))
400 		return;
401 
402 	sk->sk_state_change(sk);
403 	if (sk->sk_shutdown == SHUTDOWN_MASK ||
404 	    sk->sk_state == TCP_CLOSE)
405 		sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
406 	else
407 		sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
408 }
409 
410 static bool mptcp_pending_data_fin_ack(struct sock *sk)
411 {
412 	struct mptcp_sock *msk = mptcp_sk(sk);
413 
414 	return !__mptcp_check_fallback(msk) &&
415 	       ((1 << sk->sk_state) &
416 		(TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
417 	       msk->write_seq == READ_ONCE(msk->snd_una);
418 }
419 
420 static void mptcp_check_data_fin_ack(struct sock *sk)
421 {
422 	struct mptcp_sock *msk = mptcp_sk(sk);
423 
424 	/* Look for an acknowledged DATA_FIN */
425 	if (mptcp_pending_data_fin_ack(sk)) {
426 		WRITE_ONCE(msk->snd_data_fin_enable, 0);
427 
428 		switch (sk->sk_state) {
429 		case TCP_FIN_WAIT1:
430 			inet_sk_state_store(sk, TCP_FIN_WAIT2);
431 			break;
432 		case TCP_CLOSING:
433 		case TCP_LAST_ACK:
434 			inet_sk_state_store(sk, TCP_CLOSE);
435 			break;
436 		}
437 
438 		mptcp_close_wake_up(sk);
439 	}
440 }
441 
442 static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
443 {
444 	struct mptcp_sock *msk = mptcp_sk(sk);
445 
446 	if (READ_ONCE(msk->rcv_data_fin) &&
447 	    ((1 << sk->sk_state) &
448 	     (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
449 		u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
450 
451 		if (msk->ack_seq == rcv_data_fin_seq) {
452 			if (seq)
453 				*seq = rcv_data_fin_seq;
454 
455 			return true;
456 		}
457 	}
458 
459 	return false;
460 }
461 
462 static void mptcp_set_datafin_timeout(const struct sock *sk)
463 {
464 	struct inet_connection_sock *icsk = inet_csk(sk);
465 	u32 retransmits;
466 
467 	retransmits = min_t(u32, icsk->icsk_retransmits,
468 			    ilog2(TCP_RTO_MAX / TCP_RTO_MIN));
469 
470 	mptcp_sk(sk)->timer_ival = TCP_RTO_MIN << retransmits;
471 }
472 
473 static void __mptcp_set_timeout(struct sock *sk, long tout)
474 {
475 	mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
476 }
477 
478 static long mptcp_timeout_from_subflow(const struct mptcp_subflow_context *subflow)
479 {
480 	const struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
481 
482 	return inet_csk(ssk)->icsk_pending && !subflow->stale_count ?
483 	       inet_csk(ssk)->icsk_timeout - jiffies : 0;
484 }
485 
486 static void mptcp_set_timeout(struct sock *sk)
487 {
488 	struct mptcp_subflow_context *subflow;
489 	long tout = 0;
490 
491 	mptcp_for_each_subflow(mptcp_sk(sk), subflow)
492 		tout = max(tout, mptcp_timeout_from_subflow(subflow));
493 	__mptcp_set_timeout(sk, tout);
494 }
495 
496 static inline bool tcp_can_send_ack(const struct sock *ssk)
497 {
498 	return !((1 << inet_sk_state_load(ssk)) &
499 	       (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
500 }
501 
502 void __mptcp_subflow_send_ack(struct sock *ssk)
503 {
504 	if (tcp_can_send_ack(ssk))
505 		tcp_send_ack(ssk);
506 }
507 
508 static void mptcp_subflow_send_ack(struct sock *ssk)
509 {
510 	bool slow;
511 
512 	slow = lock_sock_fast(ssk);
513 	__mptcp_subflow_send_ack(ssk);
514 	unlock_sock_fast(ssk, slow);
515 }
516 
517 static void mptcp_send_ack(struct mptcp_sock *msk)
518 {
519 	struct mptcp_subflow_context *subflow;
520 
521 	mptcp_for_each_subflow(msk, subflow)
522 		mptcp_subflow_send_ack(mptcp_subflow_tcp_sock(subflow));
523 }
524 
525 static void mptcp_subflow_cleanup_rbuf(struct sock *ssk)
526 {
527 	bool slow;
528 
529 	slow = lock_sock_fast(ssk);
530 	if (tcp_can_send_ack(ssk))
531 		tcp_cleanup_rbuf(ssk, 1);
532 	unlock_sock_fast(ssk, slow);
533 }
534 
535 static bool mptcp_subflow_could_cleanup(const struct sock *ssk, bool rx_empty)
536 {
537 	const struct inet_connection_sock *icsk = inet_csk(ssk);
538 	u8 ack_pending = READ_ONCE(icsk->icsk_ack.pending);
539 	const struct tcp_sock *tp = tcp_sk(ssk);
540 
541 	return (ack_pending & ICSK_ACK_SCHED) &&
542 		((READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->rcv_wup) >
543 		  READ_ONCE(icsk->icsk_ack.rcv_mss)) ||
544 		 (rx_empty && ack_pending &
545 			      (ICSK_ACK_PUSHED2 | ICSK_ACK_PUSHED)));
546 }
547 
548 static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
549 {
550 	int old_space = READ_ONCE(msk->old_wspace);
551 	struct mptcp_subflow_context *subflow;
552 	struct sock *sk = (struct sock *)msk;
553 	int space =  __mptcp_space(sk);
554 	bool cleanup, rx_empty;
555 
556 	cleanup = (space > 0) && (space >= (old_space << 1));
557 	rx_empty = !__mptcp_rmem(sk);
558 
559 	mptcp_for_each_subflow(msk, subflow) {
560 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
561 
562 		if (cleanup || mptcp_subflow_could_cleanup(ssk, rx_empty))
563 			mptcp_subflow_cleanup_rbuf(ssk);
564 	}
565 }
566 
567 static bool mptcp_check_data_fin(struct sock *sk)
568 {
569 	struct mptcp_sock *msk = mptcp_sk(sk);
570 	u64 rcv_data_fin_seq;
571 	bool ret = false;
572 
573 	if (__mptcp_check_fallback(msk))
574 		return ret;
575 
576 	/* Need to ack a DATA_FIN received from a peer while this side
577 	 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
578 	 * msk->rcv_data_fin was set when parsing the incoming options
579 	 * at the subflow level and the msk lock was not held, so this
580 	 * is the first opportunity to act on the DATA_FIN and change
581 	 * the msk state.
582 	 *
583 	 * If we are caught up to the sequence number of the incoming
584 	 * DATA_FIN, send the DATA_ACK now and do state transition.  If
585 	 * not caught up, do nothing and let the recv code send DATA_ACK
586 	 * when catching up.
587 	 */
588 
589 	if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
590 		WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
591 		WRITE_ONCE(msk->rcv_data_fin, 0);
592 
593 		sk->sk_shutdown |= RCV_SHUTDOWN;
594 		smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
595 
596 		switch (sk->sk_state) {
597 		case TCP_ESTABLISHED:
598 			inet_sk_state_store(sk, TCP_CLOSE_WAIT);
599 			break;
600 		case TCP_FIN_WAIT1:
601 			inet_sk_state_store(sk, TCP_CLOSING);
602 			break;
603 		case TCP_FIN_WAIT2:
604 			inet_sk_state_store(sk, TCP_CLOSE);
605 			break;
606 		default:
607 			/* Other states not expected */
608 			WARN_ON_ONCE(1);
609 			break;
610 		}
611 
612 		ret = true;
613 		mptcp_send_ack(msk);
614 		mptcp_close_wake_up(sk);
615 	}
616 	return ret;
617 }
618 
619 static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
620 					   struct sock *ssk,
621 					   unsigned int *bytes)
622 {
623 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
624 	struct sock *sk = (struct sock *)msk;
625 	unsigned int moved = 0;
626 	bool more_data_avail;
627 	struct tcp_sock *tp;
628 	bool done = false;
629 	int sk_rbuf;
630 
631 	sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
632 
633 	if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
634 		int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
635 
636 		if (unlikely(ssk_rbuf > sk_rbuf)) {
637 			WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
638 			sk_rbuf = ssk_rbuf;
639 		}
640 	}
641 
642 	pr_debug("msk=%p ssk=%p", msk, ssk);
643 	tp = tcp_sk(ssk);
644 	do {
645 		u32 map_remaining, offset;
646 		u32 seq = tp->copied_seq;
647 		struct sk_buff *skb;
648 		bool fin;
649 
650 		/* try to move as much data as available */
651 		map_remaining = subflow->map_data_len -
652 				mptcp_subflow_get_map_offset(subflow);
653 
654 		skb = skb_peek(&ssk->sk_receive_queue);
655 		if (!skb) {
656 			/* With racing move_skbs_to_msk() and __mptcp_move_skbs(),
657 			 * a different CPU can have already processed the pending
658 			 * data, stop here or we can enter an infinite loop
659 			 */
660 			if (!moved)
661 				done = true;
662 			break;
663 		}
664 
665 		if (__mptcp_check_fallback(msk)) {
666 			/* Under fallback skbs have no MPTCP extension and TCP could
667 			 * collapse them between the dummy map creation and the
668 			 * current dequeue. Be sure to adjust the map size.
669 			 */
670 			map_remaining = skb->len;
671 			subflow->map_data_len = skb->len;
672 		}
673 
674 		offset = seq - TCP_SKB_CB(skb)->seq;
675 		fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
676 		if (fin) {
677 			done = true;
678 			seq++;
679 		}
680 
681 		if (offset < skb->len) {
682 			size_t len = skb->len - offset;
683 
684 			if (tp->urg_data)
685 				done = true;
686 
687 			if (__mptcp_move_skb(msk, ssk, skb, offset, len))
688 				moved += len;
689 			seq += len;
690 
691 			if (WARN_ON_ONCE(map_remaining < len))
692 				break;
693 		} else {
694 			WARN_ON_ONCE(!fin);
695 			sk_eat_skb(ssk, skb);
696 			done = true;
697 		}
698 
699 		WRITE_ONCE(tp->copied_seq, seq);
700 		more_data_avail = mptcp_subflow_data_available(ssk);
701 
702 		if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
703 			done = true;
704 			break;
705 		}
706 	} while (more_data_avail);
707 
708 	*bytes += moved;
709 	return done;
710 }
711 
712 static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
713 {
714 	struct sock *sk = (struct sock *)msk;
715 	struct sk_buff *skb, *tail;
716 	bool moved = false;
717 	struct rb_node *p;
718 	u64 end_seq;
719 
720 	p = rb_first(&msk->out_of_order_queue);
721 	pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
722 	while (p) {
723 		skb = rb_to_skb(p);
724 		if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
725 			break;
726 
727 		p = rb_next(p);
728 		rb_erase(&skb->rbnode, &msk->out_of_order_queue);
729 
730 		if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
731 				      msk->ack_seq))) {
732 			mptcp_drop(sk, skb);
733 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
734 			continue;
735 		}
736 
737 		end_seq = MPTCP_SKB_CB(skb)->end_seq;
738 		tail = skb_peek_tail(&sk->sk_receive_queue);
739 		if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
740 			int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
741 
742 			/* skip overlapping data, if any */
743 			pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
744 				 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
745 				 delta);
746 			MPTCP_SKB_CB(skb)->offset += delta;
747 			MPTCP_SKB_CB(skb)->map_seq += delta;
748 			__skb_queue_tail(&sk->sk_receive_queue, skb);
749 		}
750 		msk->ack_seq = end_seq;
751 		moved = true;
752 	}
753 	return moved;
754 }
755 
756 /* In most cases we will be able to lock the mptcp socket.  If its already
757  * owned, we need to defer to the work queue to avoid ABBA deadlock.
758  */
759 static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
760 {
761 	struct sock *sk = (struct sock *)msk;
762 	unsigned int moved = 0;
763 
764 	__mptcp_move_skbs_from_subflow(msk, ssk, &moved);
765 	__mptcp_ofo_queue(msk);
766 	if (unlikely(ssk->sk_err)) {
767 		if (!sock_owned_by_user(sk))
768 			__mptcp_error_report(sk);
769 		else
770 			__set_bit(MPTCP_ERROR_REPORT,  &msk->cb_flags);
771 	}
772 
773 	/* If the moves have caught up with the DATA_FIN sequence number
774 	 * it's time to ack the DATA_FIN and change socket state, but
775 	 * this is not a good place to change state. Let the workqueue
776 	 * do it.
777 	 */
778 	if (mptcp_pending_data_fin(sk, NULL))
779 		mptcp_schedule_work(sk);
780 	return moved > 0;
781 }
782 
783 void mptcp_data_ready(struct sock *sk, struct sock *ssk)
784 {
785 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
786 	struct mptcp_sock *msk = mptcp_sk(sk);
787 	int sk_rbuf, ssk_rbuf;
788 
789 	/* The peer can send data while we are shutting down this
790 	 * subflow at msk destruction time, but we must avoid enqueuing
791 	 * more data to the msk receive queue
792 	 */
793 	if (unlikely(subflow->disposable))
794 		return;
795 
796 	ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
797 	sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
798 	if (unlikely(ssk_rbuf > sk_rbuf))
799 		sk_rbuf = ssk_rbuf;
800 
801 	/* over limit? can't append more skbs to msk, Also, no need to wake-up*/
802 	if (__mptcp_rmem(sk) > sk_rbuf) {
803 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RCVPRUNED);
804 		return;
805 	}
806 
807 	/* Wake-up the reader only for in-sequence data */
808 	mptcp_data_lock(sk);
809 	if (move_skbs_to_msk(msk, ssk))
810 		sk->sk_data_ready(sk);
811 
812 	mptcp_data_unlock(sk);
813 }
814 
815 static bool __mptcp_finish_join(struct mptcp_sock *msk, struct sock *ssk)
816 {
817 	struct sock *sk = (struct sock *)msk;
818 
819 	if (sk->sk_state != TCP_ESTABLISHED)
820 		return false;
821 
822 	/* attach to msk socket only after we are sure we will deal with it
823 	 * at close time
824 	 */
825 	if (sk->sk_socket && !ssk->sk_socket)
826 		mptcp_sock_graft(ssk, sk->sk_socket);
827 
828 	mptcp_sockopt_sync_locked(msk, ssk);
829 	return true;
830 }
831 
832 static void __mptcp_flush_join_list(struct sock *sk)
833 {
834 	struct mptcp_subflow_context *tmp, *subflow;
835 	struct mptcp_sock *msk = mptcp_sk(sk);
836 
837 	list_for_each_entry_safe(subflow, tmp, &msk->join_list, node) {
838 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
839 		bool slow = lock_sock_fast(ssk);
840 
841 		list_move_tail(&subflow->node, &msk->conn_list);
842 		if (!__mptcp_finish_join(msk, ssk))
843 			mptcp_subflow_reset(ssk);
844 		unlock_sock_fast(ssk, slow);
845 	}
846 }
847 
848 static bool mptcp_timer_pending(struct sock *sk)
849 {
850 	return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
851 }
852 
853 static void mptcp_reset_timer(struct sock *sk)
854 {
855 	struct inet_connection_sock *icsk = inet_csk(sk);
856 	unsigned long tout;
857 
858 	/* prevent rescheduling on close */
859 	if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
860 		return;
861 
862 	tout = mptcp_sk(sk)->timer_ival;
863 	sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
864 }
865 
866 bool mptcp_schedule_work(struct sock *sk)
867 {
868 	if (inet_sk_state_load(sk) != TCP_CLOSE &&
869 	    schedule_work(&mptcp_sk(sk)->work)) {
870 		/* each subflow already holds a reference to the sk, and the
871 		 * workqueue is invoked by a subflow, so sk can't go away here.
872 		 */
873 		sock_hold(sk);
874 		return true;
875 	}
876 	return false;
877 }
878 
879 void mptcp_subflow_eof(struct sock *sk)
880 {
881 	if (!test_and_set_bit(MPTCP_WORK_EOF, &mptcp_sk(sk)->flags))
882 		mptcp_schedule_work(sk);
883 }
884 
885 static void mptcp_check_for_eof(struct mptcp_sock *msk)
886 {
887 	struct mptcp_subflow_context *subflow;
888 	struct sock *sk = (struct sock *)msk;
889 	int receivers = 0;
890 
891 	mptcp_for_each_subflow(msk, subflow)
892 		receivers += !subflow->rx_eof;
893 	if (receivers)
894 		return;
895 
896 	if (!(sk->sk_shutdown & RCV_SHUTDOWN)) {
897 		/* hopefully temporary hack: propagate shutdown status
898 		 * to msk, when all subflows agree on it
899 		 */
900 		sk->sk_shutdown |= RCV_SHUTDOWN;
901 
902 		smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
903 		sk->sk_data_ready(sk);
904 	}
905 
906 	switch (sk->sk_state) {
907 	case TCP_ESTABLISHED:
908 		inet_sk_state_store(sk, TCP_CLOSE_WAIT);
909 		break;
910 	case TCP_FIN_WAIT1:
911 		inet_sk_state_store(sk, TCP_CLOSING);
912 		break;
913 	case TCP_FIN_WAIT2:
914 		inet_sk_state_store(sk, TCP_CLOSE);
915 		break;
916 	default:
917 		return;
918 	}
919 	mptcp_close_wake_up(sk);
920 }
921 
922 static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
923 {
924 	struct mptcp_subflow_context *subflow;
925 
926 	msk_owned_by_me(msk);
927 
928 	mptcp_for_each_subflow(msk, subflow) {
929 		if (READ_ONCE(subflow->data_avail))
930 			return mptcp_subflow_tcp_sock(subflow);
931 	}
932 
933 	return NULL;
934 }
935 
936 static bool mptcp_skb_can_collapse_to(u64 write_seq,
937 				      const struct sk_buff *skb,
938 				      const struct mptcp_ext *mpext)
939 {
940 	if (!tcp_skb_can_collapse_to(skb))
941 		return false;
942 
943 	/* can collapse only if MPTCP level sequence is in order and this
944 	 * mapping has not been xmitted yet
945 	 */
946 	return mpext && mpext->data_seq + mpext->data_len == write_seq &&
947 	       !mpext->frozen;
948 }
949 
950 /* we can append data to the given data frag if:
951  * - there is space available in the backing page_frag
952  * - the data frag tail matches the current page_frag free offset
953  * - the data frag end sequence number matches the current write seq
954  */
955 static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
956 				       const struct page_frag *pfrag,
957 				       const struct mptcp_data_frag *df)
958 {
959 	return df && pfrag->page == df->page &&
960 		pfrag->size - pfrag->offset > 0 &&
961 		pfrag->offset == (df->offset + df->data_len) &&
962 		df->data_seq + df->data_len == msk->write_seq;
963 }
964 
965 static void dfrag_uncharge(struct sock *sk, int len)
966 {
967 	sk_mem_uncharge(sk, len);
968 	sk_wmem_queued_add(sk, -len);
969 }
970 
971 static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
972 {
973 	int len = dfrag->data_len + dfrag->overhead;
974 
975 	list_del(&dfrag->list);
976 	dfrag_uncharge(sk, len);
977 	put_page(dfrag->page);
978 }
979 
980 static void __mptcp_clean_una(struct sock *sk)
981 {
982 	struct mptcp_sock *msk = mptcp_sk(sk);
983 	struct mptcp_data_frag *dtmp, *dfrag;
984 	u64 snd_una;
985 
986 	/* on fallback we just need to ignore snd_una, as this is really
987 	 * plain TCP
988 	 */
989 	if (__mptcp_check_fallback(msk))
990 		msk->snd_una = READ_ONCE(msk->snd_nxt);
991 
992 	snd_una = msk->snd_una;
993 	list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
994 		if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
995 			break;
996 
997 		if (unlikely(dfrag == msk->first_pending)) {
998 			/* in recovery mode can see ack after the current snd head */
999 			if (WARN_ON_ONCE(!msk->recovery))
1000 				break;
1001 
1002 			WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1003 		}
1004 
1005 		dfrag_clear(sk, dfrag);
1006 	}
1007 
1008 	dfrag = mptcp_rtx_head(sk);
1009 	if (dfrag && after64(snd_una, dfrag->data_seq)) {
1010 		u64 delta = snd_una - dfrag->data_seq;
1011 
1012 		/* prevent wrap around in recovery mode */
1013 		if (unlikely(delta > dfrag->already_sent)) {
1014 			if (WARN_ON_ONCE(!msk->recovery))
1015 				goto out;
1016 			if (WARN_ON_ONCE(delta > dfrag->data_len))
1017 				goto out;
1018 			dfrag->already_sent += delta - dfrag->already_sent;
1019 		}
1020 
1021 		dfrag->data_seq += delta;
1022 		dfrag->offset += delta;
1023 		dfrag->data_len -= delta;
1024 		dfrag->already_sent -= delta;
1025 
1026 		dfrag_uncharge(sk, delta);
1027 	}
1028 
1029 	/* all retransmitted data acked, recovery completed */
1030 	if (unlikely(msk->recovery) && after64(msk->snd_una, msk->recovery_snd_nxt))
1031 		msk->recovery = false;
1032 
1033 out:
1034 	if (snd_una == READ_ONCE(msk->snd_nxt) &&
1035 	    snd_una == READ_ONCE(msk->write_seq)) {
1036 		if (mptcp_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
1037 			mptcp_stop_timer(sk);
1038 	} else {
1039 		mptcp_reset_timer(sk);
1040 	}
1041 }
1042 
1043 static void __mptcp_clean_una_wakeup(struct sock *sk)
1044 {
1045 	lockdep_assert_held_once(&sk->sk_lock.slock);
1046 
1047 	__mptcp_clean_una(sk);
1048 	mptcp_write_space(sk);
1049 }
1050 
1051 static void mptcp_clean_una_wakeup(struct sock *sk)
1052 {
1053 	mptcp_data_lock(sk);
1054 	__mptcp_clean_una_wakeup(sk);
1055 	mptcp_data_unlock(sk);
1056 }
1057 
1058 static void mptcp_enter_memory_pressure(struct sock *sk)
1059 {
1060 	struct mptcp_subflow_context *subflow;
1061 	struct mptcp_sock *msk = mptcp_sk(sk);
1062 	bool first = true;
1063 
1064 	sk_stream_moderate_sndbuf(sk);
1065 	mptcp_for_each_subflow(msk, subflow) {
1066 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1067 
1068 		if (first)
1069 			tcp_enter_memory_pressure(ssk);
1070 		sk_stream_moderate_sndbuf(ssk);
1071 		first = false;
1072 	}
1073 }
1074 
1075 /* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1076  * data
1077  */
1078 static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1079 {
1080 	if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1081 					pfrag, sk->sk_allocation)))
1082 		return true;
1083 
1084 	mptcp_enter_memory_pressure(sk);
1085 	return false;
1086 }
1087 
1088 static struct mptcp_data_frag *
1089 mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1090 		      int orig_offset)
1091 {
1092 	int offset = ALIGN(orig_offset, sizeof(long));
1093 	struct mptcp_data_frag *dfrag;
1094 
1095 	dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1096 	dfrag->data_len = 0;
1097 	dfrag->data_seq = msk->write_seq;
1098 	dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1099 	dfrag->offset = offset + sizeof(struct mptcp_data_frag);
1100 	dfrag->already_sent = 0;
1101 	dfrag->page = pfrag->page;
1102 
1103 	return dfrag;
1104 }
1105 
1106 struct mptcp_sendmsg_info {
1107 	int mss_now;
1108 	int size_goal;
1109 	u16 limit;
1110 	u16 sent;
1111 	unsigned int flags;
1112 	bool data_lock_held;
1113 };
1114 
1115 static int mptcp_check_allowed_size(const struct mptcp_sock *msk, struct sock *ssk,
1116 				    u64 data_seq, int avail_size)
1117 {
1118 	u64 window_end = mptcp_wnd_end(msk);
1119 	u64 mptcp_snd_wnd;
1120 
1121 	if (__mptcp_check_fallback(msk))
1122 		return avail_size;
1123 
1124 	mptcp_snd_wnd = window_end - data_seq;
1125 	avail_size = min_t(unsigned int, mptcp_snd_wnd, avail_size);
1126 
1127 	if (unlikely(tcp_sk(ssk)->snd_wnd < mptcp_snd_wnd)) {
1128 		tcp_sk(ssk)->snd_wnd = min_t(u64, U32_MAX, mptcp_snd_wnd);
1129 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_SNDWNDSHARED);
1130 	}
1131 
1132 	return avail_size;
1133 }
1134 
1135 static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1136 {
1137 	struct skb_ext *mpext = __skb_ext_alloc(gfp);
1138 
1139 	if (!mpext)
1140 		return false;
1141 	__skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1142 	return true;
1143 }
1144 
1145 static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
1146 {
1147 	struct sk_buff *skb;
1148 
1149 	skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
1150 	if (likely(skb)) {
1151 		if (likely(__mptcp_add_ext(skb, gfp))) {
1152 			skb_reserve(skb, MAX_TCP_HEADER);
1153 			skb->ip_summed = CHECKSUM_PARTIAL;
1154 			INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
1155 			return skb;
1156 		}
1157 		__kfree_skb(skb);
1158 	} else {
1159 		mptcp_enter_memory_pressure(sk);
1160 	}
1161 	return NULL;
1162 }
1163 
1164 static struct sk_buff *__mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
1165 {
1166 	struct sk_buff *skb;
1167 
1168 	skb = __mptcp_do_alloc_tx_skb(sk, gfp);
1169 	if (!skb)
1170 		return NULL;
1171 
1172 	if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1173 		tcp_skb_entail(ssk, skb);
1174 		return skb;
1175 	}
1176 	tcp_skb_tsorted_anchor_cleanup(skb);
1177 	kfree_skb(skb);
1178 	return NULL;
1179 }
1180 
1181 static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, bool data_lock_held)
1182 {
1183 	gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
1184 
1185 	return __mptcp_alloc_tx_skb(sk, ssk, gfp);
1186 }
1187 
1188 /* note: this always recompute the csum on the whole skb, even
1189  * if we just appended a single frag. More status info needed
1190  */
1191 static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
1192 {
1193 	struct mptcp_ext *mpext = mptcp_get_ext(skb);
1194 	__wsum csum = ~csum_unfold(mpext->csum);
1195 	int offset = skb->len - added;
1196 
1197 	mpext->csum = csum_fold(csum_block_add(csum, skb_checksum(skb, offset, added, 0), offset));
1198 }
1199 
1200 static void mptcp_update_infinite_map(struct mptcp_sock *msk,
1201 				      struct sock *ssk,
1202 				      struct mptcp_ext *mpext)
1203 {
1204 	if (!mpext)
1205 		return;
1206 
1207 	mpext->infinite_map = 1;
1208 	mpext->data_len = 0;
1209 
1210 	MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPTX);
1211 	mptcp_subflow_ctx(ssk)->send_infinite_map = 0;
1212 	pr_fallback(msk);
1213 	mptcp_do_fallback(ssk);
1214 }
1215 
1216 static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
1217 			      struct mptcp_data_frag *dfrag,
1218 			      struct mptcp_sendmsg_info *info)
1219 {
1220 	u64 data_seq = dfrag->data_seq + info->sent;
1221 	int offset = dfrag->offset + info->sent;
1222 	struct mptcp_sock *msk = mptcp_sk(sk);
1223 	bool zero_window_probe = false;
1224 	struct mptcp_ext *mpext = NULL;
1225 	bool can_coalesce = false;
1226 	bool reuse_skb = true;
1227 	struct sk_buff *skb;
1228 	size_t copy;
1229 	int i;
1230 
1231 	pr_debug("msk=%p ssk=%p sending dfrag at seq=%llu len=%u already sent=%u",
1232 		 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1233 
1234 	if (WARN_ON_ONCE(info->sent > info->limit ||
1235 			 info->limit > dfrag->data_len))
1236 		return 0;
1237 
1238 	if (unlikely(!__tcp_can_send(ssk)))
1239 		return -EAGAIN;
1240 
1241 	/* compute send limit */
1242 	info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
1243 	copy = info->size_goal;
1244 
1245 	skb = tcp_write_queue_tail(ssk);
1246 	if (skb && copy > skb->len) {
1247 		/* Limit the write to the size available in the
1248 		 * current skb, if any, so that we create at most a new skb.
1249 		 * Explicitly tells TCP internals to avoid collapsing on later
1250 		 * queue management operation, to avoid breaking the ext <->
1251 		 * SSN association set here
1252 		 */
1253 		mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1254 		if (!mptcp_skb_can_collapse_to(data_seq, skb, mpext)) {
1255 			TCP_SKB_CB(skb)->eor = 1;
1256 			goto alloc_skb;
1257 		}
1258 
1259 		i = skb_shinfo(skb)->nr_frags;
1260 		can_coalesce = skb_can_coalesce(skb, i, dfrag->page, offset);
1261 		if (!can_coalesce && i >= READ_ONCE(sysctl_max_skb_frags)) {
1262 			tcp_mark_push(tcp_sk(ssk), skb);
1263 			goto alloc_skb;
1264 		}
1265 
1266 		copy -= skb->len;
1267 	} else {
1268 alloc_skb:
1269 		skb = mptcp_alloc_tx_skb(sk, ssk, info->data_lock_held);
1270 		if (!skb)
1271 			return -ENOMEM;
1272 
1273 		i = skb_shinfo(skb)->nr_frags;
1274 		reuse_skb = false;
1275 		mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1276 	}
1277 
1278 	/* Zero window and all data acked? Probe. */
1279 	copy = mptcp_check_allowed_size(msk, ssk, data_seq, copy);
1280 	if (copy == 0) {
1281 		u64 snd_una = READ_ONCE(msk->snd_una);
1282 
1283 		if (snd_una != msk->snd_nxt) {
1284 			tcp_remove_empty_skb(ssk);
1285 			return 0;
1286 		}
1287 
1288 		zero_window_probe = true;
1289 		data_seq = snd_una - 1;
1290 		copy = 1;
1291 
1292 		/* all mptcp-level data is acked, no skbs should be present into the
1293 		 * ssk write queue
1294 		 */
1295 		WARN_ON_ONCE(reuse_skb);
1296 	}
1297 
1298 	copy = min_t(size_t, copy, info->limit - info->sent);
1299 	if (!sk_wmem_schedule(ssk, copy)) {
1300 		tcp_remove_empty_skb(ssk);
1301 		return -ENOMEM;
1302 	}
1303 
1304 	if (can_coalesce) {
1305 		skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1306 	} else {
1307 		get_page(dfrag->page);
1308 		skb_fill_page_desc(skb, i, dfrag->page, offset, copy);
1309 	}
1310 
1311 	skb->len += copy;
1312 	skb->data_len += copy;
1313 	skb->truesize += copy;
1314 	sk_wmem_queued_add(ssk, copy);
1315 	sk_mem_charge(ssk, copy);
1316 	WRITE_ONCE(tcp_sk(ssk)->write_seq, tcp_sk(ssk)->write_seq + copy);
1317 	TCP_SKB_CB(skb)->end_seq += copy;
1318 	tcp_skb_pcount_set(skb, 0);
1319 
1320 	/* on skb reuse we just need to update the DSS len */
1321 	if (reuse_skb) {
1322 		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1323 		mpext->data_len += copy;
1324 		WARN_ON_ONCE(zero_window_probe);
1325 		goto out;
1326 	}
1327 
1328 	memset(mpext, 0, sizeof(*mpext));
1329 	mpext->data_seq = data_seq;
1330 	mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
1331 	mpext->data_len = copy;
1332 	mpext->use_map = 1;
1333 	mpext->dsn64 = 1;
1334 
1335 	pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1336 		 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1337 		 mpext->dsn64);
1338 
1339 	if (zero_window_probe) {
1340 		mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1341 		mpext->frozen = 1;
1342 		if (READ_ONCE(msk->csum_enabled))
1343 			mptcp_update_data_checksum(skb, copy);
1344 		tcp_push_pending_frames(ssk);
1345 		return 0;
1346 	}
1347 out:
1348 	if (READ_ONCE(msk->csum_enabled))
1349 		mptcp_update_data_checksum(skb, copy);
1350 	if (mptcp_subflow_ctx(ssk)->send_infinite_map)
1351 		mptcp_update_infinite_map(msk, ssk, mpext);
1352 	trace_mptcp_sendmsg_frag(mpext);
1353 	mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1354 	return copy;
1355 }
1356 
1357 #define MPTCP_SEND_BURST_SIZE		((1 << 16) - \
1358 					 sizeof(struct tcphdr) - \
1359 					 MAX_TCP_OPTION_SPACE - \
1360 					 sizeof(struct ipv6hdr) - \
1361 					 sizeof(struct frag_hdr))
1362 
1363 struct subflow_send_info {
1364 	struct sock *ssk;
1365 	u64 linger_time;
1366 };
1367 
1368 void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow)
1369 {
1370 	if (!subflow->stale)
1371 		return;
1372 
1373 	subflow->stale = 0;
1374 	MPTCP_INC_STATS(sock_net(mptcp_subflow_tcp_sock(subflow)), MPTCP_MIB_SUBFLOWRECOVER);
1375 }
1376 
1377 bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
1378 {
1379 	if (unlikely(subflow->stale)) {
1380 		u32 rcv_tstamp = READ_ONCE(tcp_sk(mptcp_subflow_tcp_sock(subflow))->rcv_tstamp);
1381 
1382 		if (subflow->stale_rcv_tstamp == rcv_tstamp)
1383 			return false;
1384 
1385 		mptcp_subflow_set_active(subflow);
1386 	}
1387 	return __mptcp_subflow_active(subflow);
1388 }
1389 
1390 #define SSK_MODE_ACTIVE	0
1391 #define SSK_MODE_BACKUP	1
1392 #define SSK_MODE_MAX	2
1393 
1394 /* implement the mptcp packet scheduler;
1395  * returns the subflow that will transmit the next DSS
1396  * additionally updates the rtx timeout
1397  */
1398 static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
1399 {
1400 	struct subflow_send_info send_info[SSK_MODE_MAX];
1401 	struct mptcp_subflow_context *subflow;
1402 	struct sock *sk = (struct sock *)msk;
1403 	u32 pace, burst, wmem;
1404 	int i, nr_active = 0;
1405 	struct sock *ssk;
1406 	u64 linger_time;
1407 	long tout = 0;
1408 
1409 	msk_owned_by_me(msk);
1410 
1411 	if (__mptcp_check_fallback(msk)) {
1412 		if (!msk->first)
1413 			return NULL;
1414 		return __tcp_can_send(msk->first) &&
1415 		       sk_stream_memory_free(msk->first) ? msk->first : NULL;
1416 	}
1417 
1418 	/* re-use last subflow, if the burst allow that */
1419 	if (msk->last_snd && msk->snd_burst > 0 &&
1420 	    sk_stream_memory_free(msk->last_snd) &&
1421 	    mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
1422 		mptcp_set_timeout(sk);
1423 		return msk->last_snd;
1424 	}
1425 
1426 	/* pick the subflow with the lower wmem/wspace ratio */
1427 	for (i = 0; i < SSK_MODE_MAX; ++i) {
1428 		send_info[i].ssk = NULL;
1429 		send_info[i].linger_time = -1;
1430 	}
1431 
1432 	mptcp_for_each_subflow(msk, subflow) {
1433 		trace_mptcp_subflow_get_send(subflow);
1434 		ssk =  mptcp_subflow_tcp_sock(subflow);
1435 		if (!mptcp_subflow_active(subflow))
1436 			continue;
1437 
1438 		tout = max(tout, mptcp_timeout_from_subflow(subflow));
1439 		nr_active += !subflow->backup;
1440 		pace = subflow->avg_pacing_rate;
1441 		if (unlikely(!pace)) {
1442 			/* init pacing rate from socket */
1443 			subflow->avg_pacing_rate = READ_ONCE(ssk->sk_pacing_rate);
1444 			pace = subflow->avg_pacing_rate;
1445 			if (!pace)
1446 				continue;
1447 		}
1448 
1449 		linger_time = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32, pace);
1450 		if (linger_time < send_info[subflow->backup].linger_time) {
1451 			send_info[subflow->backup].ssk = ssk;
1452 			send_info[subflow->backup].linger_time = linger_time;
1453 		}
1454 	}
1455 	__mptcp_set_timeout(sk, tout);
1456 
1457 	/* pick the best backup if no other subflow is active */
1458 	if (!nr_active)
1459 		send_info[SSK_MODE_ACTIVE].ssk = send_info[SSK_MODE_BACKUP].ssk;
1460 
1461 	/* According to the blest algorithm, to avoid HoL blocking for the
1462 	 * faster flow, we need to:
1463 	 * - estimate the faster flow linger time
1464 	 * - use the above to estimate the amount of byte transferred
1465 	 *   by the faster flow
1466 	 * - check that the amount of queued data is greter than the above,
1467 	 *   otherwise do not use the picked, slower, subflow
1468 	 * We select the subflow with the shorter estimated time to flush
1469 	 * the queued mem, which basically ensure the above. We just need
1470 	 * to check that subflow has a non empty cwin.
1471 	 */
1472 	ssk = send_info[SSK_MODE_ACTIVE].ssk;
1473 	if (!ssk || !sk_stream_memory_free(ssk))
1474 		return NULL;
1475 
1476 	burst = min_t(int, MPTCP_SEND_BURST_SIZE, mptcp_wnd_end(msk) - msk->snd_nxt);
1477 	wmem = READ_ONCE(ssk->sk_wmem_queued);
1478 	if (!burst) {
1479 		msk->last_snd = NULL;
1480 		return ssk;
1481 	}
1482 
1483 	subflow = mptcp_subflow_ctx(ssk);
1484 	subflow->avg_pacing_rate = div_u64((u64)subflow->avg_pacing_rate * wmem +
1485 					   READ_ONCE(ssk->sk_pacing_rate) * burst,
1486 					   burst + wmem);
1487 	msk->last_snd = ssk;
1488 	msk->snd_burst = burst;
1489 	return ssk;
1490 }
1491 
1492 static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
1493 {
1494 	tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1495 	release_sock(ssk);
1496 }
1497 
1498 static void mptcp_update_post_push(struct mptcp_sock *msk,
1499 				   struct mptcp_data_frag *dfrag,
1500 				   u32 sent)
1501 {
1502 	u64 snd_nxt_new = dfrag->data_seq;
1503 
1504 	dfrag->already_sent += sent;
1505 
1506 	msk->snd_burst -= sent;
1507 
1508 	snd_nxt_new += dfrag->already_sent;
1509 
1510 	/* snd_nxt_new can be smaller than snd_nxt in case mptcp
1511 	 * is recovering after a failover. In that event, this re-sends
1512 	 * old segments.
1513 	 *
1514 	 * Thus compute snd_nxt_new candidate based on
1515 	 * the dfrag->data_seq that was sent and the data
1516 	 * that has been handed to the subflow for transmission
1517 	 * and skip update in case it was old dfrag.
1518 	 */
1519 	if (likely(after64(snd_nxt_new, msk->snd_nxt)))
1520 		msk->snd_nxt = snd_nxt_new;
1521 }
1522 
1523 void mptcp_check_and_set_pending(struct sock *sk)
1524 {
1525 	if (mptcp_send_head(sk))
1526 		mptcp_sk(sk)->push_pending |= BIT(MPTCP_PUSH_PENDING);
1527 }
1528 
1529 void __mptcp_push_pending(struct sock *sk, unsigned int flags)
1530 {
1531 	struct sock *prev_ssk = NULL, *ssk = NULL;
1532 	struct mptcp_sock *msk = mptcp_sk(sk);
1533 	struct mptcp_sendmsg_info info = {
1534 				.flags = flags,
1535 	};
1536 	bool do_check_data_fin = false;
1537 	struct mptcp_data_frag *dfrag;
1538 	int len;
1539 
1540 	while ((dfrag = mptcp_send_head(sk))) {
1541 		info.sent = dfrag->already_sent;
1542 		info.limit = dfrag->data_len;
1543 		len = dfrag->data_len - dfrag->already_sent;
1544 		while (len > 0) {
1545 			int ret = 0;
1546 
1547 			prev_ssk = ssk;
1548 			ssk = mptcp_subflow_get_send(msk);
1549 
1550 			/* First check. If the ssk has changed since
1551 			 * the last round, release prev_ssk
1552 			 */
1553 			if (ssk != prev_ssk && prev_ssk)
1554 				mptcp_push_release(prev_ssk, &info);
1555 			if (!ssk)
1556 				goto out;
1557 
1558 			/* Need to lock the new subflow only if different
1559 			 * from the previous one, otherwise we are still
1560 			 * helding the relevant lock
1561 			 */
1562 			if (ssk != prev_ssk)
1563 				lock_sock(ssk);
1564 
1565 			ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1566 			if (ret <= 0) {
1567 				if (ret == -EAGAIN)
1568 					continue;
1569 				mptcp_push_release(ssk, &info);
1570 				goto out;
1571 			}
1572 
1573 			do_check_data_fin = true;
1574 			info.sent += ret;
1575 			len -= ret;
1576 
1577 			mptcp_update_post_push(msk, dfrag, ret);
1578 		}
1579 		WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1580 	}
1581 
1582 	/* at this point we held the socket lock for the last subflow we used */
1583 	if (ssk)
1584 		mptcp_push_release(ssk, &info);
1585 
1586 out:
1587 	/* ensure the rtx timer is running */
1588 	if (!mptcp_timer_pending(sk))
1589 		mptcp_reset_timer(sk);
1590 	if (do_check_data_fin)
1591 		__mptcp_check_send_data_fin(sk);
1592 }
1593 
1594 static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, bool first)
1595 {
1596 	struct mptcp_sock *msk = mptcp_sk(sk);
1597 	struct mptcp_sendmsg_info info = {
1598 		.data_lock_held = true,
1599 	};
1600 	struct mptcp_data_frag *dfrag;
1601 	struct sock *xmit_ssk;
1602 	int len, copied = 0;
1603 
1604 	info.flags = 0;
1605 	while ((dfrag = mptcp_send_head(sk))) {
1606 		info.sent = dfrag->already_sent;
1607 		info.limit = dfrag->data_len;
1608 		len = dfrag->data_len - dfrag->already_sent;
1609 		while (len > 0) {
1610 			int ret = 0;
1611 
1612 			/* check for a different subflow usage only after
1613 			 * spooling the first chunk of data
1614 			 */
1615 			xmit_ssk = first ? ssk : mptcp_subflow_get_send(msk);
1616 			if (!xmit_ssk)
1617 				goto out;
1618 			if (xmit_ssk != ssk) {
1619 				mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk),
1620 						       MPTCP_DELEGATE_SEND);
1621 				goto out;
1622 			}
1623 
1624 			ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1625 			if (ret <= 0)
1626 				goto out;
1627 
1628 			info.sent += ret;
1629 			copied += ret;
1630 			len -= ret;
1631 			first = false;
1632 
1633 			mptcp_update_post_push(msk, dfrag, ret);
1634 		}
1635 		WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1636 	}
1637 
1638 out:
1639 	/* __mptcp_alloc_tx_skb could have released some wmem and we are
1640 	 * not going to flush it via release_sock()
1641 	 */
1642 	if (copied) {
1643 		tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1644 			 info.size_goal);
1645 		if (!mptcp_timer_pending(sk))
1646 			mptcp_reset_timer(sk);
1647 
1648 		if (msk->snd_data_fin_enable &&
1649 		    msk->snd_nxt + 1 == msk->write_seq)
1650 			mptcp_schedule_work(sk);
1651 	}
1652 }
1653 
1654 static void mptcp_set_nospace(struct sock *sk)
1655 {
1656 	/* enable autotune */
1657 	set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1658 
1659 	/* will be cleared on avail space */
1660 	set_bit(MPTCP_NOSPACE, &mptcp_sk(sk)->flags);
1661 }
1662 
1663 static int mptcp_disconnect(struct sock *sk, int flags);
1664 
1665 static int mptcp_sendmsg_fastopen(struct sock *sk, struct sock *ssk, struct msghdr *msg,
1666 				  size_t len, int *copied_syn)
1667 {
1668 	unsigned int saved_flags = msg->msg_flags;
1669 	struct mptcp_sock *msk = mptcp_sk(sk);
1670 	int ret;
1671 
1672 	lock_sock(ssk);
1673 	msg->msg_flags |= MSG_DONTWAIT;
1674 	msk->connect_flags = O_NONBLOCK;
1675 	msk->fastopening = 1;
1676 	ret = tcp_sendmsg_fastopen(ssk, msg, copied_syn, len, NULL);
1677 	msk->fastopening = 0;
1678 	msg->msg_flags = saved_flags;
1679 	release_sock(ssk);
1680 
1681 	/* do the blocking bits of inet_stream_connect outside the ssk socket lock */
1682 	if (ret == -EINPROGRESS && !(msg->msg_flags & MSG_DONTWAIT)) {
1683 		ret = __inet_stream_connect(sk->sk_socket, msg->msg_name,
1684 					    msg->msg_namelen, msg->msg_flags, 1);
1685 
1686 		/* Keep the same behaviour of plain TCP: zero the copied bytes in
1687 		 * case of any error, except timeout or signal
1688 		 */
1689 		if (ret && ret != -EINPROGRESS && ret != -ERESTARTSYS && ret != -EINTR)
1690 			*copied_syn = 0;
1691 	} else if (ret && ret != -EINPROGRESS) {
1692 		mptcp_disconnect(sk, 0);
1693 	}
1694 
1695 	return ret;
1696 }
1697 
1698 static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1699 {
1700 	struct mptcp_sock *msk = mptcp_sk(sk);
1701 	struct page_frag *pfrag;
1702 	struct socket *ssock;
1703 	size_t copied = 0;
1704 	int ret = 0;
1705 	long timeo;
1706 
1707 	/* silently ignore everything else */
1708 	msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_FASTOPEN;
1709 
1710 	lock_sock(sk);
1711 
1712 	ssock = __mptcp_nmpc_socket(msk);
1713 	if (unlikely(ssock && (inet_sk(ssock->sk)->defer_connect ||
1714 			       msg->msg_flags & MSG_FASTOPEN))) {
1715 		int copied_syn = 0;
1716 
1717 		ret = mptcp_sendmsg_fastopen(sk, ssock->sk, msg, len, &copied_syn);
1718 		copied += copied_syn;
1719 		if (ret == -EINPROGRESS && copied_syn > 0)
1720 			goto out;
1721 		else if (ret)
1722 			goto do_error;
1723 	}
1724 
1725 	timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1726 
1727 	if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1728 		ret = sk_stream_wait_connect(sk, &timeo);
1729 		if (ret)
1730 			goto do_error;
1731 	}
1732 
1733 	ret = -EPIPE;
1734 	if (unlikely(sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)))
1735 		goto do_error;
1736 
1737 	pfrag = sk_page_frag(sk);
1738 
1739 	while (msg_data_left(msg)) {
1740 		int total_ts, frag_truesize = 0;
1741 		struct mptcp_data_frag *dfrag;
1742 		bool dfrag_collapsed;
1743 		size_t psize, offset;
1744 
1745 		/* reuse tail pfrag, if possible, or carve a new one from the
1746 		 * page allocator
1747 		 */
1748 		dfrag = mptcp_pending_tail(sk);
1749 		dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1750 		if (!dfrag_collapsed) {
1751 			if (!sk_stream_memory_free(sk))
1752 				goto wait_for_memory;
1753 
1754 			if (!mptcp_page_frag_refill(sk, pfrag))
1755 				goto wait_for_memory;
1756 
1757 			dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1758 			frag_truesize = dfrag->overhead;
1759 		}
1760 
1761 		/* we do not bound vs wspace, to allow a single packet.
1762 		 * memory accounting will prevent execessive memory usage
1763 		 * anyway
1764 		 */
1765 		offset = dfrag->offset + dfrag->data_len;
1766 		psize = pfrag->size - offset;
1767 		psize = min_t(size_t, psize, msg_data_left(msg));
1768 		total_ts = psize + frag_truesize;
1769 
1770 		if (!sk_wmem_schedule(sk, total_ts))
1771 			goto wait_for_memory;
1772 
1773 		if (copy_page_from_iter(dfrag->page, offset, psize,
1774 					&msg->msg_iter) != psize) {
1775 			ret = -EFAULT;
1776 			goto do_error;
1777 		}
1778 
1779 		/* data successfully copied into the write queue */
1780 		sk->sk_forward_alloc -= total_ts;
1781 		copied += psize;
1782 		dfrag->data_len += psize;
1783 		frag_truesize += psize;
1784 		pfrag->offset += frag_truesize;
1785 		WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1786 
1787 		/* charge data on mptcp pending queue to the msk socket
1788 		 * Note: we charge such data both to sk and ssk
1789 		 */
1790 		sk_wmem_queued_add(sk, frag_truesize);
1791 		if (!dfrag_collapsed) {
1792 			get_page(dfrag->page);
1793 			list_add_tail(&dfrag->list, &msk->rtx_queue);
1794 			if (!msk->first_pending)
1795 				WRITE_ONCE(msk->first_pending, dfrag);
1796 		}
1797 		pr_debug("msk=%p dfrag at seq=%llu len=%u sent=%u new=%d", msk,
1798 			 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1799 			 !dfrag_collapsed);
1800 
1801 		continue;
1802 
1803 wait_for_memory:
1804 		mptcp_set_nospace(sk);
1805 		__mptcp_push_pending(sk, msg->msg_flags);
1806 		ret = sk_stream_wait_memory(sk, &timeo);
1807 		if (ret)
1808 			goto do_error;
1809 	}
1810 
1811 	if (copied)
1812 		__mptcp_push_pending(sk, msg->msg_flags);
1813 
1814 out:
1815 	release_sock(sk);
1816 	return copied;
1817 
1818 do_error:
1819 	if (copied)
1820 		goto out;
1821 
1822 	copied = sk_stream_error(sk, msg->msg_flags, ret);
1823 	goto out;
1824 }
1825 
1826 static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1827 				struct msghdr *msg,
1828 				size_t len, int flags,
1829 				struct scm_timestamping_internal *tss,
1830 				int *cmsg_flags)
1831 {
1832 	struct sk_buff *skb, *tmp;
1833 	int copied = 0;
1834 
1835 	skb_queue_walk_safe(&msk->receive_queue, skb, tmp) {
1836 		u32 offset = MPTCP_SKB_CB(skb)->offset;
1837 		u32 data_len = skb->len - offset;
1838 		u32 count = min_t(size_t, len - copied, data_len);
1839 		int err;
1840 
1841 		if (!(flags & MSG_TRUNC)) {
1842 			err = skb_copy_datagram_msg(skb, offset, msg, count);
1843 			if (unlikely(err < 0)) {
1844 				if (!copied)
1845 					return err;
1846 				break;
1847 			}
1848 		}
1849 
1850 		if (MPTCP_SKB_CB(skb)->has_rxtstamp) {
1851 			tcp_update_recv_tstamps(skb, tss);
1852 			*cmsg_flags |= MPTCP_CMSG_TS;
1853 		}
1854 
1855 		copied += count;
1856 
1857 		if (count < data_len) {
1858 			if (!(flags & MSG_PEEK)) {
1859 				MPTCP_SKB_CB(skb)->offset += count;
1860 				MPTCP_SKB_CB(skb)->map_seq += count;
1861 			}
1862 			break;
1863 		}
1864 
1865 		if (!(flags & MSG_PEEK)) {
1866 			/* we will bulk release the skb memory later */
1867 			skb->destructor = NULL;
1868 			WRITE_ONCE(msk->rmem_released, msk->rmem_released + skb->truesize);
1869 			__skb_unlink(skb, &msk->receive_queue);
1870 			__kfree_skb(skb);
1871 		}
1872 
1873 		if (copied >= len)
1874 			break;
1875 	}
1876 
1877 	return copied;
1878 }
1879 
1880 /* receive buffer autotuning.  See tcp_rcv_space_adjust for more information.
1881  *
1882  * Only difference: Use highest rtt estimate of the subflows in use.
1883  */
1884 static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1885 {
1886 	struct mptcp_subflow_context *subflow;
1887 	struct sock *sk = (struct sock *)msk;
1888 	u32 time, advmss = 1;
1889 	u64 rtt_us, mstamp;
1890 
1891 	msk_owned_by_me(msk);
1892 
1893 	if (copied <= 0)
1894 		return;
1895 
1896 	msk->rcvq_space.copied += copied;
1897 
1898 	mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1899 	time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1900 
1901 	rtt_us = msk->rcvq_space.rtt_us;
1902 	if (rtt_us && time < (rtt_us >> 3))
1903 		return;
1904 
1905 	rtt_us = 0;
1906 	mptcp_for_each_subflow(msk, subflow) {
1907 		const struct tcp_sock *tp;
1908 		u64 sf_rtt_us;
1909 		u32 sf_advmss;
1910 
1911 		tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1912 
1913 		sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1914 		sf_advmss = READ_ONCE(tp->advmss);
1915 
1916 		rtt_us = max(sf_rtt_us, rtt_us);
1917 		advmss = max(sf_advmss, advmss);
1918 	}
1919 
1920 	msk->rcvq_space.rtt_us = rtt_us;
1921 	if (time < (rtt_us >> 3) || rtt_us == 0)
1922 		return;
1923 
1924 	if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1925 		goto new_measure;
1926 
1927 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf) &&
1928 	    !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1929 		int rcvmem, rcvbuf;
1930 		u64 rcvwin, grow;
1931 
1932 		rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1933 
1934 		grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
1935 
1936 		do_div(grow, msk->rcvq_space.space);
1937 		rcvwin += (grow << 1);
1938 
1939 		rcvmem = SKB_TRUESIZE(advmss + MAX_TCP_HEADER);
1940 		while (tcp_win_from_space(sk, rcvmem) < advmss)
1941 			rcvmem += 128;
1942 
1943 		do_div(rcvwin, advmss);
1944 		rcvbuf = min_t(u64, rcvwin * rcvmem,
1945 			       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));
1946 
1947 		if (rcvbuf > sk->sk_rcvbuf) {
1948 			u32 window_clamp;
1949 
1950 			window_clamp = tcp_win_from_space(sk, rcvbuf);
1951 			WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
1952 
1953 			/* Make subflows follow along.  If we do not do this, we
1954 			 * get drops at subflow level if skbs can't be moved to
1955 			 * the mptcp rx queue fast enough (announced rcv_win can
1956 			 * exceed ssk->sk_rcvbuf).
1957 			 */
1958 			mptcp_for_each_subflow(msk, subflow) {
1959 				struct sock *ssk;
1960 				bool slow;
1961 
1962 				ssk = mptcp_subflow_tcp_sock(subflow);
1963 				slow = lock_sock_fast(ssk);
1964 				WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
1965 				tcp_sk(ssk)->window_clamp = window_clamp;
1966 				tcp_cleanup_rbuf(ssk, 1);
1967 				unlock_sock_fast(ssk, slow);
1968 			}
1969 		}
1970 	}
1971 
1972 	msk->rcvq_space.space = msk->rcvq_space.copied;
1973 new_measure:
1974 	msk->rcvq_space.copied = 0;
1975 	msk->rcvq_space.time = mstamp;
1976 }
1977 
1978 static void __mptcp_update_rmem(struct sock *sk)
1979 {
1980 	struct mptcp_sock *msk = mptcp_sk(sk);
1981 
1982 	if (!msk->rmem_released)
1983 		return;
1984 
1985 	atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
1986 	mptcp_rmem_uncharge(sk, msk->rmem_released);
1987 	WRITE_ONCE(msk->rmem_released, 0);
1988 }
1989 
1990 static void __mptcp_splice_receive_queue(struct sock *sk)
1991 {
1992 	struct mptcp_sock *msk = mptcp_sk(sk);
1993 
1994 	skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
1995 }
1996 
1997 static bool __mptcp_move_skbs(struct mptcp_sock *msk)
1998 {
1999 	struct sock *sk = (struct sock *)msk;
2000 	unsigned int moved = 0;
2001 	bool ret, done;
2002 
2003 	do {
2004 		struct sock *ssk = mptcp_subflow_recv_lookup(msk);
2005 		bool slowpath;
2006 
2007 		/* we can have data pending in the subflows only if the msk
2008 		 * receive buffer was full at subflow_data_ready() time,
2009 		 * that is an unlikely slow path.
2010 		 */
2011 		if (likely(!ssk))
2012 			break;
2013 
2014 		slowpath = lock_sock_fast(ssk);
2015 		mptcp_data_lock(sk);
2016 		__mptcp_update_rmem(sk);
2017 		done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
2018 		mptcp_data_unlock(sk);
2019 
2020 		if (unlikely(ssk->sk_err))
2021 			__mptcp_error_report(sk);
2022 		unlock_sock_fast(ssk, slowpath);
2023 	} while (!done);
2024 
2025 	/* acquire the data lock only if some input data is pending */
2026 	ret = moved > 0;
2027 	if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
2028 	    !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
2029 		mptcp_data_lock(sk);
2030 		__mptcp_update_rmem(sk);
2031 		ret |= __mptcp_ofo_queue(msk);
2032 		__mptcp_splice_receive_queue(sk);
2033 		mptcp_data_unlock(sk);
2034 	}
2035 	if (ret)
2036 		mptcp_check_data_fin((struct sock *)msk);
2037 	return !skb_queue_empty(&msk->receive_queue);
2038 }
2039 
2040 static unsigned int mptcp_inq_hint(const struct sock *sk)
2041 {
2042 	const struct mptcp_sock *msk = mptcp_sk(sk);
2043 	const struct sk_buff *skb;
2044 
2045 	skb = skb_peek(&msk->receive_queue);
2046 	if (skb) {
2047 		u64 hint_val = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
2048 
2049 		if (hint_val >= INT_MAX)
2050 			return INT_MAX;
2051 
2052 		return (unsigned int)hint_val;
2053 	}
2054 
2055 	if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
2056 		return 1;
2057 
2058 	return 0;
2059 }
2060 
2061 static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2062 			 int flags, int *addr_len)
2063 {
2064 	struct mptcp_sock *msk = mptcp_sk(sk);
2065 	struct scm_timestamping_internal tss;
2066 	int copied = 0, cmsg_flags = 0;
2067 	int target;
2068 	long timeo;
2069 
2070 	/* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
2071 	if (unlikely(flags & MSG_ERRQUEUE))
2072 		return inet_recv_error(sk, msg, len, addr_len);
2073 
2074 	lock_sock(sk);
2075 	if (unlikely(sk->sk_state == TCP_LISTEN)) {
2076 		copied = -ENOTCONN;
2077 		goto out_err;
2078 	}
2079 
2080 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2081 
2082 	len = min_t(size_t, len, INT_MAX);
2083 	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2084 
2085 	if (unlikely(msk->recvmsg_inq))
2086 		cmsg_flags = MPTCP_CMSG_INQ;
2087 
2088 	while (copied < len) {
2089 		int bytes_read;
2090 
2091 		bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied, flags, &tss, &cmsg_flags);
2092 		if (unlikely(bytes_read < 0)) {
2093 			if (!copied)
2094 				copied = bytes_read;
2095 			goto out_err;
2096 		}
2097 
2098 		copied += bytes_read;
2099 
2100 		/* be sure to advertise window change */
2101 		mptcp_cleanup_rbuf(msk);
2102 
2103 		if (skb_queue_empty(&msk->receive_queue) && __mptcp_move_skbs(msk))
2104 			continue;
2105 
2106 		/* only the master socket status is relevant here. The exit
2107 		 * conditions mirror closely tcp_recvmsg()
2108 		 */
2109 		if (copied >= target)
2110 			break;
2111 
2112 		if (copied) {
2113 			if (sk->sk_err ||
2114 			    sk->sk_state == TCP_CLOSE ||
2115 			    (sk->sk_shutdown & RCV_SHUTDOWN) ||
2116 			    !timeo ||
2117 			    signal_pending(current))
2118 				break;
2119 		} else {
2120 			if (sk->sk_err) {
2121 				copied = sock_error(sk);
2122 				break;
2123 			}
2124 
2125 			if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2126 				mptcp_check_for_eof(msk);
2127 
2128 			if (sk->sk_shutdown & RCV_SHUTDOWN) {
2129 				/* race breaker: the shutdown could be after the
2130 				 * previous receive queue check
2131 				 */
2132 				if (__mptcp_move_skbs(msk))
2133 					continue;
2134 				break;
2135 			}
2136 
2137 			if (sk->sk_state == TCP_CLOSE) {
2138 				copied = -ENOTCONN;
2139 				break;
2140 			}
2141 
2142 			if (!timeo) {
2143 				copied = -EAGAIN;
2144 				break;
2145 			}
2146 
2147 			if (signal_pending(current)) {
2148 				copied = sock_intr_errno(timeo);
2149 				break;
2150 			}
2151 		}
2152 
2153 		pr_debug("block timeout %ld", timeo);
2154 		sk_wait_data(sk, &timeo, NULL);
2155 	}
2156 
2157 out_err:
2158 	if (cmsg_flags && copied >= 0) {
2159 		if (cmsg_flags & MPTCP_CMSG_TS)
2160 			tcp_recv_timestamp(msg, sk, &tss);
2161 
2162 		if (cmsg_flags & MPTCP_CMSG_INQ) {
2163 			unsigned int inq = mptcp_inq_hint(sk);
2164 
2165 			put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2166 		}
2167 	}
2168 
2169 	pr_debug("msk=%p rx queue empty=%d:%d copied=%d",
2170 		 msk, skb_queue_empty_lockless(&sk->sk_receive_queue),
2171 		 skb_queue_empty(&msk->receive_queue), copied);
2172 	if (!(flags & MSG_PEEK))
2173 		mptcp_rcv_space_adjust(msk, copied);
2174 
2175 	release_sock(sk);
2176 	return copied;
2177 }
2178 
2179 static void mptcp_retransmit_timer(struct timer_list *t)
2180 {
2181 	struct inet_connection_sock *icsk = from_timer(icsk, t,
2182 						       icsk_retransmit_timer);
2183 	struct sock *sk = &icsk->icsk_inet.sk;
2184 	struct mptcp_sock *msk = mptcp_sk(sk);
2185 
2186 	bh_lock_sock(sk);
2187 	if (!sock_owned_by_user(sk)) {
2188 		/* we need a process context to retransmit */
2189 		if (!test_and_set_bit(MPTCP_WORK_RTX, &msk->flags))
2190 			mptcp_schedule_work(sk);
2191 	} else {
2192 		/* delegate our work to tcp_release_cb() */
2193 		__set_bit(MPTCP_RETRANSMIT, &msk->cb_flags);
2194 	}
2195 	bh_unlock_sock(sk);
2196 	sock_put(sk);
2197 }
2198 
2199 static void mptcp_timeout_timer(struct timer_list *t)
2200 {
2201 	struct sock *sk = from_timer(sk, t, sk_timer);
2202 
2203 	mptcp_schedule_work(sk);
2204 	sock_put(sk);
2205 }
2206 
2207 /* Find an idle subflow.  Return NULL if there is unacked data at tcp
2208  * level.
2209  *
2210  * A backup subflow is returned only if that is the only kind available.
2211  */
2212 static struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)
2213 {
2214 	struct sock *backup = NULL, *pick = NULL;
2215 	struct mptcp_subflow_context *subflow;
2216 	int min_stale_count = INT_MAX;
2217 
2218 	msk_owned_by_me(msk);
2219 
2220 	if (__mptcp_check_fallback(msk))
2221 		return NULL;
2222 
2223 	mptcp_for_each_subflow(msk, subflow) {
2224 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2225 
2226 		if (!__mptcp_subflow_active(subflow))
2227 			continue;
2228 
2229 		/* still data outstanding at TCP level? skip this */
2230 		if (!tcp_rtx_and_write_queues_empty(ssk)) {
2231 			mptcp_pm_subflow_chk_stale(msk, ssk);
2232 			min_stale_count = min_t(int, min_stale_count, subflow->stale_count);
2233 			continue;
2234 		}
2235 
2236 		if (subflow->backup) {
2237 			if (!backup)
2238 				backup = ssk;
2239 			continue;
2240 		}
2241 
2242 		if (!pick)
2243 			pick = ssk;
2244 	}
2245 
2246 	if (pick)
2247 		return pick;
2248 
2249 	/* use backup only if there are no progresses anywhere */
2250 	return min_stale_count > 1 ? backup : NULL;
2251 }
2252 
2253 static void mptcp_dispose_initial_subflow(struct mptcp_sock *msk)
2254 {
2255 	if (msk->subflow) {
2256 		iput(SOCK_INODE(msk->subflow));
2257 		msk->subflow = NULL;
2258 	}
2259 }
2260 
2261 bool __mptcp_retransmit_pending_data(struct sock *sk)
2262 {
2263 	struct mptcp_data_frag *cur, *rtx_head;
2264 	struct mptcp_sock *msk = mptcp_sk(sk);
2265 
2266 	if (__mptcp_check_fallback(msk))
2267 		return false;
2268 
2269 	if (tcp_rtx_and_write_queues_empty(sk))
2270 		return false;
2271 
2272 	/* the closing socket has some data untransmitted and/or unacked:
2273 	 * some data in the mptcp rtx queue has not really xmitted yet.
2274 	 * keep it simple and re-inject the whole mptcp level rtx queue
2275 	 */
2276 	mptcp_data_lock(sk);
2277 	__mptcp_clean_una_wakeup(sk);
2278 	rtx_head = mptcp_rtx_head(sk);
2279 	if (!rtx_head) {
2280 		mptcp_data_unlock(sk);
2281 		return false;
2282 	}
2283 
2284 	msk->recovery_snd_nxt = msk->snd_nxt;
2285 	msk->recovery = true;
2286 	mptcp_data_unlock(sk);
2287 
2288 	msk->first_pending = rtx_head;
2289 	msk->snd_burst = 0;
2290 
2291 	/* be sure to clear the "sent status" on all re-injected fragments */
2292 	list_for_each_entry(cur, &msk->rtx_queue, list) {
2293 		if (!cur->already_sent)
2294 			break;
2295 		cur->already_sent = 0;
2296 	}
2297 
2298 	return true;
2299 }
2300 
2301 /* flags for __mptcp_close_ssk() */
2302 #define MPTCP_CF_PUSH		BIT(1)
2303 #define MPTCP_CF_FASTCLOSE	BIT(2)
2304 
2305 /* subflow sockets can be either outgoing (connect) or incoming
2306  * (accept).
2307  *
2308  * Outgoing subflows use in-kernel sockets.
2309  * Incoming subflows do not have their own 'struct socket' allocated,
2310  * so we need to use tcp_close() after detaching them from the mptcp
2311  * parent socket.
2312  */
2313 static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2314 			      struct mptcp_subflow_context *subflow,
2315 			      unsigned int flags)
2316 {
2317 	struct mptcp_sock *msk = mptcp_sk(sk);
2318 	bool need_push, dispose_it;
2319 
2320 	dispose_it = !msk->subflow || ssk != msk->subflow->sk;
2321 	if (dispose_it)
2322 		list_del(&subflow->node);
2323 
2324 	lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
2325 
2326 	if (flags & MPTCP_CF_FASTCLOSE) {
2327 		/* be sure to force the tcp_disconnect() path,
2328 		 * to generate the egress reset
2329 		 */
2330 		ssk->sk_lingertime = 0;
2331 		sock_set_flag(ssk, SOCK_LINGER);
2332 		subflow->send_fastclose = 1;
2333 	}
2334 
2335 	need_push = (flags & MPTCP_CF_PUSH) && __mptcp_retransmit_pending_data(sk);
2336 	if (!dispose_it) {
2337 		tcp_disconnect(ssk, 0);
2338 		msk->subflow->state = SS_UNCONNECTED;
2339 		mptcp_subflow_ctx_reset(subflow);
2340 		release_sock(ssk);
2341 
2342 		goto out;
2343 	}
2344 
2345 	subflow->disposable = 1;
2346 
2347 	/* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2348 	 * the ssk has been already destroyed, we just need to release the
2349 	 * reference owned by msk;
2350 	 */
2351 	if (!inet_csk(ssk)->icsk_ulp_ops) {
2352 		WARN_ON_ONCE(!sock_flag(ssk, SOCK_DEAD));
2353 		kfree_rcu(subflow, rcu);
2354 	} else if (msk->in_accept_queue && msk->first == ssk) {
2355 		/* if the first subflow moved to a close state, e.g. due to
2356 		 * incoming reset and we reach here before inet_child_forget()
2357 		 * the TCP stack could later try to close it via
2358 		 * inet_csk_listen_stop(), or deliver it to the user space via
2359 		 * accept().
2360 		 * We can't delete the subflow - or risk a double free - nor let
2361 		 * the msk survive - or will be leaked in the non accept scenario:
2362 		 * fallback and let TCP cope with the subflow cleanup.
2363 		 */
2364 		WARN_ON_ONCE(sock_flag(ssk, SOCK_DEAD));
2365 		mptcp_subflow_drop_ctx(ssk);
2366 	} else {
2367 		/* otherwise tcp will dispose of the ssk and subflow ctx */
2368 		if (ssk->sk_state == TCP_LISTEN)
2369 			mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CLOSED);
2370 
2371 		__tcp_close(ssk, 0);
2372 
2373 		/* close acquired an extra ref */
2374 		__sock_put(ssk);
2375 	}
2376 	release_sock(ssk);
2377 
2378 	sock_put(ssk);
2379 
2380 	if (ssk == msk->first)
2381 		msk->first = NULL;
2382 
2383 out:
2384 	if (ssk == msk->last_snd)
2385 		msk->last_snd = NULL;
2386 
2387 	if (need_push)
2388 		__mptcp_push_pending(sk, 0);
2389 }
2390 
2391 void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2392 		     struct mptcp_subflow_context *subflow)
2393 {
2394 	if (sk->sk_state == TCP_ESTABLISHED)
2395 		mptcp_event(MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
2396 
2397 	/* subflow aborted before reaching the fully_established status
2398 	 * attempt the creation of the next subflow
2399 	 */
2400 	mptcp_pm_subflow_check_next(mptcp_sk(sk), ssk, subflow);
2401 
2402 	__mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_PUSH);
2403 }
2404 
2405 static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2406 {
2407 	return 0;
2408 }
2409 
2410 static void __mptcp_close_subflow(struct sock *sk)
2411 {
2412 	struct mptcp_subflow_context *subflow, *tmp;
2413 	struct mptcp_sock *msk = mptcp_sk(sk);
2414 
2415 	might_sleep();
2416 
2417 	mptcp_for_each_subflow_safe(msk, subflow, tmp) {
2418 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2419 
2420 		if (inet_sk_state_load(ssk) != TCP_CLOSE)
2421 			continue;
2422 
2423 		/* 'subflow_data_ready' will re-sched once rx queue is empty */
2424 		if (!skb_queue_empty_lockless(&ssk->sk_receive_queue))
2425 			continue;
2426 
2427 		mptcp_close_ssk(sk, ssk, subflow);
2428 	}
2429 
2430 	/* if the MPC subflow has been closed before the msk is accepted,
2431 	 * msk will never be accept-ed, close it now
2432 	 */
2433 	if (!msk->first && msk->in_accept_queue) {
2434 		sock_set_flag(sk, SOCK_DEAD);
2435 		inet_sk_state_store(sk, TCP_CLOSE);
2436 	}
2437 }
2438 
2439 static bool mptcp_check_close_timeout(const struct sock *sk)
2440 {
2441 	s32 delta = tcp_jiffies32 - inet_csk(sk)->icsk_mtup.probe_timestamp;
2442 	struct mptcp_subflow_context *subflow;
2443 
2444 	if (delta >= TCP_TIMEWAIT_LEN)
2445 		return true;
2446 
2447 	/* if all subflows are in closed status don't bother with additional
2448 	 * timeout
2449 	 */
2450 	mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
2451 		if (inet_sk_state_load(mptcp_subflow_tcp_sock(subflow)) !=
2452 		    TCP_CLOSE)
2453 			return false;
2454 	}
2455 	return true;
2456 }
2457 
2458 static void mptcp_check_fastclose(struct mptcp_sock *msk)
2459 {
2460 	struct mptcp_subflow_context *subflow, *tmp;
2461 	struct sock *sk = (struct sock *)msk;
2462 
2463 	if (likely(!READ_ONCE(msk->rcv_fastclose)))
2464 		return;
2465 
2466 	mptcp_token_destroy(msk);
2467 
2468 	mptcp_for_each_subflow_safe(msk, subflow, tmp) {
2469 		struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2470 		bool slow;
2471 
2472 		slow = lock_sock_fast(tcp_sk);
2473 		if (tcp_sk->sk_state != TCP_CLOSE) {
2474 			tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2475 			tcp_set_state(tcp_sk, TCP_CLOSE);
2476 		}
2477 		unlock_sock_fast(tcp_sk, slow);
2478 	}
2479 
2480 	/* Mirror the tcp_reset() error propagation */
2481 	switch (sk->sk_state) {
2482 	case TCP_SYN_SENT:
2483 		sk->sk_err = ECONNREFUSED;
2484 		break;
2485 	case TCP_CLOSE_WAIT:
2486 		sk->sk_err = EPIPE;
2487 		break;
2488 	case TCP_CLOSE:
2489 		return;
2490 	default:
2491 		sk->sk_err = ECONNRESET;
2492 	}
2493 
2494 	inet_sk_state_store(sk, TCP_CLOSE);
2495 	sk->sk_shutdown = SHUTDOWN_MASK;
2496 	smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
2497 	set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2498 
2499 	/* the calling mptcp_worker will properly destroy the socket */
2500 	if (sock_flag(sk, SOCK_DEAD))
2501 		return;
2502 
2503 	sk->sk_state_change(sk);
2504 	sk_error_report(sk);
2505 }
2506 
2507 static void __mptcp_retrans(struct sock *sk)
2508 {
2509 	struct mptcp_sock *msk = mptcp_sk(sk);
2510 	struct mptcp_sendmsg_info info = {};
2511 	struct mptcp_data_frag *dfrag;
2512 	size_t copied = 0;
2513 	struct sock *ssk;
2514 	int ret;
2515 
2516 	mptcp_clean_una_wakeup(sk);
2517 
2518 	/* first check ssk: need to kick "stale" logic */
2519 	ssk = mptcp_subflow_get_retrans(msk);
2520 	dfrag = mptcp_rtx_head(sk);
2521 	if (!dfrag) {
2522 		if (mptcp_data_fin_enabled(msk)) {
2523 			struct inet_connection_sock *icsk = inet_csk(sk);
2524 
2525 			icsk->icsk_retransmits++;
2526 			mptcp_set_datafin_timeout(sk);
2527 			mptcp_send_ack(msk);
2528 
2529 			goto reset_timer;
2530 		}
2531 
2532 		if (!mptcp_send_head(sk))
2533 			return;
2534 
2535 		goto reset_timer;
2536 	}
2537 
2538 	if (!ssk)
2539 		goto reset_timer;
2540 
2541 	lock_sock(ssk);
2542 
2543 	/* limit retransmission to the bytes already sent on some subflows */
2544 	info.sent = 0;
2545 	info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len : dfrag->already_sent;
2546 	while (info.sent < info.limit) {
2547 		ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
2548 		if (ret <= 0)
2549 			break;
2550 
2551 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
2552 		copied += ret;
2553 		info.sent += ret;
2554 	}
2555 	if (copied) {
2556 		dfrag->already_sent = max(dfrag->already_sent, info.sent);
2557 		tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2558 			 info.size_goal);
2559 		WRITE_ONCE(msk->allow_infinite_fallback, false);
2560 	}
2561 
2562 	release_sock(ssk);
2563 
2564 reset_timer:
2565 	mptcp_check_and_set_pending(sk);
2566 
2567 	if (!mptcp_timer_pending(sk))
2568 		mptcp_reset_timer(sk);
2569 }
2570 
2571 /* schedule the timeout timer for the relevant event: either close timeout
2572  * or mp_fail timeout. The close timeout takes precedence on the mp_fail one
2573  */
2574 void mptcp_reset_timeout(struct mptcp_sock *msk, unsigned long fail_tout)
2575 {
2576 	struct sock *sk = (struct sock *)msk;
2577 	unsigned long timeout, close_timeout;
2578 
2579 	if (!fail_tout && !sock_flag(sk, SOCK_DEAD))
2580 		return;
2581 
2582 	close_timeout = inet_csk(sk)->icsk_mtup.probe_timestamp - tcp_jiffies32 + jiffies + TCP_TIMEWAIT_LEN;
2583 
2584 	/* the close timeout takes precedence on the fail one, and here at least one of
2585 	 * them is active
2586 	 */
2587 	timeout = sock_flag(sk, SOCK_DEAD) ? close_timeout : fail_tout;
2588 
2589 	sk_reset_timer(sk, &sk->sk_timer, timeout);
2590 }
2591 
2592 static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
2593 {
2594 	struct sock *ssk = msk->first;
2595 	bool slow;
2596 
2597 	if (!ssk)
2598 		return;
2599 
2600 	pr_debug("MP_FAIL doesn't respond, reset the subflow");
2601 
2602 	slow = lock_sock_fast(ssk);
2603 	mptcp_subflow_reset(ssk);
2604 	WRITE_ONCE(mptcp_subflow_ctx(ssk)->fail_tout, 0);
2605 	unlock_sock_fast(ssk, slow);
2606 
2607 	mptcp_reset_timeout(msk, 0);
2608 }
2609 
2610 static void mptcp_do_fastclose(struct sock *sk)
2611 {
2612 	struct mptcp_subflow_context *subflow, *tmp;
2613 	struct mptcp_sock *msk = mptcp_sk(sk);
2614 
2615 	mptcp_for_each_subflow_safe(msk, subflow, tmp)
2616 		__mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow),
2617 				  subflow, MPTCP_CF_FASTCLOSE);
2618 }
2619 
2620 static void mptcp_worker(struct work_struct *work)
2621 {
2622 	struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
2623 	struct sock *sk = (struct sock *)msk;
2624 	unsigned long fail_tout;
2625 	int state;
2626 
2627 	lock_sock(sk);
2628 	state = sk->sk_state;
2629 	if (unlikely(state == TCP_CLOSE))
2630 		goto unlock;
2631 
2632 	mptcp_check_data_fin_ack(sk);
2633 
2634 	mptcp_check_fastclose(msk);
2635 
2636 	mptcp_pm_nl_work(msk);
2637 
2638 	if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2639 		mptcp_check_for_eof(msk);
2640 
2641 	__mptcp_check_send_data_fin(sk);
2642 	mptcp_check_data_fin(sk);
2643 
2644 	if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2645 		__mptcp_close_subflow(sk);
2646 
2647 	/* There is no point in keeping around an orphaned sk timedout or
2648 	 * closed, but we need the msk around to reply to incoming DATA_FIN,
2649 	 * even if it is orphaned and in FIN_WAIT2 state
2650 	 */
2651 	if (sock_flag(sk, SOCK_DEAD)) {
2652 		if (mptcp_check_close_timeout(sk)) {
2653 			inet_sk_state_store(sk, TCP_CLOSE);
2654 			mptcp_do_fastclose(sk);
2655 		}
2656 		if (sk->sk_state == TCP_CLOSE) {
2657 			__mptcp_destroy_sock(sk);
2658 			goto unlock;
2659 		}
2660 	}
2661 
2662 	if (test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2663 		__mptcp_retrans(sk);
2664 
2665 	fail_tout = msk->first ? READ_ONCE(mptcp_subflow_ctx(msk->first)->fail_tout) : 0;
2666 	if (fail_tout && time_after(jiffies, fail_tout))
2667 		mptcp_mp_fail_no_response(msk);
2668 
2669 unlock:
2670 	release_sock(sk);
2671 	sock_put(sk);
2672 }
2673 
2674 static int __mptcp_init_sock(struct sock *sk)
2675 {
2676 	struct mptcp_sock *msk = mptcp_sk(sk);
2677 
2678 	INIT_LIST_HEAD(&msk->conn_list);
2679 	INIT_LIST_HEAD(&msk->join_list);
2680 	INIT_LIST_HEAD(&msk->rtx_queue);
2681 	INIT_WORK(&msk->work, mptcp_worker);
2682 	__skb_queue_head_init(&msk->receive_queue);
2683 	msk->out_of_order_queue = RB_ROOT;
2684 	msk->first_pending = NULL;
2685 	msk->rmem_fwd_alloc = 0;
2686 	WRITE_ONCE(msk->rmem_released, 0);
2687 	msk->timer_ival = TCP_RTO_MIN;
2688 
2689 	msk->first = NULL;
2690 	inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
2691 	WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
2692 	WRITE_ONCE(msk->allow_infinite_fallback, true);
2693 	msk->recovery = false;
2694 
2695 	mptcp_pm_data_init(msk);
2696 
2697 	/* re-use the csk retrans timer for MPTCP-level retrans */
2698 	timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
2699 	timer_setup(&sk->sk_timer, mptcp_timeout_timer, 0);
2700 
2701 	return 0;
2702 }
2703 
2704 static void mptcp_ca_reset(struct sock *sk)
2705 {
2706 	struct inet_connection_sock *icsk = inet_csk(sk);
2707 
2708 	tcp_assign_congestion_control(sk);
2709 	strcpy(mptcp_sk(sk)->ca_name, icsk->icsk_ca_ops->name);
2710 
2711 	/* no need to keep a reference to the ops, the name will suffice */
2712 	tcp_cleanup_congestion_control(sk);
2713 	icsk->icsk_ca_ops = NULL;
2714 }
2715 
2716 static int mptcp_init_sock(struct sock *sk)
2717 {
2718 	struct net *net = sock_net(sk);
2719 	int ret;
2720 
2721 	ret = __mptcp_init_sock(sk);
2722 	if (ret)
2723 		return ret;
2724 
2725 	if (!mptcp_is_enabled(net))
2726 		return -ENOPROTOOPT;
2727 
2728 	if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2729 		return -ENOMEM;
2730 
2731 	ret = __mptcp_socket_create(mptcp_sk(sk));
2732 	if (ret)
2733 		return ret;
2734 
2735 	set_bit(SOCK_CUSTOM_SOCKOPT, &sk->sk_socket->flags);
2736 
2737 	/* fetch the ca name; do it outside __mptcp_init_sock(), so that clone will
2738 	 * propagate the correct value
2739 	 */
2740 	mptcp_ca_reset(sk);
2741 
2742 	sk_sockets_allocated_inc(sk);
2743 	sk->sk_rcvbuf = READ_ONCE(net->ipv4.sysctl_tcp_rmem[1]);
2744 	sk->sk_sndbuf = READ_ONCE(net->ipv4.sysctl_tcp_wmem[1]);
2745 
2746 	return 0;
2747 }
2748 
2749 static void __mptcp_clear_xmit(struct sock *sk)
2750 {
2751 	struct mptcp_sock *msk = mptcp_sk(sk);
2752 	struct mptcp_data_frag *dtmp, *dfrag;
2753 
2754 	WRITE_ONCE(msk->first_pending, NULL);
2755 	list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
2756 		dfrag_clear(sk, dfrag);
2757 }
2758 
2759 void mptcp_cancel_work(struct sock *sk)
2760 {
2761 	struct mptcp_sock *msk = mptcp_sk(sk);
2762 
2763 	if (cancel_work_sync(&msk->work))
2764 		__sock_put(sk);
2765 }
2766 
2767 void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
2768 {
2769 	lock_sock(ssk);
2770 
2771 	switch (ssk->sk_state) {
2772 	case TCP_LISTEN:
2773 		if (!(how & RCV_SHUTDOWN))
2774 			break;
2775 		fallthrough;
2776 	case TCP_SYN_SENT:
2777 		tcp_disconnect(ssk, O_NONBLOCK);
2778 		break;
2779 	default:
2780 		if (__mptcp_check_fallback(mptcp_sk(sk))) {
2781 			pr_debug("Fallback");
2782 			ssk->sk_shutdown |= how;
2783 			tcp_shutdown(ssk, how);
2784 		} else {
2785 			pr_debug("Sending DATA_FIN on subflow %p", ssk);
2786 			tcp_send_ack(ssk);
2787 			if (!mptcp_timer_pending(sk))
2788 				mptcp_reset_timer(sk);
2789 		}
2790 		break;
2791 	}
2792 
2793 	release_sock(ssk);
2794 }
2795 
2796 static const unsigned char new_state[16] = {
2797 	/* current state:     new state:      action:	*/
2798 	[0 /* (Invalid) */] = TCP_CLOSE,
2799 	[TCP_ESTABLISHED]   = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2800 	[TCP_SYN_SENT]      = TCP_CLOSE,
2801 	[TCP_SYN_RECV]      = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2802 	[TCP_FIN_WAIT1]     = TCP_FIN_WAIT1,
2803 	[TCP_FIN_WAIT2]     = TCP_FIN_WAIT2,
2804 	[TCP_TIME_WAIT]     = TCP_CLOSE,	/* should not happen ! */
2805 	[TCP_CLOSE]         = TCP_CLOSE,
2806 	[TCP_CLOSE_WAIT]    = TCP_LAST_ACK  | TCP_ACTION_FIN,
2807 	[TCP_LAST_ACK]      = TCP_LAST_ACK,
2808 	[TCP_LISTEN]        = TCP_CLOSE,
2809 	[TCP_CLOSING]       = TCP_CLOSING,
2810 	[TCP_NEW_SYN_RECV]  = TCP_CLOSE,	/* should not happen ! */
2811 };
2812 
2813 static int mptcp_close_state(struct sock *sk)
2814 {
2815 	int next = (int)new_state[sk->sk_state];
2816 	int ns = next & TCP_STATE_MASK;
2817 
2818 	inet_sk_state_store(sk, ns);
2819 
2820 	return next & TCP_ACTION_FIN;
2821 }
2822 
2823 static void __mptcp_check_send_data_fin(struct sock *sk)
2824 {
2825 	struct mptcp_subflow_context *subflow;
2826 	struct mptcp_sock *msk = mptcp_sk(sk);
2827 
2828 	pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2829 		 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2830 		 msk->snd_nxt, msk->write_seq);
2831 
2832 	/* we still need to enqueue subflows or not really shutting down,
2833 	 * skip this
2834 	 */
2835 	if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2836 	    mptcp_send_head(sk))
2837 		return;
2838 
2839 	WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2840 
2841 	/* fallback socket will not get data_fin/ack, can move to the next
2842 	 * state now
2843 	 */
2844 	if (__mptcp_check_fallback(msk)) {
2845 		WRITE_ONCE(msk->snd_una, msk->write_seq);
2846 		if ((1 << sk->sk_state) & (TCPF_CLOSING | TCPF_LAST_ACK)) {
2847 			inet_sk_state_store(sk, TCP_CLOSE);
2848 			mptcp_close_wake_up(sk);
2849 		} else if (sk->sk_state == TCP_FIN_WAIT1) {
2850 			inet_sk_state_store(sk, TCP_FIN_WAIT2);
2851 		}
2852 	}
2853 
2854 	mptcp_for_each_subflow(msk, subflow) {
2855 		struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2856 
2857 		mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
2858 	}
2859 }
2860 
2861 static void __mptcp_wr_shutdown(struct sock *sk)
2862 {
2863 	struct mptcp_sock *msk = mptcp_sk(sk);
2864 
2865 	pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2866 		 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2867 		 !!mptcp_send_head(sk));
2868 
2869 	/* will be ignored by fallback sockets */
2870 	WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2871 	WRITE_ONCE(msk->snd_data_fin_enable, 1);
2872 
2873 	__mptcp_check_send_data_fin(sk);
2874 }
2875 
2876 static void __mptcp_destroy_sock(struct sock *sk)
2877 {
2878 	struct mptcp_sock *msk = mptcp_sk(sk);
2879 
2880 	pr_debug("msk=%p", msk);
2881 
2882 	might_sleep();
2883 
2884 	mptcp_stop_timer(sk);
2885 	sk_stop_timer(sk, &sk->sk_timer);
2886 	msk->pm.status = 0;
2887 
2888 	sk->sk_prot->destroy(sk);
2889 
2890 	WARN_ON_ONCE(msk->rmem_fwd_alloc);
2891 	WARN_ON_ONCE(msk->rmem_released);
2892 	sk_stream_kill_queues(sk);
2893 	xfrm_sk_free_policy(sk);
2894 
2895 	sock_put(sk);
2896 }
2897 
2898 static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
2899 {
2900 	/* Concurrent splices from sk_receive_queue into receive_queue will
2901 	 * always show at least one non-empty queue when checked in this order.
2902 	 */
2903 	if (skb_queue_empty_lockless(&((struct sock *)msk)->sk_receive_queue) &&
2904 	    skb_queue_empty_lockless(&msk->receive_queue))
2905 		return 0;
2906 
2907 	return EPOLLIN | EPOLLRDNORM;
2908 }
2909 
2910 static void mptcp_listen_inuse_dec(struct sock *sk)
2911 {
2912 	if (inet_sk_state_load(sk) == TCP_LISTEN)
2913 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2914 }
2915 
2916 bool __mptcp_close(struct sock *sk, long timeout)
2917 {
2918 	struct mptcp_subflow_context *subflow;
2919 	struct mptcp_sock *msk = mptcp_sk(sk);
2920 	bool do_cancel_work = false;
2921 	int subflows_alive = 0;
2922 
2923 	sk->sk_shutdown = SHUTDOWN_MASK;
2924 
2925 	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
2926 		mptcp_listen_inuse_dec(sk);
2927 		inet_sk_state_store(sk, TCP_CLOSE);
2928 		goto cleanup;
2929 	}
2930 
2931 	if (mptcp_check_readable(msk)) {
2932 		/* the msk has read data, do the MPTCP equivalent of TCP reset */
2933 		inet_sk_state_store(sk, TCP_CLOSE);
2934 		mptcp_do_fastclose(sk);
2935 	} else if (mptcp_close_state(sk)) {
2936 		__mptcp_wr_shutdown(sk);
2937 	}
2938 
2939 	sk_stream_wait_close(sk, timeout);
2940 
2941 cleanup:
2942 	/* orphan all the subflows */
2943 	inet_csk(sk)->icsk_mtup.probe_timestamp = tcp_jiffies32;
2944 	mptcp_for_each_subflow(msk, subflow) {
2945 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2946 		bool slow = lock_sock_fast_nested(ssk);
2947 
2948 		subflows_alive += ssk->sk_state != TCP_CLOSE;
2949 
2950 		/* since the close timeout takes precedence on the fail one,
2951 		 * cancel the latter
2952 		 */
2953 		if (ssk == msk->first)
2954 			subflow->fail_tout = 0;
2955 
2956 		/* detach from the parent socket, but allow data_ready to
2957 		 * push incoming data into the mptcp stack, to properly ack it
2958 		 */
2959 		ssk->sk_socket = NULL;
2960 		ssk->sk_wq = NULL;
2961 		unlock_sock_fast(ssk, slow);
2962 	}
2963 	sock_orphan(sk);
2964 
2965 	/* all the subflows are closed, only timeout can change the msk
2966 	 * state, let's not keep resources busy for no reasons
2967 	 */
2968 	if (subflows_alive == 0)
2969 		inet_sk_state_store(sk, TCP_CLOSE);
2970 
2971 	sock_hold(sk);
2972 	pr_debug("msk=%p state=%d", sk, sk->sk_state);
2973 	if (msk->token)
2974 		mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
2975 
2976 	if (sk->sk_state == TCP_CLOSE) {
2977 		__mptcp_destroy_sock(sk);
2978 		do_cancel_work = true;
2979 	} else {
2980 		mptcp_reset_timeout(msk, 0);
2981 	}
2982 
2983 	return do_cancel_work;
2984 }
2985 
2986 static void mptcp_close(struct sock *sk, long timeout)
2987 {
2988 	bool do_cancel_work;
2989 
2990 	lock_sock(sk);
2991 
2992 	do_cancel_work = __mptcp_close(sk, timeout);
2993 	release_sock(sk);
2994 	if (do_cancel_work)
2995 		mptcp_cancel_work(sk);
2996 
2997 	sock_put(sk);
2998 }
2999 
3000 void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
3001 {
3002 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3003 	const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
3004 	struct ipv6_pinfo *msk6 = inet6_sk(msk);
3005 
3006 	msk->sk_v6_daddr = ssk->sk_v6_daddr;
3007 	msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
3008 
3009 	if (msk6 && ssk6) {
3010 		msk6->saddr = ssk6->saddr;
3011 		msk6->flow_label = ssk6->flow_label;
3012 	}
3013 #endif
3014 
3015 	inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
3016 	inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
3017 	inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
3018 	inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
3019 	inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
3020 	inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
3021 }
3022 
3023 static int mptcp_disconnect(struct sock *sk, int flags)
3024 {
3025 	struct mptcp_sock *msk = mptcp_sk(sk);
3026 
3027 	/* We are on the fastopen error path. We can't call straight into the
3028 	 * subflows cleanup code due to lock nesting (we are already under
3029 	 * msk->firstsocket lock). Do nothing and leave the cleanup to the
3030 	 * caller.
3031 	 */
3032 	if (msk->fastopening)
3033 		return 0;
3034 
3035 	mptcp_listen_inuse_dec(sk);
3036 	inet_sk_state_store(sk, TCP_CLOSE);
3037 
3038 	mptcp_stop_timer(sk);
3039 	sk_stop_timer(sk, &sk->sk_timer);
3040 
3041 	if (msk->token)
3042 		mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
3043 
3044 	/* msk->subflow is still intact, the following will not free the first
3045 	 * subflow
3046 	 */
3047 	mptcp_destroy_common(msk, MPTCP_CF_FASTCLOSE);
3048 	msk->last_snd = NULL;
3049 	WRITE_ONCE(msk->flags, 0);
3050 	msk->cb_flags = 0;
3051 	msk->push_pending = 0;
3052 	msk->recovery = false;
3053 	msk->can_ack = false;
3054 	msk->fully_established = false;
3055 	msk->rcv_data_fin = false;
3056 	msk->snd_data_fin_enable = false;
3057 	msk->rcv_fastclose = false;
3058 	msk->use_64bit_ack = false;
3059 	WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
3060 	mptcp_pm_data_reset(msk);
3061 	mptcp_ca_reset(sk);
3062 
3063 	sk->sk_shutdown = 0;
3064 	sk_error_report(sk);
3065 	return 0;
3066 }
3067 
3068 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3069 static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
3070 {
3071 	unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
3072 
3073 	return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
3074 }
3075 #endif
3076 
3077 struct sock *mptcp_sk_clone(const struct sock *sk,
3078 			    const struct mptcp_options_received *mp_opt,
3079 			    struct request_sock *req)
3080 {
3081 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
3082 	struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
3083 	struct mptcp_sock *msk;
3084 
3085 	if (!nsk)
3086 		return NULL;
3087 
3088 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3089 	if (nsk->sk_family == AF_INET6)
3090 		inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
3091 #endif
3092 
3093 	__mptcp_init_sock(nsk);
3094 
3095 	msk = mptcp_sk(nsk);
3096 	msk->local_key = subflow_req->local_key;
3097 	msk->token = subflow_req->token;
3098 	msk->subflow = NULL;
3099 	msk->in_accept_queue = 1;
3100 	WRITE_ONCE(msk->fully_established, false);
3101 	if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
3102 		WRITE_ONCE(msk->csum_enabled, true);
3103 
3104 	msk->write_seq = subflow_req->idsn + 1;
3105 	msk->snd_nxt = msk->write_seq;
3106 	msk->snd_una = msk->write_seq;
3107 	msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
3108 	msk->setsockopt_seq = mptcp_sk(sk)->setsockopt_seq;
3109 
3110 	sock_reset_flag(nsk, SOCK_RCU_FREE);
3111 	/* will be fully established after successful MPC subflow creation */
3112 	inet_sk_state_store(nsk, TCP_SYN_RECV);
3113 
3114 	security_inet_csk_clone(nsk, req);
3115 	bh_unlock_sock(nsk);
3116 
3117 	/* note: the newly allocated socket refcount is 2 now */
3118 	return nsk;
3119 }
3120 
3121 void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
3122 {
3123 	const struct tcp_sock *tp = tcp_sk(ssk);
3124 
3125 	msk->rcvq_space.copied = 0;
3126 	msk->rcvq_space.rtt_us = 0;
3127 
3128 	msk->rcvq_space.time = tp->tcp_mstamp;
3129 
3130 	/* initial rcv_space offering made to peer */
3131 	msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
3132 				      TCP_INIT_CWND * tp->advmss);
3133 	if (msk->rcvq_space.space == 0)
3134 		msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
3135 
3136 	WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
3137 }
3138 
3139 static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
3140 				 bool kern)
3141 {
3142 	struct mptcp_sock *msk = mptcp_sk(sk);
3143 	struct socket *listener;
3144 	struct sock *newsk;
3145 
3146 	listener = __mptcp_nmpc_socket(msk);
3147 	if (WARN_ON_ONCE(!listener)) {
3148 		*err = -EINVAL;
3149 		return NULL;
3150 	}
3151 
3152 	pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
3153 	newsk = inet_csk_accept(listener->sk, flags, err, kern);
3154 	if (!newsk)
3155 		return NULL;
3156 
3157 	pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
3158 	if (sk_is_mptcp(newsk)) {
3159 		struct mptcp_subflow_context *subflow;
3160 		struct sock *new_mptcp_sock;
3161 
3162 		subflow = mptcp_subflow_ctx(newsk);
3163 		new_mptcp_sock = subflow->conn;
3164 
3165 		/* is_mptcp should be false if subflow->conn is missing, see
3166 		 * subflow_syn_recv_sock()
3167 		 */
3168 		if (WARN_ON_ONCE(!new_mptcp_sock)) {
3169 			tcp_sk(newsk)->is_mptcp = 0;
3170 			goto out;
3171 		}
3172 
3173 		newsk = new_mptcp_sock;
3174 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
3175 	} else {
3176 		MPTCP_INC_STATS(sock_net(sk),
3177 				MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
3178 	}
3179 
3180 out:
3181 	newsk->sk_kern_sock = kern;
3182 	return newsk;
3183 }
3184 
3185 void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags)
3186 {
3187 	struct mptcp_subflow_context *subflow, *tmp;
3188 	struct sock *sk = (struct sock *)msk;
3189 
3190 	__mptcp_clear_xmit(sk);
3191 
3192 	/* join list will be eventually flushed (with rst) at sock lock release time */
3193 	mptcp_for_each_subflow_safe(msk, subflow, tmp)
3194 		__mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow), subflow, flags);
3195 
3196 	/* move to sk_receive_queue, sk_stream_kill_queues will purge it */
3197 	mptcp_data_lock(sk);
3198 	skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
3199 	__skb_queue_purge(&sk->sk_receive_queue);
3200 	skb_rbtree_purge(&msk->out_of_order_queue);
3201 	mptcp_data_unlock(sk);
3202 
3203 	/* move all the rx fwd alloc into the sk_mem_reclaim_final in
3204 	 * inet_sock_destruct() will dispose it
3205 	 */
3206 	sk->sk_forward_alloc += msk->rmem_fwd_alloc;
3207 	msk->rmem_fwd_alloc = 0;
3208 	mptcp_token_destroy(msk);
3209 	mptcp_pm_free_anno_list(msk);
3210 	mptcp_free_local_addr_list(msk);
3211 }
3212 
3213 static void mptcp_destroy(struct sock *sk)
3214 {
3215 	struct mptcp_sock *msk = mptcp_sk(sk);
3216 
3217 	/* clears msk->subflow, allowing the following to close
3218 	 * even the initial subflow
3219 	 */
3220 	mptcp_dispose_initial_subflow(msk);
3221 	mptcp_destroy_common(msk, 0);
3222 	sk_sockets_allocated_dec(sk);
3223 }
3224 
3225 void __mptcp_data_acked(struct sock *sk)
3226 {
3227 	if (!sock_owned_by_user(sk))
3228 		__mptcp_clean_una(sk);
3229 	else
3230 		__set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->cb_flags);
3231 
3232 	if (mptcp_pending_data_fin_ack(sk))
3233 		mptcp_schedule_work(sk);
3234 }
3235 
3236 void __mptcp_check_push(struct sock *sk, struct sock *ssk)
3237 {
3238 	if (!mptcp_send_head(sk))
3239 		return;
3240 
3241 	if (!sock_owned_by_user(sk))
3242 		__mptcp_subflow_push_pending(sk, ssk, false);
3243 	else
3244 		__set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
3245 }
3246 
3247 #define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
3248 				      BIT(MPTCP_RETRANSMIT) | \
3249 				      BIT(MPTCP_FLUSH_JOIN_LIST))
3250 
3251 /* processes deferred events and flush wmem */
3252 static void mptcp_release_cb(struct sock *sk)
3253 	__must_hold(&sk->sk_lock.slock)
3254 {
3255 	struct mptcp_sock *msk = mptcp_sk(sk);
3256 
3257 	for (;;) {
3258 		unsigned long flags = (msk->cb_flags & MPTCP_FLAGS_PROCESS_CTX_NEED) |
3259 				      msk->push_pending;
3260 		if (!flags)
3261 			break;
3262 
3263 		/* the following actions acquire the subflow socket lock
3264 		 *
3265 		 * 1) can't be invoked in atomic scope
3266 		 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
3267 		 *    datapath acquires the msk socket spinlock while helding
3268 		 *    the subflow socket lock
3269 		 */
3270 		msk->push_pending = 0;
3271 		msk->cb_flags &= ~flags;
3272 		spin_unlock_bh(&sk->sk_lock.slock);
3273 		if (flags & BIT(MPTCP_FLUSH_JOIN_LIST))
3274 			__mptcp_flush_join_list(sk);
3275 		if (flags & BIT(MPTCP_PUSH_PENDING))
3276 			__mptcp_push_pending(sk, 0);
3277 		if (flags & BIT(MPTCP_RETRANSMIT))
3278 			__mptcp_retrans(sk);
3279 
3280 		cond_resched();
3281 		spin_lock_bh(&sk->sk_lock.slock);
3282 	}
3283 
3284 	if (__test_and_clear_bit(MPTCP_CLEAN_UNA, &msk->cb_flags))
3285 		__mptcp_clean_una_wakeup(sk);
3286 	if (unlikely(&msk->cb_flags)) {
3287 		/* be sure to set the current sk state before tacking actions
3288 		 * depending on sk_state, that is processing MPTCP_ERROR_REPORT
3289 		 */
3290 		if (__test_and_clear_bit(MPTCP_CONNECTED, &msk->cb_flags))
3291 			__mptcp_set_connected(sk);
3292 		if (__test_and_clear_bit(MPTCP_ERROR_REPORT, &msk->cb_flags))
3293 			__mptcp_error_report(sk);
3294 		if (__test_and_clear_bit(MPTCP_RESET_SCHEDULER, &msk->cb_flags))
3295 			msk->last_snd = NULL;
3296 	}
3297 
3298 	__mptcp_update_rmem(sk);
3299 }
3300 
3301 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
3302  * TCP can't schedule delack timer before the subflow is fully established.
3303  * MPTCP uses the delack timer to do 3rd ack retransmissions
3304  */
3305 static void schedule_3rdack_retransmission(struct sock *ssk)
3306 {
3307 	struct inet_connection_sock *icsk = inet_csk(ssk);
3308 	struct tcp_sock *tp = tcp_sk(ssk);
3309 	unsigned long timeout;
3310 
3311 	if (mptcp_subflow_ctx(ssk)->fully_established)
3312 		return;
3313 
3314 	/* reschedule with a timeout above RTT, as we must look only for drop */
3315 	if (tp->srtt_us)
3316 		timeout = usecs_to_jiffies(tp->srtt_us >> (3 - 1));
3317 	else
3318 		timeout = TCP_TIMEOUT_INIT;
3319 	timeout += jiffies;
3320 
3321 	WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
3322 	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
3323 	icsk->icsk_ack.timeout = timeout;
3324 	sk_reset_timer(ssk, &icsk->icsk_delack_timer, timeout);
3325 }
3326 
3327 void mptcp_subflow_process_delegated(struct sock *ssk)
3328 {
3329 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3330 	struct sock *sk = subflow->conn;
3331 
3332 	if (test_bit(MPTCP_DELEGATE_SEND, &subflow->delegated_status)) {
3333 		mptcp_data_lock(sk);
3334 		if (!sock_owned_by_user(sk))
3335 			__mptcp_subflow_push_pending(sk, ssk, true);
3336 		else
3337 			__set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
3338 		mptcp_data_unlock(sk);
3339 		mptcp_subflow_delegated_done(subflow, MPTCP_DELEGATE_SEND);
3340 	}
3341 	if (test_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status)) {
3342 		schedule_3rdack_retransmission(ssk);
3343 		mptcp_subflow_delegated_done(subflow, MPTCP_DELEGATE_ACK);
3344 	}
3345 }
3346 
3347 static int mptcp_hash(struct sock *sk)
3348 {
3349 	/* should never be called,
3350 	 * we hash the TCP subflows not the master socket
3351 	 */
3352 	WARN_ON_ONCE(1);
3353 	return 0;
3354 }
3355 
3356 static void mptcp_unhash(struct sock *sk)
3357 {
3358 	/* called from sk_common_release(), but nothing to do here */
3359 }
3360 
3361 static int mptcp_get_port(struct sock *sk, unsigned short snum)
3362 {
3363 	struct mptcp_sock *msk = mptcp_sk(sk);
3364 	struct socket *ssock;
3365 
3366 	ssock = __mptcp_nmpc_socket(msk);
3367 	pr_debug("msk=%p, subflow=%p", msk, ssock);
3368 	if (WARN_ON_ONCE(!ssock))
3369 		return -EINVAL;
3370 
3371 	return inet_csk_get_port(ssock->sk, snum);
3372 }
3373 
3374 void mptcp_finish_connect(struct sock *ssk)
3375 {
3376 	struct mptcp_subflow_context *subflow;
3377 	struct mptcp_sock *msk;
3378 	struct sock *sk;
3379 
3380 	subflow = mptcp_subflow_ctx(ssk);
3381 	sk = subflow->conn;
3382 	msk = mptcp_sk(sk);
3383 
3384 	pr_debug("msk=%p, token=%u", sk, subflow->token);
3385 
3386 	subflow->map_seq = subflow->iasn;
3387 	subflow->map_subflow_seq = 1;
3388 
3389 	/* the socket is not connected yet, no msk/subflow ops can access/race
3390 	 * accessing the field below
3391 	 */
3392 	WRITE_ONCE(msk->local_key, subflow->local_key);
3393 	WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
3394 	WRITE_ONCE(msk->snd_nxt, msk->write_seq);
3395 	WRITE_ONCE(msk->snd_una, msk->write_seq);
3396 
3397 	mptcp_pm_new_connection(msk, ssk, 0);
3398 
3399 	mptcp_rcv_space_init(msk, ssk);
3400 }
3401 
3402 void mptcp_sock_graft(struct sock *sk, struct socket *parent)
3403 {
3404 	write_lock_bh(&sk->sk_callback_lock);
3405 	rcu_assign_pointer(sk->sk_wq, &parent->wq);
3406 	sk_set_socket(sk, parent);
3407 	sk->sk_uid = SOCK_INODE(parent)->i_uid;
3408 	write_unlock_bh(&sk->sk_callback_lock);
3409 }
3410 
3411 bool mptcp_finish_join(struct sock *ssk)
3412 {
3413 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3414 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3415 	struct sock *parent = (void *)msk;
3416 	bool ret = true;
3417 
3418 	pr_debug("msk=%p, subflow=%p", msk, subflow);
3419 
3420 	/* mptcp socket already closing? */
3421 	if (!mptcp_is_fully_established(parent)) {
3422 		subflow->reset_reason = MPTCP_RST_EMPTCP;
3423 		return false;
3424 	}
3425 
3426 	if (!list_empty(&subflow->node))
3427 		goto out;
3428 
3429 	if (!mptcp_pm_allow_new_subflow(msk))
3430 		goto err_prohibited;
3431 
3432 	/* active connections are already on conn_list.
3433 	 * If we can't acquire msk socket lock here, let the release callback
3434 	 * handle it
3435 	 */
3436 	mptcp_data_lock(parent);
3437 	if (!sock_owned_by_user(parent)) {
3438 		ret = __mptcp_finish_join(msk, ssk);
3439 		if (ret) {
3440 			sock_hold(ssk);
3441 			list_add_tail(&subflow->node, &msk->conn_list);
3442 		}
3443 	} else {
3444 		sock_hold(ssk);
3445 		list_add_tail(&subflow->node, &msk->join_list);
3446 		__set_bit(MPTCP_FLUSH_JOIN_LIST, &msk->cb_flags);
3447 	}
3448 	mptcp_data_unlock(parent);
3449 
3450 	if (!ret) {
3451 err_prohibited:
3452 		subflow->reset_reason = MPTCP_RST_EPROHIBIT;
3453 		return false;
3454 	}
3455 
3456 	subflow->map_seq = READ_ONCE(msk->ack_seq);
3457 	WRITE_ONCE(msk->allow_infinite_fallback, false);
3458 
3459 out:
3460 	mptcp_event(MPTCP_EVENT_SUB_ESTABLISHED, msk, ssk, GFP_ATOMIC);
3461 	return true;
3462 }
3463 
3464 static void mptcp_shutdown(struct sock *sk, int how)
3465 {
3466 	pr_debug("sk=%p, how=%d", sk, how);
3467 
3468 	if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3469 		__mptcp_wr_shutdown(sk);
3470 }
3471 
3472 static int mptcp_forward_alloc_get(const struct sock *sk)
3473 {
3474 	return sk->sk_forward_alloc + mptcp_sk(sk)->rmem_fwd_alloc;
3475 }
3476 
3477 static int mptcp_ioctl_outq(const struct mptcp_sock *msk, u64 v)
3478 {
3479 	const struct sock *sk = (void *)msk;
3480 	u64 delta;
3481 
3482 	if (sk->sk_state == TCP_LISTEN)
3483 		return -EINVAL;
3484 
3485 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
3486 		return 0;
3487 
3488 	delta = msk->write_seq - v;
3489 	if (__mptcp_check_fallback(msk) && msk->first) {
3490 		struct tcp_sock *tp = tcp_sk(msk->first);
3491 
3492 		/* the first subflow is disconnected after close - see
3493 		 * __mptcp_close_ssk(). tcp_disconnect() moves the write_seq
3494 		 * so ignore that status, too.
3495 		 */
3496 		if (!((1 << msk->first->sk_state) &
3497 		      (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)))
3498 			delta += READ_ONCE(tp->write_seq) - tp->snd_una;
3499 	}
3500 	if (delta > INT_MAX)
3501 		delta = INT_MAX;
3502 
3503 	return (int)delta;
3504 }
3505 
3506 static int mptcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
3507 {
3508 	struct mptcp_sock *msk = mptcp_sk(sk);
3509 	bool slow;
3510 	int answ;
3511 
3512 	switch (cmd) {
3513 	case SIOCINQ:
3514 		if (sk->sk_state == TCP_LISTEN)
3515 			return -EINVAL;
3516 
3517 		lock_sock(sk);
3518 		__mptcp_move_skbs(msk);
3519 		answ = mptcp_inq_hint(sk);
3520 		release_sock(sk);
3521 		break;
3522 	case SIOCOUTQ:
3523 		slow = lock_sock_fast(sk);
3524 		answ = mptcp_ioctl_outq(msk, READ_ONCE(msk->snd_una));
3525 		unlock_sock_fast(sk, slow);
3526 		break;
3527 	case SIOCOUTQNSD:
3528 		slow = lock_sock_fast(sk);
3529 		answ = mptcp_ioctl_outq(msk, msk->snd_nxt);
3530 		unlock_sock_fast(sk, slow);
3531 		break;
3532 	default:
3533 		return -ENOIOCTLCMD;
3534 	}
3535 
3536 	return put_user(answ, (int __user *)arg);
3537 }
3538 
3539 static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3540 					 struct mptcp_subflow_context *subflow)
3541 {
3542 	subflow->request_mptcp = 0;
3543 	__mptcp_do_fallback(msk);
3544 }
3545 
3546 static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
3547 {
3548 	struct mptcp_subflow_context *subflow;
3549 	struct mptcp_sock *msk = mptcp_sk(sk);
3550 	struct socket *ssock;
3551 	int err = -EINVAL;
3552 
3553 	ssock = __mptcp_nmpc_socket(msk);
3554 	if (!ssock)
3555 		return -EINVAL;
3556 
3557 	mptcp_token_destroy(msk);
3558 	inet_sk_state_store(sk, TCP_SYN_SENT);
3559 	subflow = mptcp_subflow_ctx(ssock->sk);
3560 #ifdef CONFIG_TCP_MD5SIG
3561 	/* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3562 	 * TCP option space.
3563 	 */
3564 	if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
3565 		mptcp_subflow_early_fallback(msk, subflow);
3566 #endif
3567 	if (subflow->request_mptcp && mptcp_token_new_connect(ssock->sk)) {
3568 		MPTCP_INC_STATS(sock_net(ssock->sk), MPTCP_MIB_TOKENFALLBACKINIT);
3569 		mptcp_subflow_early_fallback(msk, subflow);
3570 	}
3571 	if (likely(!__mptcp_check_fallback(msk)))
3572 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVE);
3573 
3574 	/* if reaching here via the fastopen/sendmsg path, the caller already
3575 	 * acquired the subflow socket lock, too.
3576 	 */
3577 	if (msk->fastopening)
3578 		err = __inet_stream_connect(ssock, uaddr, addr_len, msk->connect_flags, 1);
3579 	else
3580 		err = inet_stream_connect(ssock, uaddr, addr_len, msk->connect_flags);
3581 	inet_sk(sk)->defer_connect = inet_sk(ssock->sk)->defer_connect;
3582 
3583 	/* on successful connect, the msk state will be moved to established by
3584 	 * subflow_finish_connect()
3585 	 */
3586 	if (unlikely(err && err != -EINPROGRESS)) {
3587 		inet_sk_state_store(sk, inet_sk_state_load(ssock->sk));
3588 		return err;
3589 	}
3590 
3591 	mptcp_copy_inaddrs(sk, ssock->sk);
3592 
3593 	/* unblocking connect, mptcp-level inet_stream_connect will error out
3594 	 * without changing the socket state, update it here.
3595 	 */
3596 	if (err == -EINPROGRESS)
3597 		sk->sk_socket->state = ssock->state;
3598 	return err;
3599 }
3600 
3601 static struct proto mptcp_prot = {
3602 	.name		= "MPTCP",
3603 	.owner		= THIS_MODULE,
3604 	.init		= mptcp_init_sock,
3605 	.connect	= mptcp_connect,
3606 	.disconnect	= mptcp_disconnect,
3607 	.close		= mptcp_close,
3608 	.accept		= mptcp_accept,
3609 	.setsockopt	= mptcp_setsockopt,
3610 	.getsockopt	= mptcp_getsockopt,
3611 	.shutdown	= mptcp_shutdown,
3612 	.destroy	= mptcp_destroy,
3613 	.sendmsg	= mptcp_sendmsg,
3614 	.ioctl		= mptcp_ioctl,
3615 	.recvmsg	= mptcp_recvmsg,
3616 	.release_cb	= mptcp_release_cb,
3617 	.hash		= mptcp_hash,
3618 	.unhash		= mptcp_unhash,
3619 	.get_port	= mptcp_get_port,
3620 	.forward_alloc_get	= mptcp_forward_alloc_get,
3621 	.sockets_allocated	= &mptcp_sockets_allocated,
3622 
3623 	.memory_allocated	= &tcp_memory_allocated,
3624 	.per_cpu_fw_alloc	= &tcp_memory_per_cpu_fw_alloc,
3625 
3626 	.memory_pressure	= &tcp_memory_pressure,
3627 	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
3628 	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
3629 	.sysctl_mem	= sysctl_tcp_mem,
3630 	.obj_size	= sizeof(struct mptcp_sock),
3631 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
3632 	.no_autobind	= true,
3633 };
3634 
3635 static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3636 {
3637 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
3638 	struct socket *ssock;
3639 	int err;
3640 
3641 	lock_sock(sock->sk);
3642 	ssock = __mptcp_nmpc_socket(msk);
3643 	if (!ssock) {
3644 		err = -EINVAL;
3645 		goto unlock;
3646 	}
3647 
3648 	err = ssock->ops->bind(ssock, uaddr, addr_len);
3649 	if (!err)
3650 		mptcp_copy_inaddrs(sock->sk, ssock->sk);
3651 
3652 unlock:
3653 	release_sock(sock->sk);
3654 	return err;
3655 }
3656 
3657 static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
3658 				int addr_len, int flags)
3659 {
3660 	int ret;
3661 
3662 	lock_sock(sock->sk);
3663 	mptcp_sk(sock->sk)->connect_flags = flags;
3664 	ret = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
3665 	release_sock(sock->sk);
3666 	return ret;
3667 }
3668 
3669 static int mptcp_listen(struct socket *sock, int backlog)
3670 {
3671 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
3672 	struct sock *sk = sock->sk;
3673 	struct socket *ssock;
3674 	int err;
3675 
3676 	pr_debug("msk=%p", msk);
3677 
3678 	lock_sock(sk);
3679 	ssock = __mptcp_nmpc_socket(msk);
3680 	if (!ssock) {
3681 		err = -EINVAL;
3682 		goto unlock;
3683 	}
3684 
3685 	mptcp_token_destroy(msk);
3686 	inet_sk_state_store(sk, TCP_LISTEN);
3687 	sock_set_flag(sk, SOCK_RCU_FREE);
3688 
3689 	err = ssock->ops->listen(ssock, backlog);
3690 	inet_sk_state_store(sk, inet_sk_state_load(ssock->sk));
3691 	if (!err) {
3692 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
3693 		mptcp_copy_inaddrs(sk, ssock->sk);
3694 	}
3695 
3696 	mptcp_event_pm_listener(ssock->sk, MPTCP_EVENT_LISTENER_CREATED);
3697 
3698 unlock:
3699 	release_sock(sk);
3700 	return err;
3701 }
3702 
3703 static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3704 			       int flags, bool kern)
3705 {
3706 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
3707 	struct socket *ssock;
3708 	int err;
3709 
3710 	pr_debug("msk=%p", msk);
3711 
3712 	ssock = __mptcp_nmpc_socket(msk);
3713 	if (!ssock)
3714 		return -EINVAL;
3715 
3716 	err = ssock->ops->accept(sock, newsock, flags, kern);
3717 	if (err == 0 && !mptcp_is_tcpsk(newsock->sk)) {
3718 		struct mptcp_sock *msk = mptcp_sk(newsock->sk);
3719 		struct mptcp_subflow_context *subflow;
3720 		struct sock *newsk = newsock->sk;
3721 
3722 		set_bit(SOCK_CUSTOM_SOCKOPT, &newsock->flags);
3723 		msk->in_accept_queue = 0;
3724 
3725 		lock_sock(newsk);
3726 
3727 		/* set ssk->sk_socket of accept()ed flows to mptcp socket.
3728 		 * This is needed so NOSPACE flag can be set from tcp stack.
3729 		 */
3730 		mptcp_for_each_subflow(msk, subflow) {
3731 			struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3732 
3733 			if (!ssk->sk_socket)
3734 				mptcp_sock_graft(ssk, newsock);
3735 		}
3736 		release_sock(newsk);
3737 	}
3738 
3739 	return err;
3740 }
3741 
3742 static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3743 {
3744 	struct sock *sk = (struct sock *)msk;
3745 
3746 	if (unlikely(sk->sk_shutdown & SEND_SHUTDOWN))
3747 		return EPOLLOUT | EPOLLWRNORM;
3748 
3749 	if (sk_stream_is_writeable(sk))
3750 		return EPOLLOUT | EPOLLWRNORM;
3751 
3752 	mptcp_set_nospace(sk);
3753 	smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3754 	if (sk_stream_is_writeable(sk))
3755 		return EPOLLOUT | EPOLLWRNORM;
3756 
3757 	return 0;
3758 }
3759 
3760 static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3761 			   struct poll_table_struct *wait)
3762 {
3763 	struct sock *sk = sock->sk;
3764 	struct mptcp_sock *msk;
3765 	__poll_t mask = 0;
3766 	int state;
3767 
3768 	msk = mptcp_sk(sk);
3769 	sock_poll_wait(file, sock, wait);
3770 
3771 	state = inet_sk_state_load(sk);
3772 	pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
3773 	if (state == TCP_LISTEN) {
3774 		if (WARN_ON_ONCE(!msk->subflow || !msk->subflow->sk))
3775 			return 0;
3776 
3777 		return inet_csk_listen_poll(msk->subflow->sk);
3778 	}
3779 
3780 	if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3781 		mask |= mptcp_check_readable(msk);
3782 		mask |= mptcp_check_writeable(msk);
3783 	} else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) {
3784 		/* cf tcp_poll() note about TFO */
3785 		mask |= EPOLLOUT | EPOLLWRNORM;
3786 	}
3787 	if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
3788 		mask |= EPOLLHUP;
3789 	if (sk->sk_shutdown & RCV_SHUTDOWN)
3790 		mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3791 
3792 	/* This barrier is coupled with smp_wmb() in __mptcp_error_report() */
3793 	smp_rmb();
3794 	if (sk->sk_err)
3795 		mask |= EPOLLERR;
3796 
3797 	return mask;
3798 }
3799 
3800 static const struct proto_ops mptcp_stream_ops = {
3801 	.family		   = PF_INET,
3802 	.owner		   = THIS_MODULE,
3803 	.release	   = inet_release,
3804 	.bind		   = mptcp_bind,
3805 	.connect	   = mptcp_stream_connect,
3806 	.socketpair	   = sock_no_socketpair,
3807 	.accept		   = mptcp_stream_accept,
3808 	.getname	   = inet_getname,
3809 	.poll		   = mptcp_poll,
3810 	.ioctl		   = inet_ioctl,
3811 	.gettstamp	   = sock_gettstamp,
3812 	.listen		   = mptcp_listen,
3813 	.shutdown	   = inet_shutdown,
3814 	.setsockopt	   = sock_common_setsockopt,
3815 	.getsockopt	   = sock_common_getsockopt,
3816 	.sendmsg	   = inet_sendmsg,
3817 	.recvmsg	   = inet_recvmsg,
3818 	.mmap		   = sock_no_mmap,
3819 	.sendpage	   = inet_sendpage,
3820 };
3821 
3822 static struct inet_protosw mptcp_protosw = {
3823 	.type		= SOCK_STREAM,
3824 	.protocol	= IPPROTO_MPTCP,
3825 	.prot		= &mptcp_prot,
3826 	.ops		= &mptcp_stream_ops,
3827 	.flags		= INET_PROTOSW_ICSK,
3828 };
3829 
3830 static int mptcp_napi_poll(struct napi_struct *napi, int budget)
3831 {
3832 	struct mptcp_delegated_action *delegated;
3833 	struct mptcp_subflow_context *subflow;
3834 	int work_done = 0;
3835 
3836 	delegated = container_of(napi, struct mptcp_delegated_action, napi);
3837 	while ((subflow = mptcp_subflow_delegated_next(delegated)) != NULL) {
3838 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3839 
3840 		bh_lock_sock_nested(ssk);
3841 		if (!sock_owned_by_user(ssk) &&
3842 		    mptcp_subflow_has_delegated_action(subflow))
3843 			mptcp_subflow_process_delegated(ssk);
3844 		/* ... elsewhere tcp_release_cb_override already processed
3845 		 * the action or will do at next release_sock().
3846 		 * In both case must dequeue the subflow here - on the same
3847 		 * CPU that scheduled it.
3848 		 */
3849 		bh_unlock_sock(ssk);
3850 		sock_put(ssk);
3851 
3852 		if (++work_done == budget)
3853 			return budget;
3854 	}
3855 
3856 	/* always provide a 0 'work_done' argument, so that napi_complete_done
3857 	 * will not try accessing the NULL napi->dev ptr
3858 	 */
3859 	napi_complete_done(napi, 0);
3860 	return work_done;
3861 }
3862 
3863 void __init mptcp_proto_init(void)
3864 {
3865 	struct mptcp_delegated_action *delegated;
3866 	int cpu;
3867 
3868 	mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
3869 
3870 	if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
3871 		panic("Failed to allocate MPTCP pcpu counter\n");
3872 
3873 	init_dummy_netdev(&mptcp_napi_dev);
3874 	for_each_possible_cpu(cpu) {
3875 		delegated = per_cpu_ptr(&mptcp_delegated_actions, cpu);
3876 		INIT_LIST_HEAD(&delegated->head);
3877 		netif_napi_add_tx(&mptcp_napi_dev, &delegated->napi,
3878 				  mptcp_napi_poll);
3879 		napi_enable(&delegated->napi);
3880 	}
3881 
3882 	mptcp_subflow_init();
3883 	mptcp_pm_init();
3884 	mptcp_token_init();
3885 
3886 	if (proto_register(&mptcp_prot, 1) != 0)
3887 		panic("Failed to register MPTCP proto.\n");
3888 
3889 	inet_register_protosw(&mptcp_protosw);
3890 
3891 	BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
3892 }
3893 
3894 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3895 static const struct proto_ops mptcp_v6_stream_ops = {
3896 	.family		   = PF_INET6,
3897 	.owner		   = THIS_MODULE,
3898 	.release	   = inet6_release,
3899 	.bind		   = mptcp_bind,
3900 	.connect	   = mptcp_stream_connect,
3901 	.socketpair	   = sock_no_socketpair,
3902 	.accept		   = mptcp_stream_accept,
3903 	.getname	   = inet6_getname,
3904 	.poll		   = mptcp_poll,
3905 	.ioctl		   = inet6_ioctl,
3906 	.gettstamp	   = sock_gettstamp,
3907 	.listen		   = mptcp_listen,
3908 	.shutdown	   = inet_shutdown,
3909 	.setsockopt	   = sock_common_setsockopt,
3910 	.getsockopt	   = sock_common_getsockopt,
3911 	.sendmsg	   = inet6_sendmsg,
3912 	.recvmsg	   = inet6_recvmsg,
3913 	.mmap		   = sock_no_mmap,
3914 	.sendpage	   = inet_sendpage,
3915 #ifdef CONFIG_COMPAT
3916 	.compat_ioctl	   = inet6_compat_ioctl,
3917 #endif
3918 };
3919 
3920 static struct proto mptcp_v6_prot;
3921 
3922 static struct inet_protosw mptcp_v6_protosw = {
3923 	.type		= SOCK_STREAM,
3924 	.protocol	= IPPROTO_MPTCP,
3925 	.prot		= &mptcp_v6_prot,
3926 	.ops		= &mptcp_v6_stream_ops,
3927 	.flags		= INET_PROTOSW_ICSK,
3928 };
3929 
3930 int __init mptcp_proto_v6_init(void)
3931 {
3932 	int err;
3933 
3934 	mptcp_v6_prot = mptcp_prot;
3935 	strcpy(mptcp_v6_prot.name, "MPTCPv6");
3936 	mptcp_v6_prot.slab = NULL;
3937 	mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
3938 
3939 	err = proto_register(&mptcp_v6_prot, 1);
3940 	if (err)
3941 		return err;
3942 
3943 	err = inet6_register_protosw(&mptcp_v6_protosw);
3944 	if (err)
3945 		proto_unregister(&mptcp_v6_prot);
3946 
3947 	return err;
3948 }
3949 #endif
3950